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> |
1da177e4 | 72 | #include <linux/bitops.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> |
f1083048 | 80 | #include <linux/sched/mm.h> |
4a3e2f71 | 81 | #include <linux/mutex.h> |
1da177e4 LT |
82 | #include <linux/string.h> |
83 | #include <linux/mm.h> | |
84 | #include <linux/socket.h> | |
85 | #include <linux/sockios.h> | |
86 | #include <linux/errno.h> | |
87 | #include <linux/interrupt.h> | |
88 | #include <linux/if_ether.h> | |
89 | #include <linux/netdevice.h> | |
90 | #include <linux/etherdevice.h> | |
0187bdfb | 91 | #include <linux/ethtool.h> |
1da177e4 | 92 | #include <linux/skbuff.h> |
a7862b45 | 93 | #include <linux/bpf.h> |
b5cdae32 | 94 | #include <linux/bpf_trace.h> |
457c4cbc | 95 | #include <net/net_namespace.h> |
1da177e4 | 96 | #include <net/sock.h> |
02d62e86 | 97 | #include <net/busy_poll.h> |
1da177e4 | 98 | #include <linux/rtnetlink.h> |
1da177e4 | 99 | #include <linux/stat.h> |
1da177e4 | 100 | #include <net/dst.h> |
fc4099f1 | 101 | #include <net/dst_metadata.h> |
1da177e4 | 102 | #include <net/pkt_sched.h> |
87d83093 | 103 | #include <net/pkt_cls.h> |
1da177e4 | 104 | #include <net/checksum.h> |
44540960 | 105 | #include <net/xfrm.h> |
1da177e4 LT |
106 | #include <linux/highmem.h> |
107 | #include <linux/init.h> | |
1da177e4 | 108 | #include <linux/module.h> |
1da177e4 LT |
109 | #include <linux/netpoll.h> |
110 | #include <linux/rcupdate.h> | |
111 | #include <linux/delay.h> | |
1da177e4 | 112 | #include <net/iw_handler.h> |
1da177e4 | 113 | #include <asm/current.h> |
5bdb9886 | 114 | #include <linux/audit.h> |
db217334 | 115 | #include <linux/dmaengine.h> |
f6a78bfc | 116 | #include <linux/err.h> |
c7fa9d18 | 117 | #include <linux/ctype.h> |
723e98b7 | 118 | #include <linux/if_arp.h> |
6de329e2 | 119 | #include <linux/if_vlan.h> |
8f0f2223 | 120 | #include <linux/ip.h> |
ad55dcaf | 121 | #include <net/ip.h> |
25cd9ba0 | 122 | #include <net/mpls.h> |
8f0f2223 DM |
123 | #include <linux/ipv6.h> |
124 | #include <linux/in.h> | |
b6b2fed1 DM |
125 | #include <linux/jhash.h> |
126 | #include <linux/random.h> | |
9cbc1cb8 | 127 | #include <trace/events/napi.h> |
cf66ba58 | 128 | #include <trace/events/net.h> |
07dc22e7 | 129 | #include <trace/events/skb.h> |
caeda9b9 | 130 | #include <linux/inetdevice.h> |
c445477d | 131 | #include <linux/cpu_rmap.h> |
c5905afb | 132 | #include <linux/static_key.h> |
af12fa6e | 133 | #include <linux/hashtable.h> |
60877a32 | 134 | #include <linux/vmalloc.h> |
529d0489 | 135 | #include <linux/if_macvlan.h> |
e7fd2885 | 136 | #include <linux/errqueue.h> |
3b47d303 | 137 | #include <linux/hrtimer.h> |
357b6cc5 | 138 | #include <linux/netfilter_ingress.h> |
40e4e713 | 139 | #include <linux/crash_dump.h> |
b72b5bf6 | 140 | #include <linux/sctp.h> |
ae847f40 | 141 | #include <net/udp_tunnel.h> |
6621dd29 | 142 | #include <linux/net_namespace.h> |
aaa5d90b | 143 | #include <linux/indirect_call_wrapper.h> |
af3836df | 144 | #include <net/devlink.h> |
1da177e4 | 145 | |
342709ef PE |
146 | #include "net-sysfs.h" |
147 | ||
d565b0a1 HX |
148 | #define MAX_GRO_SKBS 8 |
149 | ||
5d38a079 HX |
150 | /* This should be increased if a protocol with a bigger head is added. */ |
151 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
152 | ||
1da177e4 | 153 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 154 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
155 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
156 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 157 | static struct list_head offload_base __read_mostly; |
1da177e4 | 158 | |
ae78dbfa | 159 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 | 160 | static int call_netdevice_notifiers_info(unsigned long val, |
54951194 | 161 | struct netdev_notifier_info *info); |
26372605 PM |
162 | static int call_netdevice_notifiers_extack(unsigned long val, |
163 | struct net_device *dev, | |
164 | struct netlink_ext_ack *extack); | |
90b602f8 | 165 | static struct napi_struct *napi_by_id(unsigned int napi_id); |
ae78dbfa | 166 | |
1da177e4 | 167 | /* |
7562f876 | 168 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
169 | * semaphore. |
170 | * | |
c6d14c84 | 171 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
172 | * |
173 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 174 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
175 | * actual updates. This allows pure readers to access the list even |
176 | * while a writer is preparing to update it. | |
177 | * | |
178 | * To put it another way, dev_base_lock is held for writing only to | |
179 | * protect against pure readers; the rtnl semaphore provides the | |
180 | * protection against other writers. | |
181 | * | |
182 | * See, for example usages, register_netdevice() and | |
183 | * unregister_netdevice(), which must be called with the rtnl | |
184 | * semaphore held. | |
185 | */ | |
1da177e4 | 186 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
187 | EXPORT_SYMBOL(dev_base_lock); |
188 | ||
6c557001 FW |
189 | static DEFINE_MUTEX(ifalias_mutex); |
190 | ||
af12fa6e ET |
191 | /* protects napi_hash addition/deletion and napi_gen_id */ |
192 | static DEFINE_SPINLOCK(napi_hash_lock); | |
193 | ||
52bd2d62 | 194 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 195 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 196 | |
18afa4b0 | 197 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 198 | |
4e985ada TG |
199 | static inline void dev_base_seq_inc(struct net *net) |
200 | { | |
643aa9cb | 201 | while (++net->dev_base_seq == 0) |
202 | ; | |
4e985ada TG |
203 | } |
204 | ||
881d966b | 205 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 206 | { |
8387ff25 | 207 | unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); |
95c96174 | 208 | |
08e9897d | 209 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
210 | } |
211 | ||
881d966b | 212 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 213 | { |
7c28bd0b | 214 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
215 | } |
216 | ||
e36fa2f7 | 217 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
218 | { |
219 | #ifdef CONFIG_RPS | |
e36fa2f7 | 220 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
221 | #endif |
222 | } | |
223 | ||
e36fa2f7 | 224 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
225 | { |
226 | #ifdef CONFIG_RPS | |
e36fa2f7 | 227 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
228 | #endif |
229 | } | |
230 | ||
ff927412 JP |
231 | static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, |
232 | const char *name) | |
233 | { | |
234 | struct netdev_name_node *name_node; | |
235 | ||
236 | name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); | |
237 | if (!name_node) | |
238 | return NULL; | |
239 | INIT_HLIST_NODE(&name_node->hlist); | |
240 | name_node->dev = dev; | |
241 | name_node->name = name; | |
242 | return name_node; | |
243 | } | |
244 | ||
245 | static struct netdev_name_node * | |
246 | netdev_name_node_head_alloc(struct net_device *dev) | |
247 | { | |
36fbf1e5 JP |
248 | struct netdev_name_node *name_node; |
249 | ||
250 | name_node = netdev_name_node_alloc(dev, dev->name); | |
251 | if (!name_node) | |
252 | return NULL; | |
253 | INIT_LIST_HEAD(&name_node->list); | |
254 | return name_node; | |
ff927412 JP |
255 | } |
256 | ||
257 | static void netdev_name_node_free(struct netdev_name_node *name_node) | |
258 | { | |
259 | kfree(name_node); | |
260 | } | |
261 | ||
262 | static void netdev_name_node_add(struct net *net, | |
263 | struct netdev_name_node *name_node) | |
264 | { | |
265 | hlist_add_head_rcu(&name_node->hlist, | |
266 | dev_name_hash(net, name_node->name)); | |
267 | } | |
268 | ||
269 | static void netdev_name_node_del(struct netdev_name_node *name_node) | |
270 | { | |
271 | hlist_del_rcu(&name_node->hlist); | |
272 | } | |
273 | ||
274 | static struct netdev_name_node *netdev_name_node_lookup(struct net *net, | |
275 | const char *name) | |
276 | { | |
277 | struct hlist_head *head = dev_name_hash(net, name); | |
278 | struct netdev_name_node *name_node; | |
279 | ||
280 | hlist_for_each_entry(name_node, head, hlist) | |
281 | if (!strcmp(name_node->name, name)) | |
282 | return name_node; | |
283 | return NULL; | |
284 | } | |
285 | ||
286 | static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, | |
287 | const char *name) | |
288 | { | |
289 | struct hlist_head *head = dev_name_hash(net, name); | |
290 | struct netdev_name_node *name_node; | |
291 | ||
292 | hlist_for_each_entry_rcu(name_node, head, hlist) | |
293 | if (!strcmp(name_node->name, name)) | |
294 | return name_node; | |
295 | return NULL; | |
296 | } | |
297 | ||
36fbf1e5 JP |
298 | int netdev_name_node_alt_create(struct net_device *dev, const char *name) |
299 | { | |
300 | struct netdev_name_node *name_node; | |
301 | struct net *net = dev_net(dev); | |
302 | ||
303 | name_node = netdev_name_node_lookup(net, name); | |
304 | if (name_node) | |
305 | return -EEXIST; | |
306 | name_node = netdev_name_node_alloc(dev, name); | |
307 | if (!name_node) | |
308 | return -ENOMEM; | |
309 | netdev_name_node_add(net, name_node); | |
310 | /* The node that holds dev->name acts as a head of per-device list. */ | |
311 | list_add_tail(&name_node->list, &dev->name_node->list); | |
312 | ||
313 | return 0; | |
314 | } | |
315 | EXPORT_SYMBOL(netdev_name_node_alt_create); | |
316 | ||
317 | static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) | |
318 | { | |
319 | list_del(&name_node->list); | |
320 | netdev_name_node_del(name_node); | |
321 | kfree(name_node->name); | |
322 | netdev_name_node_free(name_node); | |
323 | } | |
324 | ||
325 | int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) | |
326 | { | |
327 | struct netdev_name_node *name_node; | |
328 | struct net *net = dev_net(dev); | |
329 | ||
330 | name_node = netdev_name_node_lookup(net, name); | |
331 | if (!name_node) | |
332 | return -ENOENT; | |
e08ad805 ED |
333 | /* lookup might have found our primary name or a name belonging |
334 | * to another device. | |
335 | */ | |
336 | if (name_node == dev->name_node || name_node->dev != dev) | |
337 | return -EINVAL; | |
338 | ||
36fbf1e5 JP |
339 | __netdev_name_node_alt_destroy(name_node); |
340 | ||
341 | return 0; | |
342 | } | |
343 | EXPORT_SYMBOL(netdev_name_node_alt_destroy); | |
344 | ||
345 | static void netdev_name_node_alt_flush(struct net_device *dev) | |
346 | { | |
347 | struct netdev_name_node *name_node, *tmp; | |
348 | ||
349 | list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) | |
350 | __netdev_name_node_alt_destroy(name_node); | |
351 | } | |
352 | ||
ce286d32 | 353 | /* Device list insertion */ |
53759be9 | 354 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 355 | { |
c346dca1 | 356 | struct net *net = dev_net(dev); |
ce286d32 EB |
357 | |
358 | ASSERT_RTNL(); | |
359 | ||
360 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 361 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
ff927412 | 362 | netdev_name_node_add(net, dev->name_node); |
fb699dfd ED |
363 | hlist_add_head_rcu(&dev->index_hlist, |
364 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 365 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
366 | |
367 | dev_base_seq_inc(net); | |
ce286d32 EB |
368 | } |
369 | ||
fb699dfd ED |
370 | /* Device list removal |
371 | * caller must respect a RCU grace period before freeing/reusing dev | |
372 | */ | |
ce286d32 EB |
373 | static void unlist_netdevice(struct net_device *dev) |
374 | { | |
375 | ASSERT_RTNL(); | |
376 | ||
377 | /* Unlink dev from the device chain */ | |
378 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 379 | list_del_rcu(&dev->dev_list); |
ff927412 | 380 | netdev_name_node_del(dev->name_node); |
fb699dfd | 381 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 382 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
383 | |
384 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
385 | } |
386 | ||
1da177e4 LT |
387 | /* |
388 | * Our notifier list | |
389 | */ | |
390 | ||
f07d5b94 | 391 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
392 | |
393 | /* | |
394 | * Device drivers call our routines to queue packets here. We empty the | |
395 | * queue in the local softnet handler. | |
396 | */ | |
bea3348e | 397 | |
9958da05 | 398 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 399 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 400 | |
1da177e4 | 401 | /******************************************************************************* |
eb13da1a | 402 | * |
403 | * Protocol management and registration routines | |
404 | * | |
405 | *******************************************************************************/ | |
1da177e4 | 406 | |
1da177e4 | 407 | |
1da177e4 LT |
408 | /* |
409 | * Add a protocol ID to the list. Now that the input handler is | |
410 | * smarter we can dispense with all the messy stuff that used to be | |
411 | * here. | |
412 | * | |
413 | * BEWARE!!! Protocol handlers, mangling input packets, | |
414 | * MUST BE last in hash buckets and checking protocol handlers | |
415 | * MUST start from promiscuous ptype_all chain in net_bh. | |
416 | * It is true now, do not change it. | |
417 | * Explanation follows: if protocol handler, mangling packet, will | |
418 | * be the first on list, it is not able to sense, that packet | |
419 | * is cloned and should be copied-on-write, so that it will | |
420 | * change it and subsequent readers will get broken packet. | |
421 | * --ANK (980803) | |
422 | */ | |
423 | ||
c07b68e8 ED |
424 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
425 | { | |
426 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 427 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 428 | else |
7866a621 SN |
429 | return pt->dev ? &pt->dev->ptype_specific : |
430 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
431 | } |
432 | ||
1da177e4 LT |
433 | /** |
434 | * dev_add_pack - add packet handler | |
435 | * @pt: packet type declaration | |
436 | * | |
437 | * Add a protocol handler to the networking stack. The passed &packet_type | |
438 | * is linked into kernel lists and may not be freed until it has been | |
439 | * removed from the kernel lists. | |
440 | * | |
4ec93edb | 441 | * This call does not sleep therefore it can not |
1da177e4 LT |
442 | * guarantee all CPU's that are in middle of receiving packets |
443 | * will see the new packet type (until the next received packet). | |
444 | */ | |
445 | ||
446 | void dev_add_pack(struct packet_type *pt) | |
447 | { | |
c07b68e8 | 448 | struct list_head *head = ptype_head(pt); |
1da177e4 | 449 | |
c07b68e8 ED |
450 | spin_lock(&ptype_lock); |
451 | list_add_rcu(&pt->list, head); | |
452 | spin_unlock(&ptype_lock); | |
1da177e4 | 453 | } |
d1b19dff | 454 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 455 | |
1da177e4 LT |
456 | /** |
457 | * __dev_remove_pack - remove packet handler | |
458 | * @pt: packet type declaration | |
459 | * | |
460 | * Remove a protocol handler that was previously added to the kernel | |
461 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
462 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 463 | * returns. |
1da177e4 LT |
464 | * |
465 | * The packet type might still be in use by receivers | |
466 | * and must not be freed until after all the CPU's have gone | |
467 | * through a quiescent state. | |
468 | */ | |
469 | void __dev_remove_pack(struct packet_type *pt) | |
470 | { | |
c07b68e8 | 471 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
472 | struct packet_type *pt1; |
473 | ||
c07b68e8 | 474 | spin_lock(&ptype_lock); |
1da177e4 LT |
475 | |
476 | list_for_each_entry(pt1, head, list) { | |
477 | if (pt == pt1) { | |
478 | list_del_rcu(&pt->list); | |
479 | goto out; | |
480 | } | |
481 | } | |
482 | ||
7b6cd1ce | 483 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 484 | out: |
c07b68e8 | 485 | spin_unlock(&ptype_lock); |
1da177e4 | 486 | } |
d1b19dff ED |
487 | EXPORT_SYMBOL(__dev_remove_pack); |
488 | ||
1da177e4 LT |
489 | /** |
490 | * dev_remove_pack - remove packet handler | |
491 | * @pt: packet type declaration | |
492 | * | |
493 | * Remove a protocol handler that was previously added to the kernel | |
494 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
495 | * from the kernel lists and can be freed or reused once this function | |
496 | * returns. | |
497 | * | |
498 | * This call sleeps to guarantee that no CPU is looking at the packet | |
499 | * type after return. | |
500 | */ | |
501 | void dev_remove_pack(struct packet_type *pt) | |
502 | { | |
503 | __dev_remove_pack(pt); | |
4ec93edb | 504 | |
1da177e4 LT |
505 | synchronize_net(); |
506 | } | |
d1b19dff | 507 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 508 | |
62532da9 VY |
509 | |
510 | /** | |
511 | * dev_add_offload - register offload handlers | |
512 | * @po: protocol offload declaration | |
513 | * | |
514 | * Add protocol offload handlers to the networking stack. The passed | |
515 | * &proto_offload is linked into kernel lists and may not be freed until | |
516 | * it has been removed from the kernel lists. | |
517 | * | |
518 | * This call does not sleep therefore it can not | |
519 | * guarantee all CPU's that are in middle of receiving packets | |
520 | * will see the new offload handlers (until the next received packet). | |
521 | */ | |
522 | void dev_add_offload(struct packet_offload *po) | |
523 | { | |
bdef7de4 | 524 | struct packet_offload *elem; |
62532da9 VY |
525 | |
526 | spin_lock(&offload_lock); | |
bdef7de4 DM |
527 | list_for_each_entry(elem, &offload_base, list) { |
528 | if (po->priority < elem->priority) | |
529 | break; | |
530 | } | |
531 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
532 | spin_unlock(&offload_lock); |
533 | } | |
534 | EXPORT_SYMBOL(dev_add_offload); | |
535 | ||
536 | /** | |
537 | * __dev_remove_offload - remove offload handler | |
538 | * @po: packet offload declaration | |
539 | * | |
540 | * Remove a protocol offload handler that was previously added to the | |
541 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
542 | * is removed from the kernel lists and can be freed or reused once this | |
543 | * function returns. | |
544 | * | |
545 | * The packet type might still be in use by receivers | |
546 | * and must not be freed until after all the CPU's have gone | |
547 | * through a quiescent state. | |
548 | */ | |
1d143d9f | 549 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
550 | { |
551 | struct list_head *head = &offload_base; | |
552 | struct packet_offload *po1; | |
553 | ||
c53aa505 | 554 | spin_lock(&offload_lock); |
62532da9 VY |
555 | |
556 | list_for_each_entry(po1, head, list) { | |
557 | if (po == po1) { | |
558 | list_del_rcu(&po->list); | |
559 | goto out; | |
560 | } | |
561 | } | |
562 | ||
563 | pr_warn("dev_remove_offload: %p not found\n", po); | |
564 | out: | |
c53aa505 | 565 | spin_unlock(&offload_lock); |
62532da9 | 566 | } |
62532da9 VY |
567 | |
568 | /** | |
569 | * dev_remove_offload - remove packet offload handler | |
570 | * @po: packet offload declaration | |
571 | * | |
572 | * Remove a packet offload handler that was previously added to the kernel | |
573 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
574 | * removed from the kernel lists and can be freed or reused once this | |
575 | * function returns. | |
576 | * | |
577 | * This call sleeps to guarantee that no CPU is looking at the packet | |
578 | * type after return. | |
579 | */ | |
580 | void dev_remove_offload(struct packet_offload *po) | |
581 | { | |
582 | __dev_remove_offload(po); | |
583 | ||
584 | synchronize_net(); | |
585 | } | |
586 | EXPORT_SYMBOL(dev_remove_offload); | |
587 | ||
1da177e4 | 588 | /****************************************************************************** |
eb13da1a | 589 | * |
590 | * Device Boot-time Settings Routines | |
591 | * | |
592 | ******************************************************************************/ | |
1da177e4 LT |
593 | |
594 | /* Boot time configuration table */ | |
595 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
596 | ||
597 | /** | |
598 | * netdev_boot_setup_add - add new setup entry | |
599 | * @name: name of the device | |
600 | * @map: configured settings for the device | |
601 | * | |
602 | * Adds new setup entry to the dev_boot_setup list. The function | |
603 | * returns 0 on error and 1 on success. This is a generic routine to | |
604 | * all netdevices. | |
605 | */ | |
606 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
607 | { | |
608 | struct netdev_boot_setup *s; | |
609 | int i; | |
610 | ||
611 | s = dev_boot_setup; | |
612 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
613 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
614 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 615 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
616 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
617 | break; | |
618 | } | |
619 | } | |
620 | ||
621 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
622 | } | |
623 | ||
624 | /** | |
722c9a0c | 625 | * netdev_boot_setup_check - check boot time settings |
626 | * @dev: the netdevice | |
1da177e4 | 627 | * |
722c9a0c | 628 | * Check boot time settings for the device. |
629 | * The found settings are set for the device to be used | |
630 | * later in the device probing. | |
631 | * Returns 0 if no settings found, 1 if they are. | |
1da177e4 LT |
632 | */ |
633 | int netdev_boot_setup_check(struct net_device *dev) | |
634 | { | |
635 | struct netdev_boot_setup *s = dev_boot_setup; | |
636 | int i; | |
637 | ||
638 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
639 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 640 | !strcmp(dev->name, s[i].name)) { |
722c9a0c | 641 | dev->irq = s[i].map.irq; |
642 | dev->base_addr = s[i].map.base_addr; | |
643 | dev->mem_start = s[i].map.mem_start; | |
644 | dev->mem_end = s[i].map.mem_end; | |
1da177e4 LT |
645 | return 1; |
646 | } | |
647 | } | |
648 | return 0; | |
649 | } | |
d1b19dff | 650 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
651 | |
652 | ||
653 | /** | |
722c9a0c | 654 | * netdev_boot_base - get address from boot time settings |
655 | * @prefix: prefix for network device | |
656 | * @unit: id for network device | |
657 | * | |
658 | * Check boot time settings for the base address of device. | |
659 | * The found settings are set for the device to be used | |
660 | * later in the device probing. | |
661 | * Returns 0 if no settings found. | |
1da177e4 LT |
662 | */ |
663 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
664 | { | |
665 | const struct netdev_boot_setup *s = dev_boot_setup; | |
666 | char name[IFNAMSIZ]; | |
667 | int i; | |
668 | ||
669 | sprintf(name, "%s%d", prefix, unit); | |
670 | ||
671 | /* | |
672 | * If device already registered then return base of 1 | |
673 | * to indicate not to probe for this interface | |
674 | */ | |
881d966b | 675 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
676 | return 1; |
677 | ||
678 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
679 | if (!strcmp(name, s[i].name)) | |
680 | return s[i].map.base_addr; | |
681 | return 0; | |
682 | } | |
683 | ||
684 | /* | |
685 | * Saves at boot time configured settings for any netdevice. | |
686 | */ | |
687 | int __init netdev_boot_setup(char *str) | |
688 | { | |
689 | int ints[5]; | |
690 | struct ifmap map; | |
691 | ||
692 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
693 | if (!str || !*str) | |
694 | return 0; | |
695 | ||
696 | /* Save settings */ | |
697 | memset(&map, 0, sizeof(map)); | |
698 | if (ints[0] > 0) | |
699 | map.irq = ints[1]; | |
700 | if (ints[0] > 1) | |
701 | map.base_addr = ints[2]; | |
702 | if (ints[0] > 2) | |
703 | map.mem_start = ints[3]; | |
704 | if (ints[0] > 3) | |
705 | map.mem_end = ints[4]; | |
706 | ||
707 | /* Add new entry to the list */ | |
708 | return netdev_boot_setup_add(str, &map); | |
709 | } | |
710 | ||
711 | __setup("netdev=", netdev_boot_setup); | |
712 | ||
713 | /******************************************************************************* | |
eb13da1a | 714 | * |
715 | * Device Interface Subroutines | |
716 | * | |
717 | *******************************************************************************/ | |
1da177e4 | 718 | |
a54acb3a ND |
719 | /** |
720 | * dev_get_iflink - get 'iflink' value of a interface | |
721 | * @dev: targeted interface | |
722 | * | |
723 | * Indicates the ifindex the interface is linked to. | |
724 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
725 | */ | |
726 | ||
727 | int dev_get_iflink(const struct net_device *dev) | |
728 | { | |
729 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
730 | return dev->netdev_ops->ndo_get_iflink(dev); | |
731 | ||
7a66bbc9 | 732 | return dev->ifindex; |
a54acb3a ND |
733 | } |
734 | EXPORT_SYMBOL(dev_get_iflink); | |
735 | ||
fc4099f1 PS |
736 | /** |
737 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
738 | * @dev: targeted interface | |
739 | * @skb: The packet. | |
740 | * | |
741 | * For better visibility of tunnel traffic OVS needs to retrieve | |
742 | * egress tunnel information for a packet. Following API allows | |
743 | * user to get this info. | |
744 | */ | |
745 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
746 | { | |
747 | struct ip_tunnel_info *info; | |
748 | ||
749 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
750 | return -EINVAL; | |
751 | ||
752 | info = skb_tunnel_info_unclone(skb); | |
753 | if (!info) | |
754 | return -ENOMEM; | |
755 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
756 | return -EINVAL; | |
757 | ||
758 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
759 | } | |
760 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
761 | ||
1da177e4 LT |
762 | /** |
763 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 764 | * @net: the applicable net namespace |
1da177e4 LT |
765 | * @name: name to find |
766 | * | |
767 | * Find an interface by name. Must be called under RTNL semaphore | |
768 | * or @dev_base_lock. If the name is found a pointer to the device | |
769 | * is returned. If the name is not found then %NULL is returned. The | |
770 | * reference counters are not incremented so the caller must be | |
771 | * careful with locks. | |
772 | */ | |
773 | ||
881d966b | 774 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 775 | { |
ff927412 | 776 | struct netdev_name_node *node_name; |
1da177e4 | 777 | |
ff927412 JP |
778 | node_name = netdev_name_node_lookup(net, name); |
779 | return node_name ? node_name->dev : NULL; | |
1da177e4 | 780 | } |
d1b19dff | 781 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 782 | |
72c9528b | 783 | /** |
722c9a0c | 784 | * dev_get_by_name_rcu - find a device by its name |
785 | * @net: the applicable net namespace | |
786 | * @name: name to find | |
787 | * | |
788 | * Find an interface by name. | |
789 | * If the name is found a pointer to the device is returned. | |
790 | * If the name is not found then %NULL is returned. | |
791 | * The reference counters are not incremented so the caller must be | |
792 | * careful with locks. The caller must hold RCU lock. | |
72c9528b ED |
793 | */ |
794 | ||
795 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
796 | { | |
ff927412 | 797 | struct netdev_name_node *node_name; |
72c9528b | 798 | |
ff927412 JP |
799 | node_name = netdev_name_node_lookup_rcu(net, name); |
800 | return node_name ? node_name->dev : NULL; | |
72c9528b ED |
801 | } |
802 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
803 | ||
1da177e4 LT |
804 | /** |
805 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 806 | * @net: the applicable net namespace |
1da177e4 LT |
807 | * @name: name to find |
808 | * | |
809 | * Find an interface by name. This can be called from any | |
810 | * context and does its own locking. The returned handle has | |
811 | * the usage count incremented and the caller must use dev_put() to | |
812 | * release it when it is no longer needed. %NULL is returned if no | |
813 | * matching device is found. | |
814 | */ | |
815 | ||
881d966b | 816 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
817 | { |
818 | struct net_device *dev; | |
819 | ||
72c9528b ED |
820 | rcu_read_lock(); |
821 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
822 | if (dev) |
823 | dev_hold(dev); | |
72c9528b | 824 | rcu_read_unlock(); |
1da177e4 LT |
825 | return dev; |
826 | } | |
d1b19dff | 827 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
828 | |
829 | /** | |
830 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 831 | * @net: the applicable net namespace |
1da177e4 LT |
832 | * @ifindex: index of device |
833 | * | |
834 | * Search for an interface by index. Returns %NULL if the device | |
835 | * is not found or a pointer to the device. The device has not | |
836 | * had its reference counter increased so the caller must be careful | |
837 | * about locking. The caller must hold either the RTNL semaphore | |
838 | * or @dev_base_lock. | |
839 | */ | |
840 | ||
881d966b | 841 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 842 | { |
0bd8d536 ED |
843 | struct net_device *dev; |
844 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 845 | |
b67bfe0d | 846 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
847 | if (dev->ifindex == ifindex) |
848 | return dev; | |
0bd8d536 | 849 | |
1da177e4 LT |
850 | return NULL; |
851 | } | |
d1b19dff | 852 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 853 | |
fb699dfd ED |
854 | /** |
855 | * dev_get_by_index_rcu - find a device by its ifindex | |
856 | * @net: the applicable net namespace | |
857 | * @ifindex: index of device | |
858 | * | |
859 | * Search for an interface by index. Returns %NULL if the device | |
860 | * is not found or a pointer to the device. The device has not | |
861 | * had its reference counter increased so the caller must be careful | |
862 | * about locking. The caller must hold RCU lock. | |
863 | */ | |
864 | ||
865 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
866 | { | |
fb699dfd ED |
867 | struct net_device *dev; |
868 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
869 | ||
b67bfe0d | 870 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
871 | if (dev->ifindex == ifindex) |
872 | return dev; | |
873 | ||
874 | return NULL; | |
875 | } | |
876 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
877 | ||
1da177e4 LT |
878 | |
879 | /** | |
880 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 881 | * @net: the applicable net namespace |
1da177e4 LT |
882 | * @ifindex: index of device |
883 | * | |
884 | * Search for an interface by index. Returns NULL if the device | |
885 | * is not found or a pointer to the device. The device returned has | |
886 | * had a reference added and the pointer is safe until the user calls | |
887 | * dev_put to indicate they have finished with it. | |
888 | */ | |
889 | ||
881d966b | 890 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
891 | { |
892 | struct net_device *dev; | |
893 | ||
fb699dfd ED |
894 | rcu_read_lock(); |
895 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
896 | if (dev) |
897 | dev_hold(dev); | |
fb699dfd | 898 | rcu_read_unlock(); |
1da177e4 LT |
899 | return dev; |
900 | } | |
d1b19dff | 901 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 902 | |
90b602f8 ML |
903 | /** |
904 | * dev_get_by_napi_id - find a device by napi_id | |
905 | * @napi_id: ID of the NAPI struct | |
906 | * | |
907 | * Search for an interface by NAPI ID. Returns %NULL if the device | |
908 | * is not found or a pointer to the device. The device has not had | |
909 | * its reference counter increased so the caller must be careful | |
910 | * about locking. The caller must hold RCU lock. | |
911 | */ | |
912 | ||
913 | struct net_device *dev_get_by_napi_id(unsigned int napi_id) | |
914 | { | |
915 | struct napi_struct *napi; | |
916 | ||
917 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
918 | ||
919 | if (napi_id < MIN_NAPI_ID) | |
920 | return NULL; | |
921 | ||
922 | napi = napi_by_id(napi_id); | |
923 | ||
924 | return napi ? napi->dev : NULL; | |
925 | } | |
926 | EXPORT_SYMBOL(dev_get_by_napi_id); | |
927 | ||
5dbe7c17 NS |
928 | /** |
929 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
930 | * @net: network namespace | |
931 | * @name: a pointer to the buffer where the name will be stored. | |
932 | * @ifindex: the ifindex of the interface to get the name from. | |
933 | * | |
934 | * The use of raw_seqcount_begin() and cond_resched() before | |
935 | * retrying is required as we want to give the writers a chance | |
2da2b32f | 936 | * to complete when CONFIG_PREEMPTION is not set. |
5dbe7c17 NS |
937 | */ |
938 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
939 | { | |
940 | struct net_device *dev; | |
941 | unsigned int seq; | |
942 | ||
943 | retry: | |
944 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
945 | rcu_read_lock(); | |
946 | dev = dev_get_by_index_rcu(net, ifindex); | |
947 | if (!dev) { | |
948 | rcu_read_unlock(); | |
949 | return -ENODEV; | |
950 | } | |
951 | ||
952 | strcpy(name, dev->name); | |
953 | rcu_read_unlock(); | |
954 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
955 | cond_resched(); | |
956 | goto retry; | |
957 | } | |
958 | ||
959 | return 0; | |
960 | } | |
961 | ||
1da177e4 | 962 | /** |
941666c2 | 963 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 964 | * @net: the applicable net namespace |
1da177e4 LT |
965 | * @type: media type of device |
966 | * @ha: hardware address | |
967 | * | |
968 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
969 | * is not found or a pointer to the device. |
970 | * The caller must hold RCU or RTNL. | |
941666c2 | 971 | * The returned device has not had its ref count increased |
1da177e4 LT |
972 | * and the caller must therefore be careful about locking |
973 | * | |
1da177e4 LT |
974 | */ |
975 | ||
941666c2 ED |
976 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
977 | const char *ha) | |
1da177e4 LT |
978 | { |
979 | struct net_device *dev; | |
980 | ||
941666c2 | 981 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
982 | if (dev->type == type && |
983 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
984 | return dev; |
985 | ||
986 | return NULL; | |
1da177e4 | 987 | } |
941666c2 | 988 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 989 | |
881d966b | 990 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
991 | { |
992 | struct net_device *dev; | |
993 | ||
4e9cac2b | 994 | ASSERT_RTNL(); |
881d966b | 995 | for_each_netdev(net, dev) |
4e9cac2b | 996 | if (dev->type == type) |
7562f876 PE |
997 | return dev; |
998 | ||
999 | return NULL; | |
4e9cac2b | 1000 | } |
4e9cac2b PM |
1001 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
1002 | ||
881d966b | 1003 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 1004 | { |
99fe3c39 | 1005 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 1006 | |
99fe3c39 ED |
1007 | rcu_read_lock(); |
1008 | for_each_netdev_rcu(net, dev) | |
1009 | if (dev->type == type) { | |
1010 | dev_hold(dev); | |
1011 | ret = dev; | |
1012 | break; | |
1013 | } | |
1014 | rcu_read_unlock(); | |
1015 | return ret; | |
1da177e4 | 1016 | } |
1da177e4 LT |
1017 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
1018 | ||
1019 | /** | |
6c555490 | 1020 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 1021 | * @net: the applicable net namespace |
1da177e4 LT |
1022 | * @if_flags: IFF_* values |
1023 | * @mask: bitmask of bits in if_flags to check | |
1024 | * | |
1025 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 1026 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 1027 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
1028 | */ |
1029 | ||
6c555490 WC |
1030 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
1031 | unsigned short mask) | |
1da177e4 | 1032 | { |
7562f876 | 1033 | struct net_device *dev, *ret; |
1da177e4 | 1034 | |
6c555490 WC |
1035 | ASSERT_RTNL(); |
1036 | ||
7562f876 | 1037 | ret = NULL; |
6c555490 | 1038 | for_each_netdev(net, dev) { |
1da177e4 | 1039 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 1040 | ret = dev; |
1da177e4 LT |
1041 | break; |
1042 | } | |
1043 | } | |
7562f876 | 1044 | return ret; |
1da177e4 | 1045 | } |
6c555490 | 1046 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
1047 | |
1048 | /** | |
1049 | * dev_valid_name - check if name is okay for network device | |
1050 | * @name: name string | |
1051 | * | |
1052 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
1053 | * to allow sysfs to work. We also disallow any kind of |
1054 | * whitespace. | |
1da177e4 | 1055 | */ |
95f050bf | 1056 | bool dev_valid_name(const char *name) |
1da177e4 | 1057 | { |
c7fa9d18 | 1058 | if (*name == '\0') |
95f050bf | 1059 | return false; |
a9d48205 | 1060 | if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) |
95f050bf | 1061 | return false; |
c7fa9d18 | 1062 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1063 | return false; |
c7fa9d18 DM |
1064 | |
1065 | while (*name) { | |
a4176a93 | 1066 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1067 | return false; |
c7fa9d18 DM |
1068 | name++; |
1069 | } | |
95f050bf | 1070 | return true; |
1da177e4 | 1071 | } |
d1b19dff | 1072 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1073 | |
1074 | /** | |
b267b179 EB |
1075 | * __dev_alloc_name - allocate a name for a device |
1076 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1077 | * @name: name format string |
b267b179 | 1078 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1079 | * |
1080 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1081 | * id. It scans list of devices to build up a free map, then chooses |
1082 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1083 | * while allocating the name and adding the device in order to avoid | |
1084 | * duplicates. | |
1085 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1086 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1087 | */ |
1088 | ||
b267b179 | 1089 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1090 | { |
1091 | int i = 0; | |
1da177e4 LT |
1092 | const char *p; |
1093 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1094 | unsigned long *inuse; |
1da177e4 LT |
1095 | struct net_device *d; |
1096 | ||
93809105 RV |
1097 | if (!dev_valid_name(name)) |
1098 | return -EINVAL; | |
1099 | ||
51f299dd | 1100 | p = strchr(name, '%'); |
1da177e4 LT |
1101 | if (p) { |
1102 | /* | |
1103 | * Verify the string as this thing may have come from | |
1104 | * the user. There must be either one "%d" and no other "%" | |
1105 | * characters. | |
1106 | */ | |
1107 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1108 | return -EINVAL; | |
1109 | ||
1110 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1111 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1112 | if (!inuse) |
1113 | return -ENOMEM; | |
1114 | ||
881d966b | 1115 | for_each_netdev(net, d) { |
1da177e4 LT |
1116 | if (!sscanf(d->name, name, &i)) |
1117 | continue; | |
1118 | if (i < 0 || i >= max_netdevices) | |
1119 | continue; | |
1120 | ||
1121 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1122 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1123 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1124 | set_bit(i, inuse); | |
1125 | } | |
1126 | ||
1127 | i = find_first_zero_bit(inuse, max_netdevices); | |
1128 | free_page((unsigned long) inuse); | |
1129 | } | |
1130 | ||
6224abda | 1131 | snprintf(buf, IFNAMSIZ, name, i); |
b267b179 | 1132 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1133 | return i; |
1da177e4 LT |
1134 | |
1135 | /* It is possible to run out of possible slots | |
1136 | * when the name is long and there isn't enough space left | |
1137 | * for the digits, or if all bits are used. | |
1138 | */ | |
029b6d14 | 1139 | return -ENFILE; |
1da177e4 LT |
1140 | } |
1141 | ||
2c88b855 RV |
1142 | static int dev_alloc_name_ns(struct net *net, |
1143 | struct net_device *dev, | |
1144 | const char *name) | |
1145 | { | |
1146 | char buf[IFNAMSIZ]; | |
1147 | int ret; | |
1148 | ||
c46d7642 | 1149 | BUG_ON(!net); |
2c88b855 RV |
1150 | ret = __dev_alloc_name(net, name, buf); |
1151 | if (ret >= 0) | |
1152 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1153 | return ret; | |
1da177e4 LT |
1154 | } |
1155 | ||
b267b179 EB |
1156 | /** |
1157 | * dev_alloc_name - allocate a name for a device | |
1158 | * @dev: device | |
1159 | * @name: name format string | |
1160 | * | |
1161 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1162 | * id. It scans list of devices to build up a free map, then chooses | |
1163 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1164 | * while allocating the name and adding the device in order to avoid | |
1165 | * duplicates. | |
1166 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1167 | * Returns the number of the unit assigned or a negative errno code. | |
1168 | */ | |
1169 | ||
1170 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1171 | { | |
c46d7642 | 1172 | return dev_alloc_name_ns(dev_net(dev), dev, name); |
b267b179 | 1173 | } |
d1b19dff | 1174 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1175 | |
bacb7e18 ED |
1176 | static int dev_get_valid_name(struct net *net, struct net_device *dev, |
1177 | const char *name) | |
828de4f6 | 1178 | { |
55a5ec9b DM |
1179 | BUG_ON(!net); |
1180 | ||
1181 | if (!dev_valid_name(name)) | |
1182 | return -EINVAL; | |
1183 | ||
1184 | if (strchr(name, '%')) | |
1185 | return dev_alloc_name_ns(net, dev, name); | |
1186 | else if (__dev_get_by_name(net, name)) | |
1187 | return -EEXIST; | |
1188 | else if (dev->name != name) | |
1189 | strlcpy(dev->name, name, IFNAMSIZ); | |
1190 | ||
1191 | return 0; | |
d9031024 | 1192 | } |
1da177e4 LT |
1193 | |
1194 | /** | |
1195 | * dev_change_name - change name of a device | |
1196 | * @dev: device | |
1197 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1198 | * | |
1199 | * Change name of a device, can pass format strings "eth%d". | |
1200 | * for wildcarding. | |
1201 | */ | |
cf04a4c7 | 1202 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1203 | { |
238fa362 | 1204 | unsigned char old_assign_type; |
fcc5a03a | 1205 | char oldname[IFNAMSIZ]; |
1da177e4 | 1206 | int err = 0; |
fcc5a03a | 1207 | int ret; |
881d966b | 1208 | struct net *net; |
1da177e4 LT |
1209 | |
1210 | ASSERT_RTNL(); | |
c346dca1 | 1211 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1212 | |
c346dca1 | 1213 | net = dev_net(dev); |
8065a779 SWL |
1214 | |
1215 | /* Some auto-enslaved devices e.g. failover slaves are | |
1216 | * special, as userspace might rename the device after | |
1217 | * the interface had been brought up and running since | |
1218 | * the point kernel initiated auto-enslavement. Allow | |
1219 | * live name change even when these slave devices are | |
1220 | * up and running. | |
1221 | * | |
1222 | * Typically, users of these auto-enslaving devices | |
1223 | * don't actually care about slave name change, as | |
1224 | * they are supposed to operate on master interface | |
1225 | * directly. | |
1226 | */ | |
1227 | if (dev->flags & IFF_UP && | |
1228 | likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) | |
1da177e4 LT |
1229 | return -EBUSY; |
1230 | ||
30e6c9fa | 1231 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1232 | |
1233 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1234 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1235 | return 0; |
c91f6df2 | 1236 | } |
c8d90dca | 1237 | |
fcc5a03a HX |
1238 | memcpy(oldname, dev->name, IFNAMSIZ); |
1239 | ||
828de4f6 | 1240 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1241 | if (err < 0) { |
30e6c9fa | 1242 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1243 | return err; |
c91f6df2 | 1244 | } |
1da177e4 | 1245 | |
6fe82a39 VF |
1246 | if (oldname[0] && !strchr(oldname, '%')) |
1247 | netdev_info(dev, "renamed from %s\n", oldname); | |
1248 | ||
238fa362 TG |
1249 | old_assign_type = dev->name_assign_type; |
1250 | dev->name_assign_type = NET_NAME_RENAMED; | |
1251 | ||
fcc5a03a | 1252 | rollback: |
a1b3f594 EB |
1253 | ret = device_rename(&dev->dev, dev->name); |
1254 | if (ret) { | |
1255 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1256 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1257 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1258 | return ret; |
dcc99773 | 1259 | } |
7f988eab | 1260 | |
30e6c9fa | 1261 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1262 | |
5bb025fa VF |
1263 | netdev_adjacent_rename_links(dev, oldname); |
1264 | ||
7f988eab | 1265 | write_lock_bh(&dev_base_lock); |
ff927412 | 1266 | netdev_name_node_del(dev->name_node); |
72c9528b ED |
1267 | write_unlock_bh(&dev_base_lock); |
1268 | ||
1269 | synchronize_rcu(); | |
1270 | ||
1271 | write_lock_bh(&dev_base_lock); | |
ff927412 | 1272 | netdev_name_node_add(net, dev->name_node); |
7f988eab HX |
1273 | write_unlock_bh(&dev_base_lock); |
1274 | ||
056925ab | 1275 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1276 | ret = notifier_to_errno(ret); |
1277 | ||
1278 | if (ret) { | |
91e9c07b ED |
1279 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1280 | if (err >= 0) { | |
fcc5a03a | 1281 | err = ret; |
30e6c9fa | 1282 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1283 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1284 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1285 | dev->name_assign_type = old_assign_type; |
1286 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1287 | goto rollback; |
91e9c07b | 1288 | } else { |
7b6cd1ce | 1289 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1290 | dev->name, ret); |
fcc5a03a HX |
1291 | } |
1292 | } | |
1da177e4 LT |
1293 | |
1294 | return err; | |
1295 | } | |
1296 | ||
0b815a1a SH |
1297 | /** |
1298 | * dev_set_alias - change ifalias of a device | |
1299 | * @dev: device | |
1300 | * @alias: name up to IFALIASZ | |
f0db275a | 1301 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1302 | * |
1303 | * Set ifalias for a device, | |
1304 | */ | |
1305 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1306 | { | |
6c557001 | 1307 | struct dev_ifalias *new_alias = NULL; |
0b815a1a SH |
1308 | |
1309 | if (len >= IFALIASZ) | |
1310 | return -EINVAL; | |
1311 | ||
6c557001 FW |
1312 | if (len) { |
1313 | new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); | |
1314 | if (!new_alias) | |
1315 | return -ENOMEM; | |
1316 | ||
1317 | memcpy(new_alias->ifalias, alias, len); | |
1318 | new_alias->ifalias[len] = 0; | |
96ca4a2c OH |
1319 | } |
1320 | ||
6c557001 | 1321 | mutex_lock(&ifalias_mutex); |
e3f0d761 PM |
1322 | new_alias = rcu_replace_pointer(dev->ifalias, new_alias, |
1323 | mutex_is_locked(&ifalias_mutex)); | |
6c557001 FW |
1324 | mutex_unlock(&ifalias_mutex); |
1325 | ||
1326 | if (new_alias) | |
1327 | kfree_rcu(new_alias, rcuhead); | |
0b815a1a | 1328 | |
0b815a1a SH |
1329 | return len; |
1330 | } | |
0fe554a4 | 1331 | EXPORT_SYMBOL(dev_set_alias); |
0b815a1a | 1332 | |
6c557001 FW |
1333 | /** |
1334 | * dev_get_alias - get ifalias of a device | |
1335 | * @dev: device | |
20e88320 | 1336 | * @name: buffer to store name of ifalias |
6c557001 FW |
1337 | * @len: size of buffer |
1338 | * | |
1339 | * get ifalias for a device. Caller must make sure dev cannot go | |
1340 | * away, e.g. rcu read lock or own a reference count to device. | |
1341 | */ | |
1342 | int dev_get_alias(const struct net_device *dev, char *name, size_t len) | |
1343 | { | |
1344 | const struct dev_ifalias *alias; | |
1345 | int ret = 0; | |
1346 | ||
1347 | rcu_read_lock(); | |
1348 | alias = rcu_dereference(dev->ifalias); | |
1349 | if (alias) | |
1350 | ret = snprintf(name, len, "%s", alias->ifalias); | |
1351 | rcu_read_unlock(); | |
1352 | ||
1353 | return ret; | |
1354 | } | |
0b815a1a | 1355 | |
d8a33ac4 | 1356 | /** |
3041a069 | 1357 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1358 | * @dev: device to cause notification |
1359 | * | |
1360 | * Called to indicate a device has changed features. | |
1361 | */ | |
1362 | void netdev_features_change(struct net_device *dev) | |
1363 | { | |
056925ab | 1364 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1365 | } |
1366 | EXPORT_SYMBOL(netdev_features_change); | |
1367 | ||
1da177e4 LT |
1368 | /** |
1369 | * netdev_state_change - device changes state | |
1370 | * @dev: device to cause notification | |
1371 | * | |
1372 | * Called to indicate a device has changed state. This function calls | |
1373 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1374 | * to the routing socket. | |
1375 | */ | |
1376 | void netdev_state_change(struct net_device *dev) | |
1377 | { | |
1378 | if (dev->flags & IFF_UP) { | |
51d0c047 DA |
1379 | struct netdev_notifier_change_info change_info = { |
1380 | .info.dev = dev, | |
1381 | }; | |
54951194 | 1382 | |
51d0c047 | 1383 | call_netdevice_notifiers_info(NETDEV_CHANGE, |
54951194 | 1384 | &change_info.info); |
7f294054 | 1385 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1386 | } |
1387 | } | |
d1b19dff | 1388 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1389 | |
ee89bab1 | 1390 | /** |
722c9a0c | 1391 | * netdev_notify_peers - notify network peers about existence of @dev |
1392 | * @dev: network device | |
ee89bab1 AW |
1393 | * |
1394 | * Generate traffic such that interested network peers are aware of | |
1395 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1396 | * a device wants to inform the rest of the network about some sort of | |
1397 | * reconfiguration such as a failover event or virtual machine | |
1398 | * migration. | |
1399 | */ | |
1400 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1401 | { |
ee89bab1 AW |
1402 | rtnl_lock(); |
1403 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
37c343b4 | 1404 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); |
ee89bab1 | 1405 | rtnl_unlock(); |
c1da4ac7 | 1406 | } |
ee89bab1 | 1407 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1408 | |
40c900aa | 1409 | static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
1da177e4 | 1410 | { |
d314774c | 1411 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1412 | int ret; |
1da177e4 | 1413 | |
e46b66bc BH |
1414 | ASSERT_RTNL(); |
1415 | ||
1da177e4 LT |
1416 | if (!netif_device_present(dev)) |
1417 | return -ENODEV; | |
1418 | ||
ca99ca14 NH |
1419 | /* Block netpoll from trying to do any rx path servicing. |
1420 | * If we don't do this there is a chance ndo_poll_controller | |
1421 | * or ndo_poll may be running while we open the device | |
1422 | */ | |
66b5552f | 1423 | netpoll_poll_disable(dev); |
ca99ca14 | 1424 | |
40c900aa | 1425 | ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); |
3b8bcfd5 JB |
1426 | ret = notifier_to_errno(ret); |
1427 | if (ret) | |
1428 | return ret; | |
1429 | ||
1da177e4 | 1430 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1431 | |
d314774c SH |
1432 | if (ops->ndo_validate_addr) |
1433 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1434 | |
d314774c SH |
1435 | if (!ret && ops->ndo_open) |
1436 | ret = ops->ndo_open(dev); | |
1da177e4 | 1437 | |
66b5552f | 1438 | netpoll_poll_enable(dev); |
ca99ca14 | 1439 | |
bada339b JG |
1440 | if (ret) |
1441 | clear_bit(__LINK_STATE_START, &dev->state); | |
1442 | else { | |
1da177e4 | 1443 | dev->flags |= IFF_UP; |
4417da66 | 1444 | dev_set_rx_mode(dev); |
1da177e4 | 1445 | dev_activate(dev); |
7bf23575 | 1446 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1447 | } |
bada339b | 1448 | |
1da177e4 LT |
1449 | return ret; |
1450 | } | |
1451 | ||
1452 | /** | |
bd380811 | 1453 | * dev_open - prepare an interface for use. |
00f54e68 PM |
1454 | * @dev: device to open |
1455 | * @extack: netlink extended ack | |
1da177e4 | 1456 | * |
bd380811 PM |
1457 | * Takes a device from down to up state. The device's private open |
1458 | * function is invoked and then the multicast lists are loaded. Finally | |
1459 | * the device is moved into the up state and a %NETDEV_UP message is | |
1460 | * sent to the netdev notifier chain. | |
1461 | * | |
1462 | * Calling this function on an active interface is a nop. On a failure | |
1463 | * a negative errno code is returned. | |
1da177e4 | 1464 | */ |
00f54e68 | 1465 | int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
bd380811 PM |
1466 | { |
1467 | int ret; | |
1468 | ||
bd380811 PM |
1469 | if (dev->flags & IFF_UP) |
1470 | return 0; | |
1471 | ||
40c900aa | 1472 | ret = __dev_open(dev, extack); |
bd380811 PM |
1473 | if (ret < 0) |
1474 | return ret; | |
1475 | ||
7f294054 | 1476 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1477 | call_netdevice_notifiers(NETDEV_UP, dev); |
1478 | ||
1479 | return ret; | |
1480 | } | |
1481 | EXPORT_SYMBOL(dev_open); | |
1482 | ||
7051b88a | 1483 | static void __dev_close_many(struct list_head *head) |
1da177e4 | 1484 | { |
44345724 | 1485 | struct net_device *dev; |
e46b66bc | 1486 | |
bd380811 | 1487 | ASSERT_RTNL(); |
9d5010db DM |
1488 | might_sleep(); |
1489 | ||
5cde2829 | 1490 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1491 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1492 | netpoll_poll_disable(dev); |
3f4df206 | 1493 | |
44345724 | 1494 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1495 | |
44345724 | 1496 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1497 | |
44345724 OP |
1498 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1499 | * can be even on different cpu. So just clear netif_running(). | |
1500 | * | |
1501 | * dev->stop() will invoke napi_disable() on all of it's | |
1502 | * napi_struct instances on this device. | |
1503 | */ | |
4e857c58 | 1504 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1505 | } |
1da177e4 | 1506 | |
44345724 | 1507 | dev_deactivate_many(head); |
d8b2a4d2 | 1508 | |
5cde2829 | 1509 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1510 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1511 | |
44345724 OP |
1512 | /* |
1513 | * Call the device specific close. This cannot fail. | |
1514 | * Only if device is UP | |
1515 | * | |
1516 | * We allow it to be called even after a DETACH hot-plug | |
1517 | * event. | |
1518 | */ | |
1519 | if (ops->ndo_stop) | |
1520 | ops->ndo_stop(dev); | |
1521 | ||
44345724 | 1522 | dev->flags &= ~IFF_UP; |
66b5552f | 1523 | netpoll_poll_enable(dev); |
44345724 | 1524 | } |
44345724 OP |
1525 | } |
1526 | ||
7051b88a | 1527 | static void __dev_close(struct net_device *dev) |
44345724 OP |
1528 | { |
1529 | LIST_HEAD(single); | |
1530 | ||
5cde2829 | 1531 | list_add(&dev->close_list, &single); |
7051b88a | 1532 | __dev_close_many(&single); |
f87e6f47 | 1533 | list_del(&single); |
44345724 OP |
1534 | } |
1535 | ||
7051b88a | 1536 | void dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1537 | { |
1538 | struct net_device *dev, *tmp; | |
1da177e4 | 1539 | |
5cde2829 EB |
1540 | /* Remove the devices that don't need to be closed */ |
1541 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1542 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1543 | list_del_init(&dev->close_list); |
44345724 OP |
1544 | |
1545 | __dev_close_many(head); | |
1da177e4 | 1546 | |
5cde2829 | 1547 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1548 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1549 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1550 | if (unlink) |
1551 | list_del_init(&dev->close_list); | |
44345724 | 1552 | } |
bd380811 | 1553 | } |
99c4a26a | 1554 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1555 | |
1556 | /** | |
1557 | * dev_close - shutdown an interface. | |
1558 | * @dev: device to shutdown | |
1559 | * | |
1560 | * This function moves an active device into down state. A | |
1561 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1562 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1563 | * chain. | |
1564 | */ | |
7051b88a | 1565 | void dev_close(struct net_device *dev) |
bd380811 | 1566 | { |
e14a5993 ED |
1567 | if (dev->flags & IFF_UP) { |
1568 | LIST_HEAD(single); | |
1da177e4 | 1569 | |
5cde2829 | 1570 | list_add(&dev->close_list, &single); |
99c4a26a | 1571 | dev_close_many(&single, true); |
e14a5993 ED |
1572 | list_del(&single); |
1573 | } | |
1da177e4 | 1574 | } |
d1b19dff | 1575 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1576 | |
1577 | ||
0187bdfb BH |
1578 | /** |
1579 | * dev_disable_lro - disable Large Receive Offload on a device | |
1580 | * @dev: device | |
1581 | * | |
1582 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1583 | * called under RTNL. This is needed if received packets may be | |
1584 | * forwarded to another interface. | |
1585 | */ | |
1586 | void dev_disable_lro(struct net_device *dev) | |
1587 | { | |
fbe168ba MK |
1588 | struct net_device *lower_dev; |
1589 | struct list_head *iter; | |
529d0489 | 1590 | |
bc5787c6 MM |
1591 | dev->wanted_features &= ~NETIF_F_LRO; |
1592 | netdev_update_features(dev); | |
27660515 | 1593 | |
22d5969f MM |
1594 | if (unlikely(dev->features & NETIF_F_LRO)) |
1595 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1596 | |
1597 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1598 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1599 | } |
1600 | EXPORT_SYMBOL(dev_disable_lro); | |
1601 | ||
56f5aa77 MC |
1602 | /** |
1603 | * dev_disable_gro_hw - disable HW Generic Receive Offload on a device | |
1604 | * @dev: device | |
1605 | * | |
1606 | * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be | |
1607 | * called under RTNL. This is needed if Generic XDP is installed on | |
1608 | * the device. | |
1609 | */ | |
1610 | static void dev_disable_gro_hw(struct net_device *dev) | |
1611 | { | |
1612 | dev->wanted_features &= ~NETIF_F_GRO_HW; | |
1613 | netdev_update_features(dev); | |
1614 | ||
1615 | if (unlikely(dev->features & NETIF_F_GRO_HW)) | |
1616 | netdev_WARN(dev, "failed to disable GRO_HW!\n"); | |
1617 | } | |
1618 | ||
ede2762d KT |
1619 | const char *netdev_cmd_to_name(enum netdev_cmd cmd) |
1620 | { | |
1621 | #define N(val) \ | |
1622 | case NETDEV_##val: \ | |
1623 | return "NETDEV_" __stringify(val); | |
1624 | switch (cmd) { | |
1625 | N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) | |
1626 | N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) | |
1627 | N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) | |
1628 | N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) | |
1629 | N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) | |
1630 | N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) | |
1631 | N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) | |
9daae9bd GP |
1632 | N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) |
1633 | N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) | |
1570415f | 1634 | N(PRE_CHANGEADDR) |
3f5ecd8a | 1635 | } |
ede2762d KT |
1636 | #undef N |
1637 | return "UNKNOWN_NETDEV_EVENT"; | |
1638 | } | |
1639 | EXPORT_SYMBOL_GPL(netdev_cmd_to_name); | |
1640 | ||
351638e7 JP |
1641 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1642 | struct net_device *dev) | |
1643 | { | |
51d0c047 DA |
1644 | struct netdev_notifier_info info = { |
1645 | .dev = dev, | |
1646 | }; | |
351638e7 | 1647 | |
351638e7 JP |
1648 | return nb->notifier_call(nb, val, &info); |
1649 | } | |
0187bdfb | 1650 | |
afa0df59 JP |
1651 | static int call_netdevice_register_notifiers(struct notifier_block *nb, |
1652 | struct net_device *dev) | |
1653 | { | |
1654 | int err; | |
1655 | ||
1656 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); | |
1657 | err = notifier_to_errno(err); | |
1658 | if (err) | |
1659 | return err; | |
1660 | ||
1661 | if (!(dev->flags & IFF_UP)) | |
1662 | return 0; | |
1663 | ||
1664 | call_netdevice_notifier(nb, NETDEV_UP, dev); | |
1665 | return 0; | |
1666 | } | |
1667 | ||
1668 | static void call_netdevice_unregister_notifiers(struct notifier_block *nb, | |
1669 | struct net_device *dev) | |
1670 | { | |
1671 | if (dev->flags & IFF_UP) { | |
1672 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, | |
1673 | dev); | |
1674 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
1675 | } | |
1676 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); | |
1677 | } | |
1678 | ||
1679 | static int call_netdevice_register_net_notifiers(struct notifier_block *nb, | |
1680 | struct net *net) | |
1681 | { | |
1682 | struct net_device *dev; | |
1683 | int err; | |
1684 | ||
1685 | for_each_netdev(net, dev) { | |
1686 | err = call_netdevice_register_notifiers(nb, dev); | |
1687 | if (err) | |
1688 | goto rollback; | |
1689 | } | |
1690 | return 0; | |
1691 | ||
1692 | rollback: | |
1693 | for_each_netdev_continue_reverse(net, dev) | |
1694 | call_netdevice_unregister_notifiers(nb, dev); | |
1695 | return err; | |
1696 | } | |
1697 | ||
1698 | static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, | |
1699 | struct net *net) | |
1700 | { | |
1701 | struct net_device *dev; | |
1702 | ||
1703 | for_each_netdev(net, dev) | |
1704 | call_netdevice_unregister_notifiers(nb, dev); | |
1705 | } | |
1706 | ||
881d966b EB |
1707 | static int dev_boot_phase = 1; |
1708 | ||
1da177e4 | 1709 | /** |
722c9a0c | 1710 | * register_netdevice_notifier - register a network notifier block |
1711 | * @nb: notifier | |
1da177e4 | 1712 | * |
722c9a0c | 1713 | * Register a notifier to be called when network device events occur. |
1714 | * The notifier passed is linked into the kernel structures and must | |
1715 | * not be reused until it has been unregistered. A negative errno code | |
1716 | * is returned on a failure. | |
1da177e4 | 1717 | * |
722c9a0c | 1718 | * When registered all registration and up events are replayed |
1719 | * to the new notifier to allow device to have a race free | |
1720 | * view of the network device list. | |
1da177e4 LT |
1721 | */ |
1722 | ||
1723 | int register_netdevice_notifier(struct notifier_block *nb) | |
1724 | { | |
881d966b | 1725 | struct net *net; |
1da177e4 LT |
1726 | int err; |
1727 | ||
328fbe74 KT |
1728 | /* Close race with setup_net() and cleanup_net() */ |
1729 | down_write(&pernet_ops_rwsem); | |
1da177e4 | 1730 | rtnl_lock(); |
f07d5b94 | 1731 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1732 | if (err) |
1733 | goto unlock; | |
881d966b EB |
1734 | if (dev_boot_phase) |
1735 | goto unlock; | |
1736 | for_each_net(net) { | |
afa0df59 JP |
1737 | err = call_netdevice_register_net_notifiers(nb, net); |
1738 | if (err) | |
1739 | goto rollback; | |
1da177e4 | 1740 | } |
fcc5a03a HX |
1741 | |
1742 | unlock: | |
1da177e4 | 1743 | rtnl_unlock(); |
328fbe74 | 1744 | up_write(&pernet_ops_rwsem); |
1da177e4 | 1745 | return err; |
fcc5a03a HX |
1746 | |
1747 | rollback: | |
afa0df59 JP |
1748 | for_each_net_continue_reverse(net) |
1749 | call_netdevice_unregister_net_notifiers(nb, net); | |
c67625a1 PE |
1750 | |
1751 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1752 | goto unlock; |
1da177e4 | 1753 | } |
d1b19dff | 1754 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1755 | |
1756 | /** | |
722c9a0c | 1757 | * unregister_netdevice_notifier - unregister a network notifier block |
1758 | * @nb: notifier | |
1da177e4 | 1759 | * |
722c9a0c | 1760 | * Unregister a notifier previously registered by |
1761 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1762 | * kernel structures and may then be reused. A negative errno code | |
1763 | * is returned on a failure. | |
7d3d43da | 1764 | * |
722c9a0c | 1765 | * After unregistering unregister and down device events are synthesized |
1766 | * for all devices on the device list to the removed notifier to remove | |
1767 | * the need for special case cleanup code. | |
1da177e4 LT |
1768 | */ |
1769 | ||
1770 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1771 | { | |
7d3d43da | 1772 | struct net *net; |
9f514950 HX |
1773 | int err; |
1774 | ||
328fbe74 KT |
1775 | /* Close race with setup_net() and cleanup_net() */ |
1776 | down_write(&pernet_ops_rwsem); | |
9f514950 | 1777 | rtnl_lock(); |
f07d5b94 | 1778 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1779 | if (err) |
1780 | goto unlock; | |
1781 | ||
48b3a137 JP |
1782 | for_each_net(net) |
1783 | call_netdevice_unregister_net_notifiers(nb, net); | |
1784 | ||
7d3d43da | 1785 | unlock: |
9f514950 | 1786 | rtnl_unlock(); |
328fbe74 | 1787 | up_write(&pernet_ops_rwsem); |
9f514950 | 1788 | return err; |
1da177e4 | 1789 | } |
d1b19dff | 1790 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1791 | |
1f637703 JP |
1792 | static int __register_netdevice_notifier_net(struct net *net, |
1793 | struct notifier_block *nb, | |
1794 | bool ignore_call_fail) | |
1795 | { | |
1796 | int err; | |
1797 | ||
1798 | err = raw_notifier_chain_register(&net->netdev_chain, nb); | |
1799 | if (err) | |
1800 | return err; | |
1801 | if (dev_boot_phase) | |
1802 | return 0; | |
1803 | ||
1804 | err = call_netdevice_register_net_notifiers(nb, net); | |
1805 | if (err && !ignore_call_fail) | |
1806 | goto chain_unregister; | |
1807 | ||
1808 | return 0; | |
1809 | ||
1810 | chain_unregister: | |
1811 | raw_notifier_chain_unregister(&net->netdev_chain, nb); | |
1812 | return err; | |
1813 | } | |
1814 | ||
1815 | static int __unregister_netdevice_notifier_net(struct net *net, | |
1816 | struct notifier_block *nb) | |
1817 | { | |
1818 | int err; | |
1819 | ||
1820 | err = raw_notifier_chain_unregister(&net->netdev_chain, nb); | |
1821 | if (err) | |
1822 | return err; | |
1823 | ||
1824 | call_netdevice_unregister_net_notifiers(nb, net); | |
1825 | return 0; | |
1826 | } | |
1827 | ||
a30c7b42 JP |
1828 | /** |
1829 | * register_netdevice_notifier_net - register a per-netns network notifier block | |
1830 | * @net: network namespace | |
1831 | * @nb: notifier | |
1832 | * | |
1833 | * Register a notifier to be called when network device events occur. | |
1834 | * The notifier passed is linked into the kernel structures and must | |
1835 | * not be reused until it has been unregistered. A negative errno code | |
1836 | * is returned on a failure. | |
1837 | * | |
1838 | * When registered all registration and up events are replayed | |
1839 | * to the new notifier to allow device to have a race free | |
1840 | * view of the network device list. | |
1841 | */ | |
1842 | ||
1843 | int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) | |
1844 | { | |
1845 | int err; | |
1846 | ||
1847 | rtnl_lock(); | |
1f637703 | 1848 | err = __register_netdevice_notifier_net(net, nb, false); |
a30c7b42 JP |
1849 | rtnl_unlock(); |
1850 | return err; | |
a30c7b42 JP |
1851 | } |
1852 | EXPORT_SYMBOL(register_netdevice_notifier_net); | |
1853 | ||
1854 | /** | |
1855 | * unregister_netdevice_notifier_net - unregister a per-netns | |
1856 | * network notifier block | |
1857 | * @net: network namespace | |
1858 | * @nb: notifier | |
1859 | * | |
1860 | * Unregister a notifier previously registered by | |
1861 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1862 | * kernel structures and may then be reused. A negative errno code | |
1863 | * is returned on a failure. | |
1864 | * | |
1865 | * After unregistering unregister and down device events are synthesized | |
1866 | * for all devices on the device list to the removed notifier to remove | |
1867 | * the need for special case cleanup code. | |
1868 | */ | |
1869 | ||
1870 | int unregister_netdevice_notifier_net(struct net *net, | |
1871 | struct notifier_block *nb) | |
1872 | { | |
1873 | int err; | |
1874 | ||
1875 | rtnl_lock(); | |
1f637703 | 1876 | err = __unregister_netdevice_notifier_net(net, nb); |
a30c7b42 JP |
1877 | rtnl_unlock(); |
1878 | return err; | |
1879 | } | |
1880 | EXPORT_SYMBOL(unregister_netdevice_notifier_net); | |
a30c7b42 | 1881 | |
93642e14 JP |
1882 | int register_netdevice_notifier_dev_net(struct net_device *dev, |
1883 | struct notifier_block *nb, | |
1884 | struct netdev_net_notifier *nn) | |
1885 | { | |
1886 | int err; | |
a30c7b42 | 1887 | |
93642e14 JP |
1888 | rtnl_lock(); |
1889 | err = __register_netdevice_notifier_net(dev_net(dev), nb, false); | |
1890 | if (!err) { | |
1891 | nn->nb = nb; | |
1892 | list_add(&nn->list, &dev->net_notifier_list); | |
1893 | } | |
a30c7b42 JP |
1894 | rtnl_unlock(); |
1895 | return err; | |
1896 | } | |
93642e14 JP |
1897 | EXPORT_SYMBOL(register_netdevice_notifier_dev_net); |
1898 | ||
1899 | int unregister_netdevice_notifier_dev_net(struct net_device *dev, | |
1900 | struct notifier_block *nb, | |
1901 | struct netdev_net_notifier *nn) | |
1902 | { | |
1903 | int err; | |
1904 | ||
1905 | rtnl_lock(); | |
1906 | list_del(&nn->list); | |
1907 | err = __unregister_netdevice_notifier_net(dev_net(dev), nb); | |
1908 | rtnl_unlock(); | |
1909 | return err; | |
1910 | } | |
1911 | EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); | |
1912 | ||
1913 | static void move_netdevice_notifiers_dev_net(struct net_device *dev, | |
1914 | struct net *net) | |
1915 | { | |
1916 | struct netdev_net_notifier *nn; | |
1917 | ||
1918 | list_for_each_entry(nn, &dev->net_notifier_list, list) { | |
1919 | __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); | |
1920 | __register_netdevice_notifier_net(net, nn->nb, true); | |
1921 | } | |
1922 | } | |
a30c7b42 | 1923 | |
351638e7 JP |
1924 | /** |
1925 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1926 | * @val: value passed unmodified to notifier function | |
351638e7 JP |
1927 | * @info: notifier information data |
1928 | * | |
1929 | * Call all network notifier blocks. Parameters and return value | |
1930 | * are as for raw_notifier_call_chain(). | |
1931 | */ | |
1932 | ||
1d143d9f | 1933 | static int call_netdevice_notifiers_info(unsigned long val, |
1d143d9f | 1934 | struct netdev_notifier_info *info) |
351638e7 | 1935 | { |
a30c7b42 JP |
1936 | struct net *net = dev_net(info->dev); |
1937 | int ret; | |
1938 | ||
351638e7 | 1939 | ASSERT_RTNL(); |
a30c7b42 JP |
1940 | |
1941 | /* Run per-netns notifier block chain first, then run the global one. | |
1942 | * Hopefully, one day, the global one is going to be removed after | |
1943 | * all notifier block registrators get converted to be per-netns. | |
1944 | */ | |
1945 | ret = raw_notifier_call_chain(&net->netdev_chain, val, info); | |
1946 | if (ret & NOTIFY_STOP_MASK) | |
1947 | return ret; | |
351638e7 JP |
1948 | return raw_notifier_call_chain(&netdev_chain, val, info); |
1949 | } | |
351638e7 | 1950 | |
26372605 PM |
1951 | static int call_netdevice_notifiers_extack(unsigned long val, |
1952 | struct net_device *dev, | |
1953 | struct netlink_ext_ack *extack) | |
1954 | { | |
1955 | struct netdev_notifier_info info = { | |
1956 | .dev = dev, | |
1957 | .extack = extack, | |
1958 | }; | |
1959 | ||
1960 | return call_netdevice_notifiers_info(val, &info); | |
1961 | } | |
1962 | ||
1da177e4 LT |
1963 | /** |
1964 | * call_netdevice_notifiers - call all network notifier blocks | |
1965 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1966 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1967 | * |
1968 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1969 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1970 | */ |
1971 | ||
ad7379d4 | 1972 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1973 | { |
26372605 | 1974 | return call_netdevice_notifiers_extack(val, dev, NULL); |
1da177e4 | 1975 | } |
edf947f1 | 1976 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1977 | |
af7d6cce SD |
1978 | /** |
1979 | * call_netdevice_notifiers_mtu - call all network notifier blocks | |
1980 | * @val: value passed unmodified to notifier function | |
1981 | * @dev: net_device pointer passed unmodified to notifier function | |
1982 | * @arg: additional u32 argument passed to the notifier function | |
1983 | * | |
1984 | * Call all network notifier blocks. Parameters and return value | |
1985 | * are as for raw_notifier_call_chain(). | |
1986 | */ | |
1987 | static int call_netdevice_notifiers_mtu(unsigned long val, | |
1988 | struct net_device *dev, u32 arg) | |
1989 | { | |
1990 | struct netdev_notifier_info_ext info = { | |
1991 | .info.dev = dev, | |
1992 | .ext.mtu = arg, | |
1993 | }; | |
1994 | ||
1995 | BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); | |
1996 | ||
1997 | return call_netdevice_notifiers_info(val, &info.info); | |
1998 | } | |
1999 | ||
1cf51900 | 2000 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 2001 | static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); |
4577139b DB |
2002 | |
2003 | void net_inc_ingress_queue(void) | |
2004 | { | |
aabf6772 | 2005 | static_branch_inc(&ingress_needed_key); |
4577139b DB |
2006 | } |
2007 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
2008 | ||
2009 | void net_dec_ingress_queue(void) | |
2010 | { | |
aabf6772 | 2011 | static_branch_dec(&ingress_needed_key); |
4577139b DB |
2012 | } |
2013 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
2014 | #endif | |
2015 | ||
1f211a1b | 2016 | #ifdef CONFIG_NET_EGRESS |
aabf6772 | 2017 | static DEFINE_STATIC_KEY_FALSE(egress_needed_key); |
1f211a1b DB |
2018 | |
2019 | void net_inc_egress_queue(void) | |
2020 | { | |
aabf6772 | 2021 | static_branch_inc(&egress_needed_key); |
1f211a1b DB |
2022 | } |
2023 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
2024 | ||
2025 | void net_dec_egress_queue(void) | |
2026 | { | |
aabf6772 | 2027 | static_branch_dec(&egress_needed_key); |
1f211a1b DB |
2028 | } |
2029 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
2030 | #endif | |
2031 | ||
39e83922 | 2032 | static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); |
e9666d10 | 2033 | #ifdef CONFIG_JUMP_LABEL |
b90e5794 | 2034 | static atomic_t netstamp_needed_deferred; |
13baa00a | 2035 | static atomic_t netstamp_wanted; |
5fa8bbda | 2036 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 2037 | { |
b90e5794 | 2038 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 2039 | int wanted; |
b90e5794 | 2040 | |
13baa00a ED |
2041 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
2042 | if (wanted > 0) | |
39e83922 | 2043 | static_branch_enable(&netstamp_needed_key); |
13baa00a | 2044 | else |
39e83922 | 2045 | static_branch_disable(&netstamp_needed_key); |
5fa8bbda ED |
2046 | } |
2047 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 2048 | #endif |
5fa8bbda ED |
2049 | |
2050 | void net_enable_timestamp(void) | |
2051 | { | |
e9666d10 | 2052 | #ifdef CONFIG_JUMP_LABEL |
13baa00a ED |
2053 | int wanted; |
2054 | ||
2055 | while (1) { | |
2056 | wanted = atomic_read(&netstamp_wanted); | |
2057 | if (wanted <= 0) | |
2058 | break; | |
2059 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) | |
2060 | return; | |
2061 | } | |
2062 | atomic_inc(&netstamp_needed_deferred); | |
2063 | schedule_work(&netstamp_work); | |
2064 | #else | |
39e83922 | 2065 | static_branch_inc(&netstamp_needed_key); |
13baa00a | 2066 | #endif |
1da177e4 | 2067 | } |
d1b19dff | 2068 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
2069 | |
2070 | void net_disable_timestamp(void) | |
2071 | { | |
e9666d10 | 2072 | #ifdef CONFIG_JUMP_LABEL |
13baa00a ED |
2073 | int wanted; |
2074 | ||
2075 | while (1) { | |
2076 | wanted = atomic_read(&netstamp_wanted); | |
2077 | if (wanted <= 1) | |
2078 | break; | |
2079 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) | |
2080 | return; | |
2081 | } | |
2082 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
2083 | schedule_work(&netstamp_work); |
2084 | #else | |
39e83922 | 2085 | static_branch_dec(&netstamp_needed_key); |
5fa8bbda | 2086 | #endif |
1da177e4 | 2087 | } |
d1b19dff | 2088 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 2089 | |
3b098e2d | 2090 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 2091 | { |
2456e855 | 2092 | skb->tstamp = 0; |
39e83922 | 2093 | if (static_branch_unlikely(&netstamp_needed_key)) |
a61bbcf2 | 2094 | __net_timestamp(skb); |
1da177e4 LT |
2095 | } |
2096 | ||
39e83922 DB |
2097 | #define net_timestamp_check(COND, SKB) \ |
2098 | if (static_branch_unlikely(&netstamp_needed_key)) { \ | |
2099 | if ((COND) && !(SKB)->tstamp) \ | |
2100 | __net_timestamp(SKB); \ | |
2101 | } \ | |
3b098e2d | 2102 | |
f4b05d27 | 2103 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 DL |
2104 | { |
2105 | unsigned int len; | |
2106 | ||
2107 | if (!(dev->flags & IFF_UP)) | |
2108 | return false; | |
2109 | ||
2110 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
2111 | if (skb->len <= len) | |
2112 | return true; | |
2113 | ||
2114 | /* if TSO is enabled, we don't care about the length as the packet | |
2115 | * could be forwarded without being segmented before | |
2116 | */ | |
2117 | if (skb_is_gso(skb)) | |
2118 | return true; | |
2119 | ||
2120 | return false; | |
2121 | } | |
1ee481fb | 2122 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 2123 | |
a0265d28 HX |
2124 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
2125 | { | |
4e3264d2 | 2126 | int ret = ____dev_forward_skb(dev, skb); |
a0265d28 | 2127 | |
4e3264d2 MKL |
2128 | if (likely(!ret)) { |
2129 | skb->protocol = eth_type_trans(skb, dev); | |
2130 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
2131 | } | |
a0265d28 | 2132 | |
4e3264d2 | 2133 | return ret; |
a0265d28 HX |
2134 | } |
2135 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
2136 | ||
44540960 AB |
2137 | /** |
2138 | * dev_forward_skb - loopback an skb to another netif | |
2139 | * | |
2140 | * @dev: destination network device | |
2141 | * @skb: buffer to forward | |
2142 | * | |
2143 | * return values: | |
2144 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 2145 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
2146 | * |
2147 | * dev_forward_skb can be used for injecting an skb from the | |
2148 | * start_xmit function of one device into the receive queue | |
2149 | * of another device. | |
2150 | * | |
2151 | * The receiving device may be in another namespace, so | |
2152 | * we have to clear all information in the skb that could | |
2153 | * impact namespace isolation. | |
2154 | */ | |
2155 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
2156 | { | |
a0265d28 | 2157 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
2158 | } |
2159 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
2160 | ||
71d9dec2 CG |
2161 | static inline int deliver_skb(struct sk_buff *skb, |
2162 | struct packet_type *pt_prev, | |
2163 | struct net_device *orig_dev) | |
2164 | { | |
1f8b977a | 2165 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
1080e512 | 2166 | return -ENOMEM; |
63354797 | 2167 | refcount_inc(&skb->users); |
71d9dec2 CG |
2168 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
2169 | } | |
2170 | ||
7866a621 SN |
2171 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
2172 | struct packet_type **pt, | |
fbcb2170 JP |
2173 | struct net_device *orig_dev, |
2174 | __be16 type, | |
7866a621 SN |
2175 | struct list_head *ptype_list) |
2176 | { | |
2177 | struct packet_type *ptype, *pt_prev = *pt; | |
2178 | ||
2179 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
2180 | if (ptype->type != type) | |
2181 | continue; | |
2182 | if (pt_prev) | |
fbcb2170 | 2183 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
2184 | pt_prev = ptype; |
2185 | } | |
2186 | *pt = pt_prev; | |
2187 | } | |
2188 | ||
c0de08d0 EL |
2189 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
2190 | { | |
a3d744e9 | 2191 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
2192 | return false; |
2193 | ||
2194 | if (ptype->id_match) | |
2195 | return ptype->id_match(ptype, skb->sk); | |
2196 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
2197 | return true; | |
2198 | ||
2199 | return false; | |
2200 | } | |
2201 | ||
9f9a742d MR |
2202 | /** |
2203 | * dev_nit_active - return true if any network interface taps are in use | |
2204 | * | |
2205 | * @dev: network device to check for the presence of taps | |
2206 | */ | |
2207 | bool dev_nit_active(struct net_device *dev) | |
2208 | { | |
2209 | return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); | |
2210 | } | |
2211 | EXPORT_SYMBOL_GPL(dev_nit_active); | |
2212 | ||
1da177e4 LT |
2213 | /* |
2214 | * Support routine. Sends outgoing frames to any network | |
2215 | * taps currently in use. | |
2216 | */ | |
2217 | ||
74b20582 | 2218 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
2219 | { |
2220 | struct packet_type *ptype; | |
71d9dec2 CG |
2221 | struct sk_buff *skb2 = NULL; |
2222 | struct packet_type *pt_prev = NULL; | |
7866a621 | 2223 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 2224 | |
1da177e4 | 2225 | rcu_read_lock(); |
7866a621 SN |
2226 | again: |
2227 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
fa788d98 VW |
2228 | if (ptype->ignore_outgoing) |
2229 | continue; | |
2230 | ||
1da177e4 LT |
2231 | /* Never send packets back to the socket |
2232 | * they originated from - MvS (miquels@drinkel.ow.org) | |
2233 | */ | |
7866a621 SN |
2234 | if (skb_loop_sk(ptype, skb)) |
2235 | continue; | |
71d9dec2 | 2236 | |
7866a621 SN |
2237 | if (pt_prev) { |
2238 | deliver_skb(skb2, pt_prev, skb->dev); | |
2239 | pt_prev = ptype; | |
2240 | continue; | |
2241 | } | |
1da177e4 | 2242 | |
7866a621 SN |
2243 | /* need to clone skb, done only once */ |
2244 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
2245 | if (!skb2) | |
2246 | goto out_unlock; | |
70978182 | 2247 | |
7866a621 | 2248 | net_timestamp_set(skb2); |
1da177e4 | 2249 | |
7866a621 SN |
2250 | /* skb->nh should be correctly |
2251 | * set by sender, so that the second statement is | |
2252 | * just protection against buggy protocols. | |
2253 | */ | |
2254 | skb_reset_mac_header(skb2); | |
2255 | ||
2256 | if (skb_network_header(skb2) < skb2->data || | |
2257 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
2258 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
2259 | ntohs(skb2->protocol), | |
2260 | dev->name); | |
2261 | skb_reset_network_header(skb2); | |
1da177e4 | 2262 | } |
7866a621 SN |
2263 | |
2264 | skb2->transport_header = skb2->network_header; | |
2265 | skb2->pkt_type = PACKET_OUTGOING; | |
2266 | pt_prev = ptype; | |
2267 | } | |
2268 | ||
2269 | if (ptype_list == &ptype_all) { | |
2270 | ptype_list = &dev->ptype_all; | |
2271 | goto again; | |
1da177e4 | 2272 | } |
7866a621 | 2273 | out_unlock: |
581fe0ea WB |
2274 | if (pt_prev) { |
2275 | if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) | |
2276 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
2277 | else | |
2278 | kfree_skb(skb2); | |
2279 | } | |
1da177e4 LT |
2280 | rcu_read_unlock(); |
2281 | } | |
74b20582 | 2282 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 2283 | |
2c53040f BH |
2284 | /** |
2285 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
2286 | * @dev: Network device |
2287 | * @txq: number of queues available | |
2288 | * | |
2289 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
2290 | * valid. To resolve this verify the tc mapping remains valid and if | |
2291 | * not NULL the mapping. With no priorities mapping to this | |
2292 | * offset/count pair it will no longer be used. In the worst case TC0 | |
2293 | * is invalid nothing can be done so disable priority mappings. If is | |
2294 | * expected that drivers will fix this mapping if they can before | |
2295 | * calling netif_set_real_num_tx_queues. | |
2296 | */ | |
bb134d22 | 2297 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
2298 | { |
2299 | int i; | |
2300 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2301 | ||
2302 | /* If TC0 is invalidated disable TC mapping */ | |
2303 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 2304 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
2305 | dev->num_tc = 0; |
2306 | return; | |
2307 | } | |
2308 | ||
2309 | /* Invalidated prio to tc mappings set to TC0 */ | |
2310 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
2311 | int q = netdev_get_prio_tc_map(dev, i); | |
2312 | ||
2313 | tc = &dev->tc_to_txq[q]; | |
2314 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
2315 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
2316 | i, q); | |
4f57c087 JF |
2317 | netdev_set_prio_tc_map(dev, i, 0); |
2318 | } | |
2319 | } | |
2320 | } | |
2321 | ||
8d059b0f AD |
2322 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
2323 | { | |
2324 | if (dev->num_tc) { | |
2325 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2326 | int i; | |
2327 | ||
ffcfe25b | 2328 | /* walk through the TCs and see if it falls into any of them */ |
8d059b0f AD |
2329 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { |
2330 | if ((txq - tc->offset) < tc->count) | |
2331 | return i; | |
2332 | } | |
2333 | ||
ffcfe25b | 2334 | /* didn't find it, just return -1 to indicate no match */ |
8d059b0f AD |
2335 | return -1; |
2336 | } | |
2337 | ||
2338 | return 0; | |
2339 | } | |
8a5f2166 | 2340 | EXPORT_SYMBOL(netdev_txq_to_tc); |
8d059b0f | 2341 | |
537c00de | 2342 | #ifdef CONFIG_XPS |
04157469 AN |
2343 | struct static_key xps_needed __read_mostly; |
2344 | EXPORT_SYMBOL(xps_needed); | |
2345 | struct static_key xps_rxqs_needed __read_mostly; | |
2346 | EXPORT_SYMBOL(xps_rxqs_needed); | |
537c00de AD |
2347 | static DEFINE_MUTEX(xps_map_mutex); |
2348 | #define xmap_dereference(P) \ | |
2349 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
2350 | ||
6234f874 AD |
2351 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
2352 | int tci, u16 index) | |
537c00de | 2353 | { |
10cdc3f3 AD |
2354 | struct xps_map *map = NULL; |
2355 | int pos; | |
537c00de | 2356 | |
10cdc3f3 | 2357 | if (dev_maps) |
80d19669 | 2358 | map = xmap_dereference(dev_maps->attr_map[tci]); |
6234f874 AD |
2359 | if (!map) |
2360 | return false; | |
537c00de | 2361 | |
6234f874 AD |
2362 | for (pos = map->len; pos--;) { |
2363 | if (map->queues[pos] != index) | |
2364 | continue; | |
2365 | ||
2366 | if (map->len > 1) { | |
2367 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2368 | break; |
537c00de | 2369 | } |
6234f874 | 2370 | |
80d19669 | 2371 | RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); |
6234f874 AD |
2372 | kfree_rcu(map, rcu); |
2373 | return false; | |
537c00de AD |
2374 | } |
2375 | ||
6234f874 | 2376 | return true; |
10cdc3f3 AD |
2377 | } |
2378 | ||
6234f874 AD |
2379 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2380 | struct xps_dev_maps *dev_maps, | |
2381 | int cpu, u16 offset, u16 count) | |
2382 | { | |
184c449f AD |
2383 | int num_tc = dev->num_tc ? : 1; |
2384 | bool active = false; | |
2385 | int tci; | |
6234f874 | 2386 | |
184c449f AD |
2387 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2388 | int i, j; | |
2389 | ||
2390 | for (i = count, j = offset; i--; j++) { | |
6358d49a | 2391 | if (!remove_xps_queue(dev_maps, tci, j)) |
184c449f AD |
2392 | break; |
2393 | } | |
2394 | ||
2395 | active |= i < 0; | |
6234f874 AD |
2396 | } |
2397 | ||
184c449f | 2398 | return active; |
6234f874 AD |
2399 | } |
2400 | ||
867d0ad4 SD |
2401 | static void reset_xps_maps(struct net_device *dev, |
2402 | struct xps_dev_maps *dev_maps, | |
2403 | bool is_rxqs_map) | |
2404 | { | |
2405 | if (is_rxqs_map) { | |
2406 | static_key_slow_dec_cpuslocked(&xps_rxqs_needed); | |
2407 | RCU_INIT_POINTER(dev->xps_rxqs_map, NULL); | |
2408 | } else { | |
2409 | RCU_INIT_POINTER(dev->xps_cpus_map, NULL); | |
2410 | } | |
2411 | static_key_slow_dec_cpuslocked(&xps_needed); | |
2412 | kfree_rcu(dev_maps, rcu); | |
2413 | } | |
2414 | ||
80d19669 AN |
2415 | static void clean_xps_maps(struct net_device *dev, const unsigned long *mask, |
2416 | struct xps_dev_maps *dev_maps, unsigned int nr_ids, | |
2417 | u16 offset, u16 count, bool is_rxqs_map) | |
2418 | { | |
2419 | bool active = false; | |
2420 | int i, j; | |
2421 | ||
2422 | for (j = -1; j = netif_attrmask_next(j, mask, nr_ids), | |
2423 | j < nr_ids;) | |
2424 | active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, | |
2425 | count); | |
867d0ad4 SD |
2426 | if (!active) |
2427 | reset_xps_maps(dev, dev_maps, is_rxqs_map); | |
80d19669 | 2428 | |
f28c020f SD |
2429 | if (!is_rxqs_map) { |
2430 | for (i = offset + (count - 1); count--; i--) { | |
2431 | netdev_queue_numa_node_write( | |
2432 | netdev_get_tx_queue(dev, i), | |
2433 | NUMA_NO_NODE); | |
80d19669 | 2434 | } |
80d19669 AN |
2435 | } |
2436 | } | |
2437 | ||
6234f874 AD |
2438 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, |
2439 | u16 count) | |
10cdc3f3 | 2440 | { |
80d19669 | 2441 | const unsigned long *possible_mask = NULL; |
10cdc3f3 | 2442 | struct xps_dev_maps *dev_maps; |
80d19669 | 2443 | unsigned int nr_ids; |
10cdc3f3 | 2444 | |
04157469 AN |
2445 | if (!static_key_false(&xps_needed)) |
2446 | return; | |
10cdc3f3 | 2447 | |
4d99f660 | 2448 | cpus_read_lock(); |
04157469 | 2449 | mutex_lock(&xps_map_mutex); |
10cdc3f3 | 2450 | |
04157469 AN |
2451 | if (static_key_false(&xps_rxqs_needed)) { |
2452 | dev_maps = xmap_dereference(dev->xps_rxqs_map); | |
2453 | if (dev_maps) { | |
2454 | nr_ids = dev->num_rx_queues; | |
2455 | clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, | |
2456 | offset, count, true); | |
2457 | } | |
537c00de AD |
2458 | } |
2459 | ||
80d19669 AN |
2460 | dev_maps = xmap_dereference(dev->xps_cpus_map); |
2461 | if (!dev_maps) | |
2462 | goto out_no_maps; | |
2463 | ||
2464 | if (num_possible_cpus() > 1) | |
2465 | possible_mask = cpumask_bits(cpu_possible_mask); | |
2466 | nr_ids = nr_cpu_ids; | |
2467 | clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count, | |
2468 | false); | |
024e9679 | 2469 | |
537c00de AD |
2470 | out_no_maps: |
2471 | mutex_unlock(&xps_map_mutex); | |
4d99f660 | 2472 | cpus_read_unlock(); |
537c00de AD |
2473 | } |
2474 | ||
6234f874 AD |
2475 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2476 | { | |
2477 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2478 | } | |
2479 | ||
80d19669 AN |
2480 | static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, |
2481 | u16 index, bool is_rxqs_map) | |
01c5f864 AD |
2482 | { |
2483 | struct xps_map *new_map; | |
2484 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2485 | int i, pos; | |
2486 | ||
2487 | for (pos = 0; map && pos < map->len; pos++) { | |
2488 | if (map->queues[pos] != index) | |
2489 | continue; | |
2490 | return map; | |
2491 | } | |
2492 | ||
80d19669 | 2493 | /* Need to add tx-queue to this CPU's/rx-queue's existing map */ |
01c5f864 AD |
2494 | if (map) { |
2495 | if (pos < map->alloc_len) | |
2496 | return map; | |
2497 | ||
2498 | alloc_len = map->alloc_len * 2; | |
2499 | } | |
2500 | ||
80d19669 AN |
2501 | /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's |
2502 | * map | |
2503 | */ | |
2504 | if (is_rxqs_map) | |
2505 | new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); | |
2506 | else | |
2507 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2508 | cpu_to_node(attr_index)); | |
01c5f864 AD |
2509 | if (!new_map) |
2510 | return NULL; | |
2511 | ||
2512 | for (i = 0; i < pos; i++) | |
2513 | new_map->queues[i] = map->queues[i]; | |
2514 | new_map->alloc_len = alloc_len; | |
2515 | new_map->len = pos; | |
2516 | ||
2517 | return new_map; | |
2518 | } | |
2519 | ||
4d99f660 | 2520 | /* Must be called under cpus_read_lock */ |
80d19669 AN |
2521 | int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, |
2522 | u16 index, bool is_rxqs_map) | |
537c00de | 2523 | { |
80d19669 | 2524 | const unsigned long *online_mask = NULL, *possible_mask = NULL; |
01c5f864 | 2525 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
80d19669 | 2526 | int i, j, tci, numa_node_id = -2; |
184c449f | 2527 | int maps_sz, num_tc = 1, tc = 0; |
537c00de | 2528 | struct xps_map *map, *new_map; |
01c5f864 | 2529 | bool active = false; |
80d19669 | 2530 | unsigned int nr_ids; |
537c00de | 2531 | |
184c449f | 2532 | if (dev->num_tc) { |
ffcfe25b | 2533 | /* Do not allow XPS on subordinate device directly */ |
184c449f | 2534 | num_tc = dev->num_tc; |
ffcfe25b AD |
2535 | if (num_tc < 0) |
2536 | return -EINVAL; | |
2537 | ||
2538 | /* If queue belongs to subordinate dev use its map */ | |
2539 | dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; | |
2540 | ||
184c449f AD |
2541 | tc = netdev_txq_to_tc(dev, index); |
2542 | if (tc < 0) | |
2543 | return -EINVAL; | |
2544 | } | |
2545 | ||
537c00de | 2546 | mutex_lock(&xps_map_mutex); |
80d19669 AN |
2547 | if (is_rxqs_map) { |
2548 | maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); | |
2549 | dev_maps = xmap_dereference(dev->xps_rxqs_map); | |
2550 | nr_ids = dev->num_rx_queues; | |
2551 | } else { | |
2552 | maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); | |
2553 | if (num_possible_cpus() > 1) { | |
2554 | online_mask = cpumask_bits(cpu_online_mask); | |
2555 | possible_mask = cpumask_bits(cpu_possible_mask); | |
2556 | } | |
2557 | dev_maps = xmap_dereference(dev->xps_cpus_map); | |
2558 | nr_ids = nr_cpu_ids; | |
2559 | } | |
537c00de | 2560 | |
80d19669 AN |
2561 | if (maps_sz < L1_CACHE_BYTES) |
2562 | maps_sz = L1_CACHE_BYTES; | |
537c00de | 2563 | |
01c5f864 | 2564 | /* allocate memory for queue storage */ |
80d19669 AN |
2565 | for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), |
2566 | j < nr_ids;) { | |
01c5f864 AD |
2567 | if (!new_dev_maps) |
2568 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2569 | if (!new_dev_maps) { |
2570 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2571 | return -ENOMEM; |
2bb60cb9 | 2572 | } |
01c5f864 | 2573 | |
80d19669 AN |
2574 | tci = j * num_tc + tc; |
2575 | map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) : | |
01c5f864 AD |
2576 | NULL; |
2577 | ||
80d19669 | 2578 | map = expand_xps_map(map, j, index, is_rxqs_map); |
01c5f864 AD |
2579 | if (!map) |
2580 | goto error; | |
2581 | ||
80d19669 | 2582 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); |
01c5f864 AD |
2583 | } |
2584 | ||
2585 | if (!new_dev_maps) | |
2586 | goto out_no_new_maps; | |
2587 | ||
867d0ad4 SD |
2588 | if (!dev_maps) { |
2589 | /* Increment static keys at most once per type */ | |
2590 | static_key_slow_inc_cpuslocked(&xps_needed); | |
2591 | if (is_rxqs_map) | |
2592 | static_key_slow_inc_cpuslocked(&xps_rxqs_needed); | |
2593 | } | |
04157469 | 2594 | |
80d19669 AN |
2595 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2596 | j < nr_ids;) { | |
184c449f | 2597 | /* copy maps belonging to foreign traffic classes */ |
80d19669 | 2598 | for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) { |
184c449f | 2599 | /* fill in the new device map from the old device map */ |
80d19669 AN |
2600 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2601 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
184c449f AD |
2602 | } |
2603 | ||
2604 | /* We need to explicitly update tci as prevous loop | |
2605 | * could break out early if dev_maps is NULL. | |
2606 | */ | |
80d19669 | 2607 | tci = j * num_tc + tc; |
184c449f | 2608 | |
80d19669 AN |
2609 | if (netif_attr_test_mask(j, mask, nr_ids) && |
2610 | netif_attr_test_online(j, online_mask, nr_ids)) { | |
2611 | /* add tx-queue to CPU/rx-queue maps */ | |
01c5f864 AD |
2612 | int pos = 0; |
2613 | ||
80d19669 | 2614 | map = xmap_dereference(new_dev_maps->attr_map[tci]); |
01c5f864 AD |
2615 | while ((pos < map->len) && (map->queues[pos] != index)) |
2616 | pos++; | |
2617 | ||
2618 | if (pos == map->len) | |
2619 | map->queues[map->len++] = index; | |
537c00de | 2620 | #ifdef CONFIG_NUMA |
80d19669 AN |
2621 | if (!is_rxqs_map) { |
2622 | if (numa_node_id == -2) | |
2623 | numa_node_id = cpu_to_node(j); | |
2624 | else if (numa_node_id != cpu_to_node(j)) | |
2625 | numa_node_id = -1; | |
2626 | } | |
537c00de | 2627 | #endif |
01c5f864 AD |
2628 | } else if (dev_maps) { |
2629 | /* fill in the new device map from the old device map */ | |
80d19669 AN |
2630 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2631 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
537c00de | 2632 | } |
01c5f864 | 2633 | |
184c449f AD |
2634 | /* copy maps belonging to foreign traffic classes */ |
2635 | for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { | |
2636 | /* fill in the new device map from the old device map */ | |
80d19669 AN |
2637 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2638 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
184c449f | 2639 | } |
537c00de AD |
2640 | } |
2641 | ||
80d19669 AN |
2642 | if (is_rxqs_map) |
2643 | rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps); | |
2644 | else | |
2645 | rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps); | |
01c5f864 | 2646 | |
537c00de | 2647 | /* Cleanup old maps */ |
184c449f AD |
2648 | if (!dev_maps) |
2649 | goto out_no_old_maps; | |
2650 | ||
80d19669 AN |
2651 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2652 | j < nr_ids;) { | |
2653 | for (i = num_tc, tci = j * num_tc; i--; tci++) { | |
2654 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
2655 | map = xmap_dereference(dev_maps->attr_map[tci]); | |
01c5f864 AD |
2656 | if (map && map != new_map) |
2657 | kfree_rcu(map, rcu); | |
2658 | } | |
537c00de AD |
2659 | } |
2660 | ||
184c449f AD |
2661 | kfree_rcu(dev_maps, rcu); |
2662 | ||
2663 | out_no_old_maps: | |
01c5f864 AD |
2664 | dev_maps = new_dev_maps; |
2665 | active = true; | |
537c00de | 2666 | |
01c5f864 | 2667 | out_no_new_maps: |
80d19669 AN |
2668 | if (!is_rxqs_map) { |
2669 | /* update Tx queue numa node */ | |
2670 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), | |
2671 | (numa_node_id >= 0) ? | |
2672 | numa_node_id : NUMA_NO_NODE); | |
2673 | } | |
537c00de | 2674 | |
01c5f864 AD |
2675 | if (!dev_maps) |
2676 | goto out_no_maps; | |
2677 | ||
80d19669 AN |
2678 | /* removes tx-queue from unused CPUs/rx-queues */ |
2679 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), | |
2680 | j < nr_ids;) { | |
2681 | for (i = tc, tci = j * num_tc; i--; tci++) | |
184c449f | 2682 | active |= remove_xps_queue(dev_maps, tci, index); |
80d19669 AN |
2683 | if (!netif_attr_test_mask(j, mask, nr_ids) || |
2684 | !netif_attr_test_online(j, online_mask, nr_ids)) | |
184c449f AD |
2685 | active |= remove_xps_queue(dev_maps, tci, index); |
2686 | for (i = num_tc - tc, tci++; --i; tci++) | |
2687 | active |= remove_xps_queue(dev_maps, tci, index); | |
01c5f864 AD |
2688 | } |
2689 | ||
2690 | /* free map if not active */ | |
867d0ad4 SD |
2691 | if (!active) |
2692 | reset_xps_maps(dev, dev_maps, is_rxqs_map); | |
01c5f864 AD |
2693 | |
2694 | out_no_maps: | |
537c00de AD |
2695 | mutex_unlock(&xps_map_mutex); |
2696 | ||
2697 | return 0; | |
2698 | error: | |
01c5f864 | 2699 | /* remove any maps that we added */ |
80d19669 AN |
2700 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2701 | j < nr_ids;) { | |
2702 | for (i = num_tc, tci = j * num_tc; i--; tci++) { | |
2703 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
184c449f | 2704 | map = dev_maps ? |
80d19669 | 2705 | xmap_dereference(dev_maps->attr_map[tci]) : |
184c449f AD |
2706 | NULL; |
2707 | if (new_map && new_map != map) | |
2708 | kfree(new_map); | |
2709 | } | |
01c5f864 AD |
2710 | } |
2711 | ||
537c00de AD |
2712 | mutex_unlock(&xps_map_mutex); |
2713 | ||
537c00de AD |
2714 | kfree(new_dev_maps); |
2715 | return -ENOMEM; | |
2716 | } | |
4d99f660 | 2717 | EXPORT_SYMBOL_GPL(__netif_set_xps_queue); |
80d19669 AN |
2718 | |
2719 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, | |
2720 | u16 index) | |
2721 | { | |
4d99f660 AV |
2722 | int ret; |
2723 | ||
2724 | cpus_read_lock(); | |
2725 | ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false); | |
2726 | cpus_read_unlock(); | |
2727 | ||
2728 | return ret; | |
80d19669 | 2729 | } |
537c00de AD |
2730 | EXPORT_SYMBOL(netif_set_xps_queue); |
2731 | ||
2732 | #endif | |
ffcfe25b AD |
2733 | static void netdev_unbind_all_sb_channels(struct net_device *dev) |
2734 | { | |
2735 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2736 | ||
2737 | /* Unbind any subordinate channels */ | |
2738 | while (txq-- != &dev->_tx[0]) { | |
2739 | if (txq->sb_dev) | |
2740 | netdev_unbind_sb_channel(dev, txq->sb_dev); | |
2741 | } | |
2742 | } | |
2743 | ||
9cf1f6a8 AD |
2744 | void netdev_reset_tc(struct net_device *dev) |
2745 | { | |
6234f874 AD |
2746 | #ifdef CONFIG_XPS |
2747 | netif_reset_xps_queues_gt(dev, 0); | |
2748 | #endif | |
ffcfe25b AD |
2749 | netdev_unbind_all_sb_channels(dev); |
2750 | ||
2751 | /* Reset TC configuration of device */ | |
9cf1f6a8 AD |
2752 | dev->num_tc = 0; |
2753 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2754 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2755 | } | |
2756 | EXPORT_SYMBOL(netdev_reset_tc); | |
2757 | ||
2758 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2759 | { | |
2760 | if (tc >= dev->num_tc) | |
2761 | return -EINVAL; | |
2762 | ||
6234f874 AD |
2763 | #ifdef CONFIG_XPS |
2764 | netif_reset_xps_queues(dev, offset, count); | |
2765 | #endif | |
9cf1f6a8 AD |
2766 | dev->tc_to_txq[tc].count = count; |
2767 | dev->tc_to_txq[tc].offset = offset; | |
2768 | return 0; | |
2769 | } | |
2770 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2771 | ||
2772 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2773 | { | |
2774 | if (num_tc > TC_MAX_QUEUE) | |
2775 | return -EINVAL; | |
2776 | ||
6234f874 AD |
2777 | #ifdef CONFIG_XPS |
2778 | netif_reset_xps_queues_gt(dev, 0); | |
2779 | #endif | |
ffcfe25b AD |
2780 | netdev_unbind_all_sb_channels(dev); |
2781 | ||
9cf1f6a8 AD |
2782 | dev->num_tc = num_tc; |
2783 | return 0; | |
2784 | } | |
2785 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2786 | ||
ffcfe25b AD |
2787 | void netdev_unbind_sb_channel(struct net_device *dev, |
2788 | struct net_device *sb_dev) | |
2789 | { | |
2790 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2791 | ||
2792 | #ifdef CONFIG_XPS | |
2793 | netif_reset_xps_queues_gt(sb_dev, 0); | |
2794 | #endif | |
2795 | memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); | |
2796 | memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); | |
2797 | ||
2798 | while (txq-- != &dev->_tx[0]) { | |
2799 | if (txq->sb_dev == sb_dev) | |
2800 | txq->sb_dev = NULL; | |
2801 | } | |
2802 | } | |
2803 | EXPORT_SYMBOL(netdev_unbind_sb_channel); | |
2804 | ||
2805 | int netdev_bind_sb_channel_queue(struct net_device *dev, | |
2806 | struct net_device *sb_dev, | |
2807 | u8 tc, u16 count, u16 offset) | |
2808 | { | |
2809 | /* Make certain the sb_dev and dev are already configured */ | |
2810 | if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) | |
2811 | return -EINVAL; | |
2812 | ||
2813 | /* We cannot hand out queues we don't have */ | |
2814 | if ((offset + count) > dev->real_num_tx_queues) | |
2815 | return -EINVAL; | |
2816 | ||
2817 | /* Record the mapping */ | |
2818 | sb_dev->tc_to_txq[tc].count = count; | |
2819 | sb_dev->tc_to_txq[tc].offset = offset; | |
2820 | ||
2821 | /* Provide a way for Tx queue to find the tc_to_txq map or | |
2822 | * XPS map for itself. | |
2823 | */ | |
2824 | while (count--) | |
2825 | netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; | |
2826 | ||
2827 | return 0; | |
2828 | } | |
2829 | EXPORT_SYMBOL(netdev_bind_sb_channel_queue); | |
2830 | ||
2831 | int netdev_set_sb_channel(struct net_device *dev, u16 channel) | |
2832 | { | |
2833 | /* Do not use a multiqueue device to represent a subordinate channel */ | |
2834 | if (netif_is_multiqueue(dev)) | |
2835 | return -ENODEV; | |
2836 | ||
2837 | /* We allow channels 1 - 32767 to be used for subordinate channels. | |
2838 | * Channel 0 is meant to be "native" mode and used only to represent | |
2839 | * the main root device. We allow writing 0 to reset the device back | |
2840 | * to normal mode after being used as a subordinate channel. | |
2841 | */ | |
2842 | if (channel > S16_MAX) | |
2843 | return -EINVAL; | |
2844 | ||
2845 | dev->num_tc = -channel; | |
2846 | ||
2847 | return 0; | |
2848 | } | |
2849 | EXPORT_SYMBOL(netdev_set_sb_channel); | |
2850 | ||
f0796d5c JF |
2851 | /* |
2852 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
3a053b1a | 2853 | * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. |
f0796d5c | 2854 | */ |
e6484930 | 2855 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2856 | { |
ac5b7019 | 2857 | bool disabling; |
1d24eb48 TH |
2858 | int rc; |
2859 | ||
ac5b7019 JK |
2860 | disabling = txq < dev->real_num_tx_queues; |
2861 | ||
e6484930 TH |
2862 | if (txq < 1 || txq > dev->num_tx_queues) |
2863 | return -EINVAL; | |
f0796d5c | 2864 | |
5c56580b BH |
2865 | if (dev->reg_state == NETREG_REGISTERED || |
2866 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2867 | ASSERT_RTNL(); |
2868 | ||
1d24eb48 TH |
2869 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2870 | txq); | |
bf264145 TH |
2871 | if (rc) |
2872 | return rc; | |
2873 | ||
4f57c087 JF |
2874 | if (dev->num_tc) |
2875 | netif_setup_tc(dev, txq); | |
2876 | ||
ac5b7019 JK |
2877 | dev->real_num_tx_queues = txq; |
2878 | ||
2879 | if (disabling) { | |
2880 | synchronize_net(); | |
e6484930 | 2881 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2882 | #ifdef CONFIG_XPS |
2883 | netif_reset_xps_queues_gt(dev, txq); | |
2884 | #endif | |
2885 | } | |
ac5b7019 JK |
2886 | } else { |
2887 | dev->real_num_tx_queues = txq; | |
f0796d5c | 2888 | } |
e6484930 | 2889 | |
e6484930 | 2890 | return 0; |
f0796d5c JF |
2891 | } |
2892 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2893 | |
a953be53 | 2894 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2895 | /** |
2896 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2897 | * @dev: Network device | |
2898 | * @rxq: Actual number of RX queues | |
2899 | * | |
2900 | * This must be called either with the rtnl_lock held or before | |
2901 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2902 | * negative error code. If called before registration, it always |
2903 | * succeeds. | |
62fe0b40 BH |
2904 | */ |
2905 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2906 | { | |
2907 | int rc; | |
2908 | ||
bd25fa7b TH |
2909 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2910 | return -EINVAL; | |
2911 | ||
62fe0b40 BH |
2912 | if (dev->reg_state == NETREG_REGISTERED) { |
2913 | ASSERT_RTNL(); | |
2914 | ||
62fe0b40 BH |
2915 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2916 | rxq); | |
2917 | if (rc) | |
2918 | return rc; | |
62fe0b40 BH |
2919 | } |
2920 | ||
2921 | dev->real_num_rx_queues = rxq; | |
2922 | return 0; | |
2923 | } | |
2924 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2925 | #endif | |
2926 | ||
2c53040f BH |
2927 | /** |
2928 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2929 | * |
2930 | * This routine should set an upper limit on the number of RSS queues | |
2931 | * used by default by multiqueue devices. | |
2932 | */ | |
a55b138b | 2933 | int netif_get_num_default_rss_queues(void) |
16917b87 | 2934 | { |
40e4e713 HS |
2935 | return is_kdump_kernel() ? |
2936 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
2937 | } |
2938 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2939 | ||
3bcb846c | 2940 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 2941 | { |
def82a1d JP |
2942 | struct softnet_data *sd; |
2943 | unsigned long flags; | |
56079431 | 2944 | |
def82a1d | 2945 | local_irq_save(flags); |
903ceff7 | 2946 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2947 | q->next_sched = NULL; |
2948 | *sd->output_queue_tailp = q; | |
2949 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2950 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2951 | local_irq_restore(flags); | |
2952 | } | |
2953 | ||
2954 | void __netif_schedule(struct Qdisc *q) | |
2955 | { | |
2956 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2957 | __netif_reschedule(q); | |
56079431 DV |
2958 | } |
2959 | EXPORT_SYMBOL(__netif_schedule); | |
2960 | ||
e6247027 ED |
2961 | struct dev_kfree_skb_cb { |
2962 | enum skb_free_reason reason; | |
2963 | }; | |
2964 | ||
2965 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2966 | { |
e6247027 ED |
2967 | return (struct dev_kfree_skb_cb *)skb->cb; |
2968 | } | |
2969 | ||
46e5da40 JF |
2970 | void netif_schedule_queue(struct netdev_queue *txq) |
2971 | { | |
2972 | rcu_read_lock(); | |
5be5515a | 2973 | if (!netif_xmit_stopped(txq)) { |
46e5da40 JF |
2974 | struct Qdisc *q = rcu_dereference(txq->qdisc); |
2975 | ||
2976 | __netif_schedule(q); | |
2977 | } | |
2978 | rcu_read_unlock(); | |
2979 | } | |
2980 | EXPORT_SYMBOL(netif_schedule_queue); | |
2981 | ||
46e5da40 JF |
2982 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
2983 | { | |
2984 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2985 | struct Qdisc *q; | |
2986 | ||
2987 | rcu_read_lock(); | |
2988 | q = rcu_dereference(dev_queue->qdisc); | |
2989 | __netif_schedule(q); | |
2990 | rcu_read_unlock(); | |
2991 | } | |
2992 | } | |
2993 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2994 | ||
e6247027 | 2995 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2996 | { |
e6247027 | 2997 | unsigned long flags; |
56079431 | 2998 | |
9899886d MJ |
2999 | if (unlikely(!skb)) |
3000 | return; | |
3001 | ||
63354797 | 3002 | if (likely(refcount_read(&skb->users) == 1)) { |
e6247027 | 3003 | smp_rmb(); |
63354797 RE |
3004 | refcount_set(&skb->users, 0); |
3005 | } else if (likely(!refcount_dec_and_test(&skb->users))) { | |
e6247027 | 3006 | return; |
bea3348e | 3007 | } |
e6247027 ED |
3008 | get_kfree_skb_cb(skb)->reason = reason; |
3009 | local_irq_save(flags); | |
3010 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
3011 | __this_cpu_write(softnet_data.completion_queue, skb); | |
3012 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
3013 | local_irq_restore(flags); | |
56079431 | 3014 | } |
e6247027 | 3015 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 3016 | |
e6247027 | 3017 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
3018 | { |
3019 | if (in_irq() || irqs_disabled()) | |
e6247027 | 3020 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
3021 | else |
3022 | dev_kfree_skb(skb); | |
3023 | } | |
e6247027 | 3024 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
3025 | |
3026 | ||
bea3348e SH |
3027 | /** |
3028 | * netif_device_detach - mark device as removed | |
3029 | * @dev: network device | |
3030 | * | |
3031 | * Mark device as removed from system and therefore no longer available. | |
3032 | */ | |
56079431 DV |
3033 | void netif_device_detach(struct net_device *dev) |
3034 | { | |
3035 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
3036 | netif_running(dev)) { | |
d543103a | 3037 | netif_tx_stop_all_queues(dev); |
56079431 DV |
3038 | } |
3039 | } | |
3040 | EXPORT_SYMBOL(netif_device_detach); | |
3041 | ||
bea3348e SH |
3042 | /** |
3043 | * netif_device_attach - mark device as attached | |
3044 | * @dev: network device | |
3045 | * | |
3046 | * Mark device as attached from system and restart if needed. | |
3047 | */ | |
56079431 DV |
3048 | void netif_device_attach(struct net_device *dev) |
3049 | { | |
3050 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
3051 | netif_running(dev)) { | |
d543103a | 3052 | netif_tx_wake_all_queues(dev); |
4ec93edb | 3053 | __netdev_watchdog_up(dev); |
56079431 DV |
3054 | } |
3055 | } | |
3056 | EXPORT_SYMBOL(netif_device_attach); | |
3057 | ||
5605c762 JP |
3058 | /* |
3059 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
3060 | * to be used as a distribution range. | |
3061 | */ | |
eadec877 AD |
3062 | static u16 skb_tx_hash(const struct net_device *dev, |
3063 | const struct net_device *sb_dev, | |
3064 | struct sk_buff *skb) | |
5605c762 JP |
3065 | { |
3066 | u32 hash; | |
3067 | u16 qoffset = 0; | |
1b837d48 | 3068 | u16 qcount = dev->real_num_tx_queues; |
5605c762 | 3069 | |
eadec877 AD |
3070 | if (dev->num_tc) { |
3071 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
3072 | ||
3073 | qoffset = sb_dev->tc_to_txq[tc].offset; | |
3074 | qcount = sb_dev->tc_to_txq[tc].count; | |
3075 | } | |
3076 | ||
5605c762 JP |
3077 | if (skb_rx_queue_recorded(skb)) { |
3078 | hash = skb_get_rx_queue(skb); | |
6e11d157 AN |
3079 | if (hash >= qoffset) |
3080 | hash -= qoffset; | |
1b837d48 AD |
3081 | while (unlikely(hash >= qcount)) |
3082 | hash -= qcount; | |
eadec877 | 3083 | return hash + qoffset; |
5605c762 JP |
3084 | } |
3085 | ||
3086 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
3087 | } | |
5605c762 | 3088 | |
36c92474 BH |
3089 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
3090 | { | |
84d15ae5 | 3091 | static const netdev_features_t null_features; |
36c92474 | 3092 | struct net_device *dev = skb->dev; |
88ad4175 | 3093 | const char *name = ""; |
36c92474 | 3094 | |
c846ad9b BG |
3095 | if (!net_ratelimit()) |
3096 | return; | |
3097 | ||
88ad4175 BM |
3098 | if (dev) { |
3099 | if (dev->dev.parent) | |
3100 | name = dev_driver_string(dev->dev.parent); | |
3101 | else | |
3102 | name = netdev_name(dev); | |
3103 | } | |
6413139d WB |
3104 | skb_dump(KERN_WARNING, skb, false); |
3105 | WARN(1, "%s: caps=(%pNF, %pNF)\n", | |
88ad4175 | 3106 | name, dev ? &dev->features : &null_features, |
6413139d | 3107 | skb->sk ? &skb->sk->sk_route_caps : &null_features); |
36c92474 BH |
3108 | } |
3109 | ||
1da177e4 LT |
3110 | /* |
3111 | * Invalidate hardware checksum when packet is to be mangled, and | |
3112 | * complete checksum manually on outgoing path. | |
3113 | */ | |
84fa7933 | 3114 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 3115 | { |
d3bc23e7 | 3116 | __wsum csum; |
663ead3b | 3117 | int ret = 0, offset; |
1da177e4 | 3118 | |
84fa7933 | 3119 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
3120 | goto out_set_summed; |
3121 | ||
3122 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
3123 | skb_warn_bad_offload(skb); |
3124 | return -EINVAL; | |
1da177e4 LT |
3125 | } |
3126 | ||
cef401de ED |
3127 | /* Before computing a checksum, we should make sure no frag could |
3128 | * be modified by an external entity : checksum could be wrong. | |
3129 | */ | |
3130 | if (skb_has_shared_frag(skb)) { | |
3131 | ret = __skb_linearize(skb); | |
3132 | if (ret) | |
3133 | goto out; | |
3134 | } | |
3135 | ||
55508d60 | 3136 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
3137 | BUG_ON(offset >= skb_headlen(skb)); |
3138 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
3139 | ||
3140 | offset += skb->csum_offset; | |
3141 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
3142 | ||
8211fbfa HK |
3143 | ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); |
3144 | if (ret) | |
3145 | goto out; | |
1da177e4 | 3146 | |
4f2e4ad5 | 3147 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 3148 | out_set_summed: |
1da177e4 | 3149 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 3150 | out: |
1da177e4 LT |
3151 | return ret; |
3152 | } | |
d1b19dff | 3153 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 3154 | |
b72b5bf6 DC |
3155 | int skb_crc32c_csum_help(struct sk_buff *skb) |
3156 | { | |
3157 | __le32 crc32c_csum; | |
3158 | int ret = 0, offset, start; | |
3159 | ||
3160 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
3161 | goto out; | |
3162 | ||
3163 | if (unlikely(skb_is_gso(skb))) | |
3164 | goto out; | |
3165 | ||
3166 | /* Before computing a checksum, we should make sure no frag could | |
3167 | * be modified by an external entity : checksum could be wrong. | |
3168 | */ | |
3169 | if (unlikely(skb_has_shared_frag(skb))) { | |
3170 | ret = __skb_linearize(skb); | |
3171 | if (ret) | |
3172 | goto out; | |
3173 | } | |
3174 | start = skb_checksum_start_offset(skb); | |
3175 | offset = start + offsetof(struct sctphdr, checksum); | |
3176 | if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { | |
3177 | ret = -EINVAL; | |
3178 | goto out; | |
3179 | } | |
8211fbfa HK |
3180 | |
3181 | ret = skb_ensure_writable(skb, offset + sizeof(__le32)); | |
3182 | if (ret) | |
3183 | goto out; | |
3184 | ||
b72b5bf6 DC |
3185 | crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, |
3186 | skb->len - start, ~(__u32)0, | |
3187 | crc32c_csum_stub)); | |
3188 | *(__le32 *)(skb->data + offset) = crc32c_csum; | |
3189 | skb->ip_summed = CHECKSUM_NONE; | |
dba00306 | 3190 | skb->csum_not_inet = 0; |
b72b5bf6 DC |
3191 | out: |
3192 | return ret; | |
3193 | } | |
3194 | ||
53d6471c | 3195 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 3196 | { |
252e3346 | 3197 | __be16 type = skb->protocol; |
f6a78bfc | 3198 | |
19acc327 PS |
3199 | /* Tunnel gso handlers can set protocol to ethernet. */ |
3200 | if (type == htons(ETH_P_TEB)) { | |
3201 | struct ethhdr *eth; | |
3202 | ||
3203 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
3204 | return 0; | |
3205 | ||
1dfe82eb | 3206 | eth = (struct ethhdr *)skb->data; |
19acc327 PS |
3207 | type = eth->h_proto; |
3208 | } | |
3209 | ||
d4bcef3f | 3210 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
3211 | } |
3212 | ||
3213 | /** | |
3214 | * skb_mac_gso_segment - mac layer segmentation handler. | |
3215 | * @skb: buffer to segment | |
3216 | * @features: features for the output path (see dev->features) | |
3217 | */ | |
3218 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
3219 | netdev_features_t features) | |
3220 | { | |
3221 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
3222 | struct packet_offload *ptype; | |
53d6471c VY |
3223 | int vlan_depth = skb->mac_len; |
3224 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
3225 | |
3226 | if (unlikely(!type)) | |
3227 | return ERR_PTR(-EINVAL); | |
3228 | ||
53d6471c | 3229 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
3230 | |
3231 | rcu_read_lock(); | |
22061d80 | 3232 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 3233 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 3234 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
3235 | break; |
3236 | } | |
3237 | } | |
3238 | rcu_read_unlock(); | |
3239 | ||
98e399f8 | 3240 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 3241 | |
f6a78bfc HX |
3242 | return segs; |
3243 | } | |
05e8ef4a PS |
3244 | EXPORT_SYMBOL(skb_mac_gso_segment); |
3245 | ||
3246 | ||
3247 | /* openvswitch calls this on rx path, so we need a different check. | |
3248 | */ | |
3249 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
3250 | { | |
3251 | if (tx_path) | |
0c19f846 WB |
3252 | return skb->ip_summed != CHECKSUM_PARTIAL && |
3253 | skb->ip_summed != CHECKSUM_UNNECESSARY; | |
6e7bc478 ED |
3254 | |
3255 | return skb->ip_summed == CHECKSUM_NONE; | |
05e8ef4a PS |
3256 | } |
3257 | ||
3258 | /** | |
3259 | * __skb_gso_segment - Perform segmentation on skb. | |
3260 | * @skb: buffer to segment | |
3261 | * @features: features for the output path (see dev->features) | |
3262 | * @tx_path: whether it is called in TX path | |
3263 | * | |
3264 | * This function segments the given skb and returns a list of segments. | |
3265 | * | |
3266 | * It may return NULL if the skb requires no segmentation. This is | |
3267 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 | 3268 | * |
a08e7fd9 | 3269 | * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. |
05e8ef4a PS |
3270 | */ |
3271 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
3272 | netdev_features_t features, bool tx_path) | |
3273 | { | |
b2504a5d ED |
3274 | struct sk_buff *segs; |
3275 | ||
05e8ef4a PS |
3276 | if (unlikely(skb_needs_check(skb, tx_path))) { |
3277 | int err; | |
3278 | ||
b2504a5d | 3279 | /* We're going to init ->check field in TCP or UDP header */ |
a40e0a66 | 3280 | err = skb_cow_head(skb, 0); |
3281 | if (err < 0) | |
05e8ef4a PS |
3282 | return ERR_PTR(err); |
3283 | } | |
3284 | ||
802ab55a AD |
3285 | /* Only report GSO partial support if it will enable us to |
3286 | * support segmentation on this frame without needing additional | |
3287 | * work. | |
3288 | */ | |
3289 | if (features & NETIF_F_GSO_PARTIAL) { | |
3290 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
3291 | struct net_device *dev = skb->dev; | |
3292 | ||
3293 | partial_features |= dev->features & dev->gso_partial_features; | |
3294 | if (!skb_gso_ok(skb, features | partial_features)) | |
3295 | features &= ~NETIF_F_GSO_PARTIAL; | |
3296 | } | |
3297 | ||
a08e7fd9 | 3298 | BUILD_BUG_ON(SKB_GSO_CB_OFFSET + |
9207f9d4 KK |
3299 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); |
3300 | ||
68c33163 | 3301 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
3302 | SKB_GSO_CB(skb)->encap_level = 0; |
3303 | ||
05e8ef4a PS |
3304 | skb_reset_mac_header(skb); |
3305 | skb_reset_mac_len(skb); | |
3306 | ||
b2504a5d ED |
3307 | segs = skb_mac_gso_segment(skb, features); |
3308 | ||
3a1296a3 | 3309 | if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) |
b2504a5d ED |
3310 | skb_warn_bad_offload(skb); |
3311 | ||
3312 | return segs; | |
05e8ef4a | 3313 | } |
12b0004d | 3314 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 3315 | |
fb286bb2 HX |
3316 | /* Take action when hardware reception checksum errors are detected. */ |
3317 | #ifdef CONFIG_BUG | |
7fe50ac8 | 3318 | void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) |
fb286bb2 HX |
3319 | { |
3320 | if (net_ratelimit()) { | |
7b6cd1ce | 3321 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
6413139d | 3322 | skb_dump(KERN_ERR, skb, true); |
fb286bb2 HX |
3323 | dump_stack(); |
3324 | } | |
3325 | } | |
3326 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
3327 | #endif | |
3328 | ||
ab74cfeb | 3329 | /* XXX: check that highmem exists at all on the given machine. */ |
c1e756bf | 3330 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 3331 | { |
3d3a8533 | 3332 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 3333 | int i; |
f4563a75 | 3334 | |
5acbbd42 | 3335 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
3336 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
3337 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 3338 | |
ea2ab693 | 3339 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 3340 | return 1; |
ea2ab693 | 3341 | } |
5acbbd42 | 3342 | } |
3d3a8533 | 3343 | #endif |
1da177e4 LT |
3344 | return 0; |
3345 | } | |
1da177e4 | 3346 | |
3b392ddb SH |
3347 | /* If MPLS offload request, verify we are testing hardware MPLS features |
3348 | * instead of standard features for the netdev. | |
3349 | */ | |
d0edc7bf | 3350 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
3351 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
3352 | netdev_features_t features, | |
3353 | __be16 type) | |
3354 | { | |
25cd9ba0 | 3355 | if (eth_p_mpls(type)) |
3b392ddb SH |
3356 | features &= skb->dev->mpls_features; |
3357 | ||
3358 | return features; | |
3359 | } | |
3360 | #else | |
3361 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
3362 | netdev_features_t features, | |
3363 | __be16 type) | |
3364 | { | |
3365 | return features; | |
3366 | } | |
3367 | #endif | |
3368 | ||
c8f44aff | 3369 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 3370 | netdev_features_t features) |
f01a5236 | 3371 | { |
53d6471c | 3372 | int tmp; |
3b392ddb SH |
3373 | __be16 type; |
3374 | ||
3375 | type = skb_network_protocol(skb, &tmp); | |
3376 | features = net_mpls_features(skb, features, type); | |
53d6471c | 3377 | |
c0d680e5 | 3378 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 3379 | !can_checksum_protocol(features, type)) { |
996e8021 | 3380 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 3381 | } |
7be2c82c ED |
3382 | if (illegal_highdma(skb->dev, skb)) |
3383 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
3384 | |
3385 | return features; | |
3386 | } | |
3387 | ||
e38f3025 TM |
3388 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
3389 | struct net_device *dev, | |
3390 | netdev_features_t features) | |
3391 | { | |
3392 | return features; | |
3393 | } | |
3394 | EXPORT_SYMBOL(passthru_features_check); | |
3395 | ||
7ce23672 | 3396 | static netdev_features_t dflt_features_check(struct sk_buff *skb, |
8cb65d00 TM |
3397 | struct net_device *dev, |
3398 | netdev_features_t features) | |
3399 | { | |
3400 | return vlan_features_check(skb, features); | |
3401 | } | |
3402 | ||
cbc53e08 AD |
3403 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
3404 | struct net_device *dev, | |
3405 | netdev_features_t features) | |
3406 | { | |
3407 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
3408 | ||
3409 | if (gso_segs > dev->gso_max_segs) | |
3410 | return features & ~NETIF_F_GSO_MASK; | |
3411 | ||
802ab55a AD |
3412 | /* Support for GSO partial features requires software |
3413 | * intervention before we can actually process the packets | |
3414 | * so we need to strip support for any partial features now | |
3415 | * and we can pull them back in after we have partially | |
3416 | * segmented the frame. | |
3417 | */ | |
3418 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
3419 | features &= ~dev->gso_partial_features; | |
3420 | ||
3421 | /* Make sure to clear the IPv4 ID mangling feature if the | |
3422 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
3423 | */ |
3424 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
3425 | struct iphdr *iph = skb->encapsulation ? | |
3426 | inner_ip_hdr(skb) : ip_hdr(skb); | |
3427 | ||
3428 | if (!(iph->frag_off & htons(IP_DF))) | |
3429 | features &= ~NETIF_F_TSO_MANGLEID; | |
3430 | } | |
3431 | ||
3432 | return features; | |
3433 | } | |
3434 | ||
c1e756bf | 3435 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 3436 | { |
5f35227e | 3437 | struct net_device *dev = skb->dev; |
fcbeb976 | 3438 | netdev_features_t features = dev->features; |
58e998c6 | 3439 | |
cbc53e08 AD |
3440 | if (skb_is_gso(skb)) |
3441 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 3442 | |
5f35227e JG |
3443 | /* If encapsulation offload request, verify we are testing |
3444 | * hardware encapsulation features instead of standard | |
3445 | * features for the netdev | |
3446 | */ | |
3447 | if (skb->encapsulation) | |
3448 | features &= dev->hw_enc_features; | |
3449 | ||
f5a7fb88 TM |
3450 | if (skb_vlan_tagged(skb)) |
3451 | features = netdev_intersect_features(features, | |
3452 | dev->vlan_features | | |
3453 | NETIF_F_HW_VLAN_CTAG_TX | | |
3454 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 3455 | |
5f35227e JG |
3456 | if (dev->netdev_ops->ndo_features_check) |
3457 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
3458 | features); | |
8cb65d00 TM |
3459 | else |
3460 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 3461 | |
c1e756bf | 3462 | return harmonize_features(skb, features); |
58e998c6 | 3463 | } |
c1e756bf | 3464 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 3465 | |
2ea25513 | 3466 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 3467 | struct netdev_queue *txq, bool more) |
f6a78bfc | 3468 | { |
2ea25513 DM |
3469 | unsigned int len; |
3470 | int rc; | |
00829823 | 3471 | |
9f9a742d | 3472 | if (dev_nit_active(dev)) |
2ea25513 | 3473 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 3474 | |
2ea25513 DM |
3475 | len = skb->len; |
3476 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 3477 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 3478 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 3479 | |
2ea25513 DM |
3480 | return rc; |
3481 | } | |
7b9c6090 | 3482 | |
8dcda22a DM |
3483 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
3484 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
3485 | { |
3486 | struct sk_buff *skb = first; | |
3487 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 3488 | |
7f2e870f DM |
3489 | while (skb) { |
3490 | struct sk_buff *next = skb->next; | |
fc70fb64 | 3491 | |
a8305bff | 3492 | skb_mark_not_on_list(skb); |
95f6b3dd | 3493 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
3494 | if (unlikely(!dev_xmit_complete(rc))) { |
3495 | skb->next = next; | |
3496 | goto out; | |
3497 | } | |
6afff0ca | 3498 | |
7f2e870f | 3499 | skb = next; |
fe60faa5 | 3500 | if (netif_tx_queue_stopped(txq) && skb) { |
7f2e870f DM |
3501 | rc = NETDEV_TX_BUSY; |
3502 | break; | |
9ccb8975 | 3503 | } |
7f2e870f | 3504 | } |
9ccb8975 | 3505 | |
7f2e870f DM |
3506 | out: |
3507 | *ret = rc; | |
3508 | return skb; | |
3509 | } | |
b40863c6 | 3510 | |
1ff0dc94 ED |
3511 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
3512 | netdev_features_t features) | |
f6a78bfc | 3513 | { |
df8a39de | 3514 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
3515 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
3516 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
3517 | return skb; |
3518 | } | |
f6a78bfc | 3519 | |
43c26a1a DC |
3520 | int skb_csum_hwoffload_help(struct sk_buff *skb, |
3521 | const netdev_features_t features) | |
3522 | { | |
3523 | if (unlikely(skb->csum_not_inet)) | |
3524 | return !!(features & NETIF_F_SCTP_CRC) ? 0 : | |
3525 | skb_crc32c_csum_help(skb); | |
3526 | ||
3527 | return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb); | |
3528 | } | |
3529 | EXPORT_SYMBOL(skb_csum_hwoffload_help); | |
3530 | ||
f53c7239 | 3531 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) |
eae3f88e DM |
3532 | { |
3533 | netdev_features_t features; | |
f6a78bfc | 3534 | |
eae3f88e DM |
3535 | features = netif_skb_features(skb); |
3536 | skb = validate_xmit_vlan(skb, features); | |
3537 | if (unlikely(!skb)) | |
3538 | goto out_null; | |
7b9c6090 | 3539 | |
ebf4e808 IL |
3540 | skb = sk_validate_xmit_skb(skb, dev); |
3541 | if (unlikely(!skb)) | |
3542 | goto out_null; | |
3543 | ||
8b86a61d | 3544 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
3545 | struct sk_buff *segs; |
3546 | ||
3547 | segs = skb_gso_segment(skb, features); | |
cecda693 | 3548 | if (IS_ERR(segs)) { |
af6dabc9 | 3549 | goto out_kfree_skb; |
cecda693 JW |
3550 | } else if (segs) { |
3551 | consume_skb(skb); | |
3552 | skb = segs; | |
f6a78bfc | 3553 | } |
eae3f88e DM |
3554 | } else { |
3555 | if (skb_needs_linearize(skb, features) && | |
3556 | __skb_linearize(skb)) | |
3557 | goto out_kfree_skb; | |
4ec93edb | 3558 | |
eae3f88e DM |
3559 | /* If packet is not checksummed and device does not |
3560 | * support checksumming for this protocol, complete | |
3561 | * checksumming here. | |
3562 | */ | |
3563 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
3564 | if (skb->encapsulation) | |
3565 | skb_set_inner_transport_header(skb, | |
3566 | skb_checksum_start_offset(skb)); | |
3567 | else | |
3568 | skb_set_transport_header(skb, | |
3569 | skb_checksum_start_offset(skb)); | |
43c26a1a | 3570 | if (skb_csum_hwoffload_help(skb, features)) |
eae3f88e | 3571 | goto out_kfree_skb; |
7b9c6090 | 3572 | } |
0c772159 | 3573 | } |
7b9c6090 | 3574 | |
f53c7239 | 3575 | skb = validate_xmit_xfrm(skb, features, again); |
3dca3f38 | 3576 | |
eae3f88e | 3577 | return skb; |
fc70fb64 | 3578 | |
f6a78bfc HX |
3579 | out_kfree_skb: |
3580 | kfree_skb(skb); | |
eae3f88e | 3581 | out_null: |
d21fd63e | 3582 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3583 | return NULL; |
3584 | } | |
6afff0ca | 3585 | |
f53c7239 | 3586 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) |
55a93b3e ED |
3587 | { |
3588 | struct sk_buff *next, *head = NULL, *tail; | |
3589 | ||
bec3cfdc | 3590 | for (; skb != NULL; skb = next) { |
55a93b3e | 3591 | next = skb->next; |
a8305bff | 3592 | skb_mark_not_on_list(skb); |
bec3cfdc ED |
3593 | |
3594 | /* in case skb wont be segmented, point to itself */ | |
3595 | skb->prev = skb; | |
3596 | ||
f53c7239 | 3597 | skb = validate_xmit_skb(skb, dev, again); |
bec3cfdc ED |
3598 | if (!skb) |
3599 | continue; | |
55a93b3e | 3600 | |
bec3cfdc ED |
3601 | if (!head) |
3602 | head = skb; | |
3603 | else | |
3604 | tail->next = skb; | |
3605 | /* If skb was segmented, skb->prev points to | |
3606 | * the last segment. If not, it still contains skb. | |
3607 | */ | |
3608 | tail = skb->prev; | |
55a93b3e ED |
3609 | } |
3610 | return head; | |
f6a78bfc | 3611 | } |
104ba78c | 3612 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3613 | |
1def9238 ED |
3614 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3615 | { | |
3616 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3617 | ||
3618 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3619 | ||
3620 | /* To get more precise estimation of bytes sent on wire, | |
3621 | * we add to pkt_len the headers size of all segments | |
3622 | */ | |
a0dce875 | 3623 | if (shinfo->gso_size && skb_transport_header_was_set(skb)) { |
757b8b1d | 3624 | unsigned int hdr_len; |
15e5a030 | 3625 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3626 | |
757b8b1d ED |
3627 | /* mac layer + network layer */ |
3628 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3629 | ||
3630 | /* + transport layer */ | |
7c68d1a6 ED |
3631 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { |
3632 | const struct tcphdr *th; | |
3633 | struct tcphdr _tcphdr; | |
3634 | ||
3635 | th = skb_header_pointer(skb, skb_transport_offset(skb), | |
3636 | sizeof(_tcphdr), &_tcphdr); | |
3637 | if (likely(th)) | |
3638 | hdr_len += __tcp_hdrlen(th); | |
3639 | } else { | |
3640 | struct udphdr _udphdr; | |
3641 | ||
3642 | if (skb_header_pointer(skb, skb_transport_offset(skb), | |
3643 | sizeof(_udphdr), &_udphdr)) | |
3644 | hdr_len += sizeof(struct udphdr); | |
3645 | } | |
15e5a030 JW |
3646 | |
3647 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3648 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3649 | shinfo->gso_size); | |
3650 | ||
3651 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3652 | } |
3653 | } | |
3654 | ||
bbd8a0d3 KK |
3655 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3656 | struct net_device *dev, | |
3657 | struct netdev_queue *txq) | |
3658 | { | |
3659 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3660 | struct sk_buff *to_free = NULL; |
a2da570d | 3661 | bool contended; |
bbd8a0d3 KK |
3662 | int rc; |
3663 | ||
a2da570d | 3664 | qdisc_calculate_pkt_len(skb, q); |
6b3ba914 JF |
3665 | |
3666 | if (q->flags & TCQ_F_NOLOCK) { | |
379349e9 PA |
3667 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
3668 | qdisc_run(q); | |
6b3ba914 JF |
3669 | |
3670 | if (unlikely(to_free)) | |
3671 | kfree_skb_list(to_free); | |
3672 | return rc; | |
3673 | } | |
3674 | ||
79640a4c ED |
3675 | /* |
3676 | * Heuristic to force contended enqueues to serialize on a | |
3677 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3678 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3679 | * often and dequeue packets faster. |
79640a4c | 3680 | */ |
a2da570d | 3681 | contended = qdisc_is_running(q); |
79640a4c ED |
3682 | if (unlikely(contended)) |
3683 | spin_lock(&q->busylock); | |
3684 | ||
bbd8a0d3 KK |
3685 | spin_lock(root_lock); |
3686 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3687 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3688 | rc = NET_XMIT_DROP; |
3689 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3690 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3691 | /* |
3692 | * This is a work-conserving queue; there are no old skbs | |
3693 | * waiting to be sent out; and the qdisc is not running - | |
3694 | * xmit the skb directly. | |
3695 | */ | |
bfe0d029 | 3696 | |
bfe0d029 ED |
3697 | qdisc_bstats_update(q, skb); |
3698 | ||
55a93b3e | 3699 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3700 | if (unlikely(contended)) { |
3701 | spin_unlock(&q->busylock); | |
3702 | contended = false; | |
3703 | } | |
bbd8a0d3 | 3704 | __qdisc_run(q); |
6c148184 | 3705 | } |
bbd8a0d3 | 3706 | |
6c148184 | 3707 | qdisc_run_end(q); |
bbd8a0d3 KK |
3708 | rc = NET_XMIT_SUCCESS; |
3709 | } else { | |
520ac30f | 3710 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3711 | if (qdisc_run_begin(q)) { |
3712 | if (unlikely(contended)) { | |
3713 | spin_unlock(&q->busylock); | |
3714 | contended = false; | |
3715 | } | |
3716 | __qdisc_run(q); | |
6c148184 | 3717 | qdisc_run_end(q); |
79640a4c | 3718 | } |
bbd8a0d3 KK |
3719 | } |
3720 | spin_unlock(root_lock); | |
520ac30f ED |
3721 | if (unlikely(to_free)) |
3722 | kfree_skb_list(to_free); | |
79640a4c ED |
3723 | if (unlikely(contended)) |
3724 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3725 | return rc; |
3726 | } | |
3727 | ||
86f8515f | 3728 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3729 | static void skb_update_prio(struct sk_buff *skb) |
3730 | { | |
4dcb31d4 ED |
3731 | const struct netprio_map *map; |
3732 | const struct sock *sk; | |
3733 | unsigned int prioidx; | |
5bc1421e | 3734 | |
4dcb31d4 ED |
3735 | if (skb->priority) |
3736 | return; | |
3737 | map = rcu_dereference_bh(skb->dev->priomap); | |
3738 | if (!map) | |
3739 | return; | |
3740 | sk = skb_to_full_sk(skb); | |
3741 | if (!sk) | |
3742 | return; | |
91c68ce2 | 3743 | |
4dcb31d4 ED |
3744 | prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); |
3745 | ||
3746 | if (prioidx < map->priomap_len) | |
3747 | skb->priority = map->priomap[prioidx]; | |
5bc1421e NH |
3748 | } |
3749 | #else | |
3750 | #define skb_update_prio(skb) | |
3751 | #endif | |
3752 | ||
95603e22 MM |
3753 | /** |
3754 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3755 | * @net: network namespace this loopback is happening in |
3756 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3757 | * @skb: buffer to transmit |
3758 | */ | |
0c4b51f0 | 3759 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3760 | { |
3761 | skb_reset_mac_header(skb); | |
3762 | __skb_pull(skb, skb_network_offset(skb)); | |
3763 | skb->pkt_type = PACKET_LOOPBACK; | |
3764 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3765 | WARN_ON(!skb_dst(skb)); | |
3766 | skb_dst_force(skb); | |
3767 | netif_rx_ni(skb); | |
3768 | return 0; | |
3769 | } | |
3770 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3771 | ||
1f211a1b DB |
3772 | #ifdef CONFIG_NET_EGRESS |
3773 | static struct sk_buff * | |
3774 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3775 | { | |
46209401 | 3776 | struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); |
1f211a1b DB |
3777 | struct tcf_result cl_res; |
3778 | ||
46209401 | 3779 | if (!miniq) |
1f211a1b DB |
3780 | return skb; |
3781 | ||
8dc07fdb | 3782 | /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ |
46209401 | 3783 | mini_qdisc_bstats_cpu_update(miniq, skb); |
1f211a1b | 3784 | |
46209401 | 3785 | switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { |
1f211a1b DB |
3786 | case TC_ACT_OK: |
3787 | case TC_ACT_RECLASSIFY: | |
3788 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3789 | break; | |
3790 | case TC_ACT_SHOT: | |
46209401 | 3791 | mini_qdisc_qstats_cpu_drop(miniq); |
1f211a1b | 3792 | *ret = NET_XMIT_DROP; |
7e2c3aea DB |
3793 | kfree_skb(skb); |
3794 | return NULL; | |
1f211a1b DB |
3795 | case TC_ACT_STOLEN: |
3796 | case TC_ACT_QUEUED: | |
e25ea21f | 3797 | case TC_ACT_TRAP: |
1f211a1b | 3798 | *ret = NET_XMIT_SUCCESS; |
7e2c3aea | 3799 | consume_skb(skb); |
1f211a1b DB |
3800 | return NULL; |
3801 | case TC_ACT_REDIRECT: | |
3802 | /* No need to push/pop skb's mac_header here on egress! */ | |
3803 | skb_do_redirect(skb); | |
3804 | *ret = NET_XMIT_SUCCESS; | |
3805 | return NULL; | |
3806 | default: | |
3807 | break; | |
3808 | } | |
357b6cc5 | 3809 | |
1f211a1b DB |
3810 | return skb; |
3811 | } | |
3812 | #endif /* CONFIG_NET_EGRESS */ | |
3813 | ||
fc9bab24 AN |
3814 | #ifdef CONFIG_XPS |
3815 | static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, | |
3816 | struct xps_dev_maps *dev_maps, unsigned int tci) | |
3817 | { | |
3818 | struct xps_map *map; | |
3819 | int queue_index = -1; | |
3820 | ||
3821 | if (dev->num_tc) { | |
3822 | tci *= dev->num_tc; | |
3823 | tci += netdev_get_prio_tc_map(dev, skb->priority); | |
3824 | } | |
3825 | ||
3826 | map = rcu_dereference(dev_maps->attr_map[tci]); | |
3827 | if (map) { | |
3828 | if (map->len == 1) | |
3829 | queue_index = map->queues[0]; | |
3830 | else | |
3831 | queue_index = map->queues[reciprocal_scale( | |
3832 | skb_get_hash(skb), map->len)]; | |
3833 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3834 | queue_index = -1; | |
3835 | } | |
3836 | return queue_index; | |
3837 | } | |
3838 | #endif | |
3839 | ||
eadec877 AD |
3840 | static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, |
3841 | struct sk_buff *skb) | |
638b2a69 JP |
3842 | { |
3843 | #ifdef CONFIG_XPS | |
3844 | struct xps_dev_maps *dev_maps; | |
fc9bab24 | 3845 | struct sock *sk = skb->sk; |
638b2a69 JP |
3846 | int queue_index = -1; |
3847 | ||
04157469 AN |
3848 | if (!static_key_false(&xps_needed)) |
3849 | return -1; | |
3850 | ||
638b2a69 | 3851 | rcu_read_lock(); |
fc9bab24 AN |
3852 | if (!static_key_false(&xps_rxqs_needed)) |
3853 | goto get_cpus_map; | |
3854 | ||
eadec877 | 3855 | dev_maps = rcu_dereference(sb_dev->xps_rxqs_map); |
638b2a69 | 3856 | if (dev_maps) { |
fc9bab24 | 3857 | int tci = sk_rx_queue_get(sk); |
184c449f | 3858 | |
fc9bab24 AN |
3859 | if (tci >= 0 && tci < dev->num_rx_queues) |
3860 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
3861 | tci); | |
3862 | } | |
184c449f | 3863 | |
fc9bab24 AN |
3864 | get_cpus_map: |
3865 | if (queue_index < 0) { | |
eadec877 | 3866 | dev_maps = rcu_dereference(sb_dev->xps_cpus_map); |
fc9bab24 AN |
3867 | if (dev_maps) { |
3868 | unsigned int tci = skb->sender_cpu - 1; | |
3869 | ||
3870 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
3871 | tci); | |
638b2a69 JP |
3872 | } |
3873 | } | |
3874 | rcu_read_unlock(); | |
3875 | ||
3876 | return queue_index; | |
3877 | #else | |
3878 | return -1; | |
3879 | #endif | |
3880 | } | |
3881 | ||
a4ea8a3d | 3882 | u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, |
a350ecce | 3883 | struct net_device *sb_dev) |
a4ea8a3d AD |
3884 | { |
3885 | return 0; | |
3886 | } | |
3887 | EXPORT_SYMBOL(dev_pick_tx_zero); | |
3888 | ||
3889 | u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, | |
a350ecce | 3890 | struct net_device *sb_dev) |
a4ea8a3d AD |
3891 | { |
3892 | return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; | |
3893 | } | |
3894 | EXPORT_SYMBOL(dev_pick_tx_cpu_id); | |
3895 | ||
b71b5837 PA |
3896 | u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, |
3897 | struct net_device *sb_dev) | |
638b2a69 JP |
3898 | { |
3899 | struct sock *sk = skb->sk; | |
3900 | int queue_index = sk_tx_queue_get(sk); | |
3901 | ||
eadec877 AD |
3902 | sb_dev = sb_dev ? : dev; |
3903 | ||
638b2a69 JP |
3904 | if (queue_index < 0 || skb->ooo_okay || |
3905 | queue_index >= dev->real_num_tx_queues) { | |
eadec877 | 3906 | int new_index = get_xps_queue(dev, sb_dev, skb); |
f4563a75 | 3907 | |
638b2a69 | 3908 | if (new_index < 0) |
eadec877 | 3909 | new_index = skb_tx_hash(dev, sb_dev, skb); |
638b2a69 JP |
3910 | |
3911 | if (queue_index != new_index && sk && | |
004a5d01 | 3912 | sk_fullsock(sk) && |
638b2a69 JP |
3913 | rcu_access_pointer(sk->sk_dst_cache)) |
3914 | sk_tx_queue_set(sk, new_index); | |
3915 | ||
3916 | queue_index = new_index; | |
3917 | } | |
3918 | ||
3919 | return queue_index; | |
3920 | } | |
b71b5837 | 3921 | EXPORT_SYMBOL(netdev_pick_tx); |
638b2a69 | 3922 | |
4bd97d51 PA |
3923 | struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, |
3924 | struct sk_buff *skb, | |
3925 | struct net_device *sb_dev) | |
638b2a69 JP |
3926 | { |
3927 | int queue_index = 0; | |
3928 | ||
3929 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
3930 | u32 sender_cpu = skb->sender_cpu - 1; |
3931 | ||
3932 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
3933 | skb->sender_cpu = raw_smp_processor_id() + 1; |
3934 | #endif | |
3935 | ||
3936 | if (dev->real_num_tx_queues != 1) { | |
3937 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 3938 | |
638b2a69 | 3939 | if (ops->ndo_select_queue) |
a350ecce | 3940 | queue_index = ops->ndo_select_queue(dev, skb, sb_dev); |
638b2a69 | 3941 | else |
4bd97d51 | 3942 | queue_index = netdev_pick_tx(dev, skb, sb_dev); |
638b2a69 | 3943 | |
d584527c | 3944 | queue_index = netdev_cap_txqueue(dev, queue_index); |
638b2a69 JP |
3945 | } |
3946 | ||
3947 | skb_set_queue_mapping(skb, queue_index); | |
3948 | return netdev_get_tx_queue(dev, queue_index); | |
3949 | } | |
3950 | ||
d29f749e | 3951 | /** |
9d08dd3d | 3952 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3953 | * @skb: buffer to transmit |
eadec877 | 3954 | * @sb_dev: suboordinate device used for L2 forwarding offload |
d29f749e DJ |
3955 | * |
3956 | * Queue a buffer for transmission to a network device. The caller must | |
3957 | * have set the device and priority and built the buffer before calling | |
3958 | * this function. The function can be called from an interrupt. | |
3959 | * | |
3960 | * A negative errno code is returned on a failure. A success does not | |
3961 | * guarantee the frame will be transmitted as it may be dropped due | |
3962 | * to congestion or traffic shaping. | |
3963 | * | |
3964 | * ----------------------------------------------------------------------------------- | |
3965 | * I notice this method can also return errors from the queue disciplines, | |
3966 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3967 | * be positive. | |
3968 | * | |
3969 | * Regardless of the return value, the skb is consumed, so it is currently | |
3970 | * difficult to retry a send to this method. (You can bump the ref count | |
3971 | * before sending to hold a reference for retry if you are careful.) | |
3972 | * | |
3973 | * When calling this method, interrupts MUST be enabled. This is because | |
3974 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3975 | * --BLG | |
3976 | */ | |
eadec877 | 3977 | static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) |
1da177e4 LT |
3978 | { |
3979 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3980 | struct netdev_queue *txq; |
1da177e4 LT |
3981 | struct Qdisc *q; |
3982 | int rc = -ENOMEM; | |
f53c7239 | 3983 | bool again = false; |
1da177e4 | 3984 | |
6d1ccff6 ED |
3985 | skb_reset_mac_header(skb); |
3986 | ||
e7fd2885 WB |
3987 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3988 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3989 | ||
4ec93edb YH |
3990 | /* Disable soft irqs for various locks below. Also |
3991 | * stops preemption for RCU. | |
1da177e4 | 3992 | */ |
4ec93edb | 3993 | rcu_read_lock_bh(); |
1da177e4 | 3994 | |
5bc1421e NH |
3995 | skb_update_prio(skb); |
3996 | ||
1f211a1b DB |
3997 | qdisc_pkt_len_init(skb); |
3998 | #ifdef CONFIG_NET_CLS_ACT | |
8dc07fdb | 3999 | skb->tc_at_ingress = 0; |
357b6cc5 | 4000 | # ifdef CONFIG_NET_EGRESS |
aabf6772 | 4001 | if (static_branch_unlikely(&egress_needed_key)) { |
1f211a1b DB |
4002 | skb = sch_handle_egress(skb, &rc, dev); |
4003 | if (!skb) | |
4004 | goto out; | |
4005 | } | |
357b6cc5 | 4006 | # endif |
1f211a1b | 4007 | #endif |
02875878 ED |
4008 | /* If device/qdisc don't need skb->dst, release it right now while |
4009 | * its hot in this cpu cache. | |
4010 | */ | |
4011 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
4012 | skb_dst_drop(skb); | |
4013 | else | |
4014 | skb_dst_force(skb); | |
4015 | ||
4bd97d51 | 4016 | txq = netdev_core_pick_tx(dev, skb, sb_dev); |
a898def2 | 4017 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 4018 | |
cf66ba58 | 4019 | trace_net_dev_queue(skb); |
1da177e4 | 4020 | if (q->enqueue) { |
bbd8a0d3 | 4021 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 4022 | goto out; |
1da177e4 LT |
4023 | } |
4024 | ||
4025 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 4026 | * loopback, all the sorts of tunnels... |
1da177e4 | 4027 | |
eb13da1a | 4028 | * Really, it is unlikely that netif_tx_lock protection is necessary |
4029 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
4030 | * counters.) | |
4031 | * However, it is possible, that they rely on protection | |
4032 | * made by us here. | |
1da177e4 | 4033 | |
eb13da1a | 4034 | * Check this and shot the lock. It is not prone from deadlocks. |
4035 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
4036 | */ |
4037 | if (dev->flags & IFF_UP) { | |
4038 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
4039 | ||
c773e847 | 4040 | if (txq->xmit_lock_owner != cpu) { |
97cdcf37 | 4041 | if (dev_xmit_recursion()) |
745e20f1 ED |
4042 | goto recursion_alert; |
4043 | ||
f53c7239 | 4044 | skb = validate_xmit_skb(skb, dev, &again); |
1f59533f | 4045 | if (!skb) |
d21fd63e | 4046 | goto out; |
1f59533f | 4047 | |
c773e847 | 4048 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 4049 | |
73466498 | 4050 | if (!netif_xmit_stopped(txq)) { |
97cdcf37 | 4051 | dev_xmit_recursion_inc(); |
ce93718f | 4052 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
97cdcf37 | 4053 | dev_xmit_recursion_dec(); |
572a9d7b | 4054 | if (dev_xmit_complete(rc)) { |
c773e847 | 4055 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
4056 | goto out; |
4057 | } | |
4058 | } | |
c773e847 | 4059 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
4060 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
4061 | dev->name); | |
1da177e4 LT |
4062 | } else { |
4063 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
4064 | * unfortunately |
4065 | */ | |
4066 | recursion_alert: | |
e87cc472 JP |
4067 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
4068 | dev->name); | |
1da177e4 LT |
4069 | } |
4070 | } | |
4071 | ||
4072 | rc = -ENETDOWN; | |
d4828d85 | 4073 | rcu_read_unlock_bh(); |
1da177e4 | 4074 | |
015f0688 | 4075 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 4076 | kfree_skb_list(skb); |
1da177e4 LT |
4077 | return rc; |
4078 | out: | |
d4828d85 | 4079 | rcu_read_unlock_bh(); |
1da177e4 LT |
4080 | return rc; |
4081 | } | |
f663dd9a | 4082 | |
2b4aa3ce | 4083 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
4084 | { |
4085 | return __dev_queue_xmit(skb, NULL); | |
4086 | } | |
2b4aa3ce | 4087 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 4088 | |
eadec877 | 4089 | int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) |
f663dd9a | 4090 | { |
eadec877 | 4091 | return __dev_queue_xmit(skb, sb_dev); |
f663dd9a JW |
4092 | } |
4093 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
4094 | ||
865b03f2 MK |
4095 | int dev_direct_xmit(struct sk_buff *skb, u16 queue_id) |
4096 | { | |
4097 | struct net_device *dev = skb->dev; | |
4098 | struct sk_buff *orig_skb = skb; | |
4099 | struct netdev_queue *txq; | |
4100 | int ret = NETDEV_TX_BUSY; | |
4101 | bool again = false; | |
4102 | ||
4103 | if (unlikely(!netif_running(dev) || | |
4104 | !netif_carrier_ok(dev))) | |
4105 | goto drop; | |
4106 | ||
4107 | skb = validate_xmit_skb_list(skb, dev, &again); | |
4108 | if (skb != orig_skb) | |
4109 | goto drop; | |
4110 | ||
4111 | skb_set_queue_mapping(skb, queue_id); | |
4112 | txq = skb_get_tx_queue(dev, skb); | |
4113 | ||
4114 | local_bh_disable(); | |
4115 | ||
4116 | HARD_TX_LOCK(dev, txq, smp_processor_id()); | |
4117 | if (!netif_xmit_frozen_or_drv_stopped(txq)) | |
4118 | ret = netdev_start_xmit(skb, dev, txq, false); | |
4119 | HARD_TX_UNLOCK(dev, txq); | |
4120 | ||
4121 | local_bh_enable(); | |
4122 | ||
4123 | if (!dev_xmit_complete(ret)) | |
4124 | kfree_skb(skb); | |
4125 | ||
4126 | return ret; | |
4127 | drop: | |
4128 | atomic_long_inc(&dev->tx_dropped); | |
4129 | kfree_skb_list(skb); | |
4130 | return NET_XMIT_DROP; | |
4131 | } | |
4132 | EXPORT_SYMBOL(dev_direct_xmit); | |
1da177e4 | 4133 | |
eb13da1a | 4134 | /************************************************************************* |
4135 | * Receiver routines | |
4136 | *************************************************************************/ | |
1da177e4 | 4137 | |
6b2bedc3 | 4138 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
4139 | EXPORT_SYMBOL(netdev_max_backlog); |
4140 | ||
3b098e2d | 4141 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 | 4142 | int netdev_budget __read_mostly = 300; |
a4837980 KK |
4143 | /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ |
4144 | unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; | |
3d48b53f MT |
4145 | int weight_p __read_mostly = 64; /* old backlog weight */ |
4146 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
4147 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
4148 | int dev_rx_weight __read_mostly = 64; | |
4149 | int dev_tx_weight __read_mostly = 64; | |
323ebb61 EC |
4150 | /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ |
4151 | int gro_normal_batch __read_mostly = 8; | |
1da177e4 | 4152 | |
eecfd7c4 ED |
4153 | /* Called with irq disabled */ |
4154 | static inline void ____napi_schedule(struct softnet_data *sd, | |
4155 | struct napi_struct *napi) | |
4156 | { | |
4157 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
4158 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4159 | } | |
4160 | ||
bfb564e7 KK |
4161 | #ifdef CONFIG_RPS |
4162 | ||
4163 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 4164 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 4165 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
4166 | u32 rps_cpu_mask __read_mostly; |
4167 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 4168 | |
dc05360f | 4169 | struct static_key_false rps_needed __read_mostly; |
3df97ba8 | 4170 | EXPORT_SYMBOL(rps_needed); |
dc05360f | 4171 | struct static_key_false rfs_needed __read_mostly; |
13bfff25 | 4172 | EXPORT_SYMBOL(rfs_needed); |
adc9300e | 4173 | |
c445477d BH |
4174 | static struct rps_dev_flow * |
4175 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4176 | struct rps_dev_flow *rflow, u16 next_cpu) | |
4177 | { | |
a31196b0 | 4178 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
4179 | #ifdef CONFIG_RFS_ACCEL |
4180 | struct netdev_rx_queue *rxqueue; | |
4181 | struct rps_dev_flow_table *flow_table; | |
4182 | struct rps_dev_flow *old_rflow; | |
4183 | u32 flow_id; | |
4184 | u16 rxq_index; | |
4185 | int rc; | |
4186 | ||
4187 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
4188 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
4189 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
4190 | goto out; |
4191 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
4192 | if (rxq_index == skb_get_rx_queue(skb)) | |
4193 | goto out; | |
4194 | ||
4195 | rxqueue = dev->_rx + rxq_index; | |
4196 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4197 | if (!flow_table) | |
4198 | goto out; | |
61b905da | 4199 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
4200 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
4201 | rxq_index, flow_id); | |
4202 | if (rc < 0) | |
4203 | goto out; | |
4204 | old_rflow = rflow; | |
4205 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
4206 | rflow->filter = rc; |
4207 | if (old_rflow->filter == rflow->filter) | |
4208 | old_rflow->filter = RPS_NO_FILTER; | |
4209 | out: | |
4210 | #endif | |
4211 | rflow->last_qtail = | |
09994d1b | 4212 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
4213 | } |
4214 | ||
09994d1b | 4215 | rflow->cpu = next_cpu; |
c445477d BH |
4216 | return rflow; |
4217 | } | |
4218 | ||
bfb564e7 KK |
4219 | /* |
4220 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
4221 | * CPU from the RPS map of the receiving queue for a given skb. | |
4222 | * rcu_read_lock must be held on entry. | |
4223 | */ | |
4224 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4225 | struct rps_dev_flow **rflowp) | |
4226 | { | |
567e4b79 ED |
4227 | const struct rps_sock_flow_table *sock_flow_table; |
4228 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 4229 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 4230 | struct rps_map *map; |
bfb564e7 | 4231 | int cpu = -1; |
567e4b79 | 4232 | u32 tcpu; |
61b905da | 4233 | u32 hash; |
bfb564e7 KK |
4234 | |
4235 | if (skb_rx_queue_recorded(skb)) { | |
4236 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 4237 | |
62fe0b40 BH |
4238 | if (unlikely(index >= dev->real_num_rx_queues)) { |
4239 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4240 | "%s received packet on queue %u, but number " | |
4241 | "of RX queues is %u\n", | |
4242 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
4243 | goto done; |
4244 | } | |
567e4b79 ED |
4245 | rxqueue += index; |
4246 | } | |
bfb564e7 | 4247 | |
567e4b79 ED |
4248 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
4249 | ||
4250 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 4251 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 4252 | if (!flow_table && !map) |
bfb564e7 KK |
4253 | goto done; |
4254 | ||
2d47b459 | 4255 | skb_reset_network_header(skb); |
61b905da TH |
4256 | hash = skb_get_hash(skb); |
4257 | if (!hash) | |
bfb564e7 KK |
4258 | goto done; |
4259 | ||
fec5e652 TH |
4260 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
4261 | if (flow_table && sock_flow_table) { | |
fec5e652 | 4262 | struct rps_dev_flow *rflow; |
567e4b79 ED |
4263 | u32 next_cpu; |
4264 | u32 ident; | |
4265 | ||
4266 | /* First check into global flow table if there is a match */ | |
4267 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
4268 | if ((ident ^ hash) & ~rps_cpu_mask) | |
4269 | goto try_rps; | |
fec5e652 | 4270 | |
567e4b79 ED |
4271 | next_cpu = ident & rps_cpu_mask; |
4272 | ||
4273 | /* OK, now we know there is a match, | |
4274 | * we can look at the local (per receive queue) flow table | |
4275 | */ | |
61b905da | 4276 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
4277 | tcpu = rflow->cpu; |
4278 | ||
fec5e652 TH |
4279 | /* |
4280 | * If the desired CPU (where last recvmsg was done) is | |
4281 | * different from current CPU (one in the rx-queue flow | |
4282 | * table entry), switch if one of the following holds: | |
a31196b0 | 4283 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
4284 | * - Current CPU is offline. |
4285 | * - The current CPU's queue tail has advanced beyond the | |
4286 | * last packet that was enqueued using this table entry. | |
4287 | * This guarantees that all previous packets for the flow | |
4288 | * have been dequeued, thus preserving in order delivery. | |
4289 | */ | |
4290 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 4291 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 4292 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
4293 | rflow->last_qtail)) >= 0)) { |
4294 | tcpu = next_cpu; | |
c445477d | 4295 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 4296 | } |
c445477d | 4297 | |
a31196b0 | 4298 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
4299 | *rflowp = rflow; |
4300 | cpu = tcpu; | |
4301 | goto done; | |
4302 | } | |
4303 | } | |
4304 | ||
567e4b79 ED |
4305 | try_rps: |
4306 | ||
0a9627f2 | 4307 | if (map) { |
8fc54f68 | 4308 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
4309 | if (cpu_online(tcpu)) { |
4310 | cpu = tcpu; | |
4311 | goto done; | |
4312 | } | |
4313 | } | |
4314 | ||
4315 | done: | |
0a9627f2 TH |
4316 | return cpu; |
4317 | } | |
4318 | ||
c445477d BH |
4319 | #ifdef CONFIG_RFS_ACCEL |
4320 | ||
4321 | /** | |
4322 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
4323 | * @dev: Device on which the filter was set | |
4324 | * @rxq_index: RX queue index | |
4325 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
4326 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
4327 | * | |
4328 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
4329 | * this function for each installed filter and remove the filters for | |
4330 | * which it returns %true. | |
4331 | */ | |
4332 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
4333 | u32 flow_id, u16 filter_id) | |
4334 | { | |
4335 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
4336 | struct rps_dev_flow_table *flow_table; | |
4337 | struct rps_dev_flow *rflow; | |
4338 | bool expire = true; | |
a31196b0 | 4339 | unsigned int cpu; |
c445477d BH |
4340 | |
4341 | rcu_read_lock(); | |
4342 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4343 | if (flow_table && flow_id <= flow_table->mask) { | |
4344 | rflow = &flow_table->flows[flow_id]; | |
6aa7de05 | 4345 | cpu = READ_ONCE(rflow->cpu); |
a31196b0 | 4346 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
4347 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
4348 | rflow->last_qtail) < | |
4349 | (int)(10 * flow_table->mask))) | |
4350 | expire = false; | |
4351 | } | |
4352 | rcu_read_unlock(); | |
4353 | return expire; | |
4354 | } | |
4355 | EXPORT_SYMBOL(rps_may_expire_flow); | |
4356 | ||
4357 | #endif /* CONFIG_RFS_ACCEL */ | |
4358 | ||
0a9627f2 | 4359 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 4360 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 4361 | { |
e36fa2f7 ED |
4362 | struct softnet_data *sd = data; |
4363 | ||
eecfd7c4 | 4364 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 4365 | sd->received_rps++; |
0a9627f2 | 4366 | } |
e36fa2f7 | 4367 | |
fec5e652 | 4368 | #endif /* CONFIG_RPS */ |
0a9627f2 | 4369 | |
e36fa2f7 ED |
4370 | /* |
4371 | * Check if this softnet_data structure is another cpu one | |
4372 | * If yes, queue it to our IPI list and return 1 | |
4373 | * If no, return 0 | |
4374 | */ | |
4375 | static int rps_ipi_queued(struct softnet_data *sd) | |
4376 | { | |
4377 | #ifdef CONFIG_RPS | |
903ceff7 | 4378 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
4379 | |
4380 | if (sd != mysd) { | |
4381 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
4382 | mysd->rps_ipi_list = sd; | |
4383 | ||
4384 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4385 | return 1; | |
4386 | } | |
4387 | #endif /* CONFIG_RPS */ | |
4388 | return 0; | |
4389 | } | |
4390 | ||
99bbc707 WB |
4391 | #ifdef CONFIG_NET_FLOW_LIMIT |
4392 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
4393 | #endif | |
4394 | ||
4395 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
4396 | { | |
4397 | #ifdef CONFIG_NET_FLOW_LIMIT | |
4398 | struct sd_flow_limit *fl; | |
4399 | struct softnet_data *sd; | |
4400 | unsigned int old_flow, new_flow; | |
4401 | ||
4402 | if (qlen < (netdev_max_backlog >> 1)) | |
4403 | return false; | |
4404 | ||
903ceff7 | 4405 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
4406 | |
4407 | rcu_read_lock(); | |
4408 | fl = rcu_dereference(sd->flow_limit); | |
4409 | if (fl) { | |
3958afa1 | 4410 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
4411 | old_flow = fl->history[fl->history_head]; |
4412 | fl->history[fl->history_head] = new_flow; | |
4413 | ||
4414 | fl->history_head++; | |
4415 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
4416 | ||
4417 | if (likely(fl->buckets[old_flow])) | |
4418 | fl->buckets[old_flow]--; | |
4419 | ||
4420 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
4421 | fl->count++; | |
4422 | rcu_read_unlock(); | |
4423 | return true; | |
4424 | } | |
4425 | } | |
4426 | rcu_read_unlock(); | |
4427 | #endif | |
4428 | return false; | |
4429 | } | |
4430 | ||
0a9627f2 TH |
4431 | /* |
4432 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
4433 | * queue (may be a remote CPU queue). | |
4434 | */ | |
fec5e652 TH |
4435 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
4436 | unsigned int *qtail) | |
0a9627f2 | 4437 | { |
e36fa2f7 | 4438 | struct softnet_data *sd; |
0a9627f2 | 4439 | unsigned long flags; |
99bbc707 | 4440 | unsigned int qlen; |
0a9627f2 | 4441 | |
e36fa2f7 | 4442 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
4443 | |
4444 | local_irq_save(flags); | |
0a9627f2 | 4445 | |
e36fa2f7 | 4446 | rps_lock(sd); |
e9e4dd32 JA |
4447 | if (!netif_running(skb->dev)) |
4448 | goto drop; | |
99bbc707 WB |
4449 | qlen = skb_queue_len(&sd->input_pkt_queue); |
4450 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 4451 | if (qlen) { |
0a9627f2 | 4452 | enqueue: |
e36fa2f7 | 4453 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 4454 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 4455 | rps_unlock(sd); |
152102c7 | 4456 | local_irq_restore(flags); |
0a9627f2 TH |
4457 | return NET_RX_SUCCESS; |
4458 | } | |
4459 | ||
ebda37c2 ED |
4460 | /* Schedule NAPI for backlog device |
4461 | * We can use non atomic operation since we own the queue lock | |
4462 | */ | |
4463 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 4464 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 4465 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
4466 | } |
4467 | goto enqueue; | |
4468 | } | |
4469 | ||
e9e4dd32 | 4470 | drop: |
dee42870 | 4471 | sd->dropped++; |
e36fa2f7 | 4472 | rps_unlock(sd); |
0a9627f2 | 4473 | |
0a9627f2 TH |
4474 | local_irq_restore(flags); |
4475 | ||
caf586e5 | 4476 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
4477 | kfree_skb(skb); |
4478 | return NET_RX_DROP; | |
4479 | } | |
1da177e4 | 4480 | |
e817f856 JDB |
4481 | static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) |
4482 | { | |
4483 | struct net_device *dev = skb->dev; | |
4484 | struct netdev_rx_queue *rxqueue; | |
4485 | ||
4486 | rxqueue = dev->_rx; | |
4487 | ||
4488 | if (skb_rx_queue_recorded(skb)) { | |
4489 | u16 index = skb_get_rx_queue(skb); | |
4490 | ||
4491 | if (unlikely(index >= dev->real_num_rx_queues)) { | |
4492 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4493 | "%s received packet on queue %u, but number " | |
4494 | "of RX queues is %u\n", | |
4495 | dev->name, index, dev->real_num_rx_queues); | |
4496 | ||
4497 | return rxqueue; /* Return first rxqueue */ | |
4498 | } | |
4499 | rxqueue += index; | |
4500 | } | |
4501 | return rxqueue; | |
4502 | } | |
4503 | ||
d4455169 | 4504 | static u32 netif_receive_generic_xdp(struct sk_buff *skb, |
02671e23 | 4505 | struct xdp_buff *xdp, |
d4455169 JF |
4506 | struct bpf_prog *xdp_prog) |
4507 | { | |
e817f856 | 4508 | struct netdev_rx_queue *rxqueue; |
198d83bb | 4509 | void *orig_data, *orig_data_end; |
de8f3a83 | 4510 | u32 metalen, act = XDP_DROP; |
29724956 JDB |
4511 | __be16 orig_eth_type; |
4512 | struct ethhdr *eth; | |
4513 | bool orig_bcast; | |
d4455169 JF |
4514 | int hlen, off; |
4515 | u32 mac_len; | |
4516 | ||
4517 | /* Reinjected packets coming from act_mirred or similar should | |
4518 | * not get XDP generic processing. | |
4519 | */ | |
2c64605b | 4520 | if (skb_is_redirected(skb)) |
d4455169 JF |
4521 | return XDP_PASS; |
4522 | ||
de8f3a83 DB |
4523 | /* XDP packets must be linear and must have sufficient headroom |
4524 | * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also | |
4525 | * native XDP provides, thus we need to do it here as well. | |
4526 | */ | |
ad1e03b2 | 4527 | if (skb_cloned(skb) || skb_is_nonlinear(skb) || |
de8f3a83 DB |
4528 | skb_headroom(skb) < XDP_PACKET_HEADROOM) { |
4529 | int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); | |
4530 | int troom = skb->tail + skb->data_len - skb->end; | |
4531 | ||
4532 | /* In case we have to go down the path and also linearize, | |
4533 | * then lets do the pskb_expand_head() work just once here. | |
4534 | */ | |
4535 | if (pskb_expand_head(skb, | |
4536 | hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, | |
4537 | troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) | |
4538 | goto do_drop; | |
2d17d8d7 | 4539 | if (skb_linearize(skb)) |
de8f3a83 DB |
4540 | goto do_drop; |
4541 | } | |
d4455169 JF |
4542 | |
4543 | /* The XDP program wants to see the packet starting at the MAC | |
4544 | * header. | |
4545 | */ | |
4546 | mac_len = skb->data - skb_mac_header(skb); | |
4547 | hlen = skb_headlen(skb) + mac_len; | |
02671e23 BT |
4548 | xdp->data = skb->data - mac_len; |
4549 | xdp->data_meta = xdp->data; | |
4550 | xdp->data_end = xdp->data + hlen; | |
4551 | xdp->data_hard_start = skb->data - skb_headroom(skb); | |
4552 | orig_data_end = xdp->data_end; | |
4553 | orig_data = xdp->data; | |
29724956 JDB |
4554 | eth = (struct ethhdr *)xdp->data; |
4555 | orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); | |
4556 | orig_eth_type = eth->h_proto; | |
d4455169 | 4557 | |
e817f856 | 4558 | rxqueue = netif_get_rxqueue(skb); |
02671e23 | 4559 | xdp->rxq = &rxqueue->xdp_rxq; |
e817f856 | 4560 | |
02671e23 | 4561 | act = bpf_prog_run_xdp(xdp_prog, xdp); |
d4455169 | 4562 | |
065af355 | 4563 | /* check if bpf_xdp_adjust_head was used */ |
02671e23 | 4564 | off = xdp->data - orig_data; |
065af355 JDB |
4565 | if (off) { |
4566 | if (off > 0) | |
4567 | __skb_pull(skb, off); | |
4568 | else if (off < 0) | |
4569 | __skb_push(skb, -off); | |
4570 | ||
4571 | skb->mac_header += off; | |
4572 | skb_reset_network_header(skb); | |
4573 | } | |
d4455169 | 4574 | |
198d83bb NS |
4575 | /* check if bpf_xdp_adjust_tail was used. it can only "shrink" |
4576 | * pckt. | |
4577 | */ | |
02671e23 | 4578 | off = orig_data_end - xdp->data_end; |
f7613120 | 4579 | if (off != 0) { |
02671e23 | 4580 | skb_set_tail_pointer(skb, xdp->data_end - xdp->data); |
f7613120 | 4581 | skb->len -= off; |
02671e23 | 4582 | |
f7613120 | 4583 | } |
198d83bb | 4584 | |
29724956 JDB |
4585 | /* check if XDP changed eth hdr such SKB needs update */ |
4586 | eth = (struct ethhdr *)xdp->data; | |
4587 | if ((orig_eth_type != eth->h_proto) || | |
4588 | (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { | |
4589 | __skb_push(skb, ETH_HLEN); | |
4590 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4591 | } | |
4592 | ||
d4455169 | 4593 | switch (act) { |
6103aa96 | 4594 | case XDP_REDIRECT: |
d4455169 JF |
4595 | case XDP_TX: |
4596 | __skb_push(skb, mac_len); | |
de8f3a83 | 4597 | break; |
d4455169 | 4598 | case XDP_PASS: |
02671e23 | 4599 | metalen = xdp->data - xdp->data_meta; |
de8f3a83 DB |
4600 | if (metalen) |
4601 | skb_metadata_set(skb, metalen); | |
d4455169 | 4602 | break; |
d4455169 JF |
4603 | default: |
4604 | bpf_warn_invalid_xdp_action(act); | |
4605 | /* fall through */ | |
4606 | case XDP_ABORTED: | |
4607 | trace_xdp_exception(skb->dev, xdp_prog, act); | |
4608 | /* fall through */ | |
4609 | case XDP_DROP: | |
4610 | do_drop: | |
4611 | kfree_skb(skb); | |
4612 | break; | |
4613 | } | |
4614 | ||
4615 | return act; | |
4616 | } | |
4617 | ||
4618 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
4619 | * network taps in order to match in-driver-XDP behavior. | |
4620 | */ | |
7c497478 | 4621 | void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) |
d4455169 JF |
4622 | { |
4623 | struct net_device *dev = skb->dev; | |
4624 | struct netdev_queue *txq; | |
4625 | bool free_skb = true; | |
4626 | int cpu, rc; | |
4627 | ||
4bd97d51 | 4628 | txq = netdev_core_pick_tx(dev, skb, NULL); |
d4455169 JF |
4629 | cpu = smp_processor_id(); |
4630 | HARD_TX_LOCK(dev, txq, cpu); | |
4631 | if (!netif_xmit_stopped(txq)) { | |
4632 | rc = netdev_start_xmit(skb, dev, txq, 0); | |
4633 | if (dev_xmit_complete(rc)) | |
4634 | free_skb = false; | |
4635 | } | |
4636 | HARD_TX_UNLOCK(dev, txq); | |
4637 | if (free_skb) { | |
4638 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
4639 | kfree_skb(skb); | |
4640 | } | |
4641 | } | |
4642 | ||
02786475 | 4643 | static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); |
d4455169 | 4644 | |
7c497478 | 4645 | int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) |
d4455169 | 4646 | { |
d4455169 | 4647 | if (xdp_prog) { |
02671e23 BT |
4648 | struct xdp_buff xdp; |
4649 | u32 act; | |
6103aa96 | 4650 | int err; |
d4455169 | 4651 | |
02671e23 | 4652 | act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); |
d4455169 | 4653 | if (act != XDP_PASS) { |
6103aa96 JF |
4654 | switch (act) { |
4655 | case XDP_REDIRECT: | |
2facaad6 | 4656 | err = xdp_do_generic_redirect(skb->dev, skb, |
02671e23 | 4657 | &xdp, xdp_prog); |
6103aa96 JF |
4658 | if (err) |
4659 | goto out_redir; | |
02671e23 | 4660 | break; |
6103aa96 | 4661 | case XDP_TX: |
d4455169 | 4662 | generic_xdp_tx(skb, xdp_prog); |
6103aa96 JF |
4663 | break; |
4664 | } | |
d4455169 JF |
4665 | return XDP_DROP; |
4666 | } | |
4667 | } | |
4668 | return XDP_PASS; | |
6103aa96 | 4669 | out_redir: |
6103aa96 JF |
4670 | kfree_skb(skb); |
4671 | return XDP_DROP; | |
d4455169 | 4672 | } |
7c497478 | 4673 | EXPORT_SYMBOL_GPL(do_xdp_generic); |
d4455169 | 4674 | |
ae78dbfa | 4675 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 4676 | { |
b0e28f1e | 4677 | int ret; |
1da177e4 | 4678 | |
588f0330 | 4679 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 4680 | |
cf66ba58 | 4681 | trace_netif_rx(skb); |
d4455169 | 4682 | |
df334545 | 4683 | #ifdef CONFIG_RPS |
dc05360f | 4684 | if (static_branch_unlikely(&rps_needed)) { |
fec5e652 | 4685 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
4686 | int cpu; |
4687 | ||
cece1945 | 4688 | preempt_disable(); |
b0e28f1e | 4689 | rcu_read_lock(); |
fec5e652 TH |
4690 | |
4691 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
4692 | if (cpu < 0) |
4693 | cpu = smp_processor_id(); | |
fec5e652 TH |
4694 | |
4695 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4696 | ||
b0e28f1e | 4697 | rcu_read_unlock(); |
cece1945 | 4698 | preempt_enable(); |
adc9300e ED |
4699 | } else |
4700 | #endif | |
fec5e652 TH |
4701 | { |
4702 | unsigned int qtail; | |
f4563a75 | 4703 | |
fec5e652 TH |
4704 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); |
4705 | put_cpu(); | |
4706 | } | |
b0e28f1e | 4707 | return ret; |
1da177e4 | 4708 | } |
ae78dbfa BH |
4709 | |
4710 | /** | |
4711 | * netif_rx - post buffer to the network code | |
4712 | * @skb: buffer to post | |
4713 | * | |
4714 | * This function receives a packet from a device driver and queues it for | |
4715 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
4716 | * may be dropped during processing for congestion control or by the | |
4717 | * protocol layers. | |
4718 | * | |
4719 | * return values: | |
4720 | * NET_RX_SUCCESS (no congestion) | |
4721 | * NET_RX_DROP (packet was dropped) | |
4722 | * | |
4723 | */ | |
4724 | ||
4725 | int netif_rx(struct sk_buff *skb) | |
4726 | { | |
b0e3f1bd GB |
4727 | int ret; |
4728 | ||
ae78dbfa BH |
4729 | trace_netif_rx_entry(skb); |
4730 | ||
b0e3f1bd GB |
4731 | ret = netif_rx_internal(skb); |
4732 | trace_netif_rx_exit(ret); | |
4733 | ||
4734 | return ret; | |
ae78dbfa | 4735 | } |
d1b19dff | 4736 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
4737 | |
4738 | int netif_rx_ni(struct sk_buff *skb) | |
4739 | { | |
4740 | int err; | |
4741 | ||
ae78dbfa BH |
4742 | trace_netif_rx_ni_entry(skb); |
4743 | ||
1da177e4 | 4744 | preempt_disable(); |
ae78dbfa | 4745 | err = netif_rx_internal(skb); |
1da177e4 LT |
4746 | if (local_softirq_pending()) |
4747 | do_softirq(); | |
4748 | preempt_enable(); | |
b0e3f1bd | 4749 | trace_netif_rx_ni_exit(err); |
1da177e4 LT |
4750 | |
4751 | return err; | |
4752 | } | |
1da177e4 LT |
4753 | EXPORT_SYMBOL(netif_rx_ni); |
4754 | ||
0766f788 | 4755 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 4756 | { |
903ceff7 | 4757 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
4758 | |
4759 | if (sd->completion_queue) { | |
4760 | struct sk_buff *clist; | |
4761 | ||
4762 | local_irq_disable(); | |
4763 | clist = sd->completion_queue; | |
4764 | sd->completion_queue = NULL; | |
4765 | local_irq_enable(); | |
4766 | ||
4767 | while (clist) { | |
4768 | struct sk_buff *skb = clist; | |
f4563a75 | 4769 | |
1da177e4 LT |
4770 | clist = clist->next; |
4771 | ||
63354797 | 4772 | WARN_ON(refcount_read(&skb->users)); |
e6247027 ED |
4773 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
4774 | trace_consume_skb(skb); | |
4775 | else | |
4776 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
4777 | |
4778 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
4779 | __kfree_skb(skb); | |
4780 | else | |
4781 | __kfree_skb_defer(skb); | |
1da177e4 | 4782 | } |
15fad714 JDB |
4783 | |
4784 | __kfree_skb_flush(); | |
1da177e4 LT |
4785 | } |
4786 | ||
4787 | if (sd->output_queue) { | |
37437bb2 | 4788 | struct Qdisc *head; |
1da177e4 LT |
4789 | |
4790 | local_irq_disable(); | |
4791 | head = sd->output_queue; | |
4792 | sd->output_queue = NULL; | |
a9cbd588 | 4793 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
4794 | local_irq_enable(); |
4795 | ||
4796 | while (head) { | |
37437bb2 | 4797 | struct Qdisc *q = head; |
6b3ba914 | 4798 | spinlock_t *root_lock = NULL; |
37437bb2 | 4799 | |
1da177e4 LT |
4800 | head = head->next_sched; |
4801 | ||
6b3ba914 JF |
4802 | if (!(q->flags & TCQ_F_NOLOCK)) { |
4803 | root_lock = qdisc_lock(q); | |
4804 | spin_lock(root_lock); | |
4805 | } | |
3bcb846c ED |
4806 | /* We need to make sure head->next_sched is read |
4807 | * before clearing __QDISC_STATE_SCHED | |
4808 | */ | |
4809 | smp_mb__before_atomic(); | |
4810 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
4811 | qdisc_run(q); | |
6b3ba914 JF |
4812 | if (root_lock) |
4813 | spin_unlock(root_lock); | |
1da177e4 LT |
4814 | } |
4815 | } | |
f53c7239 SK |
4816 | |
4817 | xfrm_dev_backlog(sd); | |
1da177e4 LT |
4818 | } |
4819 | ||
181402a5 | 4820 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
4821 | /* This hook is defined here for ATM LANE */ |
4822 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
4823 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 4824 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 4825 | #endif |
1da177e4 | 4826 | |
1f211a1b DB |
4827 | static inline struct sk_buff * |
4828 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
4829 | struct net_device *orig_dev) | |
f697c3e8 | 4830 | { |
e7582bab | 4831 | #ifdef CONFIG_NET_CLS_ACT |
46209401 | 4832 | struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); |
d2788d34 | 4833 | struct tcf_result cl_res; |
24824a09 | 4834 | |
c9e99fd0 DB |
4835 | /* If there's at least one ingress present somewhere (so |
4836 | * we get here via enabled static key), remaining devices | |
4837 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 4838 | * out here. |
c9e99fd0 | 4839 | */ |
46209401 | 4840 | if (!miniq) |
4577139b | 4841 | return skb; |
46209401 | 4842 | |
f697c3e8 HX |
4843 | if (*pt_prev) { |
4844 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4845 | *pt_prev = NULL; | |
1da177e4 LT |
4846 | } |
4847 | ||
3365495c | 4848 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
8dc07fdb | 4849 | skb->tc_at_ingress = 1; |
46209401 | 4850 | mini_qdisc_bstats_cpu_update(miniq, skb); |
c9e99fd0 | 4851 | |
7d17c544 PB |
4852 | switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list, |
4853 | &cl_res, false)) { | |
d2788d34 DB |
4854 | case TC_ACT_OK: |
4855 | case TC_ACT_RECLASSIFY: | |
4856 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
4857 | break; | |
4858 | case TC_ACT_SHOT: | |
46209401 | 4859 | mini_qdisc_qstats_cpu_drop(miniq); |
8a3a4c6e ED |
4860 | kfree_skb(skb); |
4861 | return NULL; | |
d2788d34 DB |
4862 | case TC_ACT_STOLEN: |
4863 | case TC_ACT_QUEUED: | |
e25ea21f | 4864 | case TC_ACT_TRAP: |
8a3a4c6e | 4865 | consume_skb(skb); |
d2788d34 | 4866 | return NULL; |
27b29f63 AS |
4867 | case TC_ACT_REDIRECT: |
4868 | /* skb_mac_header check was done by cls/act_bpf, so | |
4869 | * we can safely push the L2 header back before | |
4870 | * redirecting to another netdev | |
4871 | */ | |
4872 | __skb_push(skb, skb->mac_len); | |
4873 | skb_do_redirect(skb); | |
4874 | return NULL; | |
720f22fe | 4875 | case TC_ACT_CONSUMED: |
cd11b164 | 4876 | return NULL; |
d2788d34 DB |
4877 | default: |
4878 | break; | |
f697c3e8 | 4879 | } |
e7582bab | 4880 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
4881 | return skb; |
4882 | } | |
1da177e4 | 4883 | |
24b27fc4 MB |
4884 | /** |
4885 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
4886 | * @dev: device to check | |
4887 | * | |
4888 | * Check if a receive handler is already registered for a given device. | |
4889 | * Return true if there one. | |
4890 | * | |
4891 | * The caller must hold the rtnl_mutex. | |
4892 | */ | |
4893 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
4894 | { | |
4895 | ASSERT_RTNL(); | |
4896 | return dev && rtnl_dereference(dev->rx_handler); | |
4897 | } | |
4898 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
4899 | ||
ab95bfe0 JP |
4900 | /** |
4901 | * netdev_rx_handler_register - register receive handler | |
4902 | * @dev: device to register a handler for | |
4903 | * @rx_handler: receive handler to register | |
93e2c32b | 4904 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 4905 | * |
e227867f | 4906 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
4907 | * called from __netif_receive_skb. A negative errno code is returned |
4908 | * on a failure. | |
4909 | * | |
4910 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
4911 | * |
4912 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
4913 | */ |
4914 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
4915 | rx_handler_func_t *rx_handler, |
4916 | void *rx_handler_data) | |
ab95bfe0 | 4917 | { |
1b7cd004 | 4918 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
4919 | return -EBUSY; |
4920 | ||
f5426250 PA |
4921 | if (dev->priv_flags & IFF_NO_RX_HANDLER) |
4922 | return -EINVAL; | |
4923 | ||
00cfec37 | 4924 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 4925 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
4926 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
4927 | ||
4928 | return 0; | |
4929 | } | |
4930 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
4931 | ||
4932 | /** | |
4933 | * netdev_rx_handler_unregister - unregister receive handler | |
4934 | * @dev: device to unregister a handler from | |
4935 | * | |
166ec369 | 4936 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
4937 | * |
4938 | * The caller must hold the rtnl_mutex. | |
4939 | */ | |
4940 | void netdev_rx_handler_unregister(struct net_device *dev) | |
4941 | { | |
4942 | ||
4943 | ASSERT_RTNL(); | |
a9b3cd7f | 4944 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
4945 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
4946 | * section has a guarantee to see a non NULL rx_handler_data | |
4947 | * as well. | |
4948 | */ | |
4949 | synchronize_net(); | |
a9b3cd7f | 4950 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
4951 | } |
4952 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
4953 | ||
b4b9e355 MG |
4954 | /* |
4955 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
4956 | * the special handling of PFMEMALLOC skbs. | |
4957 | */ | |
4958 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
4959 | { | |
4960 | switch (skb->protocol) { | |
2b8837ae JP |
4961 | case htons(ETH_P_ARP): |
4962 | case htons(ETH_P_IP): | |
4963 | case htons(ETH_P_IPV6): | |
4964 | case htons(ETH_P_8021Q): | |
4965 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
4966 | return true; |
4967 | default: | |
4968 | return false; | |
4969 | } | |
4970 | } | |
4971 | ||
e687ad60 PN |
4972 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
4973 | int *ret, struct net_device *orig_dev) | |
4974 | { | |
4975 | if (nf_hook_ingress_active(skb)) { | |
2c1e2703 AC |
4976 | int ingress_retval; |
4977 | ||
e687ad60 PN |
4978 | if (*pt_prev) { |
4979 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4980 | *pt_prev = NULL; | |
4981 | } | |
4982 | ||
2c1e2703 AC |
4983 | rcu_read_lock(); |
4984 | ingress_retval = nf_hook_ingress(skb); | |
4985 | rcu_read_unlock(); | |
4986 | return ingress_retval; | |
e687ad60 PN |
4987 | } |
4988 | return 0; | |
4989 | } | |
e687ad60 | 4990 | |
88eb1944 EC |
4991 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, |
4992 | struct packet_type **ppt_prev) | |
1da177e4 LT |
4993 | { |
4994 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 4995 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 4996 | struct net_device *orig_dev; |
8a4eb573 | 4997 | bool deliver_exact = false; |
1da177e4 | 4998 | int ret = NET_RX_DROP; |
252e3346 | 4999 | __be16 type; |
1da177e4 | 5000 | |
588f0330 | 5001 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 5002 | |
cf66ba58 | 5003 | trace_netif_receive_skb(skb); |
9b22ea56 | 5004 | |
cc9bd5ce | 5005 | orig_dev = skb->dev; |
8f903c70 | 5006 | |
c1d2bbe1 | 5007 | skb_reset_network_header(skb); |
fda55eca ED |
5008 | if (!skb_transport_header_was_set(skb)) |
5009 | skb_reset_transport_header(skb); | |
0b5c9db1 | 5010 | skb_reset_mac_len(skb); |
1da177e4 LT |
5011 | |
5012 | pt_prev = NULL; | |
5013 | ||
63d8ea7f | 5014 | another_round: |
b6858177 | 5015 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
5016 | |
5017 | __this_cpu_inc(softnet_data.processed); | |
5018 | ||
458bf2f2 SH |
5019 | if (static_branch_unlikely(&generic_xdp_needed_key)) { |
5020 | int ret2; | |
5021 | ||
5022 | preempt_disable(); | |
5023 | ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
5024 | preempt_enable(); | |
5025 | ||
5026 | if (ret2 != XDP_PASS) | |
5027 | return NET_RX_DROP; | |
5028 | skb_reset_mac_len(skb); | |
5029 | } | |
5030 | ||
8ad227ff PM |
5031 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
5032 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 5033 | skb = skb_vlan_untag(skb); |
bcc6d479 | 5034 | if (unlikely(!skb)) |
2c17d27c | 5035 | goto out; |
bcc6d479 JP |
5036 | } |
5037 | ||
e7246e12 WB |
5038 | if (skb_skip_tc_classify(skb)) |
5039 | goto skip_classify; | |
1da177e4 | 5040 | |
9754e293 | 5041 | if (pfmemalloc) |
b4b9e355 MG |
5042 | goto skip_taps; |
5043 | ||
1da177e4 | 5044 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
5045 | if (pt_prev) |
5046 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5047 | pt_prev = ptype; | |
5048 | } | |
5049 | ||
5050 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
5051 | if (pt_prev) | |
5052 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5053 | pt_prev = ptype; | |
1da177e4 LT |
5054 | } |
5055 | ||
b4b9e355 | 5056 | skip_taps: |
1cf51900 | 5057 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 5058 | if (static_branch_unlikely(&ingress_needed_key)) { |
1f211a1b | 5059 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); |
4577139b | 5060 | if (!skb) |
2c17d27c | 5061 | goto out; |
e687ad60 PN |
5062 | |
5063 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 5064 | goto out; |
4577139b | 5065 | } |
1cf51900 | 5066 | #endif |
2c64605b | 5067 | skb_reset_redirect(skb); |
e7246e12 | 5068 | skip_classify: |
9754e293 | 5069 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
5070 | goto drop; |
5071 | ||
df8a39de | 5072 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
5073 | if (pt_prev) { |
5074 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5075 | pt_prev = NULL; | |
5076 | } | |
48cc32d3 | 5077 | if (vlan_do_receive(&skb)) |
2425717b JF |
5078 | goto another_round; |
5079 | else if (unlikely(!skb)) | |
2c17d27c | 5080 | goto out; |
2425717b JF |
5081 | } |
5082 | ||
48cc32d3 | 5083 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
5084 | if (rx_handler) { |
5085 | if (pt_prev) { | |
5086 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5087 | pt_prev = NULL; | |
5088 | } | |
8a4eb573 JP |
5089 | switch (rx_handler(&skb)) { |
5090 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 5091 | ret = NET_RX_SUCCESS; |
2c17d27c | 5092 | goto out; |
8a4eb573 | 5093 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 5094 | goto another_round; |
8a4eb573 JP |
5095 | case RX_HANDLER_EXACT: |
5096 | deliver_exact = true; | |
5097 | case RX_HANDLER_PASS: | |
5098 | break; | |
5099 | default: | |
5100 | BUG(); | |
5101 | } | |
ab95bfe0 | 5102 | } |
1da177e4 | 5103 | |
df8a39de | 5104 | if (unlikely(skb_vlan_tag_present(skb))) { |
36b2f61a GV |
5105 | check_vlan_id: |
5106 | if (skb_vlan_tag_get_id(skb)) { | |
5107 | /* Vlan id is non 0 and vlan_do_receive() above couldn't | |
5108 | * find vlan device. | |
5109 | */ | |
d4b812de | 5110 | skb->pkt_type = PACKET_OTHERHOST; |
36b2f61a GV |
5111 | } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
5112 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
5113 | /* Outer header is 802.1P with vlan 0, inner header is | |
5114 | * 802.1Q or 802.1AD and vlan_do_receive() above could | |
5115 | * not find vlan dev for vlan id 0. | |
5116 | */ | |
5117 | __vlan_hwaccel_clear_tag(skb); | |
5118 | skb = skb_vlan_untag(skb); | |
5119 | if (unlikely(!skb)) | |
5120 | goto out; | |
5121 | if (vlan_do_receive(&skb)) | |
5122 | /* After stripping off 802.1P header with vlan 0 | |
5123 | * vlan dev is found for inner header. | |
5124 | */ | |
5125 | goto another_round; | |
5126 | else if (unlikely(!skb)) | |
5127 | goto out; | |
5128 | else | |
5129 | /* We have stripped outer 802.1P vlan 0 header. | |
5130 | * But could not find vlan dev. | |
5131 | * check again for vlan id to set OTHERHOST. | |
5132 | */ | |
5133 | goto check_vlan_id; | |
5134 | } | |
d4b812de ED |
5135 | /* Note: we might in the future use prio bits |
5136 | * and set skb->priority like in vlan_do_receive() | |
5137 | * For the time being, just ignore Priority Code Point | |
5138 | */ | |
b1817524 | 5139 | __vlan_hwaccel_clear_tag(skb); |
d4b812de | 5140 | } |
48cc32d3 | 5141 | |
7866a621 SN |
5142 | type = skb->protocol; |
5143 | ||
63d8ea7f | 5144 | /* deliver only exact match when indicated */ |
7866a621 SN |
5145 | if (likely(!deliver_exact)) { |
5146 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
5147 | &ptype_base[ntohs(type) & | |
5148 | PTYPE_HASH_MASK]); | |
5149 | } | |
1f3c8804 | 5150 | |
7866a621 SN |
5151 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
5152 | &orig_dev->ptype_specific); | |
5153 | ||
5154 | if (unlikely(skb->dev != orig_dev)) { | |
5155 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
5156 | &skb->dev->ptype_specific); | |
1da177e4 LT |
5157 | } |
5158 | ||
5159 | if (pt_prev) { | |
1f8b977a | 5160 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
0e698bf6 | 5161 | goto drop; |
88eb1944 | 5162 | *ppt_prev = pt_prev; |
1da177e4 | 5163 | } else { |
b4b9e355 | 5164 | drop: |
6e7333d3 JW |
5165 | if (!deliver_exact) |
5166 | atomic_long_inc(&skb->dev->rx_dropped); | |
5167 | else | |
5168 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
5169 | kfree_skb(skb); |
5170 | /* Jamal, now you will not able to escape explaining | |
5171 | * me how you were going to use this. :-) | |
5172 | */ | |
5173 | ret = NET_RX_DROP; | |
5174 | } | |
5175 | ||
2c17d27c | 5176 | out: |
9754e293 DM |
5177 | return ret; |
5178 | } | |
5179 | ||
88eb1944 EC |
5180 | static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) |
5181 | { | |
5182 | struct net_device *orig_dev = skb->dev; | |
5183 | struct packet_type *pt_prev = NULL; | |
5184 | int ret; | |
5185 | ||
5186 | ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev); | |
5187 | if (pt_prev) | |
f5737cba PA |
5188 | ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, |
5189 | skb->dev, pt_prev, orig_dev); | |
88eb1944 EC |
5190 | return ret; |
5191 | } | |
5192 | ||
1c601d82 JDB |
5193 | /** |
5194 | * netif_receive_skb_core - special purpose version of netif_receive_skb | |
5195 | * @skb: buffer to process | |
5196 | * | |
5197 | * More direct receive version of netif_receive_skb(). It should | |
5198 | * only be used by callers that have a need to skip RPS and Generic XDP. | |
2de9780f | 5199 | * Caller must also take care of handling if ``(page_is_)pfmemalloc``. |
1c601d82 JDB |
5200 | * |
5201 | * This function may only be called from softirq context and interrupts | |
5202 | * should be enabled. | |
5203 | * | |
5204 | * Return values (usually ignored): | |
5205 | * NET_RX_SUCCESS: no congestion | |
5206 | * NET_RX_DROP: packet was dropped | |
5207 | */ | |
5208 | int netif_receive_skb_core(struct sk_buff *skb) | |
5209 | { | |
5210 | int ret; | |
5211 | ||
5212 | rcu_read_lock(); | |
88eb1944 | 5213 | ret = __netif_receive_skb_one_core(skb, false); |
1c601d82 JDB |
5214 | rcu_read_unlock(); |
5215 | ||
5216 | return ret; | |
5217 | } | |
5218 | EXPORT_SYMBOL(netif_receive_skb_core); | |
5219 | ||
88eb1944 EC |
5220 | static inline void __netif_receive_skb_list_ptype(struct list_head *head, |
5221 | struct packet_type *pt_prev, | |
5222 | struct net_device *orig_dev) | |
4ce0017a EC |
5223 | { |
5224 | struct sk_buff *skb, *next; | |
5225 | ||
88eb1944 EC |
5226 | if (!pt_prev) |
5227 | return; | |
5228 | if (list_empty(head)) | |
5229 | return; | |
17266ee9 | 5230 | if (pt_prev->list_func != NULL) |
fdf71426 PA |
5231 | INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, |
5232 | ip_list_rcv, head, pt_prev, orig_dev); | |
17266ee9 | 5233 | else |
9a5a90d1 AL |
5234 | list_for_each_entry_safe(skb, next, head, list) { |
5235 | skb_list_del_init(skb); | |
fdf71426 | 5236 | pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
9a5a90d1 | 5237 | } |
88eb1944 EC |
5238 | } |
5239 | ||
5240 | static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) | |
5241 | { | |
5242 | /* Fast-path assumptions: | |
5243 | * - There is no RX handler. | |
5244 | * - Only one packet_type matches. | |
5245 | * If either of these fails, we will end up doing some per-packet | |
5246 | * processing in-line, then handling the 'last ptype' for the whole | |
5247 | * sublist. This can't cause out-of-order delivery to any single ptype, | |
5248 | * because the 'last ptype' must be constant across the sublist, and all | |
5249 | * other ptypes are handled per-packet. | |
5250 | */ | |
5251 | /* Current (common) ptype of sublist */ | |
5252 | struct packet_type *pt_curr = NULL; | |
5253 | /* Current (common) orig_dev of sublist */ | |
5254 | struct net_device *od_curr = NULL; | |
5255 | struct list_head sublist; | |
5256 | struct sk_buff *skb, *next; | |
5257 | ||
9af86f93 | 5258 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5259 | list_for_each_entry_safe(skb, next, head, list) { |
5260 | struct net_device *orig_dev = skb->dev; | |
5261 | struct packet_type *pt_prev = NULL; | |
5262 | ||
22f6bbb7 | 5263 | skb_list_del_init(skb); |
88eb1944 | 5264 | __netif_receive_skb_core(skb, pfmemalloc, &pt_prev); |
9af86f93 EC |
5265 | if (!pt_prev) |
5266 | continue; | |
88eb1944 EC |
5267 | if (pt_curr != pt_prev || od_curr != orig_dev) { |
5268 | /* dispatch old sublist */ | |
88eb1944 EC |
5269 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
5270 | /* start new sublist */ | |
9af86f93 | 5271 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5272 | pt_curr = pt_prev; |
5273 | od_curr = orig_dev; | |
5274 | } | |
9af86f93 | 5275 | list_add_tail(&skb->list, &sublist); |
88eb1944 EC |
5276 | } |
5277 | ||
5278 | /* dispatch final sublist */ | |
9af86f93 | 5279 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
4ce0017a EC |
5280 | } |
5281 | ||
9754e293 DM |
5282 | static int __netif_receive_skb(struct sk_buff *skb) |
5283 | { | |
5284 | int ret; | |
5285 | ||
5286 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 5287 | unsigned int noreclaim_flag; |
9754e293 DM |
5288 | |
5289 | /* | |
5290 | * PFMEMALLOC skbs are special, they should | |
5291 | * - be delivered to SOCK_MEMALLOC sockets only | |
5292 | * - stay away from userspace | |
5293 | * - have bounded memory usage | |
5294 | * | |
5295 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
5296 | * context down to all allocation sites. | |
5297 | */ | |
f1083048 | 5298 | noreclaim_flag = memalloc_noreclaim_save(); |
88eb1944 | 5299 | ret = __netif_receive_skb_one_core(skb, true); |
f1083048 | 5300 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 | 5301 | } else |
88eb1944 | 5302 | ret = __netif_receive_skb_one_core(skb, false); |
9754e293 | 5303 | |
1da177e4 LT |
5304 | return ret; |
5305 | } | |
0a9627f2 | 5306 | |
4ce0017a EC |
5307 | static void __netif_receive_skb_list(struct list_head *head) |
5308 | { | |
5309 | unsigned long noreclaim_flag = 0; | |
5310 | struct sk_buff *skb, *next; | |
5311 | bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ | |
5312 | ||
5313 | list_for_each_entry_safe(skb, next, head, list) { | |
5314 | if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { | |
5315 | struct list_head sublist; | |
5316 | ||
5317 | /* Handle the previous sublist */ | |
5318 | list_cut_before(&sublist, head, &skb->list); | |
b9f463d6 EC |
5319 | if (!list_empty(&sublist)) |
5320 | __netif_receive_skb_list_core(&sublist, pfmemalloc); | |
4ce0017a EC |
5321 | pfmemalloc = !pfmemalloc; |
5322 | /* See comments in __netif_receive_skb */ | |
5323 | if (pfmemalloc) | |
5324 | noreclaim_flag = memalloc_noreclaim_save(); | |
5325 | else | |
5326 | memalloc_noreclaim_restore(noreclaim_flag); | |
5327 | } | |
5328 | } | |
5329 | /* Handle the remaining sublist */ | |
b9f463d6 EC |
5330 | if (!list_empty(head)) |
5331 | __netif_receive_skb_list_core(head, pfmemalloc); | |
4ce0017a EC |
5332 | /* Restore pflags */ |
5333 | if (pfmemalloc) | |
5334 | memalloc_noreclaim_restore(noreclaim_flag); | |
5335 | } | |
5336 | ||
f4e63525 | 5337 | static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) |
b5cdae32 | 5338 | { |
58038695 | 5339 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); |
b5cdae32 DM |
5340 | struct bpf_prog *new = xdp->prog; |
5341 | int ret = 0; | |
5342 | ||
5343 | switch (xdp->command) { | |
58038695 | 5344 | case XDP_SETUP_PROG: |
b5cdae32 DM |
5345 | rcu_assign_pointer(dev->xdp_prog, new); |
5346 | if (old) | |
5347 | bpf_prog_put(old); | |
5348 | ||
5349 | if (old && !new) { | |
02786475 | 5350 | static_branch_dec(&generic_xdp_needed_key); |
b5cdae32 | 5351 | } else if (new && !old) { |
02786475 | 5352 | static_branch_inc(&generic_xdp_needed_key); |
b5cdae32 | 5353 | dev_disable_lro(dev); |
56f5aa77 | 5354 | dev_disable_gro_hw(dev); |
b5cdae32 DM |
5355 | } |
5356 | break; | |
b5cdae32 DM |
5357 | |
5358 | case XDP_QUERY_PROG: | |
58038695 | 5359 | xdp->prog_id = old ? old->aux->id : 0; |
b5cdae32 DM |
5360 | break; |
5361 | ||
5362 | default: | |
5363 | ret = -EINVAL; | |
5364 | break; | |
5365 | } | |
5366 | ||
5367 | return ret; | |
5368 | } | |
5369 | ||
ae78dbfa | 5370 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 5371 | { |
2c17d27c JA |
5372 | int ret; |
5373 | ||
588f0330 | 5374 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 5375 | |
c1f19b51 RC |
5376 | if (skb_defer_rx_timestamp(skb)) |
5377 | return NET_RX_SUCCESS; | |
5378 | ||
bbbe211c | 5379 | rcu_read_lock(); |
df334545 | 5380 | #ifdef CONFIG_RPS |
dc05360f | 5381 | if (static_branch_unlikely(&rps_needed)) { |
3b098e2d | 5382 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 5383 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 5384 | |
3b098e2d ED |
5385 | if (cpu >= 0) { |
5386 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
5387 | rcu_read_unlock(); | |
adc9300e | 5388 | return ret; |
3b098e2d | 5389 | } |
fec5e652 | 5390 | } |
1e94d72f | 5391 | #endif |
2c17d27c JA |
5392 | ret = __netif_receive_skb(skb); |
5393 | rcu_read_unlock(); | |
5394 | return ret; | |
0a9627f2 | 5395 | } |
ae78dbfa | 5396 | |
7da517a3 EC |
5397 | static void netif_receive_skb_list_internal(struct list_head *head) |
5398 | { | |
7da517a3 | 5399 | struct sk_buff *skb, *next; |
8c057efa | 5400 | struct list_head sublist; |
7da517a3 | 5401 | |
8c057efa | 5402 | INIT_LIST_HEAD(&sublist); |
7da517a3 EC |
5403 | list_for_each_entry_safe(skb, next, head, list) { |
5404 | net_timestamp_check(netdev_tstamp_prequeue, skb); | |
22f6bbb7 | 5405 | skb_list_del_init(skb); |
8c057efa EC |
5406 | if (!skb_defer_rx_timestamp(skb)) |
5407 | list_add_tail(&skb->list, &sublist); | |
7da517a3 | 5408 | } |
8c057efa | 5409 | list_splice_init(&sublist, head); |
7da517a3 | 5410 | |
7da517a3 EC |
5411 | rcu_read_lock(); |
5412 | #ifdef CONFIG_RPS | |
dc05360f | 5413 | if (static_branch_unlikely(&rps_needed)) { |
7da517a3 EC |
5414 | list_for_each_entry_safe(skb, next, head, list) { |
5415 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
5416 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
5417 | ||
5418 | if (cpu >= 0) { | |
8c057efa | 5419 | /* Will be handled, remove from list */ |
22f6bbb7 | 5420 | skb_list_del_init(skb); |
8c057efa | 5421 | enqueue_to_backlog(skb, cpu, &rflow->last_qtail); |
7da517a3 EC |
5422 | } |
5423 | } | |
5424 | } | |
5425 | #endif | |
5426 | __netif_receive_skb_list(head); | |
5427 | rcu_read_unlock(); | |
5428 | } | |
5429 | ||
ae78dbfa BH |
5430 | /** |
5431 | * netif_receive_skb - process receive buffer from network | |
5432 | * @skb: buffer to process | |
5433 | * | |
5434 | * netif_receive_skb() is the main receive data processing function. | |
5435 | * It always succeeds. The buffer may be dropped during processing | |
5436 | * for congestion control or by the protocol layers. | |
5437 | * | |
5438 | * This function may only be called from softirq context and interrupts | |
5439 | * should be enabled. | |
5440 | * | |
5441 | * Return values (usually ignored): | |
5442 | * NET_RX_SUCCESS: no congestion | |
5443 | * NET_RX_DROP: packet was dropped | |
5444 | */ | |
04eb4489 | 5445 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa | 5446 | { |
b0e3f1bd GB |
5447 | int ret; |
5448 | ||
ae78dbfa BH |
5449 | trace_netif_receive_skb_entry(skb); |
5450 | ||
b0e3f1bd GB |
5451 | ret = netif_receive_skb_internal(skb); |
5452 | trace_netif_receive_skb_exit(ret); | |
5453 | ||
5454 | return ret; | |
ae78dbfa | 5455 | } |
04eb4489 | 5456 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 5457 | |
f6ad8c1b EC |
5458 | /** |
5459 | * netif_receive_skb_list - process many receive buffers from network | |
5460 | * @head: list of skbs to process. | |
5461 | * | |
7da517a3 EC |
5462 | * Since return value of netif_receive_skb() is normally ignored, and |
5463 | * wouldn't be meaningful for a list, this function returns void. | |
f6ad8c1b EC |
5464 | * |
5465 | * This function may only be called from softirq context and interrupts | |
5466 | * should be enabled. | |
5467 | */ | |
5468 | void netif_receive_skb_list(struct list_head *head) | |
5469 | { | |
7da517a3 | 5470 | struct sk_buff *skb; |
f6ad8c1b | 5471 | |
b9f463d6 EC |
5472 | if (list_empty(head)) |
5473 | return; | |
b0e3f1bd GB |
5474 | if (trace_netif_receive_skb_list_entry_enabled()) { |
5475 | list_for_each_entry(skb, head, list) | |
5476 | trace_netif_receive_skb_list_entry(skb); | |
5477 | } | |
7da517a3 | 5478 | netif_receive_skb_list_internal(head); |
b0e3f1bd | 5479 | trace_netif_receive_skb_list_exit(0); |
f6ad8c1b EC |
5480 | } |
5481 | EXPORT_SYMBOL(netif_receive_skb_list); | |
5482 | ||
41852497 | 5483 | DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
5484 | |
5485 | /* Network device is going away, flush any packets still pending */ | |
5486 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 5487 | { |
6e583ce5 | 5488 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
5489 | struct softnet_data *sd; |
5490 | ||
5491 | local_bh_disable(); | |
5492 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 5493 | |
145dd5f9 | 5494 | local_irq_disable(); |
e36fa2f7 | 5495 | rps_lock(sd); |
6e7676c1 | 5496 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 5497 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 5498 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 5499 | kfree_skb(skb); |
76cc8b13 | 5500 | input_queue_head_incr(sd); |
6e583ce5 | 5501 | } |
6e7676c1 | 5502 | } |
e36fa2f7 | 5503 | rps_unlock(sd); |
145dd5f9 | 5504 | local_irq_enable(); |
6e7676c1 CG |
5505 | |
5506 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 5507 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
5508 | __skb_unlink(skb, &sd->process_queue); |
5509 | kfree_skb(skb); | |
76cc8b13 | 5510 | input_queue_head_incr(sd); |
6e7676c1 CG |
5511 | } |
5512 | } | |
145dd5f9 PA |
5513 | local_bh_enable(); |
5514 | } | |
5515 | ||
41852497 | 5516 | static void flush_all_backlogs(void) |
145dd5f9 PA |
5517 | { |
5518 | unsigned int cpu; | |
5519 | ||
5520 | get_online_cpus(); | |
5521 | ||
41852497 ED |
5522 | for_each_online_cpu(cpu) |
5523 | queue_work_on(cpu, system_highpri_wq, | |
5524 | per_cpu_ptr(&flush_works, cpu)); | |
145dd5f9 PA |
5525 | |
5526 | for_each_online_cpu(cpu) | |
41852497 | 5527 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 PA |
5528 | |
5529 | put_online_cpus(); | |
6e583ce5 SH |
5530 | } |
5531 | ||
c8079432 MM |
5532 | /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ |
5533 | static void gro_normal_list(struct napi_struct *napi) | |
5534 | { | |
5535 | if (!napi->rx_count) | |
5536 | return; | |
5537 | netif_receive_skb_list_internal(&napi->rx_list); | |
5538 | INIT_LIST_HEAD(&napi->rx_list); | |
5539 | napi->rx_count = 0; | |
5540 | } | |
5541 | ||
5542 | /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, | |
5543 | * pass the whole batch up to the stack. | |
5544 | */ | |
5545 | static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb) | |
5546 | { | |
5547 | list_add_tail(&skb->list, &napi->rx_list); | |
5548 | if (++napi->rx_count >= gro_normal_batch) | |
5549 | gro_normal_list(napi); | |
5550 | } | |
5551 | ||
aaa5d90b PA |
5552 | INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int)); |
5553 | INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int)); | |
c8079432 | 5554 | static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 | 5555 | { |
22061d80 | 5556 | struct packet_offload *ptype; |
d565b0a1 | 5557 | __be16 type = skb->protocol; |
22061d80 | 5558 | struct list_head *head = &offload_base; |
d565b0a1 HX |
5559 | int err = -ENOENT; |
5560 | ||
c3c7c254 ED |
5561 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
5562 | ||
fc59f9a3 HX |
5563 | if (NAPI_GRO_CB(skb)->count == 1) { |
5564 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 5565 | goto out; |
fc59f9a3 | 5566 | } |
d565b0a1 HX |
5567 | |
5568 | rcu_read_lock(); | |
5569 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 5570 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
5571 | continue; |
5572 | ||
aaa5d90b PA |
5573 | err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, |
5574 | ipv6_gro_complete, inet_gro_complete, | |
5575 | skb, 0); | |
d565b0a1 HX |
5576 | break; |
5577 | } | |
5578 | rcu_read_unlock(); | |
5579 | ||
5580 | if (err) { | |
5581 | WARN_ON(&ptype->list == head); | |
5582 | kfree_skb(skb); | |
5583 | return NET_RX_SUCCESS; | |
5584 | } | |
5585 | ||
5586 | out: | |
c8079432 MM |
5587 | gro_normal_one(napi, skb); |
5588 | return NET_RX_SUCCESS; | |
d565b0a1 HX |
5589 | } |
5590 | ||
6312fe77 | 5591 | static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, |
07d78363 | 5592 | bool flush_old) |
d565b0a1 | 5593 | { |
6312fe77 | 5594 | struct list_head *head = &napi->gro_hash[index].list; |
d4546c25 | 5595 | struct sk_buff *skb, *p; |
2e71a6f8 | 5596 | |
07d78363 | 5597 | list_for_each_entry_safe_reverse(skb, p, head, list) { |
2e71a6f8 ED |
5598 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) |
5599 | return; | |
992cba7e | 5600 | skb_list_del_init(skb); |
c8079432 | 5601 | napi_gro_complete(napi, skb); |
6312fe77 | 5602 | napi->gro_hash[index].count--; |
d565b0a1 | 5603 | } |
d9f37d01 LR |
5604 | |
5605 | if (!napi->gro_hash[index].count) | |
5606 | __clear_bit(index, &napi->gro_bitmask); | |
d565b0a1 | 5607 | } |
07d78363 | 5608 | |
6312fe77 | 5609 | /* napi->gro_hash[].list contains packets ordered by age. |
07d78363 DM |
5610 | * youngest packets at the head of it. |
5611 | * Complete skbs in reverse order to reduce latencies. | |
5612 | */ | |
5613 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
5614 | { | |
42519ede ED |
5615 | unsigned long bitmask = napi->gro_bitmask; |
5616 | unsigned int i, base = ~0U; | |
07d78363 | 5617 | |
42519ede ED |
5618 | while ((i = ffs(bitmask)) != 0) { |
5619 | bitmask >>= i; | |
5620 | base += i; | |
5621 | __napi_gro_flush_chain(napi, base, flush_old); | |
d9f37d01 | 5622 | } |
07d78363 | 5623 | } |
86cac58b | 5624 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 5625 | |
07d78363 DM |
5626 | static struct list_head *gro_list_prepare(struct napi_struct *napi, |
5627 | struct sk_buff *skb) | |
89c5fa33 | 5628 | { |
89c5fa33 | 5629 | unsigned int maclen = skb->dev->hard_header_len; |
0b4cec8c | 5630 | u32 hash = skb_get_hash_raw(skb); |
07d78363 | 5631 | struct list_head *head; |
d4546c25 | 5632 | struct sk_buff *p; |
89c5fa33 | 5633 | |
6312fe77 | 5634 | head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list; |
07d78363 | 5635 | list_for_each_entry(p, head, list) { |
89c5fa33 ED |
5636 | unsigned long diffs; |
5637 | ||
0b4cec8c TH |
5638 | NAPI_GRO_CB(p)->flush = 0; |
5639 | ||
5640 | if (hash != skb_get_hash_raw(p)) { | |
5641 | NAPI_GRO_CB(p)->same_flow = 0; | |
5642 | continue; | |
5643 | } | |
5644 | ||
89c5fa33 | 5645 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
b1817524 MM |
5646 | diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); |
5647 | if (skb_vlan_tag_present(p)) | |
fc5141cb | 5648 | diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); |
ce87fc6c | 5649 | diffs |= skb_metadata_dst_cmp(p, skb); |
de8f3a83 | 5650 | diffs |= skb_metadata_differs(p, skb); |
89c5fa33 ED |
5651 | if (maclen == ETH_HLEN) |
5652 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 5653 | skb_mac_header(skb)); |
89c5fa33 ED |
5654 | else if (!diffs) |
5655 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 5656 | skb_mac_header(skb), |
89c5fa33 ED |
5657 | maclen); |
5658 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 | 5659 | } |
07d78363 DM |
5660 | |
5661 | return head; | |
89c5fa33 ED |
5662 | } |
5663 | ||
299603e8 JC |
5664 | static void skb_gro_reset_offset(struct sk_buff *skb) |
5665 | { | |
5666 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
5667 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
5668 | ||
5669 | NAPI_GRO_CB(skb)->data_offset = 0; | |
5670 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
5671 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
5672 | ||
8aef998d | 5673 | if (!skb_headlen(skb) && pinfo->nr_frags && |
299603e8 JC |
5674 | !PageHighMem(skb_frag_page(frag0))) { |
5675 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
7cfd5fd5 ED |
5676 | NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, |
5677 | skb_frag_size(frag0), | |
5678 | skb->end - skb->tail); | |
89c5fa33 ED |
5679 | } |
5680 | } | |
5681 | ||
a50e233c ED |
5682 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
5683 | { | |
5684 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
5685 | ||
5686 | BUG_ON(skb->end - skb->tail < grow); | |
5687 | ||
5688 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
5689 | ||
5690 | skb->data_len -= grow; | |
5691 | skb->tail += grow; | |
5692 | ||
b54c9d5b | 5693 | skb_frag_off_add(&pinfo->frags[0], grow); |
a50e233c ED |
5694 | skb_frag_size_sub(&pinfo->frags[0], grow); |
5695 | ||
5696 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
5697 | skb_frag_unref(skb, 0); | |
5698 | memmove(pinfo->frags, pinfo->frags + 1, | |
5699 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
5700 | } | |
5701 | } | |
5702 | ||
c8079432 | 5703 | static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) |
07d78363 | 5704 | { |
6312fe77 | 5705 | struct sk_buff *oldest; |
07d78363 | 5706 | |
6312fe77 | 5707 | oldest = list_last_entry(head, struct sk_buff, list); |
07d78363 | 5708 | |
6312fe77 | 5709 | /* We are called with head length >= MAX_GRO_SKBS, so this is |
07d78363 DM |
5710 | * impossible. |
5711 | */ | |
5712 | if (WARN_ON_ONCE(!oldest)) | |
5713 | return; | |
5714 | ||
d9f37d01 LR |
5715 | /* Do not adjust napi->gro_hash[].count, caller is adding a new |
5716 | * SKB to the chain. | |
07d78363 | 5717 | */ |
ece23711 | 5718 | skb_list_del_init(oldest); |
c8079432 | 5719 | napi_gro_complete(napi, oldest); |
07d78363 DM |
5720 | } |
5721 | ||
aaa5d90b PA |
5722 | INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *, |
5723 | struct sk_buff *)); | |
5724 | INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *, | |
5725 | struct sk_buff *)); | |
bb728820 | 5726 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 | 5727 | { |
6312fe77 | 5728 | u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); |
d4546c25 | 5729 | struct list_head *head = &offload_base; |
22061d80 | 5730 | struct packet_offload *ptype; |
d565b0a1 | 5731 | __be16 type = skb->protocol; |
07d78363 | 5732 | struct list_head *gro_head; |
d4546c25 | 5733 | struct sk_buff *pp = NULL; |
5b252f0c | 5734 | enum gro_result ret; |
d4546c25 | 5735 | int same_flow; |
a50e233c | 5736 | int grow; |
d565b0a1 | 5737 | |
b5cdae32 | 5738 | if (netif_elide_gro(skb->dev)) |
d565b0a1 HX |
5739 | goto normal; |
5740 | ||
07d78363 | 5741 | gro_head = gro_list_prepare(napi, skb); |
89c5fa33 | 5742 | |
d565b0a1 HX |
5743 | rcu_read_lock(); |
5744 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 5745 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
5746 | continue; |
5747 | ||
86911732 | 5748 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 5749 | skb_reset_mac_len(skb); |
d565b0a1 | 5750 | NAPI_GRO_CB(skb)->same_flow = 0; |
d61d072e | 5751 | NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); |
5d38a079 | 5752 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 5753 | NAPI_GRO_CB(skb)->encap_mark = 0; |
fcd91dd4 | 5754 | NAPI_GRO_CB(skb)->recursion_counter = 0; |
a0ca153f | 5755 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 5756 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 5757 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 5758 | |
662880f4 TH |
5759 | /* Setup for GRO checksum validation */ |
5760 | switch (skb->ip_summed) { | |
5761 | case CHECKSUM_COMPLETE: | |
5762 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
5763 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
5764 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
5765 | break; | |
5766 | case CHECKSUM_UNNECESSARY: | |
5767 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
5768 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
5769 | break; | |
5770 | default: | |
5771 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
5772 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
5773 | } | |
d565b0a1 | 5774 | |
aaa5d90b PA |
5775 | pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, |
5776 | ipv6_gro_receive, inet_gro_receive, | |
5777 | gro_head, skb); | |
d565b0a1 HX |
5778 | break; |
5779 | } | |
5780 | rcu_read_unlock(); | |
5781 | ||
5782 | if (&ptype->list == head) | |
5783 | goto normal; | |
5784 | ||
45586c70 | 5785 | if (PTR_ERR(pp) == -EINPROGRESS) { |
25393d3f SK |
5786 | ret = GRO_CONSUMED; |
5787 | goto ok; | |
5788 | } | |
5789 | ||
0da2afd5 | 5790 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 5791 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 5792 | |
d565b0a1 | 5793 | if (pp) { |
992cba7e | 5794 | skb_list_del_init(pp); |
c8079432 | 5795 | napi_gro_complete(napi, pp); |
6312fe77 | 5796 | napi->gro_hash[hash].count--; |
d565b0a1 HX |
5797 | } |
5798 | ||
0da2afd5 | 5799 | if (same_flow) |
d565b0a1 HX |
5800 | goto ok; |
5801 | ||
600adc18 | 5802 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 5803 | goto normal; |
d565b0a1 | 5804 | |
6312fe77 | 5805 | if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) { |
c8079432 | 5806 | gro_flush_oldest(napi, gro_head); |
600adc18 | 5807 | } else { |
6312fe77 | 5808 | napi->gro_hash[hash].count++; |
600adc18 | 5809 | } |
d565b0a1 | 5810 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 5811 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 5812 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 5813 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
07d78363 | 5814 | list_add(&skb->list, gro_head); |
5d0d9be8 | 5815 | ret = GRO_HELD; |
d565b0a1 | 5816 | |
ad0f9904 | 5817 | pull: |
a50e233c ED |
5818 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
5819 | if (grow > 0) | |
5820 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 5821 | ok: |
d9f37d01 LR |
5822 | if (napi->gro_hash[hash].count) { |
5823 | if (!test_bit(hash, &napi->gro_bitmask)) | |
5824 | __set_bit(hash, &napi->gro_bitmask); | |
5825 | } else if (test_bit(hash, &napi->gro_bitmask)) { | |
5826 | __clear_bit(hash, &napi->gro_bitmask); | |
5827 | } | |
5828 | ||
5d0d9be8 | 5829 | return ret; |
d565b0a1 HX |
5830 | |
5831 | normal: | |
ad0f9904 HX |
5832 | ret = GRO_NORMAL; |
5833 | goto pull; | |
5d38a079 | 5834 | } |
96e93eab | 5835 | |
bf5a755f JC |
5836 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
5837 | { | |
5838 | struct list_head *offload_head = &offload_base; | |
5839 | struct packet_offload *ptype; | |
5840 | ||
5841 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
5842 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
5843 | continue; | |
5844 | return ptype; | |
5845 | } | |
5846 | return NULL; | |
5847 | } | |
e27a2f83 | 5848 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
5849 | |
5850 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
5851 | { | |
5852 | struct list_head *offload_head = &offload_base; | |
5853 | struct packet_offload *ptype; | |
5854 | ||
5855 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
5856 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
5857 | continue; | |
5858 | return ptype; | |
5859 | } | |
5860 | return NULL; | |
5861 | } | |
e27a2f83 | 5862 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 5863 | |
e44699d2 MK |
5864 | static void napi_skb_free_stolen_head(struct sk_buff *skb) |
5865 | { | |
5866 | skb_dst_drop(skb); | |
174e2381 | 5867 | skb_ext_put(skb); |
e44699d2 MK |
5868 | kmem_cache_free(skbuff_head_cache, skb); |
5869 | } | |
5870 | ||
6570bc79 AL |
5871 | static gro_result_t napi_skb_finish(struct napi_struct *napi, |
5872 | struct sk_buff *skb, | |
5873 | gro_result_t ret) | |
5d38a079 | 5874 | { |
5d0d9be8 HX |
5875 | switch (ret) { |
5876 | case GRO_NORMAL: | |
6570bc79 | 5877 | gro_normal_one(napi, skb); |
c7c4b3b6 | 5878 | break; |
5d38a079 | 5879 | |
5d0d9be8 | 5880 | case GRO_DROP: |
5d38a079 HX |
5881 | kfree_skb(skb); |
5882 | break; | |
5b252f0c | 5883 | |
daa86548 | 5884 | case GRO_MERGED_FREE: |
e44699d2 MK |
5885 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
5886 | napi_skb_free_stolen_head(skb); | |
5887 | else | |
d7e8883c | 5888 | __kfree_skb(skb); |
daa86548 ED |
5889 | break; |
5890 | ||
5b252f0c BH |
5891 | case GRO_HELD: |
5892 | case GRO_MERGED: | |
25393d3f | 5893 | case GRO_CONSUMED: |
5b252f0c | 5894 | break; |
5d38a079 HX |
5895 | } |
5896 | ||
c7c4b3b6 | 5897 | return ret; |
5d0d9be8 | 5898 | } |
5d0d9be8 | 5899 | |
c7c4b3b6 | 5900 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 5901 | { |
b0e3f1bd GB |
5902 | gro_result_t ret; |
5903 | ||
93f93a44 | 5904 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 5905 | trace_napi_gro_receive_entry(skb); |
86911732 | 5906 | |
a50e233c ED |
5907 | skb_gro_reset_offset(skb); |
5908 | ||
6570bc79 | 5909 | ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); |
b0e3f1bd GB |
5910 | trace_napi_gro_receive_exit(ret); |
5911 | ||
5912 | return ret; | |
d565b0a1 HX |
5913 | } |
5914 | EXPORT_SYMBOL(napi_gro_receive); | |
5915 | ||
d0c2b0d2 | 5916 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 5917 | { |
93a35f59 ED |
5918 | if (unlikely(skb->pfmemalloc)) { |
5919 | consume_skb(skb); | |
5920 | return; | |
5921 | } | |
96e93eab | 5922 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
5923 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
5924 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
b1817524 | 5925 | __vlan_hwaccel_clear_tag(skb); |
66c46d74 | 5926 | skb->dev = napi->dev; |
6d152e23 | 5927 | skb->skb_iif = 0; |
33d9a2c7 ED |
5928 | |
5929 | /* eth_type_trans() assumes pkt_type is PACKET_HOST */ | |
5930 | skb->pkt_type = PACKET_HOST; | |
5931 | ||
c3caf119 JC |
5932 | skb->encapsulation = 0; |
5933 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 5934 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
174e2381 | 5935 | skb_ext_reset(skb); |
96e93eab HX |
5936 | |
5937 | napi->skb = skb; | |
5938 | } | |
96e93eab | 5939 | |
76620aaf | 5940 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 5941 | { |
5d38a079 | 5942 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
5943 | |
5944 | if (!skb) { | |
fd11a83d | 5945 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
5946 | if (skb) { |
5947 | napi->skb = skb; | |
5948 | skb_mark_napi_id(skb, napi); | |
5949 | } | |
80595d59 | 5950 | } |
96e93eab HX |
5951 | return skb; |
5952 | } | |
76620aaf | 5953 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 5954 | |
a50e233c ED |
5955 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
5956 | struct sk_buff *skb, | |
5957 | gro_result_t ret) | |
96e93eab | 5958 | { |
5d0d9be8 HX |
5959 | switch (ret) { |
5960 | case GRO_NORMAL: | |
a50e233c ED |
5961 | case GRO_HELD: |
5962 | __skb_push(skb, ETH_HLEN); | |
5963 | skb->protocol = eth_type_trans(skb, skb->dev); | |
323ebb61 EC |
5964 | if (ret == GRO_NORMAL) |
5965 | gro_normal_one(napi, skb); | |
86911732 | 5966 | break; |
5d38a079 | 5967 | |
5d0d9be8 | 5968 | case GRO_DROP: |
5d0d9be8 HX |
5969 | napi_reuse_skb(napi, skb); |
5970 | break; | |
5b252f0c | 5971 | |
e44699d2 MK |
5972 | case GRO_MERGED_FREE: |
5973 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) | |
5974 | napi_skb_free_stolen_head(skb); | |
5975 | else | |
5976 | napi_reuse_skb(napi, skb); | |
5977 | break; | |
5978 | ||
5b252f0c | 5979 | case GRO_MERGED: |
25393d3f | 5980 | case GRO_CONSUMED: |
5b252f0c | 5981 | break; |
5d0d9be8 | 5982 | } |
5d38a079 | 5983 | |
c7c4b3b6 | 5984 | return ret; |
5d38a079 | 5985 | } |
5d0d9be8 | 5986 | |
a50e233c ED |
5987 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
5988 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
5989 | * We copy ethernet header into skb->data to have a common layout. | |
5990 | */ | |
4adb9c4a | 5991 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
5992 | { |
5993 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
5994 | const struct ethhdr *eth; |
5995 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
5996 | |
5997 | napi->skb = NULL; | |
5998 | ||
a50e233c ED |
5999 | skb_reset_mac_header(skb); |
6000 | skb_gro_reset_offset(skb); | |
6001 | ||
a50e233c ED |
6002 | if (unlikely(skb_gro_header_hard(skb, hlen))) { |
6003 | eth = skb_gro_header_slow(skb, hlen, 0); | |
6004 | if (unlikely(!eth)) { | |
4da46ceb AC |
6005 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
6006 | __func__, napi->dev->name); | |
a50e233c ED |
6007 | napi_reuse_skb(napi, skb); |
6008 | return NULL; | |
6009 | } | |
6010 | } else { | |
a4270d67 | 6011 | eth = (const struct ethhdr *)skb->data; |
a50e233c ED |
6012 | gro_pull_from_frag0(skb, hlen); |
6013 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
6014 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 6015 | } |
a50e233c ED |
6016 | __skb_pull(skb, hlen); |
6017 | ||
6018 | /* | |
6019 | * This works because the only protocols we care about don't require | |
6020 | * special handling. | |
6021 | * We'll fix it up properly in napi_frags_finish() | |
6022 | */ | |
6023 | skb->protocol = eth->h_proto; | |
76620aaf | 6024 | |
76620aaf HX |
6025 | return skb; |
6026 | } | |
76620aaf | 6027 | |
c7c4b3b6 | 6028 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 6029 | { |
b0e3f1bd | 6030 | gro_result_t ret; |
76620aaf | 6031 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
6032 | |
6033 | if (!skb) | |
c7c4b3b6 | 6034 | return GRO_DROP; |
5d0d9be8 | 6035 | |
ae78dbfa BH |
6036 | trace_napi_gro_frags_entry(skb); |
6037 | ||
b0e3f1bd GB |
6038 | ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
6039 | trace_napi_gro_frags_exit(ret); | |
6040 | ||
6041 | return ret; | |
5d0d9be8 | 6042 | } |
5d38a079 HX |
6043 | EXPORT_SYMBOL(napi_gro_frags); |
6044 | ||
573e8fca TH |
6045 | /* Compute the checksum from gro_offset and return the folded value |
6046 | * after adding in any pseudo checksum. | |
6047 | */ | |
6048 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
6049 | { | |
6050 | __wsum wsum; | |
6051 | __sum16 sum; | |
6052 | ||
6053 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
6054 | ||
6055 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
6056 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
14641931 | 6057 | /* See comments in __skb_checksum_complete(). */ |
573e8fca TH |
6058 | if (likely(!sum)) { |
6059 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
6060 | !skb->csum_complete_sw) | |
7fe50ac8 | 6061 | netdev_rx_csum_fault(skb->dev, skb); |
573e8fca TH |
6062 | } |
6063 | ||
6064 | NAPI_GRO_CB(skb)->csum = wsum; | |
6065 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
6066 | ||
6067 | return sum; | |
6068 | } | |
6069 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
6070 | ||
773fc8f6 | 6071 | static void net_rps_send_ipi(struct softnet_data *remsd) |
6072 | { | |
6073 | #ifdef CONFIG_RPS | |
6074 | while (remsd) { | |
6075 | struct softnet_data *next = remsd->rps_ipi_next; | |
6076 | ||
6077 | if (cpu_online(remsd->cpu)) | |
6078 | smp_call_function_single_async(remsd->cpu, &remsd->csd); | |
6079 | remsd = next; | |
6080 | } | |
6081 | #endif | |
6082 | } | |
6083 | ||
e326bed2 | 6084 | /* |
855abcf0 | 6085 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
6086 | * Note: called with local irq disabled, but exits with local irq enabled. |
6087 | */ | |
6088 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
6089 | { | |
6090 | #ifdef CONFIG_RPS | |
6091 | struct softnet_data *remsd = sd->rps_ipi_list; | |
6092 | ||
6093 | if (remsd) { | |
6094 | sd->rps_ipi_list = NULL; | |
6095 | ||
6096 | local_irq_enable(); | |
6097 | ||
6098 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
773fc8f6 | 6099 | net_rps_send_ipi(remsd); |
e326bed2 ED |
6100 | } else |
6101 | #endif | |
6102 | local_irq_enable(); | |
6103 | } | |
6104 | ||
d75b1ade ED |
6105 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
6106 | { | |
6107 | #ifdef CONFIG_RPS | |
6108 | return sd->rps_ipi_list != NULL; | |
6109 | #else | |
6110 | return false; | |
6111 | #endif | |
6112 | } | |
6113 | ||
bea3348e | 6114 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 6115 | { |
eecfd7c4 | 6116 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
6117 | bool again = true; |
6118 | int work = 0; | |
1da177e4 | 6119 | |
e326bed2 ED |
6120 | /* Check if we have pending ipi, its better to send them now, |
6121 | * not waiting net_rx_action() end. | |
6122 | */ | |
d75b1ade | 6123 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
6124 | local_irq_disable(); |
6125 | net_rps_action_and_irq_enable(sd); | |
6126 | } | |
d75b1ade | 6127 | |
3d48b53f | 6128 | napi->weight = dev_rx_weight; |
145dd5f9 | 6129 | while (again) { |
1da177e4 | 6130 | struct sk_buff *skb; |
6e7676c1 CG |
6131 | |
6132 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 6133 | rcu_read_lock(); |
6e7676c1 | 6134 | __netif_receive_skb(skb); |
2c17d27c | 6135 | rcu_read_unlock(); |
76cc8b13 | 6136 | input_queue_head_incr(sd); |
145dd5f9 | 6137 | if (++work >= quota) |
76cc8b13 | 6138 | return work; |
145dd5f9 | 6139 | |
6e7676c1 | 6140 | } |
1da177e4 | 6141 | |
145dd5f9 | 6142 | local_irq_disable(); |
e36fa2f7 | 6143 | rps_lock(sd); |
11ef7a89 | 6144 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
6145 | /* |
6146 | * Inline a custom version of __napi_complete(). | |
6147 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
6148 | * and NAPI_STATE_SCHED is the only possible flag set |
6149 | * on backlog. | |
6150 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
6151 | * and we dont need an smp_mb() memory barrier. |
6152 | */ | |
eecfd7c4 | 6153 | napi->state = 0; |
145dd5f9 PA |
6154 | again = false; |
6155 | } else { | |
6156 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
6157 | &sd->process_queue); | |
bea3348e | 6158 | } |
e36fa2f7 | 6159 | rps_unlock(sd); |
145dd5f9 | 6160 | local_irq_enable(); |
6e7676c1 | 6161 | } |
1da177e4 | 6162 | |
bea3348e SH |
6163 | return work; |
6164 | } | |
1da177e4 | 6165 | |
bea3348e SH |
6166 | /** |
6167 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 6168 | * @n: entry to schedule |
bea3348e | 6169 | * |
bc9ad166 ED |
6170 | * The entry's receive function will be scheduled to run. |
6171 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 6172 | */ |
b5606c2d | 6173 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
6174 | { |
6175 | unsigned long flags; | |
1da177e4 | 6176 | |
bea3348e | 6177 | local_irq_save(flags); |
903ceff7 | 6178 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 6179 | local_irq_restore(flags); |
1da177e4 | 6180 | } |
bea3348e SH |
6181 | EXPORT_SYMBOL(__napi_schedule); |
6182 | ||
39e6c820 ED |
6183 | /** |
6184 | * napi_schedule_prep - check if napi can be scheduled | |
6185 | * @n: napi context | |
6186 | * | |
6187 | * Test if NAPI routine is already running, and if not mark | |
6188 | * it as running. This is used as a condition variable | |
6189 | * insure only one NAPI poll instance runs. We also make | |
6190 | * sure there is no pending NAPI disable. | |
6191 | */ | |
6192 | bool napi_schedule_prep(struct napi_struct *n) | |
6193 | { | |
6194 | unsigned long val, new; | |
6195 | ||
6196 | do { | |
6197 | val = READ_ONCE(n->state); | |
6198 | if (unlikely(val & NAPIF_STATE_DISABLE)) | |
6199 | return false; | |
6200 | new = val | NAPIF_STATE_SCHED; | |
6201 | ||
6202 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
6203 | * This was suggested by Alexander Duyck, as compiler | |
6204 | * emits better code than : | |
6205 | * if (val & NAPIF_STATE_SCHED) | |
6206 | * new |= NAPIF_STATE_MISSED; | |
6207 | */ | |
6208 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
6209 | NAPIF_STATE_MISSED; | |
6210 | } while (cmpxchg(&n->state, val, new) != val); | |
6211 | ||
6212 | return !(val & NAPIF_STATE_SCHED); | |
6213 | } | |
6214 | EXPORT_SYMBOL(napi_schedule_prep); | |
6215 | ||
bc9ad166 ED |
6216 | /** |
6217 | * __napi_schedule_irqoff - schedule for receive | |
6218 | * @n: entry to schedule | |
6219 | * | |
6220 | * Variant of __napi_schedule() assuming hard irqs are masked | |
6221 | */ | |
6222 | void __napi_schedule_irqoff(struct napi_struct *n) | |
6223 | { | |
6224 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
6225 | } | |
6226 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
6227 | ||
364b6055 | 6228 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 6229 | { |
6f8b12d6 ED |
6230 | unsigned long flags, val, new, timeout = 0; |
6231 | bool ret = true; | |
d565b0a1 HX |
6232 | |
6233 | /* | |
217f6974 ED |
6234 | * 1) Don't let napi dequeue from the cpu poll list |
6235 | * just in case its running on a different cpu. | |
6236 | * 2) If we are busy polling, do nothing here, we have | |
6237 | * the guarantee we will be called later. | |
d565b0a1 | 6238 | */ |
217f6974 ED |
6239 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
6240 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 6241 | return false; |
d565b0a1 | 6242 | |
6f8b12d6 ED |
6243 | if (work_done) { |
6244 | if (n->gro_bitmask) | |
7e417a66 ED |
6245 | timeout = READ_ONCE(n->dev->gro_flush_timeout); |
6246 | n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); | |
6f8b12d6 ED |
6247 | } |
6248 | if (n->defer_hard_irqs_count > 0) { | |
6249 | n->defer_hard_irqs_count--; | |
7e417a66 | 6250 | timeout = READ_ONCE(n->dev->gro_flush_timeout); |
6f8b12d6 ED |
6251 | if (timeout) |
6252 | ret = false; | |
6253 | } | |
6254 | if (n->gro_bitmask) { | |
605108ac PA |
6255 | /* When the NAPI instance uses a timeout and keeps postponing |
6256 | * it, we need to bound somehow the time packets are kept in | |
6257 | * the GRO layer | |
6258 | */ | |
6259 | napi_gro_flush(n, !!timeout); | |
3b47d303 | 6260 | } |
c8079432 MM |
6261 | |
6262 | gro_normal_list(n); | |
6263 | ||
02c1602e | 6264 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
6265 | /* If n->poll_list is not empty, we need to mask irqs */ |
6266 | local_irq_save(flags); | |
02c1602e | 6267 | list_del_init(&n->poll_list); |
d75b1ade ED |
6268 | local_irq_restore(flags); |
6269 | } | |
39e6c820 ED |
6270 | |
6271 | do { | |
6272 | val = READ_ONCE(n->state); | |
6273 | ||
6274 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); | |
6275 | ||
6276 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); | |
6277 | ||
6278 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
6279 | * because we will call napi->poll() one more time. | |
6280 | * This C code was suggested by Alexander Duyck to help gcc. | |
6281 | */ | |
6282 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
6283 | NAPIF_STATE_SCHED; | |
6284 | } while (cmpxchg(&n->state, val, new) != val); | |
6285 | ||
6286 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
6287 | __napi_schedule(n); | |
6288 | return false; | |
6289 | } | |
6290 | ||
6f8b12d6 ED |
6291 | if (timeout) |
6292 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
6293 | HRTIMER_MODE_REL_PINNED); | |
6294 | return ret; | |
d565b0a1 | 6295 | } |
3b47d303 | 6296 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 6297 | |
af12fa6e | 6298 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 6299 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
6300 | { |
6301 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
6302 | struct napi_struct *napi; | |
6303 | ||
6304 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
6305 | if (napi->napi_id == napi_id) | |
6306 | return napi; | |
6307 | ||
6308 | return NULL; | |
6309 | } | |
02d62e86 ED |
6310 | |
6311 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
217f6974 | 6312 | |
ce6aea93 | 6313 | #define BUSY_POLL_BUDGET 8 |
217f6974 ED |
6314 | |
6315 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) | |
6316 | { | |
6317 | int rc; | |
6318 | ||
39e6c820 ED |
6319 | /* Busy polling means there is a high chance device driver hard irq |
6320 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
6321 | * set in napi_schedule_prep(). | |
6322 | * Since we are about to call napi->poll() once more, we can safely | |
6323 | * clear NAPI_STATE_MISSED. | |
6324 | * | |
6325 | * Note: x86 could use a single "lock and ..." instruction | |
6326 | * to perform these two clear_bit() | |
6327 | */ | |
6328 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
6329 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
6330 | ||
6331 | local_bh_disable(); | |
6332 | ||
6333 | /* All we really want here is to re-enable device interrupts. | |
6334 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
6335 | */ | |
6336 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
323ebb61 EC |
6337 | /* We can't gro_normal_list() here, because napi->poll() might have |
6338 | * rearmed the napi (napi_complete_done()) in which case it could | |
6339 | * already be running on another CPU. | |
6340 | */ | |
1e22391e | 6341 | trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); |
217f6974 | 6342 | netpoll_poll_unlock(have_poll_lock); |
323ebb61 EC |
6343 | if (rc == BUSY_POLL_BUDGET) { |
6344 | /* As the whole budget was spent, we still own the napi so can | |
6345 | * safely handle the rx_list. | |
6346 | */ | |
6347 | gro_normal_list(napi); | |
217f6974 | 6348 | __napi_schedule(napi); |
323ebb61 | 6349 | } |
217f6974 | 6350 | local_bh_enable(); |
217f6974 ED |
6351 | } |
6352 | ||
7db6b048 SS |
6353 | void napi_busy_loop(unsigned int napi_id, |
6354 | bool (*loop_end)(void *, unsigned long), | |
6355 | void *loop_end_arg) | |
02d62e86 | 6356 | { |
7db6b048 | 6357 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 6358 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 6359 | void *have_poll_lock = NULL; |
02d62e86 | 6360 | struct napi_struct *napi; |
217f6974 ED |
6361 | |
6362 | restart: | |
217f6974 | 6363 | napi_poll = NULL; |
02d62e86 | 6364 | |
2a028ecb | 6365 | rcu_read_lock(); |
02d62e86 | 6366 | |
545cd5e5 | 6367 | napi = napi_by_id(napi_id); |
02d62e86 ED |
6368 | if (!napi) |
6369 | goto out; | |
6370 | ||
217f6974 ED |
6371 | preempt_disable(); |
6372 | for (;;) { | |
2b5cd0df AD |
6373 | int work = 0; |
6374 | ||
2a028ecb | 6375 | local_bh_disable(); |
217f6974 ED |
6376 | if (!napi_poll) { |
6377 | unsigned long val = READ_ONCE(napi->state); | |
6378 | ||
6379 | /* If multiple threads are competing for this napi, | |
6380 | * we avoid dirtying napi->state as much as we can. | |
6381 | */ | |
6382 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
6383 | NAPIF_STATE_IN_BUSY_POLL)) | |
6384 | goto count; | |
6385 | if (cmpxchg(&napi->state, val, | |
6386 | val | NAPIF_STATE_IN_BUSY_POLL | | |
6387 | NAPIF_STATE_SCHED) != val) | |
6388 | goto count; | |
6389 | have_poll_lock = netpoll_poll_lock(napi); | |
6390 | napi_poll = napi->poll; | |
6391 | } | |
2b5cd0df AD |
6392 | work = napi_poll(napi, BUSY_POLL_BUDGET); |
6393 | trace_napi_poll(napi, work, BUSY_POLL_BUDGET); | |
323ebb61 | 6394 | gro_normal_list(napi); |
217f6974 | 6395 | count: |
2b5cd0df | 6396 | if (work > 0) |
7db6b048 | 6397 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 6398 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
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 ED |
6404 | if (unlikely(need_resched())) { |
6405 | if (napi_poll) | |
6406 | busy_poll_stop(napi, have_poll_lock); | |
6407 | preempt_enable(); | |
6408 | rcu_read_unlock(); | |
6409 | cond_resched(); | |
7db6b048 | 6410 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 6411 | return; |
217f6974 ED |
6412 | goto restart; |
6413 | } | |
6cdf89b1 | 6414 | cpu_relax(); |
217f6974 ED |
6415 | } |
6416 | if (napi_poll) | |
6417 | busy_poll_stop(napi, have_poll_lock); | |
6418 | preempt_enable(); | |
02d62e86 | 6419 | out: |
2a028ecb | 6420 | rcu_read_unlock(); |
02d62e86 | 6421 | } |
7db6b048 | 6422 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
6423 | |
6424 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 6425 | |
149d6ad8 | 6426 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 6427 | { |
d64b5e85 ED |
6428 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
6429 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 6430 | return; |
af12fa6e | 6431 | |
52bd2d62 | 6432 | spin_lock(&napi_hash_lock); |
af12fa6e | 6433 | |
545cd5e5 | 6434 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 6435 | do { |
545cd5e5 AD |
6436 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
6437 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
6438 | } while (napi_by_id(napi_gen_id)); |
6439 | napi->napi_id = napi_gen_id; | |
af12fa6e | 6440 | |
52bd2d62 ED |
6441 | hlist_add_head_rcu(&napi->napi_hash_node, |
6442 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 6443 | |
52bd2d62 | 6444 | spin_unlock(&napi_hash_lock); |
af12fa6e | 6445 | } |
af12fa6e ET |
6446 | |
6447 | /* Warning : caller is responsible to make sure rcu grace period | |
6448 | * is respected before freeing memory containing @napi | |
6449 | */ | |
34cbe27e | 6450 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 6451 | { |
34cbe27e ED |
6452 | bool rcu_sync_needed = false; |
6453 | ||
af12fa6e ET |
6454 | spin_lock(&napi_hash_lock); |
6455 | ||
34cbe27e ED |
6456 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
6457 | rcu_sync_needed = true; | |
af12fa6e | 6458 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 6459 | } |
af12fa6e | 6460 | spin_unlock(&napi_hash_lock); |
34cbe27e | 6461 | return rcu_sync_needed; |
af12fa6e ET |
6462 | } |
6463 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
6464 | ||
3b47d303 ED |
6465 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
6466 | { | |
6467 | struct napi_struct *napi; | |
6468 | ||
6469 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
6470 | |
6471 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
6472 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
6473 | */ | |
6f8b12d6 | 6474 | if (!napi_disable_pending(napi) && |
39e6c820 ED |
6475 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) |
6476 | __napi_schedule_irqoff(napi); | |
3b47d303 ED |
6477 | |
6478 | return HRTIMER_NORESTART; | |
6479 | } | |
6480 | ||
7c4ec749 | 6481 | static void init_gro_hash(struct napi_struct *napi) |
d565b0a1 | 6482 | { |
07d78363 DM |
6483 | int i; |
6484 | ||
6312fe77 LR |
6485 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6486 | INIT_LIST_HEAD(&napi->gro_hash[i].list); | |
6487 | napi->gro_hash[i].count = 0; | |
6488 | } | |
7c4ec749 DM |
6489 | napi->gro_bitmask = 0; |
6490 | } | |
6491 | ||
6492 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
6493 | int (*poll)(struct napi_struct *, int), int weight) | |
6494 | { | |
6495 | INIT_LIST_HEAD(&napi->poll_list); | |
6496 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); | |
6497 | napi->timer.function = napi_watchdog; | |
6498 | init_gro_hash(napi); | |
5d38a079 | 6499 | napi->skb = NULL; |
323ebb61 EC |
6500 | INIT_LIST_HEAD(&napi->rx_list); |
6501 | napi->rx_count = 0; | |
d565b0a1 | 6502 | napi->poll = poll; |
82dc3c63 | 6503 | if (weight > NAPI_POLL_WEIGHT) |
bf29e9e9 QC |
6504 | netdev_err_once(dev, "%s() called with weight %d\n", __func__, |
6505 | weight); | |
d565b0a1 HX |
6506 | napi->weight = weight; |
6507 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 6508 | napi->dev = dev; |
5d38a079 | 6509 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
6510 | napi->poll_owner = -1; |
6511 | #endif | |
6512 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 6513 | napi_hash_add(napi); |
d565b0a1 HX |
6514 | } |
6515 | EXPORT_SYMBOL(netif_napi_add); | |
6516 | ||
3b47d303 ED |
6517 | void napi_disable(struct napi_struct *n) |
6518 | { | |
6519 | might_sleep(); | |
6520 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
6521 | ||
6522 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
6523 | msleep(1); | |
2d8bff12 NH |
6524 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
6525 | msleep(1); | |
3b47d303 ED |
6526 | |
6527 | hrtimer_cancel(&n->timer); | |
6528 | ||
6529 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
6530 | } | |
6531 | EXPORT_SYMBOL(napi_disable); | |
6532 | ||
07d78363 | 6533 | static void flush_gro_hash(struct napi_struct *napi) |
d4546c25 | 6534 | { |
07d78363 | 6535 | int i; |
d4546c25 | 6536 | |
07d78363 DM |
6537 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6538 | struct sk_buff *skb, *n; | |
6539 | ||
6312fe77 | 6540 | list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) |
07d78363 | 6541 | kfree_skb(skb); |
6312fe77 | 6542 | napi->gro_hash[i].count = 0; |
07d78363 | 6543 | } |
d4546c25 DM |
6544 | } |
6545 | ||
93d05d4a | 6546 | /* Must be called in process context */ |
d565b0a1 HX |
6547 | void netif_napi_del(struct napi_struct *napi) |
6548 | { | |
93d05d4a ED |
6549 | might_sleep(); |
6550 | if (napi_hash_del(napi)) | |
6551 | synchronize_net(); | |
d7b06636 | 6552 | list_del_init(&napi->dev_list); |
76620aaf | 6553 | napi_free_frags(napi); |
d565b0a1 | 6554 | |
07d78363 | 6555 | flush_gro_hash(napi); |
d9f37d01 | 6556 | napi->gro_bitmask = 0; |
d565b0a1 HX |
6557 | } |
6558 | EXPORT_SYMBOL(netif_napi_del); | |
6559 | ||
726ce70e HX |
6560 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
6561 | { | |
6562 | void *have; | |
6563 | int work, weight; | |
6564 | ||
6565 | list_del_init(&n->poll_list); | |
6566 | ||
6567 | have = netpoll_poll_lock(n); | |
6568 | ||
6569 | weight = n->weight; | |
6570 | ||
6571 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
6572 | * with netpoll's poll_napi(). Only the entity which | |
6573 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
6574 | * actually make the ->poll() call. Therefore we avoid | |
6575 | * accidentally calling ->poll() when NAPI is not scheduled. | |
6576 | */ | |
6577 | work = 0; | |
6578 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
6579 | work = n->poll(n, weight); | |
1db19db7 | 6580 | trace_napi_poll(n, work, weight); |
726ce70e HX |
6581 | } |
6582 | ||
6583 | WARN_ON_ONCE(work > weight); | |
6584 | ||
6585 | if (likely(work < weight)) | |
6586 | goto out_unlock; | |
6587 | ||
6588 | /* Drivers must not modify the NAPI state if they | |
6589 | * consume the entire weight. In such cases this code | |
6590 | * still "owns" the NAPI instance and therefore can | |
6591 | * move the instance around on the list at-will. | |
6592 | */ | |
6593 | if (unlikely(napi_disable_pending(n))) { | |
6594 | napi_complete(n); | |
6595 | goto out_unlock; | |
6596 | } | |
6597 | ||
d9f37d01 | 6598 | if (n->gro_bitmask) { |
726ce70e HX |
6599 | /* flush too old packets |
6600 | * If HZ < 1000, flush all packets. | |
6601 | */ | |
6602 | napi_gro_flush(n, HZ >= 1000); | |
6603 | } | |
6604 | ||
c8079432 MM |
6605 | gro_normal_list(n); |
6606 | ||
001ce546 HX |
6607 | /* Some drivers may have called napi_schedule |
6608 | * prior to exhausting their budget. | |
6609 | */ | |
6610 | if (unlikely(!list_empty(&n->poll_list))) { | |
6611 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
6612 | n->dev ? n->dev->name : "backlog"); | |
6613 | goto out_unlock; | |
6614 | } | |
6615 | ||
726ce70e HX |
6616 | list_add_tail(&n->poll_list, repoll); |
6617 | ||
6618 | out_unlock: | |
6619 | netpoll_poll_unlock(have); | |
6620 | ||
6621 | return work; | |
6622 | } | |
6623 | ||
0766f788 | 6624 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 6625 | { |
903ceff7 | 6626 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e MW |
6627 | unsigned long time_limit = jiffies + |
6628 | usecs_to_jiffies(netdev_budget_usecs); | |
51b0bded | 6629 | int budget = netdev_budget; |
d75b1ade ED |
6630 | LIST_HEAD(list); |
6631 | LIST_HEAD(repoll); | |
53fb95d3 | 6632 | |
1da177e4 | 6633 | local_irq_disable(); |
d75b1ade ED |
6634 | list_splice_init(&sd->poll_list, &list); |
6635 | local_irq_enable(); | |
1da177e4 | 6636 | |
ceb8d5bf | 6637 | for (;;) { |
bea3348e | 6638 | struct napi_struct *n; |
1da177e4 | 6639 | |
ceb8d5bf HX |
6640 | if (list_empty(&list)) { |
6641 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
f52dffe0 | 6642 | goto out; |
ceb8d5bf HX |
6643 | break; |
6644 | } | |
6645 | ||
6bd373eb HX |
6646 | n = list_first_entry(&list, struct napi_struct, poll_list); |
6647 | budget -= napi_poll(n, &repoll); | |
6648 | ||
d75b1ade | 6649 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
6650 | * Allow this to run for 2 jiffies since which will allow |
6651 | * an average latency of 1.5/HZ. | |
bea3348e | 6652 | */ |
ceb8d5bf HX |
6653 | if (unlikely(budget <= 0 || |
6654 | time_after_eq(jiffies, time_limit))) { | |
6655 | sd->time_squeeze++; | |
6656 | break; | |
6657 | } | |
1da177e4 | 6658 | } |
d75b1ade | 6659 | |
d75b1ade ED |
6660 | local_irq_disable(); |
6661 | ||
6662 | list_splice_tail_init(&sd->poll_list, &list); | |
6663 | list_splice_tail(&repoll, &list); | |
6664 | list_splice(&list, &sd->poll_list); | |
6665 | if (!list_empty(&sd->poll_list)) | |
6666 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6667 | ||
e326bed2 | 6668 | net_rps_action_and_irq_enable(sd); |
f52dffe0 ED |
6669 | out: |
6670 | __kfree_skb_flush(); | |
1da177e4 LT |
6671 | } |
6672 | ||
aa9d8560 | 6673 | struct netdev_adjacent { |
9ff162a8 | 6674 | struct net_device *dev; |
5d261913 VF |
6675 | |
6676 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 6677 | bool master; |
5d261913 | 6678 | |
32b6d34f TY |
6679 | /* lookup ignore flag */ |
6680 | bool ignore; | |
6681 | ||
5d261913 VF |
6682 | /* counter for the number of times this device was added to us */ |
6683 | u16 ref_nr; | |
6684 | ||
402dae96 VF |
6685 | /* private field for the users */ |
6686 | void *private; | |
6687 | ||
9ff162a8 JP |
6688 | struct list_head list; |
6689 | struct rcu_head rcu; | |
9ff162a8 JP |
6690 | }; |
6691 | ||
6ea29da1 | 6692 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 6693 | struct list_head *adj_list) |
9ff162a8 | 6694 | { |
5d261913 | 6695 | struct netdev_adjacent *adj; |
5d261913 | 6696 | |
2f268f12 | 6697 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
6698 | if (adj->dev == adj_dev) |
6699 | return adj; | |
9ff162a8 JP |
6700 | } |
6701 | return NULL; | |
6702 | } | |
6703 | ||
32b6d34f | 6704 | static int ____netdev_has_upper_dev(struct net_device *upper_dev, void *data) |
f1170fd4 DA |
6705 | { |
6706 | struct net_device *dev = data; | |
6707 | ||
6708 | return upper_dev == dev; | |
6709 | } | |
6710 | ||
9ff162a8 JP |
6711 | /** |
6712 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
6713 | * @dev: device | |
6714 | * @upper_dev: upper device to check | |
6715 | * | |
6716 | * Find out if a device is linked to specified upper device and return true | |
6717 | * in case it is. Note that this checks only immediate upper device, | |
6718 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
6719 | */ | |
6720 | bool netdev_has_upper_dev(struct net_device *dev, | |
6721 | struct net_device *upper_dev) | |
6722 | { | |
6723 | ASSERT_RTNL(); | |
6724 | ||
32b6d34f | 6725 | return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, |
f1170fd4 | 6726 | upper_dev); |
9ff162a8 JP |
6727 | } |
6728 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
6729 | ||
1a3f060c DA |
6730 | /** |
6731 | * netdev_has_upper_dev_all - Check if device is linked to an upper device | |
6732 | * @dev: device | |
6733 | * @upper_dev: upper device to check | |
6734 | * | |
6735 | * Find out if a device is linked to specified upper device and return true | |
6736 | * in case it is. Note that this checks the entire upper device chain. | |
6737 | * The caller must hold rcu lock. | |
6738 | */ | |
6739 | ||
1a3f060c DA |
6740 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
6741 | struct net_device *upper_dev) | |
6742 | { | |
32b6d34f | 6743 | return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, |
1a3f060c DA |
6744 | upper_dev); |
6745 | } | |
6746 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
6747 | ||
9ff162a8 JP |
6748 | /** |
6749 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
6750 | * @dev: device | |
6751 | * | |
6752 | * Find out if a device is linked to an upper device and return true in case | |
6753 | * it is. The caller must hold the RTNL lock. | |
6754 | */ | |
25cc72a3 | 6755 | bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
6756 | { |
6757 | ASSERT_RTNL(); | |
6758 | ||
f1170fd4 | 6759 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 6760 | } |
25cc72a3 | 6761 | EXPORT_SYMBOL(netdev_has_any_upper_dev); |
9ff162a8 JP |
6762 | |
6763 | /** | |
6764 | * netdev_master_upper_dev_get - Get master upper device | |
6765 | * @dev: device | |
6766 | * | |
6767 | * Find a master upper device and return pointer to it or NULL in case | |
6768 | * it's not there. The caller must hold the RTNL lock. | |
6769 | */ | |
6770 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
6771 | { | |
aa9d8560 | 6772 | struct netdev_adjacent *upper; |
9ff162a8 JP |
6773 | |
6774 | ASSERT_RTNL(); | |
6775 | ||
2f268f12 | 6776 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
6777 | return NULL; |
6778 | ||
2f268f12 | 6779 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 6780 | struct netdev_adjacent, list); |
9ff162a8 JP |
6781 | if (likely(upper->master)) |
6782 | return upper->dev; | |
6783 | return NULL; | |
6784 | } | |
6785 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
6786 | ||
32b6d34f TY |
6787 | static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) |
6788 | { | |
6789 | struct netdev_adjacent *upper; | |
6790 | ||
6791 | ASSERT_RTNL(); | |
6792 | ||
6793 | if (list_empty(&dev->adj_list.upper)) | |
6794 | return NULL; | |
6795 | ||
6796 | upper = list_first_entry(&dev->adj_list.upper, | |
6797 | struct netdev_adjacent, list); | |
6798 | if (likely(upper->master) && !upper->ignore) | |
6799 | return upper->dev; | |
6800 | return NULL; | |
6801 | } | |
6802 | ||
0f524a80 DA |
6803 | /** |
6804 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
6805 | * @dev: device | |
6806 | * | |
6807 | * Find out if a device is linked to a lower device and return true in case | |
6808 | * it is. The caller must hold the RTNL lock. | |
6809 | */ | |
6810 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
6811 | { | |
6812 | ASSERT_RTNL(); | |
6813 | ||
6814 | return !list_empty(&dev->adj_list.lower); | |
6815 | } | |
6816 | ||
b6ccba4c VF |
6817 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
6818 | { | |
6819 | struct netdev_adjacent *adj; | |
6820 | ||
6821 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
6822 | ||
6823 | return adj->private; | |
6824 | } | |
6825 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
6826 | ||
44a40855 VY |
6827 | /** |
6828 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
6829 | * @dev: device | |
6830 | * @iter: list_head ** of the current position | |
6831 | * | |
6832 | * Gets the next device from the dev's upper list, starting from iter | |
6833 | * position. The caller must hold RCU read lock. | |
6834 | */ | |
6835 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
6836 | struct list_head **iter) | |
6837 | { | |
6838 | struct netdev_adjacent *upper; | |
6839 | ||
6840 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
6841 | ||
6842 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6843 | ||
6844 | if (&upper->list == &dev->adj_list.upper) | |
6845 | return NULL; | |
6846 | ||
6847 | *iter = &upper->list; | |
6848 | ||
6849 | return upper->dev; | |
6850 | } | |
6851 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
6852 | ||
32b6d34f TY |
6853 | static struct net_device *__netdev_next_upper_dev(struct net_device *dev, |
6854 | struct list_head **iter, | |
6855 | bool *ignore) | |
5343da4c TY |
6856 | { |
6857 | struct netdev_adjacent *upper; | |
6858 | ||
6859 | upper = list_entry((*iter)->next, struct netdev_adjacent, list); | |
6860 | ||
6861 | if (&upper->list == &dev->adj_list.upper) | |
6862 | return NULL; | |
6863 | ||
6864 | *iter = &upper->list; | |
32b6d34f | 6865 | *ignore = upper->ignore; |
5343da4c TY |
6866 | |
6867 | return upper->dev; | |
6868 | } | |
6869 | ||
1a3f060c DA |
6870 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
6871 | struct list_head **iter) | |
6872 | { | |
6873 | struct netdev_adjacent *upper; | |
6874 | ||
6875 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
6876 | ||
6877 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6878 | ||
6879 | if (&upper->list == &dev->adj_list.upper) | |
6880 | return NULL; | |
6881 | ||
6882 | *iter = &upper->list; | |
6883 | ||
6884 | return upper->dev; | |
6885 | } | |
6886 | ||
32b6d34f TY |
6887 | static int __netdev_walk_all_upper_dev(struct net_device *dev, |
6888 | int (*fn)(struct net_device *dev, | |
6889 | void *data), | |
6890 | void *data) | |
5343da4c TY |
6891 | { |
6892 | struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
6893 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
6894 | int ret, cur = 0; | |
32b6d34f | 6895 | bool ignore; |
5343da4c TY |
6896 | |
6897 | now = dev; | |
6898 | iter = &dev->adj_list.upper; | |
6899 | ||
6900 | while (1) { | |
6901 | if (now != dev) { | |
6902 | ret = fn(now, data); | |
6903 | if (ret) | |
6904 | return ret; | |
6905 | } | |
6906 | ||
6907 | next = NULL; | |
6908 | while (1) { | |
32b6d34f | 6909 | udev = __netdev_next_upper_dev(now, &iter, &ignore); |
5343da4c TY |
6910 | if (!udev) |
6911 | break; | |
32b6d34f TY |
6912 | if (ignore) |
6913 | continue; | |
5343da4c TY |
6914 | |
6915 | next = udev; | |
6916 | niter = &udev->adj_list.upper; | |
6917 | dev_stack[cur] = now; | |
6918 | iter_stack[cur++] = iter; | |
6919 | break; | |
6920 | } | |
6921 | ||
6922 | if (!next) { | |
6923 | if (!cur) | |
6924 | return 0; | |
6925 | next = dev_stack[--cur]; | |
6926 | niter = iter_stack[cur]; | |
6927 | } | |
6928 | ||
6929 | now = next; | |
6930 | iter = niter; | |
6931 | } | |
6932 | ||
6933 | return 0; | |
6934 | } | |
6935 | ||
1a3f060c DA |
6936 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, |
6937 | int (*fn)(struct net_device *dev, | |
6938 | void *data), | |
6939 | void *data) | |
6940 | { | |
5343da4c TY |
6941 | struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; |
6942 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
6943 | int ret, cur = 0; | |
1a3f060c | 6944 | |
5343da4c TY |
6945 | now = dev; |
6946 | iter = &dev->adj_list.upper; | |
1a3f060c | 6947 | |
5343da4c TY |
6948 | while (1) { |
6949 | if (now != dev) { | |
6950 | ret = fn(now, data); | |
6951 | if (ret) | |
6952 | return ret; | |
6953 | } | |
6954 | ||
6955 | next = NULL; | |
6956 | while (1) { | |
6957 | udev = netdev_next_upper_dev_rcu(now, &iter); | |
6958 | if (!udev) | |
6959 | break; | |
6960 | ||
6961 | next = udev; | |
6962 | niter = &udev->adj_list.upper; | |
6963 | dev_stack[cur] = now; | |
6964 | iter_stack[cur++] = iter; | |
6965 | break; | |
6966 | } | |
6967 | ||
6968 | if (!next) { | |
6969 | if (!cur) | |
6970 | return 0; | |
6971 | next = dev_stack[--cur]; | |
6972 | niter = iter_stack[cur]; | |
6973 | } | |
6974 | ||
6975 | now = next; | |
6976 | iter = niter; | |
1a3f060c DA |
6977 | } |
6978 | ||
6979 | return 0; | |
6980 | } | |
6981 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
6982 | ||
32b6d34f TY |
6983 | static bool __netdev_has_upper_dev(struct net_device *dev, |
6984 | struct net_device *upper_dev) | |
6985 | { | |
6986 | ASSERT_RTNL(); | |
6987 | ||
6988 | return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, | |
6989 | upper_dev); | |
6990 | } | |
6991 | ||
31088a11 VF |
6992 | /** |
6993 | * netdev_lower_get_next_private - Get the next ->private from the | |
6994 | * lower neighbour list | |
6995 | * @dev: device | |
6996 | * @iter: list_head ** of the current position | |
6997 | * | |
6998 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
6999 | * list, starting from iter position. The caller must hold either hold the | |
7000 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 7001 | * list will remain unchanged. |
31088a11 VF |
7002 | */ |
7003 | void *netdev_lower_get_next_private(struct net_device *dev, | |
7004 | struct list_head **iter) | |
7005 | { | |
7006 | struct netdev_adjacent *lower; | |
7007 | ||
7008 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
7009 | ||
7010 | if (&lower->list == &dev->adj_list.lower) | |
7011 | return NULL; | |
7012 | ||
6859e7df | 7013 | *iter = lower->list.next; |
31088a11 VF |
7014 | |
7015 | return lower->private; | |
7016 | } | |
7017 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
7018 | ||
7019 | /** | |
7020 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
7021 | * lower neighbour list, RCU | |
7022 | * variant | |
7023 | * @dev: device | |
7024 | * @iter: list_head ** of the current position | |
7025 | * | |
7026 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
7027 | * list, starting from iter position. The caller must hold RCU read lock. | |
7028 | */ | |
7029 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
7030 | struct list_head **iter) | |
7031 | { | |
7032 | struct netdev_adjacent *lower; | |
7033 | ||
7034 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
7035 | ||
7036 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7037 | ||
7038 | if (&lower->list == &dev->adj_list.lower) | |
7039 | return NULL; | |
7040 | ||
6859e7df | 7041 | *iter = &lower->list; |
31088a11 VF |
7042 | |
7043 | return lower->private; | |
7044 | } | |
7045 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
7046 | ||
4085ebe8 VY |
7047 | /** |
7048 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
7049 | * list | |
7050 | * @dev: device | |
7051 | * @iter: list_head ** of the current position | |
7052 | * | |
7053 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
7054 | * list, starting from iter position. The caller must hold RTNL lock or | |
7055 | * its own locking that guarantees that the neighbour lower | |
b469139e | 7056 | * list will remain unchanged. |
4085ebe8 VY |
7057 | */ |
7058 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
7059 | { | |
7060 | struct netdev_adjacent *lower; | |
7061 | ||
cfdd28be | 7062 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
7063 | |
7064 | if (&lower->list == &dev->adj_list.lower) | |
7065 | return NULL; | |
7066 | ||
cfdd28be | 7067 | *iter = lower->list.next; |
4085ebe8 VY |
7068 | |
7069 | return lower->dev; | |
7070 | } | |
7071 | EXPORT_SYMBOL(netdev_lower_get_next); | |
7072 | ||
1a3f060c DA |
7073 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
7074 | struct list_head **iter) | |
7075 | { | |
7076 | struct netdev_adjacent *lower; | |
7077 | ||
46b5ab1a | 7078 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
7079 | |
7080 | if (&lower->list == &dev->adj_list.lower) | |
7081 | return NULL; | |
7082 | ||
46b5ab1a | 7083 | *iter = &lower->list; |
1a3f060c DA |
7084 | |
7085 | return lower->dev; | |
7086 | } | |
7087 | ||
32b6d34f TY |
7088 | static struct net_device *__netdev_next_lower_dev(struct net_device *dev, |
7089 | struct list_head **iter, | |
7090 | bool *ignore) | |
7091 | { | |
7092 | struct netdev_adjacent *lower; | |
7093 | ||
7094 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
7095 | ||
7096 | if (&lower->list == &dev->adj_list.lower) | |
7097 | return NULL; | |
7098 | ||
7099 | *iter = &lower->list; | |
7100 | *ignore = lower->ignore; | |
7101 | ||
7102 | return lower->dev; | |
7103 | } | |
7104 | ||
1a3f060c DA |
7105 | int netdev_walk_all_lower_dev(struct net_device *dev, |
7106 | int (*fn)(struct net_device *dev, | |
7107 | void *data), | |
7108 | void *data) | |
7109 | { | |
5343da4c TY |
7110 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; |
7111 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7112 | int ret, cur = 0; | |
1a3f060c | 7113 | |
5343da4c TY |
7114 | now = dev; |
7115 | iter = &dev->adj_list.lower; | |
1a3f060c | 7116 | |
5343da4c TY |
7117 | while (1) { |
7118 | if (now != dev) { | |
7119 | ret = fn(now, data); | |
7120 | if (ret) | |
7121 | return ret; | |
7122 | } | |
7123 | ||
7124 | next = NULL; | |
7125 | while (1) { | |
7126 | ldev = netdev_next_lower_dev(now, &iter); | |
7127 | if (!ldev) | |
7128 | break; | |
7129 | ||
7130 | next = ldev; | |
7131 | niter = &ldev->adj_list.lower; | |
7132 | dev_stack[cur] = now; | |
7133 | iter_stack[cur++] = iter; | |
7134 | break; | |
7135 | } | |
7136 | ||
7137 | if (!next) { | |
7138 | if (!cur) | |
7139 | return 0; | |
7140 | next = dev_stack[--cur]; | |
7141 | niter = iter_stack[cur]; | |
7142 | } | |
7143 | ||
7144 | now = next; | |
7145 | iter = niter; | |
1a3f060c DA |
7146 | } |
7147 | ||
7148 | return 0; | |
7149 | } | |
7150 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
7151 | ||
32b6d34f TY |
7152 | static int __netdev_walk_all_lower_dev(struct net_device *dev, |
7153 | int (*fn)(struct net_device *dev, | |
7154 | void *data), | |
7155 | void *data) | |
7156 | { | |
7157 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7158 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7159 | int ret, cur = 0; | |
7160 | bool ignore; | |
7161 | ||
7162 | now = dev; | |
7163 | iter = &dev->adj_list.lower; | |
7164 | ||
7165 | while (1) { | |
7166 | if (now != dev) { | |
7167 | ret = fn(now, data); | |
7168 | if (ret) | |
7169 | return ret; | |
7170 | } | |
7171 | ||
7172 | next = NULL; | |
7173 | while (1) { | |
7174 | ldev = __netdev_next_lower_dev(now, &iter, &ignore); | |
7175 | if (!ldev) | |
7176 | break; | |
7177 | if (ignore) | |
7178 | continue; | |
7179 | ||
7180 | next = ldev; | |
7181 | niter = &ldev->adj_list.lower; | |
7182 | dev_stack[cur] = now; | |
7183 | iter_stack[cur++] = iter; | |
7184 | break; | |
7185 | } | |
7186 | ||
7187 | if (!next) { | |
7188 | if (!cur) | |
7189 | return 0; | |
7190 | next = dev_stack[--cur]; | |
7191 | niter = iter_stack[cur]; | |
7192 | } | |
7193 | ||
7194 | now = next; | |
7195 | iter = niter; | |
7196 | } | |
7197 | ||
7198 | return 0; | |
7199 | } | |
7200 | ||
7151affe TY |
7201 | struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
7202 | struct list_head **iter) | |
1a3f060c DA |
7203 | { |
7204 | struct netdev_adjacent *lower; | |
7205 | ||
7206 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7207 | if (&lower->list == &dev->adj_list.lower) | |
7208 | return NULL; | |
7209 | ||
7210 | *iter = &lower->list; | |
7211 | ||
7212 | return lower->dev; | |
7213 | } | |
7151affe | 7214 | EXPORT_SYMBOL(netdev_next_lower_dev_rcu); |
1a3f060c | 7215 | |
5343da4c TY |
7216 | static u8 __netdev_upper_depth(struct net_device *dev) |
7217 | { | |
7218 | struct net_device *udev; | |
7219 | struct list_head *iter; | |
7220 | u8 max_depth = 0; | |
32b6d34f | 7221 | bool ignore; |
5343da4c TY |
7222 | |
7223 | for (iter = &dev->adj_list.upper, | |
32b6d34f | 7224 | udev = __netdev_next_upper_dev(dev, &iter, &ignore); |
5343da4c | 7225 | udev; |
32b6d34f TY |
7226 | udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { |
7227 | if (ignore) | |
7228 | continue; | |
5343da4c TY |
7229 | if (max_depth < udev->upper_level) |
7230 | max_depth = udev->upper_level; | |
7231 | } | |
7232 | ||
7233 | return max_depth; | |
7234 | } | |
7235 | ||
7236 | static u8 __netdev_lower_depth(struct net_device *dev) | |
1a3f060c DA |
7237 | { |
7238 | struct net_device *ldev; | |
7239 | struct list_head *iter; | |
5343da4c | 7240 | u8 max_depth = 0; |
32b6d34f | 7241 | bool ignore; |
1a3f060c DA |
7242 | |
7243 | for (iter = &dev->adj_list.lower, | |
32b6d34f | 7244 | ldev = __netdev_next_lower_dev(dev, &iter, &ignore); |
1a3f060c | 7245 | ldev; |
32b6d34f TY |
7246 | ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { |
7247 | if (ignore) | |
7248 | continue; | |
5343da4c TY |
7249 | if (max_depth < ldev->lower_level) |
7250 | max_depth = ldev->lower_level; | |
7251 | } | |
1a3f060c | 7252 | |
5343da4c TY |
7253 | return max_depth; |
7254 | } | |
7255 | ||
7256 | static int __netdev_update_upper_level(struct net_device *dev, void *data) | |
7257 | { | |
7258 | dev->upper_level = __netdev_upper_depth(dev) + 1; | |
7259 | return 0; | |
7260 | } | |
7261 | ||
7262 | static int __netdev_update_lower_level(struct net_device *dev, void *data) | |
7263 | { | |
7264 | dev->lower_level = __netdev_lower_depth(dev) + 1; | |
7265 | return 0; | |
7266 | } | |
7267 | ||
7268 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
7269 | int (*fn)(struct net_device *dev, | |
7270 | void *data), | |
7271 | void *data) | |
7272 | { | |
7273 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7274 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7275 | int ret, cur = 0; | |
7276 | ||
7277 | now = dev; | |
7278 | iter = &dev->adj_list.lower; | |
7279 | ||
7280 | while (1) { | |
7281 | if (now != dev) { | |
7282 | ret = fn(now, data); | |
7283 | if (ret) | |
7284 | return ret; | |
7285 | } | |
7286 | ||
7287 | next = NULL; | |
7288 | while (1) { | |
7289 | ldev = netdev_next_lower_dev_rcu(now, &iter); | |
7290 | if (!ldev) | |
7291 | break; | |
7292 | ||
7293 | next = ldev; | |
7294 | niter = &ldev->adj_list.lower; | |
7295 | dev_stack[cur] = now; | |
7296 | iter_stack[cur++] = iter; | |
7297 | break; | |
7298 | } | |
7299 | ||
7300 | if (!next) { | |
7301 | if (!cur) | |
7302 | return 0; | |
7303 | next = dev_stack[--cur]; | |
7304 | niter = iter_stack[cur]; | |
7305 | } | |
7306 | ||
7307 | now = next; | |
7308 | iter = niter; | |
1a3f060c DA |
7309 | } |
7310 | ||
7311 | return 0; | |
7312 | } | |
7313 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
7314 | ||
e001bfad | 7315 | /** |
7316 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
7317 | * lower neighbour list, RCU | |
7318 | * variant | |
7319 | * @dev: device | |
7320 | * | |
7321 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
7322 | * list. The caller must hold RCU read lock. | |
7323 | */ | |
7324 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
7325 | { | |
7326 | struct netdev_adjacent *lower; | |
7327 | ||
7328 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
7329 | struct netdev_adjacent, list); | |
7330 | if (lower) | |
7331 | return lower->private; | |
7332 | return NULL; | |
7333 | } | |
7334 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
7335 | ||
9ff162a8 JP |
7336 | /** |
7337 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
7338 | * @dev: device | |
7339 | * | |
7340 | * Find a master upper device and return pointer to it or NULL in case | |
7341 | * it's not there. The caller must hold the RCU read lock. | |
7342 | */ | |
7343 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
7344 | { | |
aa9d8560 | 7345 | struct netdev_adjacent *upper; |
9ff162a8 | 7346 | |
2f268f12 | 7347 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 7348 | struct netdev_adjacent, list); |
9ff162a8 JP |
7349 | if (upper && likely(upper->master)) |
7350 | return upper->dev; | |
7351 | return NULL; | |
7352 | } | |
7353 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
7354 | ||
0a59f3a9 | 7355 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
7356 | struct net_device *adj_dev, |
7357 | struct list_head *dev_list) | |
7358 | { | |
7359 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 7360 | |
3ee32707 VF |
7361 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
7362 | "upper_%s" : "lower_%s", adj_dev->name); | |
7363 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
7364 | linkname); | |
7365 | } | |
0a59f3a9 | 7366 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
7367 | char *name, |
7368 | struct list_head *dev_list) | |
7369 | { | |
7370 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 7371 | |
3ee32707 VF |
7372 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
7373 | "upper_%s" : "lower_%s", name); | |
7374 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
7375 | } | |
7376 | ||
7ce64c79 AF |
7377 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
7378 | struct net_device *adj_dev, | |
7379 | struct list_head *dev_list) | |
7380 | { | |
7381 | return (dev_list == &dev->adj_list.upper || | |
7382 | dev_list == &dev->adj_list.lower) && | |
7383 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
7384 | } | |
3ee32707 | 7385 | |
5d261913 VF |
7386 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
7387 | struct net_device *adj_dev, | |
7863c054 | 7388 | struct list_head *dev_list, |
402dae96 | 7389 | void *private, bool master) |
5d261913 VF |
7390 | { |
7391 | struct netdev_adjacent *adj; | |
842d67a7 | 7392 | int ret; |
5d261913 | 7393 | |
6ea29da1 | 7394 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
7395 | |
7396 | if (adj) { | |
790510d9 | 7397 | adj->ref_nr += 1; |
67b62f98 DA |
7398 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
7399 | dev->name, adj_dev->name, adj->ref_nr); | |
7400 | ||
5d261913 VF |
7401 | return 0; |
7402 | } | |
7403 | ||
7404 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
7405 | if (!adj) | |
7406 | return -ENOMEM; | |
7407 | ||
7408 | adj->dev = adj_dev; | |
7409 | adj->master = master; | |
790510d9 | 7410 | adj->ref_nr = 1; |
402dae96 | 7411 | adj->private = private; |
32b6d34f | 7412 | adj->ignore = false; |
5d261913 | 7413 | dev_hold(adj_dev); |
2f268f12 | 7414 | |
67b62f98 DA |
7415 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
7416 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 7417 | |
7ce64c79 | 7418 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 7419 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
7420 | if (ret) |
7421 | goto free_adj; | |
7422 | } | |
7423 | ||
7863c054 | 7424 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
7425 | if (master) { |
7426 | ret = sysfs_create_link(&(dev->dev.kobj), | |
7427 | &(adj_dev->dev.kobj), "master"); | |
7428 | if (ret) | |
5831d66e | 7429 | goto remove_symlinks; |
842d67a7 | 7430 | |
7863c054 | 7431 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 7432 | } else { |
7863c054 | 7433 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 7434 | } |
5d261913 VF |
7435 | |
7436 | return 0; | |
842d67a7 | 7437 | |
5831d66e | 7438 | remove_symlinks: |
7ce64c79 | 7439 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 7440 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
7441 | free_adj: |
7442 | kfree(adj); | |
974daef7 | 7443 | dev_put(adj_dev); |
842d67a7 VF |
7444 | |
7445 | return ret; | |
5d261913 VF |
7446 | } |
7447 | ||
1d143d9f | 7448 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
7449 | struct net_device *adj_dev, | |
93409033 | 7450 | u16 ref_nr, |
1d143d9f | 7451 | struct list_head *dev_list) |
5d261913 VF |
7452 | { |
7453 | struct netdev_adjacent *adj; | |
7454 | ||
67b62f98 DA |
7455 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
7456 | dev->name, adj_dev->name, ref_nr); | |
7457 | ||
6ea29da1 | 7458 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 7459 | |
2f268f12 | 7460 | if (!adj) { |
67b62f98 | 7461 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 7462 | dev->name, adj_dev->name); |
67b62f98 DA |
7463 | WARN_ON(1); |
7464 | return; | |
2f268f12 | 7465 | } |
5d261913 | 7466 | |
93409033 | 7467 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
7468 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
7469 | dev->name, adj_dev->name, ref_nr, | |
7470 | adj->ref_nr - ref_nr); | |
93409033 | 7471 | adj->ref_nr -= ref_nr; |
5d261913 VF |
7472 | return; |
7473 | } | |
7474 | ||
842d67a7 VF |
7475 | if (adj->master) |
7476 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
7477 | ||
7ce64c79 | 7478 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 7479 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 7480 | |
5d261913 | 7481 | list_del_rcu(&adj->list); |
67b62f98 | 7482 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 7483 | adj_dev->name, dev->name, adj_dev->name); |
5d261913 VF |
7484 | dev_put(adj_dev); |
7485 | kfree_rcu(adj, rcu); | |
7486 | } | |
7487 | ||
1d143d9f | 7488 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
7489 | struct net_device *upper_dev, | |
7490 | struct list_head *up_list, | |
7491 | struct list_head *down_list, | |
7492 | void *private, bool master) | |
5d261913 VF |
7493 | { |
7494 | int ret; | |
7495 | ||
790510d9 | 7496 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 7497 | private, master); |
5d261913 VF |
7498 | if (ret) |
7499 | return ret; | |
7500 | ||
790510d9 | 7501 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 7502 | private, false); |
5d261913 | 7503 | if (ret) { |
790510d9 | 7504 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
7505 | return ret; |
7506 | } | |
7507 | ||
7508 | return 0; | |
7509 | } | |
7510 | ||
1d143d9f | 7511 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
7512 | struct net_device *upper_dev, | |
93409033 | 7513 | u16 ref_nr, |
1d143d9f | 7514 | struct list_head *up_list, |
7515 | struct list_head *down_list) | |
5d261913 | 7516 | { |
93409033 AC |
7517 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
7518 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
7519 | } |
7520 | ||
1d143d9f | 7521 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
7522 | struct net_device *upper_dev, | |
7523 | void *private, bool master) | |
2f268f12 | 7524 | { |
f1170fd4 DA |
7525 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
7526 | &dev->adj_list.upper, | |
7527 | &upper_dev->adj_list.lower, | |
7528 | private, master); | |
5d261913 VF |
7529 | } |
7530 | ||
1d143d9f | 7531 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
7532 | struct net_device *upper_dev) | |
2f268f12 | 7533 | { |
93409033 | 7534 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
7535 | &dev->adj_list.upper, |
7536 | &upper_dev->adj_list.lower); | |
7537 | } | |
5d261913 | 7538 | |
9ff162a8 | 7539 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 7540 | struct net_device *upper_dev, bool master, |
42ab19ee DA |
7541 | void *upper_priv, void *upper_info, |
7542 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7543 | { |
51d0c047 DA |
7544 | struct netdev_notifier_changeupper_info changeupper_info = { |
7545 | .info = { | |
7546 | .dev = dev, | |
42ab19ee | 7547 | .extack = extack, |
51d0c047 DA |
7548 | }, |
7549 | .upper_dev = upper_dev, | |
7550 | .master = master, | |
7551 | .linking = true, | |
7552 | .upper_info = upper_info, | |
7553 | }; | |
50d629e7 | 7554 | struct net_device *master_dev; |
5d261913 | 7555 | int ret = 0; |
9ff162a8 JP |
7556 | |
7557 | ASSERT_RTNL(); | |
7558 | ||
7559 | if (dev == upper_dev) | |
7560 | return -EBUSY; | |
7561 | ||
7562 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
32b6d34f | 7563 | if (__netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
7564 | return -EBUSY; |
7565 | ||
5343da4c TY |
7566 | if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) |
7567 | return -EMLINK; | |
7568 | ||
50d629e7 | 7569 | if (!master) { |
32b6d34f | 7570 | if (__netdev_has_upper_dev(dev, upper_dev)) |
50d629e7 MM |
7571 | return -EEXIST; |
7572 | } else { | |
32b6d34f | 7573 | master_dev = __netdev_master_upper_dev_get(dev); |
50d629e7 MM |
7574 | if (master_dev) |
7575 | return master_dev == upper_dev ? -EEXIST : -EBUSY; | |
7576 | } | |
9ff162a8 | 7577 | |
51d0c047 | 7578 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
7579 | &changeupper_info.info); |
7580 | ret = notifier_to_errno(ret); | |
7581 | if (ret) | |
7582 | return ret; | |
7583 | ||
6dffb044 | 7584 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 7585 | master); |
5d261913 VF |
7586 | if (ret) |
7587 | return ret; | |
9ff162a8 | 7588 | |
51d0c047 | 7589 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
b03804e7 IS |
7590 | &changeupper_info.info); |
7591 | ret = notifier_to_errno(ret); | |
7592 | if (ret) | |
f1170fd4 | 7593 | goto rollback; |
b03804e7 | 7594 | |
5343da4c | 7595 | __netdev_update_upper_level(dev, NULL); |
32b6d34f | 7596 | __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); |
5343da4c TY |
7597 | |
7598 | __netdev_update_lower_level(upper_dev, NULL); | |
32b6d34f TY |
7599 | __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, |
7600 | NULL); | |
5343da4c | 7601 | |
9ff162a8 | 7602 | return 0; |
5d261913 | 7603 | |
f1170fd4 | 7604 | rollback: |
2f268f12 | 7605 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
7606 | |
7607 | return ret; | |
9ff162a8 JP |
7608 | } |
7609 | ||
7610 | /** | |
7611 | * netdev_upper_dev_link - Add a link to the upper device | |
7612 | * @dev: device | |
7613 | * @upper_dev: new upper device | |
7a006d59 | 7614 | * @extack: netlink extended ack |
9ff162a8 JP |
7615 | * |
7616 | * Adds a link to device which is upper to this one. The caller must hold | |
7617 | * the RTNL lock. On a failure a negative errno code is returned. | |
7618 | * On success the reference counts are adjusted and the function | |
7619 | * returns zero. | |
7620 | */ | |
7621 | int netdev_upper_dev_link(struct net_device *dev, | |
42ab19ee DA |
7622 | struct net_device *upper_dev, |
7623 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7624 | { |
42ab19ee DA |
7625 | return __netdev_upper_dev_link(dev, upper_dev, false, |
7626 | NULL, NULL, extack); | |
9ff162a8 JP |
7627 | } |
7628 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
7629 | ||
7630 | /** | |
7631 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
7632 | * @dev: device | |
7633 | * @upper_dev: new upper device | |
6dffb044 | 7634 | * @upper_priv: upper device private |
29bf24af | 7635 | * @upper_info: upper info to be passed down via notifier |
7a006d59 | 7636 | * @extack: netlink extended ack |
9ff162a8 JP |
7637 | * |
7638 | * Adds a link to device which is upper to this one. In this case, only | |
7639 | * one master upper device can be linked, although other non-master devices | |
7640 | * might be linked as well. The caller must hold the RTNL lock. | |
7641 | * On a failure a negative errno code is returned. On success the reference | |
7642 | * counts are adjusted and the function returns zero. | |
7643 | */ | |
7644 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 7645 | struct net_device *upper_dev, |
42ab19ee DA |
7646 | void *upper_priv, void *upper_info, |
7647 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7648 | { |
29bf24af | 7649 | return __netdev_upper_dev_link(dev, upper_dev, true, |
42ab19ee | 7650 | upper_priv, upper_info, extack); |
9ff162a8 JP |
7651 | } |
7652 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
7653 | ||
7654 | /** | |
7655 | * netdev_upper_dev_unlink - Removes a link to upper device | |
7656 | * @dev: device | |
7657 | * @upper_dev: new upper device | |
7658 | * | |
7659 | * Removes a link to device which is upper to this one. The caller must hold | |
7660 | * the RTNL lock. | |
7661 | */ | |
7662 | void netdev_upper_dev_unlink(struct net_device *dev, | |
7663 | struct net_device *upper_dev) | |
7664 | { | |
51d0c047 DA |
7665 | struct netdev_notifier_changeupper_info changeupper_info = { |
7666 | .info = { | |
7667 | .dev = dev, | |
7668 | }, | |
7669 | .upper_dev = upper_dev, | |
7670 | .linking = false, | |
7671 | }; | |
f4563a75 | 7672 | |
9ff162a8 JP |
7673 | ASSERT_RTNL(); |
7674 | ||
0e4ead9d | 7675 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; |
0e4ead9d | 7676 | |
51d0c047 | 7677 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
7678 | &changeupper_info.info); |
7679 | ||
2f268f12 | 7680 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 7681 | |
51d0c047 | 7682 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
0e4ead9d | 7683 | &changeupper_info.info); |
5343da4c TY |
7684 | |
7685 | __netdev_update_upper_level(dev, NULL); | |
32b6d34f | 7686 | __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); |
5343da4c TY |
7687 | |
7688 | __netdev_update_lower_level(upper_dev, NULL); | |
32b6d34f TY |
7689 | __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, |
7690 | NULL); | |
9ff162a8 JP |
7691 | } |
7692 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
7693 | ||
32b6d34f TY |
7694 | static void __netdev_adjacent_dev_set(struct net_device *upper_dev, |
7695 | struct net_device *lower_dev, | |
7696 | bool val) | |
7697 | { | |
7698 | struct netdev_adjacent *adj; | |
7699 | ||
7700 | adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); | |
7701 | if (adj) | |
7702 | adj->ignore = val; | |
7703 | ||
7704 | adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); | |
7705 | if (adj) | |
7706 | adj->ignore = val; | |
7707 | } | |
7708 | ||
7709 | static void netdev_adjacent_dev_disable(struct net_device *upper_dev, | |
7710 | struct net_device *lower_dev) | |
7711 | { | |
7712 | __netdev_adjacent_dev_set(upper_dev, lower_dev, true); | |
7713 | } | |
7714 | ||
7715 | static void netdev_adjacent_dev_enable(struct net_device *upper_dev, | |
7716 | struct net_device *lower_dev) | |
7717 | { | |
7718 | __netdev_adjacent_dev_set(upper_dev, lower_dev, false); | |
7719 | } | |
7720 | ||
7721 | int netdev_adjacent_change_prepare(struct net_device *old_dev, | |
7722 | struct net_device *new_dev, | |
7723 | struct net_device *dev, | |
7724 | struct netlink_ext_ack *extack) | |
7725 | { | |
7726 | int err; | |
7727 | ||
7728 | if (!new_dev) | |
7729 | return 0; | |
7730 | ||
7731 | if (old_dev && new_dev != old_dev) | |
7732 | netdev_adjacent_dev_disable(dev, old_dev); | |
7733 | ||
7734 | err = netdev_upper_dev_link(new_dev, dev, extack); | |
7735 | if (err) { | |
7736 | if (old_dev && new_dev != old_dev) | |
7737 | netdev_adjacent_dev_enable(dev, old_dev); | |
7738 | return err; | |
7739 | } | |
7740 | ||
7741 | return 0; | |
7742 | } | |
7743 | EXPORT_SYMBOL(netdev_adjacent_change_prepare); | |
7744 | ||
7745 | void netdev_adjacent_change_commit(struct net_device *old_dev, | |
7746 | struct net_device *new_dev, | |
7747 | struct net_device *dev) | |
7748 | { | |
7749 | if (!new_dev || !old_dev) | |
7750 | return; | |
7751 | ||
7752 | if (new_dev == old_dev) | |
7753 | return; | |
7754 | ||
7755 | netdev_adjacent_dev_enable(dev, old_dev); | |
7756 | netdev_upper_dev_unlink(old_dev, dev); | |
7757 | } | |
7758 | EXPORT_SYMBOL(netdev_adjacent_change_commit); | |
7759 | ||
7760 | void netdev_adjacent_change_abort(struct net_device *old_dev, | |
7761 | struct net_device *new_dev, | |
7762 | struct net_device *dev) | |
7763 | { | |
7764 | if (!new_dev) | |
7765 | return; | |
7766 | ||
7767 | if (old_dev && new_dev != old_dev) | |
7768 | netdev_adjacent_dev_enable(dev, old_dev); | |
7769 | ||
7770 | netdev_upper_dev_unlink(new_dev, dev); | |
7771 | } | |
7772 | EXPORT_SYMBOL(netdev_adjacent_change_abort); | |
7773 | ||
61bd3857 MS |
7774 | /** |
7775 | * netdev_bonding_info_change - Dispatch event about slave change | |
7776 | * @dev: device | |
4a26e453 | 7777 | * @bonding_info: info to dispatch |
61bd3857 MS |
7778 | * |
7779 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
7780 | * The caller must hold the RTNL lock. | |
7781 | */ | |
7782 | void netdev_bonding_info_change(struct net_device *dev, | |
7783 | struct netdev_bonding_info *bonding_info) | |
7784 | { | |
51d0c047 DA |
7785 | struct netdev_notifier_bonding_info info = { |
7786 | .info.dev = dev, | |
7787 | }; | |
61bd3857 MS |
7788 | |
7789 | memcpy(&info.bonding_info, bonding_info, | |
7790 | sizeof(struct netdev_bonding_info)); | |
51d0c047 | 7791 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, |
61bd3857 MS |
7792 | &info.info); |
7793 | } | |
7794 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
7795 | ||
2ce1ee17 | 7796 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
7797 | { |
7798 | struct netdev_adjacent *iter; | |
7799 | ||
7800 | struct net *net = dev_net(dev); | |
7801 | ||
7802 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 7803 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7804 | continue; |
7805 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7806 | &iter->dev->adj_list.lower); | |
7807 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
7808 | &dev->adj_list.upper); | |
7809 | } | |
7810 | ||
7811 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 7812 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7813 | continue; |
7814 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7815 | &iter->dev->adj_list.upper); | |
7816 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
7817 | &dev->adj_list.lower); | |
7818 | } | |
7819 | } | |
7820 | ||
2ce1ee17 | 7821 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
7822 | { |
7823 | struct netdev_adjacent *iter; | |
7824 | ||
7825 | struct net *net = dev_net(dev); | |
7826 | ||
7827 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 7828 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7829 | continue; |
7830 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
7831 | &iter->dev->adj_list.lower); | |
7832 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
7833 | &dev->adj_list.upper); | |
7834 | } | |
7835 | ||
7836 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 7837 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7838 | continue; |
7839 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
7840 | &iter->dev->adj_list.upper); | |
7841 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
7842 | &dev->adj_list.lower); | |
7843 | } | |
7844 | } | |
7845 | ||
5bb025fa | 7846 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 7847 | { |
5bb025fa | 7848 | struct netdev_adjacent *iter; |
402dae96 | 7849 | |
4c75431a AF |
7850 | struct net *net = dev_net(dev); |
7851 | ||
5bb025fa | 7852 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 7853 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 7854 | continue; |
5bb025fa VF |
7855 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
7856 | &iter->dev->adj_list.lower); | |
7857 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7858 | &iter->dev->adj_list.lower); | |
7859 | } | |
402dae96 | 7860 | |
5bb025fa | 7861 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 7862 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 7863 | continue; |
5bb025fa VF |
7864 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
7865 | &iter->dev->adj_list.upper); | |
7866 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7867 | &iter->dev->adj_list.upper); | |
7868 | } | |
402dae96 | 7869 | } |
402dae96 VF |
7870 | |
7871 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
7872 | struct net_device *lower_dev) | |
7873 | { | |
7874 | struct netdev_adjacent *lower; | |
7875 | ||
7876 | if (!lower_dev) | |
7877 | return NULL; | |
6ea29da1 | 7878 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
7879 | if (!lower) |
7880 | return NULL; | |
7881 | ||
7882 | return lower->private; | |
7883 | } | |
7884 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
7885 | ||
4085ebe8 | 7886 | |
04d48266 JP |
7887 | /** |
7888 | * netdev_lower_change - Dispatch event about lower device state change | |
7889 | * @lower_dev: device | |
7890 | * @lower_state_info: state to dispatch | |
7891 | * | |
7892 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
7893 | * The caller must hold the RTNL lock. | |
7894 | */ | |
7895 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
7896 | void *lower_state_info) | |
7897 | { | |
51d0c047 DA |
7898 | struct netdev_notifier_changelowerstate_info changelowerstate_info = { |
7899 | .info.dev = lower_dev, | |
7900 | }; | |
04d48266 JP |
7901 | |
7902 | ASSERT_RTNL(); | |
7903 | changelowerstate_info.lower_state_info = lower_state_info; | |
51d0c047 | 7904 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, |
04d48266 JP |
7905 | &changelowerstate_info.info); |
7906 | } | |
7907 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
7908 | ||
b6c40d68 PM |
7909 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
7910 | { | |
d314774c SH |
7911 | const struct net_device_ops *ops = dev->netdev_ops; |
7912 | ||
d2615bf4 | 7913 | if (ops->ndo_change_rx_flags) |
d314774c | 7914 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
7915 | } |
7916 | ||
991fb3f7 | 7917 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 7918 | { |
b536db93 | 7919 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
7920 | kuid_t uid; |
7921 | kgid_t gid; | |
1da177e4 | 7922 | |
24023451 PM |
7923 | ASSERT_RTNL(); |
7924 | ||
dad9b335 WC |
7925 | dev->flags |= IFF_PROMISC; |
7926 | dev->promiscuity += inc; | |
7927 | if (dev->promiscuity == 0) { | |
7928 | /* | |
7929 | * Avoid overflow. | |
7930 | * If inc causes overflow, untouch promisc and return error. | |
7931 | */ | |
7932 | if (inc < 0) | |
7933 | dev->flags &= ~IFF_PROMISC; | |
7934 | else { | |
7935 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
7936 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
7937 | dev->name); | |
dad9b335 WC |
7938 | return -EOVERFLOW; |
7939 | } | |
7940 | } | |
52609c0b | 7941 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
7942 | pr_info("device %s %s promiscuous mode\n", |
7943 | dev->name, | |
7944 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
7945 | if (audit_enabled) { |
7946 | current_uid_gid(&uid, &gid); | |
cdfb6b34 RGB |
7947 | audit_log(audit_context(), GFP_ATOMIC, |
7948 | AUDIT_ANOM_PROMISCUOUS, | |
7949 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
7950 | dev->name, (dev->flags & IFF_PROMISC), | |
7951 | (old_flags & IFF_PROMISC), | |
7952 | from_kuid(&init_user_ns, audit_get_loginuid(current)), | |
7953 | from_kuid(&init_user_ns, uid), | |
7954 | from_kgid(&init_user_ns, gid), | |
7955 | audit_get_sessionid(current)); | |
8192b0c4 | 7956 | } |
24023451 | 7957 | |
b6c40d68 | 7958 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 7959 | } |
991fb3f7 ND |
7960 | if (notify) |
7961 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 7962 | return 0; |
1da177e4 LT |
7963 | } |
7964 | ||
4417da66 PM |
7965 | /** |
7966 | * dev_set_promiscuity - update promiscuity count on a device | |
7967 | * @dev: device | |
7968 | * @inc: modifier | |
7969 | * | |
7970 | * Add or remove promiscuity from a device. While the count in the device | |
7971 | * remains above zero the interface remains promiscuous. Once it hits zero | |
7972 | * the device reverts back to normal filtering operation. A negative inc | |
7973 | * value is used to drop promiscuity on the device. | |
dad9b335 | 7974 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 7975 | */ |
dad9b335 | 7976 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 7977 | { |
b536db93 | 7978 | unsigned int old_flags = dev->flags; |
dad9b335 | 7979 | int err; |
4417da66 | 7980 | |
991fb3f7 | 7981 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 7982 | if (err < 0) |
dad9b335 | 7983 | return err; |
4417da66 PM |
7984 | if (dev->flags != old_flags) |
7985 | dev_set_rx_mode(dev); | |
dad9b335 | 7986 | return err; |
4417da66 | 7987 | } |
d1b19dff | 7988 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 7989 | |
991fb3f7 | 7990 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 7991 | { |
991fb3f7 | 7992 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 7993 | |
24023451 PM |
7994 | ASSERT_RTNL(); |
7995 | ||
1da177e4 | 7996 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
7997 | dev->allmulti += inc; |
7998 | if (dev->allmulti == 0) { | |
7999 | /* | |
8000 | * Avoid overflow. | |
8001 | * If inc causes overflow, untouch allmulti and return error. | |
8002 | */ | |
8003 | if (inc < 0) | |
8004 | dev->flags &= ~IFF_ALLMULTI; | |
8005 | else { | |
8006 | dev->allmulti -= inc; | |
7b6cd1ce JP |
8007 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
8008 | dev->name); | |
dad9b335 WC |
8009 | return -EOVERFLOW; |
8010 | } | |
8011 | } | |
24023451 | 8012 | if (dev->flags ^ old_flags) { |
b6c40d68 | 8013 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 8014 | dev_set_rx_mode(dev); |
991fb3f7 ND |
8015 | if (notify) |
8016 | __dev_notify_flags(dev, old_flags, | |
8017 | dev->gflags ^ old_gflags); | |
24023451 | 8018 | } |
dad9b335 | 8019 | return 0; |
4417da66 | 8020 | } |
991fb3f7 ND |
8021 | |
8022 | /** | |
8023 | * dev_set_allmulti - update allmulti count on a device | |
8024 | * @dev: device | |
8025 | * @inc: modifier | |
8026 | * | |
8027 | * Add or remove reception of all multicast frames to a device. While the | |
8028 | * count in the device remains above zero the interface remains listening | |
8029 | * to all interfaces. Once it hits zero the device reverts back to normal | |
8030 | * filtering operation. A negative @inc value is used to drop the counter | |
8031 | * when releasing a resource needing all multicasts. | |
8032 | * Return 0 if successful or a negative errno code on error. | |
8033 | */ | |
8034 | ||
8035 | int dev_set_allmulti(struct net_device *dev, int inc) | |
8036 | { | |
8037 | return __dev_set_allmulti(dev, inc, true); | |
8038 | } | |
d1b19dff | 8039 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
8040 | |
8041 | /* | |
8042 | * Upload unicast and multicast address lists to device and | |
8043 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 8044 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
8045 | * are present. |
8046 | */ | |
8047 | void __dev_set_rx_mode(struct net_device *dev) | |
8048 | { | |
d314774c SH |
8049 | const struct net_device_ops *ops = dev->netdev_ops; |
8050 | ||
4417da66 PM |
8051 | /* dev_open will call this function so the list will stay sane. */ |
8052 | if (!(dev->flags&IFF_UP)) | |
8053 | return; | |
8054 | ||
8055 | if (!netif_device_present(dev)) | |
40b77c94 | 8056 | return; |
4417da66 | 8057 | |
01789349 | 8058 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
8059 | /* Unicast addresses changes may only happen under the rtnl, |
8060 | * therefore calling __dev_set_promiscuity here is safe. | |
8061 | */ | |
32e7bfc4 | 8062 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 8063 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 8064 | dev->uc_promisc = true; |
32e7bfc4 | 8065 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 8066 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 8067 | dev->uc_promisc = false; |
4417da66 | 8068 | } |
4417da66 | 8069 | } |
01789349 JP |
8070 | |
8071 | if (ops->ndo_set_rx_mode) | |
8072 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
8073 | } |
8074 | ||
8075 | void dev_set_rx_mode(struct net_device *dev) | |
8076 | { | |
b9e40857 | 8077 | netif_addr_lock_bh(dev); |
4417da66 | 8078 | __dev_set_rx_mode(dev); |
b9e40857 | 8079 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
8080 | } |
8081 | ||
f0db275a SH |
8082 | /** |
8083 | * dev_get_flags - get flags reported to userspace | |
8084 | * @dev: device | |
8085 | * | |
8086 | * Get the combination of flag bits exported through APIs to userspace. | |
8087 | */ | |
95c96174 | 8088 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 8089 | { |
95c96174 | 8090 | unsigned int flags; |
1da177e4 LT |
8091 | |
8092 | flags = (dev->flags & ~(IFF_PROMISC | | |
8093 | IFF_ALLMULTI | | |
b00055aa SR |
8094 | IFF_RUNNING | |
8095 | IFF_LOWER_UP | | |
8096 | IFF_DORMANT)) | | |
1da177e4 LT |
8097 | (dev->gflags & (IFF_PROMISC | |
8098 | IFF_ALLMULTI)); | |
8099 | ||
b00055aa SR |
8100 | if (netif_running(dev)) { |
8101 | if (netif_oper_up(dev)) | |
8102 | flags |= IFF_RUNNING; | |
8103 | if (netif_carrier_ok(dev)) | |
8104 | flags |= IFF_LOWER_UP; | |
8105 | if (netif_dormant(dev)) | |
8106 | flags |= IFF_DORMANT; | |
8107 | } | |
1da177e4 LT |
8108 | |
8109 | return flags; | |
8110 | } | |
d1b19dff | 8111 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 8112 | |
6d040321 PM |
8113 | int __dev_change_flags(struct net_device *dev, unsigned int flags, |
8114 | struct netlink_ext_ack *extack) | |
1da177e4 | 8115 | { |
b536db93 | 8116 | unsigned int old_flags = dev->flags; |
bd380811 | 8117 | int ret; |
1da177e4 | 8118 | |
24023451 PM |
8119 | ASSERT_RTNL(); |
8120 | ||
1da177e4 LT |
8121 | /* |
8122 | * Set the flags on our device. | |
8123 | */ | |
8124 | ||
8125 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
8126 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
8127 | IFF_AUTOMEDIA)) | | |
8128 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
8129 | IFF_ALLMULTI)); | |
8130 | ||
8131 | /* | |
8132 | * Load in the correct multicast list now the flags have changed. | |
8133 | */ | |
8134 | ||
b6c40d68 PM |
8135 | if ((old_flags ^ flags) & IFF_MULTICAST) |
8136 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 8137 | |
4417da66 | 8138 | dev_set_rx_mode(dev); |
1da177e4 LT |
8139 | |
8140 | /* | |
8141 | * Have we downed the interface. We handle IFF_UP ourselves | |
8142 | * according to user attempts to set it, rather than blindly | |
8143 | * setting it. | |
8144 | */ | |
8145 | ||
8146 | ret = 0; | |
7051b88a | 8147 | if ((old_flags ^ flags) & IFF_UP) { |
8148 | if (old_flags & IFF_UP) | |
8149 | __dev_close(dev); | |
8150 | else | |
40c900aa | 8151 | ret = __dev_open(dev, extack); |
7051b88a | 8152 | } |
1da177e4 | 8153 | |
1da177e4 | 8154 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 8155 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 8156 | unsigned int old_flags = dev->flags; |
d1b19dff | 8157 | |
1da177e4 | 8158 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
8159 | |
8160 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
8161 | if (dev->flags != old_flags) | |
8162 | dev_set_rx_mode(dev); | |
1da177e4 LT |
8163 | } |
8164 | ||
8165 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 8166 | * is important. Some (broken) drivers set IFF_PROMISC, when |
8167 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
8168 | */ |
8169 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
8170 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
8171 | ||
1da177e4 | 8172 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 8173 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
8174 | } |
8175 | ||
bd380811 PM |
8176 | return ret; |
8177 | } | |
8178 | ||
a528c219 ND |
8179 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
8180 | unsigned int gchanges) | |
bd380811 PM |
8181 | { |
8182 | unsigned int changes = dev->flags ^ old_flags; | |
8183 | ||
a528c219 | 8184 | if (gchanges) |
7f294054 | 8185 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 8186 | |
bd380811 PM |
8187 | if (changes & IFF_UP) { |
8188 | if (dev->flags & IFF_UP) | |
8189 | call_netdevice_notifiers(NETDEV_UP, dev); | |
8190 | else | |
8191 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
8192 | } | |
8193 | ||
8194 | if (dev->flags & IFF_UP && | |
be9efd36 | 8195 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
51d0c047 DA |
8196 | struct netdev_notifier_change_info change_info = { |
8197 | .info = { | |
8198 | .dev = dev, | |
8199 | }, | |
8200 | .flags_changed = changes, | |
8201 | }; | |
be9efd36 | 8202 | |
51d0c047 | 8203 | call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); |
be9efd36 | 8204 | } |
bd380811 PM |
8205 | } |
8206 | ||
8207 | /** | |
8208 | * dev_change_flags - change device settings | |
8209 | * @dev: device | |
8210 | * @flags: device state flags | |
567c5e13 | 8211 | * @extack: netlink extended ack |
bd380811 PM |
8212 | * |
8213 | * Change settings on device based state flags. The flags are | |
8214 | * in the userspace exported format. | |
8215 | */ | |
567c5e13 PM |
8216 | int dev_change_flags(struct net_device *dev, unsigned int flags, |
8217 | struct netlink_ext_ack *extack) | |
bd380811 | 8218 | { |
b536db93 | 8219 | int ret; |
991fb3f7 | 8220 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 | 8221 | |
6d040321 | 8222 | ret = __dev_change_flags(dev, flags, extack); |
bd380811 PM |
8223 | if (ret < 0) |
8224 | return ret; | |
8225 | ||
991fb3f7 | 8226 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 8227 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
8228 | return ret; |
8229 | } | |
d1b19dff | 8230 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 8231 | |
f51048c3 | 8232 | int __dev_set_mtu(struct net_device *dev, int new_mtu) |
2315dc91 VF |
8233 | { |
8234 | const struct net_device_ops *ops = dev->netdev_ops; | |
8235 | ||
8236 | if (ops->ndo_change_mtu) | |
8237 | return ops->ndo_change_mtu(dev, new_mtu); | |
8238 | ||
501a90c9 ED |
8239 | /* Pairs with all the lockless reads of dev->mtu in the stack */ |
8240 | WRITE_ONCE(dev->mtu, new_mtu); | |
2315dc91 VF |
8241 | return 0; |
8242 | } | |
f51048c3 | 8243 | EXPORT_SYMBOL(__dev_set_mtu); |
2315dc91 | 8244 | |
d836f5c6 ED |
8245 | int dev_validate_mtu(struct net_device *dev, int new_mtu, |
8246 | struct netlink_ext_ack *extack) | |
8247 | { | |
8248 | /* MTU must be positive, and in range */ | |
8249 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
8250 | NL_SET_ERR_MSG(extack, "mtu less than device minimum"); | |
8251 | return -EINVAL; | |
8252 | } | |
8253 | ||
8254 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
8255 | NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); | |
8256 | return -EINVAL; | |
8257 | } | |
8258 | return 0; | |
8259 | } | |
8260 | ||
f0db275a | 8261 | /** |
7a4c53be | 8262 | * dev_set_mtu_ext - Change maximum transfer unit |
f0db275a SH |
8263 | * @dev: device |
8264 | * @new_mtu: new transfer unit | |
7a4c53be | 8265 | * @extack: netlink extended ack |
f0db275a SH |
8266 | * |
8267 | * Change the maximum transfer size of the network device. | |
8268 | */ | |
7a4c53be SH |
8269 | int dev_set_mtu_ext(struct net_device *dev, int new_mtu, |
8270 | struct netlink_ext_ack *extack) | |
1da177e4 | 8271 | { |
2315dc91 | 8272 | int err, orig_mtu; |
1da177e4 LT |
8273 | |
8274 | if (new_mtu == dev->mtu) | |
8275 | return 0; | |
8276 | ||
d836f5c6 ED |
8277 | err = dev_validate_mtu(dev, new_mtu, extack); |
8278 | if (err) | |
8279 | return err; | |
1da177e4 LT |
8280 | |
8281 | if (!netif_device_present(dev)) | |
8282 | return -ENODEV; | |
8283 | ||
1d486bfb VF |
8284 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
8285 | err = notifier_to_errno(err); | |
8286 | if (err) | |
8287 | return err; | |
d314774c | 8288 | |
2315dc91 VF |
8289 | orig_mtu = dev->mtu; |
8290 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 8291 | |
2315dc91 | 8292 | if (!err) { |
af7d6cce SD |
8293 | err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
8294 | orig_mtu); | |
2315dc91 VF |
8295 | err = notifier_to_errno(err); |
8296 | if (err) { | |
8297 | /* setting mtu back and notifying everyone again, | |
8298 | * so that they have a chance to revert changes. | |
8299 | */ | |
8300 | __dev_set_mtu(dev, orig_mtu); | |
af7d6cce SD |
8301 | call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
8302 | new_mtu); | |
2315dc91 VF |
8303 | } |
8304 | } | |
1da177e4 LT |
8305 | return err; |
8306 | } | |
7a4c53be SH |
8307 | |
8308 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
8309 | { | |
8310 | struct netlink_ext_ack extack; | |
8311 | int err; | |
8312 | ||
a6bcfc89 | 8313 | memset(&extack, 0, sizeof(extack)); |
7a4c53be | 8314 | err = dev_set_mtu_ext(dev, new_mtu, &extack); |
a6bcfc89 | 8315 | if (err && extack._msg) |
7a4c53be SH |
8316 | net_err_ratelimited("%s: %s\n", dev->name, extack._msg); |
8317 | return err; | |
8318 | } | |
d1b19dff | 8319 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 8320 | |
6a643ddb CW |
8321 | /** |
8322 | * dev_change_tx_queue_len - Change TX queue length of a netdevice | |
8323 | * @dev: device | |
8324 | * @new_len: new tx queue length | |
8325 | */ | |
8326 | int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) | |
8327 | { | |
8328 | unsigned int orig_len = dev->tx_queue_len; | |
8329 | int res; | |
8330 | ||
8331 | if (new_len != (unsigned int)new_len) | |
8332 | return -ERANGE; | |
8333 | ||
8334 | if (new_len != orig_len) { | |
8335 | dev->tx_queue_len = new_len; | |
8336 | res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); | |
8337 | res = notifier_to_errno(res); | |
7effaf06 TT |
8338 | if (res) |
8339 | goto err_rollback; | |
8340 | res = dev_qdisc_change_tx_queue_len(dev); | |
8341 | if (res) | |
8342 | goto err_rollback; | |
6a643ddb CW |
8343 | } |
8344 | ||
8345 | return 0; | |
7effaf06 TT |
8346 | |
8347 | err_rollback: | |
8348 | netdev_err(dev, "refused to change device tx_queue_len\n"); | |
8349 | dev->tx_queue_len = orig_len; | |
8350 | return res; | |
6a643ddb CW |
8351 | } |
8352 | ||
cbda10fa VD |
8353 | /** |
8354 | * dev_set_group - Change group this device belongs to | |
8355 | * @dev: device | |
8356 | * @new_group: group this device should belong to | |
8357 | */ | |
8358 | void dev_set_group(struct net_device *dev, int new_group) | |
8359 | { | |
8360 | dev->group = new_group; | |
8361 | } | |
8362 | EXPORT_SYMBOL(dev_set_group); | |
8363 | ||
d59cdf94 PM |
8364 | /** |
8365 | * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. | |
8366 | * @dev: device | |
8367 | * @addr: new address | |
8368 | * @extack: netlink extended ack | |
8369 | */ | |
8370 | int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, | |
8371 | struct netlink_ext_ack *extack) | |
8372 | { | |
8373 | struct netdev_notifier_pre_changeaddr_info info = { | |
8374 | .info.dev = dev, | |
8375 | .info.extack = extack, | |
8376 | .dev_addr = addr, | |
8377 | }; | |
8378 | int rc; | |
8379 | ||
8380 | rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); | |
8381 | return notifier_to_errno(rc); | |
8382 | } | |
8383 | EXPORT_SYMBOL(dev_pre_changeaddr_notify); | |
8384 | ||
f0db275a SH |
8385 | /** |
8386 | * dev_set_mac_address - Change Media Access Control Address | |
8387 | * @dev: device | |
8388 | * @sa: new address | |
3a37a963 | 8389 | * @extack: netlink extended ack |
f0db275a SH |
8390 | * |
8391 | * Change the hardware (MAC) address of the device | |
8392 | */ | |
3a37a963 PM |
8393 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, |
8394 | struct netlink_ext_ack *extack) | |
1da177e4 | 8395 | { |
d314774c | 8396 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
8397 | int err; |
8398 | ||
d314774c | 8399 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
8400 | return -EOPNOTSUPP; |
8401 | if (sa->sa_family != dev->type) | |
8402 | return -EINVAL; | |
8403 | if (!netif_device_present(dev)) | |
8404 | return -ENODEV; | |
d59cdf94 PM |
8405 | err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); |
8406 | if (err) | |
8407 | return err; | |
d314774c | 8408 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
8409 | if (err) |
8410 | return err; | |
fbdeca2d | 8411 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 8412 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 8413 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 8414 | return 0; |
1da177e4 | 8415 | } |
d1b19dff | 8416 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 8417 | |
4bf84c35 JP |
8418 | /** |
8419 | * dev_change_carrier - Change device carrier | |
8420 | * @dev: device | |
691b3b7e | 8421 | * @new_carrier: new value |
4bf84c35 JP |
8422 | * |
8423 | * Change device carrier | |
8424 | */ | |
8425 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
8426 | { | |
8427 | const struct net_device_ops *ops = dev->netdev_ops; | |
8428 | ||
8429 | if (!ops->ndo_change_carrier) | |
8430 | return -EOPNOTSUPP; | |
8431 | if (!netif_device_present(dev)) | |
8432 | return -ENODEV; | |
8433 | return ops->ndo_change_carrier(dev, new_carrier); | |
8434 | } | |
8435 | EXPORT_SYMBOL(dev_change_carrier); | |
8436 | ||
66b52b0d JP |
8437 | /** |
8438 | * dev_get_phys_port_id - Get device physical port ID | |
8439 | * @dev: device | |
8440 | * @ppid: port ID | |
8441 | * | |
8442 | * Get device physical port ID | |
8443 | */ | |
8444 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 8445 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
8446 | { |
8447 | const struct net_device_ops *ops = dev->netdev_ops; | |
8448 | ||
8449 | if (!ops->ndo_get_phys_port_id) | |
8450 | return -EOPNOTSUPP; | |
8451 | return ops->ndo_get_phys_port_id(dev, ppid); | |
8452 | } | |
8453 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
8454 | ||
db24a904 DA |
8455 | /** |
8456 | * dev_get_phys_port_name - Get device physical port name | |
8457 | * @dev: device | |
8458 | * @name: port name | |
ed49e650 | 8459 | * @len: limit of bytes to copy to name |
db24a904 DA |
8460 | * |
8461 | * Get device physical port name | |
8462 | */ | |
8463 | int dev_get_phys_port_name(struct net_device *dev, | |
8464 | char *name, size_t len) | |
8465 | { | |
8466 | const struct net_device_ops *ops = dev->netdev_ops; | |
af3836df | 8467 | int err; |
db24a904 | 8468 | |
af3836df JP |
8469 | if (ops->ndo_get_phys_port_name) { |
8470 | err = ops->ndo_get_phys_port_name(dev, name, len); | |
8471 | if (err != -EOPNOTSUPP) | |
8472 | return err; | |
8473 | } | |
8474 | return devlink_compat_phys_port_name_get(dev, name, len); | |
db24a904 DA |
8475 | } |
8476 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
8477 | ||
d6abc596 FF |
8478 | /** |
8479 | * dev_get_port_parent_id - Get the device's port parent identifier | |
8480 | * @dev: network device | |
8481 | * @ppid: pointer to a storage for the port's parent identifier | |
8482 | * @recurse: allow/disallow recursion to lower devices | |
8483 | * | |
8484 | * Get the devices's port parent identifier | |
8485 | */ | |
8486 | int dev_get_port_parent_id(struct net_device *dev, | |
8487 | struct netdev_phys_item_id *ppid, | |
8488 | bool recurse) | |
8489 | { | |
8490 | const struct net_device_ops *ops = dev->netdev_ops; | |
8491 | struct netdev_phys_item_id first = { }; | |
8492 | struct net_device *lower_dev; | |
8493 | struct list_head *iter; | |
7e1146e8 JP |
8494 | int err; |
8495 | ||
8496 | if (ops->ndo_get_port_parent_id) { | |
8497 | err = ops->ndo_get_port_parent_id(dev, ppid); | |
8498 | if (err != -EOPNOTSUPP) | |
8499 | return err; | |
8500 | } | |
d6abc596 | 8501 | |
7e1146e8 JP |
8502 | err = devlink_compat_switch_id_get(dev, ppid); |
8503 | if (!err || err != -EOPNOTSUPP) | |
8504 | return err; | |
d6abc596 FF |
8505 | |
8506 | if (!recurse) | |
7e1146e8 | 8507 | return -EOPNOTSUPP; |
d6abc596 FF |
8508 | |
8509 | netdev_for_each_lower_dev(dev, lower_dev, iter) { | |
8510 | err = dev_get_port_parent_id(lower_dev, ppid, recurse); | |
8511 | if (err) | |
8512 | break; | |
8513 | if (!first.id_len) | |
8514 | first = *ppid; | |
8515 | else if (memcmp(&first, ppid, sizeof(*ppid))) | |
8516 | return -ENODATA; | |
8517 | } | |
8518 | ||
8519 | return err; | |
8520 | } | |
8521 | EXPORT_SYMBOL(dev_get_port_parent_id); | |
8522 | ||
8523 | /** | |
8524 | * netdev_port_same_parent_id - Indicate if two network devices have | |
8525 | * the same port parent identifier | |
8526 | * @a: first network device | |
8527 | * @b: second network device | |
8528 | */ | |
8529 | bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) | |
8530 | { | |
8531 | struct netdev_phys_item_id a_id = { }; | |
8532 | struct netdev_phys_item_id b_id = { }; | |
8533 | ||
8534 | if (dev_get_port_parent_id(a, &a_id, true) || | |
8535 | dev_get_port_parent_id(b, &b_id, true)) | |
8536 | return false; | |
8537 | ||
8538 | return netdev_phys_item_id_same(&a_id, &b_id); | |
8539 | } | |
8540 | EXPORT_SYMBOL(netdev_port_same_parent_id); | |
8541 | ||
d746d707 AK |
8542 | /** |
8543 | * dev_change_proto_down - update protocol port state information | |
8544 | * @dev: device | |
8545 | * @proto_down: new value | |
8546 | * | |
8547 | * This info can be used by switch drivers to set the phys state of the | |
8548 | * port. | |
8549 | */ | |
8550 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
8551 | { | |
8552 | const struct net_device_ops *ops = dev->netdev_ops; | |
8553 | ||
8554 | if (!ops->ndo_change_proto_down) | |
8555 | return -EOPNOTSUPP; | |
8556 | if (!netif_device_present(dev)) | |
8557 | return -ENODEV; | |
8558 | return ops->ndo_change_proto_down(dev, proto_down); | |
8559 | } | |
8560 | EXPORT_SYMBOL(dev_change_proto_down); | |
8561 | ||
b5899679 AR |
8562 | /** |
8563 | * dev_change_proto_down_generic - generic implementation for | |
8564 | * ndo_change_proto_down that sets carrier according to | |
8565 | * proto_down. | |
8566 | * | |
8567 | * @dev: device | |
8568 | * @proto_down: new value | |
8569 | */ | |
8570 | int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) | |
8571 | { | |
8572 | if (proto_down) | |
8573 | netif_carrier_off(dev); | |
8574 | else | |
8575 | netif_carrier_on(dev); | |
8576 | dev->proto_down = proto_down; | |
8577 | return 0; | |
8578 | } | |
8579 | EXPORT_SYMBOL(dev_change_proto_down_generic); | |
8580 | ||
a25717d2 JK |
8581 | u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op, |
8582 | enum bpf_netdev_command cmd) | |
d67b9cd2 | 8583 | { |
a25717d2 | 8584 | struct netdev_bpf xdp; |
d67b9cd2 | 8585 | |
a25717d2 JK |
8586 | if (!bpf_op) |
8587 | return 0; | |
118b4aa2 | 8588 | |
a25717d2 JK |
8589 | memset(&xdp, 0, sizeof(xdp)); |
8590 | xdp.command = cmd; | |
118b4aa2 | 8591 | |
a25717d2 JK |
8592 | /* Query must always succeed. */ |
8593 | WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG); | |
58038695 | 8594 | |
6b867589 | 8595 | return xdp.prog_id; |
d67b9cd2 DB |
8596 | } |
8597 | ||
f4e63525 | 8598 | static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op, |
32d60277 | 8599 | struct netlink_ext_ack *extack, u32 flags, |
d67b9cd2 DB |
8600 | struct bpf_prog *prog) |
8601 | { | |
7e6897f9 BT |
8602 | bool non_hw = !(flags & XDP_FLAGS_HW_MODE); |
8603 | struct bpf_prog *prev_prog = NULL; | |
f4e63525 | 8604 | struct netdev_bpf xdp; |
7e6897f9 BT |
8605 | int err; |
8606 | ||
8607 | if (non_hw) { | |
8608 | prev_prog = bpf_prog_by_id(__dev_xdp_query(dev, bpf_op, | |
8609 | XDP_QUERY_PROG)); | |
8610 | if (IS_ERR(prev_prog)) | |
8611 | prev_prog = NULL; | |
8612 | } | |
d67b9cd2 DB |
8613 | |
8614 | memset(&xdp, 0, sizeof(xdp)); | |
ee5d032f JK |
8615 | if (flags & XDP_FLAGS_HW_MODE) |
8616 | xdp.command = XDP_SETUP_PROG_HW; | |
8617 | else | |
8618 | xdp.command = XDP_SETUP_PROG; | |
d67b9cd2 | 8619 | xdp.extack = extack; |
32d60277 | 8620 | xdp.flags = flags; |
d67b9cd2 DB |
8621 | xdp.prog = prog; |
8622 | ||
7e6897f9 BT |
8623 | err = bpf_op(dev, &xdp); |
8624 | if (!err && non_hw) | |
8625 | bpf_prog_change_xdp(prev_prog, prog); | |
8626 | ||
8627 | if (prev_prog) | |
8628 | bpf_prog_put(prev_prog); | |
8629 | ||
8630 | return err; | |
d67b9cd2 DB |
8631 | } |
8632 | ||
bd0b2e7f JK |
8633 | static void dev_xdp_uninstall(struct net_device *dev) |
8634 | { | |
8635 | struct netdev_bpf xdp; | |
8636 | bpf_op_t ndo_bpf; | |
8637 | ||
8638 | /* Remove generic XDP */ | |
8639 | WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL)); | |
8640 | ||
8641 | /* Remove from the driver */ | |
8642 | ndo_bpf = dev->netdev_ops->ndo_bpf; | |
8643 | if (!ndo_bpf) | |
8644 | return; | |
8645 | ||
a25717d2 JK |
8646 | memset(&xdp, 0, sizeof(xdp)); |
8647 | xdp.command = XDP_QUERY_PROG; | |
8648 | WARN_ON(ndo_bpf(dev, &xdp)); | |
8649 | if (xdp.prog_id) | |
8650 | WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, | |
8651 | NULL)); | |
bd0b2e7f | 8652 | |
a25717d2 JK |
8653 | /* Remove HW offload */ |
8654 | memset(&xdp, 0, sizeof(xdp)); | |
8655 | xdp.command = XDP_QUERY_PROG_HW; | |
8656 | if (!ndo_bpf(dev, &xdp) && xdp.prog_id) | |
8657 | WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, | |
8658 | NULL)); | |
bd0b2e7f JK |
8659 | } |
8660 | ||
a7862b45 BB |
8661 | /** |
8662 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
8663 | * @dev: device | |
b5d60989 | 8664 | * @extack: netlink extended ack |
a7862b45 | 8665 | * @fd: new program fd or negative value to clear |
92234c8f | 8666 | * @expected_fd: old program fd that userspace expects to replace or clear |
85de8576 | 8667 | * @flags: xdp-related flags |
a7862b45 BB |
8668 | * |
8669 | * Set or clear a bpf program for a device | |
8670 | */ | |
ddf9f970 | 8671 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, |
92234c8f | 8672 | int fd, int expected_fd, u32 flags) |
a7862b45 BB |
8673 | { |
8674 | const struct net_device_ops *ops = dev->netdev_ops; | |
a25717d2 | 8675 | enum bpf_netdev_command query; |
92234c8f | 8676 | u32 prog_id, expected_id = 0; |
f4e63525 | 8677 | bpf_op_t bpf_op, bpf_chk; |
dfa74909 | 8678 | struct bpf_prog *prog; |
9ee963d6 | 8679 | bool offload; |
a7862b45 BB |
8680 | int err; |
8681 | ||
85de8576 DB |
8682 | ASSERT_RTNL(); |
8683 | ||
9ee963d6 JK |
8684 | offload = flags & XDP_FLAGS_HW_MODE; |
8685 | query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG; | |
a25717d2 | 8686 | |
f4e63525 | 8687 | bpf_op = bpf_chk = ops->ndo_bpf; |
01dde20c MF |
8688 | if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) { |
8689 | NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode"); | |
0489df9a | 8690 | return -EOPNOTSUPP; |
01dde20c | 8691 | } |
f4e63525 JK |
8692 | if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE)) |
8693 | bpf_op = generic_xdp_install; | |
8694 | if (bpf_op == bpf_chk) | |
8695 | bpf_chk = generic_xdp_install; | |
b5cdae32 | 8696 | |
92234c8f THJ |
8697 | prog_id = __dev_xdp_query(dev, bpf_op, query); |
8698 | if (flags & XDP_FLAGS_REPLACE) { | |
8699 | if (expected_fd >= 0) { | |
8700 | prog = bpf_prog_get_type_dev(expected_fd, | |
8701 | BPF_PROG_TYPE_XDP, | |
8702 | bpf_op == ops->ndo_bpf); | |
8703 | if (IS_ERR(prog)) | |
8704 | return PTR_ERR(prog); | |
8705 | expected_id = prog->aux->id; | |
8706 | bpf_prog_put(prog); | |
8707 | } | |
c14a9f63 | 8708 | |
92234c8f THJ |
8709 | if (prog_id != expected_id) { |
8710 | NL_SET_ERR_MSG(extack, "Active program does not match expected"); | |
8711 | return -EEXIST; | |
8712 | } | |
8713 | } | |
8714 | if (fd >= 0) { | |
9ee963d6 | 8715 | if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) { |
01dde20c | 8716 | NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time"); |
d67b9cd2 | 8717 | return -EEXIST; |
01dde20c | 8718 | } |
c14a9f63 | 8719 | |
c14a9f63 | 8720 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && prog_id) { |
01dde20c | 8721 | NL_SET_ERR_MSG(extack, "XDP program already attached"); |
d67b9cd2 | 8722 | return -EBUSY; |
01dde20c | 8723 | } |
85de8576 | 8724 | |
288b3de5 JK |
8725 | prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, |
8726 | bpf_op == ops->ndo_bpf); | |
a7862b45 BB |
8727 | if (IS_ERR(prog)) |
8728 | return PTR_ERR(prog); | |
441a3303 | 8729 | |
9ee963d6 | 8730 | if (!offload && bpf_prog_is_dev_bound(prog->aux)) { |
441a3303 JK |
8731 | NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported"); |
8732 | bpf_prog_put(prog); | |
8733 | return -EINVAL; | |
8734 | } | |
c14a9f63 | 8735 | |
aefc3e72 JK |
8736 | /* prog->aux->id may be 0 for orphaned device-bound progs */ |
8737 | if (prog->aux->id && prog->aux->id == prog_id) { | |
c14a9f63 MM |
8738 | bpf_prog_put(prog); |
8739 | return 0; | |
8740 | } | |
8741 | } else { | |
92234c8f | 8742 | if (!prog_id) |
c14a9f63 | 8743 | return 0; |
dfa74909 | 8744 | prog = NULL; |
a7862b45 BB |
8745 | } |
8746 | ||
f4e63525 | 8747 | err = dev_xdp_install(dev, bpf_op, extack, flags, prog); |
a7862b45 BB |
8748 | if (err < 0 && prog) |
8749 | bpf_prog_put(prog); | |
8750 | ||
8751 | return err; | |
8752 | } | |
a7862b45 | 8753 | |
1da177e4 LT |
8754 | /** |
8755 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 8756 | * @net: the applicable net namespace |
1da177e4 LT |
8757 | * |
8758 | * Returns a suitable unique value for a new device interface | |
8759 | * number. The caller must hold the rtnl semaphore or the | |
8760 | * dev_base_lock to be sure it remains unique. | |
8761 | */ | |
881d966b | 8762 | static int dev_new_index(struct net *net) |
1da177e4 | 8763 | { |
aa79e66e | 8764 | int ifindex = net->ifindex; |
f4563a75 | 8765 | |
1da177e4 LT |
8766 | for (;;) { |
8767 | if (++ifindex <= 0) | |
8768 | ifindex = 1; | |
881d966b | 8769 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 8770 | return net->ifindex = ifindex; |
1da177e4 LT |
8771 | } |
8772 | } | |
8773 | ||
1da177e4 | 8774 | /* Delayed registration/unregisteration */ |
3b5b34fd | 8775 | static LIST_HEAD(net_todo_list); |
200b916f | 8776 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 8777 | |
6f05f629 | 8778 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 8779 | { |
1da177e4 | 8780 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 8781 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
8782 | } |
8783 | ||
9b5e383c | 8784 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 8785 | { |
e93737b0 | 8786 | struct net_device *dev, *tmp; |
5cde2829 | 8787 | LIST_HEAD(close_head); |
9b5e383c | 8788 | |
93ee31f1 DL |
8789 | BUG_ON(dev_boot_phase); |
8790 | ASSERT_RTNL(); | |
8791 | ||
e93737b0 | 8792 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 8793 | /* Some devices call without registering |
e93737b0 KK |
8794 | * for initialization unwind. Remove those |
8795 | * devices and proceed with the remaining. | |
9b5e383c ED |
8796 | */ |
8797 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
8798 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
8799 | dev->name, dev); | |
93ee31f1 | 8800 | |
9b5e383c | 8801 | WARN_ON(1); |
e93737b0 KK |
8802 | list_del(&dev->unreg_list); |
8803 | continue; | |
9b5e383c | 8804 | } |
449f4544 | 8805 | dev->dismantle = true; |
9b5e383c | 8806 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 8807 | } |
93ee31f1 | 8808 | |
44345724 | 8809 | /* If device is running, close it first. */ |
5cde2829 EB |
8810 | list_for_each_entry(dev, head, unreg_list) |
8811 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 8812 | dev_close_many(&close_head, true); |
93ee31f1 | 8813 | |
44345724 | 8814 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
8815 | /* And unlink it from device chain. */ |
8816 | unlist_netdevice(dev); | |
93ee31f1 | 8817 | |
9b5e383c ED |
8818 | dev->reg_state = NETREG_UNREGISTERING; |
8819 | } | |
41852497 | 8820 | flush_all_backlogs(); |
93ee31f1 DL |
8821 | |
8822 | synchronize_net(); | |
8823 | ||
9b5e383c | 8824 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
8825 | struct sk_buff *skb = NULL; |
8826 | ||
9b5e383c ED |
8827 | /* Shutdown queueing discipline. */ |
8828 | dev_shutdown(dev); | |
93ee31f1 | 8829 | |
bd0b2e7f | 8830 | dev_xdp_uninstall(dev); |
93ee31f1 | 8831 | |
9b5e383c | 8832 | /* Notify protocols, that we are about to destroy |
eb13da1a | 8833 | * this device. They should clean all the things. |
8834 | */ | |
9b5e383c | 8835 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
93ee31f1 | 8836 | |
395eea6c MB |
8837 | if (!dev->rtnl_link_ops || |
8838 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
3d3ea5af | 8839 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, |
38e01b30 | 8840 | GFP_KERNEL, NULL, 0); |
395eea6c | 8841 | |
9b5e383c ED |
8842 | /* |
8843 | * Flush the unicast and multicast chains | |
8844 | */ | |
a748ee24 | 8845 | dev_uc_flush(dev); |
22bedad3 | 8846 | dev_mc_flush(dev); |
93ee31f1 | 8847 | |
36fbf1e5 | 8848 | netdev_name_node_alt_flush(dev); |
ff927412 JP |
8849 | netdev_name_node_free(dev->name_node); |
8850 | ||
9b5e383c ED |
8851 | if (dev->netdev_ops->ndo_uninit) |
8852 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 8853 | |
395eea6c MB |
8854 | if (skb) |
8855 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 8856 | |
9ff162a8 JP |
8857 | /* Notifier chain MUST detach us all upper devices. */ |
8858 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
0f524a80 | 8859 | WARN_ON(netdev_has_any_lower_dev(dev)); |
93ee31f1 | 8860 | |
9b5e383c ED |
8861 | /* Remove entries from kobject tree */ |
8862 | netdev_unregister_kobject(dev); | |
024e9679 AD |
8863 | #ifdef CONFIG_XPS |
8864 | /* Remove XPS queueing entries */ | |
8865 | netif_reset_xps_queues_gt(dev, 0); | |
8866 | #endif | |
9b5e383c | 8867 | } |
93ee31f1 | 8868 | |
850a545b | 8869 | synchronize_net(); |
395264d5 | 8870 | |
a5ee1551 | 8871 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
8872 | dev_put(dev); |
8873 | } | |
8874 | ||
8875 | static void rollback_registered(struct net_device *dev) | |
8876 | { | |
8877 | LIST_HEAD(single); | |
8878 | ||
8879 | list_add(&dev->unreg_list, &single); | |
8880 | rollback_registered_many(&single); | |
ceaaec98 | 8881 | list_del(&single); |
93ee31f1 DL |
8882 | } |
8883 | ||
fd867d51 JW |
8884 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
8885 | struct net_device *upper, netdev_features_t features) | |
8886 | { | |
8887 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
8888 | netdev_features_t feature; | |
5ba3f7d6 | 8889 | int feature_bit; |
fd867d51 | 8890 | |
3b89ea9c | 8891 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 8892 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
8893 | if (!(upper->wanted_features & feature) |
8894 | && (features & feature)) { | |
8895 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
8896 | &feature, upper->name); | |
8897 | features &= ~feature; | |
8898 | } | |
8899 | } | |
8900 | ||
8901 | return features; | |
8902 | } | |
8903 | ||
8904 | static void netdev_sync_lower_features(struct net_device *upper, | |
8905 | struct net_device *lower, netdev_features_t features) | |
8906 | { | |
8907 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
8908 | netdev_features_t feature; | |
5ba3f7d6 | 8909 | int feature_bit; |
fd867d51 | 8910 | |
3b89ea9c | 8911 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 8912 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
8913 | if (!(features & feature) && (lower->features & feature)) { |
8914 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
8915 | &feature, lower->name); | |
8916 | lower->wanted_features &= ~feature; | |
8917 | netdev_update_features(lower); | |
8918 | ||
8919 | if (unlikely(lower->features & feature)) | |
8920 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
8921 | &feature, lower->name); | |
8922 | } | |
8923 | } | |
8924 | } | |
8925 | ||
c8f44aff MM |
8926 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
8927 | netdev_features_t features) | |
b63365a2 | 8928 | { |
57422dc5 MM |
8929 | /* Fix illegal checksum combinations */ |
8930 | if ((features & NETIF_F_HW_CSUM) && | |
8931 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 8932 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
8933 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
8934 | } | |
8935 | ||
b63365a2 | 8936 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 8937 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 8938 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 8939 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
8940 | } |
8941 | ||
ec5f0615 PS |
8942 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
8943 | !(features & NETIF_F_IP_CSUM)) { | |
8944 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
8945 | features &= ~NETIF_F_TSO; | |
8946 | features &= ~NETIF_F_TSO_ECN; | |
8947 | } | |
8948 | ||
8949 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
8950 | !(features & NETIF_F_IPV6_CSUM)) { | |
8951 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
8952 | features &= ~NETIF_F_TSO6; | |
8953 | } | |
8954 | ||
b1dc497b AD |
8955 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
8956 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
8957 | features &= ~NETIF_F_TSO_MANGLEID; | |
8958 | ||
31d8b9e0 BH |
8959 | /* TSO ECN requires that TSO is present as well. */ |
8960 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
8961 | features &= ~NETIF_F_TSO_ECN; | |
8962 | ||
212b573f MM |
8963 | /* Software GSO depends on SG. */ |
8964 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 8965 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
8966 | features &= ~NETIF_F_GSO; |
8967 | } | |
8968 | ||
802ab55a AD |
8969 | /* GSO partial features require GSO partial be set */ |
8970 | if ((features & dev->gso_partial_features) && | |
8971 | !(features & NETIF_F_GSO_PARTIAL)) { | |
8972 | netdev_dbg(dev, | |
8973 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
8974 | features &= ~dev->gso_partial_features; | |
8975 | } | |
8976 | ||
fb1f5f79 MC |
8977 | if (!(features & NETIF_F_RXCSUM)) { |
8978 | /* NETIF_F_GRO_HW implies doing RXCSUM since every packet | |
8979 | * successfully merged by hardware must also have the | |
8980 | * checksum verified by hardware. If the user does not | |
8981 | * want to enable RXCSUM, logically, we should disable GRO_HW. | |
8982 | */ | |
8983 | if (features & NETIF_F_GRO_HW) { | |
8984 | netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); | |
8985 | features &= ~NETIF_F_GRO_HW; | |
8986 | } | |
8987 | } | |
8988 | ||
de8d5ab2 GP |
8989 | /* LRO/HW-GRO features cannot be combined with RX-FCS */ |
8990 | if (features & NETIF_F_RXFCS) { | |
8991 | if (features & NETIF_F_LRO) { | |
8992 | netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); | |
8993 | features &= ~NETIF_F_LRO; | |
8994 | } | |
8995 | ||
8996 | if (features & NETIF_F_GRO_HW) { | |
8997 | netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); | |
8998 | features &= ~NETIF_F_GRO_HW; | |
8999 | } | |
e6c6a929 GP |
9000 | } |
9001 | ||
b63365a2 HX |
9002 | return features; |
9003 | } | |
b63365a2 | 9004 | |
6cb6a27c | 9005 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 9006 | { |
fd867d51 | 9007 | struct net_device *upper, *lower; |
c8f44aff | 9008 | netdev_features_t features; |
fd867d51 | 9009 | struct list_head *iter; |
e7868a85 | 9010 | int err = -1; |
5455c699 | 9011 | |
87267485 MM |
9012 | ASSERT_RTNL(); |
9013 | ||
5455c699 MM |
9014 | features = netdev_get_wanted_features(dev); |
9015 | ||
9016 | if (dev->netdev_ops->ndo_fix_features) | |
9017 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
9018 | ||
9019 | /* driver might be less strict about feature dependencies */ | |
9020 | features = netdev_fix_features(dev, features); | |
9021 | ||
fd867d51 JW |
9022 | /* some features can't be enabled if they're off an an upper device */ |
9023 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
9024 | features = netdev_sync_upper_features(dev, upper, features); | |
9025 | ||
5455c699 | 9026 | if (dev->features == features) |
e7868a85 | 9027 | goto sync_lower; |
5455c699 | 9028 | |
c8f44aff MM |
9029 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
9030 | &dev->features, &features); | |
5455c699 MM |
9031 | |
9032 | if (dev->netdev_ops->ndo_set_features) | |
9033 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
9034 | else |
9035 | err = 0; | |
5455c699 | 9036 | |
6cb6a27c | 9037 | if (unlikely(err < 0)) { |
5455c699 | 9038 | netdev_err(dev, |
c8f44aff MM |
9039 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
9040 | err, &features, &dev->features); | |
17b85d29 NA |
9041 | /* return non-0 since some features might have changed and |
9042 | * it's better to fire a spurious notification than miss it | |
9043 | */ | |
9044 | return -1; | |
6cb6a27c MM |
9045 | } |
9046 | ||
e7868a85 | 9047 | sync_lower: |
fd867d51 JW |
9048 | /* some features must be disabled on lower devices when disabled |
9049 | * on an upper device (think: bonding master or bridge) | |
9050 | */ | |
9051 | netdev_for_each_lower_dev(dev, lower, iter) | |
9052 | netdev_sync_lower_features(dev, lower, features); | |
9053 | ||
ae847f40 SD |
9054 | if (!err) { |
9055 | netdev_features_t diff = features ^ dev->features; | |
9056 | ||
9057 | if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
9058 | /* udp_tunnel_{get,drop}_rx_info both need | |
9059 | * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the | |
9060 | * device, or they won't do anything. | |
9061 | * Thus we need to update dev->features | |
9062 | * *before* calling udp_tunnel_get_rx_info, | |
9063 | * but *after* calling udp_tunnel_drop_rx_info. | |
9064 | */ | |
9065 | if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
9066 | dev->features = features; | |
9067 | udp_tunnel_get_rx_info(dev); | |
9068 | } else { | |
9069 | udp_tunnel_drop_rx_info(dev); | |
9070 | } | |
9071 | } | |
9072 | ||
9daae9bd GP |
9073 | if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { |
9074 | if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { | |
9075 | dev->features = features; | |
9076 | err |= vlan_get_rx_ctag_filter_info(dev); | |
9077 | } else { | |
9078 | vlan_drop_rx_ctag_filter_info(dev); | |
9079 | } | |
9080 | } | |
9081 | ||
9082 | if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { | |
9083 | if (features & NETIF_F_HW_VLAN_STAG_FILTER) { | |
9084 | dev->features = features; | |
9085 | err |= vlan_get_rx_stag_filter_info(dev); | |
9086 | } else { | |
9087 | vlan_drop_rx_stag_filter_info(dev); | |
9088 | } | |
9089 | } | |
9090 | ||
6cb6a27c | 9091 | dev->features = features; |
ae847f40 | 9092 | } |
6cb6a27c | 9093 | |
e7868a85 | 9094 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
9095 | } |
9096 | ||
afe12cc8 MM |
9097 | /** |
9098 | * netdev_update_features - recalculate device features | |
9099 | * @dev: the device to check | |
9100 | * | |
9101 | * Recalculate dev->features set and send notifications if it | |
9102 | * has changed. Should be called after driver or hardware dependent | |
9103 | * conditions might have changed that influence the features. | |
9104 | */ | |
6cb6a27c MM |
9105 | void netdev_update_features(struct net_device *dev) |
9106 | { | |
9107 | if (__netdev_update_features(dev)) | |
9108 | netdev_features_change(dev); | |
5455c699 MM |
9109 | } |
9110 | EXPORT_SYMBOL(netdev_update_features); | |
9111 | ||
afe12cc8 MM |
9112 | /** |
9113 | * netdev_change_features - recalculate device features | |
9114 | * @dev: the device to check | |
9115 | * | |
9116 | * Recalculate dev->features set and send notifications even | |
9117 | * if they have not changed. Should be called instead of | |
9118 | * netdev_update_features() if also dev->vlan_features might | |
9119 | * have changed to allow the changes to be propagated to stacked | |
9120 | * VLAN devices. | |
9121 | */ | |
9122 | void netdev_change_features(struct net_device *dev) | |
9123 | { | |
9124 | __netdev_update_features(dev); | |
9125 | netdev_features_change(dev); | |
9126 | } | |
9127 | EXPORT_SYMBOL(netdev_change_features); | |
9128 | ||
fc4a7489 PM |
9129 | /** |
9130 | * netif_stacked_transfer_operstate - transfer operstate | |
9131 | * @rootdev: the root or lower level device to transfer state from | |
9132 | * @dev: the device to transfer operstate to | |
9133 | * | |
9134 | * Transfer operational state from root to device. This is normally | |
9135 | * called when a stacking relationship exists between the root | |
9136 | * device and the device(a leaf device). | |
9137 | */ | |
9138 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
9139 | struct net_device *dev) | |
9140 | { | |
9141 | if (rootdev->operstate == IF_OPER_DORMANT) | |
9142 | netif_dormant_on(dev); | |
9143 | else | |
9144 | netif_dormant_off(dev); | |
9145 | ||
eec517cd AL |
9146 | if (rootdev->operstate == IF_OPER_TESTING) |
9147 | netif_testing_on(dev); | |
9148 | else | |
9149 | netif_testing_off(dev); | |
9150 | ||
0575c86b ZS |
9151 | if (netif_carrier_ok(rootdev)) |
9152 | netif_carrier_on(dev); | |
9153 | else | |
9154 | netif_carrier_off(dev); | |
fc4a7489 PM |
9155 | } |
9156 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
9157 | ||
1b4bf461 ED |
9158 | static int netif_alloc_rx_queues(struct net_device *dev) |
9159 | { | |
1b4bf461 | 9160 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 9161 | struct netdev_rx_queue *rx; |
10595902 | 9162 | size_t sz = count * sizeof(*rx); |
e817f856 | 9163 | int err = 0; |
1b4bf461 | 9164 | |
bd25fa7b | 9165 | BUG_ON(count < 1); |
1b4bf461 | 9166 | |
dcda9b04 | 9167 | rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
9168 | if (!rx) |
9169 | return -ENOMEM; | |
9170 | ||
bd25fa7b TH |
9171 | dev->_rx = rx; |
9172 | ||
e817f856 | 9173 | for (i = 0; i < count; i++) { |
fe822240 | 9174 | rx[i].dev = dev; |
e817f856 JDB |
9175 | |
9176 | /* XDP RX-queue setup */ | |
9177 | err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i); | |
9178 | if (err < 0) | |
9179 | goto err_rxq_info; | |
9180 | } | |
1b4bf461 | 9181 | return 0; |
e817f856 JDB |
9182 | |
9183 | err_rxq_info: | |
9184 | /* Rollback successful reg's and free other resources */ | |
9185 | while (i--) | |
9186 | xdp_rxq_info_unreg(&rx[i].xdp_rxq); | |
141b52a9 | 9187 | kvfree(dev->_rx); |
e817f856 JDB |
9188 | dev->_rx = NULL; |
9189 | return err; | |
9190 | } | |
9191 | ||
9192 | static void netif_free_rx_queues(struct net_device *dev) | |
9193 | { | |
9194 | unsigned int i, count = dev->num_rx_queues; | |
e817f856 JDB |
9195 | |
9196 | /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ | |
9197 | if (!dev->_rx) | |
9198 | return; | |
9199 | ||
e817f856 | 9200 | for (i = 0; i < count; i++) |
82aaff2f JK |
9201 | xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); |
9202 | ||
9203 | kvfree(dev->_rx); | |
1b4bf461 ED |
9204 | } |
9205 | ||
aa942104 CG |
9206 | static void netdev_init_one_queue(struct net_device *dev, |
9207 | struct netdev_queue *queue, void *_unused) | |
9208 | { | |
9209 | /* Initialize queue lock */ | |
9210 | spin_lock_init(&queue->_xmit_lock); | |
ab92d68f | 9211 | lockdep_set_class(&queue->_xmit_lock, &dev->qdisc_xmit_lock_key); |
aa942104 | 9212 | queue->xmit_lock_owner = -1; |
b236da69 | 9213 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 9214 | queue->dev = dev; |
114cf580 TH |
9215 | #ifdef CONFIG_BQL |
9216 | dql_init(&queue->dql, HZ); | |
9217 | #endif | |
aa942104 CG |
9218 | } |
9219 | ||
60877a32 ED |
9220 | static void netif_free_tx_queues(struct net_device *dev) |
9221 | { | |
4cb28970 | 9222 | kvfree(dev->_tx); |
60877a32 ED |
9223 | } |
9224 | ||
e6484930 TH |
9225 | static int netif_alloc_netdev_queues(struct net_device *dev) |
9226 | { | |
9227 | unsigned int count = dev->num_tx_queues; | |
9228 | struct netdev_queue *tx; | |
60877a32 | 9229 | size_t sz = count * sizeof(*tx); |
e6484930 | 9230 | |
d339727c ED |
9231 | if (count < 1 || count > 0xffff) |
9232 | return -EINVAL; | |
62b5942a | 9233 | |
dcda9b04 | 9234 | tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
9235 | if (!tx) |
9236 | return -ENOMEM; | |
9237 | ||
e6484930 | 9238 | dev->_tx = tx; |
1d24eb48 | 9239 | |
e6484930 TH |
9240 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
9241 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
9242 | |
9243 | return 0; | |
e6484930 TH |
9244 | } |
9245 | ||
a2029240 DV |
9246 | void netif_tx_stop_all_queues(struct net_device *dev) |
9247 | { | |
9248 | unsigned int i; | |
9249 | ||
9250 | for (i = 0; i < dev->num_tx_queues; i++) { | |
9251 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 9252 | |
a2029240 DV |
9253 | netif_tx_stop_queue(txq); |
9254 | } | |
9255 | } | |
9256 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
9257 | ||
ab92d68f TY |
9258 | static void netdev_register_lockdep_key(struct net_device *dev) |
9259 | { | |
9260 | lockdep_register_key(&dev->qdisc_tx_busylock_key); | |
9261 | lockdep_register_key(&dev->qdisc_running_key); | |
9262 | lockdep_register_key(&dev->qdisc_xmit_lock_key); | |
9263 | lockdep_register_key(&dev->addr_list_lock_key); | |
9264 | } | |
9265 | ||
9266 | static void netdev_unregister_lockdep_key(struct net_device *dev) | |
9267 | { | |
9268 | lockdep_unregister_key(&dev->qdisc_tx_busylock_key); | |
9269 | lockdep_unregister_key(&dev->qdisc_running_key); | |
9270 | lockdep_unregister_key(&dev->qdisc_xmit_lock_key); | |
9271 | lockdep_unregister_key(&dev->addr_list_lock_key); | |
9272 | } | |
9273 | ||
9274 | void netdev_update_lockdep_key(struct net_device *dev) | |
9275 | { | |
ab92d68f | 9276 | lockdep_unregister_key(&dev->addr_list_lock_key); |
ab92d68f TY |
9277 | lockdep_register_key(&dev->addr_list_lock_key); |
9278 | ||
9279 | lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key); | |
ab92d68f TY |
9280 | } |
9281 | EXPORT_SYMBOL(netdev_update_lockdep_key); | |
9282 | ||
1da177e4 LT |
9283 | /** |
9284 | * register_netdevice - register a network device | |
9285 | * @dev: device to register | |
9286 | * | |
9287 | * Take a completed network device structure and add it to the kernel | |
9288 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
9289 | * chain. 0 is returned on success. A negative errno code is returned | |
9290 | * on a failure to set up the device, or if the name is a duplicate. | |
9291 | * | |
9292 | * Callers must hold the rtnl semaphore. You may want | |
9293 | * register_netdev() instead of this. | |
9294 | * | |
9295 | * BUGS: | |
9296 | * The locking appears insufficient to guarantee two parallel registers | |
9297 | * will not get the same name. | |
9298 | */ | |
9299 | ||
9300 | int register_netdevice(struct net_device *dev) | |
9301 | { | |
1da177e4 | 9302 | int ret; |
d314774c | 9303 | struct net *net = dev_net(dev); |
1da177e4 | 9304 | |
e283de3a FF |
9305 | BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < |
9306 | NETDEV_FEATURE_COUNT); | |
1da177e4 LT |
9307 | BUG_ON(dev_boot_phase); |
9308 | ASSERT_RTNL(); | |
9309 | ||
b17a7c17 SH |
9310 | might_sleep(); |
9311 | ||
1da177e4 LT |
9312 | /* When net_device's are persistent, this will be fatal. */ |
9313 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 9314 | BUG_ON(!net); |
1da177e4 | 9315 | |
9000edb7 JK |
9316 | ret = ethtool_check_ops(dev->ethtool_ops); |
9317 | if (ret) | |
9318 | return ret; | |
9319 | ||
f1f28aa3 | 9320 | spin_lock_init(&dev->addr_list_lock); |
ab92d68f | 9321 | lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key); |
1da177e4 | 9322 | |
828de4f6 | 9323 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
9324 | if (ret < 0) |
9325 | goto out; | |
9326 | ||
9077f052 | 9327 | ret = -ENOMEM; |
ff927412 JP |
9328 | dev->name_node = netdev_name_node_head_alloc(dev); |
9329 | if (!dev->name_node) | |
9330 | goto out; | |
9331 | ||
1da177e4 | 9332 | /* Init, if this function is available */ |
d314774c SH |
9333 | if (dev->netdev_ops->ndo_init) { |
9334 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
9335 | if (ret) { |
9336 | if (ret > 0) | |
9337 | ret = -EIO; | |
42c17fa6 | 9338 | goto err_free_name; |
1da177e4 LT |
9339 | } |
9340 | } | |
4ec93edb | 9341 | |
f646968f PM |
9342 | if (((dev->hw_features | dev->features) & |
9343 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
9344 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
9345 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
9346 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
9347 | ret = -EINVAL; | |
9348 | goto err_uninit; | |
9349 | } | |
9350 | ||
9c7dafbf PE |
9351 | ret = -EBUSY; |
9352 | if (!dev->ifindex) | |
9353 | dev->ifindex = dev_new_index(net); | |
9354 | else if (__dev_get_by_index(net, dev->ifindex)) | |
9355 | goto err_uninit; | |
9356 | ||
5455c699 MM |
9357 | /* Transfer changeable features to wanted_features and enable |
9358 | * software offloads (GSO and GRO). | |
9359 | */ | |
1a3c998f | 9360 | dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); |
14d1232f | 9361 | dev->features |= NETIF_F_SOFT_FEATURES; |
d764a122 SD |
9362 | |
9363 | if (dev->netdev_ops->ndo_udp_tunnel_add) { | |
9364 | dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
9365 | dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
9366 | } | |
9367 | ||
14d1232f | 9368 | dev->wanted_features = dev->features & dev->hw_features; |
1da177e4 | 9369 | |
cbc53e08 | 9370 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 9371 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 9372 | |
7f348a60 AD |
9373 | /* If IPv4 TCP segmentation offload is supported we should also |
9374 | * allow the device to enable segmenting the frame with the option | |
9375 | * of ignoring a static IP ID value. This doesn't enable the | |
9376 | * feature itself but allows the user to enable it later. | |
9377 | */ | |
cbc53e08 AD |
9378 | if (dev->hw_features & NETIF_F_TSO) |
9379 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
9380 | if (dev->vlan_features & NETIF_F_TSO) |
9381 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
9382 | if (dev->mpls_features & NETIF_F_TSO) | |
9383 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
9384 | if (dev->hw_enc_features & NETIF_F_TSO) | |
9385 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 9386 | |
1180e7d6 | 9387 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 9388 | */ |
1180e7d6 | 9389 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 9390 | |
ee579677 PS |
9391 | /* Make NETIF_F_SG inheritable to tunnel devices. |
9392 | */ | |
802ab55a | 9393 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 9394 | |
0d89d203 SH |
9395 | /* Make NETIF_F_SG inheritable to MPLS. |
9396 | */ | |
9397 | dev->mpls_features |= NETIF_F_SG; | |
9398 | ||
7ffbe3fd JB |
9399 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
9400 | ret = notifier_to_errno(ret); | |
9401 | if (ret) | |
9402 | goto err_uninit; | |
9403 | ||
8b41d188 | 9404 | ret = netdev_register_kobject(dev); |
cb626bf5 JH |
9405 | if (ret) { |
9406 | dev->reg_state = NETREG_UNREGISTERED; | |
7ce1b0ed | 9407 | goto err_uninit; |
cb626bf5 | 9408 | } |
b17a7c17 SH |
9409 | dev->reg_state = NETREG_REGISTERED; |
9410 | ||
6cb6a27c | 9411 | __netdev_update_features(dev); |
8e9b59b2 | 9412 | |
1da177e4 LT |
9413 | /* |
9414 | * Default initial state at registry is that the | |
9415 | * device is present. | |
9416 | */ | |
9417 | ||
9418 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
9419 | ||
8f4cccbb BH |
9420 | linkwatch_init_dev(dev); |
9421 | ||
1da177e4 | 9422 | dev_init_scheduler(dev); |
1da177e4 | 9423 | dev_hold(dev); |
ce286d32 | 9424 | list_netdevice(dev); |
7bf23575 | 9425 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 9426 | |
948b337e JP |
9427 | /* If the device has permanent device address, driver should |
9428 | * set dev_addr and also addr_assign_type should be set to | |
9429 | * NET_ADDR_PERM (default value). | |
9430 | */ | |
9431 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
9432 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
9433 | ||
1da177e4 | 9434 | /* Notify protocols, that a new device appeared. */ |
056925ab | 9435 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 9436 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
9437 | if (ret) { |
9438 | rollback_registered(dev); | |
10cc514f SAK |
9439 | rcu_barrier(); |
9440 | ||
93ee31f1 DL |
9441 | dev->reg_state = NETREG_UNREGISTERED; |
9442 | } | |
d90a909e EB |
9443 | /* |
9444 | * Prevent userspace races by waiting until the network | |
9445 | * device is fully setup before sending notifications. | |
9446 | */ | |
a2835763 PM |
9447 | if (!dev->rtnl_link_ops || |
9448 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 9449 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
9450 | |
9451 | out: | |
9452 | return ret; | |
7ce1b0ed HX |
9453 | |
9454 | err_uninit: | |
d314774c SH |
9455 | if (dev->netdev_ops->ndo_uninit) |
9456 | dev->netdev_ops->ndo_uninit(dev); | |
cf124db5 DM |
9457 | if (dev->priv_destructor) |
9458 | dev->priv_destructor(dev); | |
42c17fa6 DC |
9459 | err_free_name: |
9460 | netdev_name_node_free(dev->name_node); | |
7ce1b0ed | 9461 | goto out; |
1da177e4 | 9462 | } |
d1b19dff | 9463 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 9464 | |
937f1ba5 BH |
9465 | /** |
9466 | * init_dummy_netdev - init a dummy network device for NAPI | |
9467 | * @dev: device to init | |
9468 | * | |
9469 | * This takes a network device structure and initialize the minimum | |
9470 | * amount of fields so it can be used to schedule NAPI polls without | |
9471 | * registering a full blown interface. This is to be used by drivers | |
9472 | * that need to tie several hardware interfaces to a single NAPI | |
9473 | * poll scheduler due to HW limitations. | |
9474 | */ | |
9475 | int init_dummy_netdev(struct net_device *dev) | |
9476 | { | |
9477 | /* Clear everything. Note we don't initialize spinlocks | |
9478 | * are they aren't supposed to be taken by any of the | |
9479 | * NAPI code and this dummy netdev is supposed to be | |
9480 | * only ever used for NAPI polls | |
9481 | */ | |
9482 | memset(dev, 0, sizeof(struct net_device)); | |
9483 | ||
9484 | /* make sure we BUG if trying to hit standard | |
9485 | * register/unregister code path | |
9486 | */ | |
9487 | dev->reg_state = NETREG_DUMMY; | |
9488 | ||
937f1ba5 BH |
9489 | /* NAPI wants this */ |
9490 | INIT_LIST_HEAD(&dev->napi_list); | |
9491 | ||
9492 | /* a dummy interface is started by default */ | |
9493 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
9494 | set_bit(__LINK_STATE_START, &dev->state); | |
9495 | ||
35edfdc7 JE |
9496 | /* napi_busy_loop stats accounting wants this */ |
9497 | dev_net_set(dev, &init_net); | |
9498 | ||
29b4433d ED |
9499 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
9500 | * because users of this 'device' dont need to change | |
9501 | * its refcount. | |
9502 | */ | |
9503 | ||
937f1ba5 BH |
9504 | return 0; |
9505 | } | |
9506 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
9507 | ||
9508 | ||
1da177e4 LT |
9509 | /** |
9510 | * register_netdev - register a network device | |
9511 | * @dev: device to register | |
9512 | * | |
9513 | * Take a completed network device structure and add it to the kernel | |
9514 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
9515 | * chain. 0 is returned on success. A negative errno code is returned | |
9516 | * on a failure to set up the device, or if the name is a duplicate. | |
9517 | * | |
38b4da38 | 9518 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
9519 | * and expands the device name if you passed a format string to |
9520 | * alloc_netdev. | |
9521 | */ | |
9522 | int register_netdev(struct net_device *dev) | |
9523 | { | |
9524 | int err; | |
9525 | ||
b0f3debc KT |
9526 | if (rtnl_lock_killable()) |
9527 | return -EINTR; | |
1da177e4 | 9528 | err = register_netdevice(dev); |
1da177e4 LT |
9529 | rtnl_unlock(); |
9530 | return err; | |
9531 | } | |
9532 | EXPORT_SYMBOL(register_netdev); | |
9533 | ||
29b4433d ED |
9534 | int netdev_refcnt_read(const struct net_device *dev) |
9535 | { | |
9536 | int i, refcnt = 0; | |
9537 | ||
9538 | for_each_possible_cpu(i) | |
9539 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
9540 | return refcnt; | |
9541 | } | |
9542 | EXPORT_SYMBOL(netdev_refcnt_read); | |
9543 | ||
2c53040f | 9544 | /** |
1da177e4 | 9545 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 9546 | * @dev: target net_device |
1da177e4 LT |
9547 | * |
9548 | * This is called when unregistering network devices. | |
9549 | * | |
9550 | * Any protocol or device that holds a reference should register | |
9551 | * for netdevice notification, and cleanup and put back the | |
9552 | * reference if they receive an UNREGISTER event. | |
9553 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 9554 | * call dev_put. |
1da177e4 LT |
9555 | */ |
9556 | static void netdev_wait_allrefs(struct net_device *dev) | |
9557 | { | |
9558 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 9559 | int refcnt; |
1da177e4 | 9560 | |
e014debe ED |
9561 | linkwatch_forget_dev(dev); |
9562 | ||
1da177e4 | 9563 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
9564 | refcnt = netdev_refcnt_read(dev); |
9565 | ||
9566 | while (refcnt != 0) { | |
1da177e4 | 9567 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 9568 | rtnl_lock(); |
1da177e4 LT |
9569 | |
9570 | /* Rebroadcast unregister notification */ | |
056925ab | 9571 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 9572 | |
748e2d93 | 9573 | __rtnl_unlock(); |
0115e8e3 | 9574 | rcu_barrier(); |
748e2d93 ED |
9575 | rtnl_lock(); |
9576 | ||
1da177e4 LT |
9577 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
9578 | &dev->state)) { | |
9579 | /* We must not have linkwatch events | |
9580 | * pending on unregister. If this | |
9581 | * happens, we simply run the queue | |
9582 | * unscheduled, resulting in a noop | |
9583 | * for this device. | |
9584 | */ | |
9585 | linkwatch_run_queue(); | |
9586 | } | |
9587 | ||
6756ae4b | 9588 | __rtnl_unlock(); |
1da177e4 LT |
9589 | |
9590 | rebroadcast_time = jiffies; | |
9591 | } | |
9592 | ||
9593 | msleep(250); | |
9594 | ||
29b4433d ED |
9595 | refcnt = netdev_refcnt_read(dev); |
9596 | ||
d7c04b05 | 9597 | if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
9598 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
9599 | dev->name, refcnt); | |
1da177e4 LT |
9600 | warning_time = jiffies; |
9601 | } | |
9602 | } | |
9603 | } | |
9604 | ||
9605 | /* The sequence is: | |
9606 | * | |
9607 | * rtnl_lock(); | |
9608 | * ... | |
9609 | * register_netdevice(x1); | |
9610 | * register_netdevice(x2); | |
9611 | * ... | |
9612 | * unregister_netdevice(y1); | |
9613 | * unregister_netdevice(y2); | |
9614 | * ... | |
9615 | * rtnl_unlock(); | |
9616 | * free_netdev(y1); | |
9617 | * free_netdev(y2); | |
9618 | * | |
58ec3b4d | 9619 | * We are invoked by rtnl_unlock(). |
1da177e4 | 9620 | * This allows us to deal with problems: |
b17a7c17 | 9621 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
9622 | * without deadlocking with linkwatch via keventd. |
9623 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
9624 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
9625 | * |
9626 | * We must not return until all unregister events added during | |
9627 | * the interval the lock was held have been completed. | |
1da177e4 | 9628 | */ |
1da177e4 LT |
9629 | void netdev_run_todo(void) |
9630 | { | |
626ab0e6 | 9631 | struct list_head list; |
1da177e4 | 9632 | |
1da177e4 | 9633 | /* Snapshot list, allow later requests */ |
626ab0e6 | 9634 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
9635 | |
9636 | __rtnl_unlock(); | |
626ab0e6 | 9637 | |
0115e8e3 ED |
9638 | |
9639 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
9640 | if (!list_empty(&list)) |
9641 | rcu_barrier(); | |
9642 | ||
1da177e4 LT |
9643 | while (!list_empty(&list)) { |
9644 | struct net_device *dev | |
e5e26d75 | 9645 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
9646 | list_del(&dev->todo_list); |
9647 | ||
b17a7c17 | 9648 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 9649 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
9650 | dev->name, dev->reg_state); |
9651 | dump_stack(); | |
9652 | continue; | |
9653 | } | |
1da177e4 | 9654 | |
b17a7c17 | 9655 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 9656 | |
b17a7c17 | 9657 | netdev_wait_allrefs(dev); |
1da177e4 | 9658 | |
b17a7c17 | 9659 | /* paranoia */ |
29b4433d | 9660 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
9661 | BUG_ON(!list_empty(&dev->ptype_all)); |
9662 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
9663 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
9664 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
330c7272 | 9665 | #if IS_ENABLED(CONFIG_DECNET) |
547b792c | 9666 | WARN_ON(dev->dn_ptr); |
330c7272 | 9667 | #endif |
cf124db5 DM |
9668 | if (dev->priv_destructor) |
9669 | dev->priv_destructor(dev); | |
9670 | if (dev->needs_free_netdev) | |
9671 | free_netdev(dev); | |
9093bbb2 | 9672 | |
50624c93 EB |
9673 | /* Report a network device has been unregistered */ |
9674 | rtnl_lock(); | |
9675 | dev_net(dev)->dev_unreg_count--; | |
9676 | __rtnl_unlock(); | |
9677 | wake_up(&netdev_unregistering_wq); | |
9678 | ||
9093bbb2 SH |
9679 | /* Free network device */ |
9680 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 9681 | } |
1da177e4 LT |
9682 | } |
9683 | ||
9256645a JW |
9684 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
9685 | * all the same fields in the same order as net_device_stats, with only | |
9686 | * the type differing, but rtnl_link_stats64 may have additional fields | |
9687 | * at the end for newer counters. | |
3cfde79c | 9688 | */ |
77a1abf5 ED |
9689 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
9690 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
9691 | { |
9692 | #if BITS_PER_LONG == 64 | |
9256645a | 9693 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
9af9959e | 9694 | memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); |
9256645a JW |
9695 | /* zero out counters that only exist in rtnl_link_stats64 */ |
9696 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
9697 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 9698 | #else |
9256645a | 9699 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
9700 | const unsigned long *src = (const unsigned long *)netdev_stats; |
9701 | u64 *dst = (u64 *)stats64; | |
9702 | ||
9256645a | 9703 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
9704 | for (i = 0; i < n; i++) |
9705 | dst[i] = src[i]; | |
9256645a JW |
9706 | /* zero out counters that only exist in rtnl_link_stats64 */ |
9707 | memset((char *)stats64 + n * sizeof(u64), 0, | |
9708 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
9709 | #endif |
9710 | } | |
77a1abf5 | 9711 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 9712 | |
eeda3fd6 SH |
9713 | /** |
9714 | * dev_get_stats - get network device statistics | |
9715 | * @dev: device to get statistics from | |
28172739 | 9716 | * @storage: place to store stats |
eeda3fd6 | 9717 | * |
d7753516 BH |
9718 | * Get network statistics from device. Return @storage. |
9719 | * The device driver may provide its own method by setting | |
9720 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
9721 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 9722 | */ |
d7753516 BH |
9723 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
9724 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 9725 | { |
eeda3fd6 SH |
9726 | const struct net_device_ops *ops = dev->netdev_ops; |
9727 | ||
28172739 ED |
9728 | if (ops->ndo_get_stats64) { |
9729 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
9730 | ops->ndo_get_stats64(dev, storage); |
9731 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 9732 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
9733 | } else { |
9734 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 9735 | } |
6f64ec74 ED |
9736 | storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); |
9737 | storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); | |
9738 | storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); | |
28172739 | 9739 | return storage; |
c45d286e | 9740 | } |
eeda3fd6 | 9741 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 9742 | |
24824a09 | 9743 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 9744 | { |
24824a09 | 9745 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 9746 | |
24824a09 ED |
9747 | #ifdef CONFIG_NET_CLS_ACT |
9748 | if (queue) | |
9749 | return queue; | |
9750 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
9751 | if (!queue) | |
9752 | return NULL; | |
9753 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 9754 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
9755 | queue->qdisc_sleeping = &noop_qdisc; |
9756 | rcu_assign_pointer(dev->ingress_queue, queue); | |
9757 | #endif | |
9758 | return queue; | |
bb949fbd DM |
9759 | } |
9760 | ||
2c60db03 ED |
9761 | static const struct ethtool_ops default_ethtool_ops; |
9762 | ||
d07d7507 SG |
9763 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
9764 | const struct ethtool_ops *ops) | |
9765 | { | |
9766 | if (dev->ethtool_ops == &default_ethtool_ops) | |
9767 | dev->ethtool_ops = ops; | |
9768 | } | |
9769 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
9770 | ||
74d332c1 ED |
9771 | void netdev_freemem(struct net_device *dev) |
9772 | { | |
9773 | char *addr = (char *)dev - dev->padded; | |
9774 | ||
4cb28970 | 9775 | kvfree(addr); |
74d332c1 ED |
9776 | } |
9777 | ||
1da177e4 | 9778 | /** |
722c9a0c | 9779 | * alloc_netdev_mqs - allocate network device |
9780 | * @sizeof_priv: size of private data to allocate space for | |
9781 | * @name: device name format string | |
9782 | * @name_assign_type: origin of device name | |
9783 | * @setup: callback to initialize device | |
9784 | * @txqs: the number of TX subqueues to allocate | |
9785 | * @rxqs: the number of RX subqueues to allocate | |
9786 | * | |
9787 | * Allocates a struct net_device with private data area for driver use | |
9788 | * and performs basic initialization. Also allocates subqueue structs | |
9789 | * for each queue on the device. | |
1da177e4 | 9790 | */ |
36909ea4 | 9791 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 9792 | unsigned char name_assign_type, |
36909ea4 TH |
9793 | void (*setup)(struct net_device *), |
9794 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 9795 | { |
1da177e4 | 9796 | struct net_device *dev; |
52a59bd5 | 9797 | unsigned int alloc_size; |
1ce8e7b5 | 9798 | struct net_device *p; |
1da177e4 | 9799 | |
b6fe17d6 SH |
9800 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
9801 | ||
36909ea4 | 9802 | if (txqs < 1) { |
7b6cd1ce | 9803 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
9804 | return NULL; |
9805 | } | |
9806 | ||
36909ea4 | 9807 | if (rxqs < 1) { |
7b6cd1ce | 9808 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
9809 | return NULL; |
9810 | } | |
36909ea4 | 9811 | |
fd2ea0a7 | 9812 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
9813 | if (sizeof_priv) { |
9814 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 9815 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
9816 | alloc_size += sizeof_priv; |
9817 | } | |
9818 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 9819 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 9820 | |
dcda9b04 | 9821 | p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
62b5942a | 9822 | if (!p) |
1da177e4 | 9823 | return NULL; |
1da177e4 | 9824 | |
1ce8e7b5 | 9825 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 9826 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 9827 | |
29b4433d ED |
9828 | dev->pcpu_refcnt = alloc_percpu(int); |
9829 | if (!dev->pcpu_refcnt) | |
74d332c1 | 9830 | goto free_dev; |
ab9c73cc | 9831 | |
ab9c73cc | 9832 | if (dev_addr_init(dev)) |
29b4433d | 9833 | goto free_pcpu; |
ab9c73cc | 9834 | |
22bedad3 | 9835 | dev_mc_init(dev); |
a748ee24 | 9836 | dev_uc_init(dev); |
ccffad25 | 9837 | |
c346dca1 | 9838 | dev_net_set(dev, &init_net); |
1da177e4 | 9839 | |
ab92d68f TY |
9840 | netdev_register_lockdep_key(dev); |
9841 | ||
8d3bdbd5 | 9842 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 9843 | dev->gso_max_segs = GSO_MAX_SEGS; |
5343da4c TY |
9844 | dev->upper_level = 1; |
9845 | dev->lower_level = 1; | |
8d3bdbd5 | 9846 | |
8d3bdbd5 DM |
9847 | INIT_LIST_HEAD(&dev->napi_list); |
9848 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 9849 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 9850 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
9851 | INIT_LIST_HEAD(&dev->adj_list.upper); |
9852 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
9853 | INIT_LIST_HEAD(&dev->ptype_all); |
9854 | INIT_LIST_HEAD(&dev->ptype_specific); | |
93642e14 | 9855 | INIT_LIST_HEAD(&dev->net_notifier_list); |
59cc1f61 JK |
9856 | #ifdef CONFIG_NET_SCHED |
9857 | hash_init(dev->qdisc_hash); | |
9858 | #endif | |
02875878 | 9859 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
9860 | setup(dev); |
9861 | ||
a813104d | 9862 | if (!dev->tx_queue_len) { |
f84bb1ea | 9863 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 9864 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 9865 | } |
906470c1 | 9866 | |
36909ea4 TH |
9867 | dev->num_tx_queues = txqs; |
9868 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 9869 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 9870 | goto free_all; |
e8a0464c | 9871 | |
36909ea4 TH |
9872 | dev->num_rx_queues = rxqs; |
9873 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 9874 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 9875 | goto free_all; |
0a9627f2 | 9876 | |
1da177e4 | 9877 | strcpy(dev->name, name); |
c835a677 | 9878 | dev->name_assign_type = name_assign_type; |
cbda10fa | 9879 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
9880 | if (!dev->ethtool_ops) |
9881 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 | 9882 | |
357b6cc5 | 9883 | nf_hook_ingress_init(dev); |
e687ad60 | 9884 | |
1da177e4 | 9885 | return dev; |
ab9c73cc | 9886 | |
8d3bdbd5 DM |
9887 | free_all: |
9888 | free_netdev(dev); | |
9889 | return NULL; | |
9890 | ||
29b4433d ED |
9891 | free_pcpu: |
9892 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
9893 | free_dev: |
9894 | netdev_freemem(dev); | |
ab9c73cc | 9895 | return NULL; |
1da177e4 | 9896 | } |
36909ea4 | 9897 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
9898 | |
9899 | /** | |
722c9a0c | 9900 | * free_netdev - free network device |
9901 | * @dev: device | |
1da177e4 | 9902 | * |
722c9a0c | 9903 | * This function does the last stage of destroying an allocated device |
9904 | * interface. The reference to the device object is released. If this | |
9905 | * is the last reference then it will be freed.Must be called in process | |
9906 | * context. | |
1da177e4 LT |
9907 | */ |
9908 | void free_netdev(struct net_device *dev) | |
9909 | { | |
d565b0a1 HX |
9910 | struct napi_struct *p, *n; |
9911 | ||
93d05d4a | 9912 | might_sleep(); |
60877a32 | 9913 | netif_free_tx_queues(dev); |
e817f856 | 9914 | netif_free_rx_queues(dev); |
e8a0464c | 9915 | |
33d480ce | 9916 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 9917 | |
f001fde5 JP |
9918 | /* Flush device addresses */ |
9919 | dev_addr_flush(dev); | |
9920 | ||
d565b0a1 HX |
9921 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
9922 | netif_napi_del(p); | |
9923 | ||
29b4433d ED |
9924 | free_percpu(dev->pcpu_refcnt); |
9925 | dev->pcpu_refcnt = NULL; | |
75ccae62 THJ |
9926 | free_percpu(dev->xdp_bulkq); |
9927 | dev->xdp_bulkq = NULL; | |
29b4433d | 9928 | |
ab92d68f TY |
9929 | netdev_unregister_lockdep_key(dev); |
9930 | ||
3041a069 | 9931 | /* Compatibility with error handling in drivers */ |
1da177e4 | 9932 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 9933 | netdev_freemem(dev); |
1da177e4 LT |
9934 | return; |
9935 | } | |
9936 | ||
9937 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
9938 | dev->reg_state = NETREG_RELEASED; | |
9939 | ||
43cb76d9 GKH |
9940 | /* will free via device release */ |
9941 | put_device(&dev->dev); | |
1da177e4 | 9942 | } |
d1b19dff | 9943 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 9944 | |
f0db275a SH |
9945 | /** |
9946 | * synchronize_net - Synchronize with packet receive processing | |
9947 | * | |
9948 | * Wait for packets currently being received to be done. | |
9949 | * Does not block later packets from starting. | |
9950 | */ | |
4ec93edb | 9951 | void synchronize_net(void) |
1da177e4 LT |
9952 | { |
9953 | might_sleep(); | |
be3fc413 ED |
9954 | if (rtnl_is_locked()) |
9955 | synchronize_rcu_expedited(); | |
9956 | else | |
9957 | synchronize_rcu(); | |
1da177e4 | 9958 | } |
d1b19dff | 9959 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
9960 | |
9961 | /** | |
44a0873d | 9962 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 9963 | * @dev: device |
44a0873d | 9964 | * @head: list |
6ebfbc06 | 9965 | * |
1da177e4 | 9966 | * This function shuts down a device interface and removes it |
d59b54b1 | 9967 | * from the kernel tables. |
44a0873d | 9968 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
9969 | * |
9970 | * Callers must hold the rtnl semaphore. You may want | |
9971 | * unregister_netdev() instead of this. | |
9972 | */ | |
9973 | ||
44a0873d | 9974 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 9975 | { |
a6620712 HX |
9976 | ASSERT_RTNL(); |
9977 | ||
44a0873d | 9978 | if (head) { |
9fdce099 | 9979 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
9980 | } else { |
9981 | rollback_registered(dev); | |
9982 | /* Finish processing unregister after unlock */ | |
9983 | net_set_todo(dev); | |
9984 | } | |
1da177e4 | 9985 | } |
44a0873d | 9986 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 9987 | |
9b5e383c ED |
9988 | /** |
9989 | * unregister_netdevice_many - unregister many devices | |
9990 | * @head: list of devices | |
87757a91 ED |
9991 | * |
9992 | * Note: As most callers use a stack allocated list_head, | |
9993 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
9994 | */ |
9995 | void unregister_netdevice_many(struct list_head *head) | |
9996 | { | |
9997 | struct net_device *dev; | |
9998 | ||
9999 | if (!list_empty(head)) { | |
10000 | rollback_registered_many(head); | |
10001 | list_for_each_entry(dev, head, unreg_list) | |
10002 | net_set_todo(dev); | |
87757a91 | 10003 | list_del(head); |
9b5e383c ED |
10004 | } |
10005 | } | |
63c8099d | 10006 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 10007 | |
1da177e4 LT |
10008 | /** |
10009 | * unregister_netdev - remove device from the kernel | |
10010 | * @dev: device | |
10011 | * | |
10012 | * This function shuts down a device interface and removes it | |
d59b54b1 | 10013 | * from the kernel tables. |
1da177e4 LT |
10014 | * |
10015 | * This is just a wrapper for unregister_netdevice that takes | |
10016 | * the rtnl semaphore. In general you want to use this and not | |
10017 | * unregister_netdevice. | |
10018 | */ | |
10019 | void unregister_netdev(struct net_device *dev) | |
10020 | { | |
10021 | rtnl_lock(); | |
10022 | unregister_netdevice(dev); | |
10023 | rtnl_unlock(); | |
10024 | } | |
1da177e4 LT |
10025 | EXPORT_SYMBOL(unregister_netdev); |
10026 | ||
ce286d32 EB |
10027 | /** |
10028 | * dev_change_net_namespace - move device to different nethost namespace | |
10029 | * @dev: device | |
10030 | * @net: network namespace | |
10031 | * @pat: If not NULL name pattern to try if the current device name | |
10032 | * is already taken in the destination network namespace. | |
10033 | * | |
10034 | * This function shuts down a device interface and moves it | |
10035 | * to a new network namespace. On success 0 is returned, on | |
10036 | * a failure a netagive errno code is returned. | |
10037 | * | |
10038 | * Callers must hold the rtnl semaphore. | |
10039 | */ | |
10040 | ||
10041 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
10042 | { | |
ef6a4c88 | 10043 | struct net *net_old = dev_net(dev); |
38e01b30 | 10044 | int err, new_nsid, new_ifindex; |
ce286d32 EB |
10045 | |
10046 | ASSERT_RTNL(); | |
10047 | ||
10048 | /* Don't allow namespace local devices to be moved. */ | |
10049 | err = -EINVAL; | |
10050 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
10051 | goto out; | |
10052 | ||
10053 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
10054 | if (dev->reg_state != NETREG_REGISTERED) |
10055 | goto out; | |
10056 | ||
10057 | /* Get out if there is nothing todo */ | |
10058 | err = 0; | |
ef6a4c88 | 10059 | if (net_eq(net_old, net)) |
ce286d32 EB |
10060 | goto out; |
10061 | ||
10062 | /* Pick the destination device name, and ensure | |
10063 | * we can use it in the destination network namespace. | |
10064 | */ | |
10065 | err = -EEXIST; | |
d9031024 | 10066 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
10067 | /* We get here if we can't use the current device name */ |
10068 | if (!pat) | |
10069 | goto out; | |
7892bd08 LR |
10070 | err = dev_get_valid_name(net, dev, pat); |
10071 | if (err < 0) | |
ce286d32 EB |
10072 | goto out; |
10073 | } | |
10074 | ||
10075 | /* | |
10076 | * And now a mini version of register_netdevice unregister_netdevice. | |
10077 | */ | |
10078 | ||
10079 | /* If device is running close it first. */ | |
9b772652 | 10080 | dev_close(dev); |
ce286d32 EB |
10081 | |
10082 | /* And unlink it from device chain */ | |
ce286d32 EB |
10083 | unlist_netdevice(dev); |
10084 | ||
10085 | synchronize_net(); | |
10086 | ||
10087 | /* Shutdown queueing discipline. */ | |
10088 | dev_shutdown(dev); | |
10089 | ||
10090 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 10091 | * this device. They should clean all the things. |
10092 | * | |
10093 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
10094 | * This is wanted because this way 8021q and macvlan know | |
10095 | * the device is just moving and can keep their slaves up. | |
10096 | */ | |
ce286d32 | 10097 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 | 10098 | rcu_barrier(); |
38e01b30 | 10099 | |
d4e4fdf9 | 10100 | new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); |
38e01b30 ND |
10101 | /* If there is an ifindex conflict assign a new one */ |
10102 | if (__dev_get_by_index(net, dev->ifindex)) | |
10103 | new_ifindex = dev_new_index(net); | |
10104 | else | |
10105 | new_ifindex = dev->ifindex; | |
10106 | ||
10107 | rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, | |
10108 | new_ifindex); | |
ce286d32 EB |
10109 | |
10110 | /* | |
10111 | * Flush the unicast and multicast chains | |
10112 | */ | |
a748ee24 | 10113 | dev_uc_flush(dev); |
22bedad3 | 10114 | dev_mc_flush(dev); |
ce286d32 | 10115 | |
4e66ae2e SH |
10116 | /* Send a netdev-removed uevent to the old namespace */ |
10117 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 10118 | netdev_adjacent_del_links(dev); |
4e66ae2e | 10119 | |
93642e14 JP |
10120 | /* Move per-net netdevice notifiers that are following the netdevice */ |
10121 | move_netdevice_notifiers_dev_net(dev, net); | |
10122 | ||
ce286d32 | 10123 | /* Actually switch the network namespace */ |
c346dca1 | 10124 | dev_net_set(dev, net); |
38e01b30 | 10125 | dev->ifindex = new_ifindex; |
ce286d32 | 10126 | |
4e66ae2e SH |
10127 | /* Send a netdev-add uevent to the new namespace */ |
10128 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 10129 | netdev_adjacent_add_links(dev); |
4e66ae2e | 10130 | |
8b41d188 | 10131 | /* Fixup kobjects */ |
a1b3f594 | 10132 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 10133 | WARN_ON(err); |
ce286d32 | 10134 | |
ef6a4c88 CB |
10135 | /* Adapt owner in case owning user namespace of target network |
10136 | * namespace is different from the original one. | |
10137 | */ | |
10138 | err = netdev_change_owner(dev, net_old, net); | |
10139 | WARN_ON(err); | |
10140 | ||
ce286d32 EB |
10141 | /* Add the device back in the hashes */ |
10142 | list_netdevice(dev); | |
10143 | ||
10144 | /* Notify protocols, that a new device appeared. */ | |
10145 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
10146 | ||
d90a909e EB |
10147 | /* |
10148 | * Prevent userspace races by waiting until the network | |
10149 | * device is fully setup before sending notifications. | |
10150 | */ | |
7f294054 | 10151 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 10152 | |
ce286d32 EB |
10153 | synchronize_net(); |
10154 | err = 0; | |
10155 | out: | |
10156 | return err; | |
10157 | } | |
463d0183 | 10158 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 10159 | |
f0bf90de | 10160 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
10161 | { |
10162 | struct sk_buff **list_skb; | |
1da177e4 | 10163 | struct sk_buff *skb; |
f0bf90de | 10164 | unsigned int cpu; |
97d8b6e3 | 10165 | struct softnet_data *sd, *oldsd, *remsd = NULL; |
1da177e4 | 10166 | |
1da177e4 LT |
10167 | local_irq_disable(); |
10168 | cpu = smp_processor_id(); | |
10169 | sd = &per_cpu(softnet_data, cpu); | |
10170 | oldsd = &per_cpu(softnet_data, oldcpu); | |
10171 | ||
10172 | /* Find end of our completion_queue. */ | |
10173 | list_skb = &sd->completion_queue; | |
10174 | while (*list_skb) | |
10175 | list_skb = &(*list_skb)->next; | |
10176 | /* Append completion queue from offline CPU. */ | |
10177 | *list_skb = oldsd->completion_queue; | |
10178 | oldsd->completion_queue = NULL; | |
10179 | ||
1da177e4 | 10180 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
10181 | if (oldsd->output_queue) { |
10182 | *sd->output_queue_tailp = oldsd->output_queue; | |
10183 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
10184 | oldsd->output_queue = NULL; | |
10185 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
10186 | } | |
ac64da0b ED |
10187 | /* Append NAPI poll list from offline CPU, with one exception : |
10188 | * process_backlog() must be called by cpu owning percpu backlog. | |
10189 | * We properly handle process_queue & input_pkt_queue later. | |
10190 | */ | |
10191 | while (!list_empty(&oldsd->poll_list)) { | |
10192 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
10193 | struct napi_struct, | |
10194 | poll_list); | |
10195 | ||
10196 | list_del_init(&napi->poll_list); | |
10197 | if (napi->poll == process_backlog) | |
10198 | napi->state = 0; | |
10199 | else | |
10200 | ____napi_schedule(sd, napi); | |
264524d5 | 10201 | } |
1da177e4 LT |
10202 | |
10203 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
10204 | local_irq_enable(); | |
10205 | ||
773fc8f6 | 10206 | #ifdef CONFIG_RPS |
10207 | remsd = oldsd->rps_ipi_list; | |
10208 | oldsd->rps_ipi_list = NULL; | |
10209 | #endif | |
10210 | /* send out pending IPI's on offline CPU */ | |
10211 | net_rps_send_ipi(remsd); | |
10212 | ||
1da177e4 | 10213 | /* Process offline CPU's input_pkt_queue */ |
76cc8b13 | 10214 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 10215 | netif_rx_ni(skb); |
76cc8b13 | 10216 | input_queue_head_incr(oldsd); |
fec5e652 | 10217 | } |
ac64da0b | 10218 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 10219 | netif_rx_ni(skb); |
76cc8b13 TH |
10220 | input_queue_head_incr(oldsd); |
10221 | } | |
1da177e4 | 10222 | |
f0bf90de | 10223 | return 0; |
1da177e4 | 10224 | } |
1da177e4 | 10225 | |
7f353bf2 | 10226 | /** |
b63365a2 HX |
10227 | * netdev_increment_features - increment feature set by one |
10228 | * @all: current feature set | |
10229 | * @one: new feature set | |
10230 | * @mask: mask feature set | |
7f353bf2 HX |
10231 | * |
10232 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
10233 | * @one to the master device with current feature set @all. Will not |
10234 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 10235 | */ |
c8f44aff MM |
10236 | netdev_features_t netdev_increment_features(netdev_features_t all, |
10237 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 10238 | { |
c8cd0989 | 10239 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 10240 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 10241 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 10242 | |
a188222b | 10243 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 10244 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 10245 | |
1742f183 | 10246 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
10247 | if (all & NETIF_F_HW_CSUM) |
10248 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
10249 | |
10250 | return all; | |
10251 | } | |
b63365a2 | 10252 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 10253 | |
430f03cd | 10254 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
10255 | { |
10256 | int i; | |
10257 | struct hlist_head *hash; | |
10258 | ||
6da2ec56 | 10259 | hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); |
30d97d35 PE |
10260 | if (hash != NULL) |
10261 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
10262 | INIT_HLIST_HEAD(&hash[i]); | |
10263 | ||
10264 | return hash; | |
10265 | } | |
10266 | ||
881d966b | 10267 | /* Initialize per network namespace state */ |
4665079c | 10268 | static int __net_init netdev_init(struct net *net) |
881d966b | 10269 | { |
d9f37d01 | 10270 | BUILD_BUG_ON(GRO_HASH_BUCKETS > |
c593642c | 10271 | 8 * sizeof_field(struct napi_struct, gro_bitmask)); |
d9f37d01 | 10272 | |
734b6541 RM |
10273 | if (net != &init_net) |
10274 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 10275 | |
30d97d35 PE |
10276 | net->dev_name_head = netdev_create_hash(); |
10277 | if (net->dev_name_head == NULL) | |
10278 | goto err_name; | |
881d966b | 10279 | |
30d97d35 PE |
10280 | net->dev_index_head = netdev_create_hash(); |
10281 | if (net->dev_index_head == NULL) | |
10282 | goto err_idx; | |
881d966b | 10283 | |
a30c7b42 JP |
10284 | RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); |
10285 | ||
881d966b | 10286 | return 0; |
30d97d35 PE |
10287 | |
10288 | err_idx: | |
10289 | kfree(net->dev_name_head); | |
10290 | err_name: | |
10291 | return -ENOMEM; | |
881d966b EB |
10292 | } |
10293 | ||
f0db275a SH |
10294 | /** |
10295 | * netdev_drivername - network driver for the device | |
10296 | * @dev: network device | |
f0db275a SH |
10297 | * |
10298 | * Determine network driver for device. | |
10299 | */ | |
3019de12 | 10300 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 10301 | { |
cf04a4c7 SH |
10302 | const struct device_driver *driver; |
10303 | const struct device *parent; | |
3019de12 | 10304 | const char *empty = ""; |
6579e57b AV |
10305 | |
10306 | parent = dev->dev.parent; | |
6579e57b | 10307 | if (!parent) |
3019de12 | 10308 | return empty; |
6579e57b AV |
10309 | |
10310 | driver = parent->driver; | |
10311 | if (driver && driver->name) | |
3019de12 DM |
10312 | return driver->name; |
10313 | return empty; | |
6579e57b AV |
10314 | } |
10315 | ||
6ea754eb JP |
10316 | static void __netdev_printk(const char *level, const struct net_device *dev, |
10317 | struct va_format *vaf) | |
256df2f3 | 10318 | { |
b004ff49 | 10319 | if (dev && dev->dev.parent) { |
6ea754eb JP |
10320 | dev_printk_emit(level[1] - '0', |
10321 | dev->dev.parent, | |
10322 | "%s %s %s%s: %pV", | |
10323 | dev_driver_string(dev->dev.parent), | |
10324 | dev_name(dev->dev.parent), | |
10325 | netdev_name(dev), netdev_reg_state(dev), | |
10326 | vaf); | |
b004ff49 | 10327 | } else if (dev) { |
6ea754eb JP |
10328 | printk("%s%s%s: %pV", |
10329 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 10330 | } else { |
6ea754eb | 10331 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 10332 | } |
256df2f3 JP |
10333 | } |
10334 | ||
6ea754eb JP |
10335 | void netdev_printk(const char *level, const struct net_device *dev, |
10336 | const char *format, ...) | |
256df2f3 JP |
10337 | { |
10338 | struct va_format vaf; | |
10339 | va_list args; | |
256df2f3 JP |
10340 | |
10341 | va_start(args, format); | |
10342 | ||
10343 | vaf.fmt = format; | |
10344 | vaf.va = &args; | |
10345 | ||
6ea754eb | 10346 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 10347 | |
256df2f3 | 10348 | va_end(args); |
256df2f3 JP |
10349 | } |
10350 | EXPORT_SYMBOL(netdev_printk); | |
10351 | ||
10352 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 10353 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 10354 | { \ |
256df2f3 JP |
10355 | struct va_format vaf; \ |
10356 | va_list args; \ | |
10357 | \ | |
10358 | va_start(args, fmt); \ | |
10359 | \ | |
10360 | vaf.fmt = fmt; \ | |
10361 | vaf.va = &args; \ | |
10362 | \ | |
6ea754eb | 10363 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 10364 | \ |
256df2f3 | 10365 | va_end(args); \ |
256df2f3 JP |
10366 | } \ |
10367 | EXPORT_SYMBOL(func); | |
10368 | ||
10369 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
10370 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
10371 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
10372 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
10373 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
10374 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
10375 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
10376 | ||
4665079c | 10377 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
10378 | { |
10379 | kfree(net->dev_name_head); | |
10380 | kfree(net->dev_index_head); | |
ee21b18b VA |
10381 | if (net != &init_net) |
10382 | WARN_ON_ONCE(!list_empty(&net->dev_base_head)); | |
881d966b EB |
10383 | } |
10384 | ||
022cbae6 | 10385 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
10386 | .init = netdev_init, |
10387 | .exit = netdev_exit, | |
10388 | }; | |
10389 | ||
4665079c | 10390 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 10391 | { |
e008b5fc | 10392 | struct net_device *dev, *aux; |
ce286d32 | 10393 | /* |
e008b5fc | 10394 | * Push all migratable network devices back to the |
ce286d32 EB |
10395 | * initial network namespace |
10396 | */ | |
10397 | rtnl_lock(); | |
e008b5fc | 10398 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 10399 | int err; |
aca51397 | 10400 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
10401 | |
10402 | /* Ignore unmoveable devices (i.e. loopback) */ | |
10403 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
10404 | continue; | |
10405 | ||
e008b5fc EB |
10406 | /* Leave virtual devices for the generic cleanup */ |
10407 | if (dev->rtnl_link_ops) | |
10408 | continue; | |
d0c082ce | 10409 | |
25985edc | 10410 | /* Push remaining network devices to init_net */ |
aca51397 | 10411 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
55b40dbf JP |
10412 | if (__dev_get_by_name(&init_net, fb_name)) |
10413 | snprintf(fb_name, IFNAMSIZ, "dev%%d"); | |
aca51397 | 10414 | err = dev_change_net_namespace(dev, &init_net, fb_name); |
ce286d32 | 10415 | if (err) { |
7b6cd1ce JP |
10416 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
10417 | __func__, dev->name, err); | |
aca51397 | 10418 | BUG(); |
ce286d32 EB |
10419 | } |
10420 | } | |
10421 | rtnl_unlock(); | |
10422 | } | |
10423 | ||
50624c93 EB |
10424 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
10425 | { | |
10426 | /* Return with the rtnl_lock held when there are no network | |
10427 | * devices unregistering in any network namespace in net_list. | |
10428 | */ | |
10429 | struct net *net; | |
10430 | bool unregistering; | |
ff960a73 | 10431 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 10432 | |
ff960a73 | 10433 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 10434 | for (;;) { |
50624c93 EB |
10435 | unregistering = false; |
10436 | rtnl_lock(); | |
10437 | list_for_each_entry(net, net_list, exit_list) { | |
10438 | if (net->dev_unreg_count > 0) { | |
10439 | unregistering = true; | |
10440 | break; | |
10441 | } | |
10442 | } | |
10443 | if (!unregistering) | |
10444 | break; | |
10445 | __rtnl_unlock(); | |
ff960a73 PZ |
10446 | |
10447 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 10448 | } |
ff960a73 | 10449 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
10450 | } |
10451 | ||
04dc7f6b EB |
10452 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
10453 | { | |
10454 | /* At exit all network devices most be removed from a network | |
b595076a | 10455 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
10456 | * Do this across as many network namespaces as possible to |
10457 | * improve batching efficiency. | |
10458 | */ | |
10459 | struct net_device *dev; | |
10460 | struct net *net; | |
10461 | LIST_HEAD(dev_kill_list); | |
10462 | ||
50624c93 EB |
10463 | /* To prevent network device cleanup code from dereferencing |
10464 | * loopback devices or network devices that have been freed | |
10465 | * wait here for all pending unregistrations to complete, | |
10466 | * before unregistring the loopback device and allowing the | |
10467 | * network namespace be freed. | |
10468 | * | |
10469 | * The netdev todo list containing all network devices | |
10470 | * unregistrations that happen in default_device_exit_batch | |
10471 | * will run in the rtnl_unlock() at the end of | |
10472 | * default_device_exit_batch. | |
10473 | */ | |
10474 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
10475 | list_for_each_entry(net, net_list, exit_list) { |
10476 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 10477 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
10478 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
10479 | else | |
10480 | unregister_netdevice_queue(dev, &dev_kill_list); | |
10481 | } | |
10482 | } | |
10483 | unregister_netdevice_many(&dev_kill_list); | |
10484 | rtnl_unlock(); | |
10485 | } | |
10486 | ||
022cbae6 | 10487 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 10488 | .exit = default_device_exit, |
04dc7f6b | 10489 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
10490 | }; |
10491 | ||
1da177e4 LT |
10492 | /* |
10493 | * Initialize the DEV module. At boot time this walks the device list and | |
10494 | * unhooks any devices that fail to initialise (normally hardware not | |
10495 | * present) and leaves us with a valid list of present and active devices. | |
10496 | * | |
10497 | */ | |
10498 | ||
10499 | /* | |
10500 | * This is called single threaded during boot, so no need | |
10501 | * to take the rtnl semaphore. | |
10502 | */ | |
10503 | static int __init net_dev_init(void) | |
10504 | { | |
10505 | int i, rc = -ENOMEM; | |
10506 | ||
10507 | BUG_ON(!dev_boot_phase); | |
10508 | ||
1da177e4 LT |
10509 | if (dev_proc_init()) |
10510 | goto out; | |
10511 | ||
8b41d188 | 10512 | if (netdev_kobject_init()) |
1da177e4 LT |
10513 | goto out; |
10514 | ||
10515 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 10516 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
10517 | INIT_LIST_HEAD(&ptype_base[i]); |
10518 | ||
62532da9 VY |
10519 | INIT_LIST_HEAD(&offload_base); |
10520 | ||
881d966b EB |
10521 | if (register_pernet_subsys(&netdev_net_ops)) |
10522 | goto out; | |
1da177e4 LT |
10523 | |
10524 | /* | |
10525 | * Initialise the packet receive queues. | |
10526 | */ | |
10527 | ||
6f912042 | 10528 | for_each_possible_cpu(i) { |
41852497 | 10529 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 10530 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 10531 | |
41852497 ED |
10532 | INIT_WORK(flush, flush_backlog); |
10533 | ||
e36fa2f7 | 10534 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 10535 | skb_queue_head_init(&sd->process_queue); |
f53c7239 SK |
10536 | #ifdef CONFIG_XFRM_OFFLOAD |
10537 | skb_queue_head_init(&sd->xfrm_backlog); | |
10538 | #endif | |
e36fa2f7 | 10539 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 10540 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 10541 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
10542 | sd->csd.func = rps_trigger_softirq; |
10543 | sd->csd.info = sd; | |
e36fa2f7 | 10544 | sd->cpu = i; |
1e94d72f | 10545 | #endif |
0a9627f2 | 10546 | |
7c4ec749 | 10547 | init_gro_hash(&sd->backlog); |
e36fa2f7 ED |
10548 | sd->backlog.poll = process_backlog; |
10549 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
10550 | } |
10551 | ||
1da177e4 LT |
10552 | dev_boot_phase = 0; |
10553 | ||
505d4f73 EB |
10554 | /* The loopback device is special if any other network devices |
10555 | * is present in a network namespace the loopback device must | |
10556 | * be present. Since we now dynamically allocate and free the | |
10557 | * loopback device ensure this invariant is maintained by | |
10558 | * keeping the loopback device as the first device on the | |
10559 | * list of network devices. Ensuring the loopback devices | |
10560 | * is the first device that appears and the last network device | |
10561 | * that disappears. | |
10562 | */ | |
10563 | if (register_pernet_device(&loopback_net_ops)) | |
10564 | goto out; | |
10565 | ||
10566 | if (register_pernet_device(&default_device_ops)) | |
10567 | goto out; | |
10568 | ||
962cf36c CM |
10569 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
10570 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 10571 | |
f0bf90de SAS |
10572 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
10573 | NULL, dev_cpu_dead); | |
10574 | WARN_ON(rc < 0); | |
1da177e4 LT |
10575 | rc = 0; |
10576 | out: | |
10577 | return rc; | |
10578 | } | |
10579 | ||
10580 | subsys_initcall(net_dev_init); |