Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 | 2 | /* |
722c9a0c | 3 | * NET3 Protocol independent device support routines. |
1da177e4 | 4 | * |
1da177e4 | 5 | * Derived from the non IP parts of dev.c 1.0.19 |
722c9a0c | 6 | * Authors: Ross Biro |
1da177e4 LT |
7 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
8 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
9 | * | |
10 | * Additional Authors: | |
11 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
12 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
13 | * David Hinds <dahinds@users.sourceforge.net> | |
14 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
15 | * Adam Sulmicki <adam@cfar.umd.edu> | |
16 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
17 | * | |
18 | * Changes: | |
19 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
722c9a0c | 20 | * to 2 if register_netdev gets called |
21 | * before net_dev_init & also removed a | |
22 | * few lines of code in the process. | |
1da177e4 LT |
23 | * Alan Cox : device private ioctl copies fields back. |
24 | * Alan Cox : Transmit queue code does relevant | |
25 | * stunts to keep the queue safe. | |
26 | * Alan Cox : Fixed double lock. | |
27 | * Alan Cox : Fixed promisc NULL pointer trap | |
28 | * ???????? : Support the full private ioctl range | |
29 | * Alan Cox : Moved ioctl permission check into | |
30 | * drivers | |
31 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
32 | * Alan Cox : 100 backlog just doesn't cut it when | |
33 | * you start doing multicast video 8) | |
34 | * Alan Cox : Rewrote net_bh and list manager. | |
722c9a0c | 35 | * Alan Cox : Fix ETH_P_ALL echoback lengths. |
1da177e4 LT |
36 | * Alan Cox : Took out transmit every packet pass |
37 | * Saved a few bytes in the ioctl handler | |
38 | * Alan Cox : Network driver sets packet type before | |
39 | * calling netif_rx. Saves a function | |
40 | * call a packet. | |
41 | * Alan Cox : Hashed net_bh() | |
42 | * Richard Kooijman: Timestamp fixes. | |
43 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
44 | * Alan Cox : Device lock protection. | |
722c9a0c | 45 | * Alan Cox : Fixed nasty side effect of device close |
1da177e4 LT |
46 | * changes. |
47 | * Rudi Cilibrasi : Pass the right thing to | |
48 | * set_mac_address() | |
49 | * Dave Miller : 32bit quantity for the device lock to | |
50 | * make it work out on a Sparc. | |
51 | * Bjorn Ekwall : Added KERNELD hack. | |
52 | * Alan Cox : Cleaned up the backlog initialise. | |
53 | * Craig Metz : SIOCGIFCONF fix if space for under | |
54 | * 1 device. | |
55 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
56 | * is no device open function. | |
57 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
58 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
59 | * Cyrus Durgin : Cleaned for KMOD | |
60 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
61 | * A network device unload needs to purge | |
62 | * the backlog queue. | |
63 | * Paul Rusty Russell : SIOCSIFNAME | |
64 | * Pekka Riikonen : Netdev boot-time settings code | |
65 | * Andrew Morton : Make unregister_netdevice wait | |
722c9a0c | 66 | * indefinitely on dev->refcnt |
67 | * J Hadi Salim : - Backlog queue sampling | |
1da177e4 LT |
68 | * - netif_rx() feedback |
69 | */ | |
70 | ||
7c0f6ba6 | 71 | #include <linux/uaccess.h> |
aabb4af9 | 72 | #include <linux/bitmap.h> |
4fc268d2 | 73 | #include <linux/capability.h> |
1da177e4 LT |
74 | #include <linux/cpu.h> |
75 | #include <linux/types.h> | |
76 | #include <linux/kernel.h> | |
08e9897d | 77 | #include <linux/hash.h> |
5a0e3ad6 | 78 | #include <linux/slab.h> |
1da177e4 | 79 | #include <linux/sched.h> |
2f3c7195 | 80 | #include <linux/sched/isolation.h> |
f1083048 | 81 | #include <linux/sched/mm.h> |
dad6b977 | 82 | #include <linux/smpboot.h> |
4a3e2f71 | 83 | #include <linux/mutex.h> |
11d6011c | 84 | #include <linux/rwsem.h> |
1da177e4 LT |
85 | #include <linux/string.h> |
86 | #include <linux/mm.h> | |
87 | #include <linux/socket.h> | |
88 | #include <linux/sockios.h> | |
89 | #include <linux/errno.h> | |
90 | #include <linux/interrupt.h> | |
91 | #include <linux/if_ether.h> | |
92 | #include <linux/netdevice.h> | |
93 | #include <linux/etherdevice.h> | |
0187bdfb | 94 | #include <linux/ethtool.h> |
1da177e4 | 95 | #include <linux/skbuff.h> |
29863d41 | 96 | #include <linux/kthread.h> |
a7862b45 | 97 | #include <linux/bpf.h> |
b5cdae32 | 98 | #include <linux/bpf_trace.h> |
457c4cbc | 99 | #include <net/net_namespace.h> |
1da177e4 | 100 | #include <net/sock.h> |
02d62e86 | 101 | #include <net/busy_poll.h> |
1da177e4 | 102 | #include <linux/rtnetlink.h> |
1da177e4 | 103 | #include <linux/stat.h> |
b14a9fc4 | 104 | #include <net/dsa.h> |
1da177e4 | 105 | #include <net/dst.h> |
fc4099f1 | 106 | #include <net/dst_metadata.h> |
04f00ab2 | 107 | #include <net/gro.h> |
1da177e4 | 108 | #include <net/pkt_sched.h> |
87d83093 | 109 | #include <net/pkt_cls.h> |
1da177e4 | 110 | #include <net/checksum.h> |
44540960 | 111 | #include <net/xfrm.h> |
e420bed0 | 112 | #include <net/tcx.h> |
1da177e4 LT |
113 | #include <linux/highmem.h> |
114 | #include <linux/init.h> | |
1da177e4 | 115 | #include <linux/module.h> |
1da177e4 LT |
116 | #include <linux/netpoll.h> |
117 | #include <linux/rcupdate.h> | |
118 | #include <linux/delay.h> | |
1da177e4 | 119 | #include <net/iw_handler.h> |
1da177e4 | 120 | #include <asm/current.h> |
5bdb9886 | 121 | #include <linux/audit.h> |
db217334 | 122 | #include <linux/dmaengine.h> |
f6a78bfc | 123 | #include <linux/err.h> |
c7fa9d18 | 124 | #include <linux/ctype.h> |
723e98b7 | 125 | #include <linux/if_arp.h> |
6de329e2 | 126 | #include <linux/if_vlan.h> |
8f0f2223 | 127 | #include <linux/ip.h> |
ad55dcaf | 128 | #include <net/ip.h> |
25cd9ba0 | 129 | #include <net/mpls.h> |
8f0f2223 DM |
130 | #include <linux/ipv6.h> |
131 | #include <linux/in.h> | |
b6b2fed1 DM |
132 | #include <linux/jhash.h> |
133 | #include <linux/random.h> | |
9cbc1cb8 | 134 | #include <trace/events/napi.h> |
cf66ba58 | 135 | #include <trace/events/net.h> |
07dc22e7 | 136 | #include <trace/events/skb.h> |
70713ddd | 137 | #include <trace/events/qdisc.h> |
bf4ea1d0 | 138 | #include <trace/events/xdp.h> |
caeda9b9 | 139 | #include <linux/inetdevice.h> |
c445477d | 140 | #include <linux/cpu_rmap.h> |
c5905afb | 141 | #include <linux/static_key.h> |
af12fa6e | 142 | #include <linux/hashtable.h> |
60877a32 | 143 | #include <linux/vmalloc.h> |
529d0489 | 144 | #include <linux/if_macvlan.h> |
e7fd2885 | 145 | #include <linux/errqueue.h> |
3b47d303 | 146 | #include <linux/hrtimer.h> |
7463acfb | 147 | #include <linux/netfilter_netdev.h> |
40e4e713 | 148 | #include <linux/crash_dump.h> |
b72b5bf6 | 149 | #include <linux/sctp.h> |
ae847f40 | 150 | #include <net/udp_tunnel.h> |
6621dd29 | 151 | #include <linux/net_namespace.h> |
aaa5d90b | 152 | #include <linux/indirect_call_wrapper.h> |
af3836df | 153 | #include <net/devlink.h> |
bd869245 | 154 | #include <linux/pm_runtime.h> |
3744741a | 155 | #include <linux/prandom.h> |
127d7355 | 156 | #include <linux/once_lite.h> |
49e47a5b | 157 | #include <net/netdev_rx_queue.h> |
2b0cfa6e LB |
158 | #include <net/page_pool/types.h> |
159 | #include <net/page_pool/helpers.h> | |
490a79fa | 160 | #include <net/rps.h> |
1da177e4 | 161 | |
6264f58c | 162 | #include "dev.h" |
342709ef PE |
163 | #include "net-sysfs.h" |
164 | ||
1da177e4 | 165 | static DEFINE_SPINLOCK(ptype_lock); |
900ff8c6 | 166 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
1da177e4 | 167 | |
ae78dbfa | 168 | static int netif_rx_internal(struct sk_buff *skb); |
26372605 PM |
169 | static int call_netdevice_notifiers_extack(unsigned long val, |
170 | struct net_device *dev, | |
171 | struct netlink_ext_ack *extack); | |
ae78dbfa | 172 | |
6c557001 FW |
173 | static DEFINE_MUTEX(ifalias_mutex); |
174 | ||
af12fa6e ET |
175 | /* protects napi_hash addition/deletion and napi_gen_id */ |
176 | static DEFINE_SPINLOCK(napi_hash_lock); | |
177 | ||
52bd2d62 | 178 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 179 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 180 | |
11d6011c | 181 | static DECLARE_RWSEM(devnet_rename_sem); |
c91f6df2 | 182 | |
4e985ada TG |
183 | static inline void dev_base_seq_inc(struct net *net) |
184 | { | |
590e92cd ED |
185 | unsigned int val = net->dev_base_seq + 1; |
186 | ||
187 | WRITE_ONCE(net->dev_base_seq, val ?: 1); | |
4e985ada TG |
188 | } |
189 | ||
881d966b | 190 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 191 | { |
8387ff25 | 192 | unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); |
95c96174 | 193 | |
08e9897d | 194 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
195 | } |
196 | ||
881d966b | 197 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 198 | { |
7c28bd0b | 199 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
200 | } |
201 | ||
dad6b977 SAS |
202 | #ifndef CONFIG_PREEMPT_RT |
203 | ||
204 | static DEFINE_STATIC_KEY_FALSE(use_backlog_threads_key); | |
205 | ||
206 | static int __init setup_backlog_napi_threads(char *arg) | |
207 | { | |
208 | static_branch_enable(&use_backlog_threads_key); | |
209 | return 0; | |
210 | } | |
211 | early_param("thread_backlog_napi", setup_backlog_napi_threads); | |
212 | ||
213 | static bool use_backlog_threads(void) | |
214 | { | |
215 | return static_branch_unlikely(&use_backlog_threads_key); | |
216 | } | |
217 | ||
218 | #else | |
219 | ||
220 | static bool use_backlog_threads(void) | |
221 | { | |
222 | return true; | |
223 | } | |
224 | ||
225 | #endif | |
226 | ||
765b11f8 SAS |
227 | static inline void backlog_lock_irq_save(struct softnet_data *sd, |
228 | unsigned long *flags) | |
152102c7 | 229 | { |
80d2eefc | 230 | if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) |
e722db8d SAS |
231 | spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); |
232 | else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) | |
233 | local_irq_save(*flags); | |
152102c7 CG |
234 | } |
235 | ||
765b11f8 | 236 | static inline void backlog_lock_irq_disable(struct softnet_data *sd) |
152102c7 | 237 | { |
80d2eefc | 238 | if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) |
e722db8d SAS |
239 | spin_lock_irq(&sd->input_pkt_queue.lock); |
240 | else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) | |
241 | local_irq_disable(); | |
242 | } | |
243 | ||
765b11f8 SAS |
244 | static inline void backlog_unlock_irq_restore(struct softnet_data *sd, |
245 | unsigned long *flags) | |
e722db8d | 246 | { |
80d2eefc | 247 | if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) |
e722db8d SAS |
248 | spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); |
249 | else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) | |
250 | local_irq_restore(*flags); | |
251 | } | |
252 | ||
765b11f8 | 253 | static inline void backlog_unlock_irq_enable(struct softnet_data *sd) |
e722db8d | 254 | { |
80d2eefc | 255 | if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) |
e722db8d SAS |
256 | spin_unlock_irq(&sd->input_pkt_queue.lock); |
257 | else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) | |
258 | local_irq_enable(); | |
152102c7 CG |
259 | } |
260 | ||
ff927412 JP |
261 | static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, |
262 | const char *name) | |
263 | { | |
264 | struct netdev_name_node *name_node; | |
265 | ||
266 | name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); | |
267 | if (!name_node) | |
268 | return NULL; | |
269 | INIT_HLIST_NODE(&name_node->hlist); | |
270 | name_node->dev = dev; | |
271 | name_node->name = name; | |
272 | return name_node; | |
273 | } | |
274 | ||
275 | static struct netdev_name_node * | |
276 | netdev_name_node_head_alloc(struct net_device *dev) | |
277 | { | |
36fbf1e5 JP |
278 | struct netdev_name_node *name_node; |
279 | ||
280 | name_node = netdev_name_node_alloc(dev, dev->name); | |
281 | if (!name_node) | |
282 | return NULL; | |
283 | INIT_LIST_HEAD(&name_node->list); | |
284 | return name_node; | |
ff927412 JP |
285 | } |
286 | ||
287 | static void netdev_name_node_free(struct netdev_name_node *name_node) | |
288 | { | |
289 | kfree(name_node); | |
290 | } | |
291 | ||
292 | static void netdev_name_node_add(struct net *net, | |
293 | struct netdev_name_node *name_node) | |
294 | { | |
295 | hlist_add_head_rcu(&name_node->hlist, | |
296 | dev_name_hash(net, name_node->name)); | |
297 | } | |
298 | ||
299 | static void netdev_name_node_del(struct netdev_name_node *name_node) | |
300 | { | |
301 | hlist_del_rcu(&name_node->hlist); | |
302 | } | |
303 | ||
304 | static struct netdev_name_node *netdev_name_node_lookup(struct net *net, | |
305 | const char *name) | |
306 | { | |
307 | struct hlist_head *head = dev_name_hash(net, name); | |
308 | struct netdev_name_node *name_node; | |
309 | ||
310 | hlist_for_each_entry(name_node, head, hlist) | |
311 | if (!strcmp(name_node->name, name)) | |
312 | return name_node; | |
313 | return NULL; | |
314 | } | |
315 | ||
316 | static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, | |
317 | const char *name) | |
318 | { | |
319 | struct hlist_head *head = dev_name_hash(net, name); | |
320 | struct netdev_name_node *name_node; | |
321 | ||
322 | hlist_for_each_entry_rcu(name_node, head, hlist) | |
323 | if (!strcmp(name_node->name, name)) | |
324 | return name_node; | |
325 | return NULL; | |
326 | } | |
327 | ||
75ea27d0 AT |
328 | bool netdev_name_in_use(struct net *net, const char *name) |
329 | { | |
330 | return netdev_name_node_lookup(net, name); | |
331 | } | |
332 | EXPORT_SYMBOL(netdev_name_in_use); | |
333 | ||
36fbf1e5 JP |
334 | int netdev_name_node_alt_create(struct net_device *dev, const char *name) |
335 | { | |
336 | struct netdev_name_node *name_node; | |
337 | struct net *net = dev_net(dev); | |
338 | ||
339 | name_node = netdev_name_node_lookup(net, name); | |
340 | if (name_node) | |
341 | return -EEXIST; | |
342 | name_node = netdev_name_node_alloc(dev, name); | |
343 | if (!name_node) | |
344 | return -ENOMEM; | |
345 | netdev_name_node_add(net, name_node); | |
346 | /* The node that holds dev->name acts as a head of per-device list. */ | |
9f308313 | 347 | list_add_tail_rcu(&name_node->list, &dev->name_node->list); |
36fbf1e5 JP |
348 | |
349 | return 0; | |
350 | } | |
36fbf1e5 | 351 | |
723de3eb | 352 | static void netdev_name_node_alt_free(struct rcu_head *head) |
36fbf1e5 | 353 | { |
723de3eb JK |
354 | struct netdev_name_node *name_node = |
355 | container_of(head, struct netdev_name_node, rcu); | |
356 | ||
36fbf1e5 JP |
357 | kfree(name_node->name); |
358 | netdev_name_node_free(name_node); | |
359 | } | |
360 | ||
723de3eb JK |
361 | static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) |
362 | { | |
363 | netdev_name_node_del(name_node); | |
364 | list_del(&name_node->list); | |
365 | call_rcu(&name_node->rcu, netdev_name_node_alt_free); | |
366 | } | |
367 | ||
36fbf1e5 JP |
368 | int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) |
369 | { | |
370 | struct netdev_name_node *name_node; | |
371 | struct net *net = dev_net(dev); | |
372 | ||
373 | name_node = netdev_name_node_lookup(net, name); | |
374 | if (!name_node) | |
375 | return -ENOENT; | |
e08ad805 ED |
376 | /* lookup might have found our primary name or a name belonging |
377 | * to another device. | |
378 | */ | |
379 | if (name_node == dev->name_node || name_node->dev != dev) | |
380 | return -EINVAL; | |
381 | ||
36fbf1e5 | 382 | __netdev_name_node_alt_destroy(name_node); |
36fbf1e5 JP |
383 | return 0; |
384 | } | |
36fbf1e5 JP |
385 | |
386 | static void netdev_name_node_alt_flush(struct net_device *dev) | |
387 | { | |
388 | struct netdev_name_node *name_node, *tmp; | |
389 | ||
723de3eb JK |
390 | list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) { |
391 | list_del(&name_node->list); | |
392 | netdev_name_node_alt_free(&name_node->rcu); | |
393 | } | |
36fbf1e5 JP |
394 | } |
395 | ||
ce286d32 | 396 | /* Device list insertion */ |
53759be9 | 397 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 398 | { |
8e15aee6 | 399 | struct netdev_name_node *name_node; |
c346dca1 | 400 | struct net *net = dev_net(dev); |
ce286d32 EB |
401 | |
402 | ASSERT_RTNL(); | |
403 | ||
c6d14c84 | 404 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
ff927412 | 405 | netdev_name_node_add(net, dev->name_node); |
fb699dfd ED |
406 | hlist_add_head_rcu(&dev->index_hlist, |
407 | dev_index_hash(net, dev->ifindex)); | |
8e15aee6 JK |
408 | |
409 | netdev_for_each_altname(dev, name_node) | |
410 | netdev_name_node_add(net, name_node); | |
411 | ||
759ab1ed JK |
412 | /* We reserved the ifindex, this can't fail */ |
413 | WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL)); | |
4e985ada TG |
414 | |
415 | dev_base_seq_inc(net); | |
ce286d32 EB |
416 | } |
417 | ||
fb699dfd ED |
418 | /* Device list removal |
419 | * caller must respect a RCU grace period before freeing/reusing dev | |
420 | */ | |
e51b9624 | 421 | static void unlist_netdevice(struct net_device *dev) |
ce286d32 | 422 | { |
8e15aee6 | 423 | struct netdev_name_node *name_node; |
759ab1ed JK |
424 | struct net *net = dev_net(dev); |
425 | ||
ce286d32 EB |
426 | ASSERT_RTNL(); |
427 | ||
759ab1ed JK |
428 | xa_erase(&net->dev_by_index, dev->ifindex); |
429 | ||
8e15aee6 JK |
430 | netdev_for_each_altname(dev, name_node) |
431 | netdev_name_node_del(name_node); | |
432 | ||
ce286d32 | 433 | /* Unlink dev from the device chain */ |
c6d14c84 | 434 | list_del_rcu(&dev->dev_list); |
ff927412 | 435 | netdev_name_node_del(dev->name_node); |
fb699dfd | 436 | hlist_del_rcu(&dev->index_hlist); |
4e985ada TG |
437 | |
438 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
439 | } |
440 | ||
1da177e4 LT |
441 | /* |
442 | * Our notifier list | |
443 | */ | |
444 | ||
f07d5b94 | 445 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
446 | |
447 | /* | |
448 | * Device drivers call our routines to queue packets here. We empty the | |
449 | * queue in the local softnet handler. | |
450 | */ | |
bea3348e | 451 | |
9958da05 | 452 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 453 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 454 | |
2b0cfa6e LB |
455 | /* Page_pool has a lockless array/stack to alloc/recycle pages. |
456 | * PP consumers must pay attention to run APIs in the appropriate context | |
457 | * (e.g. NAPI context). | |
458 | */ | |
5086f0fe | 459 | static DEFINE_PER_CPU(struct page_pool *, system_page_pool); |
2b0cfa6e | 460 | |
1a33e10e CW |
461 | #ifdef CONFIG_LOCKDEP |
462 | /* | |
463 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class | |
464 | * according to dev->type | |
465 | */ | |
466 | static const unsigned short netdev_lock_type[] = { | |
467 | ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
468 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
469 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
470 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
471 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
472 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
473 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
474 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
475 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
476 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
477 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
478 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
479 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, | |
480 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
481 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
482 | ||
483 | static const char *const netdev_lock_name[] = { | |
484 | "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
485 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
486 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
487 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
488 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
489 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
490 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
491 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
492 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
493 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
494 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
495 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
496 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", | |
497 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
498 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
499 | ||
500 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
845e0ebb | 501 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
1a33e10e CW |
502 | |
503 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
504 | { | |
505 | int i; | |
506 | ||
507 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
508 | if (netdev_lock_type[i] == dev_type) | |
509 | return i; | |
510 | /* the last key is used by default */ | |
511 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
512 | } | |
513 | ||
514 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, | |
515 | unsigned short dev_type) | |
516 | { | |
517 | int i; | |
518 | ||
519 | i = netdev_lock_pos(dev_type); | |
520 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
521 | netdev_lock_name[i]); | |
522 | } | |
845e0ebb CW |
523 | |
524 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
525 | { | |
526 | int i; | |
527 | ||
528 | i = netdev_lock_pos(dev->type); | |
529 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
530 | &netdev_addr_lock_key[i], | |
531 | netdev_lock_name[i]); | |
532 | } | |
1a33e10e CW |
533 | #else |
534 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, | |
535 | unsigned short dev_type) | |
536 | { | |
537 | } | |
845e0ebb CW |
538 | |
539 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
540 | { | |
541 | } | |
1a33e10e CW |
542 | #endif |
543 | ||
1da177e4 | 544 | /******************************************************************************* |
eb13da1a | 545 | * |
546 | * Protocol management and registration routines | |
547 | * | |
548 | *******************************************************************************/ | |
1da177e4 | 549 | |
1da177e4 | 550 | |
1da177e4 LT |
551 | /* |
552 | * Add a protocol ID to the list. Now that the input handler is | |
553 | * smarter we can dispense with all the messy stuff that used to be | |
554 | * here. | |
555 | * | |
556 | * BEWARE!!! Protocol handlers, mangling input packets, | |
557 | * MUST BE last in hash buckets and checking protocol handlers | |
558 | * MUST start from promiscuous ptype_all chain in net_bh. | |
559 | * It is true now, do not change it. | |
560 | * Explanation follows: if protocol handler, mangling packet, will | |
561 | * be the first on list, it is not able to sense, that packet | |
562 | * is cloned and should be copied-on-write, so that it will | |
563 | * change it and subsequent readers will get broken packet. | |
564 | * --ANK (980803) | |
565 | */ | |
566 | ||
c07b68e8 ED |
567 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
568 | { | |
569 | if (pt->type == htons(ETH_P_ALL)) | |
0b91fa4b | 570 | return pt->dev ? &pt->dev->ptype_all : &net_hotdata.ptype_all; |
c07b68e8 | 571 | else |
7866a621 SN |
572 | return pt->dev ? &pt->dev->ptype_specific : |
573 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
574 | } |
575 | ||
1da177e4 LT |
576 | /** |
577 | * dev_add_pack - add packet handler | |
578 | * @pt: packet type declaration | |
579 | * | |
580 | * Add a protocol handler to the networking stack. The passed &packet_type | |
581 | * is linked into kernel lists and may not be freed until it has been | |
582 | * removed from the kernel lists. | |
583 | * | |
4ec93edb | 584 | * This call does not sleep therefore it can not |
1da177e4 LT |
585 | * guarantee all CPU's that are in middle of receiving packets |
586 | * will see the new packet type (until the next received packet). | |
587 | */ | |
588 | ||
589 | void dev_add_pack(struct packet_type *pt) | |
590 | { | |
c07b68e8 | 591 | struct list_head *head = ptype_head(pt); |
1da177e4 | 592 | |
c07b68e8 ED |
593 | spin_lock(&ptype_lock); |
594 | list_add_rcu(&pt->list, head); | |
595 | spin_unlock(&ptype_lock); | |
1da177e4 | 596 | } |
d1b19dff | 597 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 598 | |
1da177e4 LT |
599 | /** |
600 | * __dev_remove_pack - remove packet handler | |
601 | * @pt: packet type declaration | |
602 | * | |
603 | * Remove a protocol handler that was previously added to the kernel | |
604 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
605 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 606 | * returns. |
1da177e4 LT |
607 | * |
608 | * The packet type might still be in use by receivers | |
609 | * and must not be freed until after all the CPU's have gone | |
610 | * through a quiescent state. | |
611 | */ | |
612 | void __dev_remove_pack(struct packet_type *pt) | |
613 | { | |
c07b68e8 | 614 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
615 | struct packet_type *pt1; |
616 | ||
c07b68e8 | 617 | spin_lock(&ptype_lock); |
1da177e4 LT |
618 | |
619 | list_for_each_entry(pt1, head, list) { | |
620 | if (pt == pt1) { | |
621 | list_del_rcu(&pt->list); | |
622 | goto out; | |
623 | } | |
624 | } | |
625 | ||
7b6cd1ce | 626 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 627 | out: |
c07b68e8 | 628 | spin_unlock(&ptype_lock); |
1da177e4 | 629 | } |
d1b19dff ED |
630 | EXPORT_SYMBOL(__dev_remove_pack); |
631 | ||
1da177e4 LT |
632 | /** |
633 | * dev_remove_pack - remove packet handler | |
634 | * @pt: packet type declaration | |
635 | * | |
636 | * Remove a protocol handler that was previously added to the kernel | |
637 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
638 | * from the kernel lists and can be freed or reused once this function | |
639 | * returns. | |
640 | * | |
641 | * This call sleeps to guarantee that no CPU is looking at the packet | |
642 | * type after return. | |
643 | */ | |
644 | void dev_remove_pack(struct packet_type *pt) | |
645 | { | |
646 | __dev_remove_pack(pt); | |
4ec93edb | 647 | |
1da177e4 LT |
648 | synchronize_net(); |
649 | } | |
d1b19dff | 650 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 651 | |
62532da9 | 652 | |
1da177e4 | 653 | /******************************************************************************* |
eb13da1a | 654 | * |
655 | * Device Interface Subroutines | |
656 | * | |
657 | *******************************************************************************/ | |
1da177e4 | 658 | |
a54acb3a ND |
659 | /** |
660 | * dev_get_iflink - get 'iflink' value of a interface | |
661 | * @dev: targeted interface | |
662 | * | |
663 | * Indicates the ifindex the interface is linked to. | |
664 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
665 | */ | |
666 | ||
667 | int dev_get_iflink(const struct net_device *dev) | |
668 | { | |
669 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
670 | return dev->netdev_ops->ndo_get_iflink(dev); | |
671 | ||
e353ea9c | 672 | return READ_ONCE(dev->ifindex); |
a54acb3a ND |
673 | } |
674 | EXPORT_SYMBOL(dev_get_iflink); | |
675 | ||
fc4099f1 PS |
676 | /** |
677 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
678 | * @dev: targeted interface | |
679 | * @skb: The packet. | |
680 | * | |
681 | * For better visibility of tunnel traffic OVS needs to retrieve | |
682 | * egress tunnel information for a packet. Following API allows | |
683 | * user to get this info. | |
684 | */ | |
685 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
686 | { | |
687 | struct ip_tunnel_info *info; | |
688 | ||
689 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
690 | return -EINVAL; | |
691 | ||
692 | info = skb_tunnel_info_unclone(skb); | |
693 | if (!info) | |
694 | return -ENOMEM; | |
695 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
696 | return -EINVAL; | |
697 | ||
698 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
699 | } | |
700 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
701 | ||
ddb94eaf PNA |
702 | static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) |
703 | { | |
704 | int k = stack->num_paths++; | |
705 | ||
706 | if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) | |
707 | return NULL; | |
708 | ||
709 | return &stack->path[k]; | |
710 | } | |
711 | ||
712 | int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, | |
713 | struct net_device_path_stack *stack) | |
714 | { | |
715 | const struct net_device *last_dev; | |
716 | struct net_device_path_ctx ctx = { | |
717 | .dev = dev, | |
ddb94eaf PNA |
718 | }; |
719 | struct net_device_path *path; | |
720 | int ret = 0; | |
721 | ||
cf2df74e | 722 | memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); |
ddb94eaf PNA |
723 | stack->num_paths = 0; |
724 | while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { | |
725 | last_dev = ctx.dev; | |
726 | path = dev_fwd_path(stack); | |
727 | if (!path) | |
728 | return -1; | |
729 | ||
730 | memset(path, 0, sizeof(struct net_device_path)); | |
731 | ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); | |
732 | if (ret < 0) | |
733 | return -1; | |
734 | ||
735 | if (WARN_ON_ONCE(last_dev == ctx.dev)) | |
736 | return -1; | |
737 | } | |
a333215e FF |
738 | |
739 | if (!ctx.dev) | |
740 | return ret; | |
741 | ||
ddb94eaf PNA |
742 | path = dev_fwd_path(stack); |
743 | if (!path) | |
744 | return -1; | |
745 | path->type = DEV_PATH_ETHERNET; | |
746 | path->dev = ctx.dev; | |
747 | ||
748 | return ret; | |
749 | } | |
750 | EXPORT_SYMBOL_GPL(dev_fill_forward_path); | |
751 | ||
1da177e4 LT |
752 | /** |
753 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 754 | * @net: the applicable net namespace |
1da177e4 LT |
755 | * @name: name to find |
756 | * | |
1b3ef46c ED |
757 | * Find an interface by name. Must be called under RTNL semaphore. |
758 | * If the name is found a pointer to the device is returned. | |
759 | * If the name is not found then %NULL is returned. The | |
1da177e4 LT |
760 | * reference counters are not incremented so the caller must be |
761 | * careful with locks. | |
762 | */ | |
763 | ||
881d966b | 764 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 765 | { |
ff927412 | 766 | struct netdev_name_node *node_name; |
1da177e4 | 767 | |
ff927412 JP |
768 | node_name = netdev_name_node_lookup(net, name); |
769 | return node_name ? node_name->dev : NULL; | |
1da177e4 | 770 | } |
d1b19dff | 771 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 772 | |
72c9528b | 773 | /** |
722c9a0c | 774 | * dev_get_by_name_rcu - find a device by its name |
775 | * @net: the applicable net namespace | |
776 | * @name: name to find | |
777 | * | |
778 | * Find an interface by name. | |
779 | * If the name is found a pointer to the device is returned. | |
780 | * If the name is not found then %NULL is returned. | |
781 | * The reference counters are not incremented so the caller must be | |
782 | * careful with locks. The caller must hold RCU lock. | |
72c9528b ED |
783 | */ |
784 | ||
785 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
786 | { | |
ff927412 | 787 | struct netdev_name_node *node_name; |
72c9528b | 788 | |
ff927412 JP |
789 | node_name = netdev_name_node_lookup_rcu(net, name); |
790 | return node_name ? node_name->dev : NULL; | |
72c9528b ED |
791 | } |
792 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
793 | ||
70f7457a JK |
794 | /* Deprecated for new users, call netdev_get_by_name() instead */ |
795 | struct net_device *dev_get_by_name(struct net *net, const char *name) | |
796 | { | |
797 | struct net_device *dev; | |
798 | ||
799 | rcu_read_lock(); | |
800 | dev = dev_get_by_name_rcu(net, name); | |
801 | dev_hold(dev); | |
802 | rcu_read_unlock(); | |
803 | return dev; | |
804 | } | |
805 | EXPORT_SYMBOL(dev_get_by_name); | |
806 | ||
1da177e4 | 807 | /** |
70f7457a | 808 | * netdev_get_by_name() - find a device by its name |
c4ea43c5 | 809 | * @net: the applicable net namespace |
1da177e4 | 810 | * @name: name to find |
70f7457a JK |
811 | * @tracker: tracking object for the acquired reference |
812 | * @gfp: allocation flags for the tracker | |
1da177e4 LT |
813 | * |
814 | * Find an interface by name. This can be called from any | |
815 | * context and does its own locking. The returned handle has | |
70f7457a | 816 | * the usage count incremented and the caller must use netdev_put() to |
1da177e4 LT |
817 | * release it when it is no longer needed. %NULL is returned if no |
818 | * matching device is found. | |
819 | */ | |
70f7457a JK |
820 | struct net_device *netdev_get_by_name(struct net *net, const char *name, |
821 | netdevice_tracker *tracker, gfp_t gfp) | |
1da177e4 LT |
822 | { |
823 | struct net_device *dev; | |
824 | ||
70f7457a JK |
825 | dev = dev_get_by_name(net, name); |
826 | if (dev) | |
827 | netdev_tracker_alloc(dev, tracker, gfp); | |
1da177e4 LT |
828 | return dev; |
829 | } | |
70f7457a | 830 | EXPORT_SYMBOL(netdev_get_by_name); |
1da177e4 LT |
831 | |
832 | /** | |
833 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 834 | * @net: the applicable net namespace |
1da177e4 LT |
835 | * @ifindex: index of device |
836 | * | |
837 | * Search for an interface by index. Returns %NULL if the device | |
838 | * is not found or a pointer to the device. The device has not | |
839 | * had its reference counter increased so the caller must be careful | |
1b3ef46c | 840 | * about locking. The caller must hold the RTNL semaphore. |
1da177e4 LT |
841 | */ |
842 | ||
881d966b | 843 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 844 | { |
0bd8d536 ED |
845 | struct net_device *dev; |
846 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 847 | |
b67bfe0d | 848 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
849 | if (dev->ifindex == ifindex) |
850 | return dev; | |
0bd8d536 | 851 | |
1da177e4 LT |
852 | return NULL; |
853 | } | |
d1b19dff | 854 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 855 | |
fb699dfd ED |
856 | /** |
857 | * dev_get_by_index_rcu - find a device by its ifindex | |
858 | * @net: the applicable net namespace | |
859 | * @ifindex: index of device | |
860 | * | |
861 | * Search for an interface by index. Returns %NULL if the device | |
862 | * is not found or a pointer to the device. The device has not | |
863 | * had its reference counter increased so the caller must be careful | |
864 | * about locking. The caller must hold RCU lock. | |
865 | */ | |
866 | ||
867 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
868 | { | |
fb699dfd ED |
869 | struct net_device *dev; |
870 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
871 | ||
b67bfe0d | 872 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
873 | if (dev->ifindex == ifindex) |
874 | return dev; | |
875 | ||
876 | return NULL; | |
877 | } | |
878 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
879 | ||
70f7457a JK |
880 | /* Deprecated for new users, call netdev_get_by_index() instead */ |
881 | struct net_device *dev_get_by_index(struct net *net, int ifindex) | |
882 | { | |
883 | struct net_device *dev; | |
884 | ||
885 | rcu_read_lock(); | |
886 | dev = dev_get_by_index_rcu(net, ifindex); | |
887 | dev_hold(dev); | |
888 | rcu_read_unlock(); | |
889 | return dev; | |
890 | } | |
891 | EXPORT_SYMBOL(dev_get_by_index); | |
1da177e4 LT |
892 | |
893 | /** | |
70f7457a | 894 | * netdev_get_by_index() - find a device by its ifindex |
c4ea43c5 | 895 | * @net: the applicable net namespace |
1da177e4 | 896 | * @ifindex: index of device |
70f7457a JK |
897 | * @tracker: tracking object for the acquired reference |
898 | * @gfp: allocation flags for the tracker | |
1da177e4 LT |
899 | * |
900 | * Search for an interface by index. Returns NULL if the device | |
901 | * is not found or a pointer to the device. The device returned has | |
902 | * had a reference added and the pointer is safe until the user calls | |
70f7457a | 903 | * netdev_put() to indicate they have finished with it. |
1da177e4 | 904 | */ |
70f7457a JK |
905 | struct net_device *netdev_get_by_index(struct net *net, int ifindex, |
906 | netdevice_tracker *tracker, gfp_t gfp) | |
1da177e4 LT |
907 | { |
908 | struct net_device *dev; | |
909 | ||
70f7457a JK |
910 | dev = dev_get_by_index(net, ifindex); |
911 | if (dev) | |
912 | netdev_tracker_alloc(dev, tracker, gfp); | |
1da177e4 LT |
913 | return dev; |
914 | } | |
70f7457a | 915 | EXPORT_SYMBOL(netdev_get_by_index); |
1da177e4 | 916 | |
90b602f8 ML |
917 | /** |
918 | * dev_get_by_napi_id - find a device by napi_id | |
919 | * @napi_id: ID of the NAPI struct | |
920 | * | |
921 | * Search for an interface by NAPI ID. Returns %NULL if the device | |
922 | * is not found or a pointer to the device. The device has not had | |
923 | * its reference counter increased so the caller must be careful | |
924 | * about locking. The caller must hold RCU lock. | |
925 | */ | |
926 | ||
927 | struct net_device *dev_get_by_napi_id(unsigned int napi_id) | |
928 | { | |
929 | struct napi_struct *napi; | |
930 | ||
931 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
932 | ||
933 | if (napi_id < MIN_NAPI_ID) | |
934 | return NULL; | |
935 | ||
936 | napi = napi_by_id(napi_id); | |
937 | ||
938 | return napi ? napi->dev : NULL; | |
939 | } | |
940 | EXPORT_SYMBOL(dev_get_by_napi_id); | |
941 | ||
0840556e KI |
942 | static DEFINE_SEQLOCK(netdev_rename_lock); |
943 | ||
944 | void netdev_copy_name(struct net_device *dev, char *name) | |
945 | { | |
946 | unsigned int seq; | |
947 | ||
948 | do { | |
949 | seq = read_seqbegin(&netdev_rename_lock); | |
950 | strscpy(name, dev->name, IFNAMSIZ); | |
951 | } while (read_seqretry(&netdev_rename_lock, seq)); | |
952 | } | |
953 | ||
5dbe7c17 NS |
954 | /** |
955 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
956 | * @net: network namespace | |
957 | * @name: a pointer to the buffer where the name will be stored. | |
958 | * @ifindex: the ifindex of the interface to get the name from. | |
5dbe7c17 NS |
959 | */ |
960 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
961 | { | |
962 | struct net_device *dev; | |
11d6011c | 963 | int ret; |
5dbe7c17 | 964 | |
5dbe7c17 | 965 | rcu_read_lock(); |
11d6011c | 966 | |
5dbe7c17 NS |
967 | dev = dev_get_by_index_rcu(net, ifindex); |
968 | if (!dev) { | |
11d6011c AD |
969 | ret = -ENODEV; |
970 | goto out; | |
5dbe7c17 NS |
971 | } |
972 | ||
0840556e | 973 | netdev_copy_name(dev, name); |
5dbe7c17 | 974 | |
11d6011c AD |
975 | ret = 0; |
976 | out: | |
977 | rcu_read_unlock(); | |
11d6011c | 978 | return ret; |
5dbe7c17 NS |
979 | } |
980 | ||
1da177e4 | 981 | /** |
941666c2 | 982 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 983 | * @net: the applicable net namespace |
1da177e4 LT |
984 | * @type: media type of device |
985 | * @ha: hardware address | |
986 | * | |
987 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
988 | * is not found or a pointer to the device. |
989 | * The caller must hold RCU or RTNL. | |
941666c2 | 990 | * The returned device has not had its ref count increased |
1da177e4 LT |
991 | * and the caller must therefore be careful about locking |
992 | * | |
1da177e4 LT |
993 | */ |
994 | ||
941666c2 ED |
995 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
996 | const char *ha) | |
1da177e4 LT |
997 | { |
998 | struct net_device *dev; | |
999 | ||
941666c2 | 1000 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
1001 | if (dev->type == type && |
1002 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
1003 | return dev; |
1004 | ||
1005 | return NULL; | |
1da177e4 | 1006 | } |
941666c2 | 1007 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 1008 | |
881d966b | 1009 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 1010 | { |
99fe3c39 | 1011 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 1012 | |
99fe3c39 ED |
1013 | rcu_read_lock(); |
1014 | for_each_netdev_rcu(net, dev) | |
1015 | if (dev->type == type) { | |
1016 | dev_hold(dev); | |
1017 | ret = dev; | |
1018 | break; | |
1019 | } | |
1020 | rcu_read_unlock(); | |
1021 | return ret; | |
1da177e4 | 1022 | } |
1da177e4 LT |
1023 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
1024 | ||
1025 | /** | |
6c555490 | 1026 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 1027 | * @net: the applicable net namespace |
1da177e4 LT |
1028 | * @if_flags: IFF_* values |
1029 | * @mask: bitmask of bits in if_flags to check | |
1030 | * | |
1031 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 1032 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 1033 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
1034 | */ |
1035 | ||
6c555490 WC |
1036 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
1037 | unsigned short mask) | |
1da177e4 | 1038 | { |
7562f876 | 1039 | struct net_device *dev, *ret; |
1da177e4 | 1040 | |
6c555490 WC |
1041 | ASSERT_RTNL(); |
1042 | ||
7562f876 | 1043 | ret = NULL; |
6c555490 | 1044 | for_each_netdev(net, dev) { |
1da177e4 | 1045 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 1046 | ret = dev; |
1da177e4 LT |
1047 | break; |
1048 | } | |
1049 | } | |
7562f876 | 1050 | return ret; |
1da177e4 | 1051 | } |
6c555490 | 1052 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
1053 | |
1054 | /** | |
1055 | * dev_valid_name - check if name is okay for network device | |
1056 | * @name: name string | |
1057 | * | |
1058 | * Network device names need to be valid file names to | |
4250b75b | 1059 | * allow sysfs to work. We also disallow any kind of |
c7fa9d18 | 1060 | * whitespace. |
1da177e4 | 1061 | */ |
95f050bf | 1062 | bool dev_valid_name(const char *name) |
1da177e4 | 1063 | { |
c7fa9d18 | 1064 | if (*name == '\0') |
95f050bf | 1065 | return false; |
a9d48205 | 1066 | if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) |
95f050bf | 1067 | return false; |
c7fa9d18 | 1068 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1069 | return false; |
c7fa9d18 DM |
1070 | |
1071 | while (*name) { | |
a4176a93 | 1072 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1073 | return false; |
c7fa9d18 DM |
1074 | name++; |
1075 | } | |
95f050bf | 1076 | return true; |
1da177e4 | 1077 | } |
d1b19dff | 1078 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1079 | |
1080 | /** | |
b267b179 EB |
1081 | * __dev_alloc_name - allocate a name for a device |
1082 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1083 | * @name: name format string |
bd07063d | 1084 | * @res: result name string |
1da177e4 LT |
1085 | * |
1086 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1087 | * id. It scans list of devices to build up a free map, then chooses |
1088 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1089 | * while allocating the name and adding the device in order to avoid | |
1090 | * duplicates. | |
1091 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1092 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1093 | */ |
1094 | ||
bd07063d | 1095 | static int __dev_alloc_name(struct net *net, const char *name, char *res) |
1da177e4 LT |
1096 | { |
1097 | int i = 0; | |
1da177e4 LT |
1098 | const char *p; |
1099 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1100 | unsigned long *inuse; |
1da177e4 | 1101 | struct net_device *d; |
bd07063d | 1102 | char buf[IFNAMSIZ]; |
1da177e4 | 1103 | |
9a810468 JK |
1104 | /* Verify the string as this thing may have come from the user. |
1105 | * There must be one "%d" and no other "%" characters. | |
1106 | */ | |
51f299dd | 1107 | p = strchr(name, '%'); |
9a810468 JK |
1108 | if (!p || p[1] != 'd' || strchr(p + 2, '%')) |
1109 | return -EINVAL; | |
1110 | ||
1111 | /* Use one page as a bit array of possible slots */ | |
1112 | inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC); | |
1113 | if (!inuse) | |
1114 | return -ENOMEM; | |
1115 | ||
1116 | for_each_netdev(net, d) { | |
1117 | struct netdev_name_node *name_node; | |
1118 | ||
1119 | netdev_for_each_altname(d, name_node) { | |
1120 | if (!sscanf(name_node->name, name, &i)) | |
1da177e4 LT |
1121 | continue; |
1122 | if (i < 0 || i >= max_netdevices) | |
1123 | continue; | |
1124 | ||
9a810468 | 1125 | /* avoid cases where sscanf is not exact inverse of printf */ |
b267b179 | 1126 | snprintf(buf, IFNAMSIZ, name, i); |
9a810468 | 1127 | if (!strncmp(buf, name_node->name, IFNAMSIZ)) |
25ee1660 | 1128 | __set_bit(i, inuse); |
1da177e4 | 1129 | } |
9a810468 JK |
1130 | if (!sscanf(d->name, name, &i)) |
1131 | continue; | |
1132 | if (i < 0 || i >= max_netdevices) | |
1133 | continue; | |
1134 | ||
1135 | /* avoid cases where sscanf is not exact inverse of printf */ | |
1136 | snprintf(buf, IFNAMSIZ, name, i); | |
1137 | if (!strncmp(buf, d->name, IFNAMSIZ)) | |
1138 | __set_bit(i, inuse); | |
1da177e4 LT |
1139 | } |
1140 | ||
9a810468 JK |
1141 | i = find_first_zero_bit(inuse, max_netdevices); |
1142 | bitmap_free(inuse); | |
7ad17b04 JK |
1143 | if (i == max_netdevices) |
1144 | return -ENFILE; | |
9a810468 | 1145 | |
674e3180 GP |
1146 | /* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */ |
1147 | strscpy(buf, name, IFNAMSIZ); | |
1148 | snprintf(res, IFNAMSIZ, buf, i); | |
7ad17b04 | 1149 | return i; |
1da177e4 LT |
1150 | } |
1151 | ||
556c755a | 1152 | /* Returns negative errno or allocated unit id (see __dev_alloc_name()) */ |
311cca40 | 1153 | static int dev_prep_valid_name(struct net *net, struct net_device *dev, |
556c755a JK |
1154 | const char *want_name, char *out_name, |
1155 | int dup_errno) | |
311cca40 | 1156 | { |
311cca40 JK |
1157 | if (!dev_valid_name(want_name)) |
1158 | return -EINVAL; | |
1159 | ||
ce4cfa23 | 1160 | if (strchr(want_name, '%')) |
556c755a | 1161 | return __dev_alloc_name(net, want_name, out_name); |
ce4cfa23 JK |
1162 | |
1163 | if (netdev_name_in_use(net, want_name)) | |
556c755a | 1164 | return -dup_errno; |
ce4cfa23 | 1165 | if (out_name != want_name) |
311cca40 | 1166 | strscpy(out_name, want_name, IFNAMSIZ); |
311cca40 JK |
1167 | return 0; |
1168 | } | |
1169 | ||
b267b179 EB |
1170 | /** |
1171 | * dev_alloc_name - allocate a name for a device | |
1172 | * @dev: device | |
1173 | * @name: name format string | |
1174 | * | |
1175 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1176 | * id. It scans list of devices to build up a free map, then chooses | |
1177 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1178 | * while allocating the name and adding the device in order to avoid | |
1179 | * duplicates. | |
1180 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1181 | * Returns the number of the unit assigned or a negative errno code. | |
1182 | */ | |
1183 | ||
1184 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1185 | { | |
556c755a | 1186 | return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE); |
b267b179 | 1187 | } |
d1b19dff | 1188 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1189 | |
bacb7e18 ED |
1190 | static int dev_get_valid_name(struct net *net, struct net_device *dev, |
1191 | const char *name) | |
828de4f6 | 1192 | { |
556c755a JK |
1193 | int ret; |
1194 | ||
1195 | ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST); | |
1196 | return ret < 0 ? ret : 0; | |
d9031024 | 1197 | } |
1da177e4 LT |
1198 | |
1199 | /** | |
1200 | * dev_change_name - change name of a device | |
1201 | * @dev: device | |
1202 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1203 | * | |
1204 | * Change name of a device, can pass format strings "eth%d". | |
1205 | * for wildcarding. | |
1206 | */ | |
cf04a4c7 | 1207 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1208 | { |
238fa362 | 1209 | unsigned char old_assign_type; |
fcc5a03a | 1210 | char oldname[IFNAMSIZ]; |
1da177e4 | 1211 | int err = 0; |
fcc5a03a | 1212 | int ret; |
881d966b | 1213 | struct net *net; |
1da177e4 LT |
1214 | |
1215 | ASSERT_RTNL(); | |
c346dca1 | 1216 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1217 | |
c346dca1 | 1218 | net = dev_net(dev); |
8065a779 | 1219 | |
11d6011c | 1220 | down_write(&devnet_rename_sem); |
c91f6df2 BH |
1221 | |
1222 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
11d6011c | 1223 | up_write(&devnet_rename_sem); |
c8d90dca | 1224 | return 0; |
c91f6df2 | 1225 | } |
c8d90dca | 1226 | |
fcc5a03a HX |
1227 | memcpy(oldname, dev->name, IFNAMSIZ); |
1228 | ||
0840556e | 1229 | write_seqlock(&netdev_rename_lock); |
828de4f6 | 1230 | err = dev_get_valid_name(net, dev, newname); |
0840556e KI |
1231 | write_sequnlock(&netdev_rename_lock); |
1232 | ||
c91f6df2 | 1233 | if (err < 0) { |
11d6011c | 1234 | up_write(&devnet_rename_sem); |
d9031024 | 1235 | return err; |
c91f6df2 | 1236 | } |
1da177e4 | 1237 | |
6fe82a39 | 1238 | if (oldname[0] && !strchr(oldname, '%')) |
bd039b5e AR |
1239 | netdev_info(dev, "renamed from %s%s\n", oldname, |
1240 | dev->flags & IFF_UP ? " (while UP)" : ""); | |
6fe82a39 | 1241 | |
238fa362 | 1242 | old_assign_type = dev->name_assign_type; |
1c07dbb0 | 1243 | WRITE_ONCE(dev->name_assign_type, NET_NAME_RENAMED); |
238fa362 | 1244 | |
fcc5a03a | 1245 | rollback: |
a1b3f594 EB |
1246 | ret = device_rename(&dev->dev, dev->name); |
1247 | if (ret) { | |
1248 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1c07dbb0 | 1249 | WRITE_ONCE(dev->name_assign_type, old_assign_type); |
11d6011c | 1250 | up_write(&devnet_rename_sem); |
a1b3f594 | 1251 | return ret; |
dcc99773 | 1252 | } |
7f988eab | 1253 | |
11d6011c | 1254 | up_write(&devnet_rename_sem); |
c91f6df2 | 1255 | |
5bb025fa VF |
1256 | netdev_adjacent_rename_links(dev, oldname); |
1257 | ||
ff927412 | 1258 | netdev_name_node_del(dev->name_node); |
72c9528b | 1259 | |
4cd582ff | 1260 | synchronize_net(); |
72c9528b | 1261 | |
ff927412 | 1262 | netdev_name_node_add(net, dev->name_node); |
7f988eab | 1263 | |
056925ab | 1264 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1265 | ret = notifier_to_errno(ret); |
1266 | ||
1267 | if (ret) { | |
91e9c07b ED |
1268 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1269 | if (err >= 0) { | |
fcc5a03a | 1270 | err = ret; |
11d6011c | 1271 | down_write(&devnet_rename_sem); |
0840556e | 1272 | write_seqlock(&netdev_rename_lock); |
fcc5a03a | 1273 | memcpy(dev->name, oldname, IFNAMSIZ); |
0840556e | 1274 | write_sequnlock(&netdev_rename_lock); |
5bb025fa | 1275 | memcpy(oldname, newname, IFNAMSIZ); |
1c07dbb0 | 1276 | WRITE_ONCE(dev->name_assign_type, old_assign_type); |
238fa362 | 1277 | old_assign_type = NET_NAME_RENAMED; |
fcc5a03a | 1278 | goto rollback; |
91e9c07b | 1279 | } else { |
5b92be64 JB |
1280 | netdev_err(dev, "name change rollback failed: %d\n", |
1281 | ret); | |
fcc5a03a HX |
1282 | } |
1283 | } | |
1da177e4 LT |
1284 | |
1285 | return err; | |
1286 | } | |
1287 | ||
0b815a1a SH |
1288 | /** |
1289 | * dev_set_alias - change ifalias of a device | |
1290 | * @dev: device | |
1291 | * @alias: name up to IFALIASZ | |
f0db275a | 1292 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1293 | * |
1294 | * Set ifalias for a device, | |
1295 | */ | |
1296 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1297 | { | |
6c557001 | 1298 | struct dev_ifalias *new_alias = NULL; |
0b815a1a SH |
1299 | |
1300 | if (len >= IFALIASZ) | |
1301 | return -EINVAL; | |
1302 | ||
6c557001 FW |
1303 | if (len) { |
1304 | new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); | |
1305 | if (!new_alias) | |
1306 | return -ENOMEM; | |
1307 | ||
1308 | memcpy(new_alias->ifalias, alias, len); | |
1309 | new_alias->ifalias[len] = 0; | |
96ca4a2c OH |
1310 | } |
1311 | ||
6c557001 | 1312 | mutex_lock(&ifalias_mutex); |
e3f0d761 PM |
1313 | new_alias = rcu_replace_pointer(dev->ifalias, new_alias, |
1314 | mutex_is_locked(&ifalias_mutex)); | |
6c557001 FW |
1315 | mutex_unlock(&ifalias_mutex); |
1316 | ||
1317 | if (new_alias) | |
1318 | kfree_rcu(new_alias, rcuhead); | |
0b815a1a | 1319 | |
0b815a1a SH |
1320 | return len; |
1321 | } | |
0fe554a4 | 1322 | EXPORT_SYMBOL(dev_set_alias); |
0b815a1a | 1323 | |
6c557001 FW |
1324 | /** |
1325 | * dev_get_alias - get ifalias of a device | |
1326 | * @dev: device | |
20e88320 | 1327 | * @name: buffer to store name of ifalias |
6c557001 FW |
1328 | * @len: size of buffer |
1329 | * | |
1330 | * get ifalias for a device. Caller must make sure dev cannot go | |
1331 | * away, e.g. rcu read lock or own a reference count to device. | |
1332 | */ | |
1333 | int dev_get_alias(const struct net_device *dev, char *name, size_t len) | |
1334 | { | |
1335 | const struct dev_ifalias *alias; | |
1336 | int ret = 0; | |
1337 | ||
1338 | rcu_read_lock(); | |
1339 | alias = rcu_dereference(dev->ifalias); | |
1340 | if (alias) | |
1341 | ret = snprintf(name, len, "%s", alias->ifalias); | |
1342 | rcu_read_unlock(); | |
1343 | ||
1344 | return ret; | |
1345 | } | |
0b815a1a | 1346 | |
d8a33ac4 | 1347 | /** |
3041a069 | 1348 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1349 | * @dev: device to cause notification |
1350 | * | |
1351 | * Called to indicate a device has changed features. | |
1352 | */ | |
1353 | void netdev_features_change(struct net_device *dev) | |
1354 | { | |
056925ab | 1355 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1356 | } |
1357 | EXPORT_SYMBOL(netdev_features_change); | |
1358 | ||
1da177e4 LT |
1359 | /** |
1360 | * netdev_state_change - device changes state | |
1361 | * @dev: device to cause notification | |
1362 | * | |
1363 | * Called to indicate a device has changed state. This function calls | |
1364 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1365 | * to the routing socket. | |
1366 | */ | |
1367 | void netdev_state_change(struct net_device *dev) | |
1368 | { | |
1369 | if (dev->flags & IFF_UP) { | |
51d0c047 DA |
1370 | struct netdev_notifier_change_info change_info = { |
1371 | .info.dev = dev, | |
1372 | }; | |
54951194 | 1373 | |
51d0c047 | 1374 | call_netdevice_notifiers_info(NETDEV_CHANGE, |
54951194 | 1375 | &change_info.info); |
1d997f10 | 1376 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); |
1da177e4 LT |
1377 | } |
1378 | } | |
d1b19dff | 1379 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1380 | |
7061eb8c LP |
1381 | /** |
1382 | * __netdev_notify_peers - notify network peers about existence of @dev, | |
1383 | * to be called when rtnl lock is already held. | |
1384 | * @dev: network device | |
1385 | * | |
1386 | * Generate traffic such that interested network peers are aware of | |
1387 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1388 | * a device wants to inform the rest of the network about some sort of | |
1389 | * reconfiguration such as a failover event or virtual machine | |
1390 | * migration. | |
1391 | */ | |
1392 | void __netdev_notify_peers(struct net_device *dev) | |
1393 | { | |
1394 | ASSERT_RTNL(); | |
1395 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1396 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); | |
1397 | } | |
1398 | EXPORT_SYMBOL(__netdev_notify_peers); | |
1399 | ||
ee89bab1 | 1400 | /** |
722c9a0c | 1401 | * netdev_notify_peers - notify network peers about existence of @dev |
1402 | * @dev: network device | |
ee89bab1 AW |
1403 | * |
1404 | * Generate traffic such that interested network peers are aware of | |
1405 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1406 | * a device wants to inform the rest of the network about some sort of | |
1407 | * reconfiguration such as a failover event or virtual machine | |
1408 | * migration. | |
1409 | */ | |
1410 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1411 | { |
ee89bab1 | 1412 | rtnl_lock(); |
7061eb8c | 1413 | __netdev_notify_peers(dev); |
ee89bab1 | 1414 | rtnl_unlock(); |
c1da4ac7 | 1415 | } |
ee89bab1 | 1416 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1417 | |
29863d41 WW |
1418 | static int napi_threaded_poll(void *data); |
1419 | ||
1420 | static int napi_kthread_create(struct napi_struct *n) | |
1421 | { | |
1422 | int err = 0; | |
1423 | ||
1424 | /* Create and wake up the kthread once to put it in | |
1425 | * TASK_INTERRUPTIBLE mode to avoid the blocked task | |
1426 | * warning and work with loadavg. | |
1427 | */ | |
1428 | n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", | |
1429 | n->dev->name, n->napi_id); | |
1430 | if (IS_ERR(n->thread)) { | |
1431 | err = PTR_ERR(n->thread); | |
1432 | pr_err("kthread_run failed with err %d\n", err); | |
1433 | n->thread = NULL; | |
1434 | } | |
1435 | ||
1436 | return err; | |
1437 | } | |
1438 | ||
40c900aa | 1439 | static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
1da177e4 | 1440 | { |
d314774c | 1441 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1442 | int ret; |
1da177e4 | 1443 | |
e46b66bc | 1444 | ASSERT_RTNL(); |
d07b26f5 | 1445 | dev_addr_check(dev); |
e46b66bc | 1446 | |
bd869245 HK |
1447 | if (!netif_device_present(dev)) { |
1448 | /* may be detached because parent is runtime-suspended */ | |
1449 | if (dev->dev.parent) | |
1450 | pm_runtime_resume(dev->dev.parent); | |
1451 | if (!netif_device_present(dev)) | |
1452 | return -ENODEV; | |
1453 | } | |
1da177e4 | 1454 | |
ca99ca14 NH |
1455 | /* Block netpoll from trying to do any rx path servicing. |
1456 | * If we don't do this there is a chance ndo_poll_controller | |
1457 | * or ndo_poll may be running while we open the device | |
1458 | */ | |
66b5552f | 1459 | netpoll_poll_disable(dev); |
ca99ca14 | 1460 | |
40c900aa | 1461 | ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); |
3b8bcfd5 JB |
1462 | ret = notifier_to_errno(ret); |
1463 | if (ret) | |
1464 | return ret; | |
1465 | ||
1da177e4 | 1466 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1467 | |
d314774c SH |
1468 | if (ops->ndo_validate_addr) |
1469 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1470 | |
d314774c SH |
1471 | if (!ret && ops->ndo_open) |
1472 | ret = ops->ndo_open(dev); | |
1da177e4 | 1473 | |
66b5552f | 1474 | netpoll_poll_enable(dev); |
ca99ca14 | 1475 | |
bada339b JG |
1476 | if (ret) |
1477 | clear_bit(__LINK_STATE_START, &dev->state); | |
1478 | else { | |
1da177e4 | 1479 | dev->flags |= IFF_UP; |
4417da66 | 1480 | dev_set_rx_mode(dev); |
1da177e4 | 1481 | dev_activate(dev); |
7bf23575 | 1482 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1483 | } |
bada339b | 1484 | |
1da177e4 LT |
1485 | return ret; |
1486 | } | |
1487 | ||
1488 | /** | |
bd380811 | 1489 | * dev_open - prepare an interface for use. |
00f54e68 PM |
1490 | * @dev: device to open |
1491 | * @extack: netlink extended ack | |
1da177e4 | 1492 | * |
bd380811 PM |
1493 | * Takes a device from down to up state. The device's private open |
1494 | * function is invoked and then the multicast lists are loaded. Finally | |
1495 | * the device is moved into the up state and a %NETDEV_UP message is | |
1496 | * sent to the netdev notifier chain. | |
1497 | * | |
1498 | * Calling this function on an active interface is a nop. On a failure | |
1499 | * a negative errno code is returned. | |
1da177e4 | 1500 | */ |
00f54e68 | 1501 | int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
bd380811 PM |
1502 | { |
1503 | int ret; | |
1504 | ||
bd380811 PM |
1505 | if (dev->flags & IFF_UP) |
1506 | return 0; | |
1507 | ||
40c900aa | 1508 | ret = __dev_open(dev, extack); |
bd380811 PM |
1509 | if (ret < 0) |
1510 | return ret; | |
1511 | ||
1d997f10 | 1512 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); |
bd380811 PM |
1513 | call_netdevice_notifiers(NETDEV_UP, dev); |
1514 | ||
1515 | return ret; | |
1516 | } | |
1517 | EXPORT_SYMBOL(dev_open); | |
1518 | ||
7051b88a | 1519 | static void __dev_close_many(struct list_head *head) |
1da177e4 | 1520 | { |
44345724 | 1521 | struct net_device *dev; |
e46b66bc | 1522 | |
bd380811 | 1523 | ASSERT_RTNL(); |
9d5010db DM |
1524 | might_sleep(); |
1525 | ||
5cde2829 | 1526 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1527 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1528 | netpoll_poll_disable(dev); |
3f4df206 | 1529 | |
44345724 | 1530 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1531 | |
44345724 | 1532 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1533 | |
44345724 OP |
1534 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1535 | * can be even on different cpu. So just clear netif_running(). | |
1536 | * | |
1537 | * dev->stop() will invoke napi_disable() on all of it's | |
1538 | * napi_struct instances on this device. | |
1539 | */ | |
4e857c58 | 1540 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1541 | } |
1da177e4 | 1542 | |
44345724 | 1543 | dev_deactivate_many(head); |
d8b2a4d2 | 1544 | |
5cde2829 | 1545 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1546 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1547 | |
44345724 OP |
1548 | /* |
1549 | * Call the device specific close. This cannot fail. | |
1550 | * Only if device is UP | |
1551 | * | |
1552 | * We allow it to be called even after a DETACH hot-plug | |
1553 | * event. | |
1554 | */ | |
1555 | if (ops->ndo_stop) | |
1556 | ops->ndo_stop(dev); | |
1557 | ||
44345724 | 1558 | dev->flags &= ~IFF_UP; |
66b5552f | 1559 | netpoll_poll_enable(dev); |
44345724 | 1560 | } |
44345724 OP |
1561 | } |
1562 | ||
7051b88a | 1563 | static void __dev_close(struct net_device *dev) |
44345724 OP |
1564 | { |
1565 | LIST_HEAD(single); | |
1566 | ||
5cde2829 | 1567 | list_add(&dev->close_list, &single); |
7051b88a | 1568 | __dev_close_many(&single); |
f87e6f47 | 1569 | list_del(&single); |
44345724 OP |
1570 | } |
1571 | ||
7051b88a | 1572 | void dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1573 | { |
1574 | struct net_device *dev, *tmp; | |
1da177e4 | 1575 | |
5cde2829 EB |
1576 | /* Remove the devices that don't need to be closed */ |
1577 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1578 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1579 | list_del_init(&dev->close_list); |
44345724 OP |
1580 | |
1581 | __dev_close_many(head); | |
1da177e4 | 1582 | |
5cde2829 | 1583 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
1d997f10 | 1584 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); |
44345724 | 1585 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1586 | if (unlink) |
1587 | list_del_init(&dev->close_list); | |
44345724 | 1588 | } |
bd380811 | 1589 | } |
99c4a26a | 1590 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1591 | |
1592 | /** | |
1593 | * dev_close - shutdown an interface. | |
1594 | * @dev: device to shutdown | |
1595 | * | |
1596 | * This function moves an active device into down state. A | |
1597 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1598 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1599 | * chain. | |
1600 | */ | |
7051b88a | 1601 | void dev_close(struct net_device *dev) |
bd380811 | 1602 | { |
e14a5993 ED |
1603 | if (dev->flags & IFF_UP) { |
1604 | LIST_HEAD(single); | |
1da177e4 | 1605 | |
5cde2829 | 1606 | list_add(&dev->close_list, &single); |
99c4a26a | 1607 | dev_close_many(&single, true); |
e14a5993 ED |
1608 | list_del(&single); |
1609 | } | |
1da177e4 | 1610 | } |
d1b19dff | 1611 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1612 | |
1613 | ||
0187bdfb BH |
1614 | /** |
1615 | * dev_disable_lro - disable Large Receive Offload on a device | |
1616 | * @dev: device | |
1617 | * | |
1618 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1619 | * called under RTNL. This is needed if received packets may be | |
1620 | * forwarded to another interface. | |
1621 | */ | |
1622 | void dev_disable_lro(struct net_device *dev) | |
1623 | { | |
fbe168ba MK |
1624 | struct net_device *lower_dev; |
1625 | struct list_head *iter; | |
529d0489 | 1626 | |
bc5787c6 MM |
1627 | dev->wanted_features &= ~NETIF_F_LRO; |
1628 | netdev_update_features(dev); | |
27660515 | 1629 | |
22d5969f MM |
1630 | if (unlikely(dev->features & NETIF_F_LRO)) |
1631 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1632 | |
1633 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1634 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1635 | } |
1636 | EXPORT_SYMBOL(dev_disable_lro); | |
1637 | ||
56f5aa77 MC |
1638 | /** |
1639 | * dev_disable_gro_hw - disable HW Generic Receive Offload on a device | |
1640 | * @dev: device | |
1641 | * | |
1642 | * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be | |
1643 | * called under RTNL. This is needed if Generic XDP is installed on | |
1644 | * the device. | |
1645 | */ | |
1646 | static void dev_disable_gro_hw(struct net_device *dev) | |
1647 | { | |
1648 | dev->wanted_features &= ~NETIF_F_GRO_HW; | |
1649 | netdev_update_features(dev); | |
1650 | ||
1651 | if (unlikely(dev->features & NETIF_F_GRO_HW)) | |
1652 | netdev_WARN(dev, "failed to disable GRO_HW!\n"); | |
1653 | } | |
1654 | ||
ede2762d KT |
1655 | const char *netdev_cmd_to_name(enum netdev_cmd cmd) |
1656 | { | |
1657 | #define N(val) \ | |
1658 | case NETDEV_##val: \ | |
1659 | return "NETDEV_" __stringify(val); | |
1660 | switch (cmd) { | |
1661 | N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) | |
1662 | N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) | |
1663 | N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) | |
02a68a47 JP |
1664 | N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) |
1665 | N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) | |
1666 | N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) | |
1667 | N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) | |
9daae9bd GP |
1668 | N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) |
1669 | N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) | |
9309f97a PM |
1670 | N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) |
1671 | N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) | |
5a178186 | 1672 | N(XDP_FEAT_CHANGE) |
3f5ecd8a | 1673 | } |
ede2762d KT |
1674 | #undef N |
1675 | return "UNKNOWN_NETDEV_EVENT"; | |
1676 | } | |
1677 | EXPORT_SYMBOL_GPL(netdev_cmd_to_name); | |
1678 | ||
351638e7 JP |
1679 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1680 | struct net_device *dev) | |
1681 | { | |
51d0c047 DA |
1682 | struct netdev_notifier_info info = { |
1683 | .dev = dev, | |
1684 | }; | |
351638e7 | 1685 | |
351638e7 JP |
1686 | return nb->notifier_call(nb, val, &info); |
1687 | } | |
0187bdfb | 1688 | |
afa0df59 JP |
1689 | static int call_netdevice_register_notifiers(struct notifier_block *nb, |
1690 | struct net_device *dev) | |
1691 | { | |
1692 | int err; | |
1693 | ||
1694 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); | |
1695 | err = notifier_to_errno(err); | |
1696 | if (err) | |
1697 | return err; | |
1698 | ||
1699 | if (!(dev->flags & IFF_UP)) | |
1700 | return 0; | |
1701 | ||
1702 | call_netdevice_notifier(nb, NETDEV_UP, dev); | |
1703 | return 0; | |
1704 | } | |
1705 | ||
1706 | static void call_netdevice_unregister_notifiers(struct notifier_block *nb, | |
1707 | struct net_device *dev) | |
1708 | { | |
1709 | if (dev->flags & IFF_UP) { | |
1710 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, | |
1711 | dev); | |
1712 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
1713 | } | |
1714 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); | |
1715 | } | |
1716 | ||
1717 | static int call_netdevice_register_net_notifiers(struct notifier_block *nb, | |
1718 | struct net *net) | |
1719 | { | |
1720 | struct net_device *dev; | |
1721 | int err; | |
1722 | ||
1723 | for_each_netdev(net, dev) { | |
1724 | err = call_netdevice_register_notifiers(nb, dev); | |
1725 | if (err) | |
1726 | goto rollback; | |
1727 | } | |
1728 | return 0; | |
1729 | ||
1730 | rollback: | |
1731 | for_each_netdev_continue_reverse(net, dev) | |
1732 | call_netdevice_unregister_notifiers(nb, dev); | |
1733 | return err; | |
1734 | } | |
1735 | ||
1736 | static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, | |
1737 | struct net *net) | |
1738 | { | |
1739 | struct net_device *dev; | |
1740 | ||
1741 | for_each_netdev(net, dev) | |
1742 | call_netdevice_unregister_notifiers(nb, dev); | |
1743 | } | |
1744 | ||
881d966b EB |
1745 | static int dev_boot_phase = 1; |
1746 | ||
1da177e4 | 1747 | /** |
722c9a0c | 1748 | * register_netdevice_notifier - register a network notifier block |
1749 | * @nb: notifier | |
1da177e4 | 1750 | * |
722c9a0c | 1751 | * Register a notifier to be called when network device events occur. |
1752 | * The notifier passed is linked into the kernel structures and must | |
1753 | * not be reused until it has been unregistered. A negative errno code | |
1754 | * is returned on a failure. | |
1da177e4 | 1755 | * |
722c9a0c | 1756 | * When registered all registration and up events are replayed |
1757 | * to the new notifier to allow device to have a race free | |
1758 | * view of the network device list. | |
1da177e4 LT |
1759 | */ |
1760 | ||
1761 | int register_netdevice_notifier(struct notifier_block *nb) | |
1762 | { | |
881d966b | 1763 | struct net *net; |
1da177e4 LT |
1764 | int err; |
1765 | ||
328fbe74 KT |
1766 | /* Close race with setup_net() and cleanup_net() */ |
1767 | down_write(&pernet_ops_rwsem); | |
1da177e4 | 1768 | rtnl_lock(); |
f07d5b94 | 1769 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1770 | if (err) |
1771 | goto unlock; | |
881d966b EB |
1772 | if (dev_boot_phase) |
1773 | goto unlock; | |
1774 | for_each_net(net) { | |
afa0df59 JP |
1775 | err = call_netdevice_register_net_notifiers(nb, net); |
1776 | if (err) | |
1777 | goto rollback; | |
1da177e4 | 1778 | } |
fcc5a03a HX |
1779 | |
1780 | unlock: | |
1da177e4 | 1781 | rtnl_unlock(); |
328fbe74 | 1782 | up_write(&pernet_ops_rwsem); |
1da177e4 | 1783 | return err; |
fcc5a03a HX |
1784 | |
1785 | rollback: | |
afa0df59 JP |
1786 | for_each_net_continue_reverse(net) |
1787 | call_netdevice_unregister_net_notifiers(nb, net); | |
c67625a1 PE |
1788 | |
1789 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1790 | goto unlock; |
1da177e4 | 1791 | } |
d1b19dff | 1792 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1793 | |
1794 | /** | |
722c9a0c | 1795 | * unregister_netdevice_notifier - unregister a network notifier block |
1796 | * @nb: notifier | |
1da177e4 | 1797 | * |
722c9a0c | 1798 | * Unregister a notifier previously registered by |
1799 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1800 | * kernel structures and may then be reused. A negative errno code | |
1801 | * is returned on a failure. | |
7d3d43da | 1802 | * |
722c9a0c | 1803 | * After unregistering unregister and down device events are synthesized |
1804 | * for all devices on the device list to the removed notifier to remove | |
1805 | * the need for special case cleanup code. | |
1da177e4 LT |
1806 | */ |
1807 | ||
1808 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1809 | { | |
7d3d43da | 1810 | struct net *net; |
9f514950 HX |
1811 | int err; |
1812 | ||
328fbe74 KT |
1813 | /* Close race with setup_net() and cleanup_net() */ |
1814 | down_write(&pernet_ops_rwsem); | |
9f514950 | 1815 | rtnl_lock(); |
f07d5b94 | 1816 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1817 | if (err) |
1818 | goto unlock; | |
1819 | ||
48b3a137 JP |
1820 | for_each_net(net) |
1821 | call_netdevice_unregister_net_notifiers(nb, net); | |
1822 | ||
7d3d43da | 1823 | unlock: |
9f514950 | 1824 | rtnl_unlock(); |
328fbe74 | 1825 | up_write(&pernet_ops_rwsem); |
9f514950 | 1826 | return err; |
1da177e4 | 1827 | } |
d1b19dff | 1828 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1829 | |
1f637703 JP |
1830 | static int __register_netdevice_notifier_net(struct net *net, |
1831 | struct notifier_block *nb, | |
1832 | bool ignore_call_fail) | |
1833 | { | |
1834 | int err; | |
1835 | ||
1836 | err = raw_notifier_chain_register(&net->netdev_chain, nb); | |
1837 | if (err) | |
1838 | return err; | |
1839 | if (dev_boot_phase) | |
1840 | return 0; | |
1841 | ||
1842 | err = call_netdevice_register_net_notifiers(nb, net); | |
1843 | if (err && !ignore_call_fail) | |
1844 | goto chain_unregister; | |
1845 | ||
1846 | return 0; | |
1847 | ||
1848 | chain_unregister: | |
1849 | raw_notifier_chain_unregister(&net->netdev_chain, nb); | |
1850 | return err; | |
1851 | } | |
1852 | ||
1853 | static int __unregister_netdevice_notifier_net(struct net *net, | |
1854 | struct notifier_block *nb) | |
1855 | { | |
1856 | int err; | |
1857 | ||
1858 | err = raw_notifier_chain_unregister(&net->netdev_chain, nb); | |
1859 | if (err) | |
1860 | return err; | |
1861 | ||
1862 | call_netdevice_unregister_net_notifiers(nb, net); | |
1863 | return 0; | |
1864 | } | |
1865 | ||
a30c7b42 JP |
1866 | /** |
1867 | * register_netdevice_notifier_net - register a per-netns network notifier block | |
1868 | * @net: network namespace | |
1869 | * @nb: notifier | |
1870 | * | |
1871 | * Register a notifier to be called when network device events occur. | |
1872 | * The notifier passed is linked into the kernel structures and must | |
1873 | * not be reused until it has been unregistered. A negative errno code | |
1874 | * is returned on a failure. | |
1875 | * | |
1876 | * When registered all registration and up events are replayed | |
1877 | * to the new notifier to allow device to have a race free | |
1878 | * view of the network device list. | |
1879 | */ | |
1880 | ||
1881 | int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) | |
1882 | { | |
1883 | int err; | |
1884 | ||
1885 | rtnl_lock(); | |
1f637703 | 1886 | err = __register_netdevice_notifier_net(net, nb, false); |
a30c7b42 JP |
1887 | rtnl_unlock(); |
1888 | return err; | |
a30c7b42 JP |
1889 | } |
1890 | EXPORT_SYMBOL(register_netdevice_notifier_net); | |
1891 | ||
1892 | /** | |
1893 | * unregister_netdevice_notifier_net - unregister a per-netns | |
1894 | * network notifier block | |
1895 | * @net: network namespace | |
1896 | * @nb: notifier | |
1897 | * | |
1898 | * Unregister a notifier previously registered by | |
9054b41c | 1899 | * register_netdevice_notifier_net(). The notifier is unlinked from the |
a30c7b42 JP |
1900 | * kernel structures and may then be reused. A negative errno code |
1901 | * is returned on a failure. | |
1902 | * | |
1903 | * After unregistering unregister and down device events are synthesized | |
1904 | * for all devices on the device list to the removed notifier to remove | |
1905 | * the need for special case cleanup code. | |
1906 | */ | |
1907 | ||
1908 | int unregister_netdevice_notifier_net(struct net *net, | |
1909 | struct notifier_block *nb) | |
1910 | { | |
1911 | int err; | |
1912 | ||
1913 | rtnl_lock(); | |
1f637703 | 1914 | err = __unregister_netdevice_notifier_net(net, nb); |
a30c7b42 JP |
1915 | rtnl_unlock(); |
1916 | return err; | |
1917 | } | |
1918 | EXPORT_SYMBOL(unregister_netdevice_notifier_net); | |
a30c7b42 | 1919 | |
3e52fba0 JP |
1920 | static void __move_netdevice_notifier_net(struct net *src_net, |
1921 | struct net *dst_net, | |
1922 | struct notifier_block *nb) | |
1923 | { | |
1924 | __unregister_netdevice_notifier_net(src_net, nb); | |
1925 | __register_netdevice_notifier_net(dst_net, nb, true); | |
1926 | } | |
1927 | ||
93642e14 JP |
1928 | int register_netdevice_notifier_dev_net(struct net_device *dev, |
1929 | struct notifier_block *nb, | |
1930 | struct netdev_net_notifier *nn) | |
1931 | { | |
1932 | int err; | |
a30c7b42 | 1933 | |
93642e14 JP |
1934 | rtnl_lock(); |
1935 | err = __register_netdevice_notifier_net(dev_net(dev), nb, false); | |
1936 | if (!err) { | |
1937 | nn->nb = nb; | |
1938 | list_add(&nn->list, &dev->net_notifier_list); | |
1939 | } | |
a30c7b42 JP |
1940 | rtnl_unlock(); |
1941 | return err; | |
1942 | } | |
93642e14 JP |
1943 | EXPORT_SYMBOL(register_netdevice_notifier_dev_net); |
1944 | ||
1945 | int unregister_netdevice_notifier_dev_net(struct net_device *dev, | |
1946 | struct notifier_block *nb, | |
1947 | struct netdev_net_notifier *nn) | |
1948 | { | |
1949 | int err; | |
1950 | ||
1951 | rtnl_lock(); | |
1952 | list_del(&nn->list); | |
1953 | err = __unregister_netdevice_notifier_net(dev_net(dev), nb); | |
1954 | rtnl_unlock(); | |
1955 | return err; | |
1956 | } | |
1957 | EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); | |
1958 | ||
1959 | static void move_netdevice_notifiers_dev_net(struct net_device *dev, | |
1960 | struct net *net) | |
1961 | { | |
1962 | struct netdev_net_notifier *nn; | |
1963 | ||
3e52fba0 JP |
1964 | list_for_each_entry(nn, &dev->net_notifier_list, list) |
1965 | __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); | |
93642e14 | 1966 | } |
a30c7b42 | 1967 | |
351638e7 JP |
1968 | /** |
1969 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1970 | * @val: value passed unmodified to notifier function | |
351638e7 JP |
1971 | * @info: notifier information data |
1972 | * | |
1973 | * Call all network notifier blocks. Parameters and return value | |
1974 | * are as for raw_notifier_call_chain(). | |
1975 | */ | |
1976 | ||
88c0a6b5 VO |
1977 | int call_netdevice_notifiers_info(unsigned long val, |
1978 | struct netdev_notifier_info *info) | |
351638e7 | 1979 | { |
a30c7b42 JP |
1980 | struct net *net = dev_net(info->dev); |
1981 | int ret; | |
1982 | ||
351638e7 | 1983 | ASSERT_RTNL(); |
a30c7b42 JP |
1984 | |
1985 | /* Run per-netns notifier block chain first, then run the global one. | |
1986 | * Hopefully, one day, the global one is going to be removed after | |
1987 | * all notifier block registrators get converted to be per-netns. | |
1988 | */ | |
1989 | ret = raw_notifier_call_chain(&net->netdev_chain, val, info); | |
1990 | if (ret & NOTIFY_STOP_MASK) | |
1991 | return ret; | |
351638e7 JP |
1992 | return raw_notifier_call_chain(&netdev_chain, val, info); |
1993 | } | |
351638e7 | 1994 | |
9309f97a PM |
1995 | /** |
1996 | * call_netdevice_notifiers_info_robust - call per-netns notifier blocks | |
1997 | * for and rollback on error | |
1998 | * @val_up: value passed unmodified to notifier function | |
1999 | * @val_down: value passed unmodified to the notifier function when | |
2000 | * recovering from an error on @val_up | |
2001 | * @info: notifier information data | |
2002 | * | |
2003 | * Call all per-netns network notifier blocks, but not notifier blocks on | |
2004 | * the global notifier chain. Parameters and return value are as for | |
2005 | * raw_notifier_call_chain_robust(). | |
2006 | */ | |
2007 | ||
2008 | static int | |
2009 | call_netdevice_notifiers_info_robust(unsigned long val_up, | |
2010 | unsigned long val_down, | |
2011 | struct netdev_notifier_info *info) | |
2012 | { | |
2013 | struct net *net = dev_net(info->dev); | |
2014 | ||
2015 | ASSERT_RTNL(); | |
2016 | ||
2017 | return raw_notifier_call_chain_robust(&net->netdev_chain, | |
2018 | val_up, val_down, info); | |
2019 | } | |
2020 | ||
26372605 PM |
2021 | static int call_netdevice_notifiers_extack(unsigned long val, |
2022 | struct net_device *dev, | |
2023 | struct netlink_ext_ack *extack) | |
2024 | { | |
2025 | struct netdev_notifier_info info = { | |
2026 | .dev = dev, | |
2027 | .extack = extack, | |
2028 | }; | |
2029 | ||
2030 | return call_netdevice_notifiers_info(val, &info); | |
2031 | } | |
2032 | ||
1da177e4 LT |
2033 | /** |
2034 | * call_netdevice_notifiers - call all network notifier blocks | |
2035 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 2036 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
2037 | * |
2038 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 2039 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
2040 | */ |
2041 | ||
ad7379d4 | 2042 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 2043 | { |
26372605 | 2044 | return call_netdevice_notifiers_extack(val, dev, NULL); |
1da177e4 | 2045 | } |
edf947f1 | 2046 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 2047 | |
af7d6cce SD |
2048 | /** |
2049 | * call_netdevice_notifiers_mtu - call all network notifier blocks | |
2050 | * @val: value passed unmodified to notifier function | |
2051 | * @dev: net_device pointer passed unmodified to notifier function | |
2052 | * @arg: additional u32 argument passed to the notifier function | |
2053 | * | |
2054 | * Call all network notifier blocks. Parameters and return value | |
2055 | * are as for raw_notifier_call_chain(). | |
2056 | */ | |
2057 | static int call_netdevice_notifiers_mtu(unsigned long val, | |
2058 | struct net_device *dev, u32 arg) | |
2059 | { | |
2060 | struct netdev_notifier_info_ext info = { | |
2061 | .info.dev = dev, | |
2062 | .ext.mtu = arg, | |
2063 | }; | |
2064 | ||
2065 | BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); | |
2066 | ||
2067 | return call_netdevice_notifiers_info(val, &info.info); | |
2068 | } | |
2069 | ||
1cf51900 | 2070 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 2071 | static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); |
4577139b DB |
2072 | |
2073 | void net_inc_ingress_queue(void) | |
2074 | { | |
aabf6772 | 2075 | static_branch_inc(&ingress_needed_key); |
4577139b DB |
2076 | } |
2077 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
2078 | ||
2079 | void net_dec_ingress_queue(void) | |
2080 | { | |
aabf6772 | 2081 | static_branch_dec(&ingress_needed_key); |
4577139b DB |
2082 | } |
2083 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
2084 | #endif | |
2085 | ||
1f211a1b | 2086 | #ifdef CONFIG_NET_EGRESS |
aabf6772 | 2087 | static DEFINE_STATIC_KEY_FALSE(egress_needed_key); |
1f211a1b DB |
2088 | |
2089 | void net_inc_egress_queue(void) | |
2090 | { | |
aabf6772 | 2091 | static_branch_inc(&egress_needed_key); |
1f211a1b DB |
2092 | } |
2093 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
2094 | ||
2095 | void net_dec_egress_queue(void) | |
2096 | { | |
aabf6772 | 2097 | static_branch_dec(&egress_needed_key); |
1f211a1b DB |
2098 | } |
2099 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
2100 | #endif | |
2101 | ||
047f340b AST |
2102 | #ifdef CONFIG_NET_CLS_ACT |
2103 | DEFINE_STATIC_KEY_FALSE(tcf_bypass_check_needed_key); | |
2104 | EXPORT_SYMBOL(tcf_bypass_check_needed_key); | |
2105 | #endif | |
2106 | ||
27942a15 MKL |
2107 | DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); |
2108 | EXPORT_SYMBOL(netstamp_needed_key); | |
e9666d10 | 2109 | #ifdef CONFIG_JUMP_LABEL |
b90e5794 | 2110 | static atomic_t netstamp_needed_deferred; |
13baa00a | 2111 | static atomic_t netstamp_wanted; |
5fa8bbda | 2112 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 2113 | { |
b90e5794 | 2114 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 2115 | int wanted; |
b90e5794 | 2116 | |
13baa00a ED |
2117 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
2118 | if (wanted > 0) | |
39e83922 | 2119 | static_branch_enable(&netstamp_needed_key); |
13baa00a | 2120 | else |
39e83922 | 2121 | static_branch_disable(&netstamp_needed_key); |
5fa8bbda ED |
2122 | } |
2123 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 2124 | #endif |
5fa8bbda ED |
2125 | |
2126 | void net_enable_timestamp(void) | |
2127 | { | |
e9666d10 | 2128 | #ifdef CONFIG_JUMP_LABEL |
6af645a5 | 2129 | int wanted = atomic_read(&netstamp_wanted); |
13baa00a | 2130 | |
6af645a5 ED |
2131 | while (wanted > 0) { |
2132 | if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) | |
13baa00a ED |
2133 | return; |
2134 | } | |
2135 | atomic_inc(&netstamp_needed_deferred); | |
2136 | schedule_work(&netstamp_work); | |
2137 | #else | |
39e83922 | 2138 | static_branch_inc(&netstamp_needed_key); |
13baa00a | 2139 | #endif |
1da177e4 | 2140 | } |
d1b19dff | 2141 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
2142 | |
2143 | void net_disable_timestamp(void) | |
2144 | { | |
e9666d10 | 2145 | #ifdef CONFIG_JUMP_LABEL |
6af645a5 | 2146 | int wanted = atomic_read(&netstamp_wanted); |
13baa00a | 2147 | |
6af645a5 ED |
2148 | while (wanted > 1) { |
2149 | if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) | |
13baa00a ED |
2150 | return; |
2151 | } | |
2152 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
2153 | schedule_work(&netstamp_work); |
2154 | #else | |
39e83922 | 2155 | static_branch_dec(&netstamp_needed_key); |
5fa8bbda | 2156 | #endif |
1da177e4 | 2157 | } |
d1b19dff | 2158 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 2159 | |
3b098e2d | 2160 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 2161 | { |
2456e855 | 2162 | skb->tstamp = 0; |
27942a15 | 2163 | skb->mono_delivery_time = 0; |
39e83922 | 2164 | if (static_branch_unlikely(&netstamp_needed_key)) |
d93376f5 | 2165 | skb->tstamp = ktime_get_real(); |
1da177e4 LT |
2166 | } |
2167 | ||
39e83922 DB |
2168 | #define net_timestamp_check(COND, SKB) \ |
2169 | if (static_branch_unlikely(&netstamp_needed_key)) { \ | |
2170 | if ((COND) && !(SKB)->tstamp) \ | |
d93376f5 | 2171 | (SKB)->tstamp = ktime_get_real(); \ |
39e83922 | 2172 | } \ |
3b098e2d | 2173 | |
f4b05d27 | 2174 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 | 2175 | { |
5f7d5728 | 2176 | return __is_skb_forwardable(dev, skb, true); |
79b569f0 | 2177 | } |
1ee481fb | 2178 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 2179 | |
5f7d5728 JDB |
2180 | static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, |
2181 | bool check_mtu) | |
a0265d28 | 2182 | { |
5f7d5728 | 2183 | int ret = ____dev_forward_skb(dev, skb, check_mtu); |
a0265d28 | 2184 | |
4e3264d2 MKL |
2185 | if (likely(!ret)) { |
2186 | skb->protocol = eth_type_trans(skb, dev); | |
2187 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
2188 | } | |
a0265d28 | 2189 | |
4e3264d2 | 2190 | return ret; |
a0265d28 | 2191 | } |
5f7d5728 JDB |
2192 | |
2193 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
2194 | { | |
2195 | return __dev_forward_skb2(dev, skb, true); | |
2196 | } | |
a0265d28 HX |
2197 | EXPORT_SYMBOL_GPL(__dev_forward_skb); |
2198 | ||
44540960 AB |
2199 | /** |
2200 | * dev_forward_skb - loopback an skb to another netif | |
2201 | * | |
2202 | * @dev: destination network device | |
2203 | * @skb: buffer to forward | |
2204 | * | |
2205 | * return values: | |
2206 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 2207 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
2208 | * |
2209 | * dev_forward_skb can be used for injecting an skb from the | |
2210 | * start_xmit function of one device into the receive queue | |
2211 | * of another device. | |
2212 | * | |
2213 | * The receiving device may be in another namespace, so | |
2214 | * we have to clear all information in the skb that could | |
2215 | * impact namespace isolation. | |
2216 | */ | |
2217 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
2218 | { | |
a0265d28 | 2219 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
2220 | } |
2221 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
2222 | ||
5f7d5728 JDB |
2223 | int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) |
2224 | { | |
2225 | return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); | |
2226 | } | |
2227 | ||
71d9dec2 CG |
2228 | static inline int deliver_skb(struct sk_buff *skb, |
2229 | struct packet_type *pt_prev, | |
2230 | struct net_device *orig_dev) | |
2231 | { | |
1f8b977a | 2232 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
1080e512 | 2233 | return -ENOMEM; |
63354797 | 2234 | refcount_inc(&skb->users); |
71d9dec2 CG |
2235 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
2236 | } | |
2237 | ||
7866a621 SN |
2238 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
2239 | struct packet_type **pt, | |
fbcb2170 JP |
2240 | struct net_device *orig_dev, |
2241 | __be16 type, | |
7866a621 SN |
2242 | struct list_head *ptype_list) |
2243 | { | |
2244 | struct packet_type *ptype, *pt_prev = *pt; | |
2245 | ||
2246 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
2247 | if (ptype->type != type) | |
2248 | continue; | |
2249 | if (pt_prev) | |
fbcb2170 | 2250 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
2251 | pt_prev = ptype; |
2252 | } | |
2253 | *pt = pt_prev; | |
2254 | } | |
2255 | ||
c0de08d0 EL |
2256 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
2257 | { | |
a3d744e9 | 2258 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
2259 | return false; |
2260 | ||
2261 | if (ptype->id_match) | |
2262 | return ptype->id_match(ptype, skb->sk); | |
2263 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
2264 | return true; | |
2265 | ||
2266 | return false; | |
2267 | } | |
2268 | ||
9f9a742d MR |
2269 | /** |
2270 | * dev_nit_active - return true if any network interface taps are in use | |
2271 | * | |
2272 | * @dev: network device to check for the presence of taps | |
2273 | */ | |
2274 | bool dev_nit_active(struct net_device *dev) | |
2275 | { | |
0b91fa4b ED |
2276 | return !list_empty(&net_hotdata.ptype_all) || |
2277 | !list_empty(&dev->ptype_all); | |
9f9a742d MR |
2278 | } |
2279 | EXPORT_SYMBOL_GPL(dev_nit_active); | |
2280 | ||
1da177e4 LT |
2281 | /* |
2282 | * Support routine. Sends outgoing frames to any network | |
2283 | * taps currently in use. | |
2284 | */ | |
2285 | ||
74b20582 | 2286 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 | 2287 | { |
0b91fa4b ED |
2288 | struct list_head *ptype_list = &net_hotdata.ptype_all; |
2289 | struct packet_type *ptype, *pt_prev = NULL; | |
71d9dec2 | 2290 | struct sk_buff *skb2 = NULL; |
a61bbcf2 | 2291 | |
1da177e4 | 2292 | rcu_read_lock(); |
7866a621 SN |
2293 | again: |
2294 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
6ebfad33 | 2295 | if (READ_ONCE(ptype->ignore_outgoing)) |
fa788d98 VW |
2296 | continue; |
2297 | ||
1da177e4 LT |
2298 | /* Never send packets back to the socket |
2299 | * they originated from - MvS (miquels@drinkel.ow.org) | |
2300 | */ | |
7866a621 SN |
2301 | if (skb_loop_sk(ptype, skb)) |
2302 | continue; | |
71d9dec2 | 2303 | |
7866a621 SN |
2304 | if (pt_prev) { |
2305 | deliver_skb(skb2, pt_prev, skb->dev); | |
2306 | pt_prev = ptype; | |
2307 | continue; | |
2308 | } | |
1da177e4 | 2309 | |
7866a621 SN |
2310 | /* need to clone skb, done only once */ |
2311 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
2312 | if (!skb2) | |
2313 | goto out_unlock; | |
70978182 | 2314 | |
7866a621 | 2315 | net_timestamp_set(skb2); |
1da177e4 | 2316 | |
7866a621 SN |
2317 | /* skb->nh should be correctly |
2318 | * set by sender, so that the second statement is | |
2319 | * just protection against buggy protocols. | |
2320 | */ | |
2321 | skb_reset_mac_header(skb2); | |
2322 | ||
2323 | if (skb_network_header(skb2) < skb2->data || | |
2324 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
2325 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
2326 | ntohs(skb2->protocol), | |
2327 | dev->name); | |
2328 | skb_reset_network_header(skb2); | |
1da177e4 | 2329 | } |
7866a621 SN |
2330 | |
2331 | skb2->transport_header = skb2->network_header; | |
2332 | skb2->pkt_type = PACKET_OUTGOING; | |
2333 | pt_prev = ptype; | |
2334 | } | |
2335 | ||
0b91fa4b | 2336 | if (ptype_list == &net_hotdata.ptype_all) { |
7866a621 SN |
2337 | ptype_list = &dev->ptype_all; |
2338 | goto again; | |
1da177e4 | 2339 | } |
7866a621 | 2340 | out_unlock: |
581fe0ea WB |
2341 | if (pt_prev) { |
2342 | if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) | |
2343 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
2344 | else | |
2345 | kfree_skb(skb2); | |
2346 | } | |
1da177e4 LT |
2347 | rcu_read_unlock(); |
2348 | } | |
74b20582 | 2349 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 2350 | |
2c53040f BH |
2351 | /** |
2352 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
2353 | * @dev: Network device |
2354 | * @txq: number of queues available | |
2355 | * | |
2356 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
2357 | * valid. To resolve this verify the tc mapping remains valid and if | |
2358 | * not NULL the mapping. With no priorities mapping to this | |
2359 | * offset/count pair it will no longer be used. In the worst case TC0 | |
2360 | * is invalid nothing can be done so disable priority mappings. If is | |
2361 | * expected that drivers will fix this mapping if they can before | |
2362 | * calling netif_set_real_num_tx_queues. | |
2363 | */ | |
bb134d22 | 2364 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
2365 | { |
2366 | int i; | |
2367 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2368 | ||
2369 | /* If TC0 is invalidated disable TC mapping */ | |
2370 | if (tc->offset + tc->count > txq) { | |
5b92be64 | 2371 | netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
2372 | dev->num_tc = 0; |
2373 | return; | |
2374 | } | |
2375 | ||
2376 | /* Invalidated prio to tc mappings set to TC0 */ | |
2377 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
2378 | int q = netdev_get_prio_tc_map(dev, i); | |
2379 | ||
2380 | tc = &dev->tc_to_txq[q]; | |
2381 | if (tc->offset + tc->count > txq) { | |
5b92be64 JB |
2382 | netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
2383 | i, q); | |
4f57c087 JF |
2384 | netdev_set_prio_tc_map(dev, i, 0); |
2385 | } | |
2386 | } | |
2387 | } | |
2388 | ||
8d059b0f AD |
2389 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
2390 | { | |
2391 | if (dev->num_tc) { | |
2392 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2393 | int i; | |
2394 | ||
ffcfe25b | 2395 | /* walk through the TCs and see if it falls into any of them */ |
8d059b0f AD |
2396 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { |
2397 | if ((txq - tc->offset) < tc->count) | |
2398 | return i; | |
2399 | } | |
2400 | ||
ffcfe25b | 2401 | /* didn't find it, just return -1 to indicate no match */ |
8d059b0f AD |
2402 | return -1; |
2403 | } | |
2404 | ||
2405 | return 0; | |
2406 | } | |
8a5f2166 | 2407 | EXPORT_SYMBOL(netdev_txq_to_tc); |
8d059b0f | 2408 | |
537c00de | 2409 | #ifdef CONFIG_XPS |
5da9ace3 VO |
2410 | static struct static_key xps_needed __read_mostly; |
2411 | static struct static_key xps_rxqs_needed __read_mostly; | |
537c00de AD |
2412 | static DEFINE_MUTEX(xps_map_mutex); |
2413 | #define xmap_dereference(P) \ | |
2414 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
2415 | ||
6234f874 | 2416 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
2d05bf01 | 2417 | struct xps_dev_maps *old_maps, int tci, u16 index) |
537c00de | 2418 | { |
10cdc3f3 AD |
2419 | struct xps_map *map = NULL; |
2420 | int pos; | |
537c00de | 2421 | |
f080864a | 2422 | map = xmap_dereference(dev_maps->attr_map[tci]); |
6234f874 AD |
2423 | if (!map) |
2424 | return false; | |
537c00de | 2425 | |
6234f874 AD |
2426 | for (pos = map->len; pos--;) { |
2427 | if (map->queues[pos] != index) | |
2428 | continue; | |
2429 | ||
2430 | if (map->len > 1) { | |
2431 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2432 | break; |
537c00de | 2433 | } |
6234f874 | 2434 | |
2d05bf01 AT |
2435 | if (old_maps) |
2436 | RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); | |
80d19669 | 2437 | RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); |
6234f874 AD |
2438 | kfree_rcu(map, rcu); |
2439 | return false; | |
537c00de AD |
2440 | } |
2441 | ||
6234f874 | 2442 | return true; |
10cdc3f3 AD |
2443 | } |
2444 | ||
6234f874 AD |
2445 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2446 | struct xps_dev_maps *dev_maps, | |
2447 | int cpu, u16 offset, u16 count) | |
2448 | { | |
255c04a8 | 2449 | int num_tc = dev_maps->num_tc; |
184c449f AD |
2450 | bool active = false; |
2451 | int tci; | |
6234f874 | 2452 | |
184c449f AD |
2453 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2454 | int i, j; | |
2455 | ||
2456 | for (i = count, j = offset; i--; j++) { | |
2d05bf01 | 2457 | if (!remove_xps_queue(dev_maps, NULL, tci, j)) |
184c449f AD |
2458 | break; |
2459 | } | |
2460 | ||
2461 | active |= i < 0; | |
6234f874 AD |
2462 | } |
2463 | ||
184c449f | 2464 | return active; |
6234f874 AD |
2465 | } |
2466 | ||
867d0ad4 SD |
2467 | static void reset_xps_maps(struct net_device *dev, |
2468 | struct xps_dev_maps *dev_maps, | |
044ab86d | 2469 | enum xps_map_type type) |
867d0ad4 | 2470 | { |
867d0ad4 | 2471 | static_key_slow_dec_cpuslocked(&xps_needed); |
044ab86d AT |
2472 | if (type == XPS_RXQS) |
2473 | static_key_slow_dec_cpuslocked(&xps_rxqs_needed); | |
2474 | ||
2475 | RCU_INIT_POINTER(dev->xps_maps[type], NULL); | |
2476 | ||
867d0ad4 SD |
2477 | kfree_rcu(dev_maps, rcu); |
2478 | } | |
2479 | ||
044ab86d AT |
2480 | static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, |
2481 | u16 offset, u16 count) | |
80d19669 | 2482 | { |
044ab86d | 2483 | struct xps_dev_maps *dev_maps; |
80d19669 AN |
2484 | bool active = false; |
2485 | int i, j; | |
2486 | ||
044ab86d AT |
2487 | dev_maps = xmap_dereference(dev->xps_maps[type]); |
2488 | if (!dev_maps) | |
2489 | return; | |
2490 | ||
6f36158e AT |
2491 | for (j = 0; j < dev_maps->nr_ids; j++) |
2492 | active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); | |
867d0ad4 | 2493 | if (!active) |
044ab86d | 2494 | reset_xps_maps(dev, dev_maps, type); |
80d19669 | 2495 | |
044ab86d | 2496 | if (type == XPS_CPUS) { |
6f36158e | 2497 | for (i = offset + (count - 1); count--; i--) |
f28c020f | 2498 | netdev_queue_numa_node_write( |
6f36158e | 2499 | netdev_get_tx_queue(dev, i), NUMA_NO_NODE); |
80d19669 AN |
2500 | } |
2501 | } | |
2502 | ||
6234f874 AD |
2503 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, |
2504 | u16 count) | |
10cdc3f3 | 2505 | { |
04157469 AN |
2506 | if (!static_key_false(&xps_needed)) |
2507 | return; | |
10cdc3f3 | 2508 | |
4d99f660 | 2509 | cpus_read_lock(); |
04157469 | 2510 | mutex_lock(&xps_map_mutex); |
10cdc3f3 | 2511 | |
044ab86d AT |
2512 | if (static_key_false(&xps_rxqs_needed)) |
2513 | clean_xps_maps(dev, XPS_RXQS, offset, count); | |
80d19669 | 2514 | |
044ab86d | 2515 | clean_xps_maps(dev, XPS_CPUS, offset, count); |
024e9679 | 2516 | |
537c00de | 2517 | mutex_unlock(&xps_map_mutex); |
4d99f660 | 2518 | cpus_read_unlock(); |
537c00de AD |
2519 | } |
2520 | ||
6234f874 AD |
2521 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2522 | { | |
2523 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2524 | } | |
2525 | ||
80d19669 AN |
2526 | static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, |
2527 | u16 index, bool is_rxqs_map) | |
01c5f864 AD |
2528 | { |
2529 | struct xps_map *new_map; | |
2530 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2531 | int i, pos; | |
2532 | ||
2533 | for (pos = 0; map && pos < map->len; pos++) { | |
2534 | if (map->queues[pos] != index) | |
2535 | continue; | |
2536 | return map; | |
2537 | } | |
2538 | ||
80d19669 | 2539 | /* Need to add tx-queue to this CPU's/rx-queue's existing map */ |
01c5f864 AD |
2540 | if (map) { |
2541 | if (pos < map->alloc_len) | |
2542 | return map; | |
2543 | ||
2544 | alloc_len = map->alloc_len * 2; | |
2545 | } | |
2546 | ||
80d19669 AN |
2547 | /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's |
2548 | * map | |
2549 | */ | |
2550 | if (is_rxqs_map) | |
2551 | new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); | |
2552 | else | |
2553 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2554 | cpu_to_node(attr_index)); | |
01c5f864 AD |
2555 | if (!new_map) |
2556 | return NULL; | |
2557 | ||
2558 | for (i = 0; i < pos; i++) | |
2559 | new_map->queues[i] = map->queues[i]; | |
2560 | new_map->alloc_len = alloc_len; | |
2561 | new_map->len = pos; | |
2562 | ||
2563 | return new_map; | |
2564 | } | |
2565 | ||
402fbb99 AT |
2566 | /* Copy xps maps at a given index */ |
2567 | static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, | |
2568 | struct xps_dev_maps *new_dev_maps, int index, | |
2569 | int tc, bool skip_tc) | |
2570 | { | |
2571 | int i, tci = index * dev_maps->num_tc; | |
2572 | struct xps_map *map; | |
2573 | ||
2574 | /* copy maps belonging to foreign traffic classes */ | |
2575 | for (i = 0; i < dev_maps->num_tc; i++, tci++) { | |
2576 | if (i == tc && skip_tc) | |
2577 | continue; | |
2578 | ||
2579 | /* fill in the new device map from the old device map */ | |
2580 | map = xmap_dereference(dev_maps->attr_map[tci]); | |
2581 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
2582 | } | |
2583 | } | |
2584 | ||
4d99f660 | 2585 | /* Must be called under cpus_read_lock */ |
80d19669 | 2586 | int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, |
044ab86d | 2587 | u16 index, enum xps_map_type type) |
537c00de | 2588 | { |
2d05bf01 | 2589 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; |
6f36158e | 2590 | const unsigned long *online_mask = NULL; |
255c04a8 | 2591 | bool active = false, copy = false; |
80d19669 | 2592 | int i, j, tci, numa_node_id = -2; |
184c449f | 2593 | int maps_sz, num_tc = 1, tc = 0; |
537c00de | 2594 | struct xps_map *map, *new_map; |
80d19669 | 2595 | unsigned int nr_ids; |
537c00de | 2596 | |
5dd0dfd5 NC |
2597 | WARN_ON_ONCE(index >= dev->num_tx_queues); |
2598 | ||
184c449f | 2599 | if (dev->num_tc) { |
ffcfe25b | 2600 | /* Do not allow XPS on subordinate device directly */ |
184c449f | 2601 | num_tc = dev->num_tc; |
ffcfe25b AD |
2602 | if (num_tc < 0) |
2603 | return -EINVAL; | |
2604 | ||
2605 | /* If queue belongs to subordinate dev use its map */ | |
2606 | dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; | |
2607 | ||
184c449f AD |
2608 | tc = netdev_txq_to_tc(dev, index); |
2609 | if (tc < 0) | |
2610 | return -EINVAL; | |
2611 | } | |
2612 | ||
537c00de | 2613 | mutex_lock(&xps_map_mutex); |
044ab86d AT |
2614 | |
2615 | dev_maps = xmap_dereference(dev->xps_maps[type]); | |
2616 | if (type == XPS_RXQS) { | |
80d19669 | 2617 | maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); |
80d19669 AN |
2618 | nr_ids = dev->num_rx_queues; |
2619 | } else { | |
2620 | maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); | |
6f36158e | 2621 | if (num_possible_cpus() > 1) |
80d19669 | 2622 | online_mask = cpumask_bits(cpu_online_mask); |
80d19669 AN |
2623 | nr_ids = nr_cpu_ids; |
2624 | } | |
537c00de | 2625 | |
80d19669 AN |
2626 | if (maps_sz < L1_CACHE_BYTES) |
2627 | maps_sz = L1_CACHE_BYTES; | |
537c00de | 2628 | |
255c04a8 | 2629 | /* The old dev_maps could be larger or smaller than the one we're |
5478fcd0 AT |
2630 | * setting up now, as dev->num_tc or nr_ids could have been updated in |
2631 | * between. We could try to be smart, but let's be safe instead and only | |
2632 | * copy foreign traffic classes if the two map sizes match. | |
255c04a8 | 2633 | */ |
5478fcd0 AT |
2634 | if (dev_maps && |
2635 | dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) | |
255c04a8 AT |
2636 | copy = true; |
2637 | ||
01c5f864 | 2638 | /* allocate memory for queue storage */ |
80d19669 AN |
2639 | for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), |
2640 | j < nr_ids;) { | |
2bb60cb9 | 2641 | if (!new_dev_maps) { |
255c04a8 AT |
2642 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); |
2643 | if (!new_dev_maps) { | |
2644 | mutex_unlock(&xps_map_mutex); | |
2645 | return -ENOMEM; | |
2646 | } | |
2647 | ||
5478fcd0 | 2648 | new_dev_maps->nr_ids = nr_ids; |
255c04a8 | 2649 | new_dev_maps->num_tc = num_tc; |
2bb60cb9 | 2650 | } |
01c5f864 | 2651 | |
80d19669 | 2652 | tci = j * num_tc + tc; |
255c04a8 | 2653 | map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; |
01c5f864 | 2654 | |
044ab86d | 2655 | map = expand_xps_map(map, j, index, type == XPS_RXQS); |
01c5f864 AD |
2656 | if (!map) |
2657 | goto error; | |
2658 | ||
80d19669 | 2659 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); |
01c5f864 AD |
2660 | } |
2661 | ||
2662 | if (!new_dev_maps) | |
2663 | goto out_no_new_maps; | |
2664 | ||
867d0ad4 SD |
2665 | if (!dev_maps) { |
2666 | /* Increment static keys at most once per type */ | |
2667 | static_key_slow_inc_cpuslocked(&xps_needed); | |
044ab86d | 2668 | if (type == XPS_RXQS) |
867d0ad4 SD |
2669 | static_key_slow_inc_cpuslocked(&xps_rxqs_needed); |
2670 | } | |
04157469 | 2671 | |
6f36158e | 2672 | for (j = 0; j < nr_ids; j++) { |
402fbb99 | 2673 | bool skip_tc = false; |
184c449f | 2674 | |
80d19669 | 2675 | tci = j * num_tc + tc; |
80d19669 AN |
2676 | if (netif_attr_test_mask(j, mask, nr_ids) && |
2677 | netif_attr_test_online(j, online_mask, nr_ids)) { | |
2678 | /* add tx-queue to CPU/rx-queue maps */ | |
01c5f864 AD |
2679 | int pos = 0; |
2680 | ||
402fbb99 AT |
2681 | skip_tc = true; |
2682 | ||
80d19669 | 2683 | map = xmap_dereference(new_dev_maps->attr_map[tci]); |
01c5f864 AD |
2684 | while ((pos < map->len) && (map->queues[pos] != index)) |
2685 | pos++; | |
2686 | ||
2687 | if (pos == map->len) | |
2688 | map->queues[map->len++] = index; | |
537c00de | 2689 | #ifdef CONFIG_NUMA |
044ab86d | 2690 | if (type == XPS_CPUS) { |
80d19669 AN |
2691 | if (numa_node_id == -2) |
2692 | numa_node_id = cpu_to_node(j); | |
2693 | else if (numa_node_id != cpu_to_node(j)) | |
2694 | numa_node_id = -1; | |
2695 | } | |
537c00de | 2696 | #endif |
537c00de | 2697 | } |
01c5f864 | 2698 | |
402fbb99 AT |
2699 | if (copy) |
2700 | xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, | |
2701 | skip_tc); | |
537c00de AD |
2702 | } |
2703 | ||
044ab86d | 2704 | rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); |
01c5f864 | 2705 | |
537c00de | 2706 | /* Cleanup old maps */ |
184c449f AD |
2707 | if (!dev_maps) |
2708 | goto out_no_old_maps; | |
2709 | ||
6f36158e | 2710 | for (j = 0; j < dev_maps->nr_ids; j++) { |
255c04a8 | 2711 | for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { |
80d19669 | 2712 | map = xmap_dereference(dev_maps->attr_map[tci]); |
255c04a8 AT |
2713 | if (!map) |
2714 | continue; | |
2715 | ||
2716 | if (copy) { | |
2717 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
2718 | if (map == new_map) | |
2719 | continue; | |
2720 | } | |
2721 | ||
75b2758a | 2722 | RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); |
255c04a8 | 2723 | kfree_rcu(map, rcu); |
01c5f864 | 2724 | } |
537c00de AD |
2725 | } |
2726 | ||
2d05bf01 | 2727 | old_dev_maps = dev_maps; |
184c449f AD |
2728 | |
2729 | out_no_old_maps: | |
01c5f864 AD |
2730 | dev_maps = new_dev_maps; |
2731 | active = true; | |
537c00de | 2732 | |
01c5f864 | 2733 | out_no_new_maps: |
044ab86d | 2734 | if (type == XPS_CPUS) |
80d19669 AN |
2735 | /* update Tx queue numa node */ |
2736 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), | |
2737 | (numa_node_id >= 0) ? | |
2738 | numa_node_id : NUMA_NO_NODE); | |
537c00de | 2739 | |
01c5f864 AD |
2740 | if (!dev_maps) |
2741 | goto out_no_maps; | |
2742 | ||
80d19669 | 2743 | /* removes tx-queue from unused CPUs/rx-queues */ |
6f36158e | 2744 | for (j = 0; j < dev_maps->nr_ids; j++) { |
132f743b AT |
2745 | tci = j * dev_maps->num_tc; |
2746 | ||
2747 | for (i = 0; i < dev_maps->num_tc; i++, tci++) { | |
2748 | if (i == tc && | |
2749 | netif_attr_test_mask(j, mask, dev_maps->nr_ids) && | |
2750 | netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) | |
2751 | continue; | |
2752 | ||
2d05bf01 AT |
2753 | active |= remove_xps_queue(dev_maps, |
2754 | copy ? old_dev_maps : NULL, | |
2755 | tci, index); | |
132f743b | 2756 | } |
01c5f864 AD |
2757 | } |
2758 | ||
2d05bf01 AT |
2759 | if (old_dev_maps) |
2760 | kfree_rcu(old_dev_maps, rcu); | |
2761 | ||
01c5f864 | 2762 | /* free map if not active */ |
867d0ad4 | 2763 | if (!active) |
044ab86d | 2764 | reset_xps_maps(dev, dev_maps, type); |
01c5f864 AD |
2765 | |
2766 | out_no_maps: | |
537c00de AD |
2767 | mutex_unlock(&xps_map_mutex); |
2768 | ||
2769 | return 0; | |
2770 | error: | |
01c5f864 | 2771 | /* remove any maps that we added */ |
6f36158e | 2772 | for (j = 0; j < nr_ids; j++) { |
80d19669 AN |
2773 | for (i = num_tc, tci = j * num_tc; i--; tci++) { |
2774 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
255c04a8 | 2775 | map = copy ? |
80d19669 | 2776 | xmap_dereference(dev_maps->attr_map[tci]) : |
184c449f AD |
2777 | NULL; |
2778 | if (new_map && new_map != map) | |
2779 | kfree(new_map); | |
2780 | } | |
01c5f864 AD |
2781 | } |
2782 | ||
537c00de AD |
2783 | mutex_unlock(&xps_map_mutex); |
2784 | ||
537c00de AD |
2785 | kfree(new_dev_maps); |
2786 | return -ENOMEM; | |
2787 | } | |
4d99f660 | 2788 | EXPORT_SYMBOL_GPL(__netif_set_xps_queue); |
80d19669 AN |
2789 | |
2790 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, | |
2791 | u16 index) | |
2792 | { | |
4d99f660 AV |
2793 | int ret; |
2794 | ||
2795 | cpus_read_lock(); | |
044ab86d | 2796 | ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); |
4d99f660 AV |
2797 | cpus_read_unlock(); |
2798 | ||
2799 | return ret; | |
80d19669 | 2800 | } |
537c00de AD |
2801 | EXPORT_SYMBOL(netif_set_xps_queue); |
2802 | ||
2803 | #endif | |
ffcfe25b AD |
2804 | static void netdev_unbind_all_sb_channels(struct net_device *dev) |
2805 | { | |
2806 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2807 | ||
2808 | /* Unbind any subordinate channels */ | |
2809 | while (txq-- != &dev->_tx[0]) { | |
2810 | if (txq->sb_dev) | |
2811 | netdev_unbind_sb_channel(dev, txq->sb_dev); | |
2812 | } | |
2813 | } | |
2814 | ||
9cf1f6a8 AD |
2815 | void netdev_reset_tc(struct net_device *dev) |
2816 | { | |
6234f874 AD |
2817 | #ifdef CONFIG_XPS |
2818 | netif_reset_xps_queues_gt(dev, 0); | |
2819 | #endif | |
ffcfe25b AD |
2820 | netdev_unbind_all_sb_channels(dev); |
2821 | ||
2822 | /* Reset TC configuration of device */ | |
9cf1f6a8 AD |
2823 | dev->num_tc = 0; |
2824 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2825 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2826 | } | |
2827 | EXPORT_SYMBOL(netdev_reset_tc); | |
2828 | ||
2829 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2830 | { | |
2831 | if (tc >= dev->num_tc) | |
2832 | return -EINVAL; | |
2833 | ||
6234f874 AD |
2834 | #ifdef CONFIG_XPS |
2835 | netif_reset_xps_queues(dev, offset, count); | |
2836 | #endif | |
9cf1f6a8 AD |
2837 | dev->tc_to_txq[tc].count = count; |
2838 | dev->tc_to_txq[tc].offset = offset; | |
2839 | return 0; | |
2840 | } | |
2841 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2842 | ||
2843 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2844 | { | |
2845 | if (num_tc > TC_MAX_QUEUE) | |
2846 | return -EINVAL; | |
2847 | ||
6234f874 AD |
2848 | #ifdef CONFIG_XPS |
2849 | netif_reset_xps_queues_gt(dev, 0); | |
2850 | #endif | |
ffcfe25b AD |
2851 | netdev_unbind_all_sb_channels(dev); |
2852 | ||
9cf1f6a8 AD |
2853 | dev->num_tc = num_tc; |
2854 | return 0; | |
2855 | } | |
2856 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2857 | ||
ffcfe25b AD |
2858 | void netdev_unbind_sb_channel(struct net_device *dev, |
2859 | struct net_device *sb_dev) | |
2860 | { | |
2861 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2862 | ||
2863 | #ifdef CONFIG_XPS | |
2864 | netif_reset_xps_queues_gt(sb_dev, 0); | |
2865 | #endif | |
2866 | memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); | |
2867 | memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); | |
2868 | ||
2869 | while (txq-- != &dev->_tx[0]) { | |
2870 | if (txq->sb_dev == sb_dev) | |
2871 | txq->sb_dev = NULL; | |
2872 | } | |
2873 | } | |
2874 | EXPORT_SYMBOL(netdev_unbind_sb_channel); | |
2875 | ||
2876 | int netdev_bind_sb_channel_queue(struct net_device *dev, | |
2877 | struct net_device *sb_dev, | |
2878 | u8 tc, u16 count, u16 offset) | |
2879 | { | |
2880 | /* Make certain the sb_dev and dev are already configured */ | |
2881 | if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) | |
2882 | return -EINVAL; | |
2883 | ||
2884 | /* We cannot hand out queues we don't have */ | |
2885 | if ((offset + count) > dev->real_num_tx_queues) | |
2886 | return -EINVAL; | |
2887 | ||
2888 | /* Record the mapping */ | |
2889 | sb_dev->tc_to_txq[tc].count = count; | |
2890 | sb_dev->tc_to_txq[tc].offset = offset; | |
2891 | ||
2892 | /* Provide a way for Tx queue to find the tc_to_txq map or | |
2893 | * XPS map for itself. | |
2894 | */ | |
2895 | while (count--) | |
2896 | netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; | |
2897 | ||
2898 | return 0; | |
2899 | } | |
2900 | EXPORT_SYMBOL(netdev_bind_sb_channel_queue); | |
2901 | ||
2902 | int netdev_set_sb_channel(struct net_device *dev, u16 channel) | |
2903 | { | |
2904 | /* Do not use a multiqueue device to represent a subordinate channel */ | |
2905 | if (netif_is_multiqueue(dev)) | |
2906 | return -ENODEV; | |
2907 | ||
2908 | /* We allow channels 1 - 32767 to be used for subordinate channels. | |
2909 | * Channel 0 is meant to be "native" mode and used only to represent | |
2910 | * the main root device. We allow writing 0 to reset the device back | |
2911 | * to normal mode after being used as a subordinate channel. | |
2912 | */ | |
2913 | if (channel > S16_MAX) | |
2914 | return -EINVAL; | |
2915 | ||
2916 | dev->num_tc = -channel; | |
2917 | ||
2918 | return 0; | |
2919 | } | |
2920 | EXPORT_SYMBOL(netdev_set_sb_channel); | |
2921 | ||
f0796d5c JF |
2922 | /* |
2923 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
3a053b1a | 2924 | * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. |
f0796d5c | 2925 | */ |
e6484930 | 2926 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2927 | { |
ac5b7019 | 2928 | bool disabling; |
1d24eb48 TH |
2929 | int rc; |
2930 | ||
ac5b7019 JK |
2931 | disabling = txq < dev->real_num_tx_queues; |
2932 | ||
e6484930 TH |
2933 | if (txq < 1 || txq > dev->num_tx_queues) |
2934 | return -EINVAL; | |
f0796d5c | 2935 | |
5c56580b BH |
2936 | if (dev->reg_state == NETREG_REGISTERED || |
2937 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2938 | ASSERT_RTNL(); |
2939 | ||
1d24eb48 TH |
2940 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2941 | txq); | |
bf264145 TH |
2942 | if (rc) |
2943 | return rc; | |
2944 | ||
4f57c087 JF |
2945 | if (dev->num_tc) |
2946 | netif_setup_tc(dev, txq); | |
2947 | ||
1e080f17 JK |
2948 | dev_qdisc_change_real_num_tx(dev, txq); |
2949 | ||
ac5b7019 JK |
2950 | dev->real_num_tx_queues = txq; |
2951 | ||
2952 | if (disabling) { | |
2953 | synchronize_net(); | |
e6484930 | 2954 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2955 | #ifdef CONFIG_XPS |
2956 | netif_reset_xps_queues_gt(dev, txq); | |
2957 | #endif | |
2958 | } | |
ac5b7019 JK |
2959 | } else { |
2960 | dev->real_num_tx_queues = txq; | |
f0796d5c | 2961 | } |
e6484930 | 2962 | |
e6484930 | 2963 | return 0; |
f0796d5c JF |
2964 | } |
2965 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2966 | |
a953be53 | 2967 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2968 | /** |
2969 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2970 | * @dev: Network device | |
2971 | * @rxq: Actual number of RX queues | |
2972 | * | |
2973 | * This must be called either with the rtnl_lock held or before | |
2974 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2975 | * negative error code. If called before registration, it always |
2976 | * succeeds. | |
62fe0b40 BH |
2977 | */ |
2978 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2979 | { | |
2980 | int rc; | |
2981 | ||
bd25fa7b TH |
2982 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2983 | return -EINVAL; | |
2984 | ||
62fe0b40 BH |
2985 | if (dev->reg_state == NETREG_REGISTERED) { |
2986 | ASSERT_RTNL(); | |
2987 | ||
62fe0b40 BH |
2988 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2989 | rxq); | |
2990 | if (rc) | |
2991 | return rc; | |
62fe0b40 BH |
2992 | } |
2993 | ||
2994 | dev->real_num_rx_queues = rxq; | |
2995 | return 0; | |
2996 | } | |
2997 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2998 | #endif | |
2999 | ||
271e5b7d JK |
3000 | /** |
3001 | * netif_set_real_num_queues - set actual number of RX and TX queues used | |
3002 | * @dev: Network device | |
3003 | * @txq: Actual number of TX queues | |
3004 | * @rxq: Actual number of RX queues | |
3005 | * | |
3006 | * Set the real number of both TX and RX queues. | |
3007 | * Does nothing if the number of queues is already correct. | |
3008 | */ | |
3009 | int netif_set_real_num_queues(struct net_device *dev, | |
3010 | unsigned int txq, unsigned int rxq) | |
3011 | { | |
3012 | unsigned int old_rxq = dev->real_num_rx_queues; | |
3013 | int err; | |
3014 | ||
3015 | if (txq < 1 || txq > dev->num_tx_queues || | |
3016 | rxq < 1 || rxq > dev->num_rx_queues) | |
3017 | return -EINVAL; | |
3018 | ||
3019 | /* Start from increases, so the error path only does decreases - | |
3020 | * decreases can't fail. | |
3021 | */ | |
3022 | if (rxq > dev->real_num_rx_queues) { | |
3023 | err = netif_set_real_num_rx_queues(dev, rxq); | |
3024 | if (err) | |
3025 | return err; | |
3026 | } | |
3027 | if (txq > dev->real_num_tx_queues) { | |
3028 | err = netif_set_real_num_tx_queues(dev, txq); | |
3029 | if (err) | |
3030 | goto undo_rx; | |
3031 | } | |
3032 | if (rxq < dev->real_num_rx_queues) | |
3033 | WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); | |
3034 | if (txq < dev->real_num_tx_queues) | |
3035 | WARN_ON(netif_set_real_num_tx_queues(dev, txq)); | |
3036 | ||
3037 | return 0; | |
3038 | undo_rx: | |
3039 | WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); | |
3040 | return err; | |
3041 | } | |
3042 | EXPORT_SYMBOL(netif_set_real_num_queues); | |
3043 | ||
14d7b812 JK |
3044 | /** |
3045 | * netif_set_tso_max_size() - set the max size of TSO frames supported | |
3046 | * @dev: netdev to update | |
3047 | * @size: max skb->len of a TSO frame | |
3048 | * | |
3049 | * Set the limit on the size of TSO super-frames the device can handle. | |
7c4e983c AD |
3050 | * Unless explicitly set the stack will assume the value of |
3051 | * %GSO_LEGACY_MAX_SIZE. | |
14d7b812 JK |
3052 | */ |
3053 | void netif_set_tso_max_size(struct net_device *dev, unsigned int size) | |
3054 | { | |
7c4e983c | 3055 | dev->tso_max_size = min(GSO_MAX_SIZE, size); |
14d7b812 JK |
3056 | if (size < READ_ONCE(dev->gso_max_size)) |
3057 | netif_set_gso_max_size(dev, size); | |
9eefedd5 XL |
3058 | if (size < READ_ONCE(dev->gso_ipv4_max_size)) |
3059 | netif_set_gso_ipv4_max_size(dev, size); | |
14d7b812 JK |
3060 | } |
3061 | EXPORT_SYMBOL(netif_set_tso_max_size); | |
3062 | ||
3063 | /** | |
3064 | * netif_set_tso_max_segs() - set the max number of segs supported for TSO | |
3065 | * @dev: netdev to update | |
3066 | * @segs: max number of TCP segments | |
3067 | * | |
3068 | * Set the limit on the number of TCP segments the device can generate from | |
3069 | * a single TSO super-frame. | |
3070 | * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. | |
3071 | */ | |
3072 | void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) | |
3073 | { | |
3074 | dev->tso_max_segs = segs; | |
3075 | if (segs < READ_ONCE(dev->gso_max_segs)) | |
3076 | netif_set_gso_max_segs(dev, segs); | |
3077 | } | |
3078 | EXPORT_SYMBOL(netif_set_tso_max_segs); | |
3079 | ||
6df6398f JK |
3080 | /** |
3081 | * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper | |
3082 | * @to: netdev to update | |
3083 | * @from: netdev from which to copy the limits | |
3084 | */ | |
3085 | void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) | |
3086 | { | |
ee8b7a11 JK |
3087 | netif_set_tso_max_size(to, from->tso_max_size); |
3088 | netif_set_tso_max_segs(to, from->tso_max_segs); | |
6df6398f JK |
3089 | } |
3090 | EXPORT_SYMBOL(netif_inherit_tso_max); | |
3091 | ||
2c53040f BH |
3092 | /** |
3093 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 | 3094 | * |
046e1537 ÍH |
3095 | * Default value is the number of physical cores if there are only 1 or 2, or |
3096 | * divided by 2 if there are more. | |
16917b87 | 3097 | */ |
a55b138b | 3098 | int netif_get_num_default_rss_queues(void) |
16917b87 | 3099 | { |
046e1537 ÍH |
3100 | cpumask_var_t cpus; |
3101 | int cpu, count = 0; | |
3102 | ||
3103 | if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) | |
3104 | return 1; | |
3105 | ||
3106 | cpumask_copy(cpus, cpu_online_mask); | |
3107 | for_each_cpu(cpu, cpus) { | |
3108 | ++count; | |
3109 | cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); | |
3110 | } | |
3111 | free_cpumask_var(cpus); | |
3112 | ||
3113 | return count > 2 ? DIV_ROUND_UP(count, 2) : count; | |
16917b87 YM |
3114 | } |
3115 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
3116 | ||
3bcb846c | 3117 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 3118 | { |
def82a1d JP |
3119 | struct softnet_data *sd; |
3120 | unsigned long flags; | |
56079431 | 3121 | |
def82a1d | 3122 | local_irq_save(flags); |
903ceff7 | 3123 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
3124 | q->next_sched = NULL; |
3125 | *sd->output_queue_tailp = q; | |
3126 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
3127 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
3128 | local_irq_restore(flags); | |
3129 | } | |
3130 | ||
3131 | void __netif_schedule(struct Qdisc *q) | |
3132 | { | |
3133 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
3134 | __netif_reschedule(q); | |
56079431 DV |
3135 | } |
3136 | EXPORT_SYMBOL(__netif_schedule); | |
3137 | ||
e6247027 | 3138 | struct dev_kfree_skb_cb { |
40bbae58 | 3139 | enum skb_drop_reason reason; |
e6247027 ED |
3140 | }; |
3141 | ||
3142 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 3143 | { |
e6247027 ED |
3144 | return (struct dev_kfree_skb_cb *)skb->cb; |
3145 | } | |
3146 | ||
46e5da40 JF |
3147 | void netif_schedule_queue(struct netdev_queue *txq) |
3148 | { | |
3149 | rcu_read_lock(); | |
5be5515a | 3150 | if (!netif_xmit_stopped(txq)) { |
46e5da40 JF |
3151 | struct Qdisc *q = rcu_dereference(txq->qdisc); |
3152 | ||
3153 | __netif_schedule(q); | |
3154 | } | |
3155 | rcu_read_unlock(); | |
3156 | } | |
3157 | EXPORT_SYMBOL(netif_schedule_queue); | |
3158 | ||
46e5da40 JF |
3159 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
3160 | { | |
3161 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
3162 | struct Qdisc *q; | |
3163 | ||
3164 | rcu_read_lock(); | |
3165 | q = rcu_dereference(dev_queue->qdisc); | |
3166 | __netif_schedule(q); | |
3167 | rcu_read_unlock(); | |
3168 | } | |
3169 | } | |
3170 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
3171 | ||
40bbae58 | 3172 | void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) |
56079431 | 3173 | { |
e6247027 | 3174 | unsigned long flags; |
56079431 | 3175 | |
9899886d MJ |
3176 | if (unlikely(!skb)) |
3177 | return; | |
3178 | ||
63354797 | 3179 | if (likely(refcount_read(&skb->users) == 1)) { |
e6247027 | 3180 | smp_rmb(); |
63354797 RE |
3181 | refcount_set(&skb->users, 0); |
3182 | } else if (likely(!refcount_dec_and_test(&skb->users))) { | |
e6247027 | 3183 | return; |
bea3348e | 3184 | } |
e6247027 ED |
3185 | get_kfree_skb_cb(skb)->reason = reason; |
3186 | local_irq_save(flags); | |
3187 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
3188 | __this_cpu_write(softnet_data.completion_queue, skb); | |
3189 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
3190 | local_irq_restore(flags); | |
56079431 | 3191 | } |
40bbae58 | 3192 | EXPORT_SYMBOL(dev_kfree_skb_irq_reason); |
56079431 | 3193 | |
40bbae58 | 3194 | void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) |
56079431 | 3195 | { |
afa79d08 | 3196 | if (in_hardirq() || irqs_disabled()) |
40bbae58 | 3197 | dev_kfree_skb_irq_reason(skb, reason); |
56079431 | 3198 | else |
40bbae58 | 3199 | kfree_skb_reason(skb, reason); |
56079431 | 3200 | } |
40bbae58 | 3201 | EXPORT_SYMBOL(dev_kfree_skb_any_reason); |
56079431 DV |
3202 | |
3203 | ||
bea3348e SH |
3204 | /** |
3205 | * netif_device_detach - mark device as removed | |
3206 | * @dev: network device | |
3207 | * | |
3208 | * Mark device as removed from system and therefore no longer available. | |
3209 | */ | |
56079431 DV |
3210 | void netif_device_detach(struct net_device *dev) |
3211 | { | |
3212 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
3213 | netif_running(dev)) { | |
d543103a | 3214 | netif_tx_stop_all_queues(dev); |
56079431 DV |
3215 | } |
3216 | } | |
3217 | EXPORT_SYMBOL(netif_device_detach); | |
3218 | ||
bea3348e SH |
3219 | /** |
3220 | * netif_device_attach - mark device as attached | |
3221 | * @dev: network device | |
3222 | * | |
3223 | * Mark device as attached from system and restart if needed. | |
3224 | */ | |
56079431 DV |
3225 | void netif_device_attach(struct net_device *dev) |
3226 | { | |
3227 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
3228 | netif_running(dev)) { | |
d543103a | 3229 | netif_tx_wake_all_queues(dev); |
4ec93edb | 3230 | __netdev_watchdog_up(dev); |
56079431 DV |
3231 | } |
3232 | } | |
3233 | EXPORT_SYMBOL(netif_device_attach); | |
3234 | ||
5605c762 JP |
3235 | /* |
3236 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
3237 | * to be used as a distribution range. | |
3238 | */ | |
eadec877 AD |
3239 | static u16 skb_tx_hash(const struct net_device *dev, |
3240 | const struct net_device *sb_dev, | |
3241 | struct sk_buff *skb) | |
5605c762 JP |
3242 | { |
3243 | u32 hash; | |
3244 | u16 qoffset = 0; | |
1b837d48 | 3245 | u16 qcount = dev->real_num_tx_queues; |
5605c762 | 3246 | |
eadec877 AD |
3247 | if (dev->num_tc) { |
3248 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
3249 | ||
3250 | qoffset = sb_dev->tc_to_txq[tc].offset; | |
3251 | qcount = sb_dev->tc_to_txq[tc].count; | |
0c57eeec MC |
3252 | if (unlikely(!qcount)) { |
3253 | net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", | |
3254 | sb_dev->name, qoffset, tc); | |
3255 | qoffset = 0; | |
3256 | qcount = dev->real_num_tx_queues; | |
3257 | } | |
eadec877 AD |
3258 | } |
3259 | ||
5605c762 | 3260 | if (skb_rx_queue_recorded(skb)) { |
066b8678 | 3261 | DEBUG_NET_WARN_ON_ONCE(qcount == 0); |
5605c762 | 3262 | hash = skb_get_rx_queue(skb); |
6e11d157 AN |
3263 | if (hash >= qoffset) |
3264 | hash -= qoffset; | |
1b837d48 AD |
3265 | while (unlikely(hash >= qcount)) |
3266 | hash -= qcount; | |
eadec877 | 3267 | return hash + qoffset; |
5605c762 JP |
3268 | } |
3269 | ||
3270 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
3271 | } | |
5605c762 | 3272 | |
d457a0e3 | 3273 | void skb_warn_bad_offload(const struct sk_buff *skb) |
36c92474 | 3274 | { |
84d15ae5 | 3275 | static const netdev_features_t null_features; |
36c92474 | 3276 | struct net_device *dev = skb->dev; |
88ad4175 | 3277 | const char *name = ""; |
36c92474 | 3278 | |
c846ad9b BG |
3279 | if (!net_ratelimit()) |
3280 | return; | |
3281 | ||
88ad4175 BM |
3282 | if (dev) { |
3283 | if (dev->dev.parent) | |
3284 | name = dev_driver_string(dev->dev.parent); | |
3285 | else | |
3286 | name = netdev_name(dev); | |
3287 | } | |
6413139d WB |
3288 | skb_dump(KERN_WARNING, skb, false); |
3289 | WARN(1, "%s: caps=(%pNF, %pNF)\n", | |
88ad4175 | 3290 | name, dev ? &dev->features : &null_features, |
6413139d | 3291 | skb->sk ? &skb->sk->sk_route_caps : &null_features); |
36c92474 BH |
3292 | } |
3293 | ||
1da177e4 LT |
3294 | /* |
3295 | * Invalidate hardware checksum when packet is to be mangled, and | |
3296 | * complete checksum manually on outgoing path. | |
3297 | */ | |
84fa7933 | 3298 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 3299 | { |
d3bc23e7 | 3300 | __wsum csum; |
663ead3b | 3301 | int ret = 0, offset; |
1da177e4 | 3302 | |
84fa7933 | 3303 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
3304 | goto out_set_summed; |
3305 | ||
3aefd7d6 | 3306 | if (unlikely(skb_is_gso(skb))) { |
36c92474 BH |
3307 | skb_warn_bad_offload(skb); |
3308 | return -EINVAL; | |
1da177e4 LT |
3309 | } |
3310 | ||
cef401de ED |
3311 | /* Before computing a checksum, we should make sure no frag could |
3312 | * be modified by an external entity : checksum could be wrong. | |
3313 | */ | |
3314 | if (skb_has_shared_frag(skb)) { | |
3315 | ret = __skb_linearize(skb); | |
3316 | if (ret) | |
3317 | goto out; | |
3318 | } | |
3319 | ||
55508d60 | 3320 | offset = skb_checksum_start_offset(skb); |
d7ea0d9d | 3321 | ret = -EINVAL; |
26c29961 | 3322 | if (unlikely(offset >= skb_headlen(skb))) { |
eeee4b77 | 3323 | DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); |
26c29961 ED |
3324 | WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n", |
3325 | offset, skb_headlen(skb)); | |
d7ea0d9d | 3326 | goto out; |
eeee4b77 | 3327 | } |
a030847e HX |
3328 | csum = skb_checksum(skb, offset, skb->len - offset, 0); |
3329 | ||
3330 | offset += skb->csum_offset; | |
26c29961 | 3331 | if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) { |
eeee4b77 | 3332 | DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); |
26c29961 ED |
3333 | WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n", |
3334 | offset + sizeof(__sum16), skb_headlen(skb)); | |
d7ea0d9d | 3335 | goto out; |
eeee4b77 | 3336 | } |
8211fbfa HK |
3337 | ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); |
3338 | if (ret) | |
3339 | goto out; | |
1da177e4 | 3340 | |
4f2e4ad5 | 3341 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 3342 | out_set_summed: |
1da177e4 | 3343 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 3344 | out: |
1da177e4 LT |
3345 | return ret; |
3346 | } | |
d1b19dff | 3347 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 3348 | |
b72b5bf6 DC |
3349 | int skb_crc32c_csum_help(struct sk_buff *skb) |
3350 | { | |
3351 | __le32 crc32c_csum; | |
3352 | int ret = 0, offset, start; | |
3353 | ||
3354 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
3355 | goto out; | |
3356 | ||
3357 | if (unlikely(skb_is_gso(skb))) | |
3358 | goto out; | |
3359 | ||
3360 | /* Before computing a checksum, we should make sure no frag could | |
3361 | * be modified by an external entity : checksum could be wrong. | |
3362 | */ | |
3363 | if (unlikely(skb_has_shared_frag(skb))) { | |
3364 | ret = __skb_linearize(skb); | |
3365 | if (ret) | |
3366 | goto out; | |
3367 | } | |
3368 | start = skb_checksum_start_offset(skb); | |
3369 | offset = start + offsetof(struct sctphdr, checksum); | |
3370 | if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { | |
3371 | ret = -EINVAL; | |
3372 | goto out; | |
3373 | } | |
8211fbfa HK |
3374 | |
3375 | ret = skb_ensure_writable(skb, offset + sizeof(__le32)); | |
3376 | if (ret) | |
3377 | goto out; | |
3378 | ||
b72b5bf6 DC |
3379 | crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, |
3380 | skb->len - start, ~(__u32)0, | |
3381 | crc32c_csum_stub)); | |
3382 | *(__le32 *)(skb->data + offset) = crc32c_csum; | |
c24831a1 | 3383 | skb_reset_csum_not_inet(skb); |
b72b5bf6 DC |
3384 | out: |
3385 | return ret; | |
3386 | } | |
3387 | ||
53d6471c | 3388 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 3389 | { |
252e3346 | 3390 | __be16 type = skb->protocol; |
f6a78bfc | 3391 | |
19acc327 PS |
3392 | /* Tunnel gso handlers can set protocol to ethernet. */ |
3393 | if (type == htons(ETH_P_TEB)) { | |
3394 | struct ethhdr *eth; | |
3395 | ||
3396 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
3397 | return 0; | |
3398 | ||
1dfe82eb | 3399 | eth = (struct ethhdr *)skb->data; |
19acc327 PS |
3400 | type = eth->h_proto; |
3401 | } | |
3402 | ||
4063384e | 3403 | return vlan_get_protocol_and_depth(skb, type, depth); |
ec5f0615 PS |
3404 | } |
3405 | ||
f6a78bfc | 3406 | |
fb286bb2 HX |
3407 | /* Take action when hardware reception checksum errors are detected. */ |
3408 | #ifdef CONFIG_BUG | |
127d7355 TL |
3409 | static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) |
3410 | { | |
5b92be64 | 3411 | netdev_err(dev, "hw csum failure\n"); |
127d7355 TL |
3412 | skb_dump(KERN_ERR, skb, true); |
3413 | dump_stack(); | |
3414 | } | |
3415 | ||
7fe50ac8 | 3416 | void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) |
fb286bb2 | 3417 | { |
127d7355 | 3418 | DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); |
fb286bb2 HX |
3419 | } |
3420 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
3421 | #endif | |
3422 | ||
ab74cfeb | 3423 | /* XXX: check that highmem exists at all on the given machine. */ |
c1e756bf | 3424 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 3425 | { |
3d3a8533 | 3426 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 3427 | int i; |
f4563a75 | 3428 | |
5acbbd42 | 3429 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
3430 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
3431 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 3432 | |
ea2ab693 | 3433 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 3434 | return 1; |
ea2ab693 | 3435 | } |
5acbbd42 | 3436 | } |
3d3a8533 | 3437 | #endif |
1da177e4 LT |
3438 | return 0; |
3439 | } | |
1da177e4 | 3440 | |
3b392ddb SH |
3441 | /* If MPLS offload request, verify we are testing hardware MPLS features |
3442 | * instead of standard features for the netdev. | |
3443 | */ | |
d0edc7bf | 3444 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
3445 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
3446 | netdev_features_t features, | |
3447 | __be16 type) | |
3448 | { | |
25cd9ba0 | 3449 | if (eth_p_mpls(type)) |
3b392ddb SH |
3450 | features &= skb->dev->mpls_features; |
3451 | ||
3452 | return features; | |
3453 | } | |
3454 | #else | |
3455 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
3456 | netdev_features_t features, | |
3457 | __be16 type) | |
3458 | { | |
3459 | return features; | |
3460 | } | |
3461 | #endif | |
3462 | ||
c8f44aff | 3463 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 3464 | netdev_features_t features) |
f01a5236 | 3465 | { |
3b392ddb SH |
3466 | __be16 type; |
3467 | ||
9fc95f50 | 3468 | type = skb_network_protocol(skb, NULL); |
3b392ddb | 3469 | features = net_mpls_features(skb, features, type); |
53d6471c | 3470 | |
c0d680e5 | 3471 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 3472 | !can_checksum_protocol(features, type)) { |
996e8021 | 3473 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 3474 | } |
7be2c82c ED |
3475 | if (illegal_highdma(skb->dev, skb)) |
3476 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
3477 | |
3478 | return features; | |
3479 | } | |
3480 | ||
e38f3025 TM |
3481 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
3482 | struct net_device *dev, | |
3483 | netdev_features_t features) | |
3484 | { | |
3485 | return features; | |
3486 | } | |
3487 | EXPORT_SYMBOL(passthru_features_check); | |
3488 | ||
7ce23672 | 3489 | static netdev_features_t dflt_features_check(struct sk_buff *skb, |
8cb65d00 TM |
3490 | struct net_device *dev, |
3491 | netdev_features_t features) | |
3492 | { | |
3493 | return vlan_features_check(skb, features); | |
3494 | } | |
3495 | ||
cbc53e08 AD |
3496 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
3497 | struct net_device *dev, | |
3498 | netdev_features_t features) | |
3499 | { | |
3500 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
3501 | ||
6d872df3 | 3502 | if (gso_segs > READ_ONCE(dev->gso_max_segs)) |
cbc53e08 AD |
3503 | return features & ~NETIF_F_GSO_MASK; |
3504 | ||
24ab059d ED |
3505 | if (unlikely(skb->len >= READ_ONCE(dev->gso_max_size))) |
3506 | return features & ~NETIF_F_GSO_MASK; | |
3507 | ||
1d155dfd HK |
3508 | if (!skb_shinfo(skb)->gso_type) { |
3509 | skb_warn_bad_offload(skb); | |
3510 | return features & ~NETIF_F_GSO_MASK; | |
3511 | } | |
3512 | ||
802ab55a AD |
3513 | /* Support for GSO partial features requires software |
3514 | * intervention before we can actually process the packets | |
3515 | * so we need to strip support for any partial features now | |
3516 | * and we can pull them back in after we have partially | |
3517 | * segmented the frame. | |
3518 | */ | |
3519 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
3520 | features &= ~dev->gso_partial_features; | |
3521 | ||
3522 | /* Make sure to clear the IPv4 ID mangling feature if the | |
3523 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
3524 | */ |
3525 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
3526 | struct iphdr *iph = skb->encapsulation ? | |
3527 | inner_ip_hdr(skb) : ip_hdr(skb); | |
3528 | ||
3529 | if (!(iph->frag_off & htons(IP_DF))) | |
3530 | features &= ~NETIF_F_TSO_MANGLEID; | |
3531 | } | |
3532 | ||
3533 | return features; | |
3534 | } | |
3535 | ||
c1e756bf | 3536 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 3537 | { |
5f35227e | 3538 | struct net_device *dev = skb->dev; |
fcbeb976 | 3539 | netdev_features_t features = dev->features; |
58e998c6 | 3540 | |
cbc53e08 AD |
3541 | if (skb_is_gso(skb)) |
3542 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 3543 | |
5f35227e JG |
3544 | /* If encapsulation offload request, verify we are testing |
3545 | * hardware encapsulation features instead of standard | |
3546 | * features for the netdev | |
3547 | */ | |
3548 | if (skb->encapsulation) | |
3549 | features &= dev->hw_enc_features; | |
3550 | ||
f5a7fb88 TM |
3551 | if (skb_vlan_tagged(skb)) |
3552 | features = netdev_intersect_features(features, | |
3553 | dev->vlan_features | | |
3554 | NETIF_F_HW_VLAN_CTAG_TX | | |
3555 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 3556 | |
5f35227e JG |
3557 | if (dev->netdev_ops->ndo_features_check) |
3558 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
3559 | features); | |
8cb65d00 TM |
3560 | else |
3561 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 3562 | |
c1e756bf | 3563 | return harmonize_features(skb, features); |
58e998c6 | 3564 | } |
c1e756bf | 3565 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 3566 | |
2ea25513 | 3567 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 3568 | struct netdev_queue *txq, bool more) |
f6a78bfc | 3569 | { |
2ea25513 DM |
3570 | unsigned int len; |
3571 | int rc; | |
00829823 | 3572 | |
9f9a742d | 3573 | if (dev_nit_active(dev)) |
2ea25513 | 3574 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 3575 | |
2ea25513 DM |
3576 | len = skb->len; |
3577 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 3578 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 3579 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 3580 | |
2ea25513 DM |
3581 | return rc; |
3582 | } | |
7b9c6090 | 3583 | |
8dcda22a DM |
3584 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
3585 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
3586 | { |
3587 | struct sk_buff *skb = first; | |
3588 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 3589 | |
7f2e870f DM |
3590 | while (skb) { |
3591 | struct sk_buff *next = skb->next; | |
fc70fb64 | 3592 | |
a8305bff | 3593 | skb_mark_not_on_list(skb); |
95f6b3dd | 3594 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
3595 | if (unlikely(!dev_xmit_complete(rc))) { |
3596 | skb->next = next; | |
3597 | goto out; | |
3598 | } | |
6afff0ca | 3599 | |
7f2e870f | 3600 | skb = next; |
fe60faa5 | 3601 | if (netif_tx_queue_stopped(txq) && skb) { |
7f2e870f DM |
3602 | rc = NETDEV_TX_BUSY; |
3603 | break; | |
9ccb8975 | 3604 | } |
7f2e870f | 3605 | } |
9ccb8975 | 3606 | |
7f2e870f DM |
3607 | out: |
3608 | *ret = rc; | |
3609 | return skb; | |
3610 | } | |
b40863c6 | 3611 | |
1ff0dc94 ED |
3612 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
3613 | netdev_features_t features) | |
f6a78bfc | 3614 | { |
df8a39de | 3615 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
3616 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
3617 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
3618 | return skb; |
3619 | } | |
f6a78bfc | 3620 | |
43c26a1a DC |
3621 | int skb_csum_hwoffload_help(struct sk_buff *skb, |
3622 | const netdev_features_t features) | |
3623 | { | |
fa821170 | 3624 | if (unlikely(skb_csum_is_sctp(skb))) |
43c26a1a DC |
3625 | return !!(features & NETIF_F_SCTP_CRC) ? 0 : |
3626 | skb_crc32c_csum_help(skb); | |
3627 | ||
62fafcd6 XL |
3628 | if (features & NETIF_F_HW_CSUM) |
3629 | return 0; | |
3630 | ||
3631 | if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { | |
3632 | switch (skb->csum_offset) { | |
3633 | case offsetof(struct tcphdr, check): | |
3634 | case offsetof(struct udphdr, check): | |
3635 | return 0; | |
3636 | } | |
3637 | } | |
3638 | ||
3639 | return skb_checksum_help(skb); | |
43c26a1a DC |
3640 | } |
3641 | EXPORT_SYMBOL(skb_csum_hwoffload_help); | |
3642 | ||
f53c7239 | 3643 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) |
eae3f88e DM |
3644 | { |
3645 | netdev_features_t features; | |
f6a78bfc | 3646 | |
eae3f88e DM |
3647 | features = netif_skb_features(skb); |
3648 | skb = validate_xmit_vlan(skb, features); | |
3649 | if (unlikely(!skb)) | |
3650 | goto out_null; | |
7b9c6090 | 3651 | |
ebf4e808 IL |
3652 | skb = sk_validate_xmit_skb(skb, dev); |
3653 | if (unlikely(!skb)) | |
3654 | goto out_null; | |
3655 | ||
8b86a61d | 3656 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
3657 | struct sk_buff *segs; |
3658 | ||
3659 | segs = skb_gso_segment(skb, features); | |
cecda693 | 3660 | if (IS_ERR(segs)) { |
af6dabc9 | 3661 | goto out_kfree_skb; |
cecda693 JW |
3662 | } else if (segs) { |
3663 | consume_skb(skb); | |
3664 | skb = segs; | |
f6a78bfc | 3665 | } |
eae3f88e DM |
3666 | } else { |
3667 | if (skb_needs_linearize(skb, features) && | |
3668 | __skb_linearize(skb)) | |
3669 | goto out_kfree_skb; | |
4ec93edb | 3670 | |
eae3f88e DM |
3671 | /* If packet is not checksummed and device does not |
3672 | * support checksumming for this protocol, complete | |
3673 | * checksumming here. | |
3674 | */ | |
3675 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
3676 | if (skb->encapsulation) | |
3677 | skb_set_inner_transport_header(skb, | |
3678 | skb_checksum_start_offset(skb)); | |
3679 | else | |
3680 | skb_set_transport_header(skb, | |
3681 | skb_checksum_start_offset(skb)); | |
43c26a1a | 3682 | if (skb_csum_hwoffload_help(skb, features)) |
eae3f88e | 3683 | goto out_kfree_skb; |
7b9c6090 | 3684 | } |
0c772159 | 3685 | } |
7b9c6090 | 3686 | |
f53c7239 | 3687 | skb = validate_xmit_xfrm(skb, features, again); |
3dca3f38 | 3688 | |
eae3f88e | 3689 | return skb; |
fc70fb64 | 3690 | |
f6a78bfc HX |
3691 | out_kfree_skb: |
3692 | kfree_skb(skb); | |
eae3f88e | 3693 | out_null: |
625788b5 | 3694 | dev_core_stats_tx_dropped_inc(dev); |
eae3f88e DM |
3695 | return NULL; |
3696 | } | |
6afff0ca | 3697 | |
f53c7239 | 3698 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) |
55a93b3e ED |
3699 | { |
3700 | struct sk_buff *next, *head = NULL, *tail; | |
3701 | ||
bec3cfdc | 3702 | for (; skb != NULL; skb = next) { |
55a93b3e | 3703 | next = skb->next; |
a8305bff | 3704 | skb_mark_not_on_list(skb); |
bec3cfdc ED |
3705 | |
3706 | /* in case skb wont be segmented, point to itself */ | |
3707 | skb->prev = skb; | |
3708 | ||
f53c7239 | 3709 | skb = validate_xmit_skb(skb, dev, again); |
bec3cfdc ED |
3710 | if (!skb) |
3711 | continue; | |
55a93b3e | 3712 | |
bec3cfdc ED |
3713 | if (!head) |
3714 | head = skb; | |
3715 | else | |
3716 | tail->next = skb; | |
3717 | /* If skb was segmented, skb->prev points to | |
3718 | * the last segment. If not, it still contains skb. | |
3719 | */ | |
3720 | tail = skb->prev; | |
55a93b3e ED |
3721 | } |
3722 | return head; | |
f6a78bfc | 3723 | } |
104ba78c | 3724 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3725 | |
1def9238 ED |
3726 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3727 | { | |
3728 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3729 | ||
3730 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3731 | ||
3732 | /* To get more precise estimation of bytes sent on wire, | |
3733 | * we add to pkt_len the headers size of all segments | |
3734 | */ | |
a0dce875 | 3735 | if (shinfo->gso_size && skb_transport_header_was_set(skb)) { |
15e5a030 | 3736 | u16 gso_segs = shinfo->gso_segs; |
f5fca219 | 3737 | unsigned int hdr_len; |
1def9238 | 3738 | |
757b8b1d | 3739 | /* mac layer + network layer */ |
f5fca219 | 3740 | hdr_len = skb_transport_offset(skb); |
757b8b1d ED |
3741 | |
3742 | /* + transport layer */ | |
7c68d1a6 ED |
3743 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { |
3744 | const struct tcphdr *th; | |
3745 | struct tcphdr _tcphdr; | |
3746 | ||
f5fca219 | 3747 | th = skb_header_pointer(skb, hdr_len, |
7c68d1a6 ED |
3748 | sizeof(_tcphdr), &_tcphdr); |
3749 | if (likely(th)) | |
3750 | hdr_len += __tcp_hdrlen(th); | |
3751 | } else { | |
3752 | struct udphdr _udphdr; | |
3753 | ||
f5fca219 | 3754 | if (skb_header_pointer(skb, hdr_len, |
7c68d1a6 ED |
3755 | sizeof(_udphdr), &_udphdr)) |
3756 | hdr_len += sizeof(struct udphdr); | |
3757 | } | |
15e5a030 JW |
3758 | |
3759 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3760 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3761 | shinfo->gso_size); | |
3762 | ||
3763 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3764 | } |
3765 | } | |
3766 | ||
70713ddd QX |
3767 | static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, |
3768 | struct sk_buff **to_free, | |
3769 | struct netdev_queue *txq) | |
3770 | { | |
3771 | int rc; | |
3772 | ||
3773 | rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; | |
3774 | if (rc == NET_XMIT_SUCCESS) | |
3775 | trace_qdisc_enqueue(q, txq, skb); | |
3776 | return rc; | |
3777 | } | |
3778 | ||
bbd8a0d3 KK |
3779 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3780 | struct net_device *dev, | |
3781 | struct netdev_queue *txq) | |
3782 | { | |
3783 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3784 | struct sk_buff *to_free = NULL; |
a2da570d | 3785 | bool contended; |
bbd8a0d3 KK |
3786 | int rc; |
3787 | ||
a2da570d | 3788 | qdisc_calculate_pkt_len(skb, q); |
6b3ba914 | 3789 | |
b6a3c606 VN |
3790 | tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP); |
3791 | ||
6b3ba914 | 3792 | if (q->flags & TCQ_F_NOLOCK) { |
c4fef01b YL |
3793 | if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && |
3794 | qdisc_run_begin(q)) { | |
3795 | /* Retest nolock_qdisc_is_empty() within the protection | |
3796 | * of q->seqlock to protect from racing with requeuing. | |
3797 | */ | |
3798 | if (unlikely(!nolock_qdisc_is_empty(q))) { | |
70713ddd | 3799 | rc = dev_qdisc_enqueue(skb, q, &to_free, txq); |
c4fef01b YL |
3800 | __qdisc_run(q); |
3801 | qdisc_run_end(q); | |
3802 | ||
3803 | goto no_lock_out; | |
3804 | } | |
3805 | ||
3806 | qdisc_bstats_cpu_update(q, skb); | |
3807 | if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && | |
3808 | !nolock_qdisc_is_empty(q)) | |
3809 | __qdisc_run(q); | |
3810 | ||
3811 | qdisc_run_end(q); | |
3812 | return NET_XMIT_SUCCESS; | |
3813 | } | |
3814 | ||
70713ddd | 3815 | rc = dev_qdisc_enqueue(skb, q, &to_free, txq); |
c4fef01b | 3816 | qdisc_run(q); |
6b3ba914 | 3817 | |
c4fef01b | 3818 | no_lock_out: |
6b3ba914 | 3819 | if (unlikely(to_free)) |
7faef054 | 3820 | kfree_skb_list_reason(to_free, |
b6a3c606 | 3821 | tcf_get_drop_reason(to_free)); |
6b3ba914 JF |
3822 | return rc; |
3823 | } | |
3824 | ||
0f022d32 ED |
3825 | if (unlikely(READ_ONCE(q->owner) == smp_processor_id())) { |
3826 | kfree_skb_reason(skb, SKB_DROP_REASON_TC_RECLASSIFY_LOOP); | |
3827 | return NET_XMIT_DROP; | |
3828 | } | |
79640a4c ED |
3829 | /* |
3830 | * Heuristic to force contended enqueues to serialize on a | |
3831 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3832 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3833 | * often and dequeue packets faster. |
64445dda SAS |
3834 | * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit |
3835 | * and then other tasks will only enqueue packets. The packets will be | |
3836 | * sent after the qdisc owner is scheduled again. To prevent this | |
3837 | * scenario the task always serialize on the lock. | |
79640a4c | 3838 | */ |
a9aa5e33 | 3839 | contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); |
79640a4c ED |
3840 | if (unlikely(contended)) |
3841 | spin_lock(&q->busylock); | |
3842 | ||
bbd8a0d3 KK |
3843 | spin_lock(root_lock); |
3844 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3845 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3846 | rc = NET_XMIT_DROP; |
3847 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3848 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3849 | /* |
3850 | * This is a work-conserving queue; there are no old skbs | |
3851 | * waiting to be sent out; and the qdisc is not running - | |
3852 | * xmit the skb directly. | |
3853 | */ | |
bfe0d029 | 3854 | |
bfe0d029 ED |
3855 | qdisc_bstats_update(q, skb); |
3856 | ||
55a93b3e | 3857 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3858 | if (unlikely(contended)) { |
3859 | spin_unlock(&q->busylock); | |
3860 | contended = false; | |
3861 | } | |
bbd8a0d3 | 3862 | __qdisc_run(q); |
6c148184 | 3863 | } |
bbd8a0d3 | 3864 | |
6c148184 | 3865 | qdisc_run_end(q); |
bbd8a0d3 KK |
3866 | rc = NET_XMIT_SUCCESS; |
3867 | } else { | |
0f022d32 | 3868 | WRITE_ONCE(q->owner, smp_processor_id()); |
70713ddd | 3869 | rc = dev_qdisc_enqueue(skb, q, &to_free, txq); |
0f022d32 | 3870 | WRITE_ONCE(q->owner, -1); |
79640a4c ED |
3871 | if (qdisc_run_begin(q)) { |
3872 | if (unlikely(contended)) { | |
3873 | spin_unlock(&q->busylock); | |
3874 | contended = false; | |
3875 | } | |
3876 | __qdisc_run(q); | |
6c148184 | 3877 | qdisc_run_end(q); |
79640a4c | 3878 | } |
bbd8a0d3 KK |
3879 | } |
3880 | spin_unlock(root_lock); | |
520ac30f | 3881 | if (unlikely(to_free)) |
b6a3c606 VN |
3882 | kfree_skb_list_reason(to_free, |
3883 | tcf_get_drop_reason(to_free)); | |
79640a4c ED |
3884 | if (unlikely(contended)) |
3885 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3886 | return rc; |
3887 | } | |
3888 | ||
86f8515f | 3889 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3890 | static void skb_update_prio(struct sk_buff *skb) |
3891 | { | |
4dcb31d4 ED |
3892 | const struct netprio_map *map; |
3893 | const struct sock *sk; | |
3894 | unsigned int prioidx; | |
5bc1421e | 3895 | |
4dcb31d4 ED |
3896 | if (skb->priority) |
3897 | return; | |
3898 | map = rcu_dereference_bh(skb->dev->priomap); | |
3899 | if (!map) | |
3900 | return; | |
3901 | sk = skb_to_full_sk(skb); | |
3902 | if (!sk) | |
3903 | return; | |
91c68ce2 | 3904 | |
4dcb31d4 ED |
3905 | prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); |
3906 | ||
3907 | if (prioidx < map->priomap_len) | |
3908 | skb->priority = map->priomap[prioidx]; | |
5bc1421e NH |
3909 | } |
3910 | #else | |
3911 | #define skb_update_prio(skb) | |
3912 | #endif | |
3913 | ||
95603e22 MM |
3914 | /** |
3915 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3916 | * @net: network namespace this loopback is happening in |
3917 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3918 | * @skb: buffer to transmit |
3919 | */ | |
0c4b51f0 | 3920 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3921 | { |
3922 | skb_reset_mac_header(skb); | |
3923 | __skb_pull(skb, skb_network_offset(skb)); | |
3924 | skb->pkt_type = PACKET_LOOPBACK; | |
9122a70a CS |
3925 | if (skb->ip_summed == CHECKSUM_NONE) |
3926 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
76458fae | 3927 | DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); |
95603e22 | 3928 | skb_dst_force(skb); |
ad0a043f | 3929 | netif_rx(skb); |
95603e22 MM |
3930 | return 0; |
3931 | } | |
3932 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3933 | ||
1f211a1b | 3934 | #ifdef CONFIG_NET_EGRESS |
e420bed0 DB |
3935 | static struct netdev_queue * |
3936 | netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) | |
3937 | { | |
3938 | int qm = skb_get_queue_mapping(skb); | |
3939 | ||
3940 | return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); | |
3941 | } | |
3942 | ||
3943 | static bool netdev_xmit_txqueue_skipped(void) | |
1f211a1b | 3944 | { |
e420bed0 DB |
3945 | return __this_cpu_read(softnet_data.xmit.skip_txqueue); |
3946 | } | |
3947 | ||
3948 | void netdev_xmit_skip_txqueue(bool skip) | |
3949 | { | |
3950 | __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); | |
3951 | } | |
3952 | EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); | |
3953 | #endif /* CONFIG_NET_EGRESS */ | |
3954 | ||
3955 | #ifdef CONFIG_NET_XGRESS | |
54a59aed DB |
3956 | static int tc_run(struct tcx_entry *entry, struct sk_buff *skb, |
3957 | enum skb_drop_reason *drop_reason) | |
e420bed0 DB |
3958 | { |
3959 | int ret = TC_ACT_UNSPEC; | |
42df6e1d | 3960 | #ifdef CONFIG_NET_CLS_ACT |
e420bed0 DB |
3961 | struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); |
3962 | struct tcf_result res; | |
1f211a1b | 3963 | |
46209401 | 3964 | if (!miniq) |
e420bed0 | 3965 | return ret; |
1f211a1b | 3966 | |
047f340b AST |
3967 | if (static_branch_unlikely(&tcf_bypass_check_needed_key)) { |
3968 | if (tcf_block_bypass_sw(miniq->block)) | |
3969 | return ret; | |
3970 | } | |
3971 | ||
ec624fe7 PB |
3972 | tc_skb_cb(skb)->mru = 0; |
3973 | tc_skb_cb(skb)->post_ct = false; | |
fb278072 | 3974 | tcf_set_drop_reason(skb, *drop_reason); |
1f211a1b | 3975 | |
e420bed0 DB |
3976 | mini_qdisc_bstats_cpu_update(miniq, skb); |
3977 | ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); | |
3978 | /* Only tcf related quirks below. */ | |
3979 | switch (ret) { | |
3980 | case TC_ACT_SHOT: | |
fb278072 | 3981 | *drop_reason = tcf_get_drop_reason(skb); |
e420bed0 DB |
3982 | mini_qdisc_qstats_cpu_drop(miniq); |
3983 | break; | |
1f211a1b DB |
3984 | case TC_ACT_OK: |
3985 | case TC_ACT_RECLASSIFY: | |
e420bed0 | 3986 | skb->tc_index = TC_H_MIN(res.classid); |
1f211a1b | 3987 | break; |
e420bed0 DB |
3988 | } |
3989 | #endif /* CONFIG_NET_CLS_ACT */ | |
3990 | return ret; | |
3991 | } | |
3992 | ||
3993 | static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); | |
3994 | ||
3995 | void tcx_inc(void) | |
3996 | { | |
3997 | static_branch_inc(&tcx_needed_key); | |
3998 | } | |
3999 | ||
4000 | void tcx_dec(void) | |
4001 | { | |
4002 | static_branch_dec(&tcx_needed_key); | |
4003 | } | |
4004 | ||
4005 | static __always_inline enum tcx_action_base | |
4006 | tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, | |
4007 | const bool needs_mac) | |
4008 | { | |
4009 | const struct bpf_mprog_fp *fp; | |
4010 | const struct bpf_prog *prog; | |
4011 | int ret = TCX_NEXT; | |
4012 | ||
4013 | if (needs_mac) | |
4014 | __skb_push(skb, skb->mac_len); | |
4015 | bpf_mprog_foreach_prog(entry, fp, prog) { | |
4016 | bpf_compute_data_pointers(skb); | |
4017 | ret = bpf_prog_run(prog, skb); | |
4018 | if (ret != TCX_NEXT) | |
4019 | break; | |
4020 | } | |
4021 | if (needs_mac) | |
4022 | __skb_pull(skb, skb->mac_len); | |
4023 | return tcx_action_code(skb, ret); | |
4024 | } | |
4025 | ||
4026 | static __always_inline struct sk_buff * | |
4027 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
4028 | struct net_device *orig_dev, bool *another) | |
4029 | { | |
4030 | struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); | |
54a59aed | 4031 | enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS; |
e420bed0 DB |
4032 | int sch_ret; |
4033 | ||
4034 | if (!entry) | |
4035 | return skb; | |
4036 | if (*pt_prev) { | |
4037 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4038 | *pt_prev = NULL; | |
4039 | } | |
4040 | ||
4041 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
4042 | tcx_set_ingress(skb, true); | |
4043 | ||
4044 | if (static_branch_unlikely(&tcx_needed_key)) { | |
4045 | sch_ret = tcx_run(entry, skb, true); | |
4046 | if (sch_ret != TC_ACT_UNSPEC) | |
4047 | goto ingress_verdict; | |
4048 | } | |
54a59aed | 4049 | sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); |
e420bed0 DB |
4050 | ingress_verdict: |
4051 | switch (sch_ret) { | |
4052 | case TC_ACT_REDIRECT: | |
4053 | /* skb_mac_header check was done by BPF, so we can safely | |
4054 | * push the L2 header back before redirecting to another | |
4055 | * netdev. | |
4056 | */ | |
4057 | __skb_push(skb, skb->mac_len); | |
4058 | if (skb_do_redirect(skb) == -EAGAIN) { | |
4059 | __skb_pull(skb, skb->mac_len); | |
4060 | *another = true; | |
4061 | break; | |
4062 | } | |
4063 | *ret = NET_RX_SUCCESS; | |
4064 | return NULL; | |
1f211a1b | 4065 | case TC_ACT_SHOT: |
54a59aed | 4066 | kfree_skb_reason(skb, drop_reason); |
e420bed0 | 4067 | *ret = NET_RX_DROP; |
7e2c3aea | 4068 | return NULL; |
e420bed0 | 4069 | /* used by tc_run */ |
1f211a1b DB |
4070 | case TC_ACT_STOLEN: |
4071 | case TC_ACT_QUEUED: | |
e25ea21f | 4072 | case TC_ACT_TRAP: |
7e2c3aea | 4073 | consume_skb(skb); |
e420bed0 DB |
4074 | fallthrough; |
4075 | case TC_ACT_CONSUMED: | |
4076 | *ret = NET_RX_SUCCESS; | |
1f211a1b | 4077 | return NULL; |
e420bed0 DB |
4078 | } |
4079 | ||
4080 | return skb; | |
4081 | } | |
4082 | ||
4083 | static __always_inline struct sk_buff * | |
4084 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
4085 | { | |
4086 | struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); | |
54a59aed | 4087 | enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS; |
e420bed0 DB |
4088 | int sch_ret; |
4089 | ||
4090 | if (!entry) | |
4091 | return skb; | |
4092 | ||
4093 | /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was | |
4094 | * already set by the caller. | |
4095 | */ | |
4096 | if (static_branch_unlikely(&tcx_needed_key)) { | |
4097 | sch_ret = tcx_run(entry, skb, false); | |
4098 | if (sch_ret != TC_ACT_UNSPEC) | |
4099 | goto egress_verdict; | |
4100 | } | |
54a59aed | 4101 | sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); |
e420bed0 DB |
4102 | egress_verdict: |
4103 | switch (sch_ret) { | |
1f211a1b DB |
4104 | case TC_ACT_REDIRECT: |
4105 | /* No need to push/pop skb's mac_header here on egress! */ | |
4106 | skb_do_redirect(skb); | |
4107 | *ret = NET_XMIT_SUCCESS; | |
4108 | return NULL; | |
e420bed0 | 4109 | case TC_ACT_SHOT: |
54a59aed | 4110 | kfree_skb_reason(skb, drop_reason); |
e420bed0 DB |
4111 | *ret = NET_XMIT_DROP; |
4112 | return NULL; | |
4113 | /* used by tc_run */ | |
4114 | case TC_ACT_STOLEN: | |
4115 | case TC_ACT_QUEUED: | |
4116 | case TC_ACT_TRAP: | |
28d18b67 | 4117 | consume_skb(skb); |
3a1e2f43 DB |
4118 | fallthrough; |
4119 | case TC_ACT_CONSUMED: | |
e420bed0 DB |
4120 | *ret = NET_XMIT_SUCCESS; |
4121 | return NULL; | |
1f211a1b | 4122 | } |
357b6cc5 | 4123 | |
1f211a1b DB |
4124 | return skb; |
4125 | } | |
e420bed0 DB |
4126 | #else |
4127 | static __always_inline struct sk_buff * | |
4128 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
4129 | struct net_device *orig_dev, bool *another) | |
2f1e85b1 | 4130 | { |
e420bed0 | 4131 | return skb; |
2f1e85b1 TZ |
4132 | } |
4133 | ||
e420bed0 DB |
4134 | static __always_inline struct sk_buff * |
4135 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
2f1e85b1 | 4136 | { |
e420bed0 | 4137 | return skb; |
2f1e85b1 | 4138 | } |
e420bed0 | 4139 | #endif /* CONFIG_NET_XGRESS */ |
1f211a1b | 4140 | |
fc9bab24 AN |
4141 | #ifdef CONFIG_XPS |
4142 | static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, | |
4143 | struct xps_dev_maps *dev_maps, unsigned int tci) | |
4144 | { | |
255c04a8 | 4145 | int tc = netdev_get_prio_tc_map(dev, skb->priority); |
fc9bab24 AN |
4146 | struct xps_map *map; |
4147 | int queue_index = -1; | |
4148 | ||
5478fcd0 | 4149 | if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) |
255c04a8 AT |
4150 | return queue_index; |
4151 | ||
4152 | tci *= dev_maps->num_tc; | |
4153 | tci += tc; | |
fc9bab24 AN |
4154 | |
4155 | map = rcu_dereference(dev_maps->attr_map[tci]); | |
4156 | if (map) { | |
4157 | if (map->len == 1) | |
4158 | queue_index = map->queues[0]; | |
4159 | else | |
4160 | queue_index = map->queues[reciprocal_scale( | |
4161 | skb_get_hash(skb), map->len)]; | |
4162 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
4163 | queue_index = -1; | |
4164 | } | |
4165 | return queue_index; | |
4166 | } | |
4167 | #endif | |
4168 | ||
eadec877 AD |
4169 | static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, |
4170 | struct sk_buff *skb) | |
638b2a69 JP |
4171 | { |
4172 | #ifdef CONFIG_XPS | |
4173 | struct xps_dev_maps *dev_maps; | |
fc9bab24 | 4174 | struct sock *sk = skb->sk; |
638b2a69 JP |
4175 | int queue_index = -1; |
4176 | ||
04157469 AN |
4177 | if (!static_key_false(&xps_needed)) |
4178 | return -1; | |
4179 | ||
638b2a69 | 4180 | rcu_read_lock(); |
fc9bab24 AN |
4181 | if (!static_key_false(&xps_rxqs_needed)) |
4182 | goto get_cpus_map; | |
4183 | ||
044ab86d | 4184 | dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); |
638b2a69 | 4185 | if (dev_maps) { |
fc9bab24 | 4186 | int tci = sk_rx_queue_get(sk); |
184c449f | 4187 | |
5478fcd0 | 4188 | if (tci >= 0) |
fc9bab24 AN |
4189 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, |
4190 | tci); | |
4191 | } | |
184c449f | 4192 | |
fc9bab24 AN |
4193 | get_cpus_map: |
4194 | if (queue_index < 0) { | |
044ab86d | 4195 | dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); |
fc9bab24 AN |
4196 | if (dev_maps) { |
4197 | unsigned int tci = skb->sender_cpu - 1; | |
4198 | ||
4199 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
4200 | tci); | |
638b2a69 JP |
4201 | } |
4202 | } | |
4203 | rcu_read_unlock(); | |
4204 | ||
4205 | return queue_index; | |
4206 | #else | |
4207 | return -1; | |
4208 | #endif | |
4209 | } | |
4210 | ||
a4ea8a3d | 4211 | u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, |
a350ecce | 4212 | struct net_device *sb_dev) |
a4ea8a3d AD |
4213 | { |
4214 | return 0; | |
4215 | } | |
4216 | EXPORT_SYMBOL(dev_pick_tx_zero); | |
4217 | ||
4218 | u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, | |
a350ecce | 4219 | struct net_device *sb_dev) |
a4ea8a3d AD |
4220 | { |
4221 | return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; | |
4222 | } | |
4223 | EXPORT_SYMBOL(dev_pick_tx_cpu_id); | |
4224 | ||
b71b5837 PA |
4225 | u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, |
4226 | struct net_device *sb_dev) | |
638b2a69 JP |
4227 | { |
4228 | struct sock *sk = skb->sk; | |
4229 | int queue_index = sk_tx_queue_get(sk); | |
4230 | ||
eadec877 AD |
4231 | sb_dev = sb_dev ? : dev; |
4232 | ||
638b2a69 JP |
4233 | if (queue_index < 0 || skb->ooo_okay || |
4234 | queue_index >= dev->real_num_tx_queues) { | |
eadec877 | 4235 | int new_index = get_xps_queue(dev, sb_dev, skb); |
f4563a75 | 4236 | |
638b2a69 | 4237 | if (new_index < 0) |
eadec877 | 4238 | new_index = skb_tx_hash(dev, sb_dev, skb); |
638b2a69 JP |
4239 | |
4240 | if (queue_index != new_index && sk && | |
004a5d01 | 4241 | sk_fullsock(sk) && |
638b2a69 JP |
4242 | rcu_access_pointer(sk->sk_dst_cache)) |
4243 | sk_tx_queue_set(sk, new_index); | |
4244 | ||
4245 | queue_index = new_index; | |
4246 | } | |
4247 | ||
4248 | return queue_index; | |
4249 | } | |
b71b5837 | 4250 | EXPORT_SYMBOL(netdev_pick_tx); |
638b2a69 | 4251 | |
4bd97d51 PA |
4252 | struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, |
4253 | struct sk_buff *skb, | |
4254 | struct net_device *sb_dev) | |
638b2a69 JP |
4255 | { |
4256 | int queue_index = 0; | |
4257 | ||
4258 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
4259 | u32 sender_cpu = skb->sender_cpu - 1; |
4260 | ||
4261 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
4262 | skb->sender_cpu = raw_smp_processor_id() + 1; |
4263 | #endif | |
4264 | ||
4265 | if (dev->real_num_tx_queues != 1) { | |
4266 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 4267 | |
638b2a69 | 4268 | if (ops->ndo_select_queue) |
a350ecce | 4269 | queue_index = ops->ndo_select_queue(dev, skb, sb_dev); |
638b2a69 | 4270 | else |
4bd97d51 | 4271 | queue_index = netdev_pick_tx(dev, skb, sb_dev); |
638b2a69 | 4272 | |
d584527c | 4273 | queue_index = netdev_cap_txqueue(dev, queue_index); |
638b2a69 JP |
4274 | } |
4275 | ||
4276 | skb_set_queue_mapping(skb, queue_index); | |
4277 | return netdev_get_tx_queue(dev, queue_index); | |
4278 | } | |
4279 | ||
d29f749e | 4280 | /** |
be76955d JK |
4281 | * __dev_queue_xmit() - transmit a buffer |
4282 | * @skb: buffer to transmit | |
4283 | * @sb_dev: suboordinate device used for L2 forwarding offload | |
d29f749e | 4284 | * |
be76955d JK |
4285 | * Queue a buffer for transmission to a network device. The caller must |
4286 | * have set the device and priority and built the buffer before calling | |
4287 | * this function. The function can be called from an interrupt. | |
d29f749e | 4288 | * |
be76955d JK |
4289 | * When calling this method, interrupts MUST be enabled. This is because |
4290 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
d29f749e | 4291 | * |
be76955d JK |
4292 | * Regardless of the return value, the skb is consumed, so it is currently |
4293 | * difficult to retry a send to this method. (You can bump the ref count | |
4294 | * before sending to hold a reference for retry if you are careful.) | |
d29f749e | 4295 | * |
be76955d JK |
4296 | * Return: |
4297 | * * 0 - buffer successfully transmitted | |
4298 | * * positive qdisc return code - NET_XMIT_DROP etc. | |
4299 | * * negative errno - other errors | |
d29f749e | 4300 | */ |
c526fd8f | 4301 | int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) |
1da177e4 LT |
4302 | { |
4303 | struct net_device *dev = skb->dev; | |
2f1e85b1 | 4304 | struct netdev_queue *txq = NULL; |
1da177e4 LT |
4305 | struct Qdisc *q; |
4306 | int rc = -ENOMEM; | |
f53c7239 | 4307 | bool again = false; |
1da177e4 | 4308 | |
6d1ccff6 | 4309 | skb_reset_mac_header(skb); |
fd189422 | 4310 | skb_assert_len(skb); |
6d1ccff6 | 4311 | |
e7fd2885 | 4312 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
e7ed11ee | 4313 | __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); |
e7fd2885 | 4314 | |
4ec93edb YH |
4315 | /* Disable soft irqs for various locks below. Also |
4316 | * stops preemption for RCU. | |
1da177e4 | 4317 | */ |
4ec93edb | 4318 | rcu_read_lock_bh(); |
1da177e4 | 4319 | |
5bc1421e NH |
4320 | skb_update_prio(skb); |
4321 | ||
1f211a1b | 4322 | qdisc_pkt_len_init(skb); |
e420bed0 | 4323 | tcx_set_ingress(skb, false); |
42df6e1d | 4324 | #ifdef CONFIG_NET_EGRESS |
aabf6772 | 4325 | if (static_branch_unlikely(&egress_needed_key)) { |
42df6e1d LW |
4326 | if (nf_hook_egress_active()) { |
4327 | skb = nf_hook_egress(skb, &rc, dev); | |
4328 | if (!skb) | |
4329 | goto out; | |
4330 | } | |
2f1e85b1 TZ |
4331 | |
4332 | netdev_xmit_skip_txqueue(false); | |
4333 | ||
42df6e1d | 4334 | nf_skip_egress(skb, true); |
1f211a1b DB |
4335 | skb = sch_handle_egress(skb, &rc, dev); |
4336 | if (!skb) | |
4337 | goto out; | |
42df6e1d | 4338 | nf_skip_egress(skb, false); |
2f1e85b1 TZ |
4339 | |
4340 | if (netdev_xmit_txqueue_skipped()) | |
4341 | txq = netdev_tx_queue_mapping(dev, skb); | |
1f211a1b | 4342 | } |
1f211a1b | 4343 | #endif |
02875878 ED |
4344 | /* If device/qdisc don't need skb->dst, release it right now while |
4345 | * its hot in this cpu cache. | |
4346 | */ | |
4347 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
4348 | skb_dst_drop(skb); | |
4349 | else | |
4350 | skb_dst_force(skb); | |
4351 | ||
2f1e85b1 TZ |
4352 | if (!txq) |
4353 | txq = netdev_core_pick_tx(dev, skb, sb_dev); | |
4354 | ||
a898def2 | 4355 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 4356 | |
cf66ba58 | 4357 | trace_net_dev_queue(skb); |
1da177e4 | 4358 | if (q->enqueue) { |
bbd8a0d3 | 4359 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 4360 | goto out; |
1da177e4 LT |
4361 | } |
4362 | ||
4363 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 4364 | * loopback, all the sorts of tunnels... |
1da177e4 | 4365 | |
eb13da1a | 4366 | * Really, it is unlikely that netif_tx_lock protection is necessary |
4367 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
4368 | * counters.) | |
4369 | * However, it is possible, that they rely on protection | |
4370 | * made by us here. | |
1da177e4 | 4371 | |
eb13da1a | 4372 | * Check this and shot the lock. It is not prone from deadlocks. |
4373 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
4374 | */ |
4375 | if (dev->flags & IFF_UP) { | |
4376 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
4377 | ||
7a10d8c8 ED |
4378 | /* Other cpus might concurrently change txq->xmit_lock_owner |
4379 | * to -1 or to their cpu id, but not to our id. | |
4380 | */ | |
4381 | if (READ_ONCE(txq->xmit_lock_owner) != cpu) { | |
97cdcf37 | 4382 | if (dev_xmit_recursion()) |
745e20f1 ED |
4383 | goto recursion_alert; |
4384 | ||
f53c7239 | 4385 | skb = validate_xmit_skb(skb, dev, &again); |
1f59533f | 4386 | if (!skb) |
d21fd63e | 4387 | goto out; |
1f59533f | 4388 | |
c773e847 | 4389 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 4390 | |
73466498 | 4391 | if (!netif_xmit_stopped(txq)) { |
97cdcf37 | 4392 | dev_xmit_recursion_inc(); |
ce93718f | 4393 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
97cdcf37 | 4394 | dev_xmit_recursion_dec(); |
572a9d7b | 4395 | if (dev_xmit_complete(rc)) { |
c773e847 | 4396 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
4397 | goto out; |
4398 | } | |
4399 | } | |
c773e847 | 4400 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
4401 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
4402 | dev->name); | |
1da177e4 LT |
4403 | } else { |
4404 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
4405 | * unfortunately |
4406 | */ | |
4407 | recursion_alert: | |
e87cc472 JP |
4408 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
4409 | dev->name); | |
1da177e4 LT |
4410 | } |
4411 | } | |
4412 | ||
4413 | rc = -ENETDOWN; | |
d4828d85 | 4414 | rcu_read_unlock_bh(); |
1da177e4 | 4415 | |
625788b5 | 4416 | dev_core_stats_tx_dropped_inc(dev); |
1f59533f | 4417 | kfree_skb_list(skb); |
1da177e4 LT |
4418 | return rc; |
4419 | out: | |
d4828d85 | 4420 | rcu_read_unlock_bh(); |
1da177e4 LT |
4421 | return rc; |
4422 | } | |
c526fd8f | 4423 | EXPORT_SYMBOL(__dev_queue_xmit); |
f663dd9a | 4424 | |
36ccdf85 | 4425 | int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) |
865b03f2 MK |
4426 | { |
4427 | struct net_device *dev = skb->dev; | |
4428 | struct sk_buff *orig_skb = skb; | |
4429 | struct netdev_queue *txq; | |
4430 | int ret = NETDEV_TX_BUSY; | |
4431 | bool again = false; | |
4432 | ||
4433 | if (unlikely(!netif_running(dev) || | |
4434 | !netif_carrier_ok(dev))) | |
4435 | goto drop; | |
4436 | ||
4437 | skb = validate_xmit_skb_list(skb, dev, &again); | |
4438 | if (skb != orig_skb) | |
4439 | goto drop; | |
4440 | ||
4441 | skb_set_queue_mapping(skb, queue_id); | |
4442 | txq = skb_get_tx_queue(dev, skb); | |
4443 | ||
4444 | local_bh_disable(); | |
4445 | ||
0ad6f6e7 | 4446 | dev_xmit_recursion_inc(); |
865b03f2 MK |
4447 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
4448 | if (!netif_xmit_frozen_or_drv_stopped(txq)) | |
4449 | ret = netdev_start_xmit(skb, dev, txq, false); | |
4450 | HARD_TX_UNLOCK(dev, txq); | |
0ad6f6e7 | 4451 | dev_xmit_recursion_dec(); |
865b03f2 MK |
4452 | |
4453 | local_bh_enable(); | |
865b03f2 MK |
4454 | return ret; |
4455 | drop: | |
625788b5 | 4456 | dev_core_stats_tx_dropped_inc(dev); |
865b03f2 MK |
4457 | kfree_skb_list(skb); |
4458 | return NET_XMIT_DROP; | |
4459 | } | |
36ccdf85 | 4460 | EXPORT_SYMBOL(__dev_direct_xmit); |
1da177e4 | 4461 | |
eb13da1a | 4462 | /************************************************************************* |
4463 | * Receiver routines | |
4464 | *************************************************************************/ | |
dad6b977 | 4465 | static DEFINE_PER_CPU(struct task_struct *, backlog_napi); |
1da177e4 | 4466 | |
3d48b53f MT |
4467 | int weight_p __read_mostly = 64; /* old backlog weight */ |
4468 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
4469 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
1da177e4 | 4470 | |
eecfd7c4 ED |
4471 | /* Called with irq disabled */ |
4472 | static inline void ____napi_schedule(struct softnet_data *sd, | |
4473 | struct napi_struct *napi) | |
4474 | { | |
29863d41 WW |
4475 | struct task_struct *thread; |
4476 | ||
fbd9a2ce SAS |
4477 | lockdep_assert_irqs_disabled(); |
4478 | ||
29863d41 WW |
4479 | if (test_bit(NAPI_STATE_THREADED, &napi->state)) { |
4480 | /* Paired with smp_mb__before_atomic() in | |
5fdd2f0e WW |
4481 | * napi_enable()/dev_set_threaded(). |
4482 | * Use READ_ONCE() to guarantee a complete | |
4483 | * read on napi->thread. Only call | |
29863d41 WW |
4484 | * wake_up_process() when it's not NULL. |
4485 | */ | |
4486 | thread = READ_ONCE(napi->thread); | |
4487 | if (thread) { | |
dad6b977 SAS |
4488 | if (use_backlog_threads() && thread == raw_cpu_read(backlog_napi)) |
4489 | goto use_local_napi; | |
4490 | ||
56364c91 | 4491 | set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); |
29863d41 WW |
4492 | wake_up_process(thread); |
4493 | return; | |
4494 | } | |
4495 | } | |
4496 | ||
dad6b977 | 4497 | use_local_napi: |
eecfd7c4 | 4498 | list_add_tail(&napi->poll_list, &sd->poll_list); |
8c48eea3 | 4499 | WRITE_ONCE(napi->list_owner, smp_processor_id()); |
8b43fd3d ED |
4500 | /* If not called from net_rx_action() |
4501 | * we have to raise NET_RX_SOFTIRQ. | |
4502 | */ | |
4503 | if (!sd->in_net_rx_action) | |
4504 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
eecfd7c4 ED |
4505 | } |
4506 | ||
bfb564e7 KK |
4507 | #ifdef CONFIG_RPS |
4508 | ||
dc05360f | 4509 | struct static_key_false rps_needed __read_mostly; |
3df97ba8 | 4510 | EXPORT_SYMBOL(rps_needed); |
dc05360f | 4511 | struct static_key_false rfs_needed __read_mostly; |
13bfff25 | 4512 | EXPORT_SYMBOL(rfs_needed); |
adc9300e | 4513 | |
c445477d BH |
4514 | static struct rps_dev_flow * |
4515 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4516 | struct rps_dev_flow *rflow, u16 next_cpu) | |
4517 | { | |
a31196b0 | 4518 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
4519 | #ifdef CONFIG_RFS_ACCEL |
4520 | struct netdev_rx_queue *rxqueue; | |
4521 | struct rps_dev_flow_table *flow_table; | |
4522 | struct rps_dev_flow *old_rflow; | |
84b6823c | 4523 | u32 flow_id, head; |
c445477d BH |
4524 | u16 rxq_index; |
4525 | int rc; | |
4526 | ||
4527 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
4528 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
4529 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
4530 | goto out; |
4531 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
4532 | if (rxq_index == skb_get_rx_queue(skb)) | |
4533 | goto out; | |
4534 | ||
4535 | rxqueue = dev->_rx + rxq_index; | |
4536 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4537 | if (!flow_table) | |
4538 | goto out; | |
61b905da | 4539 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
4540 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
4541 | rxq_index, flow_id); | |
4542 | if (rc < 0) | |
4543 | goto out; | |
4544 | old_rflow = rflow; | |
4545 | rflow = &flow_table->flows[flow_id]; | |
f00bf5dc JX |
4546 | WRITE_ONCE(rflow->filter, rc); |
4547 | if (old_rflow->filter == rc) | |
4548 | WRITE_ONCE(old_rflow->filter, RPS_NO_FILTER); | |
c445477d BH |
4549 | out: |
4550 | #endif | |
84b6823c JX |
4551 | head = READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head); |
4552 | rps_input_queue_tail_save(&rflow->last_qtail, head); | |
c445477d BH |
4553 | } |
4554 | ||
f7b60cce | 4555 | WRITE_ONCE(rflow->cpu, next_cpu); |
c445477d BH |
4556 | return rflow; |
4557 | } | |
4558 | ||
bfb564e7 KK |
4559 | /* |
4560 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
4561 | * CPU from the RPS map of the receiving queue for a given skb. | |
4562 | * rcu_read_lock must be held on entry. | |
4563 | */ | |
4564 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4565 | struct rps_dev_flow **rflowp) | |
4566 | { | |
567e4b79 ED |
4567 | const struct rps_sock_flow_table *sock_flow_table; |
4568 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 4569 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 4570 | struct rps_map *map; |
bfb564e7 | 4571 | int cpu = -1; |
567e4b79 | 4572 | u32 tcpu; |
61b905da | 4573 | u32 hash; |
bfb564e7 KK |
4574 | |
4575 | if (skb_rx_queue_recorded(skb)) { | |
4576 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 4577 | |
62fe0b40 BH |
4578 | if (unlikely(index >= dev->real_num_rx_queues)) { |
4579 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4580 | "%s received packet on queue %u, but number " | |
4581 | "of RX queues is %u\n", | |
4582 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
4583 | goto done; |
4584 | } | |
567e4b79 ED |
4585 | rxqueue += index; |
4586 | } | |
bfb564e7 | 4587 | |
567e4b79 ED |
4588 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
4589 | ||
4590 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 4591 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 4592 | if (!flow_table && !map) |
bfb564e7 KK |
4593 | goto done; |
4594 | ||
2d47b459 | 4595 | skb_reset_network_header(skb); |
61b905da TH |
4596 | hash = skb_get_hash(skb); |
4597 | if (!hash) | |
bfb564e7 KK |
4598 | goto done; |
4599 | ||
ce7f49ab | 4600 | sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table); |
fec5e652 | 4601 | if (flow_table && sock_flow_table) { |
fec5e652 | 4602 | struct rps_dev_flow *rflow; |
567e4b79 ED |
4603 | u32 next_cpu; |
4604 | u32 ident; | |
4605 | ||
5c3b74a9 ED |
4606 | /* First check into global flow table if there is a match. |
4607 | * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). | |
4608 | */ | |
4609 | ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); | |
ce7f49ab | 4610 | if ((ident ^ hash) & ~net_hotdata.rps_cpu_mask) |
567e4b79 | 4611 | goto try_rps; |
fec5e652 | 4612 | |
ce7f49ab | 4613 | next_cpu = ident & net_hotdata.rps_cpu_mask; |
567e4b79 ED |
4614 | |
4615 | /* OK, now we know there is a match, | |
4616 | * we can look at the local (per receive queue) flow table | |
4617 | */ | |
61b905da | 4618 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
4619 | tcpu = rflow->cpu; |
4620 | ||
fec5e652 TH |
4621 | /* |
4622 | * If the desired CPU (where last recvmsg was done) is | |
4623 | * different from current CPU (one in the rx-queue flow | |
4624 | * table entry), switch if one of the following holds: | |
a31196b0 | 4625 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
4626 | * - Current CPU is offline. |
4627 | * - The current CPU's queue tail has advanced beyond the | |
4628 | * last packet that was enqueued using this table entry. | |
4629 | * This guarantees that all previous packets for the flow | |
4630 | * have been dequeued, thus preserving in order delivery. | |
4631 | */ | |
4632 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 4633 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
c62fdf5b | 4634 | ((int)(READ_ONCE(per_cpu(softnet_data, tcpu).input_queue_head) - |
84b6823c | 4635 | rflow->last_qtail)) >= 0)) { |
baefa31d | 4636 | tcpu = next_cpu; |
c445477d | 4637 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 4638 | } |
c445477d | 4639 | |
a31196b0 | 4640 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
4641 | *rflowp = rflow; |
4642 | cpu = tcpu; | |
4643 | goto done; | |
4644 | } | |
4645 | } | |
4646 | ||
567e4b79 ED |
4647 | try_rps: |
4648 | ||
0a9627f2 | 4649 | if (map) { |
8fc54f68 | 4650 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
4651 | if (cpu_online(tcpu)) { |
4652 | cpu = tcpu; | |
4653 | goto done; | |
4654 | } | |
4655 | } | |
4656 | ||
4657 | done: | |
0a9627f2 TH |
4658 | return cpu; |
4659 | } | |
4660 | ||
c445477d BH |
4661 | #ifdef CONFIG_RFS_ACCEL |
4662 | ||
4663 | /** | |
4664 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
4665 | * @dev: Device on which the filter was set | |
4666 | * @rxq_index: RX queue index | |
4667 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
4668 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
4669 | * | |
4670 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
4671 | * this function for each installed filter and remove the filters for | |
4672 | * which it returns %true. | |
4673 | */ | |
4674 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
4675 | u32 flow_id, u16 filter_id) | |
4676 | { | |
4677 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
4678 | struct rps_dev_flow_table *flow_table; | |
4679 | struct rps_dev_flow *rflow; | |
4680 | bool expire = true; | |
a31196b0 | 4681 | unsigned int cpu; |
c445477d BH |
4682 | |
4683 | rcu_read_lock(); | |
4684 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4685 | if (flow_table && flow_id <= flow_table->mask) { | |
4686 | rflow = &flow_table->flows[flow_id]; | |
6aa7de05 | 4687 | cpu = READ_ONCE(rflow->cpu); |
f00bf5dc | 4688 | if (READ_ONCE(rflow->filter) == filter_id && cpu < nr_cpu_ids && |
c62fdf5b | 4689 | ((int)(READ_ONCE(per_cpu(softnet_data, cpu).input_queue_head) - |
36b83ffc | 4690 | READ_ONCE(rflow->last_qtail)) < |
c445477d BH |
4691 | (int)(10 * flow_table->mask))) |
4692 | expire = false; | |
4693 | } | |
4694 | rcu_read_unlock(); | |
4695 | return expire; | |
4696 | } | |
4697 | EXPORT_SYMBOL(rps_may_expire_flow); | |
4698 | ||
4699 | #endif /* CONFIG_RFS_ACCEL */ | |
4700 | ||
0a9627f2 | 4701 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 4702 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 4703 | { |
e36fa2f7 ED |
4704 | struct softnet_data *sd = data; |
4705 | ||
eecfd7c4 | 4706 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 4707 | sd->received_rps++; |
0a9627f2 | 4708 | } |
e36fa2f7 | 4709 | |
fec5e652 | 4710 | #endif /* CONFIG_RPS */ |
0a9627f2 | 4711 | |
68822bdf | 4712 | /* Called from hardirq (IPI) context */ |
97e719a8 | 4713 | static void trigger_rx_softirq(void *data) |
68822bdf | 4714 | { |
97e719a8 ED |
4715 | struct softnet_data *sd = data; |
4716 | ||
68822bdf | 4717 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
97e719a8 | 4718 | smp_store_release(&sd->defer_ipi_scheduled, 0); |
68822bdf ED |
4719 | } |
4720 | ||
e36fa2f7 | 4721 | /* |
8fcb76b9 ED |
4722 | * After we queued a packet into sd->input_pkt_queue, |
4723 | * we need to make sure this queue is serviced soon. | |
4724 | * | |
4725 | * - If this is another cpu queue, link it to our rps_ipi_list, | |
4726 | * and make sure we will process rps_ipi_list from net_rx_action(). | |
8fcb76b9 ED |
4727 | * |
4728 | * - If this is our own queue, NAPI schedule our backlog. | |
4729 | * Note that this also raises NET_RX_SOFTIRQ. | |
e36fa2f7 | 4730 | */ |
8fcb76b9 | 4731 | static void napi_schedule_rps(struct softnet_data *sd) |
e36fa2f7 | 4732 | { |
903ceff7 | 4733 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 | 4734 | |
e722db8d | 4735 | #ifdef CONFIG_RPS |
e36fa2f7 | 4736 | if (sd != mysd) { |
dad6b977 SAS |
4737 | if (use_backlog_threads()) { |
4738 | __napi_schedule_irqoff(&sd->backlog); | |
4739 | return; | |
4740 | } | |
4741 | ||
e36fa2f7 ED |
4742 | sd->rps_ipi_next = mysd->rps_ipi_list; |
4743 | mysd->rps_ipi_list = sd; | |
4744 | ||
87eff2ec | 4745 | /* If not called from net_rx_action() or napi_threaded_poll() |
821eba96 ED |
4746 | * we have to raise NET_RX_SOFTIRQ. |
4747 | */ | |
87eff2ec | 4748 | if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) |
821eba96 | 4749 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
8fcb76b9 | 4750 | return; |
e36fa2f7 ED |
4751 | } |
4752 | #endif /* CONFIG_RPS */ | |
e722db8d | 4753 | __napi_schedule_irqoff(&mysd->backlog); |
e36fa2f7 ED |
4754 | } |
4755 | ||
80d2eefc SAS |
4756 | void kick_defer_list_purge(struct softnet_data *sd, unsigned int cpu) |
4757 | { | |
4758 | unsigned long flags; | |
4759 | ||
4760 | if (use_backlog_threads()) { | |
765b11f8 | 4761 | backlog_lock_irq_save(sd, &flags); |
80d2eefc SAS |
4762 | |
4763 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) | |
4764 | __napi_schedule_irqoff(&sd->backlog); | |
4765 | ||
765b11f8 | 4766 | backlog_unlock_irq_restore(sd, &flags); |
80d2eefc SAS |
4767 | |
4768 | } else if (!cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) { | |
4769 | smp_call_function_single_async(cpu, &sd->defer_csd); | |
4770 | } | |
4771 | } | |
4772 | ||
99bbc707 WB |
4773 | #ifdef CONFIG_NET_FLOW_LIMIT |
4774 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
4775 | #endif | |
4776 | ||
4777 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
4778 | { | |
4779 | #ifdef CONFIG_NET_FLOW_LIMIT | |
4780 | struct sd_flow_limit *fl; | |
4781 | struct softnet_data *sd; | |
4782 | unsigned int old_flow, new_flow; | |
4783 | ||
edbc666c | 4784 | if (qlen < (READ_ONCE(net_hotdata.max_backlog) >> 1)) |
99bbc707 WB |
4785 | return false; |
4786 | ||
903ceff7 | 4787 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
4788 | |
4789 | rcu_read_lock(); | |
4790 | fl = rcu_dereference(sd->flow_limit); | |
4791 | if (fl) { | |
3958afa1 | 4792 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
4793 | old_flow = fl->history[fl->history_head]; |
4794 | fl->history[fl->history_head] = new_flow; | |
4795 | ||
4796 | fl->history_head++; | |
4797 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
4798 | ||
4799 | if (likely(fl->buckets[old_flow])) | |
4800 | fl->buckets[old_flow]--; | |
4801 | ||
4802 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
4803 | fl->count++; | |
4804 | rcu_read_unlock(); | |
4805 | return true; | |
4806 | } | |
4807 | } | |
4808 | rcu_read_unlock(); | |
4809 | #endif | |
4810 | return false; | |
4811 | } | |
4812 | ||
0a9627f2 TH |
4813 | /* |
4814 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
4815 | * queue (may be a remote CPU queue). | |
4816 | */ | |
fec5e652 TH |
4817 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
4818 | unsigned int *qtail) | |
0a9627f2 | 4819 | { |
44f0bd40 | 4820 | enum skb_drop_reason reason; |
e36fa2f7 | 4821 | struct softnet_data *sd; |
0a9627f2 | 4822 | unsigned long flags; |
99bbc707 | 4823 | unsigned int qlen; |
a7ae7b0b | 4824 | int max_backlog; |
36b83ffc | 4825 | u32 tail; |
0a9627f2 | 4826 | |
95e48d86 ED |
4827 | reason = SKB_DROP_REASON_DEV_READY; |
4828 | if (!netif_running(skb->dev)) | |
4829 | goto bad_dev; | |
4830 | ||
a7ae7b0b | 4831 | reason = SKB_DROP_REASON_CPU_BACKLOG; |
e36fa2f7 | 4832 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 | 4833 | |
a7ae7b0b ED |
4834 | qlen = skb_queue_len_lockless(&sd->input_pkt_queue); |
4835 | max_backlog = READ_ONCE(net_hotdata.max_backlog); | |
4836 | if (unlikely(qlen > max_backlog)) | |
4837 | goto cpu_backlog_drop; | |
765b11f8 | 4838 | backlog_lock_irq_save(sd, &flags); |
99bbc707 | 4839 | qlen = skb_queue_len(&sd->input_pkt_queue); |
a7ae7b0b | 4840 | if (qlen <= max_backlog && !skb_flow_limit(skb, qlen)) { |
f7efd01f ED |
4841 | if (!qlen) { |
4842 | /* Schedule NAPI for backlog device. We can use | |
4843 | * non atomic operation as we own the queue lock. | |
4844 | */ | |
4845 | if (!__test_and_set_bit(NAPI_STATE_SCHED, | |
4846 | &sd->backlog.state)) | |
4847 | napi_schedule_rps(sd); | |
0a9627f2 | 4848 | } |
f7efd01f | 4849 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
36b83ffc | 4850 | tail = rps_input_queue_tail_incr(sd); |
f7efd01f | 4851 | backlog_unlock_irq_restore(sd, &flags); |
36b83ffc ED |
4852 | |
4853 | /* save the tail outside of the critical section */ | |
4854 | rps_input_queue_tail_save(qtail, tail); | |
f7efd01f | 4855 | return NET_RX_SUCCESS; |
0a9627f2 TH |
4856 | } |
4857 | ||
765b11f8 | 4858 | backlog_unlock_irq_restore(sd, &flags); |
0a9627f2 | 4859 | |
a7ae7b0b ED |
4860 | cpu_backlog_drop: |
4861 | atomic_inc(&sd->dropped); | |
95e48d86 | 4862 | bad_dev: |
625788b5 | 4863 | dev_core_stats_rx_dropped_inc(skb->dev); |
44f0bd40 | 4864 | kfree_skb_reason(skb, reason); |
0a9627f2 TH |
4865 | return NET_RX_DROP; |
4866 | } | |
1da177e4 | 4867 | |
e817f856 JDB |
4868 | static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) |
4869 | { | |
4870 | struct net_device *dev = skb->dev; | |
4871 | struct netdev_rx_queue *rxqueue; | |
4872 | ||
4873 | rxqueue = dev->_rx; | |
4874 | ||
4875 | if (skb_rx_queue_recorded(skb)) { | |
4876 | u16 index = skb_get_rx_queue(skb); | |
4877 | ||
4878 | if (unlikely(index >= dev->real_num_rx_queues)) { | |
4879 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4880 | "%s received packet on queue %u, but number " | |
4881 | "of RX queues is %u\n", | |
4882 | dev->name, index, dev->real_num_rx_queues); | |
4883 | ||
4884 | return rxqueue; /* Return first rxqueue */ | |
4885 | } | |
4886 | rxqueue += index; | |
4887 | } | |
4888 | return rxqueue; | |
4889 | } | |
4890 | ||
fe21cb91 KKD |
4891 | u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, |
4892 | struct bpf_prog *xdp_prog) | |
d4455169 | 4893 | { |
be9df4af | 4894 | void *orig_data, *orig_data_end, *hard_start; |
e817f856 | 4895 | struct netdev_rx_queue *rxqueue; |
22b60343 | 4896 | bool orig_bcast, orig_host; |
43b5169d | 4897 | u32 mac_len, frame_sz; |
29724956 JDB |
4898 | __be16 orig_eth_type; |
4899 | struct ethhdr *eth; | |
fe21cb91 | 4900 | u32 metalen, act; |
be9df4af | 4901 | int off; |
d4455169 | 4902 | |
d4455169 JF |
4903 | /* The XDP program wants to see the packet starting at the MAC |
4904 | * header. | |
4905 | */ | |
4906 | mac_len = skb->data - skb_mac_header(skb); | |
be9df4af | 4907 | hard_start = skb->data - skb_headroom(skb); |
a075767b JDB |
4908 | |
4909 | /* SKB "head" area always have tailroom for skb_shared_info */ | |
be9df4af | 4910 | frame_sz = (void *)skb_end_pointer(skb) - hard_start; |
43b5169d | 4911 | frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
a075767b | 4912 | |
be9df4af LB |
4913 | rxqueue = netif_get_rxqueue(skb); |
4914 | xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); | |
4915 | xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, | |
4916 | skb_headlen(skb) + mac_len, true); | |
e6d5dbdd LB |
4917 | if (skb_is_nonlinear(skb)) { |
4918 | skb_shinfo(skb)->xdp_frags_size = skb->data_len; | |
4919 | xdp_buff_set_frags_flag(xdp); | |
4920 | } else { | |
4921 | xdp_buff_clear_frags_flag(xdp); | |
4922 | } | |
a075767b | 4923 | |
02671e23 BT |
4924 | orig_data_end = xdp->data_end; |
4925 | orig_data = xdp->data; | |
29724956 | 4926 | eth = (struct ethhdr *)xdp->data; |
22b60343 | 4927 | orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); |
29724956 JDB |
4928 | orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); |
4929 | orig_eth_type = eth->h_proto; | |
d4455169 | 4930 | |
02671e23 | 4931 | act = bpf_prog_run_xdp(xdp_prog, xdp); |
d4455169 | 4932 | |
065af355 | 4933 | /* check if bpf_xdp_adjust_head was used */ |
02671e23 | 4934 | off = xdp->data - orig_data; |
065af355 JDB |
4935 | if (off) { |
4936 | if (off > 0) | |
4937 | __skb_pull(skb, off); | |
4938 | else if (off < 0) | |
4939 | __skb_push(skb, -off); | |
4940 | ||
4941 | skb->mac_header += off; | |
4942 | skb_reset_network_header(skb); | |
4943 | } | |
d4455169 | 4944 | |
a075767b JDB |
4945 | /* check if bpf_xdp_adjust_tail was used */ |
4946 | off = xdp->data_end - orig_data_end; | |
f7613120 | 4947 | if (off != 0) { |
02671e23 | 4948 | skb_set_tail_pointer(skb, xdp->data_end - xdp->data); |
a075767b | 4949 | skb->len += off; /* positive on grow, negative on shrink */ |
f7613120 | 4950 | } |
198d83bb | 4951 | |
e6d5dbdd LB |
4952 | /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers |
4953 | * (e.g. bpf_xdp_adjust_tail), we need to update data_len here. | |
4954 | */ | |
4955 | if (xdp_buff_has_frags(xdp)) | |
4956 | skb->data_len = skb_shinfo(skb)->xdp_frags_size; | |
4957 | else | |
4958 | skb->data_len = 0; | |
4959 | ||
29724956 JDB |
4960 | /* check if XDP changed eth hdr such SKB needs update */ |
4961 | eth = (struct ethhdr *)xdp->data; | |
4962 | if ((orig_eth_type != eth->h_proto) || | |
22b60343 MW |
4963 | (orig_host != ether_addr_equal_64bits(eth->h_dest, |
4964 | skb->dev->dev_addr)) || | |
29724956 JDB |
4965 | (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { |
4966 | __skb_push(skb, ETH_HLEN); | |
22b60343 | 4967 | skb->pkt_type = PACKET_HOST; |
29724956 JDB |
4968 | skb->protocol = eth_type_trans(skb, skb->dev); |
4969 | } | |
4970 | ||
fe21cb91 KKD |
4971 | /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull |
4972 | * before calling us again on redirect path. We do not call do_redirect | |
4973 | * as we leave that up to the caller. | |
4974 | * | |
4975 | * Caller is responsible for managing lifetime of skb (i.e. calling | |
4976 | * kfree_skb in response to actions it cannot handle/XDP_DROP). | |
4977 | */ | |
d4455169 | 4978 | switch (act) { |
6103aa96 | 4979 | case XDP_REDIRECT: |
d4455169 JF |
4980 | case XDP_TX: |
4981 | __skb_push(skb, mac_len); | |
de8f3a83 | 4982 | break; |
d4455169 | 4983 | case XDP_PASS: |
02671e23 | 4984 | metalen = xdp->data - xdp->data_meta; |
de8f3a83 DB |
4985 | if (metalen) |
4986 | skb_metadata_set(skb, metalen); | |
d4455169 | 4987 | break; |
fe21cb91 KKD |
4988 | } |
4989 | ||
4990 | return act; | |
4991 | } | |
4992 | ||
e6d5dbdd LB |
4993 | static int |
4994 | netif_skb_check_for_xdp(struct sk_buff **pskb, struct bpf_prog *prog) | |
4995 | { | |
4996 | struct sk_buff *skb = *pskb; | |
4997 | int err, hroom, troom; | |
4998 | ||
4999 | if (!skb_cow_data_for_xdp(this_cpu_read(system_page_pool), pskb, prog)) | |
5000 | return 0; | |
5001 | ||
5002 | /* In case we have to go down the path and also linearize, | |
5003 | * then lets do the pskb_expand_head() work just once here. | |
5004 | */ | |
5005 | hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); | |
5006 | troom = skb->tail + skb->data_len - skb->end; | |
5007 | err = pskb_expand_head(skb, | |
5008 | hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, | |
5009 | troom > 0 ? troom + 128 : 0, GFP_ATOMIC); | |
5010 | if (err) | |
5011 | return err; | |
5012 | ||
5013 | return skb_linearize(skb); | |
5014 | } | |
5015 | ||
4d2bb0bf | 5016 | static u32 netif_receive_generic_xdp(struct sk_buff **pskb, |
fe21cb91 KKD |
5017 | struct xdp_buff *xdp, |
5018 | struct bpf_prog *xdp_prog) | |
5019 | { | |
4d2bb0bf | 5020 | struct sk_buff *skb = *pskb; |
e6d5dbdd | 5021 | u32 mac_len, act = XDP_DROP; |
fe21cb91 KKD |
5022 | |
5023 | /* Reinjected packets coming from act_mirred or similar should | |
5024 | * not get XDP generic processing. | |
5025 | */ | |
5026 | if (skb_is_redirected(skb)) | |
5027 | return XDP_PASS; | |
5028 | ||
e6d5dbdd LB |
5029 | /* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM |
5030 | * bytes. This is the guarantee that also native XDP provides, | |
5031 | * thus we need to do it here as well. | |
fe21cb91 | 5032 | */ |
e6d5dbdd LB |
5033 | mac_len = skb->data - skb_mac_header(skb); |
5034 | __skb_push(skb, mac_len); | |
5035 | ||
fe21cb91 KKD |
5036 | if (skb_cloned(skb) || skb_is_nonlinear(skb) || |
5037 | skb_headroom(skb) < XDP_PACKET_HEADROOM) { | |
e6d5dbdd | 5038 | if (netif_skb_check_for_xdp(pskb, xdp_prog)) |
fe21cb91 KKD |
5039 | goto do_drop; |
5040 | } | |
5041 | ||
e6d5dbdd LB |
5042 | __skb_pull(*pskb, mac_len); |
5043 | ||
5044 | act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog); | |
fe21cb91 KKD |
5045 | switch (act) { |
5046 | case XDP_REDIRECT: | |
5047 | case XDP_TX: | |
5048 | case XDP_PASS: | |
5049 | break; | |
d4455169 | 5050 | default: |
e6d5dbdd | 5051 | bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act); |
df561f66 | 5052 | fallthrough; |
d4455169 | 5053 | case XDP_ABORTED: |
e6d5dbdd | 5054 | trace_xdp_exception((*pskb)->dev, xdp_prog, act); |
df561f66 | 5055 | fallthrough; |
d4455169 JF |
5056 | case XDP_DROP: |
5057 | do_drop: | |
e6d5dbdd | 5058 | kfree_skb(*pskb); |
d4455169 JF |
5059 | break; |
5060 | } | |
5061 | ||
5062 | return act; | |
5063 | } | |
5064 | ||
5065 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
1fd6e567 JA |
5066 | * network taps in order to match in-driver-XDP behavior. This also means |
5067 | * that XDP packets are able to starve other packets going through a qdisc, | |
5068 | * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX | |
5069 | * queues, so they do not have this starvation issue. | |
d4455169 | 5070 | */ |
7c497478 | 5071 | void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) |
d4455169 JF |
5072 | { |
5073 | struct net_device *dev = skb->dev; | |
5074 | struct netdev_queue *txq; | |
5075 | bool free_skb = true; | |
5076 | int cpu, rc; | |
5077 | ||
4bd97d51 | 5078 | txq = netdev_core_pick_tx(dev, skb, NULL); |
d4455169 JF |
5079 | cpu = smp_processor_id(); |
5080 | HARD_TX_LOCK(dev, txq, cpu); | |
1fd6e567 | 5081 | if (!netif_xmit_frozen_or_drv_stopped(txq)) { |
d4455169 JF |
5082 | rc = netdev_start_xmit(skb, dev, txq, 0); |
5083 | if (dev_xmit_complete(rc)) | |
5084 | free_skb = false; | |
5085 | } | |
5086 | HARD_TX_UNLOCK(dev, txq); | |
5087 | if (free_skb) { | |
5088 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
1fd6e567 | 5089 | dev_core_stats_tx_dropped_inc(dev); |
d4455169 JF |
5090 | kfree_skb(skb); |
5091 | } | |
5092 | } | |
5093 | ||
02786475 | 5094 | static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); |
d4455169 | 5095 | |
4d2bb0bf | 5096 | int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff **pskb) |
d4455169 | 5097 | { |
d4455169 | 5098 | if (xdp_prog) { |
02671e23 BT |
5099 | struct xdp_buff xdp; |
5100 | u32 act; | |
6103aa96 | 5101 | int err; |
d4455169 | 5102 | |
4d2bb0bf | 5103 | act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog); |
d4455169 | 5104 | if (act != XDP_PASS) { |
6103aa96 JF |
5105 | switch (act) { |
5106 | case XDP_REDIRECT: | |
4d2bb0bf | 5107 | err = xdp_do_generic_redirect((*pskb)->dev, *pskb, |
02671e23 | 5108 | &xdp, xdp_prog); |
6103aa96 JF |
5109 | if (err) |
5110 | goto out_redir; | |
02671e23 | 5111 | break; |
6103aa96 | 5112 | case XDP_TX: |
4d2bb0bf | 5113 | generic_xdp_tx(*pskb, xdp_prog); |
6103aa96 JF |
5114 | break; |
5115 | } | |
d4455169 JF |
5116 | return XDP_DROP; |
5117 | } | |
5118 | } | |
5119 | return XDP_PASS; | |
6103aa96 | 5120 | out_redir: |
4d2bb0bf | 5121 | kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP); |
6103aa96 | 5122 | return XDP_DROP; |
d4455169 | 5123 | } |
7c497478 | 5124 | EXPORT_SYMBOL_GPL(do_xdp_generic); |
d4455169 | 5125 | |
ae78dbfa | 5126 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 5127 | { |
b0e28f1e | 5128 | int ret; |
1da177e4 | 5129 | |
f59b5416 | 5130 | net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb); |
1da177e4 | 5131 | |
cf66ba58 | 5132 | trace_netif_rx(skb); |
d4455169 | 5133 | |
df334545 | 5134 | #ifdef CONFIG_RPS |
dc05360f | 5135 | if (static_branch_unlikely(&rps_needed)) { |
fec5e652 | 5136 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
5137 | int cpu; |
5138 | ||
5139 | rcu_read_lock(); | |
fec5e652 TH |
5140 | |
5141 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
5142 | if (cpu < 0) |
5143 | cpu = smp_processor_id(); | |
fec5e652 TH |
5144 | |
5145 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
5146 | ||
b0e28f1e | 5147 | rcu_read_unlock(); |
adc9300e ED |
5148 | } else |
5149 | #endif | |
fec5e652 TH |
5150 | { |
5151 | unsigned int qtail; | |
f4563a75 | 5152 | |
f234ae29 | 5153 | ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); |
fec5e652 | 5154 | } |
b0e28f1e | 5155 | return ret; |
1da177e4 | 5156 | } |
ae78dbfa | 5157 | |
baebdf48 SAS |
5158 | /** |
5159 | * __netif_rx - Slightly optimized version of netif_rx | |
5160 | * @skb: buffer to post | |
5161 | * | |
5162 | * This behaves as netif_rx except that it does not disable bottom halves. | |
5163 | * As a result this function may only be invoked from the interrupt context | |
5164 | * (either hard or soft interrupt). | |
5165 | */ | |
5166 | int __netif_rx(struct sk_buff *skb) | |
5167 | { | |
5168 | int ret; | |
5169 | ||
351bdbb6 | 5170 | lockdep_assert_once(hardirq_count() | softirq_count()); |
baebdf48 SAS |
5171 | |
5172 | trace_netif_rx_entry(skb); | |
5173 | ret = netif_rx_internal(skb); | |
5174 | trace_netif_rx_exit(ret); | |
5175 | return ret; | |
5176 | } | |
5177 | EXPORT_SYMBOL(__netif_rx); | |
5178 | ||
ae78dbfa BH |
5179 | /** |
5180 | * netif_rx - post buffer to the network code | |
5181 | * @skb: buffer to post | |
5182 | * | |
5183 | * This function receives a packet from a device driver and queues it for | |
baebdf48 SAS |
5184 | * the upper (protocol) levels to process via the backlog NAPI device. It |
5185 | * always succeeds. The buffer may be dropped during processing for | |
5186 | * congestion control or by the protocol layers. | |
5187 | * The network buffer is passed via the backlog NAPI device. Modern NIC | |
5188 | * driver should use NAPI and GRO. | |
167053f8 SAS |
5189 | * This function can used from interrupt and from process context. The |
5190 | * caller from process context must not disable interrupts before invoking | |
5191 | * this function. | |
ae78dbfa BH |
5192 | * |
5193 | * return values: | |
5194 | * NET_RX_SUCCESS (no congestion) | |
5195 | * NET_RX_DROP (packet was dropped) | |
5196 | * | |
5197 | */ | |
ae78dbfa BH |
5198 | int netif_rx(struct sk_buff *skb) |
5199 | { | |
167053f8 | 5200 | bool need_bh_off = !(hardirq_count() | softirq_count()); |
b0e3f1bd GB |
5201 | int ret; |
5202 | ||
167053f8 SAS |
5203 | if (need_bh_off) |
5204 | local_bh_disable(); | |
ae78dbfa | 5205 | trace_netif_rx_entry(skb); |
b0e3f1bd GB |
5206 | ret = netif_rx_internal(skb); |
5207 | trace_netif_rx_exit(ret); | |
167053f8 SAS |
5208 | if (need_bh_off) |
5209 | local_bh_enable(); | |
b0e3f1bd | 5210 | return ret; |
ae78dbfa | 5211 | } |
d1b19dff | 5212 | EXPORT_SYMBOL(netif_rx); |
1da177e4 | 5213 | |
0766f788 | 5214 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 5215 | { |
903ceff7 | 5216 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
5217 | |
5218 | if (sd->completion_queue) { | |
5219 | struct sk_buff *clist; | |
5220 | ||
5221 | local_irq_disable(); | |
5222 | clist = sd->completion_queue; | |
5223 | sd->completion_queue = NULL; | |
5224 | local_irq_enable(); | |
5225 | ||
5226 | while (clist) { | |
5227 | struct sk_buff *skb = clist; | |
f4563a75 | 5228 | |
1da177e4 LT |
5229 | clist = clist->next; |
5230 | ||
63354797 | 5231 | WARN_ON(refcount_read(&skb->users)); |
40bbae58 | 5232 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) |
dd1b5278 | 5233 | trace_consume_skb(skb, net_tx_action); |
e6247027 | 5234 | else |
c504e5c2 | 5235 | trace_kfree_skb(skb, net_tx_action, |
40bbae58 | 5236 | get_kfree_skb_cb(skb)->reason); |
15fad714 JDB |
5237 | |
5238 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
5239 | __kfree_skb(skb); | |
5240 | else | |
8fa66e4a JK |
5241 | __napi_kfree_skb(skb, |
5242 | get_kfree_skb_cb(skb)->reason); | |
1da177e4 LT |
5243 | } |
5244 | } | |
5245 | ||
5246 | if (sd->output_queue) { | |
37437bb2 | 5247 | struct Qdisc *head; |
1da177e4 LT |
5248 | |
5249 | local_irq_disable(); | |
5250 | head = sd->output_queue; | |
5251 | sd->output_queue = NULL; | |
a9cbd588 | 5252 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
5253 | local_irq_enable(); |
5254 | ||
102b55ee YL |
5255 | rcu_read_lock(); |
5256 | ||
1da177e4 | 5257 | while (head) { |
37437bb2 | 5258 | struct Qdisc *q = head; |
6b3ba914 | 5259 | spinlock_t *root_lock = NULL; |
37437bb2 | 5260 | |
1da177e4 LT |
5261 | head = head->next_sched; |
5262 | ||
3bcb846c ED |
5263 | /* We need to make sure head->next_sched is read |
5264 | * before clearing __QDISC_STATE_SCHED | |
5265 | */ | |
5266 | smp_mb__before_atomic(); | |
102b55ee YL |
5267 | |
5268 | if (!(q->flags & TCQ_F_NOLOCK)) { | |
5269 | root_lock = qdisc_lock(q); | |
5270 | spin_lock(root_lock); | |
5271 | } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, | |
5272 | &q->state))) { | |
5273 | /* There is a synchronize_net() between | |
5274 | * STATE_DEACTIVATED flag being set and | |
5275 | * qdisc_reset()/some_qdisc_is_busy() in | |
5276 | * dev_deactivate(), so we can safely bail out | |
5277 | * early here to avoid data race between | |
5278 | * qdisc_deactivate() and some_qdisc_is_busy() | |
5279 | * for lockless qdisc. | |
5280 | */ | |
5281 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
5282 | continue; | |
5283 | } | |
5284 | ||
3bcb846c ED |
5285 | clear_bit(__QDISC_STATE_SCHED, &q->state); |
5286 | qdisc_run(q); | |
6b3ba914 JF |
5287 | if (root_lock) |
5288 | spin_unlock(root_lock); | |
1da177e4 | 5289 | } |
102b55ee YL |
5290 | |
5291 | rcu_read_unlock(); | |
1da177e4 | 5292 | } |
f53c7239 SK |
5293 | |
5294 | xfrm_dev_backlog(sd); | |
1da177e4 LT |
5295 | } |
5296 | ||
181402a5 | 5297 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
5298 | /* This hook is defined here for ATM LANE */ |
5299 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
5300 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 5301 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 5302 | #endif |
1da177e4 | 5303 | |
24b27fc4 MB |
5304 | /** |
5305 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
5306 | * @dev: device to check | |
5307 | * | |
5308 | * Check if a receive handler is already registered for a given device. | |
5309 | * Return true if there one. | |
5310 | * | |
5311 | * The caller must hold the rtnl_mutex. | |
5312 | */ | |
5313 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
5314 | { | |
5315 | ASSERT_RTNL(); | |
5316 | return dev && rtnl_dereference(dev->rx_handler); | |
5317 | } | |
5318 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
5319 | ||
ab95bfe0 JP |
5320 | /** |
5321 | * netdev_rx_handler_register - register receive handler | |
5322 | * @dev: device to register a handler for | |
5323 | * @rx_handler: receive handler to register | |
93e2c32b | 5324 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 5325 | * |
e227867f | 5326 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
5327 | * called from __netif_receive_skb. A negative errno code is returned |
5328 | * on a failure. | |
5329 | * | |
5330 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
5331 | * |
5332 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
5333 | */ |
5334 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
5335 | rx_handler_func_t *rx_handler, |
5336 | void *rx_handler_data) | |
ab95bfe0 | 5337 | { |
1b7cd004 | 5338 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
5339 | return -EBUSY; |
5340 | ||
f5426250 PA |
5341 | if (dev->priv_flags & IFF_NO_RX_HANDLER) |
5342 | return -EINVAL; | |
5343 | ||
00cfec37 | 5344 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 5345 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
5346 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
5347 | ||
5348 | return 0; | |
5349 | } | |
5350 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
5351 | ||
5352 | /** | |
5353 | * netdev_rx_handler_unregister - unregister receive handler | |
5354 | * @dev: device to unregister a handler from | |
5355 | * | |
166ec369 | 5356 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
5357 | * |
5358 | * The caller must hold the rtnl_mutex. | |
5359 | */ | |
5360 | void netdev_rx_handler_unregister(struct net_device *dev) | |
5361 | { | |
5362 | ||
5363 | ASSERT_RTNL(); | |
a9b3cd7f | 5364 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
5365 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
5366 | * section has a guarantee to see a non NULL rx_handler_data | |
5367 | * as well. | |
5368 | */ | |
5369 | synchronize_net(); | |
a9b3cd7f | 5370 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
5371 | } |
5372 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
5373 | ||
b4b9e355 MG |
5374 | /* |
5375 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
5376 | * the special handling of PFMEMALLOC skbs. | |
5377 | */ | |
5378 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
5379 | { | |
5380 | switch (skb->protocol) { | |
2b8837ae JP |
5381 | case htons(ETH_P_ARP): |
5382 | case htons(ETH_P_IP): | |
5383 | case htons(ETH_P_IPV6): | |
5384 | case htons(ETH_P_8021Q): | |
5385 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
5386 | return true; |
5387 | default: | |
5388 | return false; | |
5389 | } | |
5390 | } | |
5391 | ||
e687ad60 PN |
5392 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
5393 | int *ret, struct net_device *orig_dev) | |
5394 | { | |
5395 | if (nf_hook_ingress_active(skb)) { | |
2c1e2703 AC |
5396 | int ingress_retval; |
5397 | ||
e687ad60 PN |
5398 | if (*pt_prev) { |
5399 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
5400 | *pt_prev = NULL; | |
5401 | } | |
5402 | ||
2c1e2703 AC |
5403 | rcu_read_lock(); |
5404 | ingress_retval = nf_hook_ingress(skb); | |
5405 | rcu_read_unlock(); | |
5406 | return ingress_retval; | |
e687ad60 PN |
5407 | } |
5408 | return 0; | |
5409 | } | |
e687ad60 | 5410 | |
c0bbbdc3 | 5411 | static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, |
88eb1944 | 5412 | struct packet_type **ppt_prev) |
1da177e4 LT |
5413 | { |
5414 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 5415 | rx_handler_func_t *rx_handler; |
c0bbbdc3 | 5416 | struct sk_buff *skb = *pskb; |
f2ccd8fa | 5417 | struct net_device *orig_dev; |
8a4eb573 | 5418 | bool deliver_exact = false; |
1da177e4 | 5419 | int ret = NET_RX_DROP; |
252e3346 | 5420 | __be16 type; |
1da177e4 | 5421 | |
f59b5416 | 5422 | net_timestamp_check(!READ_ONCE(net_hotdata.tstamp_prequeue), skb); |
81bbb3d4 | 5423 | |
cf66ba58 | 5424 | trace_netif_receive_skb(skb); |
9b22ea56 | 5425 | |
cc9bd5ce | 5426 | orig_dev = skb->dev; |
8f903c70 | 5427 | |
c1d2bbe1 | 5428 | skb_reset_network_header(skb); |
fda55eca ED |
5429 | if (!skb_transport_header_was_set(skb)) |
5430 | skb_reset_transport_header(skb); | |
0b5c9db1 | 5431 | skb_reset_mac_len(skb); |
1da177e4 LT |
5432 | |
5433 | pt_prev = NULL; | |
5434 | ||
63d8ea7f | 5435 | another_round: |
b6858177 | 5436 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
5437 | |
5438 | __this_cpu_inc(softnet_data.processed); | |
5439 | ||
458bf2f2 SH |
5440 | if (static_branch_unlikely(&generic_xdp_needed_key)) { |
5441 | int ret2; | |
5442 | ||
2b4cd14f | 5443 | migrate_disable(); |
4d2bb0bf LB |
5444 | ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), |
5445 | &skb); | |
2b4cd14f | 5446 | migrate_enable(); |
458bf2f2 | 5447 | |
c0bbbdc3 BS |
5448 | if (ret2 != XDP_PASS) { |
5449 | ret = NET_RX_DROP; | |
5450 | goto out; | |
5451 | } | |
458bf2f2 SH |
5452 | } |
5453 | ||
324cefaf | 5454 | if (eth_type_vlan(skb->protocol)) { |
0d5501c1 | 5455 | skb = skb_vlan_untag(skb); |
bcc6d479 | 5456 | if (unlikely(!skb)) |
2c17d27c | 5457 | goto out; |
bcc6d479 JP |
5458 | } |
5459 | ||
cd14e9b7 | 5460 | if (skb_skip_tc_classify(skb)) |
e7246e12 | 5461 | goto skip_classify; |
1da177e4 | 5462 | |
9754e293 | 5463 | if (pfmemalloc) |
b4b9e355 MG |
5464 | goto skip_taps; |
5465 | ||
0b91fa4b | 5466 | list_for_each_entry_rcu(ptype, &net_hotdata.ptype_all, list) { |
7866a621 SN |
5467 | if (pt_prev) |
5468 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5469 | pt_prev = ptype; | |
5470 | } | |
5471 | ||
5472 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
5473 | if (pt_prev) | |
5474 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5475 | pt_prev = ptype; | |
1da177e4 LT |
5476 | } |
5477 | ||
b4b9e355 | 5478 | skip_taps: |
1cf51900 | 5479 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 5480 | if (static_branch_unlikely(&ingress_needed_key)) { |
9aa1206e DB |
5481 | bool another = false; |
5482 | ||
42df6e1d | 5483 | nf_skip_egress(skb, true); |
9aa1206e DB |
5484 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, |
5485 | &another); | |
5486 | if (another) | |
5487 | goto another_round; | |
4577139b | 5488 | if (!skb) |
2c17d27c | 5489 | goto out; |
e687ad60 | 5490 | |
42df6e1d | 5491 | nf_skip_egress(skb, false); |
e687ad60 | 5492 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) |
2c17d27c | 5493 | goto out; |
cd14e9b7 | 5494 | } |
1cf51900 | 5495 | #endif |
2c64605b | 5496 | skb_reset_redirect(skb); |
e7246e12 | 5497 | skip_classify: |
9754e293 | 5498 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
5499 | goto drop; |
5500 | ||
df8a39de | 5501 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
5502 | if (pt_prev) { |
5503 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5504 | pt_prev = NULL; | |
5505 | } | |
48cc32d3 | 5506 | if (vlan_do_receive(&skb)) |
2425717b JF |
5507 | goto another_round; |
5508 | else if (unlikely(!skb)) | |
2c17d27c | 5509 | goto out; |
2425717b JF |
5510 | } |
5511 | ||
48cc32d3 | 5512 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
5513 | if (rx_handler) { |
5514 | if (pt_prev) { | |
5515 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5516 | pt_prev = NULL; | |
5517 | } | |
8a4eb573 JP |
5518 | switch (rx_handler(&skb)) { |
5519 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 5520 | ret = NET_RX_SUCCESS; |
2c17d27c | 5521 | goto out; |
8a4eb573 | 5522 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 5523 | goto another_round; |
8a4eb573 JP |
5524 | case RX_HANDLER_EXACT: |
5525 | deliver_exact = true; | |
b1866bff | 5526 | break; |
8a4eb573 JP |
5527 | case RX_HANDLER_PASS: |
5528 | break; | |
5529 | default: | |
5530 | BUG(); | |
5531 | } | |
ab95bfe0 | 5532 | } |
1da177e4 | 5533 | |
b14a9fc4 | 5534 | if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { |
36b2f61a GV |
5535 | check_vlan_id: |
5536 | if (skb_vlan_tag_get_id(skb)) { | |
5537 | /* Vlan id is non 0 and vlan_do_receive() above couldn't | |
5538 | * find vlan device. | |
5539 | */ | |
d4b812de | 5540 | skb->pkt_type = PACKET_OTHERHOST; |
324cefaf | 5541 | } else if (eth_type_vlan(skb->protocol)) { |
36b2f61a GV |
5542 | /* Outer header is 802.1P with vlan 0, inner header is |
5543 | * 802.1Q or 802.1AD and vlan_do_receive() above could | |
5544 | * not find vlan dev for vlan id 0. | |
5545 | */ | |
5546 | __vlan_hwaccel_clear_tag(skb); | |
5547 | skb = skb_vlan_untag(skb); | |
5548 | if (unlikely(!skb)) | |
5549 | goto out; | |
5550 | if (vlan_do_receive(&skb)) | |
5551 | /* After stripping off 802.1P header with vlan 0 | |
5552 | * vlan dev is found for inner header. | |
5553 | */ | |
5554 | goto another_round; | |
5555 | else if (unlikely(!skb)) | |
5556 | goto out; | |
5557 | else | |
5558 | /* We have stripped outer 802.1P vlan 0 header. | |
5559 | * But could not find vlan dev. | |
5560 | * check again for vlan id to set OTHERHOST. | |
5561 | */ | |
5562 | goto check_vlan_id; | |
5563 | } | |
d4b812de ED |
5564 | /* Note: we might in the future use prio bits |
5565 | * and set skb->priority like in vlan_do_receive() | |
5566 | * For the time being, just ignore Priority Code Point | |
5567 | */ | |
b1817524 | 5568 | __vlan_hwaccel_clear_tag(skb); |
d4b812de | 5569 | } |
48cc32d3 | 5570 | |
7866a621 SN |
5571 | type = skb->protocol; |
5572 | ||
63d8ea7f | 5573 | /* deliver only exact match when indicated */ |
7866a621 SN |
5574 | if (likely(!deliver_exact)) { |
5575 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
5576 | &ptype_base[ntohs(type) & | |
5577 | PTYPE_HASH_MASK]); | |
5578 | } | |
1f3c8804 | 5579 | |
7866a621 SN |
5580 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
5581 | &orig_dev->ptype_specific); | |
5582 | ||
5583 | if (unlikely(skb->dev != orig_dev)) { | |
5584 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
5585 | &skb->dev->ptype_specific); | |
1da177e4 LT |
5586 | } |
5587 | ||
5588 | if (pt_prev) { | |
1f8b977a | 5589 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
0e698bf6 | 5590 | goto drop; |
88eb1944 | 5591 | *ppt_prev = pt_prev; |
1da177e4 | 5592 | } else { |
b4b9e355 | 5593 | drop: |
9f8ed577 | 5594 | if (!deliver_exact) |
625788b5 | 5595 | dev_core_stats_rx_dropped_inc(skb->dev); |
9f8ed577 | 5596 | else |
625788b5 | 5597 | dev_core_stats_rx_nohandler_inc(skb->dev); |
9f8ed577 | 5598 | kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); |
1da177e4 LT |
5599 | /* Jamal, now you will not able to escape explaining |
5600 | * me how you were going to use this. :-) | |
5601 | */ | |
5602 | ret = NET_RX_DROP; | |
5603 | } | |
5604 | ||
2c17d27c | 5605 | out: |
c0bbbdc3 BS |
5606 | /* The invariant here is that if *ppt_prev is not NULL |
5607 | * then skb should also be non-NULL. | |
5608 | * | |
5609 | * Apparently *ppt_prev assignment above holds this invariant due to | |
5610 | * skb dereferencing near it. | |
5611 | */ | |
5612 | *pskb = skb; | |
9754e293 DM |
5613 | return ret; |
5614 | } | |
5615 | ||
88eb1944 EC |
5616 | static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) |
5617 | { | |
5618 | struct net_device *orig_dev = skb->dev; | |
5619 | struct packet_type *pt_prev = NULL; | |
5620 | int ret; | |
5621 | ||
c0bbbdc3 | 5622 | ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); |
88eb1944 | 5623 | if (pt_prev) |
f5737cba PA |
5624 | ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, |
5625 | skb->dev, pt_prev, orig_dev); | |
88eb1944 EC |
5626 | return ret; |
5627 | } | |
5628 | ||
1c601d82 JDB |
5629 | /** |
5630 | * netif_receive_skb_core - special purpose version of netif_receive_skb | |
5631 | * @skb: buffer to process | |
5632 | * | |
5633 | * More direct receive version of netif_receive_skb(). It should | |
5634 | * only be used by callers that have a need to skip RPS and Generic XDP. | |
2de9780f | 5635 | * Caller must also take care of handling if ``(page_is_)pfmemalloc``. |
1c601d82 JDB |
5636 | * |
5637 | * This function may only be called from softirq context and interrupts | |
5638 | * should be enabled. | |
5639 | * | |
5640 | * Return values (usually ignored): | |
5641 | * NET_RX_SUCCESS: no congestion | |
5642 | * NET_RX_DROP: packet was dropped | |
5643 | */ | |
5644 | int netif_receive_skb_core(struct sk_buff *skb) | |
5645 | { | |
5646 | int ret; | |
5647 | ||
5648 | rcu_read_lock(); | |
88eb1944 | 5649 | ret = __netif_receive_skb_one_core(skb, false); |
1c601d82 JDB |
5650 | rcu_read_unlock(); |
5651 | ||
5652 | return ret; | |
5653 | } | |
5654 | EXPORT_SYMBOL(netif_receive_skb_core); | |
5655 | ||
88eb1944 EC |
5656 | static inline void __netif_receive_skb_list_ptype(struct list_head *head, |
5657 | struct packet_type *pt_prev, | |
5658 | struct net_device *orig_dev) | |
4ce0017a EC |
5659 | { |
5660 | struct sk_buff *skb, *next; | |
5661 | ||
88eb1944 EC |
5662 | if (!pt_prev) |
5663 | return; | |
5664 | if (list_empty(head)) | |
5665 | return; | |
17266ee9 | 5666 | if (pt_prev->list_func != NULL) |
fdf71426 PA |
5667 | INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, |
5668 | ip_list_rcv, head, pt_prev, orig_dev); | |
17266ee9 | 5669 | else |
9a5a90d1 AL |
5670 | list_for_each_entry_safe(skb, next, head, list) { |
5671 | skb_list_del_init(skb); | |
fdf71426 | 5672 | pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
9a5a90d1 | 5673 | } |
88eb1944 EC |
5674 | } |
5675 | ||
5676 | static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) | |
5677 | { | |
5678 | /* Fast-path assumptions: | |
5679 | * - There is no RX handler. | |
5680 | * - Only one packet_type matches. | |
5681 | * If either of these fails, we will end up doing some per-packet | |
5682 | * processing in-line, then handling the 'last ptype' for the whole | |
5683 | * sublist. This can't cause out-of-order delivery to any single ptype, | |
5684 | * because the 'last ptype' must be constant across the sublist, and all | |
5685 | * other ptypes are handled per-packet. | |
5686 | */ | |
5687 | /* Current (common) ptype of sublist */ | |
5688 | struct packet_type *pt_curr = NULL; | |
5689 | /* Current (common) orig_dev of sublist */ | |
5690 | struct net_device *od_curr = NULL; | |
5691 | struct list_head sublist; | |
5692 | struct sk_buff *skb, *next; | |
5693 | ||
9af86f93 | 5694 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5695 | list_for_each_entry_safe(skb, next, head, list) { |
5696 | struct net_device *orig_dev = skb->dev; | |
5697 | struct packet_type *pt_prev = NULL; | |
5698 | ||
22f6bbb7 | 5699 | skb_list_del_init(skb); |
c0bbbdc3 | 5700 | __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); |
9af86f93 EC |
5701 | if (!pt_prev) |
5702 | continue; | |
88eb1944 EC |
5703 | if (pt_curr != pt_prev || od_curr != orig_dev) { |
5704 | /* dispatch old sublist */ | |
88eb1944 EC |
5705 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
5706 | /* start new sublist */ | |
9af86f93 | 5707 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5708 | pt_curr = pt_prev; |
5709 | od_curr = orig_dev; | |
5710 | } | |
9af86f93 | 5711 | list_add_tail(&skb->list, &sublist); |
88eb1944 EC |
5712 | } |
5713 | ||
5714 | /* dispatch final sublist */ | |
9af86f93 | 5715 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
4ce0017a EC |
5716 | } |
5717 | ||
9754e293 DM |
5718 | static int __netif_receive_skb(struct sk_buff *skb) |
5719 | { | |
5720 | int ret; | |
5721 | ||
5722 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 5723 | unsigned int noreclaim_flag; |
9754e293 DM |
5724 | |
5725 | /* | |
5726 | * PFMEMALLOC skbs are special, they should | |
5727 | * - be delivered to SOCK_MEMALLOC sockets only | |
5728 | * - stay away from userspace | |
5729 | * - have bounded memory usage | |
5730 | * | |
5731 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
5732 | * context down to all allocation sites. | |
5733 | */ | |
f1083048 | 5734 | noreclaim_flag = memalloc_noreclaim_save(); |
88eb1944 | 5735 | ret = __netif_receive_skb_one_core(skb, true); |
f1083048 | 5736 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 | 5737 | } else |
88eb1944 | 5738 | ret = __netif_receive_skb_one_core(skb, false); |
9754e293 | 5739 | |
1da177e4 LT |
5740 | return ret; |
5741 | } | |
0a9627f2 | 5742 | |
4ce0017a EC |
5743 | static void __netif_receive_skb_list(struct list_head *head) |
5744 | { | |
5745 | unsigned long noreclaim_flag = 0; | |
5746 | struct sk_buff *skb, *next; | |
5747 | bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ | |
5748 | ||
5749 | list_for_each_entry_safe(skb, next, head, list) { | |
5750 | if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { | |
5751 | struct list_head sublist; | |
5752 | ||
5753 | /* Handle the previous sublist */ | |
5754 | list_cut_before(&sublist, head, &skb->list); | |
b9f463d6 EC |
5755 | if (!list_empty(&sublist)) |
5756 | __netif_receive_skb_list_core(&sublist, pfmemalloc); | |
4ce0017a EC |
5757 | pfmemalloc = !pfmemalloc; |
5758 | /* See comments in __netif_receive_skb */ | |
5759 | if (pfmemalloc) | |
5760 | noreclaim_flag = memalloc_noreclaim_save(); | |
5761 | else | |
5762 | memalloc_noreclaim_restore(noreclaim_flag); | |
5763 | } | |
5764 | } | |
5765 | /* Handle the remaining sublist */ | |
b9f463d6 EC |
5766 | if (!list_empty(head)) |
5767 | __netif_receive_skb_list_core(head, pfmemalloc); | |
4ce0017a EC |
5768 | /* Restore pflags */ |
5769 | if (pfmemalloc) | |
5770 | memalloc_noreclaim_restore(noreclaim_flag); | |
5771 | } | |
5772 | ||
f4e63525 | 5773 | static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) |
b5cdae32 | 5774 | { |
58038695 | 5775 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); |
b5cdae32 DM |
5776 | struct bpf_prog *new = xdp->prog; |
5777 | int ret = 0; | |
5778 | ||
5779 | switch (xdp->command) { | |
58038695 | 5780 | case XDP_SETUP_PROG: |
b5cdae32 DM |
5781 | rcu_assign_pointer(dev->xdp_prog, new); |
5782 | if (old) | |
5783 | bpf_prog_put(old); | |
5784 | ||
5785 | if (old && !new) { | |
02786475 | 5786 | static_branch_dec(&generic_xdp_needed_key); |
b5cdae32 | 5787 | } else if (new && !old) { |
02786475 | 5788 | static_branch_inc(&generic_xdp_needed_key); |
b5cdae32 | 5789 | dev_disable_lro(dev); |
56f5aa77 | 5790 | dev_disable_gro_hw(dev); |
b5cdae32 DM |
5791 | } |
5792 | break; | |
b5cdae32 | 5793 | |
b5cdae32 DM |
5794 | default: |
5795 | ret = -EINVAL; | |
5796 | break; | |
5797 | } | |
5798 | ||
5799 | return ret; | |
5800 | } | |
5801 | ||
ae78dbfa | 5802 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 5803 | { |
2c17d27c JA |
5804 | int ret; |
5805 | ||
f59b5416 | 5806 | net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb); |
3b098e2d | 5807 | |
c1f19b51 RC |
5808 | if (skb_defer_rx_timestamp(skb)) |
5809 | return NET_RX_SUCCESS; | |
5810 | ||
bbbe211c | 5811 | rcu_read_lock(); |
df334545 | 5812 | #ifdef CONFIG_RPS |
dc05360f | 5813 | if (static_branch_unlikely(&rps_needed)) { |
3b098e2d | 5814 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 5815 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 5816 | |
3b098e2d ED |
5817 | if (cpu >= 0) { |
5818 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
5819 | rcu_read_unlock(); | |
adc9300e | 5820 | return ret; |
3b098e2d | 5821 | } |
fec5e652 | 5822 | } |
1e94d72f | 5823 | #endif |
2c17d27c JA |
5824 | ret = __netif_receive_skb(skb); |
5825 | rcu_read_unlock(); | |
5826 | return ret; | |
0a9627f2 | 5827 | } |
ae78dbfa | 5828 | |
587652bb | 5829 | void netif_receive_skb_list_internal(struct list_head *head) |
7da517a3 | 5830 | { |
7da517a3 | 5831 | struct sk_buff *skb, *next; |
8c057efa | 5832 | struct list_head sublist; |
7da517a3 | 5833 | |
8c057efa | 5834 | INIT_LIST_HEAD(&sublist); |
7da517a3 | 5835 | list_for_each_entry_safe(skb, next, head, list) { |
f59b5416 ED |
5836 | net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), |
5837 | skb); | |
22f6bbb7 | 5838 | skb_list_del_init(skb); |
8c057efa EC |
5839 | if (!skb_defer_rx_timestamp(skb)) |
5840 | list_add_tail(&skb->list, &sublist); | |
7da517a3 | 5841 | } |
8c057efa | 5842 | list_splice_init(&sublist, head); |
7da517a3 | 5843 | |
7da517a3 EC |
5844 | rcu_read_lock(); |
5845 | #ifdef CONFIG_RPS | |
dc05360f | 5846 | if (static_branch_unlikely(&rps_needed)) { |
7da517a3 EC |
5847 | list_for_each_entry_safe(skb, next, head, list) { |
5848 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
5849 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
5850 | ||
5851 | if (cpu >= 0) { | |
8c057efa | 5852 | /* Will be handled, remove from list */ |
22f6bbb7 | 5853 | skb_list_del_init(skb); |
8c057efa | 5854 | enqueue_to_backlog(skb, cpu, &rflow->last_qtail); |
7da517a3 EC |
5855 | } |
5856 | } | |
5857 | } | |
5858 | #endif | |
5859 | __netif_receive_skb_list(head); | |
5860 | rcu_read_unlock(); | |
5861 | } | |
5862 | ||
ae78dbfa BH |
5863 | /** |
5864 | * netif_receive_skb - process receive buffer from network | |
5865 | * @skb: buffer to process | |
5866 | * | |
5867 | * netif_receive_skb() is the main receive data processing function. | |
5868 | * It always succeeds. The buffer may be dropped during processing | |
5869 | * for congestion control or by the protocol layers. | |
5870 | * | |
5871 | * This function may only be called from softirq context and interrupts | |
5872 | * should be enabled. | |
5873 | * | |
5874 | * Return values (usually ignored): | |
5875 | * NET_RX_SUCCESS: no congestion | |
5876 | * NET_RX_DROP: packet was dropped | |
5877 | */ | |
04eb4489 | 5878 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa | 5879 | { |
b0e3f1bd GB |
5880 | int ret; |
5881 | ||
ae78dbfa BH |
5882 | trace_netif_receive_skb_entry(skb); |
5883 | ||
b0e3f1bd GB |
5884 | ret = netif_receive_skb_internal(skb); |
5885 | trace_netif_receive_skb_exit(ret); | |
5886 | ||
5887 | return ret; | |
ae78dbfa | 5888 | } |
04eb4489 | 5889 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 5890 | |
f6ad8c1b EC |
5891 | /** |
5892 | * netif_receive_skb_list - process many receive buffers from network | |
5893 | * @head: list of skbs to process. | |
5894 | * | |
7da517a3 EC |
5895 | * Since return value of netif_receive_skb() is normally ignored, and |
5896 | * wouldn't be meaningful for a list, this function returns void. | |
f6ad8c1b EC |
5897 | * |
5898 | * This function may only be called from softirq context and interrupts | |
5899 | * should be enabled. | |
5900 | */ | |
5901 | void netif_receive_skb_list(struct list_head *head) | |
5902 | { | |
7da517a3 | 5903 | struct sk_buff *skb; |
f6ad8c1b | 5904 | |
b9f463d6 EC |
5905 | if (list_empty(head)) |
5906 | return; | |
b0e3f1bd GB |
5907 | if (trace_netif_receive_skb_list_entry_enabled()) { |
5908 | list_for_each_entry(skb, head, list) | |
5909 | trace_netif_receive_skb_list_entry(skb); | |
5910 | } | |
7da517a3 | 5911 | netif_receive_skb_list_internal(head); |
b0e3f1bd | 5912 | trace_netif_receive_skb_list_exit(0); |
f6ad8c1b EC |
5913 | } |
5914 | EXPORT_SYMBOL(netif_receive_skb_list); | |
5915 | ||
ce1e2a77 | 5916 | static DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
5917 | |
5918 | /* Network device is going away, flush any packets still pending */ | |
5919 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 5920 | { |
6e583ce5 | 5921 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
5922 | struct softnet_data *sd; |
5923 | ||
5924 | local_bh_disable(); | |
5925 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 5926 | |
765b11f8 | 5927 | backlog_lock_irq_disable(sd); |
6e7676c1 | 5928 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 5929 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 5930 | __skb_unlink(skb, &sd->input_pkt_queue); |
7df5cb75 | 5931 | dev_kfree_skb_irq(skb); |
36b83ffc | 5932 | rps_input_queue_head_incr(sd); |
6e583ce5 | 5933 | } |
6e7676c1 | 5934 | } |
765b11f8 | 5935 | backlog_unlock_irq_enable(sd); |
6e7676c1 CG |
5936 | |
5937 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 5938 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
5939 | __skb_unlink(skb, &sd->process_queue); |
5940 | kfree_skb(skb); | |
36b83ffc | 5941 | rps_input_queue_head_incr(sd); |
6e7676c1 CG |
5942 | } |
5943 | } | |
145dd5f9 PA |
5944 | local_bh_enable(); |
5945 | } | |
5946 | ||
2de79ee2 PA |
5947 | static bool flush_required(int cpu) |
5948 | { | |
5949 | #if IS_ENABLED(CONFIG_RPS) | |
5950 | struct softnet_data *sd = &per_cpu(softnet_data, cpu); | |
5951 | bool do_flush; | |
5952 | ||
765b11f8 | 5953 | backlog_lock_irq_disable(sd); |
2de79ee2 PA |
5954 | |
5955 | /* as insertion into process_queue happens with the rps lock held, | |
5956 | * process_queue access may race only with dequeue | |
5957 | */ | |
5958 | do_flush = !skb_queue_empty(&sd->input_pkt_queue) || | |
5959 | !skb_queue_empty_lockless(&sd->process_queue); | |
765b11f8 | 5960 | backlog_unlock_irq_enable(sd); |
2de79ee2 PA |
5961 | |
5962 | return do_flush; | |
5963 | #endif | |
5964 | /* without RPS we can't safely check input_pkt_queue: during a | |
5965 | * concurrent remote skb_queue_splice() we can detect as empty both | |
5966 | * input_pkt_queue and process_queue even if the latter could end-up | |
5967 | * containing a lot of packets. | |
5968 | */ | |
5969 | return true; | |
5970 | } | |
5971 | ||
41852497 | 5972 | static void flush_all_backlogs(void) |
145dd5f9 | 5973 | { |
2de79ee2 | 5974 | static cpumask_t flush_cpus; |
145dd5f9 PA |
5975 | unsigned int cpu; |
5976 | ||
2de79ee2 PA |
5977 | /* since we are under rtnl lock protection we can use static data |
5978 | * for the cpumask and avoid allocating on stack the possibly | |
5979 | * large mask | |
5980 | */ | |
5981 | ASSERT_RTNL(); | |
5982 | ||
372bbdd5 | 5983 | cpus_read_lock(); |
145dd5f9 | 5984 | |
2de79ee2 PA |
5985 | cpumask_clear(&flush_cpus); |
5986 | for_each_online_cpu(cpu) { | |
5987 | if (flush_required(cpu)) { | |
5988 | queue_work_on(cpu, system_highpri_wq, | |
5989 | per_cpu_ptr(&flush_works, cpu)); | |
5990 | cpumask_set_cpu(cpu, &flush_cpus); | |
5991 | } | |
5992 | } | |
145dd5f9 | 5993 | |
2de79ee2 | 5994 | /* we can have in flight packet[s] on the cpus we are not flushing, |
0cbe1e57 | 5995 | * synchronize_net() in unregister_netdevice_many() will take care of |
2de79ee2 PA |
5996 | * them |
5997 | */ | |
5998 | for_each_cpu(cpu, &flush_cpus) | |
41852497 | 5999 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 | 6000 | |
372bbdd5 | 6001 | cpus_read_unlock(); |
6e583ce5 SH |
6002 | } |
6003 | ||
773fc8f6 | 6004 | static void net_rps_send_ipi(struct softnet_data *remsd) |
6005 | { | |
6006 | #ifdef CONFIG_RPS | |
6007 | while (remsd) { | |
6008 | struct softnet_data *next = remsd->rps_ipi_next; | |
6009 | ||
6010 | if (cpu_online(remsd->cpu)) | |
6011 | smp_call_function_single_async(remsd->cpu, &remsd->csd); | |
6012 | remsd = next; | |
6013 | } | |
6014 | #endif | |
6015 | } | |
6016 | ||
e326bed2 | 6017 | /* |
855abcf0 | 6018 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
6019 | * Note: called with local irq disabled, but exits with local irq enabled. |
6020 | */ | |
6021 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
6022 | { | |
6023 | #ifdef CONFIG_RPS | |
6024 | struct softnet_data *remsd = sd->rps_ipi_list; | |
6025 | ||
dad6b977 | 6026 | if (!use_backlog_threads() && remsd) { |
e326bed2 ED |
6027 | sd->rps_ipi_list = NULL; |
6028 | ||
6029 | local_irq_enable(); | |
6030 | ||
6031 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
773fc8f6 | 6032 | net_rps_send_ipi(remsd); |
e326bed2 ED |
6033 | } else |
6034 | #endif | |
6035 | local_irq_enable(); | |
6036 | } | |
6037 | ||
d75b1ade ED |
6038 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
6039 | { | |
6040 | #ifdef CONFIG_RPS | |
dad6b977 | 6041 | return !use_backlog_threads() && sd->rps_ipi_list; |
d75b1ade ED |
6042 | #else |
6043 | return false; | |
6044 | #endif | |
6045 | } | |
6046 | ||
bea3348e | 6047 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 6048 | { |
eecfd7c4 | 6049 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
6050 | bool again = true; |
6051 | int work = 0; | |
1da177e4 | 6052 | |
e326bed2 ED |
6053 | /* Check if we have pending ipi, its better to send them now, |
6054 | * not waiting net_rx_action() end. | |
6055 | */ | |
d75b1ade | 6056 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
6057 | local_irq_disable(); |
6058 | net_rps_action_and_irq_enable(sd); | |
6059 | } | |
d75b1ade | 6060 | |
71c0de9b | 6061 | napi->weight = READ_ONCE(net_hotdata.dev_rx_weight); |
145dd5f9 | 6062 | while (again) { |
1da177e4 | 6063 | struct sk_buff *skb; |
6e7676c1 CG |
6064 | |
6065 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 6066 | rcu_read_lock(); |
6e7676c1 | 6067 | __netif_receive_skb(skb); |
2c17d27c | 6068 | rcu_read_unlock(); |
c62fdf5b ED |
6069 | if (++work >= quota) { |
6070 | rps_input_queue_head_add(sd, work); | |
76cc8b13 | 6071 | return work; |
c62fdf5b | 6072 | } |
145dd5f9 | 6073 | |
6e7676c1 | 6074 | } |
1da177e4 | 6075 | |
765b11f8 | 6076 | backlog_lock_irq_disable(sd); |
11ef7a89 | 6077 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
6078 | /* |
6079 | * Inline a custom version of __napi_complete(). | |
6080 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
6081 | * and NAPI_STATE_SCHED is the only possible flag set |
6082 | * on backlog. | |
6083 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
6084 | * and we dont need an smp_mb() memory barrier. |
6085 | */ | |
dad6b977 | 6086 | napi->state &= NAPIF_STATE_THREADED; |
145dd5f9 PA |
6087 | again = false; |
6088 | } else { | |
6089 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
6090 | &sd->process_queue); | |
bea3348e | 6091 | } |
765b11f8 | 6092 | backlog_unlock_irq_enable(sd); |
6e7676c1 | 6093 | } |
1da177e4 | 6094 | |
c62fdf5b ED |
6095 | if (work) |
6096 | rps_input_queue_head_add(sd, work); | |
bea3348e SH |
6097 | return work; |
6098 | } | |
1da177e4 | 6099 | |
bea3348e SH |
6100 | /** |
6101 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 6102 | * @n: entry to schedule |
bea3348e | 6103 | * |
bc9ad166 ED |
6104 | * The entry's receive function will be scheduled to run. |
6105 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 6106 | */ |
b5606c2d | 6107 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
6108 | { |
6109 | unsigned long flags; | |
1da177e4 | 6110 | |
bea3348e | 6111 | local_irq_save(flags); |
903ceff7 | 6112 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 6113 | local_irq_restore(flags); |
1da177e4 | 6114 | } |
bea3348e SH |
6115 | EXPORT_SYMBOL(__napi_schedule); |
6116 | ||
39e6c820 ED |
6117 | /** |
6118 | * napi_schedule_prep - check if napi can be scheduled | |
6119 | * @n: napi context | |
6120 | * | |
6121 | * Test if NAPI routine is already running, and if not mark | |
ee1a4c84 | 6122 | * it as running. This is used as a condition variable to |
39e6c820 ED |
6123 | * insure only one NAPI poll instance runs. We also make |
6124 | * sure there is no pending NAPI disable. | |
6125 | */ | |
6126 | bool napi_schedule_prep(struct napi_struct *n) | |
6127 | { | |
1462160c | 6128 | unsigned long new, val = READ_ONCE(n->state); |
39e6c820 ED |
6129 | |
6130 | do { | |
39e6c820 ED |
6131 | if (unlikely(val & NAPIF_STATE_DISABLE)) |
6132 | return false; | |
6133 | new = val | NAPIF_STATE_SCHED; | |
6134 | ||
6135 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
6136 | * This was suggested by Alexander Duyck, as compiler | |
6137 | * emits better code than : | |
6138 | * if (val & NAPIF_STATE_SCHED) | |
6139 | * new |= NAPIF_STATE_MISSED; | |
6140 | */ | |
6141 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
6142 | NAPIF_STATE_MISSED; | |
1462160c | 6143 | } while (!try_cmpxchg(&n->state, &val, new)); |
39e6c820 ED |
6144 | |
6145 | return !(val & NAPIF_STATE_SCHED); | |
6146 | } | |
6147 | EXPORT_SYMBOL(napi_schedule_prep); | |
6148 | ||
bc9ad166 ED |
6149 | /** |
6150 | * __napi_schedule_irqoff - schedule for receive | |
6151 | * @n: entry to schedule | |
6152 | * | |
8380c81d SAS |
6153 | * Variant of __napi_schedule() assuming hard irqs are masked. |
6154 | * | |
6155 | * On PREEMPT_RT enabled kernels this maps to __napi_schedule() | |
6156 | * because the interrupt disabled assumption might not be true | |
6157 | * due to force-threaded interrupts and spinlock substitution. | |
bc9ad166 ED |
6158 | */ |
6159 | void __napi_schedule_irqoff(struct napi_struct *n) | |
6160 | { | |
8380c81d SAS |
6161 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) |
6162 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
6163 | else | |
6164 | __napi_schedule(n); | |
bc9ad166 ED |
6165 | } |
6166 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
6167 | ||
364b6055 | 6168 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 6169 | { |
6f8b12d6 ED |
6170 | unsigned long flags, val, new, timeout = 0; |
6171 | bool ret = true; | |
d565b0a1 HX |
6172 | |
6173 | /* | |
217f6974 ED |
6174 | * 1) Don't let napi dequeue from the cpu poll list |
6175 | * just in case its running on a different cpu. | |
6176 | * 2) If we are busy polling, do nothing here, we have | |
6177 | * the guarantee we will be called later. | |
d565b0a1 | 6178 | */ |
217f6974 ED |
6179 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
6180 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 6181 | return false; |
d565b0a1 | 6182 | |
6f8b12d6 ED |
6183 | if (work_done) { |
6184 | if (n->gro_bitmask) | |
7e417a66 ED |
6185 | timeout = READ_ONCE(n->dev->gro_flush_timeout); |
6186 | n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); | |
6f8b12d6 ED |
6187 | } |
6188 | if (n->defer_hard_irqs_count > 0) { | |
6189 | n->defer_hard_irqs_count--; | |
7e417a66 | 6190 | timeout = READ_ONCE(n->dev->gro_flush_timeout); |
6f8b12d6 ED |
6191 | if (timeout) |
6192 | ret = false; | |
6193 | } | |
6194 | if (n->gro_bitmask) { | |
605108ac PA |
6195 | /* When the NAPI instance uses a timeout and keeps postponing |
6196 | * it, we need to bound somehow the time packets are kept in | |
6197 | * the GRO layer | |
6198 | */ | |
6199 | napi_gro_flush(n, !!timeout); | |
3b47d303 | 6200 | } |
c8079432 MM |
6201 | |
6202 | gro_normal_list(n); | |
6203 | ||
02c1602e | 6204 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
6205 | /* If n->poll_list is not empty, we need to mask irqs */ |
6206 | local_irq_save(flags); | |
02c1602e | 6207 | list_del_init(&n->poll_list); |
d75b1ade ED |
6208 | local_irq_restore(flags); |
6209 | } | |
8c48eea3 | 6210 | WRITE_ONCE(n->list_owner, -1); |
39e6c820 | 6211 | |
1462160c | 6212 | val = READ_ONCE(n->state); |
39e6c820 | 6213 | do { |
39e6c820 ED |
6214 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); |
6215 | ||
7fd3253a | 6216 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | |
cb038357 | 6217 | NAPIF_STATE_SCHED_THREADED | |
7fd3253a | 6218 | NAPIF_STATE_PREFER_BUSY_POLL); |
39e6c820 ED |
6219 | |
6220 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
6221 | * because we will call napi->poll() one more time. | |
6222 | * This C code was suggested by Alexander Duyck to help gcc. | |
6223 | */ | |
6224 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
6225 | NAPIF_STATE_SCHED; | |
1462160c | 6226 | } while (!try_cmpxchg(&n->state, &val, new)); |
39e6c820 ED |
6227 | |
6228 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
6229 | __napi_schedule(n); | |
6230 | return false; | |
6231 | } | |
6232 | ||
6f8b12d6 ED |
6233 | if (timeout) |
6234 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
6235 | HRTIMER_MODE_REL_PINNED); | |
6236 | return ret; | |
d565b0a1 | 6237 | } |
3b47d303 | 6238 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 6239 | |
af12fa6e | 6240 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
27f91aaf | 6241 | struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
6242 | { |
6243 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
6244 | struct napi_struct *napi; | |
6245 | ||
6246 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
6247 | if (napi->napi_id == napi_id) | |
6248 | return napi; | |
6249 | ||
6250 | return NULL; | |
6251 | } | |
02d62e86 | 6252 | |
1200097f ED |
6253 | static void skb_defer_free_flush(struct softnet_data *sd) |
6254 | { | |
6255 | struct sk_buff *skb, *next; | |
6256 | ||
6257 | /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ | |
6258 | if (!READ_ONCE(sd->defer_list)) | |
6259 | return; | |
6260 | ||
6261 | spin_lock(&sd->defer_lock); | |
6262 | skb = sd->defer_list; | |
6263 | sd->defer_list = NULL; | |
6264 | sd->defer_count = 0; | |
6265 | spin_unlock(&sd->defer_lock); | |
6266 | ||
6267 | while (skb != NULL) { | |
6268 | next = skb->next; | |
6269 | napi_consume_skb(skb, 1); | |
6270 | skb = next; | |
6271 | } | |
6272 | } | |
6273 | ||
02d62e86 | 6274 | #if defined(CONFIG_NET_RX_BUSY_POLL) |
217f6974 | 6275 | |
7fd3253a | 6276 | static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) |
217f6974 | 6277 | { |
7fd3253a BT |
6278 | if (!skip_schedule) { |
6279 | gro_normal_list(napi); | |
6280 | __napi_schedule(napi); | |
6281 | return; | |
6282 | } | |
217f6974 | 6283 | |
7fd3253a BT |
6284 | if (napi->gro_bitmask) { |
6285 | /* flush too old packets | |
6286 | * If HZ < 1000, flush all packets. | |
6287 | */ | |
6288 | napi_gro_flush(napi, HZ >= 1000); | |
6289 | } | |
217f6974 | 6290 | |
7fd3253a BT |
6291 | gro_normal_list(napi); |
6292 | clear_bit(NAPI_STATE_SCHED, &napi->state); | |
6293 | } | |
6294 | ||
13d381b4 SR |
6295 | enum { |
6296 | NAPI_F_PREFER_BUSY_POLL = 1, | |
b4e8ae5c | 6297 | NAPI_F_END_ON_RESCHED = 2, |
13d381b4 SR |
6298 | }; |
6299 | ||
6300 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, | |
6301 | unsigned flags, u16 budget) | |
217f6974 | 6302 | { |
7fd3253a BT |
6303 | bool skip_schedule = false; |
6304 | unsigned long timeout; | |
217f6974 ED |
6305 | int rc; |
6306 | ||
39e6c820 ED |
6307 | /* Busy polling means there is a high chance device driver hard irq |
6308 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
6309 | * set in napi_schedule_prep(). | |
6310 | * Since we are about to call napi->poll() once more, we can safely | |
6311 | * clear NAPI_STATE_MISSED. | |
6312 | * | |
6313 | * Note: x86 could use a single "lock and ..." instruction | |
6314 | * to perform these two clear_bit() | |
6315 | */ | |
6316 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
6317 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
6318 | ||
6319 | local_bh_disable(); | |
6320 | ||
13d381b4 | 6321 | if (flags & NAPI_F_PREFER_BUSY_POLL) { |
7fd3253a BT |
6322 | napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); |
6323 | timeout = READ_ONCE(napi->dev->gro_flush_timeout); | |
6324 | if (napi->defer_hard_irqs_count && timeout) { | |
6325 | hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); | |
6326 | skip_schedule = true; | |
6327 | } | |
6328 | } | |
6329 | ||
217f6974 ED |
6330 | /* All we really want here is to re-enable device interrupts. |
6331 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
6332 | */ | |
7c951caf | 6333 | rc = napi->poll(napi, budget); |
323ebb61 EC |
6334 | /* We can't gro_normal_list() here, because napi->poll() might have |
6335 | * rearmed the napi (napi_complete_done()) in which case it could | |
6336 | * already be running on another CPU. | |
6337 | */ | |
7c951caf | 6338 | trace_napi_poll(napi, rc, budget); |
217f6974 | 6339 | netpoll_poll_unlock(have_poll_lock); |
7c951caf | 6340 | if (rc == budget) |
7fd3253a | 6341 | __busy_poll_stop(napi, skip_schedule); |
217f6974 | 6342 | local_bh_enable(); |
217f6974 ED |
6343 | } |
6344 | ||
13d381b4 SR |
6345 | static void __napi_busy_loop(unsigned int napi_id, |
6346 | bool (*loop_end)(void *, unsigned long), | |
6347 | void *loop_end_arg, unsigned flags, u16 budget) | |
02d62e86 | 6348 | { |
7db6b048 | 6349 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 6350 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 6351 | void *have_poll_lock = NULL; |
02d62e86 | 6352 | struct napi_struct *napi; |
217f6974 | 6353 | |
13d381b4 SR |
6354 | WARN_ON_ONCE(!rcu_read_lock_held()); |
6355 | ||
217f6974 | 6356 | restart: |
217f6974 | 6357 | napi_poll = NULL; |
02d62e86 | 6358 | |
545cd5e5 | 6359 | napi = napi_by_id(napi_id); |
02d62e86 | 6360 | if (!napi) |
13d381b4 | 6361 | return; |
02d62e86 | 6362 | |
c857946a KK |
6363 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) |
6364 | preempt_disable(); | |
217f6974 | 6365 | for (;;) { |
2b5cd0df AD |
6366 | int work = 0; |
6367 | ||
2a028ecb | 6368 | local_bh_disable(); |
217f6974 ED |
6369 | if (!napi_poll) { |
6370 | unsigned long val = READ_ONCE(napi->state); | |
6371 | ||
6372 | /* If multiple threads are competing for this napi, | |
6373 | * we avoid dirtying napi->state as much as we can. | |
6374 | */ | |
6375 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
7fd3253a | 6376 | NAPIF_STATE_IN_BUSY_POLL)) { |
13d381b4 | 6377 | if (flags & NAPI_F_PREFER_BUSY_POLL) |
7fd3253a | 6378 | set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); |
217f6974 | 6379 | goto count; |
7fd3253a | 6380 | } |
217f6974 ED |
6381 | if (cmpxchg(&napi->state, val, |
6382 | val | NAPIF_STATE_IN_BUSY_POLL | | |
7fd3253a | 6383 | NAPIF_STATE_SCHED) != val) { |
13d381b4 | 6384 | if (flags & NAPI_F_PREFER_BUSY_POLL) |
7fd3253a | 6385 | set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); |
217f6974 | 6386 | goto count; |
7fd3253a | 6387 | } |
217f6974 ED |
6388 | have_poll_lock = netpoll_poll_lock(napi); |
6389 | napi_poll = napi->poll; | |
6390 | } | |
7c951caf BT |
6391 | work = napi_poll(napi, budget); |
6392 | trace_napi_poll(napi, work, budget); | |
323ebb61 | 6393 | gro_normal_list(napi); |
217f6974 | 6394 | count: |
2b5cd0df | 6395 | if (work > 0) |
7db6b048 | 6396 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 6397 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
1200097f | 6398 | skb_defer_free_flush(this_cpu_ptr(&softnet_data)); |
2a028ecb | 6399 | local_bh_enable(); |
02d62e86 | 6400 | |
7db6b048 | 6401 | if (!loop_end || loop_end(loop_end_arg, start_time)) |
217f6974 | 6402 | break; |
02d62e86 | 6403 | |
217f6974 | 6404 | if (unlikely(need_resched())) { |
b4e8ae5c SR |
6405 | if (flags & NAPI_F_END_ON_RESCHED) |
6406 | break; | |
217f6974 | 6407 | if (napi_poll) |
13d381b4 | 6408 | busy_poll_stop(napi, have_poll_lock, flags, budget); |
c857946a KK |
6409 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) |
6410 | preempt_enable(); | |
217f6974 ED |
6411 | rcu_read_unlock(); |
6412 | cond_resched(); | |
13d381b4 | 6413 | rcu_read_lock(); |
7db6b048 | 6414 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 6415 | return; |
217f6974 ED |
6416 | goto restart; |
6417 | } | |
6cdf89b1 | 6418 | cpu_relax(); |
217f6974 ED |
6419 | } |
6420 | if (napi_poll) | |
13d381b4 | 6421 | busy_poll_stop(napi, have_poll_lock, flags, budget); |
c857946a KK |
6422 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) |
6423 | preempt_enable(); | |
13d381b4 SR |
6424 | } |
6425 | ||
b4e8ae5c SR |
6426 | void napi_busy_loop_rcu(unsigned int napi_id, |
6427 | bool (*loop_end)(void *, unsigned long), | |
6428 | void *loop_end_arg, bool prefer_busy_poll, u16 budget) | |
6429 | { | |
6430 | unsigned flags = NAPI_F_END_ON_RESCHED; | |
6431 | ||
6432 | if (prefer_busy_poll) | |
6433 | flags |= NAPI_F_PREFER_BUSY_POLL; | |
6434 | ||
6435 | __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); | |
6436 | } | |
6437 | ||
13d381b4 SR |
6438 | void napi_busy_loop(unsigned int napi_id, |
6439 | bool (*loop_end)(void *, unsigned long), | |
6440 | void *loop_end_arg, bool prefer_busy_poll, u16 budget) | |
6441 | { | |
6442 | unsigned flags = prefer_busy_poll ? NAPI_F_PREFER_BUSY_POLL : 0; | |
6443 | ||
6444 | rcu_read_lock(); | |
6445 | __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); | |
2a028ecb | 6446 | rcu_read_unlock(); |
02d62e86 | 6447 | } |
7db6b048 | 6448 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
6449 | |
6450 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 6451 | |
149d6ad8 | 6452 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 6453 | { |
4d092dd2 | 6454 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) |
52bd2d62 | 6455 | return; |
af12fa6e | 6456 | |
52bd2d62 | 6457 | spin_lock(&napi_hash_lock); |
af12fa6e | 6458 | |
545cd5e5 | 6459 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 6460 | do { |
545cd5e5 AD |
6461 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
6462 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
6463 | } while (napi_by_id(napi_gen_id)); |
6464 | napi->napi_id = napi_gen_id; | |
af12fa6e | 6465 | |
52bd2d62 ED |
6466 | hlist_add_head_rcu(&napi->napi_hash_node, |
6467 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 6468 | |
52bd2d62 | 6469 | spin_unlock(&napi_hash_lock); |
af12fa6e | 6470 | } |
af12fa6e ET |
6471 | |
6472 | /* Warning : caller is responsible to make sure rcu grace period | |
6473 | * is respected before freeing memory containing @napi | |
6474 | */ | |
5198d545 | 6475 | static void napi_hash_del(struct napi_struct *napi) |
af12fa6e ET |
6476 | { |
6477 | spin_lock(&napi_hash_lock); | |
6478 | ||
4d092dd2 | 6479 | hlist_del_init_rcu(&napi->napi_hash_node); |
5198d545 | 6480 | |
af12fa6e ET |
6481 | spin_unlock(&napi_hash_lock); |
6482 | } | |
af12fa6e | 6483 | |
3b47d303 ED |
6484 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
6485 | { | |
6486 | struct napi_struct *napi; | |
6487 | ||
6488 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
6489 | |
6490 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
6491 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
6492 | */ | |
6f8b12d6 | 6493 | if (!napi_disable_pending(napi) && |
7fd3253a BT |
6494 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { |
6495 | clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); | |
39e6c820 | 6496 | __napi_schedule_irqoff(napi); |
7fd3253a | 6497 | } |
3b47d303 ED |
6498 | |
6499 | return HRTIMER_NORESTART; | |
6500 | } | |
6501 | ||
7c4ec749 | 6502 | static void init_gro_hash(struct napi_struct *napi) |
d565b0a1 | 6503 | { |
07d78363 DM |
6504 | int i; |
6505 | ||
6312fe77 LR |
6506 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6507 | INIT_LIST_HEAD(&napi->gro_hash[i].list); | |
6508 | napi->gro_hash[i].count = 0; | |
6509 | } | |
7c4ec749 DM |
6510 | napi->gro_bitmask = 0; |
6511 | } | |
6512 | ||
5fdd2f0e WW |
6513 | int dev_set_threaded(struct net_device *dev, bool threaded) |
6514 | { | |
6515 | struct napi_struct *napi; | |
6516 | int err = 0; | |
6517 | ||
6518 | if (dev->threaded == threaded) | |
6519 | return 0; | |
6520 | ||
6521 | if (threaded) { | |
6522 | list_for_each_entry(napi, &dev->napi_list, dev_list) { | |
6523 | if (!napi->thread) { | |
6524 | err = napi_kthread_create(napi); | |
6525 | if (err) { | |
6526 | threaded = false; | |
6527 | break; | |
6528 | } | |
6529 | } | |
6530 | } | |
6531 | } | |
6532 | ||
c1742dcb | 6533 | WRITE_ONCE(dev->threaded, threaded); |
5fdd2f0e WW |
6534 | |
6535 | /* Make sure kthread is created before THREADED bit | |
6536 | * is set. | |
6537 | */ | |
6538 | smp_mb__before_atomic(); | |
6539 | ||
6540 | /* Setting/unsetting threaded mode on a napi might not immediately | |
6541 | * take effect, if the current napi instance is actively being | |
6542 | * polled. In this case, the switch between threaded mode and | |
6543 | * softirq mode will happen in the next round of napi_schedule(). | |
6544 | * This should not cause hiccups/stalls to the live traffic. | |
6545 | */ | |
274c4a6d AS |
6546 | list_for_each_entry(napi, &dev->napi_list, dev_list) |
6547 | assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); | |
5fdd2f0e WW |
6548 | |
6549 | return err; | |
6550 | } | |
8f64860f | 6551 | EXPORT_SYMBOL(dev_set_threaded); |
5fdd2f0e | 6552 | |
2a502ff0 AN |
6553 | /** |
6554 | * netif_queue_set_napi - Associate queue with the napi | |
6555 | * @dev: device to which NAPI and queue belong | |
6556 | * @queue_index: Index of queue | |
6557 | * @type: queue type as RX or TX | |
6558 | * @napi: NAPI context, pass NULL to clear previously set NAPI | |
6559 | * | |
6560 | * Set queue with its corresponding napi context. This should be done after | |
6561 | * registering the NAPI handler for the queue-vector and the queues have been | |
6562 | * mapped to the corresponding interrupt vector. | |
6563 | */ | |
6564 | void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index, | |
6565 | enum netdev_queue_type type, struct napi_struct *napi) | |
6566 | { | |
6567 | struct netdev_rx_queue *rxq; | |
6568 | struct netdev_queue *txq; | |
6569 | ||
6570 | if (WARN_ON_ONCE(napi && !napi->dev)) | |
6571 | return; | |
6572 | if (dev->reg_state >= NETREG_REGISTERED) | |
6573 | ASSERT_RTNL(); | |
6574 | ||
6575 | switch (type) { | |
6576 | case NETDEV_QUEUE_TYPE_RX: | |
6577 | rxq = __netif_get_rx_queue(dev, queue_index); | |
6578 | rxq->napi = napi; | |
6579 | return; | |
6580 | case NETDEV_QUEUE_TYPE_TX: | |
6581 | txq = netdev_get_tx_queue(dev, queue_index); | |
6582 | txq->napi = napi; | |
6583 | return; | |
6584 | default: | |
6585 | return; | |
6586 | } | |
6587 | } | |
6588 | EXPORT_SYMBOL(netif_queue_set_napi); | |
6589 | ||
58caed3d JK |
6590 | void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, |
6591 | int (*poll)(struct napi_struct *, int), int weight) | |
7c4ec749 | 6592 | { |
4d092dd2 JK |
6593 | if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) |
6594 | return; | |
6595 | ||
7c4ec749 | 6596 | INIT_LIST_HEAD(&napi->poll_list); |
4d092dd2 | 6597 | INIT_HLIST_NODE(&napi->napi_hash_node); |
7c4ec749 DM |
6598 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
6599 | napi->timer.function = napi_watchdog; | |
6600 | init_gro_hash(napi); | |
5d38a079 | 6601 | napi->skb = NULL; |
323ebb61 EC |
6602 | INIT_LIST_HEAD(&napi->rx_list); |
6603 | napi->rx_count = 0; | |
d565b0a1 | 6604 | napi->poll = poll; |
82dc3c63 | 6605 | if (weight > NAPI_POLL_WEIGHT) |
bf29e9e9 QC |
6606 | netdev_err_once(dev, "%s() called with weight %d\n", __func__, |
6607 | weight); | |
d565b0a1 | 6608 | napi->weight = weight; |
d565b0a1 | 6609 | napi->dev = dev; |
5d38a079 | 6610 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
6611 | napi->poll_owner = -1; |
6612 | #endif | |
8c48eea3 | 6613 | napi->list_owner = -1; |
d565b0a1 | 6614 | set_bit(NAPI_STATE_SCHED, &napi->state); |
96e97bc0 JK |
6615 | set_bit(NAPI_STATE_NPSVC, &napi->state); |
6616 | list_add_rcu(&napi->dev_list, &dev->napi_list); | |
93d05d4a | 6617 | napi_hash_add(napi); |
fd9ea57f | 6618 | napi_get_frags_check(napi); |
29863d41 WW |
6619 | /* Create kthread for this napi if dev->threaded is set. |
6620 | * Clear dev->threaded if kthread creation failed so that | |
6621 | * threaded mode will not be enabled in napi_enable(). | |
6622 | */ | |
6623 | if (dev->threaded && napi_kthread_create(napi)) | |
c1742dcb | 6624 | dev->threaded = false; |
26793bfb | 6625 | netif_napi_set_irq(napi, -1); |
d565b0a1 | 6626 | } |
58caed3d | 6627 | EXPORT_SYMBOL(netif_napi_add_weight); |
d565b0a1 | 6628 | |
3b47d303 ED |
6629 | void napi_disable(struct napi_struct *n) |
6630 | { | |
719c5719 JK |
6631 | unsigned long val, new; |
6632 | ||
3b47d303 ED |
6633 | might_sleep(); |
6634 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
6635 | ||
4ffa1d1c ED |
6636 | val = READ_ONCE(n->state); |
6637 | do { | |
fd896e38 | 6638 | while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { |
719c5719 | 6639 | usleep_range(20, 200); |
fd896e38 | 6640 | val = READ_ONCE(n->state); |
719c5719 JK |
6641 | } |
6642 | ||
6643 | new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; | |
6644 | new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); | |
4ffa1d1c | 6645 | } while (!try_cmpxchg(&n->state, &val, new)); |
3b47d303 ED |
6646 | |
6647 | hrtimer_cancel(&n->timer); | |
6648 | ||
6649 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
6650 | } | |
6651 | EXPORT_SYMBOL(napi_disable); | |
6652 | ||
29863d41 WW |
6653 | /** |
6654 | * napi_enable - enable NAPI scheduling | |
6655 | * @n: NAPI context | |
6656 | * | |
6657 | * Resume NAPI from being scheduled on this context. | |
6658 | * Must be paired with napi_disable. | |
6659 | */ | |
6660 | void napi_enable(struct napi_struct *n) | |
6661 | { | |
4ffa1d1c | 6662 | unsigned long new, val = READ_ONCE(n->state); |
3765996e XZ |
6663 | |
6664 | do { | |
3765996e XZ |
6665 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); |
6666 | ||
6667 | new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); | |
6668 | if (n->dev->threaded && n->thread) | |
6669 | new |= NAPIF_STATE_THREADED; | |
4ffa1d1c | 6670 | } while (!try_cmpxchg(&n->state, &val, new)); |
29863d41 WW |
6671 | } |
6672 | EXPORT_SYMBOL(napi_enable); | |
6673 | ||
07d78363 | 6674 | static void flush_gro_hash(struct napi_struct *napi) |
d4546c25 | 6675 | { |
07d78363 | 6676 | int i; |
d4546c25 | 6677 | |
07d78363 DM |
6678 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6679 | struct sk_buff *skb, *n; | |
6680 | ||
6312fe77 | 6681 | list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) |
07d78363 | 6682 | kfree_skb(skb); |
6312fe77 | 6683 | napi->gro_hash[i].count = 0; |
07d78363 | 6684 | } |
d4546c25 DM |
6685 | } |
6686 | ||
93d05d4a | 6687 | /* Must be called in process context */ |
5198d545 | 6688 | void __netif_napi_del(struct napi_struct *napi) |
d565b0a1 | 6689 | { |
4d092dd2 JK |
6690 | if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) |
6691 | return; | |
6692 | ||
5198d545 | 6693 | napi_hash_del(napi); |
5251ef82 | 6694 | list_del_rcu(&napi->dev_list); |
76620aaf | 6695 | napi_free_frags(napi); |
d565b0a1 | 6696 | |
07d78363 | 6697 | flush_gro_hash(napi); |
d9f37d01 | 6698 | napi->gro_bitmask = 0; |
29863d41 WW |
6699 | |
6700 | if (napi->thread) { | |
6701 | kthread_stop(napi->thread); | |
6702 | napi->thread = NULL; | |
6703 | } | |
d565b0a1 | 6704 | } |
5198d545 | 6705 | EXPORT_SYMBOL(__netif_napi_del); |
d565b0a1 | 6706 | |
898f8015 | 6707 | static int __napi_poll(struct napi_struct *n, bool *repoll) |
726ce70e | 6708 | { |
726ce70e HX |
6709 | int work, weight; |
6710 | ||
726ce70e HX |
6711 | weight = n->weight; |
6712 | ||
6713 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
6714 | * with netpoll's poll_napi(). Only the entity which | |
6715 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
6716 | * actually make the ->poll() call. Therefore we avoid | |
6717 | * accidentally calling ->poll() when NAPI is not scheduled. | |
6718 | */ | |
6719 | work = 0; | |
7f3eb217 | 6720 | if (napi_is_scheduled(n)) { |
726ce70e | 6721 | work = n->poll(n, weight); |
1db19db7 | 6722 | trace_napi_poll(n, work, weight); |
9a675ba5 SAS |
6723 | |
6724 | xdp_do_check_flushed(n); | |
726ce70e HX |
6725 | } |
6726 | ||
427d5838 | 6727 | if (unlikely(work > weight)) |
5b92be64 JB |
6728 | netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", |
6729 | n->poll, work, weight); | |
726ce70e HX |
6730 | |
6731 | if (likely(work < weight)) | |
898f8015 | 6732 | return work; |
726ce70e HX |
6733 | |
6734 | /* Drivers must not modify the NAPI state if they | |
6735 | * consume the entire weight. In such cases this code | |
6736 | * still "owns" the NAPI instance and therefore can | |
6737 | * move the instance around on the list at-will. | |
6738 | */ | |
6739 | if (unlikely(napi_disable_pending(n))) { | |
6740 | napi_complete(n); | |
898f8015 | 6741 | return work; |
726ce70e HX |
6742 | } |
6743 | ||
7fd3253a BT |
6744 | /* The NAPI context has more processing work, but busy-polling |
6745 | * is preferred. Exit early. | |
6746 | */ | |
6747 | if (napi_prefer_busy_poll(n)) { | |
6748 | if (napi_complete_done(n, work)) { | |
6749 | /* If timeout is not set, we need to make sure | |
6750 | * that the NAPI is re-scheduled. | |
6751 | */ | |
6752 | napi_schedule(n); | |
6753 | } | |
898f8015 | 6754 | return work; |
7fd3253a BT |
6755 | } |
6756 | ||
d9f37d01 | 6757 | if (n->gro_bitmask) { |
726ce70e HX |
6758 | /* flush too old packets |
6759 | * If HZ < 1000, flush all packets. | |
6760 | */ | |
6761 | napi_gro_flush(n, HZ >= 1000); | |
6762 | } | |
6763 | ||
c8079432 MM |
6764 | gro_normal_list(n); |
6765 | ||
001ce546 HX |
6766 | /* Some drivers may have called napi_schedule |
6767 | * prior to exhausting their budget. | |
6768 | */ | |
6769 | if (unlikely(!list_empty(&n->poll_list))) { | |
6770 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
6771 | n->dev ? n->dev->name : "backlog"); | |
898f8015 | 6772 | return work; |
001ce546 HX |
6773 | } |
6774 | ||
898f8015 FF |
6775 | *repoll = true; |
6776 | ||
6777 | return work; | |
6778 | } | |
6779 | ||
6780 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) | |
6781 | { | |
6782 | bool do_repoll = false; | |
6783 | void *have; | |
6784 | int work; | |
6785 | ||
6786 | list_del_init(&n->poll_list); | |
6787 | ||
6788 | have = netpoll_poll_lock(n); | |
6789 | ||
6790 | work = __napi_poll(n, &do_repoll); | |
6791 | ||
6792 | if (do_repoll) | |
6793 | list_add_tail(&n->poll_list, repoll); | |
726ce70e | 6794 | |
726ce70e HX |
6795 | netpoll_poll_unlock(have); |
6796 | ||
6797 | return work; | |
6798 | } | |
6799 | ||
29863d41 WW |
6800 | static int napi_thread_wait(struct napi_struct *napi) |
6801 | { | |
6802 | set_current_state(TASK_INTERRUPTIBLE); | |
6803 | ||
27f0ad71 | 6804 | while (!kthread_should_stop()) { |
cb038357 WW |
6805 | /* Testing SCHED_THREADED bit here to make sure the current |
6806 | * kthread owns this napi and could poll on this napi. | |
6807 | * Testing SCHED bit is not enough because SCHED bit might be | |
6808 | * set by some other busy poll thread or by napi_disable(). | |
6809 | */ | |
56364c91 | 6810 | if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state)) { |
29863d41 WW |
6811 | WARN_ON(!list_empty(&napi->poll_list)); |
6812 | __set_current_state(TASK_RUNNING); | |
6813 | return 0; | |
6814 | } | |
6815 | ||
6816 | schedule(); | |
6817 | set_current_state(TASK_INTERRUPTIBLE); | |
6818 | } | |
6819 | __set_current_state(TASK_RUNNING); | |
27f0ad71 | 6820 | |
29863d41 WW |
6821 | return -1; |
6822 | } | |
6823 | ||
dad6b977 | 6824 | static void napi_threaded_poll_loop(struct napi_struct *napi) |
29863d41 | 6825 | { |
a1aaee7f | 6826 | struct softnet_data *sd; |
dad6b977 | 6827 | unsigned long last_qs = jiffies; |
d6dbbb11 | 6828 | |
dad6b977 SAS |
6829 | for (;;) { |
6830 | bool repoll = false; | |
6831 | void *have; | |
29863d41 | 6832 | |
dad6b977 SAS |
6833 | local_bh_disable(); |
6834 | sd = this_cpu_ptr(&softnet_data); | |
6835 | sd->in_napi_threaded_poll = true; | |
29863d41 | 6836 | |
dad6b977 SAS |
6837 | have = netpoll_poll_lock(napi); |
6838 | __napi_poll(napi, &repoll); | |
6839 | netpoll_poll_unlock(have); | |
29863d41 | 6840 | |
dad6b977 SAS |
6841 | sd->in_napi_threaded_poll = false; |
6842 | barrier(); | |
87eff2ec | 6843 | |
dad6b977 SAS |
6844 | if (sd_has_rps_ipi_waiting(sd)) { |
6845 | local_irq_disable(); | |
6846 | net_rps_action_and_irq_enable(sd); | |
6847 | } | |
6848 | skb_defer_free_flush(sd); | |
6849 | local_bh_enable(); | |
29863d41 | 6850 | |
dad6b977 SAS |
6851 | if (!repoll) |
6852 | break; | |
29863d41 | 6853 | |
dad6b977 SAS |
6854 | rcu_softirq_qs_periodic(last_qs); |
6855 | cond_resched(); | |
29863d41 | 6856 | } |
dad6b977 SAS |
6857 | } |
6858 | ||
6859 | static int napi_threaded_poll(void *data) | |
6860 | { | |
6861 | struct napi_struct *napi = data; | |
6862 | ||
6863 | while (!napi_thread_wait(napi)) | |
6864 | napi_threaded_poll_loop(napi); | |
6865 | ||
29863d41 WW |
6866 | return 0; |
6867 | } | |
6868 | ||
0766f788 | 6869 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 6870 | { |
903ceff7 | 6871 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e | 6872 | unsigned long time_limit = jiffies + |
ae6e22f7 ED |
6873 | usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs)); |
6874 | int budget = READ_ONCE(net_hotdata.netdev_budget); | |
d75b1ade ED |
6875 | LIST_HEAD(list); |
6876 | LIST_HEAD(repoll); | |
53fb95d3 | 6877 | |
8b43fd3d | 6878 | start: |
c59647c0 | 6879 | sd->in_net_rx_action = true; |
1da177e4 | 6880 | local_irq_disable(); |
d75b1ade ED |
6881 | list_splice_init(&sd->poll_list, &list); |
6882 | local_irq_enable(); | |
1da177e4 | 6883 | |
ceb8d5bf | 6884 | for (;;) { |
bea3348e | 6885 | struct napi_struct *n; |
1da177e4 | 6886 | |
90987650 ED |
6887 | skb_defer_free_flush(sd); |
6888 | ||
ceb8d5bf | 6889 | if (list_empty(&list)) { |
8b43fd3d ED |
6890 | if (list_empty(&repoll)) { |
6891 | sd->in_net_rx_action = false; | |
6892 | barrier(); | |
6893 | /* We need to check if ____napi_schedule() | |
6894 | * had refilled poll_list while | |
6895 | * sd->in_net_rx_action was true. | |
6896 | */ | |
6897 | if (!list_empty(&sd->poll_list)) | |
6898 | goto start; | |
6899 | if (!sd_has_rps_ipi_waiting(sd)) | |
6900 | goto end; | |
6901 | } | |
ceb8d5bf HX |
6902 | break; |
6903 | } | |
6904 | ||
6bd373eb HX |
6905 | n = list_first_entry(&list, struct napi_struct, poll_list); |
6906 | budget -= napi_poll(n, &repoll); | |
6907 | ||
d75b1ade | 6908 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
6909 | * Allow this to run for 2 jiffies since which will allow |
6910 | * an average latency of 1.5/HZ. | |
bea3348e | 6911 | */ |
ceb8d5bf HX |
6912 | if (unlikely(budget <= 0 || |
6913 | time_after_eq(jiffies, time_limit))) { | |
6914 | sd->time_squeeze++; | |
6915 | break; | |
6916 | } | |
1da177e4 | 6917 | } |
d75b1ade | 6918 | |
d75b1ade ED |
6919 | local_irq_disable(); |
6920 | ||
6921 | list_splice_tail_init(&sd->poll_list, &list); | |
6922 | list_splice_tail(&repoll, &list); | |
6923 | list_splice(&list, &sd->poll_list); | |
6924 | if (!list_empty(&sd->poll_list)) | |
6925 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
c59647c0 ED |
6926 | else |
6927 | sd->in_net_rx_action = false; | |
d75b1ade | 6928 | |
e326bed2 | 6929 | net_rps_action_and_irq_enable(sd); |
90987650 | 6930 | end:; |
1da177e4 LT |
6931 | } |
6932 | ||
aa9d8560 | 6933 | struct netdev_adjacent { |
9ff162a8 | 6934 | struct net_device *dev; |
f77159a3 | 6935 | netdevice_tracker dev_tracker; |
5d261913 VF |
6936 | |
6937 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 6938 | bool master; |
5d261913 | 6939 | |
32b6d34f TY |
6940 | /* lookup ignore flag */ |
6941 | bool ignore; | |
6942 | ||
5d261913 VF |
6943 | /* counter for the number of times this device was added to us */ |
6944 | u16 ref_nr; | |
6945 | ||
402dae96 VF |
6946 | /* private field for the users */ |
6947 | void *private; | |
6948 | ||
9ff162a8 JP |
6949 | struct list_head list; |
6950 | struct rcu_head rcu; | |
9ff162a8 JP |
6951 | }; |
6952 | ||
6ea29da1 | 6953 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 6954 | struct list_head *adj_list) |
9ff162a8 | 6955 | { |
5d261913 | 6956 | struct netdev_adjacent *adj; |
5d261913 | 6957 | |
2f268f12 | 6958 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
6959 | if (adj->dev == adj_dev) |
6960 | return adj; | |
9ff162a8 JP |
6961 | } |
6962 | return NULL; | |
6963 | } | |
6964 | ||
eff74233 TY |
6965 | static int ____netdev_has_upper_dev(struct net_device *upper_dev, |
6966 | struct netdev_nested_priv *priv) | |
f1170fd4 | 6967 | { |
eff74233 | 6968 | struct net_device *dev = (struct net_device *)priv->data; |
f1170fd4 DA |
6969 | |
6970 | return upper_dev == dev; | |
6971 | } | |
6972 | ||
9ff162a8 JP |
6973 | /** |
6974 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
6975 | * @dev: device | |
6976 | * @upper_dev: upper device to check | |
6977 | * | |
6978 | * Find out if a device is linked to specified upper device and return true | |
6979 | * in case it is. Note that this checks only immediate upper device, | |
6980 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
6981 | */ | |
6982 | bool netdev_has_upper_dev(struct net_device *dev, | |
6983 | struct net_device *upper_dev) | |
6984 | { | |
eff74233 TY |
6985 | struct netdev_nested_priv priv = { |
6986 | .data = (void *)upper_dev, | |
6987 | }; | |
6988 | ||
9ff162a8 JP |
6989 | ASSERT_RTNL(); |
6990 | ||
32b6d34f | 6991 | return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, |
eff74233 | 6992 | &priv); |
9ff162a8 JP |
6993 | } |
6994 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
6995 | ||
1a3f060c | 6996 | /** |
c1639be9 | 6997 | * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device |
1a3f060c DA |
6998 | * @dev: device |
6999 | * @upper_dev: upper device to check | |
7000 | * | |
7001 | * Find out if a device is linked to specified upper device and return true | |
7002 | * in case it is. Note that this checks the entire upper device chain. | |
7003 | * The caller must hold rcu lock. | |
7004 | */ | |
7005 | ||
1a3f060c DA |
7006 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
7007 | struct net_device *upper_dev) | |
7008 | { | |
eff74233 TY |
7009 | struct netdev_nested_priv priv = { |
7010 | .data = (void *)upper_dev, | |
7011 | }; | |
7012 | ||
32b6d34f | 7013 | return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, |
eff74233 | 7014 | &priv); |
1a3f060c DA |
7015 | } |
7016 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
7017 | ||
9ff162a8 JP |
7018 | /** |
7019 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
7020 | * @dev: device | |
7021 | * | |
7022 | * Find out if a device is linked to an upper device and return true in case | |
7023 | * it is. The caller must hold the RTNL lock. | |
7024 | */ | |
25cc72a3 | 7025 | bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
7026 | { |
7027 | ASSERT_RTNL(); | |
7028 | ||
f1170fd4 | 7029 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 7030 | } |
25cc72a3 | 7031 | EXPORT_SYMBOL(netdev_has_any_upper_dev); |
9ff162a8 JP |
7032 | |
7033 | /** | |
7034 | * netdev_master_upper_dev_get - Get master upper device | |
7035 | * @dev: device | |
7036 | * | |
7037 | * Find a master upper device and return pointer to it or NULL in case | |
7038 | * it's not there. The caller must hold the RTNL lock. | |
7039 | */ | |
7040 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
7041 | { | |
aa9d8560 | 7042 | struct netdev_adjacent *upper; |
9ff162a8 JP |
7043 | |
7044 | ASSERT_RTNL(); | |
7045 | ||
2f268f12 | 7046 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
7047 | return NULL; |
7048 | ||
2f268f12 | 7049 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 7050 | struct netdev_adjacent, list); |
9ff162a8 JP |
7051 | if (likely(upper->master)) |
7052 | return upper->dev; | |
7053 | return NULL; | |
7054 | } | |
7055 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
7056 | ||
32b6d34f TY |
7057 | static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) |
7058 | { | |
7059 | struct netdev_adjacent *upper; | |
7060 | ||
7061 | ASSERT_RTNL(); | |
7062 | ||
7063 | if (list_empty(&dev->adj_list.upper)) | |
7064 | return NULL; | |
7065 | ||
7066 | upper = list_first_entry(&dev->adj_list.upper, | |
7067 | struct netdev_adjacent, list); | |
7068 | if (likely(upper->master) && !upper->ignore) | |
7069 | return upper->dev; | |
7070 | return NULL; | |
7071 | } | |
7072 | ||
0f524a80 DA |
7073 | /** |
7074 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
7075 | * @dev: device | |
7076 | * | |
7077 | * Find out if a device is linked to a lower device and return true in case | |
7078 | * it is. The caller must hold the RTNL lock. | |
7079 | */ | |
7080 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
7081 | { | |
7082 | ASSERT_RTNL(); | |
7083 | ||
7084 | return !list_empty(&dev->adj_list.lower); | |
7085 | } | |
7086 | ||
b6ccba4c VF |
7087 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
7088 | { | |
7089 | struct netdev_adjacent *adj; | |
7090 | ||
7091 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
7092 | ||
7093 | return adj->private; | |
7094 | } | |
7095 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
7096 | ||
44a40855 VY |
7097 | /** |
7098 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
7099 | * @dev: device | |
7100 | * @iter: list_head ** of the current position | |
7101 | * | |
7102 | * Gets the next device from the dev's upper list, starting from iter | |
7103 | * position. The caller must hold RCU read lock. | |
7104 | */ | |
7105 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
7106 | struct list_head **iter) | |
7107 | { | |
7108 | struct netdev_adjacent *upper; | |
7109 | ||
7110 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
7111 | ||
7112 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7113 | ||
7114 | if (&upper->list == &dev->adj_list.upper) | |
7115 | return NULL; | |
7116 | ||
7117 | *iter = &upper->list; | |
7118 | ||
7119 | return upper->dev; | |
7120 | } | |
7121 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
7122 | ||
32b6d34f TY |
7123 | static struct net_device *__netdev_next_upper_dev(struct net_device *dev, |
7124 | struct list_head **iter, | |
7125 | bool *ignore) | |
5343da4c TY |
7126 | { |
7127 | struct netdev_adjacent *upper; | |
7128 | ||
7129 | upper = list_entry((*iter)->next, struct netdev_adjacent, list); | |
7130 | ||
7131 | if (&upper->list == &dev->adj_list.upper) | |
7132 | return NULL; | |
7133 | ||
7134 | *iter = &upper->list; | |
32b6d34f | 7135 | *ignore = upper->ignore; |
5343da4c TY |
7136 | |
7137 | return upper->dev; | |
7138 | } | |
7139 | ||
1a3f060c DA |
7140 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
7141 | struct list_head **iter) | |
7142 | { | |
7143 | struct netdev_adjacent *upper; | |
7144 | ||
7145 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
7146 | ||
7147 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7148 | ||
7149 | if (&upper->list == &dev->adj_list.upper) | |
7150 | return NULL; | |
7151 | ||
7152 | *iter = &upper->list; | |
7153 | ||
7154 | return upper->dev; | |
7155 | } | |
7156 | ||
32b6d34f TY |
7157 | static int __netdev_walk_all_upper_dev(struct net_device *dev, |
7158 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7159 | struct netdev_nested_priv *priv), |
7160 | struct netdev_nested_priv *priv) | |
5343da4c TY |
7161 | { |
7162 | struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7163 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7164 | int ret, cur = 0; | |
32b6d34f | 7165 | bool ignore; |
5343da4c TY |
7166 | |
7167 | now = dev; | |
7168 | iter = &dev->adj_list.upper; | |
7169 | ||
7170 | while (1) { | |
7171 | if (now != dev) { | |
eff74233 | 7172 | ret = fn(now, priv); |
5343da4c TY |
7173 | if (ret) |
7174 | return ret; | |
7175 | } | |
7176 | ||
7177 | next = NULL; | |
7178 | while (1) { | |
32b6d34f | 7179 | udev = __netdev_next_upper_dev(now, &iter, &ignore); |
5343da4c TY |
7180 | if (!udev) |
7181 | break; | |
32b6d34f TY |
7182 | if (ignore) |
7183 | continue; | |
5343da4c TY |
7184 | |
7185 | next = udev; | |
7186 | niter = &udev->adj_list.upper; | |
7187 | dev_stack[cur] = now; | |
7188 | iter_stack[cur++] = iter; | |
7189 | break; | |
7190 | } | |
7191 | ||
7192 | if (!next) { | |
7193 | if (!cur) | |
7194 | return 0; | |
7195 | next = dev_stack[--cur]; | |
7196 | niter = iter_stack[cur]; | |
7197 | } | |
7198 | ||
7199 | now = next; | |
7200 | iter = niter; | |
7201 | } | |
7202 | ||
7203 | return 0; | |
7204 | } | |
7205 | ||
1a3f060c DA |
7206 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, |
7207 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7208 | struct netdev_nested_priv *priv), |
7209 | struct netdev_nested_priv *priv) | |
1a3f060c | 7210 | { |
5343da4c TY |
7211 | struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; |
7212 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7213 | int ret, cur = 0; | |
1a3f060c | 7214 | |
5343da4c TY |
7215 | now = dev; |
7216 | iter = &dev->adj_list.upper; | |
1a3f060c | 7217 | |
5343da4c TY |
7218 | while (1) { |
7219 | if (now != dev) { | |
eff74233 | 7220 | ret = fn(now, priv); |
5343da4c TY |
7221 | if (ret) |
7222 | return ret; | |
7223 | } | |
7224 | ||
7225 | next = NULL; | |
7226 | while (1) { | |
7227 | udev = netdev_next_upper_dev_rcu(now, &iter); | |
7228 | if (!udev) | |
7229 | break; | |
7230 | ||
7231 | next = udev; | |
7232 | niter = &udev->adj_list.upper; | |
7233 | dev_stack[cur] = now; | |
7234 | iter_stack[cur++] = iter; | |
7235 | break; | |
7236 | } | |
7237 | ||
7238 | if (!next) { | |
7239 | if (!cur) | |
7240 | return 0; | |
7241 | next = dev_stack[--cur]; | |
7242 | niter = iter_stack[cur]; | |
7243 | } | |
7244 | ||
7245 | now = next; | |
7246 | iter = niter; | |
1a3f060c DA |
7247 | } |
7248 | ||
7249 | return 0; | |
7250 | } | |
7251 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
7252 | ||
32b6d34f TY |
7253 | static bool __netdev_has_upper_dev(struct net_device *dev, |
7254 | struct net_device *upper_dev) | |
7255 | { | |
eff74233 | 7256 | struct netdev_nested_priv priv = { |
1fc70edb | 7257 | .flags = 0, |
eff74233 TY |
7258 | .data = (void *)upper_dev, |
7259 | }; | |
7260 | ||
32b6d34f TY |
7261 | ASSERT_RTNL(); |
7262 | ||
7263 | return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, | |
eff74233 | 7264 | &priv); |
32b6d34f TY |
7265 | } |
7266 | ||
31088a11 VF |
7267 | /** |
7268 | * netdev_lower_get_next_private - Get the next ->private from the | |
7269 | * lower neighbour list | |
7270 | * @dev: device | |
7271 | * @iter: list_head ** of the current position | |
7272 | * | |
7273 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
7274 | * list, starting from iter position. The caller must hold either hold the | |
7275 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 7276 | * list will remain unchanged. |
31088a11 VF |
7277 | */ |
7278 | void *netdev_lower_get_next_private(struct net_device *dev, | |
7279 | struct list_head **iter) | |
7280 | { | |
7281 | struct netdev_adjacent *lower; | |
7282 | ||
7283 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
7284 | ||
7285 | if (&lower->list == &dev->adj_list.lower) | |
7286 | return NULL; | |
7287 | ||
6859e7df | 7288 | *iter = lower->list.next; |
31088a11 VF |
7289 | |
7290 | return lower->private; | |
7291 | } | |
7292 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
7293 | ||
7294 | /** | |
7295 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
7296 | * lower neighbour list, RCU | |
7297 | * variant | |
7298 | * @dev: device | |
7299 | * @iter: list_head ** of the current position | |
7300 | * | |
7301 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
7302 | * list, starting from iter position. The caller must hold RCU read lock. | |
7303 | */ | |
7304 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
7305 | struct list_head **iter) | |
7306 | { | |
7307 | struct netdev_adjacent *lower; | |
7308 | ||
68918669 | 7309 | WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); |
31088a11 VF |
7310 | |
7311 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7312 | ||
7313 | if (&lower->list == &dev->adj_list.lower) | |
7314 | return NULL; | |
7315 | ||
6859e7df | 7316 | *iter = &lower->list; |
31088a11 VF |
7317 | |
7318 | return lower->private; | |
7319 | } | |
7320 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
7321 | ||
4085ebe8 VY |
7322 | /** |
7323 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
7324 | * list | |
7325 | * @dev: device | |
7326 | * @iter: list_head ** of the current position | |
7327 | * | |
7328 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
7329 | * list, starting from iter position. The caller must hold RTNL lock or | |
7330 | * its own locking that guarantees that the neighbour lower | |
b469139e | 7331 | * list will remain unchanged. |
4085ebe8 VY |
7332 | */ |
7333 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
7334 | { | |
7335 | struct netdev_adjacent *lower; | |
7336 | ||
cfdd28be | 7337 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
7338 | |
7339 | if (&lower->list == &dev->adj_list.lower) | |
7340 | return NULL; | |
7341 | ||
cfdd28be | 7342 | *iter = lower->list.next; |
4085ebe8 VY |
7343 | |
7344 | return lower->dev; | |
7345 | } | |
7346 | EXPORT_SYMBOL(netdev_lower_get_next); | |
7347 | ||
1a3f060c DA |
7348 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
7349 | struct list_head **iter) | |
7350 | { | |
7351 | struct netdev_adjacent *lower; | |
7352 | ||
46b5ab1a | 7353 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
7354 | |
7355 | if (&lower->list == &dev->adj_list.lower) | |
7356 | return NULL; | |
7357 | ||
46b5ab1a | 7358 | *iter = &lower->list; |
1a3f060c DA |
7359 | |
7360 | return lower->dev; | |
7361 | } | |
7362 | ||
32b6d34f TY |
7363 | static struct net_device *__netdev_next_lower_dev(struct net_device *dev, |
7364 | struct list_head **iter, | |
7365 | bool *ignore) | |
7366 | { | |
7367 | struct netdev_adjacent *lower; | |
7368 | ||
7369 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
7370 | ||
7371 | if (&lower->list == &dev->adj_list.lower) | |
7372 | return NULL; | |
7373 | ||
7374 | *iter = &lower->list; | |
7375 | *ignore = lower->ignore; | |
7376 | ||
7377 | return lower->dev; | |
7378 | } | |
7379 | ||
1a3f060c DA |
7380 | int netdev_walk_all_lower_dev(struct net_device *dev, |
7381 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7382 | struct netdev_nested_priv *priv), |
7383 | struct netdev_nested_priv *priv) | |
1a3f060c | 7384 | { |
5343da4c TY |
7385 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; |
7386 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7387 | int ret, cur = 0; | |
1a3f060c | 7388 | |
5343da4c TY |
7389 | now = dev; |
7390 | iter = &dev->adj_list.lower; | |
1a3f060c | 7391 | |
5343da4c TY |
7392 | while (1) { |
7393 | if (now != dev) { | |
eff74233 | 7394 | ret = fn(now, priv); |
5343da4c TY |
7395 | if (ret) |
7396 | return ret; | |
7397 | } | |
7398 | ||
7399 | next = NULL; | |
7400 | while (1) { | |
7401 | ldev = netdev_next_lower_dev(now, &iter); | |
7402 | if (!ldev) | |
7403 | break; | |
7404 | ||
7405 | next = ldev; | |
7406 | niter = &ldev->adj_list.lower; | |
7407 | dev_stack[cur] = now; | |
7408 | iter_stack[cur++] = iter; | |
7409 | break; | |
7410 | } | |
7411 | ||
7412 | if (!next) { | |
7413 | if (!cur) | |
7414 | return 0; | |
7415 | next = dev_stack[--cur]; | |
7416 | niter = iter_stack[cur]; | |
7417 | } | |
7418 | ||
7419 | now = next; | |
7420 | iter = niter; | |
1a3f060c DA |
7421 | } |
7422 | ||
7423 | return 0; | |
7424 | } | |
7425 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
7426 | ||
32b6d34f TY |
7427 | static int __netdev_walk_all_lower_dev(struct net_device *dev, |
7428 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7429 | struct netdev_nested_priv *priv), |
7430 | struct netdev_nested_priv *priv) | |
32b6d34f TY |
7431 | { |
7432 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7433 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7434 | int ret, cur = 0; | |
7435 | bool ignore; | |
7436 | ||
7437 | now = dev; | |
7438 | iter = &dev->adj_list.lower; | |
7439 | ||
7440 | while (1) { | |
7441 | if (now != dev) { | |
eff74233 | 7442 | ret = fn(now, priv); |
32b6d34f TY |
7443 | if (ret) |
7444 | return ret; | |
7445 | } | |
7446 | ||
7447 | next = NULL; | |
7448 | while (1) { | |
7449 | ldev = __netdev_next_lower_dev(now, &iter, &ignore); | |
7450 | if (!ldev) | |
7451 | break; | |
7452 | if (ignore) | |
7453 | continue; | |
7454 | ||
7455 | next = ldev; | |
7456 | niter = &ldev->adj_list.lower; | |
7457 | dev_stack[cur] = now; | |
7458 | iter_stack[cur++] = iter; | |
7459 | break; | |
7460 | } | |
7461 | ||
7462 | if (!next) { | |
7463 | if (!cur) | |
7464 | return 0; | |
7465 | next = dev_stack[--cur]; | |
7466 | niter = iter_stack[cur]; | |
7467 | } | |
7468 | ||
7469 | now = next; | |
7470 | iter = niter; | |
7471 | } | |
7472 | ||
7473 | return 0; | |
7474 | } | |
7475 | ||
7151affe TY |
7476 | struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
7477 | struct list_head **iter) | |
1a3f060c DA |
7478 | { |
7479 | struct netdev_adjacent *lower; | |
7480 | ||
7481 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7482 | if (&lower->list == &dev->adj_list.lower) | |
7483 | return NULL; | |
7484 | ||
7485 | *iter = &lower->list; | |
7486 | ||
7487 | return lower->dev; | |
7488 | } | |
7151affe | 7489 | EXPORT_SYMBOL(netdev_next_lower_dev_rcu); |
1a3f060c | 7490 | |
5343da4c TY |
7491 | static u8 __netdev_upper_depth(struct net_device *dev) |
7492 | { | |
7493 | struct net_device *udev; | |
7494 | struct list_head *iter; | |
7495 | u8 max_depth = 0; | |
32b6d34f | 7496 | bool ignore; |
5343da4c TY |
7497 | |
7498 | for (iter = &dev->adj_list.upper, | |
32b6d34f | 7499 | udev = __netdev_next_upper_dev(dev, &iter, &ignore); |
5343da4c | 7500 | udev; |
32b6d34f TY |
7501 | udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { |
7502 | if (ignore) | |
7503 | continue; | |
5343da4c TY |
7504 | if (max_depth < udev->upper_level) |
7505 | max_depth = udev->upper_level; | |
7506 | } | |
7507 | ||
7508 | return max_depth; | |
7509 | } | |
7510 | ||
7511 | static u8 __netdev_lower_depth(struct net_device *dev) | |
1a3f060c DA |
7512 | { |
7513 | struct net_device *ldev; | |
7514 | struct list_head *iter; | |
5343da4c | 7515 | u8 max_depth = 0; |
32b6d34f | 7516 | bool ignore; |
1a3f060c DA |
7517 | |
7518 | for (iter = &dev->adj_list.lower, | |
32b6d34f | 7519 | ldev = __netdev_next_lower_dev(dev, &iter, &ignore); |
1a3f060c | 7520 | ldev; |
32b6d34f TY |
7521 | ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { |
7522 | if (ignore) | |
7523 | continue; | |
5343da4c TY |
7524 | if (max_depth < ldev->lower_level) |
7525 | max_depth = ldev->lower_level; | |
7526 | } | |
1a3f060c | 7527 | |
5343da4c TY |
7528 | return max_depth; |
7529 | } | |
7530 | ||
eff74233 TY |
7531 | static int __netdev_update_upper_level(struct net_device *dev, |
7532 | struct netdev_nested_priv *__unused) | |
5343da4c TY |
7533 | { |
7534 | dev->upper_level = __netdev_upper_depth(dev) + 1; | |
7535 | return 0; | |
7536 | } | |
7537 | ||
f32404ae JB |
7538 | #ifdef CONFIG_LOCKDEP |
7539 | static LIST_HEAD(net_unlink_list); | |
7540 | ||
7541 | static void net_unlink_todo(struct net_device *dev) | |
7542 | { | |
7543 | if (list_empty(&dev->unlink_list)) | |
7544 | list_add_tail(&dev->unlink_list, &net_unlink_list); | |
7545 | } | |
7546 | #endif | |
7547 | ||
eff74233 | 7548 | static int __netdev_update_lower_level(struct net_device *dev, |
1fc70edb | 7549 | struct netdev_nested_priv *priv) |
5343da4c TY |
7550 | { |
7551 | dev->lower_level = __netdev_lower_depth(dev) + 1; | |
1fc70edb TY |
7552 | |
7553 | #ifdef CONFIG_LOCKDEP | |
7554 | if (!priv) | |
7555 | return 0; | |
7556 | ||
7557 | if (priv->flags & NESTED_SYNC_IMM) | |
7558 | dev->nested_level = dev->lower_level - 1; | |
7559 | if (priv->flags & NESTED_SYNC_TODO) | |
7560 | net_unlink_todo(dev); | |
7561 | #endif | |
5343da4c TY |
7562 | return 0; |
7563 | } | |
7564 | ||
7565 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
7566 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7567 | struct netdev_nested_priv *priv), |
7568 | struct netdev_nested_priv *priv) | |
5343da4c TY |
7569 | { |
7570 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7571 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7572 | int ret, cur = 0; | |
7573 | ||
7574 | now = dev; | |
7575 | iter = &dev->adj_list.lower; | |
7576 | ||
7577 | while (1) { | |
7578 | if (now != dev) { | |
eff74233 | 7579 | ret = fn(now, priv); |
5343da4c TY |
7580 | if (ret) |
7581 | return ret; | |
7582 | } | |
7583 | ||
7584 | next = NULL; | |
7585 | while (1) { | |
7586 | ldev = netdev_next_lower_dev_rcu(now, &iter); | |
7587 | if (!ldev) | |
7588 | break; | |
7589 | ||
7590 | next = ldev; | |
7591 | niter = &ldev->adj_list.lower; | |
7592 | dev_stack[cur] = now; | |
7593 | iter_stack[cur++] = iter; | |
7594 | break; | |
7595 | } | |
7596 | ||
7597 | if (!next) { | |
7598 | if (!cur) | |
7599 | return 0; | |
7600 | next = dev_stack[--cur]; | |
7601 | niter = iter_stack[cur]; | |
7602 | } | |
7603 | ||
7604 | now = next; | |
7605 | iter = niter; | |
1a3f060c DA |
7606 | } |
7607 | ||
7608 | return 0; | |
7609 | } | |
7610 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
7611 | ||
e001bfad | 7612 | /** |
7613 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
7614 | * lower neighbour list, RCU | |
7615 | * variant | |
7616 | * @dev: device | |
7617 | * | |
7618 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
7619 | * list. The caller must hold RCU read lock. | |
7620 | */ | |
7621 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
7622 | { | |
7623 | struct netdev_adjacent *lower; | |
7624 | ||
7625 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
7626 | struct netdev_adjacent, list); | |
7627 | if (lower) | |
7628 | return lower->private; | |
7629 | return NULL; | |
7630 | } | |
7631 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
7632 | ||
9ff162a8 JP |
7633 | /** |
7634 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
7635 | * @dev: device | |
7636 | * | |
7637 | * Find a master upper device and return pointer to it or NULL in case | |
7638 | * it's not there. The caller must hold the RCU read lock. | |
7639 | */ | |
7640 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
7641 | { | |
aa9d8560 | 7642 | struct netdev_adjacent *upper; |
9ff162a8 | 7643 | |
2f268f12 | 7644 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 7645 | struct netdev_adjacent, list); |
9ff162a8 JP |
7646 | if (upper && likely(upper->master)) |
7647 | return upper->dev; | |
7648 | return NULL; | |
7649 | } | |
7650 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
7651 | ||
0a59f3a9 | 7652 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
7653 | struct net_device *adj_dev, |
7654 | struct list_head *dev_list) | |
7655 | { | |
7656 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 7657 | |
3ee32707 VF |
7658 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
7659 | "upper_%s" : "lower_%s", adj_dev->name); | |
7660 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
7661 | linkname); | |
7662 | } | |
0a59f3a9 | 7663 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
7664 | char *name, |
7665 | struct list_head *dev_list) | |
7666 | { | |
7667 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 7668 | |
3ee32707 VF |
7669 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
7670 | "upper_%s" : "lower_%s", name); | |
7671 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
7672 | } | |
7673 | ||
7ce64c79 AF |
7674 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
7675 | struct net_device *adj_dev, | |
7676 | struct list_head *dev_list) | |
7677 | { | |
7678 | return (dev_list == &dev->adj_list.upper || | |
7679 | dev_list == &dev->adj_list.lower) && | |
7680 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
7681 | } | |
3ee32707 | 7682 | |
5d261913 VF |
7683 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
7684 | struct net_device *adj_dev, | |
7863c054 | 7685 | struct list_head *dev_list, |
402dae96 | 7686 | void *private, bool master) |
5d261913 VF |
7687 | { |
7688 | struct netdev_adjacent *adj; | |
842d67a7 | 7689 | int ret; |
5d261913 | 7690 | |
6ea29da1 | 7691 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
7692 | |
7693 | if (adj) { | |
790510d9 | 7694 | adj->ref_nr += 1; |
67b62f98 DA |
7695 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
7696 | dev->name, adj_dev->name, adj->ref_nr); | |
7697 | ||
5d261913 VF |
7698 | return 0; |
7699 | } | |
7700 | ||
7701 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
7702 | if (!adj) | |
7703 | return -ENOMEM; | |
7704 | ||
7705 | adj->dev = adj_dev; | |
7706 | adj->master = master; | |
790510d9 | 7707 | adj->ref_nr = 1; |
402dae96 | 7708 | adj->private = private; |
32b6d34f | 7709 | adj->ignore = false; |
d62607c3 | 7710 | netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); |
2f268f12 | 7711 | |
67b62f98 DA |
7712 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
7713 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 7714 | |
7ce64c79 | 7715 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 7716 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
7717 | if (ret) |
7718 | goto free_adj; | |
7719 | } | |
7720 | ||
7863c054 | 7721 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
7722 | if (master) { |
7723 | ret = sysfs_create_link(&(dev->dev.kobj), | |
7724 | &(adj_dev->dev.kobj), "master"); | |
7725 | if (ret) | |
5831d66e | 7726 | goto remove_symlinks; |
842d67a7 | 7727 | |
7863c054 | 7728 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 7729 | } else { |
7863c054 | 7730 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 7731 | } |
5d261913 VF |
7732 | |
7733 | return 0; | |
842d67a7 | 7734 | |
5831d66e | 7735 | remove_symlinks: |
7ce64c79 | 7736 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 7737 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 | 7738 | free_adj: |
d62607c3 | 7739 | netdev_put(adj_dev, &adj->dev_tracker); |
842d67a7 VF |
7740 | kfree(adj); |
7741 | ||
7742 | return ret; | |
5d261913 VF |
7743 | } |
7744 | ||
1d143d9f | 7745 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
7746 | struct net_device *adj_dev, | |
93409033 | 7747 | u16 ref_nr, |
1d143d9f | 7748 | struct list_head *dev_list) |
5d261913 VF |
7749 | { |
7750 | struct netdev_adjacent *adj; | |
7751 | ||
67b62f98 DA |
7752 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
7753 | dev->name, adj_dev->name, ref_nr); | |
7754 | ||
6ea29da1 | 7755 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 7756 | |
2f268f12 | 7757 | if (!adj) { |
67b62f98 | 7758 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 7759 | dev->name, adj_dev->name); |
67b62f98 DA |
7760 | WARN_ON(1); |
7761 | return; | |
2f268f12 | 7762 | } |
5d261913 | 7763 | |
93409033 | 7764 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
7765 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
7766 | dev->name, adj_dev->name, ref_nr, | |
7767 | adj->ref_nr - ref_nr); | |
93409033 | 7768 | adj->ref_nr -= ref_nr; |
5d261913 VF |
7769 | return; |
7770 | } | |
7771 | ||
842d67a7 VF |
7772 | if (adj->master) |
7773 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
7774 | ||
7ce64c79 | 7775 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 7776 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 7777 | |
5d261913 | 7778 | list_del_rcu(&adj->list); |
67b62f98 | 7779 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 7780 | adj_dev->name, dev->name, adj_dev->name); |
d62607c3 | 7781 | netdev_put(adj_dev, &adj->dev_tracker); |
5d261913 VF |
7782 | kfree_rcu(adj, rcu); |
7783 | } | |
7784 | ||
1d143d9f | 7785 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
7786 | struct net_device *upper_dev, | |
7787 | struct list_head *up_list, | |
7788 | struct list_head *down_list, | |
7789 | void *private, bool master) | |
5d261913 VF |
7790 | { |
7791 | int ret; | |
7792 | ||
790510d9 | 7793 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 7794 | private, master); |
5d261913 VF |
7795 | if (ret) |
7796 | return ret; | |
7797 | ||
790510d9 | 7798 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 7799 | private, false); |
5d261913 | 7800 | if (ret) { |
790510d9 | 7801 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
7802 | return ret; |
7803 | } | |
7804 | ||
7805 | return 0; | |
7806 | } | |
7807 | ||
1d143d9f | 7808 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
7809 | struct net_device *upper_dev, | |
93409033 | 7810 | u16 ref_nr, |
1d143d9f | 7811 | struct list_head *up_list, |
7812 | struct list_head *down_list) | |
5d261913 | 7813 | { |
93409033 AC |
7814 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
7815 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
7816 | } |
7817 | ||
1d143d9f | 7818 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
7819 | struct net_device *upper_dev, | |
7820 | void *private, bool master) | |
2f268f12 | 7821 | { |
f1170fd4 DA |
7822 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
7823 | &dev->adj_list.upper, | |
7824 | &upper_dev->adj_list.lower, | |
7825 | private, master); | |
5d261913 VF |
7826 | } |
7827 | ||
1d143d9f | 7828 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
7829 | struct net_device *upper_dev) | |
2f268f12 | 7830 | { |
93409033 | 7831 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
7832 | &dev->adj_list.upper, |
7833 | &upper_dev->adj_list.lower); | |
7834 | } | |
5d261913 | 7835 | |
9ff162a8 | 7836 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 7837 | struct net_device *upper_dev, bool master, |
42ab19ee | 7838 | void *upper_priv, void *upper_info, |
1fc70edb | 7839 | struct netdev_nested_priv *priv, |
42ab19ee | 7840 | struct netlink_ext_ack *extack) |
9ff162a8 | 7841 | { |
51d0c047 DA |
7842 | struct netdev_notifier_changeupper_info changeupper_info = { |
7843 | .info = { | |
7844 | .dev = dev, | |
42ab19ee | 7845 | .extack = extack, |
51d0c047 DA |
7846 | }, |
7847 | .upper_dev = upper_dev, | |
7848 | .master = master, | |
7849 | .linking = true, | |
7850 | .upper_info = upper_info, | |
7851 | }; | |
50d629e7 | 7852 | struct net_device *master_dev; |
5d261913 | 7853 | int ret = 0; |
9ff162a8 JP |
7854 | |
7855 | ASSERT_RTNL(); | |
7856 | ||
7857 | if (dev == upper_dev) | |
7858 | return -EBUSY; | |
7859 | ||
7860 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
32b6d34f | 7861 | if (__netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
7862 | return -EBUSY; |
7863 | ||
5343da4c TY |
7864 | if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) |
7865 | return -EMLINK; | |
7866 | ||
50d629e7 | 7867 | if (!master) { |
32b6d34f | 7868 | if (__netdev_has_upper_dev(dev, upper_dev)) |
50d629e7 MM |
7869 | return -EEXIST; |
7870 | } else { | |
32b6d34f | 7871 | master_dev = __netdev_master_upper_dev_get(dev); |
50d629e7 MM |
7872 | if (master_dev) |
7873 | return master_dev == upper_dev ? -EEXIST : -EBUSY; | |
7874 | } | |
9ff162a8 | 7875 | |
51d0c047 | 7876 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
7877 | &changeupper_info.info); |
7878 | ret = notifier_to_errno(ret); | |
7879 | if (ret) | |
7880 | return ret; | |
7881 | ||
6dffb044 | 7882 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 7883 | master); |
5d261913 VF |
7884 | if (ret) |
7885 | return ret; | |
9ff162a8 | 7886 | |
51d0c047 | 7887 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
b03804e7 IS |
7888 | &changeupper_info.info); |
7889 | ret = notifier_to_errno(ret); | |
7890 | if (ret) | |
f1170fd4 | 7891 | goto rollback; |
b03804e7 | 7892 | |
5343da4c | 7893 | __netdev_update_upper_level(dev, NULL); |
32b6d34f | 7894 | __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); |
5343da4c | 7895 | |
1fc70edb | 7896 | __netdev_update_lower_level(upper_dev, priv); |
32b6d34f | 7897 | __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, |
1fc70edb | 7898 | priv); |
5343da4c | 7899 | |
9ff162a8 | 7900 | return 0; |
5d261913 | 7901 | |
f1170fd4 | 7902 | rollback: |
2f268f12 | 7903 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
7904 | |
7905 | return ret; | |
9ff162a8 JP |
7906 | } |
7907 | ||
7908 | /** | |
7909 | * netdev_upper_dev_link - Add a link to the upper device | |
7910 | * @dev: device | |
7911 | * @upper_dev: new upper device | |
7a006d59 | 7912 | * @extack: netlink extended ack |
9ff162a8 JP |
7913 | * |
7914 | * Adds a link to device which is upper to this one. The caller must hold | |
7915 | * the RTNL lock. On a failure a negative errno code is returned. | |
7916 | * On success the reference counts are adjusted and the function | |
7917 | * returns zero. | |
7918 | */ | |
7919 | int netdev_upper_dev_link(struct net_device *dev, | |
42ab19ee DA |
7920 | struct net_device *upper_dev, |
7921 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7922 | { |
1fc70edb TY |
7923 | struct netdev_nested_priv priv = { |
7924 | .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, | |
7925 | .data = NULL, | |
7926 | }; | |
7927 | ||
42ab19ee | 7928 | return __netdev_upper_dev_link(dev, upper_dev, false, |
1fc70edb | 7929 | NULL, NULL, &priv, extack); |
9ff162a8 JP |
7930 | } |
7931 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
7932 | ||
7933 | /** | |
7934 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
7935 | * @dev: device | |
7936 | * @upper_dev: new upper device | |
6dffb044 | 7937 | * @upper_priv: upper device private |
29bf24af | 7938 | * @upper_info: upper info to be passed down via notifier |
7a006d59 | 7939 | * @extack: netlink extended ack |
9ff162a8 JP |
7940 | * |
7941 | * Adds a link to device which is upper to this one. In this case, only | |
7942 | * one master upper device can be linked, although other non-master devices | |
7943 | * might be linked as well. The caller must hold the RTNL lock. | |
7944 | * On a failure a negative errno code is returned. On success the reference | |
7945 | * counts are adjusted and the function returns zero. | |
7946 | */ | |
7947 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 7948 | struct net_device *upper_dev, |
42ab19ee DA |
7949 | void *upper_priv, void *upper_info, |
7950 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7951 | { |
1fc70edb TY |
7952 | struct netdev_nested_priv priv = { |
7953 | .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, | |
7954 | .data = NULL, | |
7955 | }; | |
7956 | ||
29bf24af | 7957 | return __netdev_upper_dev_link(dev, upper_dev, true, |
1fc70edb | 7958 | upper_priv, upper_info, &priv, extack); |
9ff162a8 JP |
7959 | } |
7960 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
7961 | ||
fe8300fd | 7962 | static void __netdev_upper_dev_unlink(struct net_device *dev, |
1fc70edb TY |
7963 | struct net_device *upper_dev, |
7964 | struct netdev_nested_priv *priv) | |
9ff162a8 | 7965 | { |
51d0c047 DA |
7966 | struct netdev_notifier_changeupper_info changeupper_info = { |
7967 | .info = { | |
7968 | .dev = dev, | |
7969 | }, | |
7970 | .upper_dev = upper_dev, | |
7971 | .linking = false, | |
7972 | }; | |
f4563a75 | 7973 | |
9ff162a8 JP |
7974 | ASSERT_RTNL(); |
7975 | ||
0e4ead9d | 7976 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; |
0e4ead9d | 7977 | |
51d0c047 | 7978 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
7979 | &changeupper_info.info); |
7980 | ||
2f268f12 | 7981 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 7982 | |
51d0c047 | 7983 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
0e4ead9d | 7984 | &changeupper_info.info); |
5343da4c TY |
7985 | |
7986 | __netdev_update_upper_level(dev, NULL); | |
32b6d34f | 7987 | __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); |
5343da4c | 7988 | |
1fc70edb | 7989 | __netdev_update_lower_level(upper_dev, priv); |
32b6d34f | 7990 | __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, |
1fc70edb | 7991 | priv); |
9ff162a8 | 7992 | } |
fe8300fd TY |
7993 | |
7994 | /** | |
7995 | * netdev_upper_dev_unlink - Removes a link to upper device | |
7996 | * @dev: device | |
7997 | * @upper_dev: new upper device | |
7998 | * | |
7999 | * Removes a link to device which is upper to this one. The caller must hold | |
8000 | * the RTNL lock. | |
8001 | */ | |
8002 | void netdev_upper_dev_unlink(struct net_device *dev, | |
8003 | struct net_device *upper_dev) | |
8004 | { | |
1fc70edb TY |
8005 | struct netdev_nested_priv priv = { |
8006 | .flags = NESTED_SYNC_TODO, | |
8007 | .data = NULL, | |
8008 | }; | |
8009 | ||
8010 | __netdev_upper_dev_unlink(dev, upper_dev, &priv); | |
9ff162a8 JP |
8011 | } |
8012 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
8013 | ||
32b6d34f TY |
8014 | static void __netdev_adjacent_dev_set(struct net_device *upper_dev, |
8015 | struct net_device *lower_dev, | |
8016 | bool val) | |
8017 | { | |
8018 | struct netdev_adjacent *adj; | |
8019 | ||
8020 | adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); | |
8021 | if (adj) | |
8022 | adj->ignore = val; | |
8023 | ||
8024 | adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); | |
8025 | if (adj) | |
8026 | adj->ignore = val; | |
8027 | } | |
8028 | ||
8029 | static void netdev_adjacent_dev_disable(struct net_device *upper_dev, | |
8030 | struct net_device *lower_dev) | |
8031 | { | |
8032 | __netdev_adjacent_dev_set(upper_dev, lower_dev, true); | |
8033 | } | |
8034 | ||
8035 | static void netdev_adjacent_dev_enable(struct net_device *upper_dev, | |
8036 | struct net_device *lower_dev) | |
8037 | { | |
8038 | __netdev_adjacent_dev_set(upper_dev, lower_dev, false); | |
8039 | } | |
8040 | ||
8041 | int netdev_adjacent_change_prepare(struct net_device *old_dev, | |
8042 | struct net_device *new_dev, | |
8043 | struct net_device *dev, | |
8044 | struct netlink_ext_ack *extack) | |
8045 | { | |
1fc70edb TY |
8046 | struct netdev_nested_priv priv = { |
8047 | .flags = 0, | |
8048 | .data = NULL, | |
8049 | }; | |
32b6d34f TY |
8050 | int err; |
8051 | ||
8052 | if (!new_dev) | |
8053 | return 0; | |
8054 | ||
8055 | if (old_dev && new_dev != old_dev) | |
8056 | netdev_adjacent_dev_disable(dev, old_dev); | |
1fc70edb TY |
8057 | err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, |
8058 | extack); | |
32b6d34f TY |
8059 | if (err) { |
8060 | if (old_dev && new_dev != old_dev) | |
8061 | netdev_adjacent_dev_enable(dev, old_dev); | |
8062 | return err; | |
8063 | } | |
8064 | ||
8065 | return 0; | |
8066 | } | |
8067 | EXPORT_SYMBOL(netdev_adjacent_change_prepare); | |
8068 | ||
8069 | void netdev_adjacent_change_commit(struct net_device *old_dev, | |
8070 | struct net_device *new_dev, | |
8071 | struct net_device *dev) | |
8072 | { | |
1fc70edb TY |
8073 | struct netdev_nested_priv priv = { |
8074 | .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, | |
8075 | .data = NULL, | |
8076 | }; | |
8077 | ||
32b6d34f TY |
8078 | if (!new_dev || !old_dev) |
8079 | return; | |
8080 | ||
8081 | if (new_dev == old_dev) | |
8082 | return; | |
8083 | ||
8084 | netdev_adjacent_dev_enable(dev, old_dev); | |
1fc70edb | 8085 | __netdev_upper_dev_unlink(old_dev, dev, &priv); |
32b6d34f TY |
8086 | } |
8087 | EXPORT_SYMBOL(netdev_adjacent_change_commit); | |
8088 | ||
8089 | void netdev_adjacent_change_abort(struct net_device *old_dev, | |
8090 | struct net_device *new_dev, | |
8091 | struct net_device *dev) | |
8092 | { | |
1fc70edb TY |
8093 | struct netdev_nested_priv priv = { |
8094 | .flags = 0, | |
8095 | .data = NULL, | |
8096 | }; | |
8097 | ||
32b6d34f TY |
8098 | if (!new_dev) |
8099 | return; | |
8100 | ||
8101 | if (old_dev && new_dev != old_dev) | |
8102 | netdev_adjacent_dev_enable(dev, old_dev); | |
8103 | ||
1fc70edb | 8104 | __netdev_upper_dev_unlink(new_dev, dev, &priv); |
32b6d34f TY |
8105 | } |
8106 | EXPORT_SYMBOL(netdev_adjacent_change_abort); | |
8107 | ||
61bd3857 MS |
8108 | /** |
8109 | * netdev_bonding_info_change - Dispatch event about slave change | |
8110 | * @dev: device | |
4a26e453 | 8111 | * @bonding_info: info to dispatch |
61bd3857 MS |
8112 | * |
8113 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
8114 | * The caller must hold the RTNL lock. | |
8115 | */ | |
8116 | void netdev_bonding_info_change(struct net_device *dev, | |
8117 | struct netdev_bonding_info *bonding_info) | |
8118 | { | |
51d0c047 DA |
8119 | struct netdev_notifier_bonding_info info = { |
8120 | .info.dev = dev, | |
8121 | }; | |
61bd3857 MS |
8122 | |
8123 | memcpy(&info.bonding_info, bonding_info, | |
8124 | sizeof(struct netdev_bonding_info)); | |
51d0c047 | 8125 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, |
61bd3857 MS |
8126 | &info.info); |
8127 | } | |
8128 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
8129 | ||
9309f97a PM |
8130 | static int netdev_offload_xstats_enable_l3(struct net_device *dev, |
8131 | struct netlink_ext_ack *extack) | |
8132 | { | |
8133 | struct netdev_notifier_offload_xstats_info info = { | |
8134 | .info.dev = dev, | |
8135 | .info.extack = extack, | |
8136 | .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, | |
8137 | }; | |
8138 | int err; | |
8139 | int rc; | |
8140 | ||
8141 | dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), | |
8142 | GFP_KERNEL); | |
8143 | if (!dev->offload_xstats_l3) | |
8144 | return -ENOMEM; | |
8145 | ||
8146 | rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, | |
8147 | NETDEV_OFFLOAD_XSTATS_DISABLE, | |
8148 | &info.info); | |
8149 | err = notifier_to_errno(rc); | |
8150 | if (err) | |
8151 | goto free_stats; | |
8152 | ||
8153 | return 0; | |
8154 | ||
8155 | free_stats: | |
8156 | kfree(dev->offload_xstats_l3); | |
8157 | dev->offload_xstats_l3 = NULL; | |
8158 | return err; | |
8159 | } | |
8160 | ||
8161 | int netdev_offload_xstats_enable(struct net_device *dev, | |
8162 | enum netdev_offload_xstats_type type, | |
8163 | struct netlink_ext_ack *extack) | |
8164 | { | |
8165 | ASSERT_RTNL(); | |
8166 | ||
8167 | if (netdev_offload_xstats_enabled(dev, type)) | |
8168 | return -EALREADY; | |
8169 | ||
8170 | switch (type) { | |
8171 | case NETDEV_OFFLOAD_XSTATS_TYPE_L3: | |
8172 | return netdev_offload_xstats_enable_l3(dev, extack); | |
8173 | } | |
8174 | ||
8175 | WARN_ON(1); | |
8176 | return -EINVAL; | |
8177 | } | |
8178 | EXPORT_SYMBOL(netdev_offload_xstats_enable); | |
8179 | ||
8180 | static void netdev_offload_xstats_disable_l3(struct net_device *dev) | |
8181 | { | |
8182 | struct netdev_notifier_offload_xstats_info info = { | |
8183 | .info.dev = dev, | |
8184 | .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, | |
8185 | }; | |
8186 | ||
8187 | call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, | |
8188 | &info.info); | |
8189 | kfree(dev->offload_xstats_l3); | |
8190 | dev->offload_xstats_l3 = NULL; | |
8191 | } | |
8192 | ||
8193 | int netdev_offload_xstats_disable(struct net_device *dev, | |
8194 | enum netdev_offload_xstats_type type) | |
8195 | { | |
8196 | ASSERT_RTNL(); | |
8197 | ||
8198 | if (!netdev_offload_xstats_enabled(dev, type)) | |
8199 | return -EALREADY; | |
8200 | ||
8201 | switch (type) { | |
8202 | case NETDEV_OFFLOAD_XSTATS_TYPE_L3: | |
8203 | netdev_offload_xstats_disable_l3(dev); | |
8204 | return 0; | |
8205 | } | |
8206 | ||
8207 | WARN_ON(1); | |
8208 | return -EINVAL; | |
8209 | } | |
8210 | EXPORT_SYMBOL(netdev_offload_xstats_disable); | |
8211 | ||
8212 | static void netdev_offload_xstats_disable_all(struct net_device *dev) | |
8213 | { | |
8214 | netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); | |
8215 | } | |
8216 | ||
8217 | static struct rtnl_hw_stats64 * | |
8218 | netdev_offload_xstats_get_ptr(const struct net_device *dev, | |
8219 | enum netdev_offload_xstats_type type) | |
8220 | { | |
8221 | switch (type) { | |
8222 | case NETDEV_OFFLOAD_XSTATS_TYPE_L3: | |
8223 | return dev->offload_xstats_l3; | |
8224 | } | |
8225 | ||
8226 | WARN_ON(1); | |
8227 | return NULL; | |
8228 | } | |
8229 | ||
8230 | bool netdev_offload_xstats_enabled(const struct net_device *dev, | |
8231 | enum netdev_offload_xstats_type type) | |
8232 | { | |
8233 | ASSERT_RTNL(); | |
8234 | ||
8235 | return netdev_offload_xstats_get_ptr(dev, type); | |
8236 | } | |
8237 | EXPORT_SYMBOL(netdev_offload_xstats_enabled); | |
8238 | ||
8239 | struct netdev_notifier_offload_xstats_ru { | |
8240 | bool used; | |
8241 | }; | |
8242 | ||
8243 | struct netdev_notifier_offload_xstats_rd { | |
8244 | struct rtnl_hw_stats64 stats; | |
8245 | bool used; | |
8246 | }; | |
8247 | ||
8248 | static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, | |
8249 | const struct rtnl_hw_stats64 *src) | |
8250 | { | |
8251 | dest->rx_packets += src->rx_packets; | |
8252 | dest->tx_packets += src->tx_packets; | |
8253 | dest->rx_bytes += src->rx_bytes; | |
8254 | dest->tx_bytes += src->tx_bytes; | |
8255 | dest->rx_errors += src->rx_errors; | |
8256 | dest->tx_errors += src->tx_errors; | |
8257 | dest->rx_dropped += src->rx_dropped; | |
8258 | dest->tx_dropped += src->tx_dropped; | |
8259 | dest->multicast += src->multicast; | |
8260 | } | |
8261 | ||
8262 | static int netdev_offload_xstats_get_used(struct net_device *dev, | |
8263 | enum netdev_offload_xstats_type type, | |
8264 | bool *p_used, | |
8265 | struct netlink_ext_ack *extack) | |
8266 | { | |
8267 | struct netdev_notifier_offload_xstats_ru report_used = {}; | |
8268 | struct netdev_notifier_offload_xstats_info info = { | |
8269 | .info.dev = dev, | |
8270 | .info.extack = extack, | |
8271 | .type = type, | |
8272 | .report_used = &report_used, | |
8273 | }; | |
8274 | int rc; | |
8275 | ||
8276 | WARN_ON(!netdev_offload_xstats_enabled(dev, type)); | |
8277 | rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, | |
8278 | &info.info); | |
8279 | *p_used = report_used.used; | |
8280 | return notifier_to_errno(rc); | |
8281 | } | |
8282 | ||
8283 | static int netdev_offload_xstats_get_stats(struct net_device *dev, | |
8284 | enum netdev_offload_xstats_type type, | |
8285 | struct rtnl_hw_stats64 *p_stats, | |
8286 | bool *p_used, | |
8287 | struct netlink_ext_ack *extack) | |
8288 | { | |
8289 | struct netdev_notifier_offload_xstats_rd report_delta = {}; | |
8290 | struct netdev_notifier_offload_xstats_info info = { | |
8291 | .info.dev = dev, | |
8292 | .info.extack = extack, | |
8293 | .type = type, | |
8294 | .report_delta = &report_delta, | |
8295 | }; | |
8296 | struct rtnl_hw_stats64 *stats; | |
8297 | int rc; | |
8298 | ||
8299 | stats = netdev_offload_xstats_get_ptr(dev, type); | |
8300 | if (WARN_ON(!stats)) | |
8301 | return -EINVAL; | |
8302 | ||
8303 | rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, | |
8304 | &info.info); | |
8305 | ||
8306 | /* Cache whatever we got, even if there was an error, otherwise the | |
8307 | * successful stats retrievals would get lost. | |
8308 | */ | |
8309 | netdev_hw_stats64_add(stats, &report_delta.stats); | |
8310 | ||
8311 | if (p_stats) | |
8312 | *p_stats = *stats; | |
8313 | *p_used = report_delta.used; | |
8314 | ||
8315 | return notifier_to_errno(rc); | |
8316 | } | |
8317 | ||
8318 | int netdev_offload_xstats_get(struct net_device *dev, | |
8319 | enum netdev_offload_xstats_type type, | |
8320 | struct rtnl_hw_stats64 *p_stats, bool *p_used, | |
8321 | struct netlink_ext_ack *extack) | |
8322 | { | |
8323 | ASSERT_RTNL(); | |
8324 | ||
8325 | if (p_stats) | |
8326 | return netdev_offload_xstats_get_stats(dev, type, p_stats, | |
8327 | p_used, extack); | |
8328 | else | |
8329 | return netdev_offload_xstats_get_used(dev, type, p_used, | |
8330 | extack); | |
8331 | } | |
8332 | EXPORT_SYMBOL(netdev_offload_xstats_get); | |
8333 | ||
8334 | void | |
8335 | netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, | |
8336 | const struct rtnl_hw_stats64 *stats) | |
8337 | { | |
8338 | report_delta->used = true; | |
8339 | netdev_hw_stats64_add(&report_delta->stats, stats); | |
8340 | } | |
8341 | EXPORT_SYMBOL(netdev_offload_xstats_report_delta); | |
8342 | ||
8343 | void | |
8344 | netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) | |
8345 | { | |
8346 | report_used->used = true; | |
8347 | } | |
8348 | EXPORT_SYMBOL(netdev_offload_xstats_report_used); | |
8349 | ||
8350 | void netdev_offload_xstats_push_delta(struct net_device *dev, | |
8351 | enum netdev_offload_xstats_type type, | |
8352 | const struct rtnl_hw_stats64 *p_stats) | |
8353 | { | |
8354 | struct rtnl_hw_stats64 *stats; | |
8355 | ||
8356 | ASSERT_RTNL(); | |
8357 | ||
8358 | stats = netdev_offload_xstats_get_ptr(dev, type); | |
8359 | if (WARN_ON(!stats)) | |
8360 | return; | |
8361 | ||
8362 | netdev_hw_stats64_add(stats, p_stats); | |
8363 | } | |
8364 | EXPORT_SYMBOL(netdev_offload_xstats_push_delta); | |
8365 | ||
cff9f12b MG |
8366 | /** |
8367 | * netdev_get_xmit_slave - Get the xmit slave of master device | |
8842500d | 8368 | * @dev: device |
cff9f12b MG |
8369 | * @skb: The packet |
8370 | * @all_slaves: assume all the slaves are active | |
8371 | * | |
8372 | * The reference counters are not incremented so the caller must be | |
8373 | * careful with locks. The caller must hold RCU lock. | |
8374 | * %NULL is returned if no slave is found. | |
8375 | */ | |
8376 | ||
8377 | struct net_device *netdev_get_xmit_slave(struct net_device *dev, | |
8378 | struct sk_buff *skb, | |
8379 | bool all_slaves) | |
8380 | { | |
8381 | const struct net_device_ops *ops = dev->netdev_ops; | |
8382 | ||
8383 | if (!ops->ndo_get_xmit_slave) | |
8384 | return NULL; | |
8385 | return ops->ndo_get_xmit_slave(dev, skb, all_slaves); | |
8386 | } | |
8387 | EXPORT_SYMBOL(netdev_get_xmit_slave); | |
8388 | ||
719a402c TT |
8389 | static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, |
8390 | struct sock *sk) | |
8391 | { | |
8392 | const struct net_device_ops *ops = dev->netdev_ops; | |
8393 | ||
8394 | if (!ops->ndo_sk_get_lower_dev) | |
8395 | return NULL; | |
8396 | return ops->ndo_sk_get_lower_dev(dev, sk); | |
8397 | } | |
8398 | ||
8399 | /** | |
8400 | * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket | |
8401 | * @dev: device | |
8402 | * @sk: the socket | |
8403 | * | |
8404 | * %NULL is returned if no lower device is found. | |
8405 | */ | |
8406 | ||
8407 | struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, | |
8408 | struct sock *sk) | |
8409 | { | |
8410 | struct net_device *lower; | |
8411 | ||
8412 | lower = netdev_sk_get_lower_dev(dev, sk); | |
8413 | while (lower) { | |
8414 | dev = lower; | |
8415 | lower = netdev_sk_get_lower_dev(dev, sk); | |
8416 | } | |
8417 | ||
8418 | return dev; | |
8419 | } | |
8420 | EXPORT_SYMBOL(netdev_sk_get_lowest_dev); | |
8421 | ||
2ce1ee17 | 8422 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
8423 | { |
8424 | struct netdev_adjacent *iter; | |
8425 | ||
8426 | struct net *net = dev_net(dev); | |
8427 | ||
8428 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 8429 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
8430 | continue; |
8431 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
8432 | &iter->dev->adj_list.lower); | |
8433 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
8434 | &dev->adj_list.upper); | |
8435 | } | |
8436 | ||
8437 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 8438 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
8439 | continue; |
8440 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
8441 | &iter->dev->adj_list.upper); | |
8442 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
8443 | &dev->adj_list.lower); | |
8444 | } | |
8445 | } | |
8446 | ||
2ce1ee17 | 8447 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
8448 | { |
8449 | struct netdev_adjacent *iter; | |
8450 | ||
8451 | struct net *net = dev_net(dev); | |
8452 | ||
8453 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 8454 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
8455 | continue; |
8456 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
8457 | &iter->dev->adj_list.lower); | |
8458 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
8459 | &dev->adj_list.upper); | |
8460 | } | |
8461 | ||
8462 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 8463 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
8464 | continue; |
8465 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
8466 | &iter->dev->adj_list.upper); | |
8467 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
8468 | &dev->adj_list.lower); | |
8469 | } | |
8470 | } | |
8471 | ||
5bb025fa | 8472 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 8473 | { |
5bb025fa | 8474 | struct netdev_adjacent *iter; |
402dae96 | 8475 | |
4c75431a AF |
8476 | struct net *net = dev_net(dev); |
8477 | ||
5bb025fa | 8478 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 8479 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 8480 | continue; |
5bb025fa VF |
8481 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
8482 | &iter->dev->adj_list.lower); | |
8483 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
8484 | &iter->dev->adj_list.lower); | |
8485 | } | |
402dae96 | 8486 | |
5bb025fa | 8487 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 8488 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 8489 | continue; |
5bb025fa VF |
8490 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
8491 | &iter->dev->adj_list.upper); | |
8492 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
8493 | &iter->dev->adj_list.upper); | |
8494 | } | |
402dae96 | 8495 | } |
402dae96 VF |
8496 | |
8497 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
8498 | struct net_device *lower_dev) | |
8499 | { | |
8500 | struct netdev_adjacent *lower; | |
8501 | ||
8502 | if (!lower_dev) | |
8503 | return NULL; | |
6ea29da1 | 8504 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
8505 | if (!lower) |
8506 | return NULL; | |
8507 | ||
8508 | return lower->private; | |
8509 | } | |
8510 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
8511 | ||
4085ebe8 | 8512 | |
04d48266 | 8513 | /** |
c1639be9 | 8514 | * netdev_lower_state_changed - Dispatch event about lower device state change |
04d48266 JP |
8515 | * @lower_dev: device |
8516 | * @lower_state_info: state to dispatch | |
8517 | * | |
8518 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
8519 | * The caller must hold the RTNL lock. | |
8520 | */ | |
8521 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
8522 | void *lower_state_info) | |
8523 | { | |
51d0c047 DA |
8524 | struct netdev_notifier_changelowerstate_info changelowerstate_info = { |
8525 | .info.dev = lower_dev, | |
8526 | }; | |
04d48266 JP |
8527 | |
8528 | ASSERT_RTNL(); | |
8529 | changelowerstate_info.lower_state_info = lower_state_info; | |
51d0c047 | 8530 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, |
04d48266 JP |
8531 | &changelowerstate_info.info); |
8532 | } | |
8533 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
8534 | ||
b6c40d68 PM |
8535 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
8536 | { | |
d314774c SH |
8537 | const struct net_device_ops *ops = dev->netdev_ops; |
8538 | ||
d2615bf4 | 8539 | if (ops->ndo_change_rx_flags) |
d314774c | 8540 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
8541 | } |
8542 | ||
991fb3f7 | 8543 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 8544 | { |
b536db93 | 8545 | unsigned int old_flags = dev->flags; |
55a2c86c | 8546 | unsigned int promiscuity, flags; |
d04a48b0 EB |
8547 | kuid_t uid; |
8548 | kgid_t gid; | |
1da177e4 | 8549 | |
24023451 PM |
8550 | ASSERT_RTNL(); |
8551 | ||
55a2c86c ED |
8552 | promiscuity = dev->promiscuity + inc; |
8553 | if (promiscuity == 0) { | |
dad9b335 WC |
8554 | /* |
8555 | * Avoid overflow. | |
8556 | * If inc causes overflow, untouch promisc and return error. | |
8557 | */ | |
55a2c86c | 8558 | if (unlikely(inc > 0)) { |
5b92be64 | 8559 | netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); |
dad9b335 WC |
8560 | return -EOVERFLOW; |
8561 | } | |
55a2c86c ED |
8562 | flags = old_flags & ~IFF_PROMISC; |
8563 | } else { | |
8564 | flags = old_flags | IFF_PROMISC; | |
dad9b335 | 8565 | } |
55a2c86c ED |
8566 | WRITE_ONCE(dev->promiscuity, promiscuity); |
8567 | if (flags != old_flags) { | |
8568 | WRITE_ONCE(dev->flags, flags); | |
3ba0bf47 JB |
8569 | netdev_info(dev, "%s promiscuous mode\n", |
8570 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
8571 | if (audit_enabled) { |
8572 | current_uid_gid(&uid, &gid); | |
cdfb6b34 RGB |
8573 | audit_log(audit_context(), GFP_ATOMIC, |
8574 | AUDIT_ANOM_PROMISCUOUS, | |
8575 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
8576 | dev->name, (dev->flags & IFF_PROMISC), | |
8577 | (old_flags & IFF_PROMISC), | |
8578 | from_kuid(&init_user_ns, audit_get_loginuid(current)), | |
8579 | from_kuid(&init_user_ns, uid), | |
8580 | from_kgid(&init_user_ns, gid), | |
8581 | audit_get_sessionid(current)); | |
8192b0c4 | 8582 | } |
24023451 | 8583 | |
b6c40d68 | 8584 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 8585 | } |
991fb3f7 | 8586 | if (notify) |
1d997f10 | 8587 | __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); |
dad9b335 | 8588 | return 0; |
1da177e4 LT |
8589 | } |
8590 | ||
4417da66 PM |
8591 | /** |
8592 | * dev_set_promiscuity - update promiscuity count on a device | |
8593 | * @dev: device | |
8594 | * @inc: modifier | |
8595 | * | |
8596 | * Add or remove promiscuity from a device. While the count in the device | |
8597 | * remains above zero the interface remains promiscuous. Once it hits zero | |
8598 | * the device reverts back to normal filtering operation. A negative inc | |
8599 | * value is used to drop promiscuity on the device. | |
dad9b335 | 8600 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 8601 | */ |
dad9b335 | 8602 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 8603 | { |
b536db93 | 8604 | unsigned int old_flags = dev->flags; |
dad9b335 | 8605 | int err; |
4417da66 | 8606 | |
991fb3f7 | 8607 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 8608 | if (err < 0) |
dad9b335 | 8609 | return err; |
4417da66 PM |
8610 | if (dev->flags != old_flags) |
8611 | dev_set_rx_mode(dev); | |
dad9b335 | 8612 | return err; |
4417da66 | 8613 | } |
d1b19dff | 8614 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 8615 | |
991fb3f7 | 8616 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 8617 | { |
991fb3f7 | 8618 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
55a2c86c | 8619 | unsigned int allmulti, flags; |
1da177e4 | 8620 | |
24023451 PM |
8621 | ASSERT_RTNL(); |
8622 | ||
55a2c86c ED |
8623 | allmulti = dev->allmulti + inc; |
8624 | if (allmulti == 0) { | |
dad9b335 WC |
8625 | /* |
8626 | * Avoid overflow. | |
8627 | * If inc causes overflow, untouch allmulti and return error. | |
8628 | */ | |
55a2c86c | 8629 | if (unlikely(inc > 0)) { |
5b92be64 | 8630 | netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); |
dad9b335 WC |
8631 | return -EOVERFLOW; |
8632 | } | |
55a2c86c ED |
8633 | flags = old_flags & ~IFF_ALLMULTI; |
8634 | } else { | |
8635 | flags = old_flags | IFF_ALLMULTI; | |
dad9b335 | 8636 | } |
55a2c86c ED |
8637 | WRITE_ONCE(dev->allmulti, allmulti); |
8638 | if (flags != old_flags) { | |
8639 | WRITE_ONCE(dev->flags, flags); | |
802dcbd6 JB |
8640 | netdev_info(dev, "%s allmulticast mode\n", |
8641 | dev->flags & IFF_ALLMULTI ? "entered" : "left"); | |
b6c40d68 | 8642 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 8643 | dev_set_rx_mode(dev); |
991fb3f7 ND |
8644 | if (notify) |
8645 | __dev_notify_flags(dev, old_flags, | |
1d997f10 | 8646 | dev->gflags ^ old_gflags, 0, NULL); |
24023451 | 8647 | } |
dad9b335 | 8648 | return 0; |
4417da66 | 8649 | } |
991fb3f7 ND |
8650 | |
8651 | /** | |
8652 | * dev_set_allmulti - update allmulti count on a device | |
8653 | * @dev: device | |
8654 | * @inc: modifier | |
8655 | * | |
8656 | * Add or remove reception of all multicast frames to a device. While the | |
8657 | * count in the device remains above zero the interface remains listening | |
8658 | * to all interfaces. Once it hits zero the device reverts back to normal | |
8659 | * filtering operation. A negative @inc value is used to drop the counter | |
8660 | * when releasing a resource needing all multicasts. | |
8661 | * Return 0 if successful or a negative errno code on error. | |
8662 | */ | |
8663 | ||
8664 | int dev_set_allmulti(struct net_device *dev, int inc) | |
8665 | { | |
8666 | return __dev_set_allmulti(dev, inc, true); | |
8667 | } | |
d1b19dff | 8668 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
8669 | |
8670 | /* | |
8671 | * Upload unicast and multicast address lists to device and | |
8672 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 8673 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
8674 | * are present. |
8675 | */ | |
8676 | void __dev_set_rx_mode(struct net_device *dev) | |
8677 | { | |
d314774c SH |
8678 | const struct net_device_ops *ops = dev->netdev_ops; |
8679 | ||
4417da66 PM |
8680 | /* dev_open will call this function so the list will stay sane. */ |
8681 | if (!(dev->flags&IFF_UP)) | |
8682 | return; | |
8683 | ||
8684 | if (!netif_device_present(dev)) | |
40b77c94 | 8685 | return; |
4417da66 | 8686 | |
01789349 | 8687 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
8688 | /* Unicast addresses changes may only happen under the rtnl, |
8689 | * therefore calling __dev_set_promiscuity here is safe. | |
8690 | */ | |
32e7bfc4 | 8691 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 8692 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 8693 | dev->uc_promisc = true; |
32e7bfc4 | 8694 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 8695 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 8696 | dev->uc_promisc = false; |
4417da66 | 8697 | } |
4417da66 | 8698 | } |
01789349 JP |
8699 | |
8700 | if (ops->ndo_set_rx_mode) | |
8701 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
8702 | } |
8703 | ||
8704 | void dev_set_rx_mode(struct net_device *dev) | |
8705 | { | |
b9e40857 | 8706 | netif_addr_lock_bh(dev); |
4417da66 | 8707 | __dev_set_rx_mode(dev); |
b9e40857 | 8708 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
8709 | } |
8710 | ||
f0db275a SH |
8711 | /** |
8712 | * dev_get_flags - get flags reported to userspace | |
8713 | * @dev: device | |
8714 | * | |
8715 | * Get the combination of flag bits exported through APIs to userspace. | |
8716 | */ | |
95c96174 | 8717 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 8718 | { |
95c96174 | 8719 | unsigned int flags; |
1da177e4 | 8720 | |
8afc7a78 | 8721 | flags = (READ_ONCE(dev->flags) & ~(IFF_PROMISC | |
1da177e4 | 8722 | IFF_ALLMULTI | |
b00055aa SR |
8723 | IFF_RUNNING | |
8724 | IFF_LOWER_UP | | |
8725 | IFF_DORMANT)) | | |
8afc7a78 | 8726 | (READ_ONCE(dev->gflags) & (IFF_PROMISC | |
1da177e4 LT |
8727 | IFF_ALLMULTI)); |
8728 | ||
b00055aa SR |
8729 | if (netif_running(dev)) { |
8730 | if (netif_oper_up(dev)) | |
8731 | flags |= IFF_RUNNING; | |
8732 | if (netif_carrier_ok(dev)) | |
8733 | flags |= IFF_LOWER_UP; | |
8734 | if (netif_dormant(dev)) | |
8735 | flags |= IFF_DORMANT; | |
8736 | } | |
1da177e4 LT |
8737 | |
8738 | return flags; | |
8739 | } | |
d1b19dff | 8740 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 8741 | |
6d040321 PM |
8742 | int __dev_change_flags(struct net_device *dev, unsigned int flags, |
8743 | struct netlink_ext_ack *extack) | |
1da177e4 | 8744 | { |
b536db93 | 8745 | unsigned int old_flags = dev->flags; |
bd380811 | 8746 | int ret; |
1da177e4 | 8747 | |
24023451 PM |
8748 | ASSERT_RTNL(); |
8749 | ||
1da177e4 LT |
8750 | /* |
8751 | * Set the flags on our device. | |
8752 | */ | |
8753 | ||
8754 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
8755 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
8756 | IFF_AUTOMEDIA)) | | |
8757 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
8758 | IFF_ALLMULTI)); | |
8759 | ||
8760 | /* | |
8761 | * Load in the correct multicast list now the flags have changed. | |
8762 | */ | |
8763 | ||
b6c40d68 PM |
8764 | if ((old_flags ^ flags) & IFF_MULTICAST) |
8765 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 8766 | |
4417da66 | 8767 | dev_set_rx_mode(dev); |
1da177e4 LT |
8768 | |
8769 | /* | |
8770 | * Have we downed the interface. We handle IFF_UP ourselves | |
8771 | * according to user attempts to set it, rather than blindly | |
8772 | * setting it. | |
8773 | */ | |
8774 | ||
8775 | ret = 0; | |
7051b88a | 8776 | if ((old_flags ^ flags) & IFF_UP) { |
8777 | if (old_flags & IFF_UP) | |
8778 | __dev_close(dev); | |
8779 | else | |
40c900aa | 8780 | ret = __dev_open(dev, extack); |
7051b88a | 8781 | } |
1da177e4 | 8782 | |
1da177e4 | 8783 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 8784 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 8785 | unsigned int old_flags = dev->flags; |
d1b19dff | 8786 | |
1da177e4 | 8787 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
8788 | |
8789 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
8790 | if (dev->flags != old_flags) | |
8791 | dev_set_rx_mode(dev); | |
1da177e4 LT |
8792 | } |
8793 | ||
8794 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 8795 | * is important. Some (broken) drivers set IFF_PROMISC, when |
8796 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
8797 | */ |
8798 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
8799 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
8800 | ||
1da177e4 | 8801 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 8802 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
8803 | } |
8804 | ||
bd380811 PM |
8805 | return ret; |
8806 | } | |
8807 | ||
a528c219 | 8808 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
1d997f10 HL |
8809 | unsigned int gchanges, u32 portid, |
8810 | const struct nlmsghdr *nlh) | |
bd380811 PM |
8811 | { |
8812 | unsigned int changes = dev->flags ^ old_flags; | |
8813 | ||
a528c219 | 8814 | if (gchanges) |
1d997f10 | 8815 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); |
a528c219 | 8816 | |
bd380811 PM |
8817 | if (changes & IFF_UP) { |
8818 | if (dev->flags & IFF_UP) | |
8819 | call_netdevice_notifiers(NETDEV_UP, dev); | |
8820 | else | |
8821 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
8822 | } | |
8823 | ||
8824 | if (dev->flags & IFF_UP && | |
be9efd36 | 8825 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
51d0c047 DA |
8826 | struct netdev_notifier_change_info change_info = { |
8827 | .info = { | |
8828 | .dev = dev, | |
8829 | }, | |
8830 | .flags_changed = changes, | |
8831 | }; | |
be9efd36 | 8832 | |
51d0c047 | 8833 | call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); |
be9efd36 | 8834 | } |
bd380811 PM |
8835 | } |
8836 | ||
8837 | /** | |
8838 | * dev_change_flags - change device settings | |
8839 | * @dev: device | |
8840 | * @flags: device state flags | |
567c5e13 | 8841 | * @extack: netlink extended ack |
bd380811 PM |
8842 | * |
8843 | * Change settings on device based state flags. The flags are | |
8844 | * in the userspace exported format. | |
8845 | */ | |
567c5e13 PM |
8846 | int dev_change_flags(struct net_device *dev, unsigned int flags, |
8847 | struct netlink_ext_ack *extack) | |
bd380811 | 8848 | { |
b536db93 | 8849 | int ret; |
991fb3f7 | 8850 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 | 8851 | |
6d040321 | 8852 | ret = __dev_change_flags(dev, flags, extack); |
bd380811 PM |
8853 | if (ret < 0) |
8854 | return ret; | |
8855 | ||
991fb3f7 | 8856 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
1d997f10 | 8857 | __dev_notify_flags(dev, old_flags, changes, 0, NULL); |
1da177e4 LT |
8858 | return ret; |
8859 | } | |
d1b19dff | 8860 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 8861 | |
f51048c3 | 8862 | int __dev_set_mtu(struct net_device *dev, int new_mtu) |
2315dc91 VF |
8863 | { |
8864 | const struct net_device_ops *ops = dev->netdev_ops; | |
8865 | ||
8866 | if (ops->ndo_change_mtu) | |
8867 | return ops->ndo_change_mtu(dev, new_mtu); | |
8868 | ||
501a90c9 ED |
8869 | /* Pairs with all the lockless reads of dev->mtu in the stack */ |
8870 | WRITE_ONCE(dev->mtu, new_mtu); | |
2315dc91 VF |
8871 | return 0; |
8872 | } | |
f51048c3 | 8873 | EXPORT_SYMBOL(__dev_set_mtu); |
2315dc91 | 8874 | |
d836f5c6 ED |
8875 | int dev_validate_mtu(struct net_device *dev, int new_mtu, |
8876 | struct netlink_ext_ack *extack) | |
8877 | { | |
8878 | /* MTU must be positive, and in range */ | |
8879 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
8880 | NL_SET_ERR_MSG(extack, "mtu less than device minimum"); | |
8881 | return -EINVAL; | |
8882 | } | |
8883 | ||
8884 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
8885 | NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); | |
8886 | return -EINVAL; | |
8887 | } | |
8888 | return 0; | |
8889 | } | |
8890 | ||
f0db275a | 8891 | /** |
7a4c53be | 8892 | * dev_set_mtu_ext - Change maximum transfer unit |
f0db275a SH |
8893 | * @dev: device |
8894 | * @new_mtu: new transfer unit | |
7a4c53be | 8895 | * @extack: netlink extended ack |
f0db275a SH |
8896 | * |
8897 | * Change the maximum transfer size of the network device. | |
8898 | */ | |
7a4c53be SH |
8899 | int dev_set_mtu_ext(struct net_device *dev, int new_mtu, |
8900 | struct netlink_ext_ack *extack) | |
1da177e4 | 8901 | { |
2315dc91 | 8902 | int err, orig_mtu; |
1da177e4 LT |
8903 | |
8904 | if (new_mtu == dev->mtu) | |
8905 | return 0; | |
8906 | ||
d836f5c6 ED |
8907 | err = dev_validate_mtu(dev, new_mtu, extack); |
8908 | if (err) | |
8909 | return err; | |
1da177e4 LT |
8910 | |
8911 | if (!netif_device_present(dev)) | |
8912 | return -ENODEV; | |
8913 | ||
1d486bfb VF |
8914 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
8915 | err = notifier_to_errno(err); | |
8916 | if (err) | |
8917 | return err; | |
d314774c | 8918 | |
2315dc91 VF |
8919 | orig_mtu = dev->mtu; |
8920 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 8921 | |
2315dc91 | 8922 | if (!err) { |
af7d6cce SD |
8923 | err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
8924 | orig_mtu); | |
2315dc91 VF |
8925 | err = notifier_to_errno(err); |
8926 | if (err) { | |
8927 | /* setting mtu back and notifying everyone again, | |
8928 | * so that they have a chance to revert changes. | |
8929 | */ | |
8930 | __dev_set_mtu(dev, orig_mtu); | |
af7d6cce SD |
8931 | call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
8932 | new_mtu); | |
2315dc91 VF |
8933 | } |
8934 | } | |
1da177e4 LT |
8935 | return err; |
8936 | } | |
7a4c53be SH |
8937 | |
8938 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
8939 | { | |
8940 | struct netlink_ext_ack extack; | |
8941 | int err; | |
8942 | ||
a6bcfc89 | 8943 | memset(&extack, 0, sizeof(extack)); |
7a4c53be | 8944 | err = dev_set_mtu_ext(dev, new_mtu, &extack); |
a6bcfc89 | 8945 | if (err && extack._msg) |
7a4c53be SH |
8946 | net_err_ratelimited("%s: %s\n", dev->name, extack._msg); |
8947 | return err; | |
8948 | } | |
d1b19dff | 8949 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 8950 | |
6a643ddb CW |
8951 | /** |
8952 | * dev_change_tx_queue_len - Change TX queue length of a netdevice | |
8953 | * @dev: device | |
8954 | * @new_len: new tx queue length | |
8955 | */ | |
8956 | int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) | |
8957 | { | |
8958 | unsigned int orig_len = dev->tx_queue_len; | |
8959 | int res; | |
8960 | ||
8961 | if (new_len != (unsigned int)new_len) | |
8962 | return -ERANGE; | |
8963 | ||
8964 | if (new_len != orig_len) { | |
ad13b5b0 | 8965 | WRITE_ONCE(dev->tx_queue_len, new_len); |
6a643ddb CW |
8966 | res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); |
8967 | res = notifier_to_errno(res); | |
7effaf06 TT |
8968 | if (res) |
8969 | goto err_rollback; | |
8970 | res = dev_qdisc_change_tx_queue_len(dev); | |
8971 | if (res) | |
8972 | goto err_rollback; | |
6a643ddb CW |
8973 | } |
8974 | ||
8975 | return 0; | |
7effaf06 TT |
8976 | |
8977 | err_rollback: | |
8978 | netdev_err(dev, "refused to change device tx_queue_len\n"); | |
ad13b5b0 | 8979 | WRITE_ONCE(dev->tx_queue_len, orig_len); |
7effaf06 | 8980 | return res; |
6a643ddb CW |
8981 | } |
8982 | ||
cbda10fa VD |
8983 | /** |
8984 | * dev_set_group - Change group this device belongs to | |
8985 | * @dev: device | |
8986 | * @new_group: group this device should belong to | |
8987 | */ | |
8988 | void dev_set_group(struct net_device *dev, int new_group) | |
8989 | { | |
8990 | dev->group = new_group; | |
8991 | } | |
cbda10fa | 8992 | |
d59cdf94 PM |
8993 | /** |
8994 | * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. | |
8995 | * @dev: device | |
8996 | * @addr: new address | |
8997 | * @extack: netlink extended ack | |
8998 | */ | |
8999 | int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, | |
9000 | struct netlink_ext_ack *extack) | |
9001 | { | |
9002 | struct netdev_notifier_pre_changeaddr_info info = { | |
9003 | .info.dev = dev, | |
9004 | .info.extack = extack, | |
9005 | .dev_addr = addr, | |
9006 | }; | |
9007 | int rc; | |
9008 | ||
9009 | rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); | |
9010 | return notifier_to_errno(rc); | |
9011 | } | |
9012 | EXPORT_SYMBOL(dev_pre_changeaddr_notify); | |
9013 | ||
f0db275a SH |
9014 | /** |
9015 | * dev_set_mac_address - Change Media Access Control Address | |
9016 | * @dev: device | |
9017 | * @sa: new address | |
3a37a963 | 9018 | * @extack: netlink extended ack |
f0db275a SH |
9019 | * |
9020 | * Change the hardware (MAC) address of the device | |
9021 | */ | |
3a37a963 PM |
9022 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, |
9023 | struct netlink_ext_ack *extack) | |
1da177e4 | 9024 | { |
d314774c | 9025 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
9026 | int err; |
9027 | ||
d314774c | 9028 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
9029 | return -EOPNOTSUPP; |
9030 | if (sa->sa_family != dev->type) | |
9031 | return -EINVAL; | |
9032 | if (!netif_device_present(dev)) | |
9033 | return -ENODEV; | |
d59cdf94 PM |
9034 | err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); |
9035 | if (err) | |
9036 | return err; | |
0ec92a8f PG |
9037 | if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { |
9038 | err = ops->ndo_set_mac_address(dev, sa); | |
9039 | if (err) | |
9040 | return err; | |
9041 | } | |
fbdeca2d | 9042 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 9043 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 9044 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 9045 | return 0; |
1da177e4 | 9046 | } |
d1b19dff | 9047 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 9048 | |
c7d52737 | 9049 | DECLARE_RWSEM(dev_addr_sem); |
3b23a32a CW |
9050 | |
9051 | int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, | |
9052 | struct netlink_ext_ack *extack) | |
9053 | { | |
9054 | int ret; | |
9055 | ||
9056 | down_write(&dev_addr_sem); | |
9057 | ret = dev_set_mac_address(dev, sa, extack); | |
9058 | up_write(&dev_addr_sem); | |
9059 | return ret; | |
9060 | } | |
9061 | EXPORT_SYMBOL(dev_set_mac_address_user); | |
9062 | ||
9063 | int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) | |
9064 | { | |
b5f0de6d | 9065 | size_t size = sizeof(sa->sa_data_min); |
3b23a32a CW |
9066 | struct net_device *dev; |
9067 | int ret = 0; | |
9068 | ||
9069 | down_read(&dev_addr_sem); | |
9070 | rcu_read_lock(); | |
9071 | ||
9072 | dev = dev_get_by_name_rcu(net, dev_name); | |
9073 | if (!dev) { | |
9074 | ret = -ENODEV; | |
9075 | goto unlock; | |
9076 | } | |
9077 | if (!dev->addr_len) | |
9078 | memset(sa->sa_data, 0, size); | |
9079 | else | |
9080 | memcpy(sa->sa_data, dev->dev_addr, | |
9081 | min_t(size_t, size, dev->addr_len)); | |
9082 | sa->sa_family = dev->type; | |
9083 | ||
9084 | unlock: | |
9085 | rcu_read_unlock(); | |
9086 | up_read(&dev_addr_sem); | |
9087 | return ret; | |
9088 | } | |
9089 | EXPORT_SYMBOL(dev_get_mac_address); | |
9090 | ||
4bf84c35 JP |
9091 | /** |
9092 | * dev_change_carrier - Change device carrier | |
9093 | * @dev: device | |
691b3b7e | 9094 | * @new_carrier: new value |
4bf84c35 JP |
9095 | * |
9096 | * Change device carrier | |
9097 | */ | |
9098 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
9099 | { | |
9100 | const struct net_device_ops *ops = dev->netdev_ops; | |
9101 | ||
9102 | if (!ops->ndo_change_carrier) | |
9103 | return -EOPNOTSUPP; | |
9104 | if (!netif_device_present(dev)) | |
9105 | return -ENODEV; | |
9106 | return ops->ndo_change_carrier(dev, new_carrier); | |
9107 | } | |
4bf84c35 | 9108 | |
66b52b0d JP |
9109 | /** |
9110 | * dev_get_phys_port_id - Get device physical port ID | |
9111 | * @dev: device | |
9112 | * @ppid: port ID | |
9113 | * | |
9114 | * Get device physical port ID | |
9115 | */ | |
9116 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 9117 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
9118 | { |
9119 | const struct net_device_ops *ops = dev->netdev_ops; | |
9120 | ||
9121 | if (!ops->ndo_get_phys_port_id) | |
9122 | return -EOPNOTSUPP; | |
9123 | return ops->ndo_get_phys_port_id(dev, ppid); | |
9124 | } | |
66b52b0d | 9125 | |
db24a904 DA |
9126 | /** |
9127 | * dev_get_phys_port_name - Get device physical port name | |
9128 | * @dev: device | |
9129 | * @name: port name | |
ed49e650 | 9130 | * @len: limit of bytes to copy to name |
db24a904 DA |
9131 | * |
9132 | * Get device physical port name | |
9133 | */ | |
9134 | int dev_get_phys_port_name(struct net_device *dev, | |
9135 | char *name, size_t len) | |
9136 | { | |
9137 | const struct net_device_ops *ops = dev->netdev_ops; | |
af3836df | 9138 | int err; |
db24a904 | 9139 | |
af3836df JP |
9140 | if (ops->ndo_get_phys_port_name) { |
9141 | err = ops->ndo_get_phys_port_name(dev, name, len); | |
9142 | if (err != -EOPNOTSUPP) | |
9143 | return err; | |
9144 | } | |
9145 | return devlink_compat_phys_port_name_get(dev, name, len); | |
db24a904 | 9146 | } |
db24a904 | 9147 | |
d6abc596 FF |
9148 | /** |
9149 | * dev_get_port_parent_id - Get the device's port parent identifier | |
9150 | * @dev: network device | |
9151 | * @ppid: pointer to a storage for the port's parent identifier | |
9152 | * @recurse: allow/disallow recursion to lower devices | |
9153 | * | |
9154 | * Get the devices's port parent identifier | |
9155 | */ | |
9156 | int dev_get_port_parent_id(struct net_device *dev, | |
9157 | struct netdev_phys_item_id *ppid, | |
9158 | bool recurse) | |
9159 | { | |
9160 | const struct net_device_ops *ops = dev->netdev_ops; | |
9161 | struct netdev_phys_item_id first = { }; | |
9162 | struct net_device *lower_dev; | |
9163 | struct list_head *iter; | |
7e1146e8 JP |
9164 | int err; |
9165 | ||
9166 | if (ops->ndo_get_port_parent_id) { | |
9167 | err = ops->ndo_get_port_parent_id(dev, ppid); | |
9168 | if (err != -EOPNOTSUPP) | |
9169 | return err; | |
9170 | } | |
d6abc596 | 9171 | |
7e1146e8 | 9172 | err = devlink_compat_switch_id_get(dev, ppid); |
c0288ae8 | 9173 | if (!recurse || err != -EOPNOTSUPP) |
7e1146e8 | 9174 | return err; |
d6abc596 | 9175 | |
d6abc596 | 9176 | netdev_for_each_lower_dev(dev, lower_dev, iter) { |
c0288ae8 | 9177 | err = dev_get_port_parent_id(lower_dev, ppid, true); |
d6abc596 FF |
9178 | if (err) |
9179 | break; | |
9180 | if (!first.id_len) | |
9181 | first = *ppid; | |
9182 | else if (memcmp(&first, ppid, sizeof(*ppid))) | |
e1b9efe6 | 9183 | return -EOPNOTSUPP; |
d6abc596 FF |
9184 | } |
9185 | ||
9186 | return err; | |
9187 | } | |
9188 | EXPORT_SYMBOL(dev_get_port_parent_id); | |
9189 | ||
9190 | /** | |
9191 | * netdev_port_same_parent_id - Indicate if two network devices have | |
9192 | * the same port parent identifier | |
9193 | * @a: first network device | |
9194 | * @b: second network device | |
9195 | */ | |
9196 | bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) | |
9197 | { | |
9198 | struct netdev_phys_item_id a_id = { }; | |
9199 | struct netdev_phys_item_id b_id = { }; | |
9200 | ||
9201 | if (dev_get_port_parent_id(a, &a_id, true) || | |
9202 | dev_get_port_parent_id(b, &b_id, true)) | |
9203 | return false; | |
9204 | ||
9205 | return netdev_phys_item_id_same(&a_id, &b_id); | |
9206 | } | |
9207 | EXPORT_SYMBOL(netdev_port_same_parent_id); | |
9208 | ||
d746d707 | 9209 | /** |
2106efda JK |
9210 | * dev_change_proto_down - set carrier according to proto_down. |
9211 | * | |
d746d707 AK |
9212 | * @dev: device |
9213 | * @proto_down: new value | |
d746d707 AK |
9214 | */ |
9215 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
9216 | { | |
2106efda | 9217 | if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) |
d746d707 AK |
9218 | return -EOPNOTSUPP; |
9219 | if (!netif_device_present(dev)) | |
9220 | return -ENODEV; | |
b5899679 AR |
9221 | if (proto_down) |
9222 | netif_carrier_off(dev); | |
9223 | else | |
9224 | netif_carrier_on(dev); | |
6890ab31 | 9225 | WRITE_ONCE(dev->proto_down, proto_down); |
b5899679 AR |
9226 | return 0; |
9227 | } | |
b5899679 | 9228 | |
829eb208 RP |
9229 | /** |
9230 | * dev_change_proto_down_reason - proto down reason | |
9231 | * | |
9232 | * @dev: device | |
9233 | * @mask: proto down mask | |
9234 | * @value: proto down value | |
9235 | */ | |
9236 | void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, | |
9237 | u32 value) | |
9238 | { | |
6890ab31 | 9239 | u32 proto_down_reason; |
829eb208 RP |
9240 | int b; |
9241 | ||
9242 | if (!mask) { | |
6890ab31 | 9243 | proto_down_reason = value; |
829eb208 | 9244 | } else { |
6890ab31 | 9245 | proto_down_reason = dev->proto_down_reason; |
829eb208 RP |
9246 | for_each_set_bit(b, &mask, 32) { |
9247 | if (value & (1 << b)) | |
6890ab31 | 9248 | proto_down_reason |= BIT(b); |
829eb208 | 9249 | else |
6890ab31 | 9250 | proto_down_reason &= ~BIT(b); |
829eb208 RP |
9251 | } |
9252 | } | |
6890ab31 | 9253 | WRITE_ONCE(dev->proto_down_reason, proto_down_reason); |
829eb208 | 9254 | } |
829eb208 | 9255 | |
aa8d3a71 AN |
9256 | struct bpf_xdp_link { |
9257 | struct bpf_link link; | |
9258 | struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ | |
9259 | int flags; | |
9260 | }; | |
9261 | ||
c8a36f19 | 9262 | static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) |
d67b9cd2 | 9263 | { |
7f0a8382 AN |
9264 | if (flags & XDP_FLAGS_HW_MODE) |
9265 | return XDP_MODE_HW; | |
9266 | if (flags & XDP_FLAGS_DRV_MODE) | |
9267 | return XDP_MODE_DRV; | |
c8a36f19 AN |
9268 | if (flags & XDP_FLAGS_SKB_MODE) |
9269 | return XDP_MODE_SKB; | |
9270 | return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; | |
7f0a8382 | 9271 | } |
d67b9cd2 | 9272 | |
7f0a8382 AN |
9273 | static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) |
9274 | { | |
9275 | switch (mode) { | |
9276 | case XDP_MODE_SKB: | |
9277 | return generic_xdp_install; | |
9278 | case XDP_MODE_DRV: | |
9279 | case XDP_MODE_HW: | |
9280 | return dev->netdev_ops->ndo_bpf; | |
9281 | default: | |
9282 | return NULL; | |
5d867245 | 9283 | } |
7f0a8382 | 9284 | } |
118b4aa2 | 9285 | |
aa8d3a71 AN |
9286 | static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, |
9287 | enum bpf_xdp_mode mode) | |
9288 | { | |
9289 | return dev->xdp_state[mode].link; | |
9290 | } | |
9291 | ||
7f0a8382 AN |
9292 | static struct bpf_prog *dev_xdp_prog(struct net_device *dev, |
9293 | enum bpf_xdp_mode mode) | |
9294 | { | |
aa8d3a71 AN |
9295 | struct bpf_xdp_link *link = dev_xdp_link(dev, mode); |
9296 | ||
9297 | if (link) | |
9298 | return link->link.prog; | |
7f0a8382 AN |
9299 | return dev->xdp_state[mode].prog; |
9300 | } | |
9301 | ||
879af96f | 9302 | u8 dev_xdp_prog_count(struct net_device *dev) |
998f1729 THJ |
9303 | { |
9304 | u8 count = 0; | |
9305 | int i; | |
9306 | ||
9307 | for (i = 0; i < __MAX_XDP_MODE; i++) | |
9308 | if (dev->xdp_state[i].prog || dev->xdp_state[i].link) | |
9309 | count++; | |
9310 | return count; | |
9311 | } | |
879af96f | 9312 | EXPORT_SYMBOL_GPL(dev_xdp_prog_count); |
998f1729 | 9313 | |
7f0a8382 AN |
9314 | u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) |
9315 | { | |
9316 | struct bpf_prog *prog = dev_xdp_prog(dev, mode); | |
118b4aa2 | 9317 | |
7f0a8382 AN |
9318 | return prog ? prog->aux->id : 0; |
9319 | } | |
58038695 | 9320 | |
aa8d3a71 AN |
9321 | static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, |
9322 | struct bpf_xdp_link *link) | |
9323 | { | |
9324 | dev->xdp_state[mode].link = link; | |
9325 | dev->xdp_state[mode].prog = NULL; | |
d67b9cd2 DB |
9326 | } |
9327 | ||
7f0a8382 AN |
9328 | static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, |
9329 | struct bpf_prog *prog) | |
9330 | { | |
aa8d3a71 | 9331 | dev->xdp_state[mode].link = NULL; |
7f0a8382 | 9332 | dev->xdp_state[mode].prog = prog; |
d67b9cd2 DB |
9333 | } |
9334 | ||
7f0a8382 AN |
9335 | static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, |
9336 | bpf_op_t bpf_op, struct netlink_ext_ack *extack, | |
9337 | u32 flags, struct bpf_prog *prog) | |
d67b9cd2 | 9338 | { |
f4e63525 | 9339 | struct netdev_bpf xdp; |
7e6897f9 BT |
9340 | int err; |
9341 | ||
d67b9cd2 | 9342 | memset(&xdp, 0, sizeof(xdp)); |
7f0a8382 | 9343 | xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; |
d67b9cd2 | 9344 | xdp.extack = extack; |
32d60277 | 9345 | xdp.flags = flags; |
d67b9cd2 DB |
9346 | xdp.prog = prog; |
9347 | ||
7f0a8382 AN |
9348 | /* Drivers assume refcnt is already incremented (i.e, prog pointer is |
9349 | * "moved" into driver), so they don't increment it on their own, but | |
9350 | * they do decrement refcnt when program is detached or replaced. | |
9351 | * Given net_device also owns link/prog, we need to bump refcnt here | |
9352 | * to prevent drivers from underflowing it. | |
9353 | */ | |
9354 | if (prog) | |
9355 | bpf_prog_inc(prog); | |
7e6897f9 | 9356 | err = bpf_op(dev, &xdp); |
7f0a8382 AN |
9357 | if (err) { |
9358 | if (prog) | |
9359 | bpf_prog_put(prog); | |
9360 | return err; | |
9361 | } | |
7e6897f9 | 9362 | |
7f0a8382 AN |
9363 | if (mode != XDP_MODE_HW) |
9364 | bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); | |
7e6897f9 | 9365 | |
7f0a8382 | 9366 | return 0; |
d67b9cd2 DB |
9367 | } |
9368 | ||
bd0b2e7f JK |
9369 | static void dev_xdp_uninstall(struct net_device *dev) |
9370 | { | |
aa8d3a71 | 9371 | struct bpf_xdp_link *link; |
7f0a8382 AN |
9372 | struct bpf_prog *prog; |
9373 | enum bpf_xdp_mode mode; | |
9374 | bpf_op_t bpf_op; | |
bd0b2e7f | 9375 | |
7f0a8382 | 9376 | ASSERT_RTNL(); |
bd0b2e7f | 9377 | |
7f0a8382 AN |
9378 | for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { |
9379 | prog = dev_xdp_prog(dev, mode); | |
9380 | if (!prog) | |
9381 | continue; | |
bd0b2e7f | 9382 | |
7f0a8382 AN |
9383 | bpf_op = dev_xdp_bpf_op(dev, mode); |
9384 | if (!bpf_op) | |
9385 | continue; | |
bd0b2e7f | 9386 | |
7f0a8382 AN |
9387 | WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); |
9388 | ||
aa8d3a71 AN |
9389 | /* auto-detach link from net device */ |
9390 | link = dev_xdp_link(dev, mode); | |
9391 | if (link) | |
9392 | link->dev = NULL; | |
9393 | else | |
9394 | bpf_prog_put(prog); | |
9395 | ||
9396 | dev_xdp_set_link(dev, mode, NULL); | |
7f0a8382 | 9397 | } |
bd0b2e7f JK |
9398 | } |
9399 | ||
d4baa936 | 9400 | static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, |
aa8d3a71 AN |
9401 | struct bpf_xdp_link *link, struct bpf_prog *new_prog, |
9402 | struct bpf_prog *old_prog, u32 flags) | |
a7862b45 | 9403 | { |
998f1729 | 9404 | unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); |
d4baa936 | 9405 | struct bpf_prog *cur_prog; |
879af96f JM |
9406 | struct net_device *upper; |
9407 | struct list_head *iter; | |
d4baa936 | 9408 | enum bpf_xdp_mode mode; |
7f0a8382 | 9409 | bpf_op_t bpf_op; |
a7862b45 BB |
9410 | int err; |
9411 | ||
85de8576 DB |
9412 | ASSERT_RTNL(); |
9413 | ||
aa8d3a71 AN |
9414 | /* either link or prog attachment, never both */ |
9415 | if (link && (new_prog || old_prog)) | |
9416 | return -EINVAL; | |
9417 | /* link supports only XDP mode flags */ | |
9418 | if (link && (flags & ~XDP_FLAGS_MODES)) { | |
9419 | NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); | |
9420 | return -EINVAL; | |
9421 | } | |
998f1729 THJ |
9422 | /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ |
9423 | if (num_modes > 1) { | |
d4baa936 AN |
9424 | NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); |
9425 | return -EINVAL; | |
9426 | } | |
998f1729 THJ |
9427 | /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ |
9428 | if (!num_modes && dev_xdp_prog_count(dev) > 1) { | |
9429 | NL_SET_ERR_MSG(extack, | |
9430 | "More than one program loaded, unset mode is ambiguous"); | |
9431 | return -EINVAL; | |
9432 | } | |
d4baa936 AN |
9433 | /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ |
9434 | if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { | |
9435 | NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); | |
9436 | return -EINVAL; | |
01dde20c | 9437 | } |
a25717d2 | 9438 | |
c8a36f19 | 9439 | mode = dev_xdp_mode(dev, flags); |
aa8d3a71 AN |
9440 | /* can't replace attached link */ |
9441 | if (dev_xdp_link(dev, mode)) { | |
9442 | NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); | |
9443 | return -EBUSY; | |
01dde20c | 9444 | } |
c14a9f63 | 9445 | |
879af96f JM |
9446 | /* don't allow if an upper device already has a program */ |
9447 | netdev_for_each_upper_dev_rcu(dev, upper, iter) { | |
9448 | if (dev_xdp_prog_count(upper) > 0) { | |
9449 | NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); | |
9450 | return -EEXIST; | |
9451 | } | |
9452 | } | |
9453 | ||
d4baa936 | 9454 | cur_prog = dev_xdp_prog(dev, mode); |
aa8d3a71 AN |
9455 | /* can't replace attached prog with link */ |
9456 | if (link && cur_prog) { | |
9457 | NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); | |
9458 | return -EBUSY; | |
9459 | } | |
d4baa936 AN |
9460 | if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { |
9461 | NL_SET_ERR_MSG(extack, "Active program does not match expected"); | |
9462 | return -EEXIST; | |
92234c8f | 9463 | } |
c14a9f63 | 9464 | |
aa8d3a71 AN |
9465 | /* put effective new program into new_prog */ |
9466 | if (link) | |
9467 | new_prog = link->link.prog; | |
85de8576 | 9468 | |
d4baa936 AN |
9469 | if (new_prog) { |
9470 | bool offload = mode == XDP_MODE_HW; | |
7f0a8382 AN |
9471 | enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB |
9472 | ? XDP_MODE_DRV : XDP_MODE_SKB; | |
441a3303 | 9473 | |
068d9d1e AN |
9474 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { |
9475 | NL_SET_ERR_MSG(extack, "XDP program already attached"); | |
9476 | return -EBUSY; | |
9477 | } | |
d4baa936 | 9478 | if (!offload && dev_xdp_prog(dev, other_mode)) { |
7f0a8382 | 9479 | NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); |
d67b9cd2 | 9480 | return -EEXIST; |
01dde20c | 9481 | } |
9d03ebc7 SF |
9482 | if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { |
9483 | NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); | |
441a3303 JK |
9484 | return -EINVAL; |
9485 | } | |
2b3486bc SF |
9486 | if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { |
9487 | NL_SET_ERR_MSG(extack, "Program bound to different device"); | |
441a3303 JK |
9488 | return -EINVAL; |
9489 | } | |
d4baa936 | 9490 | if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { |
fbee97fe | 9491 | NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); |
fbee97fe DA |
9492 | return -EINVAL; |
9493 | } | |
d4baa936 AN |
9494 | if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { |
9495 | NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); | |
92164774 LB |
9496 | return -EINVAL; |
9497 | } | |
d4baa936 | 9498 | } |
92164774 | 9499 | |
d4baa936 AN |
9500 | /* don't call drivers if the effective program didn't change */ |
9501 | if (new_prog != cur_prog) { | |
9502 | bpf_op = dev_xdp_bpf_op(dev, mode); | |
9503 | if (!bpf_op) { | |
9504 | NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); | |
9505 | return -EOPNOTSUPP; | |
c14a9f63 | 9506 | } |
a7862b45 | 9507 | |
d4baa936 AN |
9508 | err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); |
9509 | if (err) | |
9510 | return err; | |
7f0a8382 | 9511 | } |
d4baa936 | 9512 | |
aa8d3a71 AN |
9513 | if (link) |
9514 | dev_xdp_set_link(dev, mode, link); | |
9515 | else | |
9516 | dev_xdp_set_prog(dev, mode, new_prog); | |
d4baa936 AN |
9517 | if (cur_prog) |
9518 | bpf_prog_put(cur_prog); | |
a7862b45 | 9519 | |
7f0a8382 | 9520 | return 0; |
a7862b45 | 9521 | } |
a7862b45 | 9522 | |
aa8d3a71 AN |
9523 | static int dev_xdp_attach_link(struct net_device *dev, |
9524 | struct netlink_ext_ack *extack, | |
9525 | struct bpf_xdp_link *link) | |
9526 | { | |
9527 | return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); | |
9528 | } | |
9529 | ||
9530 | static int dev_xdp_detach_link(struct net_device *dev, | |
9531 | struct netlink_ext_ack *extack, | |
9532 | struct bpf_xdp_link *link) | |
9533 | { | |
9534 | enum bpf_xdp_mode mode; | |
9535 | bpf_op_t bpf_op; | |
9536 | ||
9537 | ASSERT_RTNL(); | |
9538 | ||
c8a36f19 | 9539 | mode = dev_xdp_mode(dev, link->flags); |
aa8d3a71 AN |
9540 | if (dev_xdp_link(dev, mode) != link) |
9541 | return -EINVAL; | |
9542 | ||
9543 | bpf_op = dev_xdp_bpf_op(dev, mode); | |
9544 | WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); | |
9545 | dev_xdp_set_link(dev, mode, NULL); | |
9546 | return 0; | |
9547 | } | |
9548 | ||
9549 | static void bpf_xdp_link_release(struct bpf_link *link) | |
9550 | { | |
9551 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9552 | ||
9553 | rtnl_lock(); | |
9554 | ||
9555 | /* if racing with net_device's tear down, xdp_link->dev might be | |
9556 | * already NULL, in which case link was already auto-detached | |
9557 | */ | |
73b11c2a | 9558 | if (xdp_link->dev) { |
aa8d3a71 | 9559 | WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); |
73b11c2a AN |
9560 | xdp_link->dev = NULL; |
9561 | } | |
aa8d3a71 AN |
9562 | |
9563 | rtnl_unlock(); | |
9564 | } | |
9565 | ||
73b11c2a AN |
9566 | static int bpf_xdp_link_detach(struct bpf_link *link) |
9567 | { | |
9568 | bpf_xdp_link_release(link); | |
9569 | return 0; | |
9570 | } | |
9571 | ||
aa8d3a71 AN |
9572 | static void bpf_xdp_link_dealloc(struct bpf_link *link) |
9573 | { | |
9574 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9575 | ||
9576 | kfree(xdp_link); | |
9577 | } | |
9578 | ||
c1931c97 AN |
9579 | static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, |
9580 | struct seq_file *seq) | |
9581 | { | |
9582 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9583 | u32 ifindex = 0; | |
9584 | ||
9585 | rtnl_lock(); | |
9586 | if (xdp_link->dev) | |
9587 | ifindex = xdp_link->dev->ifindex; | |
9588 | rtnl_unlock(); | |
9589 | ||
9590 | seq_printf(seq, "ifindex:\t%u\n", ifindex); | |
9591 | } | |
9592 | ||
9593 | static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, | |
9594 | struct bpf_link_info *info) | |
9595 | { | |
9596 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9597 | u32 ifindex = 0; | |
9598 | ||
9599 | rtnl_lock(); | |
9600 | if (xdp_link->dev) | |
9601 | ifindex = xdp_link->dev->ifindex; | |
9602 | rtnl_unlock(); | |
9603 | ||
9604 | info->xdp.ifindex = ifindex; | |
9605 | return 0; | |
9606 | } | |
9607 | ||
026a4c28 AN |
9608 | static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, |
9609 | struct bpf_prog *old_prog) | |
9610 | { | |
9611 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9612 | enum bpf_xdp_mode mode; | |
9613 | bpf_op_t bpf_op; | |
9614 | int err = 0; | |
9615 | ||
9616 | rtnl_lock(); | |
9617 | ||
9618 | /* link might have been auto-released already, so fail */ | |
9619 | if (!xdp_link->dev) { | |
9620 | err = -ENOLINK; | |
9621 | goto out_unlock; | |
9622 | } | |
9623 | ||
9624 | if (old_prog && link->prog != old_prog) { | |
9625 | err = -EPERM; | |
9626 | goto out_unlock; | |
9627 | } | |
9628 | old_prog = link->prog; | |
382778ed THJ |
9629 | if (old_prog->type != new_prog->type || |
9630 | old_prog->expected_attach_type != new_prog->expected_attach_type) { | |
9631 | err = -EINVAL; | |
9632 | goto out_unlock; | |
9633 | } | |
9634 | ||
026a4c28 AN |
9635 | if (old_prog == new_prog) { |
9636 | /* no-op, don't disturb drivers */ | |
9637 | bpf_prog_put(new_prog); | |
9638 | goto out_unlock; | |
9639 | } | |
9640 | ||
c8a36f19 | 9641 | mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); |
026a4c28 AN |
9642 | bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); |
9643 | err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, | |
9644 | xdp_link->flags, new_prog); | |
9645 | if (err) | |
9646 | goto out_unlock; | |
9647 | ||
9648 | old_prog = xchg(&link->prog, new_prog); | |
9649 | bpf_prog_put(old_prog); | |
9650 | ||
9651 | out_unlock: | |
9652 | rtnl_unlock(); | |
9653 | return err; | |
9654 | } | |
9655 | ||
aa8d3a71 AN |
9656 | static const struct bpf_link_ops bpf_xdp_link_lops = { |
9657 | .release = bpf_xdp_link_release, | |
9658 | .dealloc = bpf_xdp_link_dealloc, | |
73b11c2a | 9659 | .detach = bpf_xdp_link_detach, |
c1931c97 AN |
9660 | .show_fdinfo = bpf_xdp_link_show_fdinfo, |
9661 | .fill_link_info = bpf_xdp_link_fill_link_info, | |
026a4c28 | 9662 | .update_prog = bpf_xdp_link_update, |
aa8d3a71 AN |
9663 | }; |
9664 | ||
9665 | int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) | |
9666 | { | |
9667 | struct net *net = current->nsproxy->net_ns; | |
9668 | struct bpf_link_primer link_primer; | |
bf4ea1d0 | 9669 | struct netlink_ext_ack extack = {}; |
aa8d3a71 AN |
9670 | struct bpf_xdp_link *link; |
9671 | struct net_device *dev; | |
9672 | int err, fd; | |
9673 | ||
5acc7d3e | 9674 | rtnl_lock(); |
aa8d3a71 | 9675 | dev = dev_get_by_index(net, attr->link_create.target_ifindex); |
5acc7d3e XZ |
9676 | if (!dev) { |
9677 | rtnl_unlock(); | |
aa8d3a71 | 9678 | return -EINVAL; |
5acc7d3e | 9679 | } |
aa8d3a71 AN |
9680 | |
9681 | link = kzalloc(sizeof(*link), GFP_USER); | |
9682 | if (!link) { | |
9683 | err = -ENOMEM; | |
5acc7d3e | 9684 | goto unlock; |
aa8d3a71 AN |
9685 | } |
9686 | ||
9687 | bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); | |
9688 | link->dev = dev; | |
9689 | link->flags = attr->link_create.flags; | |
9690 | ||
9691 | err = bpf_link_prime(&link->link, &link_primer); | |
9692 | if (err) { | |
9693 | kfree(link); | |
5acc7d3e | 9694 | goto unlock; |
aa8d3a71 AN |
9695 | } |
9696 | ||
bf4ea1d0 | 9697 | err = dev_xdp_attach_link(dev, &extack, link); |
aa8d3a71 AN |
9698 | rtnl_unlock(); |
9699 | ||
9700 | if (err) { | |
5acc7d3e | 9701 | link->dev = NULL; |
aa8d3a71 | 9702 | bpf_link_cleanup(&link_primer); |
bf4ea1d0 | 9703 | trace_bpf_xdp_link_attach_failed(extack._msg); |
aa8d3a71 AN |
9704 | goto out_put_dev; |
9705 | } | |
9706 | ||
9707 | fd = bpf_link_settle(&link_primer); | |
9708 | /* link itself doesn't hold dev's refcnt to not complicate shutdown */ | |
9709 | dev_put(dev); | |
9710 | return fd; | |
9711 | ||
5acc7d3e XZ |
9712 | unlock: |
9713 | rtnl_unlock(); | |
9714 | ||
aa8d3a71 AN |
9715 | out_put_dev: |
9716 | dev_put(dev); | |
9717 | return err; | |
9718 | } | |
9719 | ||
d4baa936 AN |
9720 | /** |
9721 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
9722 | * @dev: device | |
9723 | * @extack: netlink extended ack | |
9724 | * @fd: new program fd or negative value to clear | |
9725 | * @expected_fd: old program fd that userspace expects to replace or clear | |
9726 | * @flags: xdp-related flags | |
9727 | * | |
9728 | * Set or clear a bpf program for a device | |
9729 | */ | |
9730 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, | |
9731 | int fd, int expected_fd, u32 flags) | |
9732 | { | |
c8a36f19 | 9733 | enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); |
d4baa936 AN |
9734 | struct bpf_prog *new_prog = NULL, *old_prog = NULL; |
9735 | int err; | |
9736 | ||
9737 | ASSERT_RTNL(); | |
9738 | ||
9739 | if (fd >= 0) { | |
9740 | new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, | |
9741 | mode != XDP_MODE_SKB); | |
9742 | if (IS_ERR(new_prog)) | |
9743 | return PTR_ERR(new_prog); | |
9744 | } | |
9745 | ||
9746 | if (expected_fd >= 0) { | |
9747 | old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, | |
9748 | mode != XDP_MODE_SKB); | |
9749 | if (IS_ERR(old_prog)) { | |
9750 | err = PTR_ERR(old_prog); | |
9751 | old_prog = NULL; | |
9752 | goto err_out; | |
c14a9f63 | 9753 | } |
a7862b45 BB |
9754 | } |
9755 | ||
aa8d3a71 | 9756 | err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); |
a7862b45 | 9757 | |
d4baa936 AN |
9758 | err_out: |
9759 | if (err && new_prog) | |
9760 | bpf_prog_put(new_prog); | |
9761 | if (old_prog) | |
9762 | bpf_prog_put(old_prog); | |
a7862b45 BB |
9763 | return err; |
9764 | } | |
a7862b45 | 9765 | |
1da177e4 | 9766 | /** |
759ab1ed JK |
9767 | * dev_index_reserve() - allocate an ifindex in a namespace |
9768 | * @net: the applicable net namespace | |
9769 | * @ifindex: requested ifindex, pass %0 to get one allocated | |
9770 | * | |
9771 | * Allocate a ifindex for a new device. Caller must either use the ifindex | |
9772 | * to store the device (via list_netdevice()) or call dev_index_release() | |
9773 | * to give the index up. | |
1da177e4 | 9774 | * |
759ab1ed | 9775 | * Return: a suitable unique value for a new device interface number or -errno. |
1da177e4 | 9776 | */ |
759ab1ed | 9777 | static int dev_index_reserve(struct net *net, u32 ifindex) |
1da177e4 | 9778 | { |
759ab1ed | 9779 | int err; |
f4563a75 | 9780 | |
956db0a1 JK |
9781 | if (ifindex > INT_MAX) { |
9782 | DEBUG_NET_WARN_ON_ONCE(1); | |
9783 | return -EINVAL; | |
9784 | } | |
9785 | ||
759ab1ed JK |
9786 | if (!ifindex) |
9787 | err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, | |
9788 | xa_limit_31b, &net->ifindex, GFP_KERNEL); | |
9789 | else | |
9790 | err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); | |
9791 | if (err < 0) | |
9792 | return err; | |
9793 | ||
9794 | return ifindex; | |
9795 | } | |
9796 | ||
9797 | static void dev_index_release(struct net *net, int ifindex) | |
9798 | { | |
9799 | /* Expect only unused indexes, unlist_netdevice() removes the used */ | |
9800 | WARN_ON(xa_erase(&net->dev_by_index, ifindex)); | |
1da177e4 LT |
9801 | } |
9802 | ||
1da177e4 | 9803 | /* Delayed registration/unregisteration */ |
0b5c21bb | 9804 | LIST_HEAD(net_todo_list); |
200b916f | 9805 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
ffabe98c | 9806 | atomic_t dev_unreg_count = ATOMIC_INIT(0); |
1da177e4 | 9807 | |
6f05f629 | 9808 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 9809 | { |
1da177e4 | 9810 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
9811 | } |
9812 | ||
fd867d51 JW |
9813 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
9814 | struct net_device *upper, netdev_features_t features) | |
9815 | { | |
9816 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
9817 | netdev_features_t feature; | |
5ba3f7d6 | 9818 | int feature_bit; |
fd867d51 | 9819 | |
3b89ea9c | 9820 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 9821 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
9822 | if (!(upper->wanted_features & feature) |
9823 | && (features & feature)) { | |
9824 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
9825 | &feature, upper->name); | |
9826 | features &= ~feature; | |
9827 | } | |
9828 | } | |
9829 | ||
9830 | return features; | |
9831 | } | |
9832 | ||
9833 | static void netdev_sync_lower_features(struct net_device *upper, | |
9834 | struct net_device *lower, netdev_features_t features) | |
9835 | { | |
9836 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
9837 | netdev_features_t feature; | |
5ba3f7d6 | 9838 | int feature_bit; |
fd867d51 | 9839 | |
3b89ea9c | 9840 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 9841 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
9842 | if (!(features & feature) && (lower->features & feature)) { |
9843 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
9844 | &feature, lower->name); | |
9845 | lower->wanted_features &= ~feature; | |
dd912306 | 9846 | __netdev_update_features(lower); |
fd867d51 JW |
9847 | |
9848 | if (unlikely(lower->features & feature)) | |
9849 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
9850 | &feature, lower->name); | |
dd912306 CW |
9851 | else |
9852 | netdev_features_change(lower); | |
fd867d51 JW |
9853 | } |
9854 | } | |
9855 | } | |
9856 | ||
c8f44aff MM |
9857 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
9858 | netdev_features_t features) | |
b63365a2 | 9859 | { |
57422dc5 MM |
9860 | /* Fix illegal checksum combinations */ |
9861 | if ((features & NETIF_F_HW_CSUM) && | |
9862 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 9863 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
9864 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
9865 | } | |
9866 | ||
b63365a2 | 9867 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 9868 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 9869 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 9870 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
9871 | } |
9872 | ||
ec5f0615 PS |
9873 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
9874 | !(features & NETIF_F_IP_CSUM)) { | |
9875 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
9876 | features &= ~NETIF_F_TSO; | |
9877 | features &= ~NETIF_F_TSO_ECN; | |
9878 | } | |
9879 | ||
9880 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
9881 | !(features & NETIF_F_IPV6_CSUM)) { | |
9882 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
9883 | features &= ~NETIF_F_TSO6; | |
9884 | } | |
9885 | ||
b1dc497b AD |
9886 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
9887 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
9888 | features &= ~NETIF_F_TSO_MANGLEID; | |
9889 | ||
31d8b9e0 BH |
9890 | /* TSO ECN requires that TSO is present as well. */ |
9891 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
9892 | features &= ~NETIF_F_TSO_ECN; | |
9893 | ||
212b573f MM |
9894 | /* Software GSO depends on SG. */ |
9895 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 9896 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
9897 | features &= ~NETIF_F_GSO; |
9898 | } | |
9899 | ||
802ab55a AD |
9900 | /* GSO partial features require GSO partial be set */ |
9901 | if ((features & dev->gso_partial_features) && | |
9902 | !(features & NETIF_F_GSO_PARTIAL)) { | |
9903 | netdev_dbg(dev, | |
9904 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
9905 | features &= ~dev->gso_partial_features; | |
9906 | } | |
9907 | ||
fb1f5f79 MC |
9908 | if (!(features & NETIF_F_RXCSUM)) { |
9909 | /* NETIF_F_GRO_HW implies doing RXCSUM since every packet | |
9910 | * successfully merged by hardware must also have the | |
9911 | * checksum verified by hardware. If the user does not | |
9912 | * want to enable RXCSUM, logically, we should disable GRO_HW. | |
9913 | */ | |
9914 | if (features & NETIF_F_GRO_HW) { | |
9915 | netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); | |
9916 | features &= ~NETIF_F_GRO_HW; | |
9917 | } | |
9918 | } | |
9919 | ||
de8d5ab2 GP |
9920 | /* LRO/HW-GRO features cannot be combined with RX-FCS */ |
9921 | if (features & NETIF_F_RXFCS) { | |
9922 | if (features & NETIF_F_LRO) { | |
9923 | netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); | |
9924 | features &= ~NETIF_F_LRO; | |
9925 | } | |
9926 | ||
9927 | if (features & NETIF_F_GRO_HW) { | |
9928 | netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); | |
9929 | features &= ~NETIF_F_GRO_HW; | |
9930 | } | |
e6c6a929 GP |
9931 | } |
9932 | ||
54b2b3ec BB |
9933 | if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { |
9934 | netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); | |
9935 | features &= ~NETIF_F_LRO; | |
9936 | } | |
9937 | ||
25537d71 TT |
9938 | if (features & NETIF_F_HW_TLS_TX) { |
9939 | bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == | |
9940 | (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
9941 | bool hw_csum = features & NETIF_F_HW_CSUM; | |
9942 | ||
9943 | if (!ip_csum && !hw_csum) { | |
9944 | netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); | |
9945 | features &= ~NETIF_F_HW_TLS_TX; | |
9946 | } | |
ae0b04b2 TT |
9947 | } |
9948 | ||
a3eb4e9d TT |
9949 | if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { |
9950 | netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); | |
9951 | features &= ~NETIF_F_HW_TLS_RX; | |
9952 | } | |
9953 | ||
b63365a2 HX |
9954 | return features; |
9955 | } | |
b63365a2 | 9956 | |
6cb6a27c | 9957 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 9958 | { |
fd867d51 | 9959 | struct net_device *upper, *lower; |
c8f44aff | 9960 | netdev_features_t features; |
fd867d51 | 9961 | struct list_head *iter; |
e7868a85 | 9962 | int err = -1; |
5455c699 | 9963 | |
87267485 MM |
9964 | ASSERT_RTNL(); |
9965 | ||
5455c699 MM |
9966 | features = netdev_get_wanted_features(dev); |
9967 | ||
9968 | if (dev->netdev_ops->ndo_fix_features) | |
9969 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
9970 | ||
9971 | /* driver might be less strict about feature dependencies */ | |
9972 | features = netdev_fix_features(dev, features); | |
9973 | ||
4250b75b | 9974 | /* some features can't be enabled if they're off on an upper device */ |
fd867d51 JW |
9975 | netdev_for_each_upper_dev_rcu(dev, upper, iter) |
9976 | features = netdev_sync_upper_features(dev, upper, features); | |
9977 | ||
5455c699 | 9978 | if (dev->features == features) |
e7868a85 | 9979 | goto sync_lower; |
5455c699 | 9980 | |
c8f44aff MM |
9981 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
9982 | &dev->features, &features); | |
5455c699 MM |
9983 | |
9984 | if (dev->netdev_ops->ndo_set_features) | |
9985 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
9986 | else |
9987 | err = 0; | |
5455c699 | 9988 | |
6cb6a27c | 9989 | if (unlikely(err < 0)) { |
5455c699 | 9990 | netdev_err(dev, |
c8f44aff MM |
9991 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
9992 | err, &features, &dev->features); | |
17b85d29 NA |
9993 | /* return non-0 since some features might have changed and |
9994 | * it's better to fire a spurious notification than miss it | |
9995 | */ | |
9996 | return -1; | |
6cb6a27c MM |
9997 | } |
9998 | ||
e7868a85 | 9999 | sync_lower: |
fd867d51 JW |
10000 | /* some features must be disabled on lower devices when disabled |
10001 | * on an upper device (think: bonding master or bridge) | |
10002 | */ | |
10003 | netdev_for_each_lower_dev(dev, lower, iter) | |
10004 | netdev_sync_lower_features(dev, lower, features); | |
10005 | ||
ae847f40 SD |
10006 | if (!err) { |
10007 | netdev_features_t diff = features ^ dev->features; | |
10008 | ||
10009 | if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
10010 | /* udp_tunnel_{get,drop}_rx_info both need | |
10011 | * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the | |
10012 | * device, or they won't do anything. | |
10013 | * Thus we need to update dev->features | |
10014 | * *before* calling udp_tunnel_get_rx_info, | |
10015 | * but *after* calling udp_tunnel_drop_rx_info. | |
10016 | */ | |
10017 | if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
10018 | dev->features = features; | |
10019 | udp_tunnel_get_rx_info(dev); | |
10020 | } else { | |
10021 | udp_tunnel_drop_rx_info(dev); | |
10022 | } | |
10023 | } | |
10024 | ||
9daae9bd GP |
10025 | if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { |
10026 | if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { | |
10027 | dev->features = features; | |
10028 | err |= vlan_get_rx_ctag_filter_info(dev); | |
10029 | } else { | |
10030 | vlan_drop_rx_ctag_filter_info(dev); | |
10031 | } | |
10032 | } | |
10033 | ||
10034 | if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { | |
10035 | if (features & NETIF_F_HW_VLAN_STAG_FILTER) { | |
10036 | dev->features = features; | |
10037 | err |= vlan_get_rx_stag_filter_info(dev); | |
10038 | } else { | |
10039 | vlan_drop_rx_stag_filter_info(dev); | |
10040 | } | |
10041 | } | |
10042 | ||
6cb6a27c | 10043 | dev->features = features; |
ae847f40 | 10044 | } |
6cb6a27c | 10045 | |
e7868a85 | 10046 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
10047 | } |
10048 | ||
afe12cc8 MM |
10049 | /** |
10050 | * netdev_update_features - recalculate device features | |
10051 | * @dev: the device to check | |
10052 | * | |
10053 | * Recalculate dev->features set and send notifications if it | |
10054 | * has changed. Should be called after driver or hardware dependent | |
10055 | * conditions might have changed that influence the features. | |
10056 | */ | |
6cb6a27c MM |
10057 | void netdev_update_features(struct net_device *dev) |
10058 | { | |
10059 | if (__netdev_update_features(dev)) | |
10060 | netdev_features_change(dev); | |
5455c699 MM |
10061 | } |
10062 | EXPORT_SYMBOL(netdev_update_features); | |
10063 | ||
afe12cc8 MM |
10064 | /** |
10065 | * netdev_change_features - recalculate device features | |
10066 | * @dev: the device to check | |
10067 | * | |
10068 | * Recalculate dev->features set and send notifications even | |
10069 | * if they have not changed. Should be called instead of | |
10070 | * netdev_update_features() if also dev->vlan_features might | |
10071 | * have changed to allow the changes to be propagated to stacked | |
10072 | * VLAN devices. | |
10073 | */ | |
10074 | void netdev_change_features(struct net_device *dev) | |
10075 | { | |
10076 | __netdev_update_features(dev); | |
10077 | netdev_features_change(dev); | |
10078 | } | |
10079 | EXPORT_SYMBOL(netdev_change_features); | |
10080 | ||
fc4a7489 PM |
10081 | /** |
10082 | * netif_stacked_transfer_operstate - transfer operstate | |
10083 | * @rootdev: the root or lower level device to transfer state from | |
10084 | * @dev: the device to transfer operstate to | |
10085 | * | |
10086 | * Transfer operational state from root to device. This is normally | |
10087 | * called when a stacking relationship exists between the root | |
10088 | * device and the device(a leaf device). | |
10089 | */ | |
10090 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
10091 | struct net_device *dev) | |
10092 | { | |
10093 | if (rootdev->operstate == IF_OPER_DORMANT) | |
10094 | netif_dormant_on(dev); | |
10095 | else | |
10096 | netif_dormant_off(dev); | |
10097 | ||
eec517cd AL |
10098 | if (rootdev->operstate == IF_OPER_TESTING) |
10099 | netif_testing_on(dev); | |
10100 | else | |
10101 | netif_testing_off(dev); | |
10102 | ||
0575c86b ZS |
10103 | if (netif_carrier_ok(rootdev)) |
10104 | netif_carrier_on(dev); | |
10105 | else | |
10106 | netif_carrier_off(dev); | |
fc4a7489 PM |
10107 | } |
10108 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
10109 | ||
1b4bf461 ED |
10110 | static int netif_alloc_rx_queues(struct net_device *dev) |
10111 | { | |
1b4bf461 | 10112 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 10113 | struct netdev_rx_queue *rx; |
10595902 | 10114 | size_t sz = count * sizeof(*rx); |
e817f856 | 10115 | int err = 0; |
1b4bf461 | 10116 | |
bd25fa7b | 10117 | BUG_ON(count < 1); |
1b4bf461 | 10118 | |
c948f51c | 10119 | rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
10120 | if (!rx) |
10121 | return -ENOMEM; | |
10122 | ||
bd25fa7b TH |
10123 | dev->_rx = rx; |
10124 | ||
e817f856 | 10125 | for (i = 0; i < count; i++) { |
fe822240 | 10126 | rx[i].dev = dev; |
e817f856 JDB |
10127 | |
10128 | /* XDP RX-queue setup */ | |
b02e5a0e | 10129 | err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); |
e817f856 JDB |
10130 | if (err < 0) |
10131 | goto err_rxq_info; | |
10132 | } | |
1b4bf461 | 10133 | return 0; |
e817f856 JDB |
10134 | |
10135 | err_rxq_info: | |
10136 | /* Rollback successful reg's and free other resources */ | |
10137 | while (i--) | |
10138 | xdp_rxq_info_unreg(&rx[i].xdp_rxq); | |
141b52a9 | 10139 | kvfree(dev->_rx); |
e817f856 JDB |
10140 | dev->_rx = NULL; |
10141 | return err; | |
10142 | } | |
10143 | ||
10144 | static void netif_free_rx_queues(struct net_device *dev) | |
10145 | { | |
10146 | unsigned int i, count = dev->num_rx_queues; | |
e817f856 JDB |
10147 | |
10148 | /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ | |
10149 | if (!dev->_rx) | |
10150 | return; | |
10151 | ||
e817f856 | 10152 | for (i = 0; i < count; i++) |
82aaff2f JK |
10153 | xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); |
10154 | ||
10155 | kvfree(dev->_rx); | |
1b4bf461 ED |
10156 | } |
10157 | ||
aa942104 CG |
10158 | static void netdev_init_one_queue(struct net_device *dev, |
10159 | struct netdev_queue *queue, void *_unused) | |
10160 | { | |
10161 | /* Initialize queue lock */ | |
10162 | spin_lock_init(&queue->_xmit_lock); | |
1a33e10e | 10163 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); |
aa942104 | 10164 | queue->xmit_lock_owner = -1; |
b236da69 | 10165 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 10166 | queue->dev = dev; |
114cf580 TH |
10167 | #ifdef CONFIG_BQL |
10168 | dql_init(&queue->dql, HZ); | |
10169 | #endif | |
aa942104 CG |
10170 | } |
10171 | ||
60877a32 ED |
10172 | static void netif_free_tx_queues(struct net_device *dev) |
10173 | { | |
4cb28970 | 10174 | kvfree(dev->_tx); |
60877a32 ED |
10175 | } |
10176 | ||
e6484930 TH |
10177 | static int netif_alloc_netdev_queues(struct net_device *dev) |
10178 | { | |
10179 | unsigned int count = dev->num_tx_queues; | |
10180 | struct netdev_queue *tx; | |
60877a32 | 10181 | size_t sz = count * sizeof(*tx); |
e6484930 | 10182 | |
d339727c ED |
10183 | if (count < 1 || count > 0xffff) |
10184 | return -EINVAL; | |
62b5942a | 10185 | |
c948f51c | 10186 | tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
10187 | if (!tx) |
10188 | return -ENOMEM; | |
10189 | ||
e6484930 | 10190 | dev->_tx = tx; |
1d24eb48 | 10191 | |
e6484930 TH |
10192 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
10193 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
10194 | |
10195 | return 0; | |
e6484930 TH |
10196 | } |
10197 | ||
a2029240 DV |
10198 | void netif_tx_stop_all_queues(struct net_device *dev) |
10199 | { | |
10200 | unsigned int i; | |
10201 | ||
10202 | for (i = 0; i < dev->num_tx_queues; i++) { | |
10203 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 10204 | |
a2029240 DV |
10205 | netif_tx_stop_queue(txq); |
10206 | } | |
10207 | } | |
10208 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
10209 | ||
34d21de9 DB |
10210 | static int netdev_do_alloc_pcpu_stats(struct net_device *dev) |
10211 | { | |
10212 | void __percpu *v; | |
10213 | ||
024ee930 PY |
10214 | /* Drivers implementing ndo_get_peer_dev must support tstat |
10215 | * accounting, so that skb_do_redirect() can bump the dev's | |
10216 | * RX stats upon network namespace switch. | |
10217 | */ | |
10218 | if (dev->netdev_ops->ndo_get_peer_dev && | |
10219 | dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) | |
10220 | return -EOPNOTSUPP; | |
10221 | ||
34d21de9 DB |
10222 | switch (dev->pcpu_stat_type) { |
10223 | case NETDEV_PCPU_STAT_NONE: | |
10224 | return 0; | |
10225 | case NETDEV_PCPU_STAT_LSTATS: | |
10226 | v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); | |
10227 | break; | |
10228 | case NETDEV_PCPU_STAT_TSTATS: | |
10229 | v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); | |
10230 | break; | |
10231 | case NETDEV_PCPU_STAT_DSTATS: | |
10232 | v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); | |
10233 | break; | |
10234 | default: | |
10235 | return -EINVAL; | |
10236 | } | |
10237 | ||
10238 | return v ? 0 : -ENOMEM; | |
10239 | } | |
10240 | ||
10241 | static void netdev_do_free_pcpu_stats(struct net_device *dev) | |
10242 | { | |
10243 | switch (dev->pcpu_stat_type) { | |
10244 | case NETDEV_PCPU_STAT_NONE: | |
10245 | return; | |
10246 | case NETDEV_PCPU_STAT_LSTATS: | |
10247 | free_percpu(dev->lstats); | |
10248 | break; | |
10249 | case NETDEV_PCPU_STAT_TSTATS: | |
10250 | free_percpu(dev->tstats); | |
10251 | break; | |
10252 | case NETDEV_PCPU_STAT_DSTATS: | |
10253 | free_percpu(dev->dstats); | |
10254 | break; | |
10255 | } | |
10256 | } | |
10257 | ||
1da177e4 | 10258 | /** |
fa926bb3 JK |
10259 | * register_netdevice() - register a network device |
10260 | * @dev: device to register | |
1da177e4 | 10261 | * |
fa926bb3 JK |
10262 | * Take a prepared network device structure and make it externally accessible. |
10263 | * A %NETDEV_REGISTER message is sent to the netdev notifier chain. | |
10264 | * Callers must hold the rtnl lock - you may want register_netdev() | |
10265 | * instead of this. | |
1da177e4 | 10266 | */ |
1da177e4 LT |
10267 | int register_netdevice(struct net_device *dev) |
10268 | { | |
1da177e4 | 10269 | int ret; |
d314774c | 10270 | struct net *net = dev_net(dev); |
1da177e4 | 10271 | |
e283de3a FF |
10272 | BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < |
10273 | NETDEV_FEATURE_COUNT); | |
1da177e4 LT |
10274 | BUG_ON(dev_boot_phase); |
10275 | ASSERT_RTNL(); | |
10276 | ||
b17a7c17 SH |
10277 | might_sleep(); |
10278 | ||
1da177e4 LT |
10279 | /* When net_device's are persistent, this will be fatal. */ |
10280 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 10281 | BUG_ON(!net); |
1da177e4 | 10282 | |
9000edb7 JK |
10283 | ret = ethtool_check_ops(dev->ethtool_ops); |
10284 | if (ret) | |
10285 | return ret; | |
10286 | ||
f1f28aa3 | 10287 | spin_lock_init(&dev->addr_list_lock); |
845e0ebb | 10288 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 10289 | |
828de4f6 | 10290 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
10291 | if (ret < 0) |
10292 | goto out; | |
10293 | ||
9077f052 | 10294 | ret = -ENOMEM; |
ff927412 JP |
10295 | dev->name_node = netdev_name_node_head_alloc(dev); |
10296 | if (!dev->name_node) | |
10297 | goto out; | |
10298 | ||
1da177e4 | 10299 | /* Init, if this function is available */ |
d314774c SH |
10300 | if (dev->netdev_ops->ndo_init) { |
10301 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
10302 | if (ret) { |
10303 | if (ret > 0) | |
10304 | ret = -EIO; | |
42c17fa6 | 10305 | goto err_free_name; |
1da177e4 LT |
10306 | } |
10307 | } | |
4ec93edb | 10308 | |
f646968f PM |
10309 | if (((dev->hw_features | dev->features) & |
10310 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
10311 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
10312 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
10313 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
10314 | ret = -EINVAL; | |
10315 | goto err_uninit; | |
10316 | } | |
10317 | ||
34d21de9 DB |
10318 | ret = netdev_do_alloc_pcpu_stats(dev); |
10319 | if (ret) | |
10320 | goto err_uninit; | |
10321 | ||
759ab1ed JK |
10322 | ret = dev_index_reserve(net, dev->ifindex); |
10323 | if (ret < 0) | |
34d21de9 | 10324 | goto err_free_pcpu; |
759ab1ed | 10325 | dev->ifindex = ret; |
9c7dafbf | 10326 | |
5455c699 MM |
10327 | /* Transfer changeable features to wanted_features and enable |
10328 | * software offloads (GSO and GRO). | |
10329 | */ | |
1a3c998f | 10330 | dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); |
14d1232f | 10331 | dev->features |= NETIF_F_SOFT_FEATURES; |
d764a122 | 10332 | |
876c4384 | 10333 | if (dev->udp_tunnel_nic_info) { |
d764a122 SD |
10334 | dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; |
10335 | dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
10336 | } | |
10337 | ||
14d1232f | 10338 | dev->wanted_features = dev->features & dev->hw_features; |
1da177e4 | 10339 | |
cbc53e08 | 10340 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 10341 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 10342 | |
7f348a60 AD |
10343 | /* If IPv4 TCP segmentation offload is supported we should also |
10344 | * allow the device to enable segmenting the frame with the option | |
10345 | * of ignoring a static IP ID value. This doesn't enable the | |
10346 | * feature itself but allows the user to enable it later. | |
10347 | */ | |
cbc53e08 AD |
10348 | if (dev->hw_features & NETIF_F_TSO) |
10349 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
10350 | if (dev->vlan_features & NETIF_F_TSO) |
10351 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
10352 | if (dev->mpls_features & NETIF_F_TSO) | |
10353 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
10354 | if (dev->hw_enc_features & NETIF_F_TSO) | |
10355 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 10356 | |
1180e7d6 | 10357 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 10358 | */ |
1180e7d6 | 10359 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 10360 | |
ee579677 PS |
10361 | /* Make NETIF_F_SG inheritable to tunnel devices. |
10362 | */ | |
802ab55a | 10363 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 10364 | |
0d89d203 SH |
10365 | /* Make NETIF_F_SG inheritable to MPLS. |
10366 | */ | |
10367 | dev->mpls_features |= NETIF_F_SG; | |
10368 | ||
7ffbe3fd JB |
10369 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
10370 | ret = notifier_to_errno(ret); | |
10371 | if (ret) | |
759ab1ed | 10372 | goto err_ifindex_release; |
7ffbe3fd | 10373 | |
8b41d188 | 10374 | ret = netdev_register_kobject(dev); |
e51b9624 | 10375 | |
4d42b37d | 10376 | WRITE_ONCE(dev->reg_state, ret ? NETREG_UNREGISTERED : NETREG_REGISTERED); |
e51b9624 | 10377 | |
cc26c266 | 10378 | if (ret) |
02a68a47 | 10379 | goto err_uninit_notify; |
b17a7c17 | 10380 | |
6cb6a27c | 10381 | __netdev_update_features(dev); |
8e9b59b2 | 10382 | |
1da177e4 LT |
10383 | /* |
10384 | * Default initial state at registry is that the | |
10385 | * device is present. | |
10386 | */ | |
10387 | ||
10388 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
10389 | ||
8f4cccbb BH |
10390 | linkwatch_init_dev(dev); |
10391 | ||
1da177e4 | 10392 | dev_init_scheduler(dev); |
b2309a71 | 10393 | |
d62607c3 | 10394 | netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); |
ce286d32 | 10395 | list_netdevice(dev); |
b2309a71 | 10396 | |
7bf23575 | 10397 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 10398 | |
948b337e JP |
10399 | /* If the device has permanent device address, driver should |
10400 | * set dev_addr and also addr_assign_type should be set to | |
10401 | * NET_ADDR_PERM (default value). | |
10402 | */ | |
10403 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
10404 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
10405 | ||
1da177e4 | 10406 | /* Notify protocols, that a new device appeared. */ |
056925ab | 10407 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 10408 | ret = notifier_to_errno(ret); |
93ee31f1 | 10409 | if (ret) { |
766b0515 JK |
10410 | /* Expect explicit free_netdev() on failure */ |
10411 | dev->needs_free_netdev = false; | |
037e56bd | 10412 | unregister_netdevice_queue(dev, NULL); |
766b0515 | 10413 | goto out; |
93ee31f1 | 10414 | } |
d90a909e EB |
10415 | /* |
10416 | * Prevent userspace races by waiting until the network | |
10417 | * device is fully setup before sending notifications. | |
10418 | */ | |
a2835763 PM |
10419 | if (!dev->rtnl_link_ops || |
10420 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
1d997f10 | 10421 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); |
1da177e4 LT |
10422 | |
10423 | out: | |
10424 | return ret; | |
7ce1b0ed | 10425 | |
02a68a47 JP |
10426 | err_uninit_notify: |
10427 | call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); | |
759ab1ed JK |
10428 | err_ifindex_release: |
10429 | dev_index_release(net, dev->ifindex); | |
34d21de9 DB |
10430 | err_free_pcpu: |
10431 | netdev_do_free_pcpu_stats(dev); | |
7ce1b0ed | 10432 | err_uninit: |
d314774c SH |
10433 | if (dev->netdev_ops->ndo_uninit) |
10434 | dev->netdev_ops->ndo_uninit(dev); | |
cf124db5 DM |
10435 | if (dev->priv_destructor) |
10436 | dev->priv_destructor(dev); | |
42c17fa6 DC |
10437 | err_free_name: |
10438 | netdev_name_node_free(dev->name_node); | |
7ce1b0ed | 10439 | goto out; |
1da177e4 | 10440 | } |
d1b19dff | 10441 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 10442 | |
c661050f BL |
10443 | /* Initialize the core of a dummy net device. |
10444 | * This is useful if you are calling this function after alloc_netdev(), | |
10445 | * since it does not memset the net_device fields. | |
937f1ba5 | 10446 | */ |
c661050f | 10447 | static void init_dummy_netdev_core(struct net_device *dev) |
937f1ba5 | 10448 | { |
937f1ba5 BH |
10449 | /* make sure we BUG if trying to hit standard |
10450 | * register/unregister code path | |
10451 | */ | |
10452 | dev->reg_state = NETREG_DUMMY; | |
10453 | ||
937f1ba5 BH |
10454 | /* NAPI wants this */ |
10455 | INIT_LIST_HEAD(&dev->napi_list); | |
10456 | ||
10457 | /* a dummy interface is started by default */ | |
10458 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
10459 | set_bit(__LINK_STATE_START, &dev->state); | |
10460 | ||
35edfdc7 JE |
10461 | /* napi_busy_loop stats accounting wants this */ |
10462 | dev_net_set(dev, &init_net); | |
10463 | ||
29b4433d ED |
10464 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
10465 | * because users of this 'device' dont need to change | |
10466 | * its refcount. | |
10467 | */ | |
937f1ba5 | 10468 | } |
937f1ba5 | 10469 | |
c661050f BL |
10470 | /** |
10471 | * init_dummy_netdev - init a dummy network device for NAPI | |
10472 | * @dev: device to init | |
10473 | * | |
10474 | * This takes a network device structure and initializes the minimum | |
10475 | * amount of fields so it can be used to schedule NAPI polls without | |
10476 | * registering a full blown interface. This is to be used by drivers | |
10477 | * that need to tie several hardware interfaces to a single NAPI | |
10478 | * poll scheduler due to HW limitations. | |
10479 | */ | |
10480 | void init_dummy_netdev(struct net_device *dev) | |
10481 | { | |
10482 | /* Clear everything. Note we don't initialize spinlocks | |
10483 | * as they aren't supposed to be taken by any of the | |
10484 | * NAPI code and this dummy netdev is supposed to be | |
10485 | * only ever used for NAPI polls | |
10486 | */ | |
10487 | memset(dev, 0, sizeof(struct net_device)); | |
10488 | init_dummy_netdev_core(dev); | |
10489 | } | |
10490 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
937f1ba5 | 10491 | |
1da177e4 LT |
10492 | /** |
10493 | * register_netdev - register a network device | |
10494 | * @dev: device to register | |
10495 | * | |
10496 | * Take a completed network device structure and add it to the kernel | |
10497 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
10498 | * chain. 0 is returned on success. A negative errno code is returned | |
10499 | * on a failure to set up the device, or if the name is a duplicate. | |
10500 | * | |
38b4da38 | 10501 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
10502 | * and expands the device name if you passed a format string to |
10503 | * alloc_netdev. | |
10504 | */ | |
10505 | int register_netdev(struct net_device *dev) | |
10506 | { | |
10507 | int err; | |
10508 | ||
b0f3debc KT |
10509 | if (rtnl_lock_killable()) |
10510 | return -EINTR; | |
1da177e4 | 10511 | err = register_netdevice(dev); |
1da177e4 LT |
10512 | rtnl_unlock(); |
10513 | return err; | |
10514 | } | |
10515 | EXPORT_SYMBOL(register_netdev); | |
10516 | ||
29b4433d ED |
10517 | int netdev_refcnt_read(const struct net_device *dev) |
10518 | { | |
919067cc | 10519 | #ifdef CONFIG_PCPU_DEV_REFCNT |
29b4433d ED |
10520 | int i, refcnt = 0; |
10521 | ||
10522 | for_each_possible_cpu(i) | |
10523 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
10524 | return refcnt; | |
919067cc ED |
10525 | #else |
10526 | return refcount_read(&dev->dev_refcnt); | |
10527 | #endif | |
29b4433d ED |
10528 | } |
10529 | EXPORT_SYMBOL(netdev_refcnt_read); | |
10530 | ||
5aa3afe1 DV |
10531 | int netdev_unregister_timeout_secs __read_mostly = 10; |
10532 | ||
de2b541b MCC |
10533 | #define WAIT_REFS_MIN_MSECS 1 |
10534 | #define WAIT_REFS_MAX_MSECS 250 | |
2c53040f | 10535 | /** |
faab39f6 JK |
10536 | * netdev_wait_allrefs_any - wait until all references are gone. |
10537 | * @list: list of net_devices to wait on | |
1da177e4 LT |
10538 | * |
10539 | * This is called when unregistering network devices. | |
10540 | * | |
10541 | * Any protocol or device that holds a reference should register | |
10542 | * for netdevice notification, and cleanup and put back the | |
10543 | * reference if they receive an UNREGISTER event. | |
10544 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 10545 | * call dev_put. |
1da177e4 | 10546 | */ |
faab39f6 | 10547 | static struct net_device *netdev_wait_allrefs_any(struct list_head *list) |
1da177e4 LT |
10548 | { |
10549 | unsigned long rebroadcast_time, warning_time; | |
faab39f6 JK |
10550 | struct net_device *dev; |
10551 | int wait = 0; | |
1da177e4 LT |
10552 | |
10553 | rebroadcast_time = warning_time = jiffies; | |
29b4433d | 10554 | |
faab39f6 JK |
10555 | list_for_each_entry(dev, list, todo_list) |
10556 | if (netdev_refcnt_read(dev) == 1) | |
10557 | return dev; | |
10558 | ||
10559 | while (true) { | |
1da177e4 | 10560 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 10561 | rtnl_lock(); |
1da177e4 LT |
10562 | |
10563 | /* Rebroadcast unregister notification */ | |
faab39f6 JK |
10564 | list_for_each_entry(dev, list, todo_list) |
10565 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
1da177e4 | 10566 | |
748e2d93 | 10567 | __rtnl_unlock(); |
0115e8e3 | 10568 | rcu_barrier(); |
748e2d93 ED |
10569 | rtnl_lock(); |
10570 | ||
faab39f6 JK |
10571 | list_for_each_entry(dev, list, todo_list) |
10572 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
10573 | &dev->state)) { | |
10574 | /* We must not have linkwatch events | |
10575 | * pending on unregister. If this | |
10576 | * happens, we simply run the queue | |
10577 | * unscheduled, resulting in a noop | |
10578 | * for this device. | |
10579 | */ | |
10580 | linkwatch_run_queue(); | |
10581 | break; | |
10582 | } | |
1da177e4 | 10583 | |
6756ae4b | 10584 | __rtnl_unlock(); |
1da177e4 LT |
10585 | |
10586 | rebroadcast_time = jiffies; | |
10587 | } | |
10588 | ||
cd42ba1c ED |
10589 | rcu_barrier(); |
10590 | ||
0e4be9e5 | 10591 | if (!wait) { |
0e4be9e5 FR |
10592 | wait = WAIT_REFS_MIN_MSECS; |
10593 | } else { | |
10594 | msleep(wait); | |
10595 | wait = min(wait << 1, WAIT_REFS_MAX_MSECS); | |
10596 | } | |
1da177e4 | 10597 | |
faab39f6 JK |
10598 | list_for_each_entry(dev, list, todo_list) |
10599 | if (netdev_refcnt_read(dev) == 1) | |
10600 | return dev; | |
29b4433d | 10601 | |
faab39f6 | 10602 | if (time_after(jiffies, warning_time + |
05e49cfc | 10603 | READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { |
faab39f6 JK |
10604 | list_for_each_entry(dev, list, todo_list) { |
10605 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", | |
10606 | dev->name, netdev_refcnt_read(dev)); | |
10607 | ref_tracker_dir_print(&dev->refcnt_tracker, 10); | |
10608 | } | |
10609 | ||
1da177e4 LT |
10610 | warning_time = jiffies; |
10611 | } | |
10612 | } | |
10613 | } | |
10614 | ||
10615 | /* The sequence is: | |
10616 | * | |
10617 | * rtnl_lock(); | |
10618 | * ... | |
10619 | * register_netdevice(x1); | |
10620 | * register_netdevice(x2); | |
10621 | * ... | |
10622 | * unregister_netdevice(y1); | |
10623 | * unregister_netdevice(y2); | |
10624 | * ... | |
10625 | * rtnl_unlock(); | |
10626 | * free_netdev(y1); | |
10627 | * free_netdev(y2); | |
10628 | * | |
58ec3b4d | 10629 | * We are invoked by rtnl_unlock(). |
1da177e4 | 10630 | * This allows us to deal with problems: |
b17a7c17 | 10631 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
10632 | * without deadlocking with linkwatch via keventd. |
10633 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
10634 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
10635 | * |
10636 | * We must not return until all unregister events added during | |
10637 | * the interval the lock was held have been completed. | |
1da177e4 | 10638 | */ |
1da177e4 LT |
10639 | void netdev_run_todo(void) |
10640 | { | |
ae68db14 | 10641 | struct net_device *dev, *tmp; |
626ab0e6 | 10642 | struct list_head list; |
ffabe98c | 10643 | int cnt; |
1fc70edb TY |
10644 | #ifdef CONFIG_LOCKDEP |
10645 | struct list_head unlink_list; | |
10646 | ||
10647 | list_replace_init(&net_unlink_list, &unlink_list); | |
10648 | ||
10649 | while (!list_empty(&unlink_list)) { | |
10650 | struct net_device *dev = list_first_entry(&unlink_list, | |
10651 | struct net_device, | |
10652 | unlink_list); | |
0e8b8d6a | 10653 | list_del_init(&dev->unlink_list); |
1fc70edb TY |
10654 | dev->nested_level = dev->lower_level - 1; |
10655 | } | |
10656 | #endif | |
1da177e4 | 10657 | |
1da177e4 | 10658 | /* Snapshot list, allow later requests */ |
626ab0e6 | 10659 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
10660 | |
10661 | __rtnl_unlock(); | |
626ab0e6 | 10662 | |
0115e8e3 | 10663 | /* Wait for rcu callbacks to finish before next phase */ |
850a545b EB |
10664 | if (!list_empty(&list)) |
10665 | rcu_barrier(); | |
10666 | ||
ae68db14 | 10667 | list_for_each_entry_safe(dev, tmp, &list, todo_list) { |
b17a7c17 | 10668 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
ae68db14 JK |
10669 | netdev_WARN(dev, "run_todo but not unregistering\n"); |
10670 | list_del(&dev->todo_list); | |
b17a7c17 SH |
10671 | continue; |
10672 | } | |
1da177e4 | 10673 | |
4d42b37d | 10674 | WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERED); |
facd15df | 10675 | linkwatch_sync_dev(dev); |
ae68db14 JK |
10676 | } |
10677 | ||
ffabe98c | 10678 | cnt = 0; |
ae68db14 | 10679 | while (!list_empty(&list)) { |
faab39f6 | 10680 | dev = netdev_wait_allrefs_any(&list); |
ae68db14 | 10681 | list_del(&dev->todo_list); |
1da177e4 | 10682 | |
b17a7c17 | 10683 | /* paranoia */ |
add2d736 | 10684 | BUG_ON(netdev_refcnt_read(dev) != 1); |
7866a621 SN |
10685 | BUG_ON(!list_empty(&dev->ptype_all)); |
10686 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
10687 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
10688 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
1202cdd6 | 10689 | |
34d21de9 | 10690 | netdev_do_free_pcpu_stats(dev); |
cf124db5 DM |
10691 | if (dev->priv_destructor) |
10692 | dev->priv_destructor(dev); | |
10693 | if (dev->needs_free_netdev) | |
10694 | free_netdev(dev); | |
9093bbb2 | 10695 | |
ffabe98c | 10696 | cnt++; |
50624c93 | 10697 | |
9093bbb2 SH |
10698 | /* Free network device */ |
10699 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 10700 | } |
ffabe98c ED |
10701 | if (cnt && atomic_sub_and_test(cnt, &dev_unreg_count)) |
10702 | wake_up(&netdev_unregistering_wq); | |
1da177e4 LT |
10703 | } |
10704 | ||
9256645a JW |
10705 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
10706 | * all the same fields in the same order as net_device_stats, with only | |
10707 | * the type differing, but rtnl_link_stats64 may have additional fields | |
10708 | * at the end for newer counters. | |
3cfde79c | 10709 | */ |
77a1abf5 ED |
10710 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
10711 | const struct net_device_stats *netdev_stats) | |
3cfde79c | 10712 | { |
6c1c5097 ED |
10713 | size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); |
10714 | const atomic_long_t *src = (atomic_long_t *)netdev_stats; | |
3cfde79c BH |
10715 | u64 *dst = (u64 *)stats64; |
10716 | ||
9256645a | 10717 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c | 10718 | for (i = 0; i < n; i++) |
9b55d3f0 | 10719 | dst[i] = (unsigned long)atomic_long_read(&src[i]); |
9256645a JW |
10720 | /* zero out counters that only exist in rtnl_link_stats64 */ |
10721 | memset((char *)stats64 + n * sizeof(u64), 0, | |
10722 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c | 10723 | } |
77a1abf5 | 10724 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 10725 | |
5247dbf1 YD |
10726 | static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc( |
10727 | struct net_device *dev) | |
625788b5 ED |
10728 | { |
10729 | struct net_device_core_stats __percpu *p; | |
10730 | ||
10731 | p = alloc_percpu_gfp(struct net_device_core_stats, | |
10732 | GFP_ATOMIC | __GFP_NOWARN); | |
10733 | ||
10734 | if (p && cmpxchg(&dev->core_stats, NULL, p)) | |
10735 | free_percpu(p); | |
10736 | ||
10737 | /* This READ_ONCE() pairs with the cmpxchg() above */ | |
6510ea97 | 10738 | return READ_ONCE(dev->core_stats); |
625788b5 | 10739 | } |
5247dbf1 YD |
10740 | |
10741 | noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset) | |
10742 | { | |
10743 | /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ | |
10744 | struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats); | |
10745 | unsigned long __percpu *field; | |
10746 | ||
10747 | if (unlikely(!p)) { | |
10748 | p = netdev_core_stats_alloc(dev); | |
10749 | if (!p) | |
10750 | return; | |
10751 | } | |
10752 | ||
10753 | field = (__force unsigned long __percpu *)((__force void *)p + offset); | |
10754 | this_cpu_inc(*field); | |
10755 | } | |
10756 | EXPORT_SYMBOL_GPL(netdev_core_stats_inc); | |
625788b5 | 10757 | |
eeda3fd6 SH |
10758 | /** |
10759 | * dev_get_stats - get network device statistics | |
10760 | * @dev: device to get statistics from | |
28172739 | 10761 | * @storage: place to store stats |
eeda3fd6 | 10762 | * |
d7753516 BH |
10763 | * Get network statistics from device. Return @storage. |
10764 | * The device driver may provide its own method by setting | |
10765 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
10766 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 10767 | */ |
d7753516 BH |
10768 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
10769 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 10770 | { |
eeda3fd6 | 10771 | const struct net_device_ops *ops = dev->netdev_ops; |
625788b5 | 10772 | const struct net_device_core_stats __percpu *p; |
eeda3fd6 | 10773 | |
28172739 ED |
10774 | if (ops->ndo_get_stats64) { |
10775 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
10776 | ops->ndo_get_stats64(dev, storage); |
10777 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 10778 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
3e2f544d BL |
10779 | } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_TSTATS) { |
10780 | dev_get_tstats64(dev, storage); | |
caf586e5 ED |
10781 | } else { |
10782 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 10783 | } |
625788b5 ED |
10784 | |
10785 | /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ | |
10786 | p = READ_ONCE(dev->core_stats); | |
10787 | if (p) { | |
10788 | const struct net_device_core_stats *core_stats; | |
10789 | int i; | |
10790 | ||
10791 | for_each_possible_cpu(i) { | |
10792 | core_stats = per_cpu_ptr(p, i); | |
6510ea97 SAS |
10793 | storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); |
10794 | storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); | |
10795 | storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); | |
0e55546b | 10796 | storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); |
625788b5 ED |
10797 | } |
10798 | } | |
28172739 | 10799 | return storage; |
c45d286e | 10800 | } |
eeda3fd6 | 10801 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 10802 | |
44fa32f0 HK |
10803 | /** |
10804 | * dev_fetch_sw_netstats - get per-cpu network device statistics | |
10805 | * @s: place to store stats | |
10806 | * @netstats: per-cpu network stats to read from | |
10807 | * | |
10808 | * Read per-cpu network statistics and populate the related fields in @s. | |
10809 | */ | |
10810 | void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, | |
10811 | const struct pcpu_sw_netstats __percpu *netstats) | |
10812 | { | |
10813 | int cpu; | |
10814 | ||
10815 | for_each_possible_cpu(cpu) { | |
9962acef | 10816 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; |
44fa32f0 | 10817 | const struct pcpu_sw_netstats *stats; |
44fa32f0 HK |
10818 | unsigned int start; |
10819 | ||
10820 | stats = per_cpu_ptr(netstats, cpu); | |
10821 | do { | |
d120d1a6 | 10822 | start = u64_stats_fetch_begin(&stats->syncp); |
9962acef ED |
10823 | rx_packets = u64_stats_read(&stats->rx_packets); |
10824 | rx_bytes = u64_stats_read(&stats->rx_bytes); | |
10825 | tx_packets = u64_stats_read(&stats->tx_packets); | |
10826 | tx_bytes = u64_stats_read(&stats->tx_bytes); | |
d120d1a6 | 10827 | } while (u64_stats_fetch_retry(&stats->syncp, start)); |
44fa32f0 | 10828 | |
9962acef ED |
10829 | s->rx_packets += rx_packets; |
10830 | s->rx_bytes += rx_bytes; | |
10831 | s->tx_packets += tx_packets; | |
10832 | s->tx_bytes += tx_bytes; | |
44fa32f0 HK |
10833 | } |
10834 | } | |
10835 | EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); | |
10836 | ||
a1839426 HK |
10837 | /** |
10838 | * dev_get_tstats64 - ndo_get_stats64 implementation | |
10839 | * @dev: device to get statistics from | |
10840 | * @s: place to store stats | |
10841 | * | |
10842 | * Populate @s from dev->stats and dev->tstats. Can be used as | |
10843 | * ndo_get_stats64() callback. | |
10844 | */ | |
10845 | void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) | |
10846 | { | |
10847 | netdev_stats_to_stats64(s, &dev->stats); | |
10848 | dev_fetch_sw_netstats(s, dev->tstats); | |
10849 | } | |
10850 | EXPORT_SYMBOL_GPL(dev_get_tstats64); | |
10851 | ||
24824a09 | 10852 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 10853 | { |
24824a09 | 10854 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 10855 | |
24824a09 ED |
10856 | #ifdef CONFIG_NET_CLS_ACT |
10857 | if (queue) | |
10858 | return queue; | |
10859 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
10860 | if (!queue) | |
10861 | return NULL; | |
10862 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 10863 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
d636fc5d | 10864 | RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); |
24824a09 ED |
10865 | rcu_assign_pointer(dev->ingress_queue, queue); |
10866 | #endif | |
10867 | return queue; | |
bb949fbd DM |
10868 | } |
10869 | ||
2c60db03 ED |
10870 | static const struct ethtool_ops default_ethtool_ops; |
10871 | ||
d07d7507 SG |
10872 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
10873 | const struct ethtool_ops *ops) | |
10874 | { | |
10875 | if (dev->ethtool_ops == &default_ethtool_ops) | |
10876 | dev->ethtool_ops = ops; | |
10877 | } | |
10878 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
10879 | ||
d9360708 HK |
10880 | /** |
10881 | * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default | |
10882 | * @dev: netdev to enable the IRQ coalescing on | |
10883 | * | |
10884 | * Sets a conservative default for SW IRQ coalescing. Users can use | |
10885 | * sysfs attributes to override the default values. | |
10886 | */ | |
10887 | void netdev_sw_irq_coalesce_default_on(struct net_device *dev) | |
10888 | { | |
10889 | WARN_ON(dev->reg_state == NETREG_REGISTERED); | |
10890 | ||
748b4428 HK |
10891 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { |
10892 | dev->gro_flush_timeout = 20000; | |
10893 | dev->napi_defer_hard_irqs = 1; | |
10894 | } | |
d9360708 HK |
10895 | } |
10896 | EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); | |
10897 | ||
74d332c1 ED |
10898 | void netdev_freemem(struct net_device *dev) |
10899 | { | |
10900 | char *addr = (char *)dev - dev->padded; | |
10901 | ||
4cb28970 | 10902 | kvfree(addr); |
74d332c1 ED |
10903 | } |
10904 | ||
1da177e4 | 10905 | /** |
722c9a0c | 10906 | * alloc_netdev_mqs - allocate network device |
10907 | * @sizeof_priv: size of private data to allocate space for | |
10908 | * @name: device name format string | |
10909 | * @name_assign_type: origin of device name | |
10910 | * @setup: callback to initialize device | |
10911 | * @txqs: the number of TX subqueues to allocate | |
10912 | * @rxqs: the number of RX subqueues to allocate | |
10913 | * | |
10914 | * Allocates a struct net_device with private data area for driver use | |
10915 | * and performs basic initialization. Also allocates subqueue structs | |
10916 | * for each queue on the device. | |
1da177e4 | 10917 | */ |
36909ea4 | 10918 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 10919 | unsigned char name_assign_type, |
36909ea4 TH |
10920 | void (*setup)(struct net_device *), |
10921 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 10922 | { |
1da177e4 | 10923 | struct net_device *dev; |
52a59bd5 | 10924 | unsigned int alloc_size; |
1ce8e7b5 | 10925 | struct net_device *p; |
1da177e4 | 10926 | |
b6fe17d6 SH |
10927 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
10928 | ||
36909ea4 | 10929 | if (txqs < 1) { |
7b6cd1ce | 10930 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
10931 | return NULL; |
10932 | } | |
10933 | ||
36909ea4 | 10934 | if (rxqs < 1) { |
7b6cd1ce | 10935 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
10936 | return NULL; |
10937 | } | |
36909ea4 | 10938 | |
fd2ea0a7 | 10939 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
10940 | if (sizeof_priv) { |
10941 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 10942 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
10943 | alloc_size += sizeof_priv; |
10944 | } | |
10945 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 10946 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 10947 | |
c948f51c | 10948 | p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); |
62b5942a | 10949 | if (!p) |
1da177e4 | 10950 | return NULL; |
1da177e4 | 10951 | |
1ce8e7b5 | 10952 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 10953 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 10954 | |
b6d7c0eb | 10955 | ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); |
919067cc | 10956 | #ifdef CONFIG_PCPU_DEV_REFCNT |
29b4433d ED |
10957 | dev->pcpu_refcnt = alloc_percpu(int); |
10958 | if (!dev->pcpu_refcnt) | |
74d332c1 | 10959 | goto free_dev; |
4c6c11ea | 10960 | __dev_hold(dev); |
add2d736 ED |
10961 | #else |
10962 | refcount_set(&dev->dev_refcnt, 1); | |
919067cc | 10963 | #endif |
ab9c73cc | 10964 | |
ab9c73cc | 10965 | if (dev_addr_init(dev)) |
29b4433d | 10966 | goto free_pcpu; |
ab9c73cc | 10967 | |
22bedad3 | 10968 | dev_mc_init(dev); |
a748ee24 | 10969 | dev_uc_init(dev); |
ccffad25 | 10970 | |
c346dca1 | 10971 | dev_net_set(dev, &init_net); |
1da177e4 | 10972 | |
7c4e983c | 10973 | dev->gso_max_size = GSO_LEGACY_MAX_SIZE; |
13ce2daa | 10974 | dev->xdp_zc_max_segs = 1; |
30b678d8 | 10975 | dev->gso_max_segs = GSO_MAX_SEGS; |
0fe79f28 | 10976 | dev->gro_max_size = GRO_LEGACY_MAX_SIZE; |
9eefedd5 XL |
10977 | dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; |
10978 | dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; | |
14d7b812 JK |
10979 | dev->tso_max_size = TSO_LEGACY_MAX_SIZE; |
10980 | dev->tso_max_segs = TSO_MAX_SEGS; | |
5343da4c TY |
10981 | dev->upper_level = 1; |
10982 | dev->lower_level = 1; | |
1fc70edb TY |
10983 | #ifdef CONFIG_LOCKDEP |
10984 | dev->nested_level = 0; | |
10985 | INIT_LIST_HEAD(&dev->unlink_list); | |
10986 | #endif | |
8d3bdbd5 | 10987 | |
8d3bdbd5 DM |
10988 | INIT_LIST_HEAD(&dev->napi_list); |
10989 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 10990 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 10991 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
10992 | INIT_LIST_HEAD(&dev->adj_list.upper); |
10993 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
10994 | INIT_LIST_HEAD(&dev->ptype_all); |
10995 | INIT_LIST_HEAD(&dev->ptype_specific); | |
93642e14 | 10996 | INIT_LIST_HEAD(&dev->net_notifier_list); |
59cc1f61 JK |
10997 | #ifdef CONFIG_NET_SCHED |
10998 | hash_init(dev->qdisc_hash); | |
10999 | #endif | |
02875878 | 11000 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
11001 | setup(dev); |
11002 | ||
a813104d | 11003 | if (!dev->tx_queue_len) { |
f84bb1ea | 11004 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 11005 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 11006 | } |
906470c1 | 11007 | |
36909ea4 TH |
11008 | dev->num_tx_queues = txqs; |
11009 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 11010 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 11011 | goto free_all; |
e8a0464c | 11012 | |
36909ea4 TH |
11013 | dev->num_rx_queues = rxqs; |
11014 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 11015 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 11016 | goto free_all; |
0a9627f2 | 11017 | |
1da177e4 | 11018 | strcpy(dev->name, name); |
c835a677 | 11019 | dev->name_assign_type = name_assign_type; |
cbda10fa | 11020 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
11021 | if (!dev->ethtool_ops) |
11022 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 | 11023 | |
17d20784 | 11024 | nf_hook_netdev_init(dev); |
e687ad60 | 11025 | |
1da177e4 | 11026 | return dev; |
ab9c73cc | 11027 | |
8d3bdbd5 DM |
11028 | free_all: |
11029 | free_netdev(dev); | |
11030 | return NULL; | |
11031 | ||
29b4433d | 11032 | free_pcpu: |
919067cc | 11033 | #ifdef CONFIG_PCPU_DEV_REFCNT |
29b4433d | 11034 | free_percpu(dev->pcpu_refcnt); |
74d332c1 | 11035 | free_dev: |
919067cc | 11036 | #endif |
74d332c1 | 11037 | netdev_freemem(dev); |
ab9c73cc | 11038 | return NULL; |
1da177e4 | 11039 | } |
36909ea4 | 11040 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
11041 | |
11042 | /** | |
722c9a0c | 11043 | * free_netdev - free network device |
11044 | * @dev: device | |
1da177e4 | 11045 | * |
722c9a0c | 11046 | * This function does the last stage of destroying an allocated device |
11047 | * interface. The reference to the device object is released. If this | |
11048 | * is the last reference then it will be freed.Must be called in process | |
11049 | * context. | |
1da177e4 LT |
11050 | */ |
11051 | void free_netdev(struct net_device *dev) | |
11052 | { | |
d565b0a1 HX |
11053 | struct napi_struct *p, *n; |
11054 | ||
93d05d4a | 11055 | might_sleep(); |
c269a24c JK |
11056 | |
11057 | /* When called immediately after register_netdevice() failed the unwind | |
11058 | * handling may still be dismantling the device. Handle that case by | |
11059 | * deferring the free. | |
11060 | */ | |
11061 | if (dev->reg_state == NETREG_UNREGISTERING) { | |
11062 | ASSERT_RTNL(); | |
11063 | dev->needs_free_netdev = true; | |
11064 | return; | |
11065 | } | |
11066 | ||
60877a32 | 11067 | netif_free_tx_queues(dev); |
e817f856 | 11068 | netif_free_rx_queues(dev); |
e8a0464c | 11069 | |
33d480ce | 11070 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 11071 | |
f001fde5 JP |
11072 | /* Flush device addresses */ |
11073 | dev_addr_flush(dev); | |
11074 | ||
d565b0a1 HX |
11075 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
11076 | netif_napi_del(p); | |
11077 | ||
4d92b95f | 11078 | ref_tracker_dir_exit(&dev->refcnt_tracker); |
919067cc | 11079 | #ifdef CONFIG_PCPU_DEV_REFCNT |
29b4433d ED |
11080 | free_percpu(dev->pcpu_refcnt); |
11081 | dev->pcpu_refcnt = NULL; | |
919067cc | 11082 | #endif |
625788b5 ED |
11083 | free_percpu(dev->core_stats); |
11084 | dev->core_stats = NULL; | |
75ccae62 THJ |
11085 | free_percpu(dev->xdp_bulkq); |
11086 | dev->xdp_bulkq = NULL; | |
29b4433d | 11087 | |
3041a069 | 11088 | /* Compatibility with error handling in drivers */ |
f8d05679 BL |
11089 | if (dev->reg_state == NETREG_UNINITIALIZED || |
11090 | dev->reg_state == NETREG_DUMMY) { | |
74d332c1 | 11091 | netdev_freemem(dev); |
1da177e4 LT |
11092 | return; |
11093 | } | |
11094 | ||
11095 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
4d42b37d | 11096 | WRITE_ONCE(dev->reg_state, NETREG_RELEASED); |
1da177e4 | 11097 | |
43cb76d9 GKH |
11098 | /* will free via device release */ |
11099 | put_device(&dev->dev); | |
1da177e4 | 11100 | } |
d1b19dff | 11101 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 11102 | |
c661050f BL |
11103 | /** |
11104 | * alloc_netdev_dummy - Allocate and initialize a dummy net device. | |
11105 | * @sizeof_priv: size of private data to allocate space for | |
11106 | * | |
11107 | * Return: the allocated net_device on success, NULL otherwise | |
11108 | */ | |
11109 | struct net_device *alloc_netdev_dummy(int sizeof_priv) | |
11110 | { | |
11111 | return alloc_netdev(sizeof_priv, "dummy#", NET_NAME_UNKNOWN, | |
11112 | init_dummy_netdev_core); | |
11113 | } | |
11114 | EXPORT_SYMBOL_GPL(alloc_netdev_dummy); | |
11115 | ||
f0db275a SH |
11116 | /** |
11117 | * synchronize_net - Synchronize with packet receive processing | |
11118 | * | |
11119 | * Wait for packets currently being received to be done. | |
11120 | * Does not block later packets from starting. | |
11121 | */ | |
4ec93edb | 11122 | void synchronize_net(void) |
1da177e4 LT |
11123 | { |
11124 | might_sleep(); | |
be3fc413 ED |
11125 | if (rtnl_is_locked()) |
11126 | synchronize_rcu_expedited(); | |
11127 | else | |
11128 | synchronize_rcu(); | |
1da177e4 | 11129 | } |
d1b19dff | 11130 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
11131 | |
11132 | /** | |
44a0873d | 11133 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 11134 | * @dev: device |
44a0873d | 11135 | * @head: list |
6ebfbc06 | 11136 | * |
1da177e4 | 11137 | * This function shuts down a device interface and removes it |
d59b54b1 | 11138 | * from the kernel tables. |
44a0873d | 11139 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
11140 | * |
11141 | * Callers must hold the rtnl semaphore. You may want | |
11142 | * unregister_netdev() instead of this. | |
11143 | */ | |
11144 | ||
44a0873d | 11145 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 11146 | { |
a6620712 HX |
11147 | ASSERT_RTNL(); |
11148 | ||
44a0873d | 11149 | if (head) { |
9fdce099 | 11150 | list_move_tail(&dev->unreg_list, head); |
44a0873d | 11151 | } else { |
037e56bd JK |
11152 | LIST_HEAD(single); |
11153 | ||
11154 | list_add(&dev->unreg_list, &single); | |
0cbe1e57 | 11155 | unregister_netdevice_many(&single); |
44a0873d | 11156 | } |
1da177e4 | 11157 | } |
44a0873d | 11158 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 11159 | |
77f4aa9a HL |
11160 | void unregister_netdevice_many_notify(struct list_head *head, |
11161 | u32 portid, const struct nlmsghdr *nlh) | |
bcfe2f1a JK |
11162 | { |
11163 | struct net_device *dev, *tmp; | |
11164 | LIST_HEAD(close_head); | |
ffabe98c | 11165 | int cnt = 0; |
bcfe2f1a JK |
11166 | |
11167 | BUG_ON(dev_boot_phase); | |
11168 | ASSERT_RTNL(); | |
11169 | ||
0cbe1e57 JK |
11170 | if (list_empty(head)) |
11171 | return; | |
11172 | ||
bcfe2f1a JK |
11173 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
11174 | /* Some devices call without registering | |
11175 | * for initialization unwind. Remove those | |
11176 | * devices and proceed with the remaining. | |
11177 | */ | |
11178 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
11179 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", | |
11180 | dev->name, dev); | |
11181 | ||
11182 | WARN_ON(1); | |
11183 | list_del(&dev->unreg_list); | |
11184 | continue; | |
11185 | } | |
11186 | dev->dismantle = true; | |
11187 | BUG_ON(dev->reg_state != NETREG_REGISTERED); | |
11188 | } | |
11189 | ||
11190 | /* If device is running, close it first. */ | |
11191 | list_for_each_entry(dev, head, unreg_list) | |
11192 | list_add_tail(&dev->close_list, &close_head); | |
11193 | dev_close_many(&close_head, true); | |
11194 | ||
11195 | list_for_each_entry(dev, head, unreg_list) { | |
11196 | /* And unlink it from device chain. */ | |
e51b9624 | 11197 | unlist_netdevice(dev); |
4d42b37d | 11198 | WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERING); |
bcfe2f1a JK |
11199 | } |
11200 | flush_all_backlogs(); | |
11201 | ||
11202 | synchronize_net(); | |
11203 | ||
11204 | list_for_each_entry(dev, head, unreg_list) { | |
11205 | struct sk_buff *skb = NULL; | |
11206 | ||
11207 | /* Shutdown queueing discipline. */ | |
11208 | dev_shutdown(dev); | |
e420bed0 | 11209 | dev_tcx_uninstall(dev); |
bcfe2f1a | 11210 | dev_xdp_uninstall(dev); |
2b3486bc | 11211 | bpf_dev_bound_netdev_unregister(dev); |
bcfe2f1a | 11212 | |
9309f97a PM |
11213 | netdev_offload_xstats_disable_all(dev); |
11214 | ||
bcfe2f1a JK |
11215 | /* Notify protocols, that we are about to destroy |
11216 | * this device. They should clean all the things. | |
11217 | */ | |
11218 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
11219 | ||
11220 | if (!dev->rtnl_link_ops || | |
11221 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
11222 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, | |
77f4aa9a | 11223 | GFP_KERNEL, NULL, 0, |
59d3efd2 | 11224 | portid, nlh); |
bcfe2f1a JK |
11225 | |
11226 | /* | |
11227 | * Flush the unicast and multicast chains | |
11228 | */ | |
11229 | dev_uc_flush(dev); | |
11230 | dev_mc_flush(dev); | |
11231 | ||
11232 | netdev_name_node_alt_flush(dev); | |
11233 | netdev_name_node_free(dev->name_node); | |
11234 | ||
02a68a47 JP |
11235 | call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); |
11236 | ||
bcfe2f1a JK |
11237 | if (dev->netdev_ops->ndo_uninit) |
11238 | dev->netdev_ops->ndo_uninit(dev); | |
11239 | ||
11240 | if (skb) | |
77f4aa9a | 11241 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); |
bcfe2f1a JK |
11242 | |
11243 | /* Notifier chain MUST detach us all upper devices. */ | |
11244 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
11245 | WARN_ON(netdev_has_any_lower_dev(dev)); | |
11246 | ||
11247 | /* Remove entries from kobject tree */ | |
11248 | netdev_unregister_kobject(dev); | |
11249 | #ifdef CONFIG_XPS | |
11250 | /* Remove XPS queueing entries */ | |
11251 | netif_reset_xps_queues_gt(dev, 0); | |
11252 | #endif | |
11253 | } | |
11254 | ||
11255 | synchronize_net(); | |
11256 | ||
11257 | list_for_each_entry(dev, head, unreg_list) { | |
d62607c3 | 11258 | netdev_put(dev, &dev->dev_registered_tracker); |
bcfe2f1a | 11259 | net_set_todo(dev); |
ffabe98c | 11260 | cnt++; |
bcfe2f1a | 11261 | } |
ffabe98c | 11262 | atomic_add(cnt, &dev_unreg_count); |
0cbe1e57 JK |
11263 | |
11264 | list_del(head); | |
bcfe2f1a | 11265 | } |
77f4aa9a HL |
11266 | |
11267 | /** | |
11268 | * unregister_netdevice_many - unregister many devices | |
11269 | * @head: list of devices | |
11270 | * | |
11271 | * Note: As most callers use a stack allocated list_head, | |
11272 | * we force a list_del() to make sure stack wont be corrupted later. | |
11273 | */ | |
11274 | void unregister_netdevice_many(struct list_head *head) | |
11275 | { | |
11276 | unregister_netdevice_many_notify(head, 0, NULL); | |
11277 | } | |
0cbe1e57 | 11278 | EXPORT_SYMBOL(unregister_netdevice_many); |
bcfe2f1a | 11279 | |
1da177e4 LT |
11280 | /** |
11281 | * unregister_netdev - remove device from the kernel | |
11282 | * @dev: device | |
11283 | * | |
11284 | * This function shuts down a device interface and removes it | |
d59b54b1 | 11285 | * from the kernel tables. |
1da177e4 LT |
11286 | * |
11287 | * This is just a wrapper for unregister_netdevice that takes | |
11288 | * the rtnl semaphore. In general you want to use this and not | |
11289 | * unregister_netdevice. | |
11290 | */ | |
11291 | void unregister_netdev(struct net_device *dev) | |
11292 | { | |
11293 | rtnl_lock(); | |
11294 | unregister_netdevice(dev); | |
11295 | rtnl_unlock(); | |
11296 | } | |
1da177e4 LT |
11297 | EXPORT_SYMBOL(unregister_netdev); |
11298 | ||
ce286d32 | 11299 | /** |
0854fa82 | 11300 | * __dev_change_net_namespace - move device to different nethost namespace |
ce286d32 EB |
11301 | * @dev: device |
11302 | * @net: network namespace | |
11303 | * @pat: If not NULL name pattern to try if the current device name | |
11304 | * is already taken in the destination network namespace. | |
eeb85a14 AV |
11305 | * @new_ifindex: If not zero, specifies device index in the target |
11306 | * namespace. | |
ce286d32 EB |
11307 | * |
11308 | * This function shuts down a device interface and moves it | |
11309 | * to a new network namespace. On success 0 is returned, on | |
11310 | * a failure a netagive errno code is returned. | |
11311 | * | |
11312 | * Callers must hold the rtnl semaphore. | |
11313 | */ | |
11314 | ||
0854fa82 AV |
11315 | int __dev_change_net_namespace(struct net_device *dev, struct net *net, |
11316 | const char *pat, int new_ifindex) | |
ce286d32 | 11317 | { |
7663d522 | 11318 | struct netdev_name_node *name_node; |
ef6a4c88 | 11319 | struct net *net_old = dev_net(dev); |
311cca40 | 11320 | char new_name[IFNAMSIZ] = {}; |
eeb85a14 | 11321 | int err, new_nsid; |
ce286d32 EB |
11322 | |
11323 | ASSERT_RTNL(); | |
11324 | ||
11325 | /* Don't allow namespace local devices to be moved. */ | |
11326 | err = -EINVAL; | |
11327 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
11328 | goto out; | |
11329 | ||
11330 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
11331 | if (dev->reg_state != NETREG_REGISTERED) |
11332 | goto out; | |
11333 | ||
11334 | /* Get out if there is nothing todo */ | |
11335 | err = 0; | |
ef6a4c88 | 11336 | if (net_eq(net_old, net)) |
ce286d32 EB |
11337 | goto out; |
11338 | ||
11339 | /* Pick the destination device name, and ensure | |
11340 | * we can use it in the destination network namespace. | |
11341 | */ | |
11342 | err = -EEXIST; | |
75ea27d0 | 11343 | if (netdev_name_in_use(net, dev->name)) { |
ce286d32 EB |
11344 | /* We get here if we can't use the current device name */ |
11345 | if (!pat) | |
11346 | goto out; | |
556c755a | 11347 | err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST); |
7892bd08 | 11348 | if (err < 0) |
ce286d32 EB |
11349 | goto out; |
11350 | } | |
7663d522 JK |
11351 | /* Check that none of the altnames conflicts. */ |
11352 | err = -EEXIST; | |
11353 | netdev_for_each_altname(dev, name_node) | |
11354 | if (netdev_name_in_use(net, name_node->name)) | |
11355 | goto out; | |
ce286d32 | 11356 | |
eeb85a14 | 11357 | /* Check that new_ifindex isn't used yet. */ |
759ab1ed JK |
11358 | if (new_ifindex) { |
11359 | err = dev_index_reserve(net, new_ifindex); | |
11360 | if (err < 0) | |
11361 | goto out; | |
11362 | } else { | |
11363 | /* If there is an ifindex conflict assign a new one */ | |
11364 | err = dev_index_reserve(net, dev->ifindex); | |
11365 | if (err == -EBUSY) | |
11366 | err = dev_index_reserve(net, 0); | |
11367 | if (err < 0) | |
11368 | goto out; | |
11369 | new_ifindex = err; | |
11370 | } | |
eeb85a14 | 11371 | |
ce286d32 EB |
11372 | /* |
11373 | * And now a mini version of register_netdevice unregister_netdevice. | |
11374 | */ | |
11375 | ||
11376 | /* If device is running close it first. */ | |
9b772652 | 11377 | dev_close(dev); |
ce286d32 EB |
11378 | |
11379 | /* And unlink it from device chain */ | |
e51b9624 | 11380 | unlist_netdevice(dev); |
ce286d32 EB |
11381 | |
11382 | synchronize_net(); | |
11383 | ||
11384 | /* Shutdown queueing discipline. */ | |
11385 | dev_shutdown(dev); | |
11386 | ||
11387 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 11388 | * this device. They should clean all the things. |
11389 | * | |
11390 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
11391 | * This is wanted because this way 8021q and macvlan know | |
11392 | * the device is just moving and can keep their slaves up. | |
11393 | */ | |
ce286d32 | 11394 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 | 11395 | rcu_barrier(); |
38e01b30 | 11396 | |
d4e4fdf9 | 11397 | new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); |
38e01b30 ND |
11398 | |
11399 | rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, | |
11400 | new_ifindex); | |
ce286d32 EB |
11401 | |
11402 | /* | |
11403 | * Flush the unicast and multicast chains | |
11404 | */ | |
a748ee24 | 11405 | dev_uc_flush(dev); |
22bedad3 | 11406 | dev_mc_flush(dev); |
ce286d32 | 11407 | |
4e66ae2e SH |
11408 | /* Send a netdev-removed uevent to the old namespace */ |
11409 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 11410 | netdev_adjacent_del_links(dev); |
4e66ae2e | 11411 | |
93642e14 JP |
11412 | /* Move per-net netdevice notifiers that are following the netdevice */ |
11413 | move_netdevice_notifiers_dev_net(dev, net); | |
11414 | ||
ce286d32 | 11415 | /* Actually switch the network namespace */ |
c346dca1 | 11416 | dev_net_set(dev, net); |
38e01b30 | 11417 | dev->ifindex = new_ifindex; |
ce286d32 | 11418 | |
0840556e KI |
11419 | if (new_name[0]) { |
11420 | /* Rename the netdev to prepared name */ | |
11421 | write_seqlock(&netdev_rename_lock); | |
311cca40 | 11422 | strscpy(dev->name, new_name, IFNAMSIZ); |
0840556e KI |
11423 | write_sequnlock(&netdev_rename_lock); |
11424 | } | |
311cca40 | 11425 | |
8b41d188 | 11426 | /* Fixup kobjects */ |
dd891b5b | 11427 | dev_set_uevent_suppress(&dev->dev, 1); |
a1b3f594 | 11428 | err = device_rename(&dev->dev, dev->name); |
dd891b5b | 11429 | dev_set_uevent_suppress(&dev->dev, 0); |
8b41d188 | 11430 | WARN_ON(err); |
ce286d32 | 11431 | |
dd891b5b JK |
11432 | /* Send a netdev-add uevent to the new namespace */ |
11433 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
11434 | netdev_adjacent_add_links(dev); | |
11435 | ||
ef6a4c88 CB |
11436 | /* Adapt owner in case owning user namespace of target network |
11437 | * namespace is different from the original one. | |
11438 | */ | |
11439 | err = netdev_change_owner(dev, net_old, net); | |
11440 | WARN_ON(err); | |
11441 | ||
ce286d32 EB |
11442 | /* Add the device back in the hashes */ |
11443 | list_netdevice(dev); | |
11444 | ||
11445 | /* Notify protocols, that a new device appeared. */ | |
11446 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
11447 | ||
d90a909e EB |
11448 | /* |
11449 | * Prevent userspace races by waiting until the network | |
11450 | * device is fully setup before sending notifications. | |
11451 | */ | |
1d997f10 | 11452 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); |
d90a909e | 11453 | |
ce286d32 EB |
11454 | synchronize_net(); |
11455 | err = 0; | |
11456 | out: | |
11457 | return err; | |
11458 | } | |
0854fa82 | 11459 | EXPORT_SYMBOL_GPL(__dev_change_net_namespace); |
ce286d32 | 11460 | |
f0bf90de | 11461 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
11462 | { |
11463 | struct sk_buff **list_skb; | |
1da177e4 | 11464 | struct sk_buff *skb; |
f0bf90de | 11465 | unsigned int cpu; |
97d8b6e3 | 11466 | struct softnet_data *sd, *oldsd, *remsd = NULL; |
1da177e4 | 11467 | |
1da177e4 LT |
11468 | local_irq_disable(); |
11469 | cpu = smp_processor_id(); | |
11470 | sd = &per_cpu(softnet_data, cpu); | |
11471 | oldsd = &per_cpu(softnet_data, oldcpu); | |
11472 | ||
11473 | /* Find end of our completion_queue. */ | |
11474 | list_skb = &sd->completion_queue; | |
11475 | while (*list_skb) | |
11476 | list_skb = &(*list_skb)->next; | |
11477 | /* Append completion queue from offline CPU. */ | |
11478 | *list_skb = oldsd->completion_queue; | |
11479 | oldsd->completion_queue = NULL; | |
11480 | ||
1da177e4 | 11481 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
11482 | if (oldsd->output_queue) { |
11483 | *sd->output_queue_tailp = oldsd->output_queue; | |
11484 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
11485 | oldsd->output_queue = NULL; | |
11486 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
11487 | } | |
ac64da0b ED |
11488 | /* Append NAPI poll list from offline CPU, with one exception : |
11489 | * process_backlog() must be called by cpu owning percpu backlog. | |
11490 | * We properly handle process_queue & input_pkt_queue later. | |
11491 | */ | |
11492 | while (!list_empty(&oldsd->poll_list)) { | |
11493 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
11494 | struct napi_struct, | |
11495 | poll_list); | |
11496 | ||
11497 | list_del_init(&napi->poll_list); | |
11498 | if (napi->poll == process_backlog) | |
dad6b977 | 11499 | napi->state &= NAPIF_STATE_THREADED; |
ac64da0b ED |
11500 | else |
11501 | ____napi_schedule(sd, napi); | |
264524d5 | 11502 | } |
1da177e4 LT |
11503 | |
11504 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
11505 | local_irq_enable(); | |
11506 | ||
dad6b977 | 11507 | if (!use_backlog_threads()) { |
773fc8f6 | 11508 | #ifdef CONFIG_RPS |
dad6b977 SAS |
11509 | remsd = oldsd->rps_ipi_list; |
11510 | oldsd->rps_ipi_list = NULL; | |
773fc8f6 | 11511 | #endif |
dad6b977 SAS |
11512 | /* send out pending IPI's on offline CPU */ |
11513 | net_rps_send_ipi(remsd); | |
11514 | } | |
773fc8f6 | 11515 | |
1da177e4 | 11516 | /* Process offline CPU's input_pkt_queue */ |
76cc8b13 | 11517 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
ad0a043f | 11518 | netif_rx(skb); |
36b83ffc | 11519 | rps_input_queue_head_incr(oldsd); |
fec5e652 | 11520 | } |
ac64da0b | 11521 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
ad0a043f | 11522 | netif_rx(skb); |
36b83ffc | 11523 | rps_input_queue_head_incr(oldsd); |
76cc8b13 | 11524 | } |
1da177e4 | 11525 | |
f0bf90de | 11526 | return 0; |
1da177e4 | 11527 | } |
1da177e4 | 11528 | |
7f353bf2 | 11529 | /** |
b63365a2 HX |
11530 | * netdev_increment_features - increment feature set by one |
11531 | * @all: current feature set | |
11532 | * @one: new feature set | |
11533 | * @mask: mask feature set | |
7f353bf2 HX |
11534 | * |
11535 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
11536 | * @one to the master device with current feature set @all. Will not |
11537 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 11538 | */ |
c8f44aff MM |
11539 | netdev_features_t netdev_increment_features(netdev_features_t all, |
11540 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 11541 | { |
c8cd0989 | 11542 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 11543 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 11544 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 11545 | |
a188222b | 11546 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 11547 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 11548 | |
1742f183 | 11549 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
11550 | if (all & NETIF_F_HW_CSUM) |
11551 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
11552 | |
11553 | return all; | |
11554 | } | |
b63365a2 | 11555 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 11556 | |
430f03cd | 11557 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
11558 | { |
11559 | int i; | |
11560 | struct hlist_head *hash; | |
11561 | ||
6da2ec56 | 11562 | hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); |
30d97d35 PE |
11563 | if (hash != NULL) |
11564 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
11565 | INIT_HLIST_HEAD(&hash[i]); | |
11566 | ||
11567 | return hash; | |
11568 | } | |
11569 | ||
881d966b | 11570 | /* Initialize per network namespace state */ |
4665079c | 11571 | static int __net_init netdev_init(struct net *net) |
881d966b | 11572 | { |
d9f37d01 | 11573 | BUILD_BUG_ON(GRO_HASH_BUCKETS > |
c593642c | 11574 | 8 * sizeof_field(struct napi_struct, gro_bitmask)); |
d9f37d01 | 11575 | |
9c1be193 | 11576 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 11577 | |
30d97d35 PE |
11578 | net->dev_name_head = netdev_create_hash(); |
11579 | if (net->dev_name_head == NULL) | |
11580 | goto err_name; | |
881d966b | 11581 | |
30d97d35 PE |
11582 | net->dev_index_head = netdev_create_hash(); |
11583 | if (net->dev_index_head == NULL) | |
11584 | goto err_idx; | |
881d966b | 11585 | |
ceaac91d | 11586 | xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); |
759ab1ed | 11587 | |
a30c7b42 JP |
11588 | RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); |
11589 | ||
881d966b | 11590 | return 0; |
30d97d35 PE |
11591 | |
11592 | err_idx: | |
11593 | kfree(net->dev_name_head); | |
11594 | err_name: | |
11595 | return -ENOMEM; | |
881d966b EB |
11596 | } |
11597 | ||
f0db275a SH |
11598 | /** |
11599 | * netdev_drivername - network driver for the device | |
11600 | * @dev: network device | |
f0db275a SH |
11601 | * |
11602 | * Determine network driver for device. | |
11603 | */ | |
3019de12 | 11604 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 11605 | { |
cf04a4c7 SH |
11606 | const struct device_driver *driver; |
11607 | const struct device *parent; | |
3019de12 | 11608 | const char *empty = ""; |
6579e57b AV |
11609 | |
11610 | parent = dev->dev.parent; | |
6579e57b | 11611 | if (!parent) |
3019de12 | 11612 | return empty; |
6579e57b AV |
11613 | |
11614 | driver = parent->driver; | |
11615 | if (driver && driver->name) | |
3019de12 DM |
11616 | return driver->name; |
11617 | return empty; | |
6579e57b AV |
11618 | } |
11619 | ||
6ea754eb JP |
11620 | static void __netdev_printk(const char *level, const struct net_device *dev, |
11621 | struct va_format *vaf) | |
256df2f3 | 11622 | { |
b004ff49 | 11623 | if (dev && dev->dev.parent) { |
6ea754eb JP |
11624 | dev_printk_emit(level[1] - '0', |
11625 | dev->dev.parent, | |
11626 | "%s %s %s%s: %pV", | |
11627 | dev_driver_string(dev->dev.parent), | |
11628 | dev_name(dev->dev.parent), | |
11629 | netdev_name(dev), netdev_reg_state(dev), | |
11630 | vaf); | |
b004ff49 | 11631 | } else if (dev) { |
6ea754eb JP |
11632 | printk("%s%s%s: %pV", |
11633 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 11634 | } else { |
6ea754eb | 11635 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 11636 | } |
256df2f3 JP |
11637 | } |
11638 | ||
6ea754eb JP |
11639 | void netdev_printk(const char *level, const struct net_device *dev, |
11640 | const char *format, ...) | |
256df2f3 JP |
11641 | { |
11642 | struct va_format vaf; | |
11643 | va_list args; | |
256df2f3 JP |
11644 | |
11645 | va_start(args, format); | |
11646 | ||
11647 | vaf.fmt = format; | |
11648 | vaf.va = &args; | |
11649 | ||
6ea754eb | 11650 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 11651 | |
256df2f3 | 11652 | va_end(args); |
256df2f3 JP |
11653 | } |
11654 | EXPORT_SYMBOL(netdev_printk); | |
11655 | ||
11656 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 11657 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 11658 | { \ |
256df2f3 JP |
11659 | struct va_format vaf; \ |
11660 | va_list args; \ | |
11661 | \ | |
11662 | va_start(args, fmt); \ | |
11663 | \ | |
11664 | vaf.fmt = fmt; \ | |
11665 | vaf.va = &args; \ | |
11666 | \ | |
6ea754eb | 11667 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 11668 | \ |
256df2f3 | 11669 | va_end(args); \ |
256df2f3 JP |
11670 | } \ |
11671 | EXPORT_SYMBOL(func); | |
11672 | ||
11673 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
11674 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
11675 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
11676 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
11677 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
11678 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
11679 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
11680 | ||
4665079c | 11681 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
11682 | { |
11683 | kfree(net->dev_name_head); | |
11684 | kfree(net->dev_index_head); | |
759ab1ed | 11685 | xa_destroy(&net->dev_by_index); |
ee21b18b VA |
11686 | if (net != &init_net) |
11687 | WARN_ON_ONCE(!list_empty(&net->dev_base_head)); | |
881d966b EB |
11688 | } |
11689 | ||
022cbae6 | 11690 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
11691 | .init = netdev_init, |
11692 | .exit = netdev_exit, | |
11693 | }; | |
11694 | ||
ee403248 | 11695 | static void __net_exit default_device_exit_net(struct net *net) |
ce286d32 | 11696 | { |
d09486a0 | 11697 | struct netdev_name_node *name_node, *tmp; |
e008b5fc | 11698 | struct net_device *dev, *aux; |
ce286d32 | 11699 | /* |
e008b5fc | 11700 | * Push all migratable network devices back to the |
ce286d32 EB |
11701 | * initial network namespace |
11702 | */ | |
ee403248 | 11703 | ASSERT_RTNL(); |
e008b5fc | 11704 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 11705 | int err; |
aca51397 | 11706 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
11707 | |
11708 | /* Ignore unmoveable devices (i.e. loopback) */ | |
11709 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
11710 | continue; | |
11711 | ||
e008b5fc | 11712 | /* Leave virtual devices for the generic cleanup */ |
3a5ca857 | 11713 | if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) |
e008b5fc | 11714 | continue; |
d0c082ce | 11715 | |
25985edc | 11716 | /* Push remaining network devices to init_net */ |
aca51397 | 11717 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
75ea27d0 | 11718 | if (netdev_name_in_use(&init_net, fb_name)) |
55b40dbf | 11719 | snprintf(fb_name, IFNAMSIZ, "dev%%d"); |
d09486a0 JK |
11720 | |
11721 | netdev_for_each_altname_safe(dev, name_node, tmp) | |
723de3eb | 11722 | if (netdev_name_in_use(&init_net, name_node->name)) |
d09486a0 | 11723 | __netdev_name_node_alt_destroy(name_node); |
d09486a0 | 11724 | |
0854fa82 | 11725 | err = dev_change_net_namespace(dev, &init_net, fb_name); |
ce286d32 | 11726 | if (err) { |
7b6cd1ce JP |
11727 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
11728 | __func__, dev->name, err); | |
aca51397 | 11729 | BUG(); |
ce286d32 EB |
11730 | } |
11731 | } | |
ce286d32 EB |
11732 | } |
11733 | ||
04dc7f6b EB |
11734 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
11735 | { | |
11736 | /* At exit all network devices most be removed from a network | |
b595076a | 11737 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
11738 | * Do this across as many network namespaces as possible to |
11739 | * improve batching efficiency. | |
11740 | */ | |
11741 | struct net_device *dev; | |
11742 | struct net *net; | |
11743 | LIST_HEAD(dev_kill_list); | |
11744 | ||
ee403248 ED |
11745 | rtnl_lock(); |
11746 | list_for_each_entry(net, net_list, exit_list) { | |
11747 | default_device_exit_net(net); | |
11748 | cond_resched(); | |
11749 | } | |
ee403248 | 11750 | |
04dc7f6b EB |
11751 | list_for_each_entry(net, net_list, exit_list) { |
11752 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 11753 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
11754 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
11755 | else | |
11756 | unregister_netdevice_queue(dev, &dev_kill_list); | |
11757 | } | |
11758 | } | |
11759 | unregister_netdevice_many(&dev_kill_list); | |
11760 | rtnl_unlock(); | |
11761 | } | |
11762 | ||
022cbae6 | 11763 | static struct pernet_operations __net_initdata default_device_ops = { |
04dc7f6b | 11764 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
11765 | }; |
11766 | ||
43a71cd6 CL |
11767 | static void __init net_dev_struct_check(void) |
11768 | { | |
11769 | /* TX read-mostly hotpath */ | |
11770 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags); | |
11771 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops); | |
11772 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops); | |
11773 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx); | |
11774 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues); | |
11775 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size); | |
11776 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size); | |
11777 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs); | |
993498e5 | 11778 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features); |
43a71cd6 CL |
11779 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc); |
11780 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu); | |
11781 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom); | |
11782 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq); | |
11783 | #ifdef CONFIG_XPS | |
11784 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps); | |
11785 | #endif | |
11786 | #ifdef CONFIG_NETFILTER_EGRESS | |
11787 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress); | |
11788 | #endif | |
11789 | #ifdef CONFIG_NET_XGRESS | |
11790 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress); | |
11791 | #endif | |
993498e5 | 11792 | CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160); |
43a71cd6 CL |
11793 | |
11794 | /* TXRX read-mostly hotpath */ | |
c353c7b7 | 11795 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, lstats); |
f6e0a498 | 11796 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, state); |
43a71cd6 CL |
11797 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags); |
11798 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len); | |
11799 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features); | |
11800 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr); | |
f6e0a498 | 11801 | CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 46); |
43a71cd6 CL |
11802 | |
11803 | /* RX read-mostly hotpath */ | |
11804 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific); | |
11805 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex); | |
11806 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues); | |
11807 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx); | |
11808 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_flush_timeout); | |
11809 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, napi_defer_hard_irqs); | |
11810 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size); | |
11811 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size); | |
11812 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler); | |
11813 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data); | |
11814 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net); | |
11815 | #ifdef CONFIG_NETPOLL | |
11816 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo); | |
11817 | #endif | |
11818 | #ifdef CONFIG_NET_XGRESS | |
11819 | CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress); | |
11820 | #endif | |
d3d344a1 | 11821 | CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 104); |
43a71cd6 CL |
11822 | } |
11823 | ||
1da177e4 LT |
11824 | /* |
11825 | * Initialize the DEV module. At boot time this walks the device list and | |
11826 | * unhooks any devices that fail to initialise (normally hardware not | |
11827 | * present) and leaves us with a valid list of present and active devices. | |
11828 | * | |
11829 | */ | |
11830 | ||
2b0cfa6e LB |
11831 | /* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */ |
11832 | #define SYSTEM_PERCPU_PAGE_POOL_SIZE ((1 << 20) / PAGE_SIZE) | |
11833 | ||
11834 | static int net_page_pool_create(int cpuid) | |
11835 | { | |
11836 | #if IS_ENABLED(CONFIG_PAGE_POOL) | |
11837 | struct page_pool_params page_pool_params = { | |
11838 | .pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE, | |
f853fa5c | 11839 | .flags = PP_FLAG_SYSTEM_POOL, |
341ee1a5 | 11840 | .nid = cpu_to_mem(cpuid), |
2b0cfa6e LB |
11841 | }; |
11842 | struct page_pool *pp_ptr; | |
11843 | ||
11844 | pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid); | |
11845 | if (IS_ERR(pp_ptr)) | |
11846 | return -ENOMEM; | |
11847 | ||
11848 | per_cpu(system_page_pool, cpuid) = pp_ptr; | |
11849 | #endif | |
11850 | return 0; | |
11851 | } | |
11852 | ||
dad6b977 SAS |
11853 | static int backlog_napi_should_run(unsigned int cpu) |
11854 | { | |
11855 | struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); | |
11856 | struct napi_struct *napi = &sd->backlog; | |
11857 | ||
11858 | return test_bit(NAPI_STATE_SCHED_THREADED, &napi->state); | |
11859 | } | |
11860 | ||
11861 | static void run_backlog_napi(unsigned int cpu) | |
11862 | { | |
11863 | struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); | |
11864 | ||
11865 | napi_threaded_poll_loop(&sd->backlog); | |
11866 | } | |
11867 | ||
11868 | static void backlog_napi_setup(unsigned int cpu) | |
11869 | { | |
11870 | struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); | |
11871 | struct napi_struct *napi = &sd->backlog; | |
11872 | ||
11873 | napi->thread = this_cpu_read(backlog_napi); | |
11874 | set_bit(NAPI_STATE_THREADED, &napi->state); | |
11875 | } | |
11876 | ||
11877 | static struct smp_hotplug_thread backlog_threads = { | |
11878 | .store = &backlog_napi, | |
11879 | .thread_should_run = backlog_napi_should_run, | |
11880 | .thread_fn = run_backlog_napi, | |
11881 | .thread_comm = "backlog_napi/%u", | |
11882 | .setup = backlog_napi_setup, | |
11883 | }; | |
11884 | ||
1da177e4 LT |
11885 | /* |
11886 | * This is called single threaded during boot, so no need | |
11887 | * to take the rtnl semaphore. | |
11888 | */ | |
11889 | static int __init net_dev_init(void) | |
11890 | { | |
11891 | int i, rc = -ENOMEM; | |
11892 | ||
11893 | BUG_ON(!dev_boot_phase); | |
11894 | ||
43a71cd6 CL |
11895 | net_dev_struct_check(); |
11896 | ||
1da177e4 LT |
11897 | if (dev_proc_init()) |
11898 | goto out; | |
11899 | ||
8b41d188 | 11900 | if (netdev_kobject_init()) |
1da177e4 LT |
11901 | goto out; |
11902 | ||
82d8a867 | 11903 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
11904 | INIT_LIST_HEAD(&ptype_base[i]); |
11905 | ||
881d966b EB |
11906 | if (register_pernet_subsys(&netdev_net_ops)) |
11907 | goto out; | |
1da177e4 LT |
11908 | |
11909 | /* | |
11910 | * Initialise the packet receive queues. | |
11911 | */ | |
11912 | ||
6f912042 | 11913 | for_each_possible_cpu(i) { |
41852497 | 11914 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 11915 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 11916 | |
41852497 ED |
11917 | INIT_WORK(flush, flush_backlog); |
11918 | ||
e36fa2f7 | 11919 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 11920 | skb_queue_head_init(&sd->process_queue); |
f53c7239 SK |
11921 | #ifdef CONFIG_XFRM_OFFLOAD |
11922 | skb_queue_head_init(&sd->xfrm_backlog); | |
11923 | #endif | |
e36fa2f7 | 11924 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 11925 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 11926 | #ifdef CONFIG_RPS |
545b8c8d | 11927 | INIT_CSD(&sd->csd, rps_trigger_softirq, sd); |
e36fa2f7 | 11928 | sd->cpu = i; |
1e94d72f | 11929 | #endif |
97e719a8 | 11930 | INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); |
68822bdf | 11931 | spin_lock_init(&sd->defer_lock); |
0a9627f2 | 11932 | |
7c4ec749 | 11933 | init_gro_hash(&sd->backlog); |
e36fa2f7 ED |
11934 | sd->backlog.poll = process_backlog; |
11935 | sd->backlog.weight = weight_p; | |
dad6b977 | 11936 | INIT_LIST_HEAD(&sd->backlog.poll_list); |
2b0cfa6e LB |
11937 | |
11938 | if (net_page_pool_create(i)) | |
11939 | goto out; | |
1da177e4 | 11940 | } |
dad6b977 SAS |
11941 | if (use_backlog_threads()) |
11942 | smpboot_register_percpu_thread(&backlog_threads); | |
1da177e4 | 11943 | |
1da177e4 LT |
11944 | dev_boot_phase = 0; |
11945 | ||
505d4f73 EB |
11946 | /* The loopback device is special if any other network devices |
11947 | * is present in a network namespace the loopback device must | |
11948 | * be present. Since we now dynamically allocate and free the | |
11949 | * loopback device ensure this invariant is maintained by | |
11950 | * keeping the loopback device as the first device on the | |
11951 | * list of network devices. Ensuring the loopback devices | |
11952 | * is the first device that appears and the last network device | |
11953 | * that disappears. | |
11954 | */ | |
11955 | if (register_pernet_device(&loopback_net_ops)) | |
11956 | goto out; | |
11957 | ||
11958 | if (register_pernet_device(&default_device_ops)) | |
11959 | goto out; | |
11960 | ||
962cf36c CM |
11961 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
11962 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 11963 | |
f0bf90de SAS |
11964 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
11965 | NULL, dev_cpu_dead); | |
11966 | WARN_ON(rc < 0); | |
1da177e4 | 11967 | rc = 0; |
2f3c7195 MT |
11968 | |
11969 | /* avoid static key IPIs to isolated CPUs */ | |
11970 | if (housekeeping_enabled(HK_TYPE_MISC)) | |
11971 | net_enable_timestamp(); | |
1da177e4 | 11972 | out: |
2b0cfa6e LB |
11973 | if (rc < 0) { |
11974 | for_each_possible_cpu(i) { | |
11975 | struct page_pool *pp_ptr; | |
11976 | ||
11977 | pp_ptr = per_cpu(system_page_pool, i); | |
11978 | if (!pp_ptr) | |
11979 | continue; | |
11980 | ||
11981 | page_pool_destroy(pp_ptr); | |
11982 | per_cpu(system_page_pool, i) = NULL; | |
11983 | } | |
11984 | } | |
11985 | ||
1da177e4 LT |
11986 | return rc; |
11987 | } | |
11988 | ||
11989 | subsys_initcall(net_dev_init); |