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> |
11d6011c | 82 | #include <linux/rwsem.h> |
1da177e4 LT |
83 | #include <linux/string.h> |
84 | #include <linux/mm.h> | |
85 | #include <linux/socket.h> | |
86 | #include <linux/sockios.h> | |
87 | #include <linux/errno.h> | |
88 | #include <linux/interrupt.h> | |
89 | #include <linux/if_ether.h> | |
90 | #include <linux/netdevice.h> | |
91 | #include <linux/etherdevice.h> | |
0187bdfb | 92 | #include <linux/ethtool.h> |
1da177e4 | 93 | #include <linux/skbuff.h> |
29863d41 | 94 | #include <linux/kthread.h> |
a7862b45 | 95 | #include <linux/bpf.h> |
b5cdae32 | 96 | #include <linux/bpf_trace.h> |
457c4cbc | 97 | #include <net/net_namespace.h> |
1da177e4 | 98 | #include <net/sock.h> |
02d62e86 | 99 | #include <net/busy_poll.h> |
1da177e4 | 100 | #include <linux/rtnetlink.h> |
1da177e4 | 101 | #include <linux/stat.h> |
b14a9fc4 | 102 | #include <net/dsa.h> |
1da177e4 | 103 | #include <net/dst.h> |
fc4099f1 | 104 | #include <net/dst_metadata.h> |
04f00ab2 | 105 | #include <net/gro.h> |
1da177e4 | 106 | #include <net/pkt_sched.h> |
87d83093 | 107 | #include <net/pkt_cls.h> |
1da177e4 | 108 | #include <net/checksum.h> |
44540960 | 109 | #include <net/xfrm.h> |
1da177e4 LT |
110 | #include <linux/highmem.h> |
111 | #include <linux/init.h> | |
1da177e4 | 112 | #include <linux/module.h> |
1da177e4 LT |
113 | #include <linux/netpoll.h> |
114 | #include <linux/rcupdate.h> | |
115 | #include <linux/delay.h> | |
1da177e4 | 116 | #include <net/iw_handler.h> |
1da177e4 | 117 | #include <asm/current.h> |
5bdb9886 | 118 | #include <linux/audit.h> |
db217334 | 119 | #include <linux/dmaengine.h> |
f6a78bfc | 120 | #include <linux/err.h> |
c7fa9d18 | 121 | #include <linux/ctype.h> |
723e98b7 | 122 | #include <linux/if_arp.h> |
6de329e2 | 123 | #include <linux/if_vlan.h> |
8f0f2223 | 124 | #include <linux/ip.h> |
ad55dcaf | 125 | #include <net/ip.h> |
25cd9ba0 | 126 | #include <net/mpls.h> |
8f0f2223 DM |
127 | #include <linux/ipv6.h> |
128 | #include <linux/in.h> | |
b6b2fed1 DM |
129 | #include <linux/jhash.h> |
130 | #include <linux/random.h> | |
9cbc1cb8 | 131 | #include <trace/events/napi.h> |
cf66ba58 | 132 | #include <trace/events/net.h> |
07dc22e7 | 133 | #include <trace/events/skb.h> |
caeda9b9 | 134 | #include <linux/inetdevice.h> |
c445477d | 135 | #include <linux/cpu_rmap.h> |
c5905afb | 136 | #include <linux/static_key.h> |
af12fa6e | 137 | #include <linux/hashtable.h> |
60877a32 | 138 | #include <linux/vmalloc.h> |
529d0489 | 139 | #include <linux/if_macvlan.h> |
e7fd2885 | 140 | #include <linux/errqueue.h> |
3b47d303 | 141 | #include <linux/hrtimer.h> |
357b6cc5 | 142 | #include <linux/netfilter_ingress.h> |
40e4e713 | 143 | #include <linux/crash_dump.h> |
b72b5bf6 | 144 | #include <linux/sctp.h> |
ae847f40 | 145 | #include <net/udp_tunnel.h> |
6621dd29 | 146 | #include <linux/net_namespace.h> |
aaa5d90b | 147 | #include <linux/indirect_call_wrapper.h> |
af3836df | 148 | #include <net/devlink.h> |
bd869245 | 149 | #include <linux/pm_runtime.h> |
3744741a | 150 | #include <linux/prandom.h> |
1da177e4 | 151 | |
342709ef PE |
152 | #include "net-sysfs.h" |
153 | ||
d565b0a1 HX |
154 | #define MAX_GRO_SKBS 8 |
155 | ||
5d38a079 HX |
156 | /* This should be increased if a protocol with a bigger head is added. */ |
157 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
158 | ||
1da177e4 | 159 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 160 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
161 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
162 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 163 | static struct list_head offload_base __read_mostly; |
1da177e4 | 164 | |
ae78dbfa | 165 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 | 166 | static int call_netdevice_notifiers_info(unsigned long val, |
54951194 | 167 | struct netdev_notifier_info *info); |
26372605 PM |
168 | static int call_netdevice_notifiers_extack(unsigned long val, |
169 | struct net_device *dev, | |
170 | struct netlink_ext_ack *extack); | |
90b602f8 | 171 | static struct napi_struct *napi_by_id(unsigned int napi_id); |
ae78dbfa | 172 | |
1da177e4 | 173 | /* |
7562f876 | 174 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
175 | * semaphore. |
176 | * | |
c6d14c84 | 177 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
178 | * |
179 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 180 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
181 | * actual updates. This allows pure readers to access the list even |
182 | * while a writer is preparing to update it. | |
183 | * | |
184 | * To put it another way, dev_base_lock is held for writing only to | |
185 | * protect against pure readers; the rtnl semaphore provides the | |
186 | * protection against other writers. | |
187 | * | |
188 | * See, for example usages, register_netdevice() and | |
189 | * unregister_netdevice(), which must be called with the rtnl | |
190 | * semaphore held. | |
191 | */ | |
1da177e4 | 192 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
193 | EXPORT_SYMBOL(dev_base_lock); |
194 | ||
6c557001 FW |
195 | static DEFINE_MUTEX(ifalias_mutex); |
196 | ||
af12fa6e ET |
197 | /* protects napi_hash addition/deletion and napi_gen_id */ |
198 | static DEFINE_SPINLOCK(napi_hash_lock); | |
199 | ||
52bd2d62 | 200 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 201 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 202 | |
11d6011c | 203 | static DECLARE_RWSEM(devnet_rename_sem); |
c91f6df2 | 204 | |
4e985ada TG |
205 | static inline void dev_base_seq_inc(struct net *net) |
206 | { | |
643aa9cb | 207 | while (++net->dev_base_seq == 0) |
208 | ; | |
4e985ada TG |
209 | } |
210 | ||
881d966b | 211 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 212 | { |
8387ff25 | 213 | unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); |
95c96174 | 214 | |
08e9897d | 215 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
216 | } |
217 | ||
881d966b | 218 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 219 | { |
7c28bd0b | 220 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
221 | } |
222 | ||
e36fa2f7 | 223 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
224 | { |
225 | #ifdef CONFIG_RPS | |
e36fa2f7 | 226 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
227 | #endif |
228 | } | |
229 | ||
e36fa2f7 | 230 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
231 | { |
232 | #ifdef CONFIG_RPS | |
e36fa2f7 | 233 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
234 | #endif |
235 | } | |
236 | ||
ff927412 JP |
237 | static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, |
238 | const char *name) | |
239 | { | |
240 | struct netdev_name_node *name_node; | |
241 | ||
242 | name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); | |
243 | if (!name_node) | |
244 | return NULL; | |
245 | INIT_HLIST_NODE(&name_node->hlist); | |
246 | name_node->dev = dev; | |
247 | name_node->name = name; | |
248 | return name_node; | |
249 | } | |
250 | ||
251 | static struct netdev_name_node * | |
252 | netdev_name_node_head_alloc(struct net_device *dev) | |
253 | { | |
36fbf1e5 JP |
254 | struct netdev_name_node *name_node; |
255 | ||
256 | name_node = netdev_name_node_alloc(dev, dev->name); | |
257 | if (!name_node) | |
258 | return NULL; | |
259 | INIT_LIST_HEAD(&name_node->list); | |
260 | return name_node; | |
ff927412 JP |
261 | } |
262 | ||
263 | static void netdev_name_node_free(struct netdev_name_node *name_node) | |
264 | { | |
265 | kfree(name_node); | |
266 | } | |
267 | ||
268 | static void netdev_name_node_add(struct net *net, | |
269 | struct netdev_name_node *name_node) | |
270 | { | |
271 | hlist_add_head_rcu(&name_node->hlist, | |
272 | dev_name_hash(net, name_node->name)); | |
273 | } | |
274 | ||
275 | static void netdev_name_node_del(struct netdev_name_node *name_node) | |
276 | { | |
277 | hlist_del_rcu(&name_node->hlist); | |
278 | } | |
279 | ||
280 | static struct netdev_name_node *netdev_name_node_lookup(struct net *net, | |
281 | const char *name) | |
282 | { | |
283 | struct hlist_head *head = dev_name_hash(net, name); | |
284 | struct netdev_name_node *name_node; | |
285 | ||
286 | hlist_for_each_entry(name_node, head, hlist) | |
287 | if (!strcmp(name_node->name, name)) | |
288 | return name_node; | |
289 | return NULL; | |
290 | } | |
291 | ||
292 | static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, | |
293 | const char *name) | |
294 | { | |
295 | struct hlist_head *head = dev_name_hash(net, name); | |
296 | struct netdev_name_node *name_node; | |
297 | ||
298 | hlist_for_each_entry_rcu(name_node, head, hlist) | |
299 | if (!strcmp(name_node->name, name)) | |
300 | return name_node; | |
301 | return NULL; | |
302 | } | |
303 | ||
36fbf1e5 JP |
304 | int netdev_name_node_alt_create(struct net_device *dev, const char *name) |
305 | { | |
306 | struct netdev_name_node *name_node; | |
307 | struct net *net = dev_net(dev); | |
308 | ||
309 | name_node = netdev_name_node_lookup(net, name); | |
310 | if (name_node) | |
311 | return -EEXIST; | |
312 | name_node = netdev_name_node_alloc(dev, name); | |
313 | if (!name_node) | |
314 | return -ENOMEM; | |
315 | netdev_name_node_add(net, name_node); | |
316 | /* The node that holds dev->name acts as a head of per-device list. */ | |
317 | list_add_tail(&name_node->list, &dev->name_node->list); | |
318 | ||
319 | return 0; | |
320 | } | |
321 | EXPORT_SYMBOL(netdev_name_node_alt_create); | |
322 | ||
323 | static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) | |
324 | { | |
325 | list_del(&name_node->list); | |
326 | netdev_name_node_del(name_node); | |
327 | kfree(name_node->name); | |
328 | netdev_name_node_free(name_node); | |
329 | } | |
330 | ||
331 | int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) | |
332 | { | |
333 | struct netdev_name_node *name_node; | |
334 | struct net *net = dev_net(dev); | |
335 | ||
336 | name_node = netdev_name_node_lookup(net, name); | |
337 | if (!name_node) | |
338 | return -ENOENT; | |
e08ad805 ED |
339 | /* lookup might have found our primary name or a name belonging |
340 | * to another device. | |
341 | */ | |
342 | if (name_node == dev->name_node || name_node->dev != dev) | |
343 | return -EINVAL; | |
344 | ||
36fbf1e5 JP |
345 | __netdev_name_node_alt_destroy(name_node); |
346 | ||
347 | return 0; | |
348 | } | |
349 | EXPORT_SYMBOL(netdev_name_node_alt_destroy); | |
350 | ||
351 | static void netdev_name_node_alt_flush(struct net_device *dev) | |
352 | { | |
353 | struct netdev_name_node *name_node, *tmp; | |
354 | ||
355 | list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) | |
356 | __netdev_name_node_alt_destroy(name_node); | |
357 | } | |
358 | ||
ce286d32 | 359 | /* Device list insertion */ |
53759be9 | 360 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 361 | { |
c346dca1 | 362 | struct net *net = dev_net(dev); |
ce286d32 EB |
363 | |
364 | ASSERT_RTNL(); | |
365 | ||
366 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 367 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
ff927412 | 368 | netdev_name_node_add(net, dev->name_node); |
fb699dfd ED |
369 | hlist_add_head_rcu(&dev->index_hlist, |
370 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 371 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
372 | |
373 | dev_base_seq_inc(net); | |
ce286d32 EB |
374 | } |
375 | ||
fb699dfd ED |
376 | /* Device list removal |
377 | * caller must respect a RCU grace period before freeing/reusing dev | |
378 | */ | |
ce286d32 EB |
379 | static void unlist_netdevice(struct net_device *dev) |
380 | { | |
381 | ASSERT_RTNL(); | |
382 | ||
383 | /* Unlink dev from the device chain */ | |
384 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 385 | list_del_rcu(&dev->dev_list); |
ff927412 | 386 | netdev_name_node_del(dev->name_node); |
fb699dfd | 387 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 388 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
389 | |
390 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
391 | } |
392 | ||
1da177e4 LT |
393 | /* |
394 | * Our notifier list | |
395 | */ | |
396 | ||
f07d5b94 | 397 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
398 | |
399 | /* | |
400 | * Device drivers call our routines to queue packets here. We empty the | |
401 | * queue in the local softnet handler. | |
402 | */ | |
bea3348e | 403 | |
9958da05 | 404 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 405 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 406 | |
1a33e10e CW |
407 | #ifdef CONFIG_LOCKDEP |
408 | /* | |
409 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class | |
410 | * according to dev->type | |
411 | */ | |
412 | static const unsigned short netdev_lock_type[] = { | |
413 | ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
414 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
415 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
416 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
417 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
418 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
419 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
420 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
421 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
422 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
423 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
424 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
425 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, | |
426 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
427 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
428 | ||
429 | static const char *const netdev_lock_name[] = { | |
430 | "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
431 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
432 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
433 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
434 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
435 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
436 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
437 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
438 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
439 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
440 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
441 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
442 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", | |
443 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
444 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
445 | ||
446 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
845e0ebb | 447 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
1a33e10e CW |
448 | |
449 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
450 | { | |
451 | int i; | |
452 | ||
453 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
454 | if (netdev_lock_type[i] == dev_type) | |
455 | return i; | |
456 | /* the last key is used by default */ | |
457 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
458 | } | |
459 | ||
460 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, | |
461 | unsigned short dev_type) | |
462 | { | |
463 | int i; | |
464 | ||
465 | i = netdev_lock_pos(dev_type); | |
466 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
467 | netdev_lock_name[i]); | |
468 | } | |
845e0ebb CW |
469 | |
470 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
471 | { | |
472 | int i; | |
473 | ||
474 | i = netdev_lock_pos(dev->type); | |
475 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
476 | &netdev_addr_lock_key[i], | |
477 | netdev_lock_name[i]); | |
478 | } | |
1a33e10e CW |
479 | #else |
480 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, | |
481 | unsigned short dev_type) | |
482 | { | |
483 | } | |
845e0ebb CW |
484 | |
485 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
486 | { | |
487 | } | |
1a33e10e CW |
488 | #endif |
489 | ||
1da177e4 | 490 | /******************************************************************************* |
eb13da1a | 491 | * |
492 | * Protocol management and registration routines | |
493 | * | |
494 | *******************************************************************************/ | |
1da177e4 | 495 | |
1da177e4 | 496 | |
1da177e4 LT |
497 | /* |
498 | * Add a protocol ID to the list. Now that the input handler is | |
499 | * smarter we can dispense with all the messy stuff that used to be | |
500 | * here. | |
501 | * | |
502 | * BEWARE!!! Protocol handlers, mangling input packets, | |
503 | * MUST BE last in hash buckets and checking protocol handlers | |
504 | * MUST start from promiscuous ptype_all chain in net_bh. | |
505 | * It is true now, do not change it. | |
506 | * Explanation follows: if protocol handler, mangling packet, will | |
507 | * be the first on list, it is not able to sense, that packet | |
508 | * is cloned and should be copied-on-write, so that it will | |
509 | * change it and subsequent readers will get broken packet. | |
510 | * --ANK (980803) | |
511 | */ | |
512 | ||
c07b68e8 ED |
513 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
514 | { | |
515 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 516 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 517 | else |
7866a621 SN |
518 | return pt->dev ? &pt->dev->ptype_specific : |
519 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
520 | } |
521 | ||
1da177e4 LT |
522 | /** |
523 | * dev_add_pack - add packet handler | |
524 | * @pt: packet type declaration | |
525 | * | |
526 | * Add a protocol handler to the networking stack. The passed &packet_type | |
527 | * is linked into kernel lists and may not be freed until it has been | |
528 | * removed from the kernel lists. | |
529 | * | |
4ec93edb | 530 | * This call does not sleep therefore it can not |
1da177e4 LT |
531 | * guarantee all CPU's that are in middle of receiving packets |
532 | * will see the new packet type (until the next received packet). | |
533 | */ | |
534 | ||
535 | void dev_add_pack(struct packet_type *pt) | |
536 | { | |
c07b68e8 | 537 | struct list_head *head = ptype_head(pt); |
1da177e4 | 538 | |
c07b68e8 ED |
539 | spin_lock(&ptype_lock); |
540 | list_add_rcu(&pt->list, head); | |
541 | spin_unlock(&ptype_lock); | |
1da177e4 | 542 | } |
d1b19dff | 543 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 544 | |
1da177e4 LT |
545 | /** |
546 | * __dev_remove_pack - remove packet handler | |
547 | * @pt: packet type declaration | |
548 | * | |
549 | * Remove a protocol handler that was previously added to the kernel | |
550 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
551 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 552 | * returns. |
1da177e4 LT |
553 | * |
554 | * The packet type might still be in use by receivers | |
555 | * and must not be freed until after all the CPU's have gone | |
556 | * through a quiescent state. | |
557 | */ | |
558 | void __dev_remove_pack(struct packet_type *pt) | |
559 | { | |
c07b68e8 | 560 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
561 | struct packet_type *pt1; |
562 | ||
c07b68e8 | 563 | spin_lock(&ptype_lock); |
1da177e4 LT |
564 | |
565 | list_for_each_entry(pt1, head, list) { | |
566 | if (pt == pt1) { | |
567 | list_del_rcu(&pt->list); | |
568 | goto out; | |
569 | } | |
570 | } | |
571 | ||
7b6cd1ce | 572 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 573 | out: |
c07b68e8 | 574 | spin_unlock(&ptype_lock); |
1da177e4 | 575 | } |
d1b19dff ED |
576 | EXPORT_SYMBOL(__dev_remove_pack); |
577 | ||
1da177e4 LT |
578 | /** |
579 | * dev_remove_pack - remove packet handler | |
580 | * @pt: packet type declaration | |
581 | * | |
582 | * Remove a protocol handler that was previously added to the kernel | |
583 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
584 | * from the kernel lists and can be freed or reused once this function | |
585 | * returns. | |
586 | * | |
587 | * This call sleeps to guarantee that no CPU is looking at the packet | |
588 | * type after return. | |
589 | */ | |
590 | void dev_remove_pack(struct packet_type *pt) | |
591 | { | |
592 | __dev_remove_pack(pt); | |
4ec93edb | 593 | |
1da177e4 LT |
594 | synchronize_net(); |
595 | } | |
d1b19dff | 596 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 597 | |
62532da9 VY |
598 | |
599 | /** | |
600 | * dev_add_offload - register offload handlers | |
601 | * @po: protocol offload declaration | |
602 | * | |
603 | * Add protocol offload handlers to the networking stack. The passed | |
604 | * &proto_offload is linked into kernel lists and may not be freed until | |
605 | * it has been removed from the kernel lists. | |
606 | * | |
607 | * This call does not sleep therefore it can not | |
608 | * guarantee all CPU's that are in middle of receiving packets | |
609 | * will see the new offload handlers (until the next received packet). | |
610 | */ | |
611 | void dev_add_offload(struct packet_offload *po) | |
612 | { | |
bdef7de4 | 613 | struct packet_offload *elem; |
62532da9 VY |
614 | |
615 | spin_lock(&offload_lock); | |
bdef7de4 DM |
616 | list_for_each_entry(elem, &offload_base, list) { |
617 | if (po->priority < elem->priority) | |
618 | break; | |
619 | } | |
620 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
621 | spin_unlock(&offload_lock); |
622 | } | |
623 | EXPORT_SYMBOL(dev_add_offload); | |
624 | ||
625 | /** | |
626 | * __dev_remove_offload - remove offload handler | |
627 | * @po: packet offload declaration | |
628 | * | |
629 | * Remove a protocol offload handler that was previously added to the | |
630 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
631 | * is removed from the kernel lists and can be freed or reused once this | |
632 | * function returns. | |
633 | * | |
634 | * The packet type might still be in use by receivers | |
635 | * and must not be freed until after all the CPU's have gone | |
636 | * through a quiescent state. | |
637 | */ | |
1d143d9f | 638 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
639 | { |
640 | struct list_head *head = &offload_base; | |
641 | struct packet_offload *po1; | |
642 | ||
c53aa505 | 643 | spin_lock(&offload_lock); |
62532da9 VY |
644 | |
645 | list_for_each_entry(po1, head, list) { | |
646 | if (po == po1) { | |
647 | list_del_rcu(&po->list); | |
648 | goto out; | |
649 | } | |
650 | } | |
651 | ||
652 | pr_warn("dev_remove_offload: %p not found\n", po); | |
653 | out: | |
c53aa505 | 654 | spin_unlock(&offload_lock); |
62532da9 | 655 | } |
62532da9 VY |
656 | |
657 | /** | |
658 | * dev_remove_offload - remove packet offload handler | |
659 | * @po: packet offload declaration | |
660 | * | |
661 | * Remove a packet offload handler that was previously added to the kernel | |
662 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
663 | * removed from the kernel lists and can be freed or reused once this | |
664 | * function returns. | |
665 | * | |
666 | * This call sleeps to guarantee that no CPU is looking at the packet | |
667 | * type after return. | |
668 | */ | |
669 | void dev_remove_offload(struct packet_offload *po) | |
670 | { | |
671 | __dev_remove_offload(po); | |
672 | ||
673 | synchronize_net(); | |
674 | } | |
675 | EXPORT_SYMBOL(dev_remove_offload); | |
676 | ||
1da177e4 | 677 | /****************************************************************************** |
eb13da1a | 678 | * |
679 | * Device Boot-time Settings Routines | |
680 | * | |
681 | ******************************************************************************/ | |
1da177e4 LT |
682 | |
683 | /* Boot time configuration table */ | |
684 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
685 | ||
686 | /** | |
687 | * netdev_boot_setup_add - add new setup entry | |
688 | * @name: name of the device | |
689 | * @map: configured settings for the device | |
690 | * | |
691 | * Adds new setup entry to the dev_boot_setup list. The function | |
692 | * returns 0 on error and 1 on success. This is a generic routine to | |
693 | * all netdevices. | |
694 | */ | |
695 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
696 | { | |
697 | struct netdev_boot_setup *s; | |
698 | int i; | |
699 | ||
700 | s = dev_boot_setup; | |
701 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
702 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
703 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 704 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
705 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
706 | break; | |
707 | } | |
708 | } | |
709 | ||
710 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
711 | } | |
712 | ||
713 | /** | |
722c9a0c | 714 | * netdev_boot_setup_check - check boot time settings |
715 | * @dev: the netdevice | |
1da177e4 | 716 | * |
722c9a0c | 717 | * Check boot time settings for the device. |
718 | * The found settings are set for the device to be used | |
719 | * later in the device probing. | |
720 | * Returns 0 if no settings found, 1 if they are. | |
1da177e4 LT |
721 | */ |
722 | int netdev_boot_setup_check(struct net_device *dev) | |
723 | { | |
724 | struct netdev_boot_setup *s = dev_boot_setup; | |
725 | int i; | |
726 | ||
727 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
728 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 729 | !strcmp(dev->name, s[i].name)) { |
722c9a0c | 730 | dev->irq = s[i].map.irq; |
731 | dev->base_addr = s[i].map.base_addr; | |
732 | dev->mem_start = s[i].map.mem_start; | |
733 | dev->mem_end = s[i].map.mem_end; | |
1da177e4 LT |
734 | return 1; |
735 | } | |
736 | } | |
737 | return 0; | |
738 | } | |
d1b19dff | 739 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
740 | |
741 | ||
742 | /** | |
722c9a0c | 743 | * netdev_boot_base - get address from boot time settings |
744 | * @prefix: prefix for network device | |
745 | * @unit: id for network device | |
746 | * | |
747 | * Check boot time settings for the base address of device. | |
748 | * The found settings are set for the device to be used | |
749 | * later in the device probing. | |
750 | * Returns 0 if no settings found. | |
1da177e4 LT |
751 | */ |
752 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
753 | { | |
754 | const struct netdev_boot_setup *s = dev_boot_setup; | |
755 | char name[IFNAMSIZ]; | |
756 | int i; | |
757 | ||
758 | sprintf(name, "%s%d", prefix, unit); | |
759 | ||
760 | /* | |
761 | * If device already registered then return base of 1 | |
762 | * to indicate not to probe for this interface | |
763 | */ | |
881d966b | 764 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
765 | return 1; |
766 | ||
767 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
768 | if (!strcmp(name, s[i].name)) | |
769 | return s[i].map.base_addr; | |
770 | return 0; | |
771 | } | |
772 | ||
773 | /* | |
774 | * Saves at boot time configured settings for any netdevice. | |
775 | */ | |
776 | int __init netdev_boot_setup(char *str) | |
777 | { | |
778 | int ints[5]; | |
779 | struct ifmap map; | |
780 | ||
781 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
782 | if (!str || !*str) | |
783 | return 0; | |
784 | ||
785 | /* Save settings */ | |
786 | memset(&map, 0, sizeof(map)); | |
787 | if (ints[0] > 0) | |
788 | map.irq = ints[1]; | |
789 | if (ints[0] > 1) | |
790 | map.base_addr = ints[2]; | |
791 | if (ints[0] > 2) | |
792 | map.mem_start = ints[3]; | |
793 | if (ints[0] > 3) | |
794 | map.mem_end = ints[4]; | |
795 | ||
796 | /* Add new entry to the list */ | |
797 | return netdev_boot_setup_add(str, &map); | |
798 | } | |
799 | ||
800 | __setup("netdev=", netdev_boot_setup); | |
801 | ||
802 | /******************************************************************************* | |
eb13da1a | 803 | * |
804 | * Device Interface Subroutines | |
805 | * | |
806 | *******************************************************************************/ | |
1da177e4 | 807 | |
a54acb3a ND |
808 | /** |
809 | * dev_get_iflink - get 'iflink' value of a interface | |
810 | * @dev: targeted interface | |
811 | * | |
812 | * Indicates the ifindex the interface is linked to. | |
813 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
814 | */ | |
815 | ||
816 | int dev_get_iflink(const struct net_device *dev) | |
817 | { | |
818 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
819 | return dev->netdev_ops->ndo_get_iflink(dev); | |
820 | ||
7a66bbc9 | 821 | return dev->ifindex; |
a54acb3a ND |
822 | } |
823 | EXPORT_SYMBOL(dev_get_iflink); | |
824 | ||
fc4099f1 PS |
825 | /** |
826 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
827 | * @dev: targeted interface | |
828 | * @skb: The packet. | |
829 | * | |
830 | * For better visibility of tunnel traffic OVS needs to retrieve | |
831 | * egress tunnel information for a packet. Following API allows | |
832 | * user to get this info. | |
833 | */ | |
834 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
835 | { | |
836 | struct ip_tunnel_info *info; | |
837 | ||
838 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
839 | return -EINVAL; | |
840 | ||
841 | info = skb_tunnel_info_unclone(skb); | |
842 | if (!info) | |
843 | return -ENOMEM; | |
844 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
845 | return -EINVAL; | |
846 | ||
847 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
848 | } | |
849 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
850 | ||
1da177e4 LT |
851 | /** |
852 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 853 | * @net: the applicable net namespace |
1da177e4 LT |
854 | * @name: name to find |
855 | * | |
856 | * Find an interface by name. Must be called under RTNL semaphore | |
857 | * or @dev_base_lock. If the name is found a pointer to the device | |
858 | * is returned. If the name is not found then %NULL is returned. The | |
859 | * reference counters are not incremented so the caller must be | |
860 | * careful with locks. | |
861 | */ | |
862 | ||
881d966b | 863 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 864 | { |
ff927412 | 865 | struct netdev_name_node *node_name; |
1da177e4 | 866 | |
ff927412 JP |
867 | node_name = netdev_name_node_lookup(net, name); |
868 | return node_name ? node_name->dev : NULL; | |
1da177e4 | 869 | } |
d1b19dff | 870 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 871 | |
72c9528b | 872 | /** |
722c9a0c | 873 | * dev_get_by_name_rcu - find a device by its name |
874 | * @net: the applicable net namespace | |
875 | * @name: name to find | |
876 | * | |
877 | * Find an interface by name. | |
878 | * If the name is found a pointer to the device is returned. | |
879 | * If the name is not found then %NULL is returned. | |
880 | * The reference counters are not incremented so the caller must be | |
881 | * careful with locks. The caller must hold RCU lock. | |
72c9528b ED |
882 | */ |
883 | ||
884 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
885 | { | |
ff927412 | 886 | struct netdev_name_node *node_name; |
72c9528b | 887 | |
ff927412 JP |
888 | node_name = netdev_name_node_lookup_rcu(net, name); |
889 | return node_name ? node_name->dev : NULL; | |
72c9528b ED |
890 | } |
891 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
892 | ||
1da177e4 LT |
893 | /** |
894 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 895 | * @net: the applicable net namespace |
1da177e4 LT |
896 | * @name: name to find |
897 | * | |
898 | * Find an interface by name. This can be called from any | |
899 | * context and does its own locking. The returned handle has | |
900 | * the usage count incremented and the caller must use dev_put() to | |
901 | * release it when it is no longer needed. %NULL is returned if no | |
902 | * matching device is found. | |
903 | */ | |
904 | ||
881d966b | 905 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
906 | { |
907 | struct net_device *dev; | |
908 | ||
72c9528b ED |
909 | rcu_read_lock(); |
910 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
911 | if (dev) |
912 | dev_hold(dev); | |
72c9528b | 913 | rcu_read_unlock(); |
1da177e4 LT |
914 | return dev; |
915 | } | |
d1b19dff | 916 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
917 | |
918 | /** | |
919 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 920 | * @net: the applicable net namespace |
1da177e4 LT |
921 | * @ifindex: index of device |
922 | * | |
923 | * Search for an interface by index. Returns %NULL if the device | |
924 | * is not found or a pointer to the device. The device has not | |
925 | * had its reference counter increased so the caller must be careful | |
926 | * about locking. The caller must hold either the RTNL semaphore | |
927 | * or @dev_base_lock. | |
928 | */ | |
929 | ||
881d966b | 930 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 931 | { |
0bd8d536 ED |
932 | struct net_device *dev; |
933 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 934 | |
b67bfe0d | 935 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
936 | if (dev->ifindex == ifindex) |
937 | return dev; | |
0bd8d536 | 938 | |
1da177e4 LT |
939 | return NULL; |
940 | } | |
d1b19dff | 941 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 942 | |
fb699dfd ED |
943 | /** |
944 | * dev_get_by_index_rcu - find a device by its ifindex | |
945 | * @net: the applicable net namespace | |
946 | * @ifindex: index of device | |
947 | * | |
948 | * Search for an interface by index. Returns %NULL if the device | |
949 | * is not found or a pointer to the device. The device has not | |
950 | * had its reference counter increased so the caller must be careful | |
951 | * about locking. The caller must hold RCU lock. | |
952 | */ | |
953 | ||
954 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
955 | { | |
fb699dfd ED |
956 | struct net_device *dev; |
957 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
958 | ||
b67bfe0d | 959 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
960 | if (dev->ifindex == ifindex) |
961 | return dev; | |
962 | ||
963 | return NULL; | |
964 | } | |
965 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
966 | ||
1da177e4 LT |
967 | |
968 | /** | |
969 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 970 | * @net: the applicable net namespace |
1da177e4 LT |
971 | * @ifindex: index of device |
972 | * | |
973 | * Search for an interface by index. Returns NULL if the device | |
974 | * is not found or a pointer to the device. The device returned has | |
975 | * had a reference added and the pointer is safe until the user calls | |
976 | * dev_put to indicate they have finished with it. | |
977 | */ | |
978 | ||
881d966b | 979 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
980 | { |
981 | struct net_device *dev; | |
982 | ||
fb699dfd ED |
983 | rcu_read_lock(); |
984 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
985 | if (dev) |
986 | dev_hold(dev); | |
fb699dfd | 987 | rcu_read_unlock(); |
1da177e4 LT |
988 | return dev; |
989 | } | |
d1b19dff | 990 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 991 | |
90b602f8 ML |
992 | /** |
993 | * dev_get_by_napi_id - find a device by napi_id | |
994 | * @napi_id: ID of the NAPI struct | |
995 | * | |
996 | * Search for an interface by NAPI ID. Returns %NULL if the device | |
997 | * is not found or a pointer to the device. The device has not had | |
998 | * its reference counter increased so the caller must be careful | |
999 | * about locking. The caller must hold RCU lock. | |
1000 | */ | |
1001 | ||
1002 | struct net_device *dev_get_by_napi_id(unsigned int napi_id) | |
1003 | { | |
1004 | struct napi_struct *napi; | |
1005 | ||
1006 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
1007 | ||
1008 | if (napi_id < MIN_NAPI_ID) | |
1009 | return NULL; | |
1010 | ||
1011 | napi = napi_by_id(napi_id); | |
1012 | ||
1013 | return napi ? napi->dev : NULL; | |
1014 | } | |
1015 | EXPORT_SYMBOL(dev_get_by_napi_id); | |
1016 | ||
5dbe7c17 NS |
1017 | /** |
1018 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
1019 | * @net: network namespace | |
1020 | * @name: a pointer to the buffer where the name will be stored. | |
1021 | * @ifindex: the ifindex of the interface to get the name from. | |
5dbe7c17 NS |
1022 | */ |
1023 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
1024 | { | |
1025 | struct net_device *dev; | |
11d6011c | 1026 | int ret; |
5dbe7c17 | 1027 | |
11d6011c | 1028 | down_read(&devnet_rename_sem); |
5dbe7c17 | 1029 | rcu_read_lock(); |
11d6011c | 1030 | |
5dbe7c17 NS |
1031 | dev = dev_get_by_index_rcu(net, ifindex); |
1032 | if (!dev) { | |
11d6011c AD |
1033 | ret = -ENODEV; |
1034 | goto out; | |
5dbe7c17 NS |
1035 | } |
1036 | ||
1037 | strcpy(name, dev->name); | |
5dbe7c17 | 1038 | |
11d6011c AD |
1039 | ret = 0; |
1040 | out: | |
1041 | rcu_read_unlock(); | |
1042 | up_read(&devnet_rename_sem); | |
1043 | return ret; | |
5dbe7c17 NS |
1044 | } |
1045 | ||
1da177e4 | 1046 | /** |
941666c2 | 1047 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 1048 | * @net: the applicable net namespace |
1da177e4 LT |
1049 | * @type: media type of device |
1050 | * @ha: hardware address | |
1051 | * | |
1052 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
1053 | * is not found or a pointer to the device. |
1054 | * The caller must hold RCU or RTNL. | |
941666c2 | 1055 | * The returned device has not had its ref count increased |
1da177e4 LT |
1056 | * and the caller must therefore be careful about locking |
1057 | * | |
1da177e4 LT |
1058 | */ |
1059 | ||
941666c2 ED |
1060 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
1061 | const char *ha) | |
1da177e4 LT |
1062 | { |
1063 | struct net_device *dev; | |
1064 | ||
941666c2 | 1065 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
1066 | if (dev->type == type && |
1067 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
1068 | return dev; |
1069 | ||
1070 | return NULL; | |
1da177e4 | 1071 | } |
941666c2 | 1072 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 1073 | |
881d966b | 1074 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 1075 | { |
99fe3c39 | 1076 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 1077 | |
99fe3c39 ED |
1078 | rcu_read_lock(); |
1079 | for_each_netdev_rcu(net, dev) | |
1080 | if (dev->type == type) { | |
1081 | dev_hold(dev); | |
1082 | ret = dev; | |
1083 | break; | |
1084 | } | |
1085 | rcu_read_unlock(); | |
1086 | return ret; | |
1da177e4 | 1087 | } |
1da177e4 LT |
1088 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
1089 | ||
1090 | /** | |
6c555490 | 1091 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 1092 | * @net: the applicable net namespace |
1da177e4 LT |
1093 | * @if_flags: IFF_* values |
1094 | * @mask: bitmask of bits in if_flags to check | |
1095 | * | |
1096 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 1097 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 1098 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
1099 | */ |
1100 | ||
6c555490 WC |
1101 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
1102 | unsigned short mask) | |
1da177e4 | 1103 | { |
7562f876 | 1104 | struct net_device *dev, *ret; |
1da177e4 | 1105 | |
6c555490 WC |
1106 | ASSERT_RTNL(); |
1107 | ||
7562f876 | 1108 | ret = NULL; |
6c555490 | 1109 | for_each_netdev(net, dev) { |
1da177e4 | 1110 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 1111 | ret = dev; |
1da177e4 LT |
1112 | break; |
1113 | } | |
1114 | } | |
7562f876 | 1115 | return ret; |
1da177e4 | 1116 | } |
6c555490 | 1117 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
1118 | |
1119 | /** | |
1120 | * dev_valid_name - check if name is okay for network device | |
1121 | * @name: name string | |
1122 | * | |
1123 | * Network device names need to be valid file names to | |
4250b75b | 1124 | * allow sysfs to work. We also disallow any kind of |
c7fa9d18 | 1125 | * whitespace. |
1da177e4 | 1126 | */ |
95f050bf | 1127 | bool dev_valid_name(const char *name) |
1da177e4 | 1128 | { |
c7fa9d18 | 1129 | if (*name == '\0') |
95f050bf | 1130 | return false; |
a9d48205 | 1131 | if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) |
95f050bf | 1132 | return false; |
c7fa9d18 | 1133 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1134 | return false; |
c7fa9d18 DM |
1135 | |
1136 | while (*name) { | |
a4176a93 | 1137 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1138 | return false; |
c7fa9d18 DM |
1139 | name++; |
1140 | } | |
95f050bf | 1141 | return true; |
1da177e4 | 1142 | } |
d1b19dff | 1143 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1144 | |
1145 | /** | |
b267b179 EB |
1146 | * __dev_alloc_name - allocate a name for a device |
1147 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1148 | * @name: name format string |
b267b179 | 1149 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1150 | * |
1151 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1152 | * id. It scans list of devices to build up a free map, then chooses |
1153 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1154 | * while allocating the name and adding the device in order to avoid | |
1155 | * duplicates. | |
1156 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1157 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1158 | */ |
1159 | ||
b267b179 | 1160 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1161 | { |
1162 | int i = 0; | |
1da177e4 LT |
1163 | const char *p; |
1164 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1165 | unsigned long *inuse; |
1da177e4 LT |
1166 | struct net_device *d; |
1167 | ||
93809105 RV |
1168 | if (!dev_valid_name(name)) |
1169 | return -EINVAL; | |
1170 | ||
51f299dd | 1171 | p = strchr(name, '%'); |
1da177e4 LT |
1172 | if (p) { |
1173 | /* | |
1174 | * Verify the string as this thing may have come from | |
1175 | * the user. There must be either one "%d" and no other "%" | |
1176 | * characters. | |
1177 | */ | |
1178 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1179 | return -EINVAL; | |
1180 | ||
1181 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1182 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1183 | if (!inuse) |
1184 | return -ENOMEM; | |
1185 | ||
881d966b | 1186 | for_each_netdev(net, d) { |
6c015a22 JB |
1187 | struct netdev_name_node *name_node; |
1188 | list_for_each_entry(name_node, &d->name_node->list, list) { | |
1189 | if (!sscanf(name_node->name, name, &i)) | |
1190 | continue; | |
1191 | if (i < 0 || i >= max_netdevices) | |
1192 | continue; | |
1193 | ||
1194 | /* avoid cases where sscanf is not exact inverse of printf */ | |
1195 | snprintf(buf, IFNAMSIZ, name, i); | |
1196 | if (!strncmp(buf, name_node->name, IFNAMSIZ)) | |
1197 | set_bit(i, inuse); | |
1198 | } | |
1da177e4 LT |
1199 | if (!sscanf(d->name, name, &i)) |
1200 | continue; | |
1201 | if (i < 0 || i >= max_netdevices) | |
1202 | continue; | |
1203 | ||
1204 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1205 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1206 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1207 | set_bit(i, inuse); | |
1208 | } | |
1209 | ||
1210 | i = find_first_zero_bit(inuse, max_netdevices); | |
1211 | free_page((unsigned long) inuse); | |
1212 | } | |
1213 | ||
6224abda | 1214 | snprintf(buf, IFNAMSIZ, name, i); |
b267b179 | 1215 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1216 | return i; |
1da177e4 LT |
1217 | |
1218 | /* It is possible to run out of possible slots | |
1219 | * when the name is long and there isn't enough space left | |
1220 | * for the digits, or if all bits are used. | |
1221 | */ | |
029b6d14 | 1222 | return -ENFILE; |
1da177e4 LT |
1223 | } |
1224 | ||
2c88b855 RV |
1225 | static int dev_alloc_name_ns(struct net *net, |
1226 | struct net_device *dev, | |
1227 | const char *name) | |
1228 | { | |
1229 | char buf[IFNAMSIZ]; | |
1230 | int ret; | |
1231 | ||
c46d7642 | 1232 | BUG_ON(!net); |
2c88b855 RV |
1233 | ret = __dev_alloc_name(net, name, buf); |
1234 | if (ret >= 0) | |
1235 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1236 | return ret; | |
1da177e4 LT |
1237 | } |
1238 | ||
b267b179 EB |
1239 | /** |
1240 | * dev_alloc_name - allocate a name for a device | |
1241 | * @dev: device | |
1242 | * @name: name format string | |
1243 | * | |
1244 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1245 | * id. It scans list of devices to build up a free map, then chooses | |
1246 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1247 | * while allocating the name and adding the device in order to avoid | |
1248 | * duplicates. | |
1249 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1250 | * Returns the number of the unit assigned or a negative errno code. | |
1251 | */ | |
1252 | ||
1253 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1254 | { | |
c46d7642 | 1255 | return dev_alloc_name_ns(dev_net(dev), dev, name); |
b267b179 | 1256 | } |
d1b19dff | 1257 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1258 | |
bacb7e18 ED |
1259 | static int dev_get_valid_name(struct net *net, struct net_device *dev, |
1260 | const char *name) | |
828de4f6 | 1261 | { |
55a5ec9b DM |
1262 | BUG_ON(!net); |
1263 | ||
1264 | if (!dev_valid_name(name)) | |
1265 | return -EINVAL; | |
1266 | ||
1267 | if (strchr(name, '%')) | |
1268 | return dev_alloc_name_ns(net, dev, name); | |
1269 | else if (__dev_get_by_name(net, name)) | |
1270 | return -EEXIST; | |
1271 | else if (dev->name != name) | |
1272 | strlcpy(dev->name, name, IFNAMSIZ); | |
1273 | ||
1274 | return 0; | |
d9031024 | 1275 | } |
1da177e4 LT |
1276 | |
1277 | /** | |
1278 | * dev_change_name - change name of a device | |
1279 | * @dev: device | |
1280 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1281 | * | |
1282 | * Change name of a device, can pass format strings "eth%d". | |
1283 | * for wildcarding. | |
1284 | */ | |
cf04a4c7 | 1285 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1286 | { |
238fa362 | 1287 | unsigned char old_assign_type; |
fcc5a03a | 1288 | char oldname[IFNAMSIZ]; |
1da177e4 | 1289 | int err = 0; |
fcc5a03a | 1290 | int ret; |
881d966b | 1291 | struct net *net; |
1da177e4 LT |
1292 | |
1293 | ASSERT_RTNL(); | |
c346dca1 | 1294 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1295 | |
c346dca1 | 1296 | net = dev_net(dev); |
8065a779 SWL |
1297 | |
1298 | /* Some auto-enslaved devices e.g. failover slaves are | |
1299 | * special, as userspace might rename the device after | |
1300 | * the interface had been brought up and running since | |
1301 | * the point kernel initiated auto-enslavement. Allow | |
1302 | * live name change even when these slave devices are | |
1303 | * up and running. | |
1304 | * | |
1305 | * Typically, users of these auto-enslaving devices | |
1306 | * don't actually care about slave name change, as | |
1307 | * they are supposed to operate on master interface | |
1308 | * directly. | |
1309 | */ | |
1310 | if (dev->flags & IFF_UP && | |
1311 | likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) | |
1da177e4 LT |
1312 | return -EBUSY; |
1313 | ||
11d6011c | 1314 | down_write(&devnet_rename_sem); |
c91f6df2 BH |
1315 | |
1316 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
11d6011c | 1317 | up_write(&devnet_rename_sem); |
c8d90dca | 1318 | return 0; |
c91f6df2 | 1319 | } |
c8d90dca | 1320 | |
fcc5a03a HX |
1321 | memcpy(oldname, dev->name, IFNAMSIZ); |
1322 | ||
828de4f6 | 1323 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1324 | if (err < 0) { |
11d6011c | 1325 | up_write(&devnet_rename_sem); |
d9031024 | 1326 | return err; |
c91f6df2 | 1327 | } |
1da177e4 | 1328 | |
6fe82a39 VF |
1329 | if (oldname[0] && !strchr(oldname, '%')) |
1330 | netdev_info(dev, "renamed from %s\n", oldname); | |
1331 | ||
238fa362 TG |
1332 | old_assign_type = dev->name_assign_type; |
1333 | dev->name_assign_type = NET_NAME_RENAMED; | |
1334 | ||
fcc5a03a | 1335 | rollback: |
a1b3f594 EB |
1336 | ret = device_rename(&dev->dev, dev->name); |
1337 | if (ret) { | |
1338 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1339 | dev->name_assign_type = old_assign_type; |
11d6011c | 1340 | up_write(&devnet_rename_sem); |
a1b3f594 | 1341 | return ret; |
dcc99773 | 1342 | } |
7f988eab | 1343 | |
11d6011c | 1344 | up_write(&devnet_rename_sem); |
c91f6df2 | 1345 | |
5bb025fa VF |
1346 | netdev_adjacent_rename_links(dev, oldname); |
1347 | ||
7f988eab | 1348 | write_lock_bh(&dev_base_lock); |
ff927412 | 1349 | netdev_name_node_del(dev->name_node); |
72c9528b ED |
1350 | write_unlock_bh(&dev_base_lock); |
1351 | ||
1352 | synchronize_rcu(); | |
1353 | ||
1354 | write_lock_bh(&dev_base_lock); | |
ff927412 | 1355 | netdev_name_node_add(net, dev->name_node); |
7f988eab HX |
1356 | write_unlock_bh(&dev_base_lock); |
1357 | ||
056925ab | 1358 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1359 | ret = notifier_to_errno(ret); |
1360 | ||
1361 | if (ret) { | |
91e9c07b ED |
1362 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1363 | if (err >= 0) { | |
fcc5a03a | 1364 | err = ret; |
11d6011c | 1365 | down_write(&devnet_rename_sem); |
fcc5a03a | 1366 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1367 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1368 | dev->name_assign_type = old_assign_type; |
1369 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1370 | goto rollback; |
91e9c07b | 1371 | } else { |
7b6cd1ce | 1372 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1373 | dev->name, ret); |
fcc5a03a HX |
1374 | } |
1375 | } | |
1da177e4 LT |
1376 | |
1377 | return err; | |
1378 | } | |
1379 | ||
0b815a1a SH |
1380 | /** |
1381 | * dev_set_alias - change ifalias of a device | |
1382 | * @dev: device | |
1383 | * @alias: name up to IFALIASZ | |
f0db275a | 1384 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1385 | * |
1386 | * Set ifalias for a device, | |
1387 | */ | |
1388 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1389 | { | |
6c557001 | 1390 | struct dev_ifalias *new_alias = NULL; |
0b815a1a SH |
1391 | |
1392 | if (len >= IFALIASZ) | |
1393 | return -EINVAL; | |
1394 | ||
6c557001 FW |
1395 | if (len) { |
1396 | new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); | |
1397 | if (!new_alias) | |
1398 | return -ENOMEM; | |
1399 | ||
1400 | memcpy(new_alias->ifalias, alias, len); | |
1401 | new_alias->ifalias[len] = 0; | |
96ca4a2c OH |
1402 | } |
1403 | ||
6c557001 | 1404 | mutex_lock(&ifalias_mutex); |
e3f0d761 PM |
1405 | new_alias = rcu_replace_pointer(dev->ifalias, new_alias, |
1406 | mutex_is_locked(&ifalias_mutex)); | |
6c557001 FW |
1407 | mutex_unlock(&ifalias_mutex); |
1408 | ||
1409 | if (new_alias) | |
1410 | kfree_rcu(new_alias, rcuhead); | |
0b815a1a | 1411 | |
0b815a1a SH |
1412 | return len; |
1413 | } | |
0fe554a4 | 1414 | EXPORT_SYMBOL(dev_set_alias); |
0b815a1a | 1415 | |
6c557001 FW |
1416 | /** |
1417 | * dev_get_alias - get ifalias of a device | |
1418 | * @dev: device | |
20e88320 | 1419 | * @name: buffer to store name of ifalias |
6c557001 FW |
1420 | * @len: size of buffer |
1421 | * | |
1422 | * get ifalias for a device. Caller must make sure dev cannot go | |
1423 | * away, e.g. rcu read lock or own a reference count to device. | |
1424 | */ | |
1425 | int dev_get_alias(const struct net_device *dev, char *name, size_t len) | |
1426 | { | |
1427 | const struct dev_ifalias *alias; | |
1428 | int ret = 0; | |
1429 | ||
1430 | rcu_read_lock(); | |
1431 | alias = rcu_dereference(dev->ifalias); | |
1432 | if (alias) | |
1433 | ret = snprintf(name, len, "%s", alias->ifalias); | |
1434 | rcu_read_unlock(); | |
1435 | ||
1436 | return ret; | |
1437 | } | |
0b815a1a | 1438 | |
d8a33ac4 | 1439 | /** |
3041a069 | 1440 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1441 | * @dev: device to cause notification |
1442 | * | |
1443 | * Called to indicate a device has changed features. | |
1444 | */ | |
1445 | void netdev_features_change(struct net_device *dev) | |
1446 | { | |
056925ab | 1447 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1448 | } |
1449 | EXPORT_SYMBOL(netdev_features_change); | |
1450 | ||
1da177e4 LT |
1451 | /** |
1452 | * netdev_state_change - device changes state | |
1453 | * @dev: device to cause notification | |
1454 | * | |
1455 | * Called to indicate a device has changed state. This function calls | |
1456 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1457 | * to the routing socket. | |
1458 | */ | |
1459 | void netdev_state_change(struct net_device *dev) | |
1460 | { | |
1461 | if (dev->flags & IFF_UP) { | |
51d0c047 DA |
1462 | struct netdev_notifier_change_info change_info = { |
1463 | .info.dev = dev, | |
1464 | }; | |
54951194 | 1465 | |
51d0c047 | 1466 | call_netdevice_notifiers_info(NETDEV_CHANGE, |
54951194 | 1467 | &change_info.info); |
7f294054 | 1468 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1469 | } |
1470 | } | |
d1b19dff | 1471 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1472 | |
7061eb8c LP |
1473 | /** |
1474 | * __netdev_notify_peers - notify network peers about existence of @dev, | |
1475 | * to be called when rtnl lock is already held. | |
1476 | * @dev: network device | |
1477 | * | |
1478 | * Generate traffic such that interested network peers are aware of | |
1479 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1480 | * a device wants to inform the rest of the network about some sort of | |
1481 | * reconfiguration such as a failover event or virtual machine | |
1482 | * migration. | |
1483 | */ | |
1484 | void __netdev_notify_peers(struct net_device *dev) | |
1485 | { | |
1486 | ASSERT_RTNL(); | |
1487 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1488 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); | |
1489 | } | |
1490 | EXPORT_SYMBOL(__netdev_notify_peers); | |
1491 | ||
ee89bab1 | 1492 | /** |
722c9a0c | 1493 | * netdev_notify_peers - notify network peers about existence of @dev |
1494 | * @dev: network device | |
ee89bab1 AW |
1495 | * |
1496 | * Generate traffic such that interested network peers are aware of | |
1497 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1498 | * a device wants to inform the rest of the network about some sort of | |
1499 | * reconfiguration such as a failover event or virtual machine | |
1500 | * migration. | |
1501 | */ | |
1502 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1503 | { |
ee89bab1 | 1504 | rtnl_lock(); |
7061eb8c | 1505 | __netdev_notify_peers(dev); |
ee89bab1 | 1506 | rtnl_unlock(); |
c1da4ac7 | 1507 | } |
ee89bab1 | 1508 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1509 | |
29863d41 WW |
1510 | static int napi_threaded_poll(void *data); |
1511 | ||
1512 | static int napi_kthread_create(struct napi_struct *n) | |
1513 | { | |
1514 | int err = 0; | |
1515 | ||
1516 | /* Create and wake up the kthread once to put it in | |
1517 | * TASK_INTERRUPTIBLE mode to avoid the blocked task | |
1518 | * warning and work with loadavg. | |
1519 | */ | |
1520 | n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", | |
1521 | n->dev->name, n->napi_id); | |
1522 | if (IS_ERR(n->thread)) { | |
1523 | err = PTR_ERR(n->thread); | |
1524 | pr_err("kthread_run failed with err %d\n", err); | |
1525 | n->thread = NULL; | |
1526 | } | |
1527 | ||
1528 | return err; | |
1529 | } | |
1530 | ||
40c900aa | 1531 | static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
1da177e4 | 1532 | { |
d314774c | 1533 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1534 | int ret; |
1da177e4 | 1535 | |
e46b66bc BH |
1536 | ASSERT_RTNL(); |
1537 | ||
bd869245 HK |
1538 | if (!netif_device_present(dev)) { |
1539 | /* may be detached because parent is runtime-suspended */ | |
1540 | if (dev->dev.parent) | |
1541 | pm_runtime_resume(dev->dev.parent); | |
1542 | if (!netif_device_present(dev)) | |
1543 | return -ENODEV; | |
1544 | } | |
1da177e4 | 1545 | |
ca99ca14 NH |
1546 | /* Block netpoll from trying to do any rx path servicing. |
1547 | * If we don't do this there is a chance ndo_poll_controller | |
1548 | * or ndo_poll may be running while we open the device | |
1549 | */ | |
66b5552f | 1550 | netpoll_poll_disable(dev); |
ca99ca14 | 1551 | |
40c900aa | 1552 | ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); |
3b8bcfd5 JB |
1553 | ret = notifier_to_errno(ret); |
1554 | if (ret) | |
1555 | return ret; | |
1556 | ||
1da177e4 | 1557 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1558 | |
d314774c SH |
1559 | if (ops->ndo_validate_addr) |
1560 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1561 | |
d314774c SH |
1562 | if (!ret && ops->ndo_open) |
1563 | ret = ops->ndo_open(dev); | |
1da177e4 | 1564 | |
66b5552f | 1565 | netpoll_poll_enable(dev); |
ca99ca14 | 1566 | |
bada339b JG |
1567 | if (ret) |
1568 | clear_bit(__LINK_STATE_START, &dev->state); | |
1569 | else { | |
1da177e4 | 1570 | dev->flags |= IFF_UP; |
4417da66 | 1571 | dev_set_rx_mode(dev); |
1da177e4 | 1572 | dev_activate(dev); |
7bf23575 | 1573 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1574 | } |
bada339b | 1575 | |
1da177e4 LT |
1576 | return ret; |
1577 | } | |
1578 | ||
1579 | /** | |
bd380811 | 1580 | * dev_open - prepare an interface for use. |
00f54e68 PM |
1581 | * @dev: device to open |
1582 | * @extack: netlink extended ack | |
1da177e4 | 1583 | * |
bd380811 PM |
1584 | * Takes a device from down to up state. The device's private open |
1585 | * function is invoked and then the multicast lists are loaded. Finally | |
1586 | * the device is moved into the up state and a %NETDEV_UP message is | |
1587 | * sent to the netdev notifier chain. | |
1588 | * | |
1589 | * Calling this function on an active interface is a nop. On a failure | |
1590 | * a negative errno code is returned. | |
1da177e4 | 1591 | */ |
00f54e68 | 1592 | int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
bd380811 PM |
1593 | { |
1594 | int ret; | |
1595 | ||
bd380811 PM |
1596 | if (dev->flags & IFF_UP) |
1597 | return 0; | |
1598 | ||
40c900aa | 1599 | ret = __dev_open(dev, extack); |
bd380811 PM |
1600 | if (ret < 0) |
1601 | return ret; | |
1602 | ||
7f294054 | 1603 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1604 | call_netdevice_notifiers(NETDEV_UP, dev); |
1605 | ||
1606 | return ret; | |
1607 | } | |
1608 | EXPORT_SYMBOL(dev_open); | |
1609 | ||
7051b88a | 1610 | static void __dev_close_many(struct list_head *head) |
1da177e4 | 1611 | { |
44345724 | 1612 | struct net_device *dev; |
e46b66bc | 1613 | |
bd380811 | 1614 | ASSERT_RTNL(); |
9d5010db DM |
1615 | might_sleep(); |
1616 | ||
5cde2829 | 1617 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1618 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1619 | netpoll_poll_disable(dev); |
3f4df206 | 1620 | |
44345724 | 1621 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1622 | |
44345724 | 1623 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1624 | |
44345724 OP |
1625 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1626 | * can be even on different cpu. So just clear netif_running(). | |
1627 | * | |
1628 | * dev->stop() will invoke napi_disable() on all of it's | |
1629 | * napi_struct instances on this device. | |
1630 | */ | |
4e857c58 | 1631 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1632 | } |
1da177e4 | 1633 | |
44345724 | 1634 | dev_deactivate_many(head); |
d8b2a4d2 | 1635 | |
5cde2829 | 1636 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1637 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1638 | |
44345724 OP |
1639 | /* |
1640 | * Call the device specific close. This cannot fail. | |
1641 | * Only if device is UP | |
1642 | * | |
1643 | * We allow it to be called even after a DETACH hot-plug | |
1644 | * event. | |
1645 | */ | |
1646 | if (ops->ndo_stop) | |
1647 | ops->ndo_stop(dev); | |
1648 | ||
44345724 | 1649 | dev->flags &= ~IFF_UP; |
66b5552f | 1650 | netpoll_poll_enable(dev); |
44345724 | 1651 | } |
44345724 OP |
1652 | } |
1653 | ||
7051b88a | 1654 | static void __dev_close(struct net_device *dev) |
44345724 OP |
1655 | { |
1656 | LIST_HEAD(single); | |
1657 | ||
5cde2829 | 1658 | list_add(&dev->close_list, &single); |
7051b88a | 1659 | __dev_close_many(&single); |
f87e6f47 | 1660 | list_del(&single); |
44345724 OP |
1661 | } |
1662 | ||
7051b88a | 1663 | void dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1664 | { |
1665 | struct net_device *dev, *tmp; | |
1da177e4 | 1666 | |
5cde2829 EB |
1667 | /* Remove the devices that don't need to be closed */ |
1668 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1669 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1670 | list_del_init(&dev->close_list); |
44345724 OP |
1671 | |
1672 | __dev_close_many(head); | |
1da177e4 | 1673 | |
5cde2829 | 1674 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1675 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1676 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1677 | if (unlink) |
1678 | list_del_init(&dev->close_list); | |
44345724 | 1679 | } |
bd380811 | 1680 | } |
99c4a26a | 1681 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1682 | |
1683 | /** | |
1684 | * dev_close - shutdown an interface. | |
1685 | * @dev: device to shutdown | |
1686 | * | |
1687 | * This function moves an active device into down state. A | |
1688 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1689 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1690 | * chain. | |
1691 | */ | |
7051b88a | 1692 | void dev_close(struct net_device *dev) |
bd380811 | 1693 | { |
e14a5993 ED |
1694 | if (dev->flags & IFF_UP) { |
1695 | LIST_HEAD(single); | |
1da177e4 | 1696 | |
5cde2829 | 1697 | list_add(&dev->close_list, &single); |
99c4a26a | 1698 | dev_close_many(&single, true); |
e14a5993 ED |
1699 | list_del(&single); |
1700 | } | |
1da177e4 | 1701 | } |
d1b19dff | 1702 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1703 | |
1704 | ||
0187bdfb BH |
1705 | /** |
1706 | * dev_disable_lro - disable Large Receive Offload on a device | |
1707 | * @dev: device | |
1708 | * | |
1709 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1710 | * called under RTNL. This is needed if received packets may be | |
1711 | * forwarded to another interface. | |
1712 | */ | |
1713 | void dev_disable_lro(struct net_device *dev) | |
1714 | { | |
fbe168ba MK |
1715 | struct net_device *lower_dev; |
1716 | struct list_head *iter; | |
529d0489 | 1717 | |
bc5787c6 MM |
1718 | dev->wanted_features &= ~NETIF_F_LRO; |
1719 | netdev_update_features(dev); | |
27660515 | 1720 | |
22d5969f MM |
1721 | if (unlikely(dev->features & NETIF_F_LRO)) |
1722 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1723 | |
1724 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1725 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1726 | } |
1727 | EXPORT_SYMBOL(dev_disable_lro); | |
1728 | ||
56f5aa77 MC |
1729 | /** |
1730 | * dev_disable_gro_hw - disable HW Generic Receive Offload on a device | |
1731 | * @dev: device | |
1732 | * | |
1733 | * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be | |
1734 | * called under RTNL. This is needed if Generic XDP is installed on | |
1735 | * the device. | |
1736 | */ | |
1737 | static void dev_disable_gro_hw(struct net_device *dev) | |
1738 | { | |
1739 | dev->wanted_features &= ~NETIF_F_GRO_HW; | |
1740 | netdev_update_features(dev); | |
1741 | ||
1742 | if (unlikely(dev->features & NETIF_F_GRO_HW)) | |
1743 | netdev_WARN(dev, "failed to disable GRO_HW!\n"); | |
1744 | } | |
1745 | ||
ede2762d KT |
1746 | const char *netdev_cmd_to_name(enum netdev_cmd cmd) |
1747 | { | |
1748 | #define N(val) \ | |
1749 | case NETDEV_##val: \ | |
1750 | return "NETDEV_" __stringify(val); | |
1751 | switch (cmd) { | |
1752 | N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) | |
1753 | N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) | |
1754 | N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) | |
1755 | N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) | |
1756 | N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) | |
1757 | N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) | |
1758 | N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) | |
9daae9bd GP |
1759 | N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) |
1760 | N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) | |
1570415f | 1761 | N(PRE_CHANGEADDR) |
3f5ecd8a | 1762 | } |
ede2762d KT |
1763 | #undef N |
1764 | return "UNKNOWN_NETDEV_EVENT"; | |
1765 | } | |
1766 | EXPORT_SYMBOL_GPL(netdev_cmd_to_name); | |
1767 | ||
351638e7 JP |
1768 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1769 | struct net_device *dev) | |
1770 | { | |
51d0c047 DA |
1771 | struct netdev_notifier_info info = { |
1772 | .dev = dev, | |
1773 | }; | |
351638e7 | 1774 | |
351638e7 JP |
1775 | return nb->notifier_call(nb, val, &info); |
1776 | } | |
0187bdfb | 1777 | |
afa0df59 JP |
1778 | static int call_netdevice_register_notifiers(struct notifier_block *nb, |
1779 | struct net_device *dev) | |
1780 | { | |
1781 | int err; | |
1782 | ||
1783 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); | |
1784 | err = notifier_to_errno(err); | |
1785 | if (err) | |
1786 | return err; | |
1787 | ||
1788 | if (!(dev->flags & IFF_UP)) | |
1789 | return 0; | |
1790 | ||
1791 | call_netdevice_notifier(nb, NETDEV_UP, dev); | |
1792 | return 0; | |
1793 | } | |
1794 | ||
1795 | static void call_netdevice_unregister_notifiers(struct notifier_block *nb, | |
1796 | struct net_device *dev) | |
1797 | { | |
1798 | if (dev->flags & IFF_UP) { | |
1799 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, | |
1800 | dev); | |
1801 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
1802 | } | |
1803 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); | |
1804 | } | |
1805 | ||
1806 | static int call_netdevice_register_net_notifiers(struct notifier_block *nb, | |
1807 | struct net *net) | |
1808 | { | |
1809 | struct net_device *dev; | |
1810 | int err; | |
1811 | ||
1812 | for_each_netdev(net, dev) { | |
1813 | err = call_netdevice_register_notifiers(nb, dev); | |
1814 | if (err) | |
1815 | goto rollback; | |
1816 | } | |
1817 | return 0; | |
1818 | ||
1819 | rollback: | |
1820 | for_each_netdev_continue_reverse(net, dev) | |
1821 | call_netdevice_unregister_notifiers(nb, dev); | |
1822 | return err; | |
1823 | } | |
1824 | ||
1825 | static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, | |
1826 | struct net *net) | |
1827 | { | |
1828 | struct net_device *dev; | |
1829 | ||
1830 | for_each_netdev(net, dev) | |
1831 | call_netdevice_unregister_notifiers(nb, dev); | |
1832 | } | |
1833 | ||
881d966b EB |
1834 | static int dev_boot_phase = 1; |
1835 | ||
1da177e4 | 1836 | /** |
722c9a0c | 1837 | * register_netdevice_notifier - register a network notifier block |
1838 | * @nb: notifier | |
1da177e4 | 1839 | * |
722c9a0c | 1840 | * Register a notifier to be called when network device events occur. |
1841 | * The notifier passed is linked into the kernel structures and must | |
1842 | * not be reused until it has been unregistered. A negative errno code | |
1843 | * is returned on a failure. | |
1da177e4 | 1844 | * |
722c9a0c | 1845 | * When registered all registration and up events are replayed |
1846 | * to the new notifier to allow device to have a race free | |
1847 | * view of the network device list. | |
1da177e4 LT |
1848 | */ |
1849 | ||
1850 | int register_netdevice_notifier(struct notifier_block *nb) | |
1851 | { | |
881d966b | 1852 | struct net *net; |
1da177e4 LT |
1853 | int err; |
1854 | ||
328fbe74 KT |
1855 | /* Close race with setup_net() and cleanup_net() */ |
1856 | down_write(&pernet_ops_rwsem); | |
1da177e4 | 1857 | rtnl_lock(); |
f07d5b94 | 1858 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1859 | if (err) |
1860 | goto unlock; | |
881d966b EB |
1861 | if (dev_boot_phase) |
1862 | goto unlock; | |
1863 | for_each_net(net) { | |
afa0df59 JP |
1864 | err = call_netdevice_register_net_notifiers(nb, net); |
1865 | if (err) | |
1866 | goto rollback; | |
1da177e4 | 1867 | } |
fcc5a03a HX |
1868 | |
1869 | unlock: | |
1da177e4 | 1870 | rtnl_unlock(); |
328fbe74 | 1871 | up_write(&pernet_ops_rwsem); |
1da177e4 | 1872 | return err; |
fcc5a03a HX |
1873 | |
1874 | rollback: | |
afa0df59 JP |
1875 | for_each_net_continue_reverse(net) |
1876 | call_netdevice_unregister_net_notifiers(nb, net); | |
c67625a1 PE |
1877 | |
1878 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1879 | goto unlock; |
1da177e4 | 1880 | } |
d1b19dff | 1881 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1882 | |
1883 | /** | |
722c9a0c | 1884 | * unregister_netdevice_notifier - unregister a network notifier block |
1885 | * @nb: notifier | |
1da177e4 | 1886 | * |
722c9a0c | 1887 | * Unregister a notifier previously registered by |
1888 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1889 | * kernel structures and may then be reused. A negative errno code | |
1890 | * is returned on a failure. | |
7d3d43da | 1891 | * |
722c9a0c | 1892 | * After unregistering unregister and down device events are synthesized |
1893 | * for all devices on the device list to the removed notifier to remove | |
1894 | * the need for special case cleanup code. | |
1da177e4 LT |
1895 | */ |
1896 | ||
1897 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1898 | { | |
7d3d43da | 1899 | struct net *net; |
9f514950 HX |
1900 | int err; |
1901 | ||
328fbe74 KT |
1902 | /* Close race with setup_net() and cleanup_net() */ |
1903 | down_write(&pernet_ops_rwsem); | |
9f514950 | 1904 | rtnl_lock(); |
f07d5b94 | 1905 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1906 | if (err) |
1907 | goto unlock; | |
1908 | ||
48b3a137 JP |
1909 | for_each_net(net) |
1910 | call_netdevice_unregister_net_notifiers(nb, net); | |
1911 | ||
7d3d43da | 1912 | unlock: |
9f514950 | 1913 | rtnl_unlock(); |
328fbe74 | 1914 | up_write(&pernet_ops_rwsem); |
9f514950 | 1915 | return err; |
1da177e4 | 1916 | } |
d1b19dff | 1917 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1918 | |
1f637703 JP |
1919 | static int __register_netdevice_notifier_net(struct net *net, |
1920 | struct notifier_block *nb, | |
1921 | bool ignore_call_fail) | |
1922 | { | |
1923 | int err; | |
1924 | ||
1925 | err = raw_notifier_chain_register(&net->netdev_chain, nb); | |
1926 | if (err) | |
1927 | return err; | |
1928 | if (dev_boot_phase) | |
1929 | return 0; | |
1930 | ||
1931 | err = call_netdevice_register_net_notifiers(nb, net); | |
1932 | if (err && !ignore_call_fail) | |
1933 | goto chain_unregister; | |
1934 | ||
1935 | return 0; | |
1936 | ||
1937 | chain_unregister: | |
1938 | raw_notifier_chain_unregister(&net->netdev_chain, nb); | |
1939 | return err; | |
1940 | } | |
1941 | ||
1942 | static int __unregister_netdevice_notifier_net(struct net *net, | |
1943 | struct notifier_block *nb) | |
1944 | { | |
1945 | int err; | |
1946 | ||
1947 | err = raw_notifier_chain_unregister(&net->netdev_chain, nb); | |
1948 | if (err) | |
1949 | return err; | |
1950 | ||
1951 | call_netdevice_unregister_net_notifiers(nb, net); | |
1952 | return 0; | |
1953 | } | |
1954 | ||
a30c7b42 JP |
1955 | /** |
1956 | * register_netdevice_notifier_net - register a per-netns network notifier block | |
1957 | * @net: network namespace | |
1958 | * @nb: notifier | |
1959 | * | |
1960 | * Register a notifier to be called when network device events occur. | |
1961 | * The notifier passed is linked into the kernel structures and must | |
1962 | * not be reused until it has been unregistered. A negative errno code | |
1963 | * is returned on a failure. | |
1964 | * | |
1965 | * When registered all registration and up events are replayed | |
1966 | * to the new notifier to allow device to have a race free | |
1967 | * view of the network device list. | |
1968 | */ | |
1969 | ||
1970 | int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) | |
1971 | { | |
1972 | int err; | |
1973 | ||
1974 | rtnl_lock(); | |
1f637703 | 1975 | err = __register_netdevice_notifier_net(net, nb, false); |
a30c7b42 JP |
1976 | rtnl_unlock(); |
1977 | return err; | |
a30c7b42 JP |
1978 | } |
1979 | EXPORT_SYMBOL(register_netdevice_notifier_net); | |
1980 | ||
1981 | /** | |
1982 | * unregister_netdevice_notifier_net - unregister a per-netns | |
1983 | * network notifier block | |
1984 | * @net: network namespace | |
1985 | * @nb: notifier | |
1986 | * | |
1987 | * Unregister a notifier previously registered by | |
1988 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1989 | * kernel structures and may then be reused. A negative errno code | |
1990 | * is returned on a failure. | |
1991 | * | |
1992 | * After unregistering unregister and down device events are synthesized | |
1993 | * for all devices on the device list to the removed notifier to remove | |
1994 | * the need for special case cleanup code. | |
1995 | */ | |
1996 | ||
1997 | int unregister_netdevice_notifier_net(struct net *net, | |
1998 | struct notifier_block *nb) | |
1999 | { | |
2000 | int err; | |
2001 | ||
2002 | rtnl_lock(); | |
1f637703 | 2003 | err = __unregister_netdevice_notifier_net(net, nb); |
a30c7b42 JP |
2004 | rtnl_unlock(); |
2005 | return err; | |
2006 | } | |
2007 | EXPORT_SYMBOL(unregister_netdevice_notifier_net); | |
a30c7b42 | 2008 | |
93642e14 JP |
2009 | int register_netdevice_notifier_dev_net(struct net_device *dev, |
2010 | struct notifier_block *nb, | |
2011 | struct netdev_net_notifier *nn) | |
2012 | { | |
2013 | int err; | |
a30c7b42 | 2014 | |
93642e14 JP |
2015 | rtnl_lock(); |
2016 | err = __register_netdevice_notifier_net(dev_net(dev), nb, false); | |
2017 | if (!err) { | |
2018 | nn->nb = nb; | |
2019 | list_add(&nn->list, &dev->net_notifier_list); | |
2020 | } | |
a30c7b42 JP |
2021 | rtnl_unlock(); |
2022 | return err; | |
2023 | } | |
93642e14 JP |
2024 | EXPORT_SYMBOL(register_netdevice_notifier_dev_net); |
2025 | ||
2026 | int unregister_netdevice_notifier_dev_net(struct net_device *dev, | |
2027 | struct notifier_block *nb, | |
2028 | struct netdev_net_notifier *nn) | |
2029 | { | |
2030 | int err; | |
2031 | ||
2032 | rtnl_lock(); | |
2033 | list_del(&nn->list); | |
2034 | err = __unregister_netdevice_notifier_net(dev_net(dev), nb); | |
2035 | rtnl_unlock(); | |
2036 | return err; | |
2037 | } | |
2038 | EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); | |
2039 | ||
2040 | static void move_netdevice_notifiers_dev_net(struct net_device *dev, | |
2041 | struct net *net) | |
2042 | { | |
2043 | struct netdev_net_notifier *nn; | |
2044 | ||
2045 | list_for_each_entry(nn, &dev->net_notifier_list, list) { | |
2046 | __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); | |
2047 | __register_netdevice_notifier_net(net, nn->nb, true); | |
2048 | } | |
2049 | } | |
a30c7b42 | 2050 | |
351638e7 JP |
2051 | /** |
2052 | * call_netdevice_notifiers_info - call all network notifier blocks | |
2053 | * @val: value passed unmodified to notifier function | |
351638e7 JP |
2054 | * @info: notifier information data |
2055 | * | |
2056 | * Call all network notifier blocks. Parameters and return value | |
2057 | * are as for raw_notifier_call_chain(). | |
2058 | */ | |
2059 | ||
1d143d9f | 2060 | static int call_netdevice_notifiers_info(unsigned long val, |
1d143d9f | 2061 | struct netdev_notifier_info *info) |
351638e7 | 2062 | { |
a30c7b42 JP |
2063 | struct net *net = dev_net(info->dev); |
2064 | int ret; | |
2065 | ||
351638e7 | 2066 | ASSERT_RTNL(); |
a30c7b42 JP |
2067 | |
2068 | /* Run per-netns notifier block chain first, then run the global one. | |
2069 | * Hopefully, one day, the global one is going to be removed after | |
2070 | * all notifier block registrators get converted to be per-netns. | |
2071 | */ | |
2072 | ret = raw_notifier_call_chain(&net->netdev_chain, val, info); | |
2073 | if (ret & NOTIFY_STOP_MASK) | |
2074 | return ret; | |
351638e7 JP |
2075 | return raw_notifier_call_chain(&netdev_chain, val, info); |
2076 | } | |
351638e7 | 2077 | |
26372605 PM |
2078 | static int call_netdevice_notifiers_extack(unsigned long val, |
2079 | struct net_device *dev, | |
2080 | struct netlink_ext_ack *extack) | |
2081 | { | |
2082 | struct netdev_notifier_info info = { | |
2083 | .dev = dev, | |
2084 | .extack = extack, | |
2085 | }; | |
2086 | ||
2087 | return call_netdevice_notifiers_info(val, &info); | |
2088 | } | |
2089 | ||
1da177e4 LT |
2090 | /** |
2091 | * call_netdevice_notifiers - call all network notifier blocks | |
2092 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 2093 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
2094 | * |
2095 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 2096 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
2097 | */ |
2098 | ||
ad7379d4 | 2099 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 2100 | { |
26372605 | 2101 | return call_netdevice_notifiers_extack(val, dev, NULL); |
1da177e4 | 2102 | } |
edf947f1 | 2103 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 2104 | |
af7d6cce SD |
2105 | /** |
2106 | * call_netdevice_notifiers_mtu - call all network notifier blocks | |
2107 | * @val: value passed unmodified to notifier function | |
2108 | * @dev: net_device pointer passed unmodified to notifier function | |
2109 | * @arg: additional u32 argument passed to the notifier function | |
2110 | * | |
2111 | * Call all network notifier blocks. Parameters and return value | |
2112 | * are as for raw_notifier_call_chain(). | |
2113 | */ | |
2114 | static int call_netdevice_notifiers_mtu(unsigned long val, | |
2115 | struct net_device *dev, u32 arg) | |
2116 | { | |
2117 | struct netdev_notifier_info_ext info = { | |
2118 | .info.dev = dev, | |
2119 | .ext.mtu = arg, | |
2120 | }; | |
2121 | ||
2122 | BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); | |
2123 | ||
2124 | return call_netdevice_notifiers_info(val, &info.info); | |
2125 | } | |
2126 | ||
1cf51900 | 2127 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 2128 | static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); |
4577139b DB |
2129 | |
2130 | void net_inc_ingress_queue(void) | |
2131 | { | |
aabf6772 | 2132 | static_branch_inc(&ingress_needed_key); |
4577139b DB |
2133 | } |
2134 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
2135 | ||
2136 | void net_dec_ingress_queue(void) | |
2137 | { | |
aabf6772 | 2138 | static_branch_dec(&ingress_needed_key); |
4577139b DB |
2139 | } |
2140 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
2141 | #endif | |
2142 | ||
1f211a1b | 2143 | #ifdef CONFIG_NET_EGRESS |
aabf6772 | 2144 | static DEFINE_STATIC_KEY_FALSE(egress_needed_key); |
1f211a1b DB |
2145 | |
2146 | void net_inc_egress_queue(void) | |
2147 | { | |
aabf6772 | 2148 | static_branch_inc(&egress_needed_key); |
1f211a1b DB |
2149 | } |
2150 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
2151 | ||
2152 | void net_dec_egress_queue(void) | |
2153 | { | |
aabf6772 | 2154 | static_branch_dec(&egress_needed_key); |
1f211a1b DB |
2155 | } |
2156 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
2157 | #endif | |
2158 | ||
39e83922 | 2159 | static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); |
e9666d10 | 2160 | #ifdef CONFIG_JUMP_LABEL |
b90e5794 | 2161 | static atomic_t netstamp_needed_deferred; |
13baa00a | 2162 | static atomic_t netstamp_wanted; |
5fa8bbda | 2163 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 2164 | { |
b90e5794 | 2165 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 2166 | int wanted; |
b90e5794 | 2167 | |
13baa00a ED |
2168 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
2169 | if (wanted > 0) | |
39e83922 | 2170 | static_branch_enable(&netstamp_needed_key); |
13baa00a | 2171 | else |
39e83922 | 2172 | static_branch_disable(&netstamp_needed_key); |
5fa8bbda ED |
2173 | } |
2174 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 2175 | #endif |
5fa8bbda ED |
2176 | |
2177 | void net_enable_timestamp(void) | |
2178 | { | |
e9666d10 | 2179 | #ifdef CONFIG_JUMP_LABEL |
13baa00a ED |
2180 | int wanted; |
2181 | ||
2182 | while (1) { | |
2183 | wanted = atomic_read(&netstamp_wanted); | |
2184 | if (wanted <= 0) | |
2185 | break; | |
2186 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) | |
2187 | return; | |
2188 | } | |
2189 | atomic_inc(&netstamp_needed_deferred); | |
2190 | schedule_work(&netstamp_work); | |
2191 | #else | |
39e83922 | 2192 | static_branch_inc(&netstamp_needed_key); |
13baa00a | 2193 | #endif |
1da177e4 | 2194 | } |
d1b19dff | 2195 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
2196 | |
2197 | void net_disable_timestamp(void) | |
2198 | { | |
e9666d10 | 2199 | #ifdef CONFIG_JUMP_LABEL |
13baa00a ED |
2200 | int wanted; |
2201 | ||
2202 | while (1) { | |
2203 | wanted = atomic_read(&netstamp_wanted); | |
2204 | if (wanted <= 1) | |
2205 | break; | |
2206 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) | |
2207 | return; | |
2208 | } | |
2209 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
2210 | schedule_work(&netstamp_work); |
2211 | #else | |
39e83922 | 2212 | static_branch_dec(&netstamp_needed_key); |
5fa8bbda | 2213 | #endif |
1da177e4 | 2214 | } |
d1b19dff | 2215 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 2216 | |
3b098e2d | 2217 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 2218 | { |
2456e855 | 2219 | skb->tstamp = 0; |
39e83922 | 2220 | if (static_branch_unlikely(&netstamp_needed_key)) |
a61bbcf2 | 2221 | __net_timestamp(skb); |
1da177e4 LT |
2222 | } |
2223 | ||
39e83922 DB |
2224 | #define net_timestamp_check(COND, SKB) \ |
2225 | if (static_branch_unlikely(&netstamp_needed_key)) { \ | |
2226 | if ((COND) && !(SKB)->tstamp) \ | |
2227 | __net_timestamp(SKB); \ | |
2228 | } \ | |
3b098e2d | 2229 | |
f4b05d27 | 2230 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 | 2231 | { |
5f7d5728 | 2232 | return __is_skb_forwardable(dev, skb, true); |
79b569f0 | 2233 | } |
1ee481fb | 2234 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 2235 | |
5f7d5728 JDB |
2236 | static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, |
2237 | bool check_mtu) | |
a0265d28 | 2238 | { |
5f7d5728 | 2239 | int ret = ____dev_forward_skb(dev, skb, check_mtu); |
a0265d28 | 2240 | |
4e3264d2 MKL |
2241 | if (likely(!ret)) { |
2242 | skb->protocol = eth_type_trans(skb, dev); | |
2243 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
2244 | } | |
a0265d28 | 2245 | |
4e3264d2 | 2246 | return ret; |
a0265d28 | 2247 | } |
5f7d5728 JDB |
2248 | |
2249 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
2250 | { | |
2251 | return __dev_forward_skb2(dev, skb, true); | |
2252 | } | |
a0265d28 HX |
2253 | EXPORT_SYMBOL_GPL(__dev_forward_skb); |
2254 | ||
44540960 AB |
2255 | /** |
2256 | * dev_forward_skb - loopback an skb to another netif | |
2257 | * | |
2258 | * @dev: destination network device | |
2259 | * @skb: buffer to forward | |
2260 | * | |
2261 | * return values: | |
2262 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 2263 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
2264 | * |
2265 | * dev_forward_skb can be used for injecting an skb from the | |
2266 | * start_xmit function of one device into the receive queue | |
2267 | * of another device. | |
2268 | * | |
2269 | * The receiving device may be in another namespace, so | |
2270 | * we have to clear all information in the skb that could | |
2271 | * impact namespace isolation. | |
2272 | */ | |
2273 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
2274 | { | |
a0265d28 | 2275 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
2276 | } |
2277 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
2278 | ||
5f7d5728 JDB |
2279 | int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) |
2280 | { | |
2281 | return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); | |
2282 | } | |
2283 | ||
71d9dec2 CG |
2284 | static inline int deliver_skb(struct sk_buff *skb, |
2285 | struct packet_type *pt_prev, | |
2286 | struct net_device *orig_dev) | |
2287 | { | |
1f8b977a | 2288 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
1080e512 | 2289 | return -ENOMEM; |
63354797 | 2290 | refcount_inc(&skb->users); |
71d9dec2 CG |
2291 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
2292 | } | |
2293 | ||
7866a621 SN |
2294 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
2295 | struct packet_type **pt, | |
fbcb2170 JP |
2296 | struct net_device *orig_dev, |
2297 | __be16 type, | |
7866a621 SN |
2298 | struct list_head *ptype_list) |
2299 | { | |
2300 | struct packet_type *ptype, *pt_prev = *pt; | |
2301 | ||
2302 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
2303 | if (ptype->type != type) | |
2304 | continue; | |
2305 | if (pt_prev) | |
fbcb2170 | 2306 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
2307 | pt_prev = ptype; |
2308 | } | |
2309 | *pt = pt_prev; | |
2310 | } | |
2311 | ||
c0de08d0 EL |
2312 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
2313 | { | |
a3d744e9 | 2314 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
2315 | return false; |
2316 | ||
2317 | if (ptype->id_match) | |
2318 | return ptype->id_match(ptype, skb->sk); | |
2319 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
2320 | return true; | |
2321 | ||
2322 | return false; | |
2323 | } | |
2324 | ||
9f9a742d MR |
2325 | /** |
2326 | * dev_nit_active - return true if any network interface taps are in use | |
2327 | * | |
2328 | * @dev: network device to check for the presence of taps | |
2329 | */ | |
2330 | bool dev_nit_active(struct net_device *dev) | |
2331 | { | |
2332 | return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); | |
2333 | } | |
2334 | EXPORT_SYMBOL_GPL(dev_nit_active); | |
2335 | ||
1da177e4 LT |
2336 | /* |
2337 | * Support routine. Sends outgoing frames to any network | |
2338 | * taps currently in use. | |
2339 | */ | |
2340 | ||
74b20582 | 2341 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
2342 | { |
2343 | struct packet_type *ptype; | |
71d9dec2 CG |
2344 | struct sk_buff *skb2 = NULL; |
2345 | struct packet_type *pt_prev = NULL; | |
7866a621 | 2346 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 2347 | |
1da177e4 | 2348 | rcu_read_lock(); |
7866a621 SN |
2349 | again: |
2350 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
fa788d98 VW |
2351 | if (ptype->ignore_outgoing) |
2352 | continue; | |
2353 | ||
1da177e4 LT |
2354 | /* Never send packets back to the socket |
2355 | * they originated from - MvS (miquels@drinkel.ow.org) | |
2356 | */ | |
7866a621 SN |
2357 | if (skb_loop_sk(ptype, skb)) |
2358 | continue; | |
71d9dec2 | 2359 | |
7866a621 SN |
2360 | if (pt_prev) { |
2361 | deliver_skb(skb2, pt_prev, skb->dev); | |
2362 | pt_prev = ptype; | |
2363 | continue; | |
2364 | } | |
1da177e4 | 2365 | |
7866a621 SN |
2366 | /* need to clone skb, done only once */ |
2367 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
2368 | if (!skb2) | |
2369 | goto out_unlock; | |
70978182 | 2370 | |
7866a621 | 2371 | net_timestamp_set(skb2); |
1da177e4 | 2372 | |
7866a621 SN |
2373 | /* skb->nh should be correctly |
2374 | * set by sender, so that the second statement is | |
2375 | * just protection against buggy protocols. | |
2376 | */ | |
2377 | skb_reset_mac_header(skb2); | |
2378 | ||
2379 | if (skb_network_header(skb2) < skb2->data || | |
2380 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
2381 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
2382 | ntohs(skb2->protocol), | |
2383 | dev->name); | |
2384 | skb_reset_network_header(skb2); | |
1da177e4 | 2385 | } |
7866a621 SN |
2386 | |
2387 | skb2->transport_header = skb2->network_header; | |
2388 | skb2->pkt_type = PACKET_OUTGOING; | |
2389 | pt_prev = ptype; | |
2390 | } | |
2391 | ||
2392 | if (ptype_list == &ptype_all) { | |
2393 | ptype_list = &dev->ptype_all; | |
2394 | goto again; | |
1da177e4 | 2395 | } |
7866a621 | 2396 | out_unlock: |
581fe0ea WB |
2397 | if (pt_prev) { |
2398 | if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) | |
2399 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
2400 | else | |
2401 | kfree_skb(skb2); | |
2402 | } | |
1da177e4 LT |
2403 | rcu_read_unlock(); |
2404 | } | |
74b20582 | 2405 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 2406 | |
2c53040f BH |
2407 | /** |
2408 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
2409 | * @dev: Network device |
2410 | * @txq: number of queues available | |
2411 | * | |
2412 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
2413 | * valid. To resolve this verify the tc mapping remains valid and if | |
2414 | * not NULL the mapping. With no priorities mapping to this | |
2415 | * offset/count pair it will no longer be used. In the worst case TC0 | |
2416 | * is invalid nothing can be done so disable priority mappings. If is | |
2417 | * expected that drivers will fix this mapping if they can before | |
2418 | * calling netif_set_real_num_tx_queues. | |
2419 | */ | |
bb134d22 | 2420 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
2421 | { |
2422 | int i; | |
2423 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2424 | ||
2425 | /* If TC0 is invalidated disable TC mapping */ | |
2426 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 2427 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
2428 | dev->num_tc = 0; |
2429 | return; | |
2430 | } | |
2431 | ||
2432 | /* Invalidated prio to tc mappings set to TC0 */ | |
2433 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
2434 | int q = netdev_get_prio_tc_map(dev, i); | |
2435 | ||
2436 | tc = &dev->tc_to_txq[q]; | |
2437 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
2438 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
2439 | i, q); | |
4f57c087 JF |
2440 | netdev_set_prio_tc_map(dev, i, 0); |
2441 | } | |
2442 | } | |
2443 | } | |
2444 | ||
8d059b0f AD |
2445 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
2446 | { | |
2447 | if (dev->num_tc) { | |
2448 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2449 | int i; | |
2450 | ||
ffcfe25b | 2451 | /* walk through the TCs and see if it falls into any of them */ |
8d059b0f AD |
2452 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { |
2453 | if ((txq - tc->offset) < tc->count) | |
2454 | return i; | |
2455 | } | |
2456 | ||
ffcfe25b | 2457 | /* didn't find it, just return -1 to indicate no match */ |
8d059b0f AD |
2458 | return -1; |
2459 | } | |
2460 | ||
2461 | return 0; | |
2462 | } | |
8a5f2166 | 2463 | EXPORT_SYMBOL(netdev_txq_to_tc); |
8d059b0f | 2464 | |
537c00de | 2465 | #ifdef CONFIG_XPS |
04157469 AN |
2466 | struct static_key xps_needed __read_mostly; |
2467 | EXPORT_SYMBOL(xps_needed); | |
2468 | struct static_key xps_rxqs_needed __read_mostly; | |
2469 | EXPORT_SYMBOL(xps_rxqs_needed); | |
537c00de AD |
2470 | static DEFINE_MUTEX(xps_map_mutex); |
2471 | #define xmap_dereference(P) \ | |
2472 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
2473 | ||
6234f874 AD |
2474 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
2475 | int tci, u16 index) | |
537c00de | 2476 | { |
10cdc3f3 AD |
2477 | struct xps_map *map = NULL; |
2478 | int pos; | |
537c00de | 2479 | |
10cdc3f3 | 2480 | if (dev_maps) |
80d19669 | 2481 | map = xmap_dereference(dev_maps->attr_map[tci]); |
6234f874 AD |
2482 | if (!map) |
2483 | return false; | |
537c00de | 2484 | |
6234f874 AD |
2485 | for (pos = map->len; pos--;) { |
2486 | if (map->queues[pos] != index) | |
2487 | continue; | |
2488 | ||
2489 | if (map->len > 1) { | |
2490 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2491 | break; |
537c00de | 2492 | } |
6234f874 | 2493 | |
80d19669 | 2494 | RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); |
6234f874 AD |
2495 | kfree_rcu(map, rcu); |
2496 | return false; | |
537c00de AD |
2497 | } |
2498 | ||
6234f874 | 2499 | return true; |
10cdc3f3 AD |
2500 | } |
2501 | ||
6234f874 AD |
2502 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2503 | struct xps_dev_maps *dev_maps, | |
2504 | int cpu, u16 offset, u16 count) | |
2505 | { | |
184c449f AD |
2506 | int num_tc = dev->num_tc ? : 1; |
2507 | bool active = false; | |
2508 | int tci; | |
6234f874 | 2509 | |
184c449f AD |
2510 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2511 | int i, j; | |
2512 | ||
2513 | for (i = count, j = offset; i--; j++) { | |
6358d49a | 2514 | if (!remove_xps_queue(dev_maps, tci, j)) |
184c449f AD |
2515 | break; |
2516 | } | |
2517 | ||
2518 | active |= i < 0; | |
6234f874 AD |
2519 | } |
2520 | ||
184c449f | 2521 | return active; |
6234f874 AD |
2522 | } |
2523 | ||
867d0ad4 SD |
2524 | static void reset_xps_maps(struct net_device *dev, |
2525 | struct xps_dev_maps *dev_maps, | |
2526 | bool is_rxqs_map) | |
2527 | { | |
2528 | if (is_rxqs_map) { | |
2529 | static_key_slow_dec_cpuslocked(&xps_rxqs_needed); | |
2530 | RCU_INIT_POINTER(dev->xps_rxqs_map, NULL); | |
2531 | } else { | |
2532 | RCU_INIT_POINTER(dev->xps_cpus_map, NULL); | |
2533 | } | |
2534 | static_key_slow_dec_cpuslocked(&xps_needed); | |
2535 | kfree_rcu(dev_maps, rcu); | |
2536 | } | |
2537 | ||
80d19669 AN |
2538 | static void clean_xps_maps(struct net_device *dev, const unsigned long *mask, |
2539 | struct xps_dev_maps *dev_maps, unsigned int nr_ids, | |
2540 | u16 offset, u16 count, bool is_rxqs_map) | |
2541 | { | |
2542 | bool active = false; | |
2543 | int i, j; | |
2544 | ||
2545 | for (j = -1; j = netif_attrmask_next(j, mask, nr_ids), | |
2546 | j < nr_ids;) | |
2547 | active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, | |
2548 | count); | |
867d0ad4 SD |
2549 | if (!active) |
2550 | reset_xps_maps(dev, dev_maps, is_rxqs_map); | |
80d19669 | 2551 | |
f28c020f SD |
2552 | if (!is_rxqs_map) { |
2553 | for (i = offset + (count - 1); count--; i--) { | |
2554 | netdev_queue_numa_node_write( | |
2555 | netdev_get_tx_queue(dev, i), | |
2556 | NUMA_NO_NODE); | |
80d19669 | 2557 | } |
80d19669 AN |
2558 | } |
2559 | } | |
2560 | ||
6234f874 AD |
2561 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, |
2562 | u16 count) | |
10cdc3f3 | 2563 | { |
80d19669 | 2564 | const unsigned long *possible_mask = NULL; |
10cdc3f3 | 2565 | struct xps_dev_maps *dev_maps; |
80d19669 | 2566 | unsigned int nr_ids; |
10cdc3f3 | 2567 | |
04157469 AN |
2568 | if (!static_key_false(&xps_needed)) |
2569 | return; | |
10cdc3f3 | 2570 | |
4d99f660 | 2571 | cpus_read_lock(); |
04157469 | 2572 | mutex_lock(&xps_map_mutex); |
10cdc3f3 | 2573 | |
04157469 AN |
2574 | if (static_key_false(&xps_rxqs_needed)) { |
2575 | dev_maps = xmap_dereference(dev->xps_rxqs_map); | |
2576 | if (dev_maps) { | |
2577 | nr_ids = dev->num_rx_queues; | |
2578 | clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, | |
2579 | offset, count, true); | |
2580 | } | |
537c00de AD |
2581 | } |
2582 | ||
80d19669 AN |
2583 | dev_maps = xmap_dereference(dev->xps_cpus_map); |
2584 | if (!dev_maps) | |
2585 | goto out_no_maps; | |
2586 | ||
2587 | if (num_possible_cpus() > 1) | |
2588 | possible_mask = cpumask_bits(cpu_possible_mask); | |
2589 | nr_ids = nr_cpu_ids; | |
2590 | clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count, | |
2591 | false); | |
024e9679 | 2592 | |
537c00de AD |
2593 | out_no_maps: |
2594 | mutex_unlock(&xps_map_mutex); | |
4d99f660 | 2595 | cpus_read_unlock(); |
537c00de AD |
2596 | } |
2597 | ||
6234f874 AD |
2598 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2599 | { | |
2600 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2601 | } | |
2602 | ||
80d19669 AN |
2603 | static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, |
2604 | u16 index, bool is_rxqs_map) | |
01c5f864 AD |
2605 | { |
2606 | struct xps_map *new_map; | |
2607 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2608 | int i, pos; | |
2609 | ||
2610 | for (pos = 0; map && pos < map->len; pos++) { | |
2611 | if (map->queues[pos] != index) | |
2612 | continue; | |
2613 | return map; | |
2614 | } | |
2615 | ||
80d19669 | 2616 | /* Need to add tx-queue to this CPU's/rx-queue's existing map */ |
01c5f864 AD |
2617 | if (map) { |
2618 | if (pos < map->alloc_len) | |
2619 | return map; | |
2620 | ||
2621 | alloc_len = map->alloc_len * 2; | |
2622 | } | |
2623 | ||
80d19669 AN |
2624 | /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's |
2625 | * map | |
2626 | */ | |
2627 | if (is_rxqs_map) | |
2628 | new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); | |
2629 | else | |
2630 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2631 | cpu_to_node(attr_index)); | |
01c5f864 AD |
2632 | if (!new_map) |
2633 | return NULL; | |
2634 | ||
2635 | for (i = 0; i < pos; i++) | |
2636 | new_map->queues[i] = map->queues[i]; | |
2637 | new_map->alloc_len = alloc_len; | |
2638 | new_map->len = pos; | |
2639 | ||
2640 | return new_map; | |
2641 | } | |
2642 | ||
4d99f660 | 2643 | /* Must be called under cpus_read_lock */ |
80d19669 AN |
2644 | int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, |
2645 | u16 index, bool is_rxqs_map) | |
537c00de | 2646 | { |
80d19669 | 2647 | const unsigned long *online_mask = NULL, *possible_mask = NULL; |
01c5f864 | 2648 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
80d19669 | 2649 | int i, j, tci, numa_node_id = -2; |
184c449f | 2650 | int maps_sz, num_tc = 1, tc = 0; |
537c00de | 2651 | struct xps_map *map, *new_map; |
01c5f864 | 2652 | bool active = false; |
80d19669 | 2653 | unsigned int nr_ids; |
537c00de | 2654 | |
184c449f | 2655 | if (dev->num_tc) { |
ffcfe25b | 2656 | /* Do not allow XPS on subordinate device directly */ |
184c449f | 2657 | num_tc = dev->num_tc; |
ffcfe25b AD |
2658 | if (num_tc < 0) |
2659 | return -EINVAL; | |
2660 | ||
2661 | /* If queue belongs to subordinate dev use its map */ | |
2662 | dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; | |
2663 | ||
184c449f AD |
2664 | tc = netdev_txq_to_tc(dev, index); |
2665 | if (tc < 0) | |
2666 | return -EINVAL; | |
2667 | } | |
2668 | ||
537c00de | 2669 | mutex_lock(&xps_map_mutex); |
80d19669 AN |
2670 | if (is_rxqs_map) { |
2671 | maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); | |
2672 | dev_maps = xmap_dereference(dev->xps_rxqs_map); | |
2673 | nr_ids = dev->num_rx_queues; | |
2674 | } else { | |
2675 | maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); | |
2676 | if (num_possible_cpus() > 1) { | |
2677 | online_mask = cpumask_bits(cpu_online_mask); | |
2678 | possible_mask = cpumask_bits(cpu_possible_mask); | |
2679 | } | |
2680 | dev_maps = xmap_dereference(dev->xps_cpus_map); | |
2681 | nr_ids = nr_cpu_ids; | |
2682 | } | |
537c00de | 2683 | |
80d19669 AN |
2684 | if (maps_sz < L1_CACHE_BYTES) |
2685 | maps_sz = L1_CACHE_BYTES; | |
537c00de | 2686 | |
01c5f864 | 2687 | /* allocate memory for queue storage */ |
80d19669 AN |
2688 | for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), |
2689 | j < nr_ids;) { | |
01c5f864 AD |
2690 | if (!new_dev_maps) |
2691 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2692 | if (!new_dev_maps) { |
2693 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2694 | return -ENOMEM; |
2bb60cb9 | 2695 | } |
01c5f864 | 2696 | |
80d19669 AN |
2697 | tci = j * num_tc + tc; |
2698 | map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) : | |
01c5f864 AD |
2699 | NULL; |
2700 | ||
80d19669 | 2701 | map = expand_xps_map(map, j, index, is_rxqs_map); |
01c5f864 AD |
2702 | if (!map) |
2703 | goto error; | |
2704 | ||
80d19669 | 2705 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); |
01c5f864 AD |
2706 | } |
2707 | ||
2708 | if (!new_dev_maps) | |
2709 | goto out_no_new_maps; | |
2710 | ||
867d0ad4 SD |
2711 | if (!dev_maps) { |
2712 | /* Increment static keys at most once per type */ | |
2713 | static_key_slow_inc_cpuslocked(&xps_needed); | |
2714 | if (is_rxqs_map) | |
2715 | static_key_slow_inc_cpuslocked(&xps_rxqs_needed); | |
2716 | } | |
04157469 | 2717 | |
80d19669 AN |
2718 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2719 | j < nr_ids;) { | |
184c449f | 2720 | /* copy maps belonging to foreign traffic classes */ |
80d19669 | 2721 | for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) { |
184c449f | 2722 | /* fill in the new device map from the old device map */ |
80d19669 AN |
2723 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2724 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
184c449f AD |
2725 | } |
2726 | ||
2727 | /* We need to explicitly update tci as prevous loop | |
2728 | * could break out early if dev_maps is NULL. | |
2729 | */ | |
80d19669 | 2730 | tci = j * num_tc + tc; |
184c449f | 2731 | |
80d19669 AN |
2732 | if (netif_attr_test_mask(j, mask, nr_ids) && |
2733 | netif_attr_test_online(j, online_mask, nr_ids)) { | |
2734 | /* add tx-queue to CPU/rx-queue maps */ | |
01c5f864 AD |
2735 | int pos = 0; |
2736 | ||
80d19669 | 2737 | map = xmap_dereference(new_dev_maps->attr_map[tci]); |
01c5f864 AD |
2738 | while ((pos < map->len) && (map->queues[pos] != index)) |
2739 | pos++; | |
2740 | ||
2741 | if (pos == map->len) | |
2742 | map->queues[map->len++] = index; | |
537c00de | 2743 | #ifdef CONFIG_NUMA |
80d19669 AN |
2744 | if (!is_rxqs_map) { |
2745 | if (numa_node_id == -2) | |
2746 | numa_node_id = cpu_to_node(j); | |
2747 | else if (numa_node_id != cpu_to_node(j)) | |
2748 | numa_node_id = -1; | |
2749 | } | |
537c00de | 2750 | #endif |
01c5f864 AD |
2751 | } else if (dev_maps) { |
2752 | /* fill in the new device map from the old device map */ | |
80d19669 AN |
2753 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2754 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
537c00de | 2755 | } |
01c5f864 | 2756 | |
184c449f AD |
2757 | /* copy maps belonging to foreign traffic classes */ |
2758 | for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { | |
2759 | /* fill in the new device map from the old device map */ | |
80d19669 AN |
2760 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2761 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
184c449f | 2762 | } |
537c00de AD |
2763 | } |
2764 | ||
80d19669 AN |
2765 | if (is_rxqs_map) |
2766 | rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps); | |
2767 | else | |
2768 | rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps); | |
01c5f864 | 2769 | |
537c00de | 2770 | /* Cleanup old maps */ |
184c449f AD |
2771 | if (!dev_maps) |
2772 | goto out_no_old_maps; | |
2773 | ||
80d19669 AN |
2774 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2775 | j < nr_ids;) { | |
2776 | for (i = num_tc, tci = j * num_tc; i--; tci++) { | |
2777 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
2778 | map = xmap_dereference(dev_maps->attr_map[tci]); | |
01c5f864 AD |
2779 | if (map && map != new_map) |
2780 | kfree_rcu(map, rcu); | |
2781 | } | |
537c00de AD |
2782 | } |
2783 | ||
184c449f AD |
2784 | kfree_rcu(dev_maps, rcu); |
2785 | ||
2786 | out_no_old_maps: | |
01c5f864 AD |
2787 | dev_maps = new_dev_maps; |
2788 | active = true; | |
537c00de | 2789 | |
01c5f864 | 2790 | out_no_new_maps: |
80d19669 AN |
2791 | if (!is_rxqs_map) { |
2792 | /* update Tx queue numa node */ | |
2793 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), | |
2794 | (numa_node_id >= 0) ? | |
2795 | numa_node_id : NUMA_NO_NODE); | |
2796 | } | |
537c00de | 2797 | |
01c5f864 AD |
2798 | if (!dev_maps) |
2799 | goto out_no_maps; | |
2800 | ||
80d19669 AN |
2801 | /* removes tx-queue from unused CPUs/rx-queues */ |
2802 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), | |
2803 | j < nr_ids;) { | |
2804 | for (i = tc, tci = j * num_tc; i--; tci++) | |
184c449f | 2805 | active |= remove_xps_queue(dev_maps, tci, index); |
80d19669 AN |
2806 | if (!netif_attr_test_mask(j, mask, nr_ids) || |
2807 | !netif_attr_test_online(j, online_mask, nr_ids)) | |
184c449f AD |
2808 | active |= remove_xps_queue(dev_maps, tci, index); |
2809 | for (i = num_tc - tc, tci++; --i; tci++) | |
2810 | active |= remove_xps_queue(dev_maps, tci, index); | |
01c5f864 AD |
2811 | } |
2812 | ||
2813 | /* free map if not active */ | |
867d0ad4 SD |
2814 | if (!active) |
2815 | reset_xps_maps(dev, dev_maps, is_rxqs_map); | |
01c5f864 AD |
2816 | |
2817 | out_no_maps: | |
537c00de AD |
2818 | mutex_unlock(&xps_map_mutex); |
2819 | ||
2820 | return 0; | |
2821 | error: | |
01c5f864 | 2822 | /* remove any maps that we added */ |
80d19669 AN |
2823 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2824 | j < nr_ids;) { | |
2825 | for (i = num_tc, tci = j * num_tc; i--; tci++) { | |
2826 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
184c449f | 2827 | map = dev_maps ? |
80d19669 | 2828 | xmap_dereference(dev_maps->attr_map[tci]) : |
184c449f AD |
2829 | NULL; |
2830 | if (new_map && new_map != map) | |
2831 | kfree(new_map); | |
2832 | } | |
01c5f864 AD |
2833 | } |
2834 | ||
537c00de AD |
2835 | mutex_unlock(&xps_map_mutex); |
2836 | ||
537c00de AD |
2837 | kfree(new_dev_maps); |
2838 | return -ENOMEM; | |
2839 | } | |
4d99f660 | 2840 | EXPORT_SYMBOL_GPL(__netif_set_xps_queue); |
80d19669 AN |
2841 | |
2842 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, | |
2843 | u16 index) | |
2844 | { | |
4d99f660 AV |
2845 | int ret; |
2846 | ||
2847 | cpus_read_lock(); | |
2848 | ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false); | |
2849 | cpus_read_unlock(); | |
2850 | ||
2851 | return ret; | |
80d19669 | 2852 | } |
537c00de AD |
2853 | EXPORT_SYMBOL(netif_set_xps_queue); |
2854 | ||
2855 | #endif | |
ffcfe25b AD |
2856 | static void netdev_unbind_all_sb_channels(struct net_device *dev) |
2857 | { | |
2858 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2859 | ||
2860 | /* Unbind any subordinate channels */ | |
2861 | while (txq-- != &dev->_tx[0]) { | |
2862 | if (txq->sb_dev) | |
2863 | netdev_unbind_sb_channel(dev, txq->sb_dev); | |
2864 | } | |
2865 | } | |
2866 | ||
9cf1f6a8 AD |
2867 | void netdev_reset_tc(struct net_device *dev) |
2868 | { | |
6234f874 AD |
2869 | #ifdef CONFIG_XPS |
2870 | netif_reset_xps_queues_gt(dev, 0); | |
2871 | #endif | |
ffcfe25b AD |
2872 | netdev_unbind_all_sb_channels(dev); |
2873 | ||
2874 | /* Reset TC configuration of device */ | |
9cf1f6a8 AD |
2875 | dev->num_tc = 0; |
2876 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2877 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2878 | } | |
2879 | EXPORT_SYMBOL(netdev_reset_tc); | |
2880 | ||
2881 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2882 | { | |
2883 | if (tc >= dev->num_tc) | |
2884 | return -EINVAL; | |
2885 | ||
6234f874 AD |
2886 | #ifdef CONFIG_XPS |
2887 | netif_reset_xps_queues(dev, offset, count); | |
2888 | #endif | |
9cf1f6a8 AD |
2889 | dev->tc_to_txq[tc].count = count; |
2890 | dev->tc_to_txq[tc].offset = offset; | |
2891 | return 0; | |
2892 | } | |
2893 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2894 | ||
2895 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2896 | { | |
2897 | if (num_tc > TC_MAX_QUEUE) | |
2898 | return -EINVAL; | |
2899 | ||
6234f874 AD |
2900 | #ifdef CONFIG_XPS |
2901 | netif_reset_xps_queues_gt(dev, 0); | |
2902 | #endif | |
ffcfe25b AD |
2903 | netdev_unbind_all_sb_channels(dev); |
2904 | ||
9cf1f6a8 AD |
2905 | dev->num_tc = num_tc; |
2906 | return 0; | |
2907 | } | |
2908 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2909 | ||
ffcfe25b AD |
2910 | void netdev_unbind_sb_channel(struct net_device *dev, |
2911 | struct net_device *sb_dev) | |
2912 | { | |
2913 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2914 | ||
2915 | #ifdef CONFIG_XPS | |
2916 | netif_reset_xps_queues_gt(sb_dev, 0); | |
2917 | #endif | |
2918 | memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); | |
2919 | memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); | |
2920 | ||
2921 | while (txq-- != &dev->_tx[0]) { | |
2922 | if (txq->sb_dev == sb_dev) | |
2923 | txq->sb_dev = NULL; | |
2924 | } | |
2925 | } | |
2926 | EXPORT_SYMBOL(netdev_unbind_sb_channel); | |
2927 | ||
2928 | int netdev_bind_sb_channel_queue(struct net_device *dev, | |
2929 | struct net_device *sb_dev, | |
2930 | u8 tc, u16 count, u16 offset) | |
2931 | { | |
2932 | /* Make certain the sb_dev and dev are already configured */ | |
2933 | if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) | |
2934 | return -EINVAL; | |
2935 | ||
2936 | /* We cannot hand out queues we don't have */ | |
2937 | if ((offset + count) > dev->real_num_tx_queues) | |
2938 | return -EINVAL; | |
2939 | ||
2940 | /* Record the mapping */ | |
2941 | sb_dev->tc_to_txq[tc].count = count; | |
2942 | sb_dev->tc_to_txq[tc].offset = offset; | |
2943 | ||
2944 | /* Provide a way for Tx queue to find the tc_to_txq map or | |
2945 | * XPS map for itself. | |
2946 | */ | |
2947 | while (count--) | |
2948 | netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; | |
2949 | ||
2950 | return 0; | |
2951 | } | |
2952 | EXPORT_SYMBOL(netdev_bind_sb_channel_queue); | |
2953 | ||
2954 | int netdev_set_sb_channel(struct net_device *dev, u16 channel) | |
2955 | { | |
2956 | /* Do not use a multiqueue device to represent a subordinate channel */ | |
2957 | if (netif_is_multiqueue(dev)) | |
2958 | return -ENODEV; | |
2959 | ||
2960 | /* We allow channels 1 - 32767 to be used for subordinate channels. | |
2961 | * Channel 0 is meant to be "native" mode and used only to represent | |
2962 | * the main root device. We allow writing 0 to reset the device back | |
2963 | * to normal mode after being used as a subordinate channel. | |
2964 | */ | |
2965 | if (channel > S16_MAX) | |
2966 | return -EINVAL; | |
2967 | ||
2968 | dev->num_tc = -channel; | |
2969 | ||
2970 | return 0; | |
2971 | } | |
2972 | EXPORT_SYMBOL(netdev_set_sb_channel); | |
2973 | ||
f0796d5c JF |
2974 | /* |
2975 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
3a053b1a | 2976 | * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. |
f0796d5c | 2977 | */ |
e6484930 | 2978 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2979 | { |
ac5b7019 | 2980 | bool disabling; |
1d24eb48 TH |
2981 | int rc; |
2982 | ||
ac5b7019 JK |
2983 | disabling = txq < dev->real_num_tx_queues; |
2984 | ||
e6484930 TH |
2985 | if (txq < 1 || txq > dev->num_tx_queues) |
2986 | return -EINVAL; | |
f0796d5c | 2987 | |
5c56580b BH |
2988 | if (dev->reg_state == NETREG_REGISTERED || |
2989 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2990 | ASSERT_RTNL(); |
2991 | ||
1d24eb48 TH |
2992 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2993 | txq); | |
bf264145 TH |
2994 | if (rc) |
2995 | return rc; | |
2996 | ||
4f57c087 JF |
2997 | if (dev->num_tc) |
2998 | netif_setup_tc(dev, txq); | |
2999 | ||
ac5b7019 JK |
3000 | dev->real_num_tx_queues = txq; |
3001 | ||
3002 | if (disabling) { | |
3003 | synchronize_net(); | |
e6484930 | 3004 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
3005 | #ifdef CONFIG_XPS |
3006 | netif_reset_xps_queues_gt(dev, txq); | |
3007 | #endif | |
3008 | } | |
ac5b7019 JK |
3009 | } else { |
3010 | dev->real_num_tx_queues = txq; | |
f0796d5c | 3011 | } |
e6484930 | 3012 | |
e6484930 | 3013 | return 0; |
f0796d5c JF |
3014 | } |
3015 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 3016 | |
a953be53 | 3017 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
3018 | /** |
3019 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
3020 | * @dev: Network device | |
3021 | * @rxq: Actual number of RX queues | |
3022 | * | |
3023 | * This must be called either with the rtnl_lock held or before | |
3024 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
3025 | * negative error code. If called before registration, it always |
3026 | * succeeds. | |
62fe0b40 BH |
3027 | */ |
3028 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
3029 | { | |
3030 | int rc; | |
3031 | ||
bd25fa7b TH |
3032 | if (rxq < 1 || rxq > dev->num_rx_queues) |
3033 | return -EINVAL; | |
3034 | ||
62fe0b40 BH |
3035 | if (dev->reg_state == NETREG_REGISTERED) { |
3036 | ASSERT_RTNL(); | |
3037 | ||
62fe0b40 BH |
3038 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
3039 | rxq); | |
3040 | if (rc) | |
3041 | return rc; | |
62fe0b40 BH |
3042 | } |
3043 | ||
3044 | dev->real_num_rx_queues = rxq; | |
3045 | return 0; | |
3046 | } | |
3047 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
3048 | #endif | |
3049 | ||
2c53040f BH |
3050 | /** |
3051 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
3052 | * |
3053 | * This routine should set an upper limit on the number of RSS queues | |
3054 | * used by default by multiqueue devices. | |
3055 | */ | |
a55b138b | 3056 | int netif_get_num_default_rss_queues(void) |
16917b87 | 3057 | { |
40e4e713 HS |
3058 | return is_kdump_kernel() ? |
3059 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
3060 | } |
3061 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
3062 | ||
3bcb846c | 3063 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 3064 | { |
def82a1d JP |
3065 | struct softnet_data *sd; |
3066 | unsigned long flags; | |
56079431 | 3067 | |
def82a1d | 3068 | local_irq_save(flags); |
903ceff7 | 3069 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
3070 | q->next_sched = NULL; |
3071 | *sd->output_queue_tailp = q; | |
3072 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
3073 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
3074 | local_irq_restore(flags); | |
3075 | } | |
3076 | ||
3077 | void __netif_schedule(struct Qdisc *q) | |
3078 | { | |
3079 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
3080 | __netif_reschedule(q); | |
56079431 DV |
3081 | } |
3082 | EXPORT_SYMBOL(__netif_schedule); | |
3083 | ||
e6247027 ED |
3084 | struct dev_kfree_skb_cb { |
3085 | enum skb_free_reason reason; | |
3086 | }; | |
3087 | ||
3088 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 3089 | { |
e6247027 ED |
3090 | return (struct dev_kfree_skb_cb *)skb->cb; |
3091 | } | |
3092 | ||
46e5da40 JF |
3093 | void netif_schedule_queue(struct netdev_queue *txq) |
3094 | { | |
3095 | rcu_read_lock(); | |
5be5515a | 3096 | if (!netif_xmit_stopped(txq)) { |
46e5da40 JF |
3097 | struct Qdisc *q = rcu_dereference(txq->qdisc); |
3098 | ||
3099 | __netif_schedule(q); | |
3100 | } | |
3101 | rcu_read_unlock(); | |
3102 | } | |
3103 | EXPORT_SYMBOL(netif_schedule_queue); | |
3104 | ||
46e5da40 JF |
3105 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
3106 | { | |
3107 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
3108 | struct Qdisc *q; | |
3109 | ||
3110 | rcu_read_lock(); | |
3111 | q = rcu_dereference(dev_queue->qdisc); | |
3112 | __netif_schedule(q); | |
3113 | rcu_read_unlock(); | |
3114 | } | |
3115 | } | |
3116 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
3117 | ||
e6247027 | 3118 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 3119 | { |
e6247027 | 3120 | unsigned long flags; |
56079431 | 3121 | |
9899886d MJ |
3122 | if (unlikely(!skb)) |
3123 | return; | |
3124 | ||
63354797 | 3125 | if (likely(refcount_read(&skb->users) == 1)) { |
e6247027 | 3126 | smp_rmb(); |
63354797 RE |
3127 | refcount_set(&skb->users, 0); |
3128 | } else if (likely(!refcount_dec_and_test(&skb->users))) { | |
e6247027 | 3129 | return; |
bea3348e | 3130 | } |
e6247027 ED |
3131 | get_kfree_skb_cb(skb)->reason = reason; |
3132 | local_irq_save(flags); | |
3133 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
3134 | __this_cpu_write(softnet_data.completion_queue, skb); | |
3135 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
3136 | local_irq_restore(flags); | |
56079431 | 3137 | } |
e6247027 | 3138 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 3139 | |
e6247027 | 3140 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
3141 | { |
3142 | if (in_irq() || irqs_disabled()) | |
e6247027 | 3143 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
3144 | else |
3145 | dev_kfree_skb(skb); | |
3146 | } | |
e6247027 | 3147 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
3148 | |
3149 | ||
bea3348e SH |
3150 | /** |
3151 | * netif_device_detach - mark device as removed | |
3152 | * @dev: network device | |
3153 | * | |
3154 | * Mark device as removed from system and therefore no longer available. | |
3155 | */ | |
56079431 DV |
3156 | void netif_device_detach(struct net_device *dev) |
3157 | { | |
3158 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
3159 | netif_running(dev)) { | |
d543103a | 3160 | netif_tx_stop_all_queues(dev); |
56079431 DV |
3161 | } |
3162 | } | |
3163 | EXPORT_SYMBOL(netif_device_detach); | |
3164 | ||
bea3348e SH |
3165 | /** |
3166 | * netif_device_attach - mark device as attached | |
3167 | * @dev: network device | |
3168 | * | |
3169 | * Mark device as attached from system and restart if needed. | |
3170 | */ | |
56079431 DV |
3171 | void netif_device_attach(struct net_device *dev) |
3172 | { | |
3173 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
3174 | netif_running(dev)) { | |
d543103a | 3175 | netif_tx_wake_all_queues(dev); |
4ec93edb | 3176 | __netdev_watchdog_up(dev); |
56079431 DV |
3177 | } |
3178 | } | |
3179 | EXPORT_SYMBOL(netif_device_attach); | |
3180 | ||
5605c762 JP |
3181 | /* |
3182 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
3183 | * to be used as a distribution range. | |
3184 | */ | |
eadec877 AD |
3185 | static u16 skb_tx_hash(const struct net_device *dev, |
3186 | const struct net_device *sb_dev, | |
3187 | struct sk_buff *skb) | |
5605c762 JP |
3188 | { |
3189 | u32 hash; | |
3190 | u16 qoffset = 0; | |
1b837d48 | 3191 | u16 qcount = dev->real_num_tx_queues; |
5605c762 | 3192 | |
eadec877 AD |
3193 | if (dev->num_tc) { |
3194 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
3195 | ||
3196 | qoffset = sb_dev->tc_to_txq[tc].offset; | |
3197 | qcount = sb_dev->tc_to_txq[tc].count; | |
3198 | } | |
3199 | ||
5605c762 JP |
3200 | if (skb_rx_queue_recorded(skb)) { |
3201 | hash = skb_get_rx_queue(skb); | |
6e11d157 AN |
3202 | if (hash >= qoffset) |
3203 | hash -= qoffset; | |
1b837d48 AD |
3204 | while (unlikely(hash >= qcount)) |
3205 | hash -= qcount; | |
eadec877 | 3206 | return hash + qoffset; |
5605c762 JP |
3207 | } |
3208 | ||
3209 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
3210 | } | |
5605c762 | 3211 | |
36c92474 BH |
3212 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
3213 | { | |
84d15ae5 | 3214 | static const netdev_features_t null_features; |
36c92474 | 3215 | struct net_device *dev = skb->dev; |
88ad4175 | 3216 | const char *name = ""; |
36c92474 | 3217 | |
c846ad9b BG |
3218 | if (!net_ratelimit()) |
3219 | return; | |
3220 | ||
88ad4175 BM |
3221 | if (dev) { |
3222 | if (dev->dev.parent) | |
3223 | name = dev_driver_string(dev->dev.parent); | |
3224 | else | |
3225 | name = netdev_name(dev); | |
3226 | } | |
6413139d WB |
3227 | skb_dump(KERN_WARNING, skb, false); |
3228 | WARN(1, "%s: caps=(%pNF, %pNF)\n", | |
88ad4175 | 3229 | name, dev ? &dev->features : &null_features, |
6413139d | 3230 | skb->sk ? &skb->sk->sk_route_caps : &null_features); |
36c92474 BH |
3231 | } |
3232 | ||
1da177e4 LT |
3233 | /* |
3234 | * Invalidate hardware checksum when packet is to be mangled, and | |
3235 | * complete checksum manually on outgoing path. | |
3236 | */ | |
84fa7933 | 3237 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 3238 | { |
d3bc23e7 | 3239 | __wsum csum; |
663ead3b | 3240 | int ret = 0, offset; |
1da177e4 | 3241 | |
84fa7933 | 3242 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
3243 | goto out_set_summed; |
3244 | ||
3aefd7d6 | 3245 | if (unlikely(skb_is_gso(skb))) { |
36c92474 BH |
3246 | skb_warn_bad_offload(skb); |
3247 | return -EINVAL; | |
1da177e4 LT |
3248 | } |
3249 | ||
cef401de ED |
3250 | /* Before computing a checksum, we should make sure no frag could |
3251 | * be modified by an external entity : checksum could be wrong. | |
3252 | */ | |
3253 | if (skb_has_shared_frag(skb)) { | |
3254 | ret = __skb_linearize(skb); | |
3255 | if (ret) | |
3256 | goto out; | |
3257 | } | |
3258 | ||
55508d60 | 3259 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
3260 | BUG_ON(offset >= skb_headlen(skb)); |
3261 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
3262 | ||
3263 | offset += skb->csum_offset; | |
3264 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
3265 | ||
8211fbfa HK |
3266 | ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); |
3267 | if (ret) | |
3268 | goto out; | |
1da177e4 | 3269 | |
4f2e4ad5 | 3270 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 3271 | out_set_summed: |
1da177e4 | 3272 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 3273 | out: |
1da177e4 LT |
3274 | return ret; |
3275 | } | |
d1b19dff | 3276 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 3277 | |
b72b5bf6 DC |
3278 | int skb_crc32c_csum_help(struct sk_buff *skb) |
3279 | { | |
3280 | __le32 crc32c_csum; | |
3281 | int ret = 0, offset, start; | |
3282 | ||
3283 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
3284 | goto out; | |
3285 | ||
3286 | if (unlikely(skb_is_gso(skb))) | |
3287 | goto out; | |
3288 | ||
3289 | /* Before computing a checksum, we should make sure no frag could | |
3290 | * be modified by an external entity : checksum could be wrong. | |
3291 | */ | |
3292 | if (unlikely(skb_has_shared_frag(skb))) { | |
3293 | ret = __skb_linearize(skb); | |
3294 | if (ret) | |
3295 | goto out; | |
3296 | } | |
3297 | start = skb_checksum_start_offset(skb); | |
3298 | offset = start + offsetof(struct sctphdr, checksum); | |
3299 | if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { | |
3300 | ret = -EINVAL; | |
3301 | goto out; | |
3302 | } | |
8211fbfa HK |
3303 | |
3304 | ret = skb_ensure_writable(skb, offset + sizeof(__le32)); | |
3305 | if (ret) | |
3306 | goto out; | |
3307 | ||
b72b5bf6 DC |
3308 | crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, |
3309 | skb->len - start, ~(__u32)0, | |
3310 | crc32c_csum_stub)); | |
3311 | *(__le32 *)(skb->data + offset) = crc32c_csum; | |
3312 | skb->ip_summed = CHECKSUM_NONE; | |
dba00306 | 3313 | skb->csum_not_inet = 0; |
b72b5bf6 DC |
3314 | out: |
3315 | return ret; | |
3316 | } | |
3317 | ||
53d6471c | 3318 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 3319 | { |
252e3346 | 3320 | __be16 type = skb->protocol; |
f6a78bfc | 3321 | |
19acc327 PS |
3322 | /* Tunnel gso handlers can set protocol to ethernet. */ |
3323 | if (type == htons(ETH_P_TEB)) { | |
3324 | struct ethhdr *eth; | |
3325 | ||
3326 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
3327 | return 0; | |
3328 | ||
1dfe82eb | 3329 | eth = (struct ethhdr *)skb->data; |
19acc327 PS |
3330 | type = eth->h_proto; |
3331 | } | |
3332 | ||
d4bcef3f | 3333 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
3334 | } |
3335 | ||
3336 | /** | |
3337 | * skb_mac_gso_segment - mac layer segmentation handler. | |
3338 | * @skb: buffer to segment | |
3339 | * @features: features for the output path (see dev->features) | |
3340 | */ | |
3341 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
3342 | netdev_features_t features) | |
3343 | { | |
3344 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
3345 | struct packet_offload *ptype; | |
53d6471c VY |
3346 | int vlan_depth = skb->mac_len; |
3347 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
3348 | |
3349 | if (unlikely(!type)) | |
3350 | return ERR_PTR(-EINVAL); | |
3351 | ||
53d6471c | 3352 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
3353 | |
3354 | rcu_read_lock(); | |
22061d80 | 3355 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 3356 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 3357 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
3358 | break; |
3359 | } | |
3360 | } | |
3361 | rcu_read_unlock(); | |
3362 | ||
98e399f8 | 3363 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 3364 | |
f6a78bfc HX |
3365 | return segs; |
3366 | } | |
05e8ef4a PS |
3367 | EXPORT_SYMBOL(skb_mac_gso_segment); |
3368 | ||
3369 | ||
3370 | /* openvswitch calls this on rx path, so we need a different check. | |
3371 | */ | |
3372 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
3373 | { | |
3374 | if (tx_path) | |
0c19f846 WB |
3375 | return skb->ip_summed != CHECKSUM_PARTIAL && |
3376 | skb->ip_summed != CHECKSUM_UNNECESSARY; | |
6e7bc478 ED |
3377 | |
3378 | return skb->ip_summed == CHECKSUM_NONE; | |
05e8ef4a PS |
3379 | } |
3380 | ||
3381 | /** | |
3382 | * __skb_gso_segment - Perform segmentation on skb. | |
3383 | * @skb: buffer to segment | |
3384 | * @features: features for the output path (see dev->features) | |
3385 | * @tx_path: whether it is called in TX path | |
3386 | * | |
3387 | * This function segments the given skb and returns a list of segments. | |
3388 | * | |
3389 | * It may return NULL if the skb requires no segmentation. This is | |
3390 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 | 3391 | * |
a08e7fd9 | 3392 | * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. |
05e8ef4a PS |
3393 | */ |
3394 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
3395 | netdev_features_t features, bool tx_path) | |
3396 | { | |
b2504a5d ED |
3397 | struct sk_buff *segs; |
3398 | ||
05e8ef4a PS |
3399 | if (unlikely(skb_needs_check(skb, tx_path))) { |
3400 | int err; | |
3401 | ||
b2504a5d | 3402 | /* We're going to init ->check field in TCP or UDP header */ |
a40e0a66 | 3403 | err = skb_cow_head(skb, 0); |
3404 | if (err < 0) | |
05e8ef4a PS |
3405 | return ERR_PTR(err); |
3406 | } | |
3407 | ||
802ab55a AD |
3408 | /* Only report GSO partial support if it will enable us to |
3409 | * support segmentation on this frame without needing additional | |
3410 | * work. | |
3411 | */ | |
3412 | if (features & NETIF_F_GSO_PARTIAL) { | |
3413 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
3414 | struct net_device *dev = skb->dev; | |
3415 | ||
3416 | partial_features |= dev->features & dev->gso_partial_features; | |
3417 | if (!skb_gso_ok(skb, features | partial_features)) | |
3418 | features &= ~NETIF_F_GSO_PARTIAL; | |
3419 | } | |
3420 | ||
a08e7fd9 | 3421 | BUILD_BUG_ON(SKB_GSO_CB_OFFSET + |
9207f9d4 KK |
3422 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); |
3423 | ||
68c33163 | 3424 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
3425 | SKB_GSO_CB(skb)->encap_level = 0; |
3426 | ||
05e8ef4a PS |
3427 | skb_reset_mac_header(skb); |
3428 | skb_reset_mac_len(skb); | |
3429 | ||
b2504a5d ED |
3430 | segs = skb_mac_gso_segment(skb, features); |
3431 | ||
3a1296a3 | 3432 | if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) |
b2504a5d ED |
3433 | skb_warn_bad_offload(skb); |
3434 | ||
3435 | return segs; | |
05e8ef4a | 3436 | } |
12b0004d | 3437 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 3438 | |
fb286bb2 HX |
3439 | /* Take action when hardware reception checksum errors are detected. */ |
3440 | #ifdef CONFIG_BUG | |
7fe50ac8 | 3441 | void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) |
fb286bb2 HX |
3442 | { |
3443 | if (net_ratelimit()) { | |
7b6cd1ce | 3444 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
6413139d | 3445 | skb_dump(KERN_ERR, skb, true); |
fb286bb2 HX |
3446 | dump_stack(); |
3447 | } | |
3448 | } | |
3449 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
3450 | #endif | |
3451 | ||
ab74cfeb | 3452 | /* XXX: check that highmem exists at all on the given machine. */ |
c1e756bf | 3453 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 3454 | { |
3d3a8533 | 3455 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 3456 | int i; |
f4563a75 | 3457 | |
5acbbd42 | 3458 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
3459 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
3460 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 3461 | |
ea2ab693 | 3462 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 3463 | return 1; |
ea2ab693 | 3464 | } |
5acbbd42 | 3465 | } |
3d3a8533 | 3466 | #endif |
1da177e4 LT |
3467 | return 0; |
3468 | } | |
1da177e4 | 3469 | |
3b392ddb SH |
3470 | /* If MPLS offload request, verify we are testing hardware MPLS features |
3471 | * instead of standard features for the netdev. | |
3472 | */ | |
d0edc7bf | 3473 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
3474 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
3475 | netdev_features_t features, | |
3476 | __be16 type) | |
3477 | { | |
25cd9ba0 | 3478 | if (eth_p_mpls(type)) |
3b392ddb SH |
3479 | features &= skb->dev->mpls_features; |
3480 | ||
3481 | return features; | |
3482 | } | |
3483 | #else | |
3484 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
3485 | netdev_features_t features, | |
3486 | __be16 type) | |
3487 | { | |
3488 | return features; | |
3489 | } | |
3490 | #endif | |
3491 | ||
c8f44aff | 3492 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 3493 | netdev_features_t features) |
f01a5236 | 3494 | { |
3b392ddb SH |
3495 | __be16 type; |
3496 | ||
9fc95f50 | 3497 | type = skb_network_protocol(skb, NULL); |
3b392ddb | 3498 | features = net_mpls_features(skb, features, type); |
53d6471c | 3499 | |
c0d680e5 | 3500 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 3501 | !can_checksum_protocol(features, type)) { |
996e8021 | 3502 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 3503 | } |
7be2c82c ED |
3504 | if (illegal_highdma(skb->dev, skb)) |
3505 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
3506 | |
3507 | return features; | |
3508 | } | |
3509 | ||
e38f3025 TM |
3510 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
3511 | struct net_device *dev, | |
3512 | netdev_features_t features) | |
3513 | { | |
3514 | return features; | |
3515 | } | |
3516 | EXPORT_SYMBOL(passthru_features_check); | |
3517 | ||
7ce23672 | 3518 | static netdev_features_t dflt_features_check(struct sk_buff *skb, |
8cb65d00 TM |
3519 | struct net_device *dev, |
3520 | netdev_features_t features) | |
3521 | { | |
3522 | return vlan_features_check(skb, features); | |
3523 | } | |
3524 | ||
cbc53e08 AD |
3525 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
3526 | struct net_device *dev, | |
3527 | netdev_features_t features) | |
3528 | { | |
3529 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
3530 | ||
3531 | if (gso_segs > dev->gso_max_segs) | |
3532 | return features & ~NETIF_F_GSO_MASK; | |
3533 | ||
1d155dfd HK |
3534 | if (!skb_shinfo(skb)->gso_type) { |
3535 | skb_warn_bad_offload(skb); | |
3536 | return features & ~NETIF_F_GSO_MASK; | |
3537 | } | |
3538 | ||
802ab55a AD |
3539 | /* Support for GSO partial features requires software |
3540 | * intervention before we can actually process the packets | |
3541 | * so we need to strip support for any partial features now | |
3542 | * and we can pull them back in after we have partially | |
3543 | * segmented the frame. | |
3544 | */ | |
3545 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
3546 | features &= ~dev->gso_partial_features; | |
3547 | ||
3548 | /* Make sure to clear the IPv4 ID mangling feature if the | |
3549 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
3550 | */ |
3551 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
3552 | struct iphdr *iph = skb->encapsulation ? | |
3553 | inner_ip_hdr(skb) : ip_hdr(skb); | |
3554 | ||
3555 | if (!(iph->frag_off & htons(IP_DF))) | |
3556 | features &= ~NETIF_F_TSO_MANGLEID; | |
3557 | } | |
3558 | ||
3559 | return features; | |
3560 | } | |
3561 | ||
c1e756bf | 3562 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 3563 | { |
5f35227e | 3564 | struct net_device *dev = skb->dev; |
fcbeb976 | 3565 | netdev_features_t features = dev->features; |
58e998c6 | 3566 | |
cbc53e08 AD |
3567 | if (skb_is_gso(skb)) |
3568 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 3569 | |
5f35227e JG |
3570 | /* If encapsulation offload request, verify we are testing |
3571 | * hardware encapsulation features instead of standard | |
3572 | * features for the netdev | |
3573 | */ | |
3574 | if (skb->encapsulation) | |
3575 | features &= dev->hw_enc_features; | |
3576 | ||
f5a7fb88 TM |
3577 | if (skb_vlan_tagged(skb)) |
3578 | features = netdev_intersect_features(features, | |
3579 | dev->vlan_features | | |
3580 | NETIF_F_HW_VLAN_CTAG_TX | | |
3581 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 3582 | |
5f35227e JG |
3583 | if (dev->netdev_ops->ndo_features_check) |
3584 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
3585 | features); | |
8cb65d00 TM |
3586 | else |
3587 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 3588 | |
c1e756bf | 3589 | return harmonize_features(skb, features); |
58e998c6 | 3590 | } |
c1e756bf | 3591 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 3592 | |
2ea25513 | 3593 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 3594 | struct netdev_queue *txq, bool more) |
f6a78bfc | 3595 | { |
2ea25513 DM |
3596 | unsigned int len; |
3597 | int rc; | |
00829823 | 3598 | |
9f9a742d | 3599 | if (dev_nit_active(dev)) |
2ea25513 | 3600 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 3601 | |
2ea25513 | 3602 | len = skb->len; |
3744741a | 3603 | PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); |
2ea25513 | 3604 | trace_net_dev_start_xmit(skb, dev); |
95f6b3dd | 3605 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 3606 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 3607 | |
2ea25513 DM |
3608 | return rc; |
3609 | } | |
7b9c6090 | 3610 | |
8dcda22a DM |
3611 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
3612 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
3613 | { |
3614 | struct sk_buff *skb = first; | |
3615 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 3616 | |
7f2e870f DM |
3617 | while (skb) { |
3618 | struct sk_buff *next = skb->next; | |
fc70fb64 | 3619 | |
a8305bff | 3620 | skb_mark_not_on_list(skb); |
95f6b3dd | 3621 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
3622 | if (unlikely(!dev_xmit_complete(rc))) { |
3623 | skb->next = next; | |
3624 | goto out; | |
3625 | } | |
6afff0ca | 3626 | |
7f2e870f | 3627 | skb = next; |
fe60faa5 | 3628 | if (netif_tx_queue_stopped(txq) && skb) { |
7f2e870f DM |
3629 | rc = NETDEV_TX_BUSY; |
3630 | break; | |
9ccb8975 | 3631 | } |
7f2e870f | 3632 | } |
9ccb8975 | 3633 | |
7f2e870f DM |
3634 | out: |
3635 | *ret = rc; | |
3636 | return skb; | |
3637 | } | |
b40863c6 | 3638 | |
1ff0dc94 ED |
3639 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
3640 | netdev_features_t features) | |
f6a78bfc | 3641 | { |
df8a39de | 3642 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
3643 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
3644 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
3645 | return skb; |
3646 | } | |
f6a78bfc | 3647 | |
43c26a1a DC |
3648 | int skb_csum_hwoffload_help(struct sk_buff *skb, |
3649 | const netdev_features_t features) | |
3650 | { | |
fa821170 | 3651 | if (unlikely(skb_csum_is_sctp(skb))) |
43c26a1a DC |
3652 | return !!(features & NETIF_F_SCTP_CRC) ? 0 : |
3653 | skb_crc32c_csum_help(skb); | |
3654 | ||
62fafcd6 XL |
3655 | if (features & NETIF_F_HW_CSUM) |
3656 | return 0; | |
3657 | ||
3658 | if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { | |
3659 | switch (skb->csum_offset) { | |
3660 | case offsetof(struct tcphdr, check): | |
3661 | case offsetof(struct udphdr, check): | |
3662 | return 0; | |
3663 | } | |
3664 | } | |
3665 | ||
3666 | return skb_checksum_help(skb); | |
43c26a1a DC |
3667 | } |
3668 | EXPORT_SYMBOL(skb_csum_hwoffload_help); | |
3669 | ||
f53c7239 | 3670 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) |
eae3f88e DM |
3671 | { |
3672 | netdev_features_t features; | |
f6a78bfc | 3673 | |
eae3f88e DM |
3674 | features = netif_skb_features(skb); |
3675 | skb = validate_xmit_vlan(skb, features); | |
3676 | if (unlikely(!skb)) | |
3677 | goto out_null; | |
7b9c6090 | 3678 | |
ebf4e808 IL |
3679 | skb = sk_validate_xmit_skb(skb, dev); |
3680 | if (unlikely(!skb)) | |
3681 | goto out_null; | |
3682 | ||
8b86a61d | 3683 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
3684 | struct sk_buff *segs; |
3685 | ||
3686 | segs = skb_gso_segment(skb, features); | |
cecda693 | 3687 | if (IS_ERR(segs)) { |
af6dabc9 | 3688 | goto out_kfree_skb; |
cecda693 JW |
3689 | } else if (segs) { |
3690 | consume_skb(skb); | |
3691 | skb = segs; | |
f6a78bfc | 3692 | } |
eae3f88e DM |
3693 | } else { |
3694 | if (skb_needs_linearize(skb, features) && | |
3695 | __skb_linearize(skb)) | |
3696 | goto out_kfree_skb; | |
4ec93edb | 3697 | |
eae3f88e DM |
3698 | /* If packet is not checksummed and device does not |
3699 | * support checksumming for this protocol, complete | |
3700 | * checksumming here. | |
3701 | */ | |
3702 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
3703 | if (skb->encapsulation) | |
3704 | skb_set_inner_transport_header(skb, | |
3705 | skb_checksum_start_offset(skb)); | |
3706 | else | |
3707 | skb_set_transport_header(skb, | |
3708 | skb_checksum_start_offset(skb)); | |
43c26a1a | 3709 | if (skb_csum_hwoffload_help(skb, features)) |
eae3f88e | 3710 | goto out_kfree_skb; |
7b9c6090 | 3711 | } |
0c772159 | 3712 | } |
7b9c6090 | 3713 | |
f53c7239 | 3714 | skb = validate_xmit_xfrm(skb, features, again); |
3dca3f38 | 3715 | |
eae3f88e | 3716 | return skb; |
fc70fb64 | 3717 | |
f6a78bfc HX |
3718 | out_kfree_skb: |
3719 | kfree_skb(skb); | |
eae3f88e | 3720 | out_null: |
d21fd63e | 3721 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3722 | return NULL; |
3723 | } | |
6afff0ca | 3724 | |
f53c7239 | 3725 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) |
55a93b3e ED |
3726 | { |
3727 | struct sk_buff *next, *head = NULL, *tail; | |
3728 | ||
bec3cfdc | 3729 | for (; skb != NULL; skb = next) { |
55a93b3e | 3730 | next = skb->next; |
a8305bff | 3731 | skb_mark_not_on_list(skb); |
bec3cfdc ED |
3732 | |
3733 | /* in case skb wont be segmented, point to itself */ | |
3734 | skb->prev = skb; | |
3735 | ||
f53c7239 | 3736 | skb = validate_xmit_skb(skb, dev, again); |
bec3cfdc ED |
3737 | if (!skb) |
3738 | continue; | |
55a93b3e | 3739 | |
bec3cfdc ED |
3740 | if (!head) |
3741 | head = skb; | |
3742 | else | |
3743 | tail->next = skb; | |
3744 | /* If skb was segmented, skb->prev points to | |
3745 | * the last segment. If not, it still contains skb. | |
3746 | */ | |
3747 | tail = skb->prev; | |
55a93b3e ED |
3748 | } |
3749 | return head; | |
f6a78bfc | 3750 | } |
104ba78c | 3751 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3752 | |
1def9238 ED |
3753 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3754 | { | |
3755 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3756 | ||
3757 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3758 | ||
3759 | /* To get more precise estimation of bytes sent on wire, | |
3760 | * we add to pkt_len the headers size of all segments | |
3761 | */ | |
a0dce875 | 3762 | if (shinfo->gso_size && skb_transport_header_was_set(skb)) { |
757b8b1d | 3763 | unsigned int hdr_len; |
15e5a030 | 3764 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3765 | |
757b8b1d ED |
3766 | /* mac layer + network layer */ |
3767 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3768 | ||
3769 | /* + transport layer */ | |
7c68d1a6 ED |
3770 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { |
3771 | const struct tcphdr *th; | |
3772 | struct tcphdr _tcphdr; | |
3773 | ||
3774 | th = skb_header_pointer(skb, skb_transport_offset(skb), | |
3775 | sizeof(_tcphdr), &_tcphdr); | |
3776 | if (likely(th)) | |
3777 | hdr_len += __tcp_hdrlen(th); | |
3778 | } else { | |
3779 | struct udphdr _udphdr; | |
3780 | ||
3781 | if (skb_header_pointer(skb, skb_transport_offset(skb), | |
3782 | sizeof(_udphdr), &_udphdr)) | |
3783 | hdr_len += sizeof(struct udphdr); | |
3784 | } | |
15e5a030 JW |
3785 | |
3786 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3787 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3788 | shinfo->gso_size); | |
3789 | ||
3790 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3791 | } |
3792 | } | |
3793 | ||
bbd8a0d3 KK |
3794 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3795 | struct net_device *dev, | |
3796 | struct netdev_queue *txq) | |
3797 | { | |
3798 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3799 | struct sk_buff *to_free = NULL; |
a2da570d | 3800 | bool contended; |
bbd8a0d3 KK |
3801 | int rc; |
3802 | ||
a2da570d | 3803 | qdisc_calculate_pkt_len(skb, q); |
6b3ba914 JF |
3804 | |
3805 | if (q->flags & TCQ_F_NOLOCK) { | |
ac5c66f2 | 3806 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
379349e9 | 3807 | qdisc_run(q); |
6b3ba914 JF |
3808 | |
3809 | if (unlikely(to_free)) | |
3810 | kfree_skb_list(to_free); | |
3811 | return rc; | |
3812 | } | |
3813 | ||
79640a4c ED |
3814 | /* |
3815 | * Heuristic to force contended enqueues to serialize on a | |
3816 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3817 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3818 | * often and dequeue packets faster. |
79640a4c | 3819 | */ |
a2da570d | 3820 | contended = qdisc_is_running(q); |
79640a4c ED |
3821 | if (unlikely(contended)) |
3822 | spin_lock(&q->busylock); | |
3823 | ||
bbd8a0d3 KK |
3824 | spin_lock(root_lock); |
3825 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3826 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3827 | rc = NET_XMIT_DROP; |
3828 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3829 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3830 | /* |
3831 | * This is a work-conserving queue; there are no old skbs | |
3832 | * waiting to be sent out; and the qdisc is not running - | |
3833 | * xmit the skb directly. | |
3834 | */ | |
bfe0d029 | 3835 | |
bfe0d029 ED |
3836 | qdisc_bstats_update(q, skb); |
3837 | ||
55a93b3e | 3838 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3839 | if (unlikely(contended)) { |
3840 | spin_unlock(&q->busylock); | |
3841 | contended = false; | |
3842 | } | |
bbd8a0d3 | 3843 | __qdisc_run(q); |
6c148184 | 3844 | } |
bbd8a0d3 | 3845 | |
6c148184 | 3846 | qdisc_run_end(q); |
bbd8a0d3 KK |
3847 | rc = NET_XMIT_SUCCESS; |
3848 | } else { | |
ac5c66f2 | 3849 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3850 | if (qdisc_run_begin(q)) { |
3851 | if (unlikely(contended)) { | |
3852 | spin_unlock(&q->busylock); | |
3853 | contended = false; | |
3854 | } | |
3855 | __qdisc_run(q); | |
6c148184 | 3856 | qdisc_run_end(q); |
79640a4c | 3857 | } |
bbd8a0d3 KK |
3858 | } |
3859 | spin_unlock(root_lock); | |
520ac30f ED |
3860 | if (unlikely(to_free)) |
3861 | kfree_skb_list(to_free); | |
79640a4c ED |
3862 | if (unlikely(contended)) |
3863 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3864 | return rc; |
3865 | } | |
3866 | ||
86f8515f | 3867 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3868 | static void skb_update_prio(struct sk_buff *skb) |
3869 | { | |
4dcb31d4 ED |
3870 | const struct netprio_map *map; |
3871 | const struct sock *sk; | |
3872 | unsigned int prioidx; | |
5bc1421e | 3873 | |
4dcb31d4 ED |
3874 | if (skb->priority) |
3875 | return; | |
3876 | map = rcu_dereference_bh(skb->dev->priomap); | |
3877 | if (!map) | |
3878 | return; | |
3879 | sk = skb_to_full_sk(skb); | |
3880 | if (!sk) | |
3881 | return; | |
91c68ce2 | 3882 | |
4dcb31d4 ED |
3883 | prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); |
3884 | ||
3885 | if (prioidx < map->priomap_len) | |
3886 | skb->priority = map->priomap[prioidx]; | |
5bc1421e NH |
3887 | } |
3888 | #else | |
3889 | #define skb_update_prio(skb) | |
3890 | #endif | |
3891 | ||
95603e22 MM |
3892 | /** |
3893 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3894 | * @net: network namespace this loopback is happening in |
3895 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3896 | * @skb: buffer to transmit |
3897 | */ | |
0c4b51f0 | 3898 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3899 | { |
3900 | skb_reset_mac_header(skb); | |
3901 | __skb_pull(skb, skb_network_offset(skb)); | |
3902 | skb->pkt_type = PACKET_LOOPBACK; | |
3903 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3904 | WARN_ON(!skb_dst(skb)); | |
3905 | skb_dst_force(skb); | |
3906 | netif_rx_ni(skb); | |
3907 | return 0; | |
3908 | } | |
3909 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3910 | ||
1f211a1b DB |
3911 | #ifdef CONFIG_NET_EGRESS |
3912 | static struct sk_buff * | |
3913 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3914 | { | |
46209401 | 3915 | struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); |
1f211a1b DB |
3916 | struct tcf_result cl_res; |
3917 | ||
46209401 | 3918 | if (!miniq) |
1f211a1b DB |
3919 | return skb; |
3920 | ||
8dc07fdb | 3921 | /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ |
aadaca9e | 3922 | qdisc_skb_cb(skb)->mru = 0; |
7baf2429 | 3923 | qdisc_skb_cb(skb)->post_ct = false; |
46209401 | 3924 | mini_qdisc_bstats_cpu_update(miniq, skb); |
1f211a1b | 3925 | |
46209401 | 3926 | switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { |
1f211a1b DB |
3927 | case TC_ACT_OK: |
3928 | case TC_ACT_RECLASSIFY: | |
3929 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3930 | break; | |
3931 | case TC_ACT_SHOT: | |
46209401 | 3932 | mini_qdisc_qstats_cpu_drop(miniq); |
1f211a1b | 3933 | *ret = NET_XMIT_DROP; |
7e2c3aea DB |
3934 | kfree_skb(skb); |
3935 | return NULL; | |
1f211a1b DB |
3936 | case TC_ACT_STOLEN: |
3937 | case TC_ACT_QUEUED: | |
e25ea21f | 3938 | case TC_ACT_TRAP: |
1f211a1b | 3939 | *ret = NET_XMIT_SUCCESS; |
7e2c3aea | 3940 | consume_skb(skb); |
1f211a1b DB |
3941 | return NULL; |
3942 | case TC_ACT_REDIRECT: | |
3943 | /* No need to push/pop skb's mac_header here on egress! */ | |
3944 | skb_do_redirect(skb); | |
3945 | *ret = NET_XMIT_SUCCESS; | |
3946 | return NULL; | |
3947 | default: | |
3948 | break; | |
3949 | } | |
357b6cc5 | 3950 | |
1f211a1b DB |
3951 | return skb; |
3952 | } | |
3953 | #endif /* CONFIG_NET_EGRESS */ | |
3954 | ||
fc9bab24 AN |
3955 | #ifdef CONFIG_XPS |
3956 | static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, | |
3957 | struct xps_dev_maps *dev_maps, unsigned int tci) | |
3958 | { | |
3959 | struct xps_map *map; | |
3960 | int queue_index = -1; | |
3961 | ||
3962 | if (dev->num_tc) { | |
3963 | tci *= dev->num_tc; | |
3964 | tci += netdev_get_prio_tc_map(dev, skb->priority); | |
3965 | } | |
3966 | ||
3967 | map = rcu_dereference(dev_maps->attr_map[tci]); | |
3968 | if (map) { | |
3969 | if (map->len == 1) | |
3970 | queue_index = map->queues[0]; | |
3971 | else | |
3972 | queue_index = map->queues[reciprocal_scale( | |
3973 | skb_get_hash(skb), map->len)]; | |
3974 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3975 | queue_index = -1; | |
3976 | } | |
3977 | return queue_index; | |
3978 | } | |
3979 | #endif | |
3980 | ||
eadec877 AD |
3981 | static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, |
3982 | struct sk_buff *skb) | |
638b2a69 JP |
3983 | { |
3984 | #ifdef CONFIG_XPS | |
3985 | struct xps_dev_maps *dev_maps; | |
fc9bab24 | 3986 | struct sock *sk = skb->sk; |
638b2a69 JP |
3987 | int queue_index = -1; |
3988 | ||
04157469 AN |
3989 | if (!static_key_false(&xps_needed)) |
3990 | return -1; | |
3991 | ||
638b2a69 | 3992 | rcu_read_lock(); |
fc9bab24 AN |
3993 | if (!static_key_false(&xps_rxqs_needed)) |
3994 | goto get_cpus_map; | |
3995 | ||
eadec877 | 3996 | dev_maps = rcu_dereference(sb_dev->xps_rxqs_map); |
638b2a69 | 3997 | if (dev_maps) { |
fc9bab24 | 3998 | int tci = sk_rx_queue_get(sk); |
184c449f | 3999 | |
fc9bab24 AN |
4000 | if (tci >= 0 && tci < dev->num_rx_queues) |
4001 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
4002 | tci); | |
4003 | } | |
184c449f | 4004 | |
fc9bab24 AN |
4005 | get_cpus_map: |
4006 | if (queue_index < 0) { | |
eadec877 | 4007 | dev_maps = rcu_dereference(sb_dev->xps_cpus_map); |
fc9bab24 AN |
4008 | if (dev_maps) { |
4009 | unsigned int tci = skb->sender_cpu - 1; | |
4010 | ||
4011 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
4012 | tci); | |
638b2a69 JP |
4013 | } |
4014 | } | |
4015 | rcu_read_unlock(); | |
4016 | ||
4017 | return queue_index; | |
4018 | #else | |
4019 | return -1; | |
4020 | #endif | |
4021 | } | |
4022 | ||
a4ea8a3d | 4023 | u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, |
a350ecce | 4024 | struct net_device *sb_dev) |
a4ea8a3d AD |
4025 | { |
4026 | return 0; | |
4027 | } | |
4028 | EXPORT_SYMBOL(dev_pick_tx_zero); | |
4029 | ||
4030 | u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, | |
a350ecce | 4031 | struct net_device *sb_dev) |
a4ea8a3d AD |
4032 | { |
4033 | return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; | |
4034 | } | |
4035 | EXPORT_SYMBOL(dev_pick_tx_cpu_id); | |
4036 | ||
b71b5837 PA |
4037 | u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, |
4038 | struct net_device *sb_dev) | |
638b2a69 JP |
4039 | { |
4040 | struct sock *sk = skb->sk; | |
4041 | int queue_index = sk_tx_queue_get(sk); | |
4042 | ||
eadec877 AD |
4043 | sb_dev = sb_dev ? : dev; |
4044 | ||
638b2a69 JP |
4045 | if (queue_index < 0 || skb->ooo_okay || |
4046 | queue_index >= dev->real_num_tx_queues) { | |
eadec877 | 4047 | int new_index = get_xps_queue(dev, sb_dev, skb); |
f4563a75 | 4048 | |
638b2a69 | 4049 | if (new_index < 0) |
eadec877 | 4050 | new_index = skb_tx_hash(dev, sb_dev, skb); |
638b2a69 JP |
4051 | |
4052 | if (queue_index != new_index && sk && | |
004a5d01 | 4053 | sk_fullsock(sk) && |
638b2a69 JP |
4054 | rcu_access_pointer(sk->sk_dst_cache)) |
4055 | sk_tx_queue_set(sk, new_index); | |
4056 | ||
4057 | queue_index = new_index; | |
4058 | } | |
4059 | ||
4060 | return queue_index; | |
4061 | } | |
b71b5837 | 4062 | EXPORT_SYMBOL(netdev_pick_tx); |
638b2a69 | 4063 | |
4bd97d51 PA |
4064 | struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, |
4065 | struct sk_buff *skb, | |
4066 | struct net_device *sb_dev) | |
638b2a69 JP |
4067 | { |
4068 | int queue_index = 0; | |
4069 | ||
4070 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
4071 | u32 sender_cpu = skb->sender_cpu - 1; |
4072 | ||
4073 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
4074 | skb->sender_cpu = raw_smp_processor_id() + 1; |
4075 | #endif | |
4076 | ||
4077 | if (dev->real_num_tx_queues != 1) { | |
4078 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 4079 | |
638b2a69 | 4080 | if (ops->ndo_select_queue) |
a350ecce | 4081 | queue_index = ops->ndo_select_queue(dev, skb, sb_dev); |
638b2a69 | 4082 | else |
4bd97d51 | 4083 | queue_index = netdev_pick_tx(dev, skb, sb_dev); |
638b2a69 | 4084 | |
d584527c | 4085 | queue_index = netdev_cap_txqueue(dev, queue_index); |
638b2a69 JP |
4086 | } |
4087 | ||
4088 | skb_set_queue_mapping(skb, queue_index); | |
4089 | return netdev_get_tx_queue(dev, queue_index); | |
4090 | } | |
4091 | ||
d29f749e | 4092 | /** |
9d08dd3d | 4093 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 4094 | * @skb: buffer to transmit |
eadec877 | 4095 | * @sb_dev: suboordinate device used for L2 forwarding offload |
d29f749e DJ |
4096 | * |
4097 | * Queue a buffer for transmission to a network device. The caller must | |
4098 | * have set the device and priority and built the buffer before calling | |
4099 | * this function. The function can be called from an interrupt. | |
4100 | * | |
4101 | * A negative errno code is returned on a failure. A success does not | |
4102 | * guarantee the frame will be transmitted as it may be dropped due | |
4103 | * to congestion or traffic shaping. | |
4104 | * | |
4105 | * ----------------------------------------------------------------------------------- | |
4106 | * I notice this method can also return errors from the queue disciplines, | |
4107 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
4108 | * be positive. | |
4109 | * | |
4110 | * Regardless of the return value, the skb is consumed, so it is currently | |
4111 | * difficult to retry a send to this method. (You can bump the ref count | |
4112 | * before sending to hold a reference for retry if you are careful.) | |
4113 | * | |
4114 | * When calling this method, interrupts MUST be enabled. This is because | |
4115 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
4116 | * --BLG | |
4117 | */ | |
eadec877 | 4118 | static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) |
1da177e4 LT |
4119 | { |
4120 | struct net_device *dev = skb->dev; | |
dc2b4847 | 4121 | struct netdev_queue *txq; |
1da177e4 LT |
4122 | struct Qdisc *q; |
4123 | int rc = -ENOMEM; | |
f53c7239 | 4124 | bool again = false; |
1da177e4 | 4125 | |
6d1ccff6 ED |
4126 | skb_reset_mac_header(skb); |
4127 | ||
e7fd2885 | 4128 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
e7ed11ee | 4129 | __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); |
e7fd2885 | 4130 | |
4ec93edb YH |
4131 | /* Disable soft irqs for various locks below. Also |
4132 | * stops preemption for RCU. | |
1da177e4 | 4133 | */ |
4ec93edb | 4134 | rcu_read_lock_bh(); |
1da177e4 | 4135 | |
5bc1421e NH |
4136 | skb_update_prio(skb); |
4137 | ||
1f211a1b DB |
4138 | qdisc_pkt_len_init(skb); |
4139 | #ifdef CONFIG_NET_CLS_ACT | |
8dc07fdb | 4140 | skb->tc_at_ingress = 0; |
357b6cc5 | 4141 | # ifdef CONFIG_NET_EGRESS |
aabf6772 | 4142 | if (static_branch_unlikely(&egress_needed_key)) { |
1f211a1b DB |
4143 | skb = sch_handle_egress(skb, &rc, dev); |
4144 | if (!skb) | |
4145 | goto out; | |
4146 | } | |
357b6cc5 | 4147 | # endif |
1f211a1b | 4148 | #endif |
02875878 ED |
4149 | /* If device/qdisc don't need skb->dst, release it right now while |
4150 | * its hot in this cpu cache. | |
4151 | */ | |
4152 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
4153 | skb_dst_drop(skb); | |
4154 | else | |
4155 | skb_dst_force(skb); | |
4156 | ||
4bd97d51 | 4157 | txq = netdev_core_pick_tx(dev, skb, sb_dev); |
a898def2 | 4158 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 4159 | |
cf66ba58 | 4160 | trace_net_dev_queue(skb); |
1da177e4 | 4161 | if (q->enqueue) { |
bbd8a0d3 | 4162 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 4163 | goto out; |
1da177e4 LT |
4164 | } |
4165 | ||
4166 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 4167 | * loopback, all the sorts of tunnels... |
1da177e4 | 4168 | |
eb13da1a | 4169 | * Really, it is unlikely that netif_tx_lock protection is necessary |
4170 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
4171 | * counters.) | |
4172 | * However, it is possible, that they rely on protection | |
4173 | * made by us here. | |
1da177e4 | 4174 | |
eb13da1a | 4175 | * Check this and shot the lock. It is not prone from deadlocks. |
4176 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
4177 | */ |
4178 | if (dev->flags & IFF_UP) { | |
4179 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
4180 | ||
c773e847 | 4181 | if (txq->xmit_lock_owner != cpu) { |
97cdcf37 | 4182 | if (dev_xmit_recursion()) |
745e20f1 ED |
4183 | goto recursion_alert; |
4184 | ||
f53c7239 | 4185 | skb = validate_xmit_skb(skb, dev, &again); |
1f59533f | 4186 | if (!skb) |
d21fd63e | 4187 | goto out; |
1f59533f | 4188 | |
3744741a | 4189 | PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); |
c773e847 | 4190 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 4191 | |
73466498 | 4192 | if (!netif_xmit_stopped(txq)) { |
97cdcf37 | 4193 | dev_xmit_recursion_inc(); |
ce93718f | 4194 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
97cdcf37 | 4195 | dev_xmit_recursion_dec(); |
572a9d7b | 4196 | if (dev_xmit_complete(rc)) { |
c773e847 | 4197 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
4198 | goto out; |
4199 | } | |
4200 | } | |
c773e847 | 4201 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
4202 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
4203 | dev->name); | |
1da177e4 LT |
4204 | } else { |
4205 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
4206 | * unfortunately |
4207 | */ | |
4208 | recursion_alert: | |
e87cc472 JP |
4209 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
4210 | dev->name); | |
1da177e4 LT |
4211 | } |
4212 | } | |
4213 | ||
4214 | rc = -ENETDOWN; | |
d4828d85 | 4215 | rcu_read_unlock_bh(); |
1da177e4 | 4216 | |
015f0688 | 4217 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 4218 | kfree_skb_list(skb); |
1da177e4 LT |
4219 | return rc; |
4220 | out: | |
d4828d85 | 4221 | rcu_read_unlock_bh(); |
1da177e4 LT |
4222 | return rc; |
4223 | } | |
f663dd9a | 4224 | |
2b4aa3ce | 4225 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
4226 | { |
4227 | return __dev_queue_xmit(skb, NULL); | |
4228 | } | |
2b4aa3ce | 4229 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 4230 | |
eadec877 | 4231 | int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) |
f663dd9a | 4232 | { |
eadec877 | 4233 | return __dev_queue_xmit(skb, sb_dev); |
f663dd9a JW |
4234 | } |
4235 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
4236 | ||
36ccdf85 | 4237 | int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) |
865b03f2 MK |
4238 | { |
4239 | struct net_device *dev = skb->dev; | |
4240 | struct sk_buff *orig_skb = skb; | |
4241 | struct netdev_queue *txq; | |
4242 | int ret = NETDEV_TX_BUSY; | |
4243 | bool again = false; | |
4244 | ||
4245 | if (unlikely(!netif_running(dev) || | |
4246 | !netif_carrier_ok(dev))) | |
4247 | goto drop; | |
4248 | ||
4249 | skb = validate_xmit_skb_list(skb, dev, &again); | |
4250 | if (skb != orig_skb) | |
4251 | goto drop; | |
4252 | ||
4253 | skb_set_queue_mapping(skb, queue_id); | |
4254 | txq = skb_get_tx_queue(dev, skb); | |
3744741a | 4255 | PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); |
865b03f2 MK |
4256 | |
4257 | local_bh_disable(); | |
4258 | ||
0ad6f6e7 | 4259 | dev_xmit_recursion_inc(); |
865b03f2 MK |
4260 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
4261 | if (!netif_xmit_frozen_or_drv_stopped(txq)) | |
4262 | ret = netdev_start_xmit(skb, dev, txq, false); | |
4263 | HARD_TX_UNLOCK(dev, txq); | |
0ad6f6e7 | 4264 | dev_xmit_recursion_dec(); |
865b03f2 MK |
4265 | |
4266 | local_bh_enable(); | |
865b03f2 MK |
4267 | return ret; |
4268 | drop: | |
4269 | atomic_long_inc(&dev->tx_dropped); | |
4270 | kfree_skb_list(skb); | |
4271 | return NET_XMIT_DROP; | |
4272 | } | |
36ccdf85 | 4273 | EXPORT_SYMBOL(__dev_direct_xmit); |
1da177e4 | 4274 | |
eb13da1a | 4275 | /************************************************************************* |
4276 | * Receiver routines | |
4277 | *************************************************************************/ | |
1da177e4 | 4278 | |
6b2bedc3 | 4279 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
4280 | EXPORT_SYMBOL(netdev_max_backlog); |
4281 | ||
3b098e2d | 4282 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 | 4283 | int netdev_budget __read_mostly = 300; |
a4837980 KK |
4284 | /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ |
4285 | unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; | |
3d48b53f MT |
4286 | int weight_p __read_mostly = 64; /* old backlog weight */ |
4287 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
4288 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
4289 | int dev_rx_weight __read_mostly = 64; | |
4290 | int dev_tx_weight __read_mostly = 64; | |
323ebb61 EC |
4291 | /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ |
4292 | int gro_normal_batch __read_mostly = 8; | |
1da177e4 | 4293 | |
eecfd7c4 ED |
4294 | /* Called with irq disabled */ |
4295 | static inline void ____napi_schedule(struct softnet_data *sd, | |
4296 | struct napi_struct *napi) | |
4297 | { | |
29863d41 WW |
4298 | struct task_struct *thread; |
4299 | ||
4300 | if (test_bit(NAPI_STATE_THREADED, &napi->state)) { | |
4301 | /* Paired with smp_mb__before_atomic() in | |
5fdd2f0e WW |
4302 | * napi_enable()/dev_set_threaded(). |
4303 | * Use READ_ONCE() to guarantee a complete | |
4304 | * read on napi->thread. Only call | |
29863d41 WW |
4305 | * wake_up_process() when it's not NULL. |
4306 | */ | |
4307 | thread = READ_ONCE(napi->thread); | |
4308 | if (thread) { | |
cb038357 WW |
4309 | /* Avoid doing set_bit() if the thread is in |
4310 | * INTERRUPTIBLE state, cause napi_thread_wait() | |
4311 | * makes sure to proceed with napi polling | |
4312 | * if the thread is explicitly woken from here. | |
4313 | */ | |
4314 | if (READ_ONCE(thread->state) != TASK_INTERRUPTIBLE) | |
4315 | set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); | |
29863d41 WW |
4316 | wake_up_process(thread); |
4317 | return; | |
4318 | } | |
4319 | } | |
4320 | ||
eecfd7c4 ED |
4321 | list_add_tail(&napi->poll_list, &sd->poll_list); |
4322 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4323 | } | |
4324 | ||
bfb564e7 KK |
4325 | #ifdef CONFIG_RPS |
4326 | ||
4327 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 4328 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 4329 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
4330 | u32 rps_cpu_mask __read_mostly; |
4331 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 4332 | |
dc05360f | 4333 | struct static_key_false rps_needed __read_mostly; |
3df97ba8 | 4334 | EXPORT_SYMBOL(rps_needed); |
dc05360f | 4335 | struct static_key_false rfs_needed __read_mostly; |
13bfff25 | 4336 | EXPORT_SYMBOL(rfs_needed); |
adc9300e | 4337 | |
c445477d BH |
4338 | static struct rps_dev_flow * |
4339 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4340 | struct rps_dev_flow *rflow, u16 next_cpu) | |
4341 | { | |
a31196b0 | 4342 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
4343 | #ifdef CONFIG_RFS_ACCEL |
4344 | struct netdev_rx_queue *rxqueue; | |
4345 | struct rps_dev_flow_table *flow_table; | |
4346 | struct rps_dev_flow *old_rflow; | |
4347 | u32 flow_id; | |
4348 | u16 rxq_index; | |
4349 | int rc; | |
4350 | ||
4351 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
4352 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
4353 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
4354 | goto out; |
4355 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
4356 | if (rxq_index == skb_get_rx_queue(skb)) | |
4357 | goto out; | |
4358 | ||
4359 | rxqueue = dev->_rx + rxq_index; | |
4360 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4361 | if (!flow_table) | |
4362 | goto out; | |
61b905da | 4363 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
4364 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
4365 | rxq_index, flow_id); | |
4366 | if (rc < 0) | |
4367 | goto out; | |
4368 | old_rflow = rflow; | |
4369 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
4370 | rflow->filter = rc; |
4371 | if (old_rflow->filter == rflow->filter) | |
4372 | old_rflow->filter = RPS_NO_FILTER; | |
4373 | out: | |
4374 | #endif | |
4375 | rflow->last_qtail = | |
09994d1b | 4376 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
4377 | } |
4378 | ||
09994d1b | 4379 | rflow->cpu = next_cpu; |
c445477d BH |
4380 | return rflow; |
4381 | } | |
4382 | ||
bfb564e7 KK |
4383 | /* |
4384 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
4385 | * CPU from the RPS map of the receiving queue for a given skb. | |
4386 | * rcu_read_lock must be held on entry. | |
4387 | */ | |
4388 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4389 | struct rps_dev_flow **rflowp) | |
4390 | { | |
567e4b79 ED |
4391 | const struct rps_sock_flow_table *sock_flow_table; |
4392 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 4393 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 4394 | struct rps_map *map; |
bfb564e7 | 4395 | int cpu = -1; |
567e4b79 | 4396 | u32 tcpu; |
61b905da | 4397 | u32 hash; |
bfb564e7 KK |
4398 | |
4399 | if (skb_rx_queue_recorded(skb)) { | |
4400 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 4401 | |
62fe0b40 BH |
4402 | if (unlikely(index >= dev->real_num_rx_queues)) { |
4403 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4404 | "%s received packet on queue %u, but number " | |
4405 | "of RX queues is %u\n", | |
4406 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
4407 | goto done; |
4408 | } | |
567e4b79 ED |
4409 | rxqueue += index; |
4410 | } | |
bfb564e7 | 4411 | |
567e4b79 ED |
4412 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
4413 | ||
4414 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 4415 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 4416 | if (!flow_table && !map) |
bfb564e7 KK |
4417 | goto done; |
4418 | ||
2d47b459 | 4419 | skb_reset_network_header(skb); |
61b905da TH |
4420 | hash = skb_get_hash(skb); |
4421 | if (!hash) | |
bfb564e7 KK |
4422 | goto done; |
4423 | ||
fec5e652 TH |
4424 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
4425 | if (flow_table && sock_flow_table) { | |
fec5e652 | 4426 | struct rps_dev_flow *rflow; |
567e4b79 ED |
4427 | u32 next_cpu; |
4428 | u32 ident; | |
4429 | ||
4430 | /* First check into global flow table if there is a match */ | |
4431 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
4432 | if ((ident ^ hash) & ~rps_cpu_mask) | |
4433 | goto try_rps; | |
fec5e652 | 4434 | |
567e4b79 ED |
4435 | next_cpu = ident & rps_cpu_mask; |
4436 | ||
4437 | /* OK, now we know there is a match, | |
4438 | * we can look at the local (per receive queue) flow table | |
4439 | */ | |
61b905da | 4440 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
4441 | tcpu = rflow->cpu; |
4442 | ||
fec5e652 TH |
4443 | /* |
4444 | * If the desired CPU (where last recvmsg was done) is | |
4445 | * different from current CPU (one in the rx-queue flow | |
4446 | * table entry), switch if one of the following holds: | |
a31196b0 | 4447 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
4448 | * - Current CPU is offline. |
4449 | * - The current CPU's queue tail has advanced beyond the | |
4450 | * last packet that was enqueued using this table entry. | |
4451 | * This guarantees that all previous packets for the flow | |
4452 | * have been dequeued, thus preserving in order delivery. | |
4453 | */ | |
4454 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 4455 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 4456 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
4457 | rflow->last_qtail)) >= 0)) { |
4458 | tcpu = next_cpu; | |
c445477d | 4459 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 4460 | } |
c445477d | 4461 | |
a31196b0 | 4462 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
4463 | *rflowp = rflow; |
4464 | cpu = tcpu; | |
4465 | goto done; | |
4466 | } | |
4467 | } | |
4468 | ||
567e4b79 ED |
4469 | try_rps: |
4470 | ||
0a9627f2 | 4471 | if (map) { |
8fc54f68 | 4472 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
4473 | if (cpu_online(tcpu)) { |
4474 | cpu = tcpu; | |
4475 | goto done; | |
4476 | } | |
4477 | } | |
4478 | ||
4479 | done: | |
0a9627f2 TH |
4480 | return cpu; |
4481 | } | |
4482 | ||
c445477d BH |
4483 | #ifdef CONFIG_RFS_ACCEL |
4484 | ||
4485 | /** | |
4486 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
4487 | * @dev: Device on which the filter was set | |
4488 | * @rxq_index: RX queue index | |
4489 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
4490 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
4491 | * | |
4492 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
4493 | * this function for each installed filter and remove the filters for | |
4494 | * which it returns %true. | |
4495 | */ | |
4496 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
4497 | u32 flow_id, u16 filter_id) | |
4498 | { | |
4499 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
4500 | struct rps_dev_flow_table *flow_table; | |
4501 | struct rps_dev_flow *rflow; | |
4502 | bool expire = true; | |
a31196b0 | 4503 | unsigned int cpu; |
c445477d BH |
4504 | |
4505 | rcu_read_lock(); | |
4506 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4507 | if (flow_table && flow_id <= flow_table->mask) { | |
4508 | rflow = &flow_table->flows[flow_id]; | |
6aa7de05 | 4509 | cpu = READ_ONCE(rflow->cpu); |
a31196b0 | 4510 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
4511 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
4512 | rflow->last_qtail) < | |
4513 | (int)(10 * flow_table->mask))) | |
4514 | expire = false; | |
4515 | } | |
4516 | rcu_read_unlock(); | |
4517 | return expire; | |
4518 | } | |
4519 | EXPORT_SYMBOL(rps_may_expire_flow); | |
4520 | ||
4521 | #endif /* CONFIG_RFS_ACCEL */ | |
4522 | ||
0a9627f2 | 4523 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 4524 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 4525 | { |
e36fa2f7 ED |
4526 | struct softnet_data *sd = data; |
4527 | ||
eecfd7c4 | 4528 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 4529 | sd->received_rps++; |
0a9627f2 | 4530 | } |
e36fa2f7 | 4531 | |
fec5e652 | 4532 | #endif /* CONFIG_RPS */ |
0a9627f2 | 4533 | |
e36fa2f7 ED |
4534 | /* |
4535 | * Check if this softnet_data structure is another cpu one | |
4536 | * If yes, queue it to our IPI list and return 1 | |
4537 | * If no, return 0 | |
4538 | */ | |
4539 | static int rps_ipi_queued(struct softnet_data *sd) | |
4540 | { | |
4541 | #ifdef CONFIG_RPS | |
903ceff7 | 4542 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
4543 | |
4544 | if (sd != mysd) { | |
4545 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
4546 | mysd->rps_ipi_list = sd; | |
4547 | ||
4548 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4549 | return 1; | |
4550 | } | |
4551 | #endif /* CONFIG_RPS */ | |
4552 | return 0; | |
4553 | } | |
4554 | ||
99bbc707 WB |
4555 | #ifdef CONFIG_NET_FLOW_LIMIT |
4556 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
4557 | #endif | |
4558 | ||
4559 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
4560 | { | |
4561 | #ifdef CONFIG_NET_FLOW_LIMIT | |
4562 | struct sd_flow_limit *fl; | |
4563 | struct softnet_data *sd; | |
4564 | unsigned int old_flow, new_flow; | |
4565 | ||
4566 | if (qlen < (netdev_max_backlog >> 1)) | |
4567 | return false; | |
4568 | ||
903ceff7 | 4569 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
4570 | |
4571 | rcu_read_lock(); | |
4572 | fl = rcu_dereference(sd->flow_limit); | |
4573 | if (fl) { | |
3958afa1 | 4574 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
4575 | old_flow = fl->history[fl->history_head]; |
4576 | fl->history[fl->history_head] = new_flow; | |
4577 | ||
4578 | fl->history_head++; | |
4579 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
4580 | ||
4581 | if (likely(fl->buckets[old_flow])) | |
4582 | fl->buckets[old_flow]--; | |
4583 | ||
4584 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
4585 | fl->count++; | |
4586 | rcu_read_unlock(); | |
4587 | return true; | |
4588 | } | |
4589 | } | |
4590 | rcu_read_unlock(); | |
4591 | #endif | |
4592 | return false; | |
4593 | } | |
4594 | ||
0a9627f2 TH |
4595 | /* |
4596 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
4597 | * queue (may be a remote CPU queue). | |
4598 | */ | |
fec5e652 TH |
4599 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
4600 | unsigned int *qtail) | |
0a9627f2 | 4601 | { |
e36fa2f7 | 4602 | struct softnet_data *sd; |
0a9627f2 | 4603 | unsigned long flags; |
99bbc707 | 4604 | unsigned int qlen; |
0a9627f2 | 4605 | |
e36fa2f7 | 4606 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
4607 | |
4608 | local_irq_save(flags); | |
0a9627f2 | 4609 | |
e36fa2f7 | 4610 | rps_lock(sd); |
e9e4dd32 JA |
4611 | if (!netif_running(skb->dev)) |
4612 | goto drop; | |
99bbc707 WB |
4613 | qlen = skb_queue_len(&sd->input_pkt_queue); |
4614 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 4615 | if (qlen) { |
0a9627f2 | 4616 | enqueue: |
e36fa2f7 | 4617 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 4618 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 4619 | rps_unlock(sd); |
152102c7 | 4620 | local_irq_restore(flags); |
0a9627f2 TH |
4621 | return NET_RX_SUCCESS; |
4622 | } | |
4623 | ||
ebda37c2 ED |
4624 | /* Schedule NAPI for backlog device |
4625 | * We can use non atomic operation since we own the queue lock | |
4626 | */ | |
4627 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 4628 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 4629 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
4630 | } |
4631 | goto enqueue; | |
4632 | } | |
4633 | ||
e9e4dd32 | 4634 | drop: |
dee42870 | 4635 | sd->dropped++; |
e36fa2f7 | 4636 | rps_unlock(sd); |
0a9627f2 | 4637 | |
0a9627f2 TH |
4638 | local_irq_restore(flags); |
4639 | ||
caf586e5 | 4640 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
4641 | kfree_skb(skb); |
4642 | return NET_RX_DROP; | |
4643 | } | |
1da177e4 | 4644 | |
e817f856 JDB |
4645 | static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) |
4646 | { | |
4647 | struct net_device *dev = skb->dev; | |
4648 | struct netdev_rx_queue *rxqueue; | |
4649 | ||
4650 | rxqueue = dev->_rx; | |
4651 | ||
4652 | if (skb_rx_queue_recorded(skb)) { | |
4653 | u16 index = skb_get_rx_queue(skb); | |
4654 | ||
4655 | if (unlikely(index >= dev->real_num_rx_queues)) { | |
4656 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4657 | "%s received packet on queue %u, but number " | |
4658 | "of RX queues is %u\n", | |
4659 | dev->name, index, dev->real_num_rx_queues); | |
4660 | ||
4661 | return rxqueue; /* Return first rxqueue */ | |
4662 | } | |
4663 | rxqueue += index; | |
4664 | } | |
4665 | return rxqueue; | |
4666 | } | |
4667 | ||
d4455169 | 4668 | static u32 netif_receive_generic_xdp(struct sk_buff *skb, |
02671e23 | 4669 | struct xdp_buff *xdp, |
d4455169 JF |
4670 | struct bpf_prog *xdp_prog) |
4671 | { | |
be9df4af | 4672 | void *orig_data, *orig_data_end, *hard_start; |
e817f856 | 4673 | struct netdev_rx_queue *rxqueue; |
de8f3a83 | 4674 | u32 metalen, act = XDP_DROP; |
43b5169d | 4675 | u32 mac_len, frame_sz; |
29724956 JDB |
4676 | __be16 orig_eth_type; |
4677 | struct ethhdr *eth; | |
4678 | bool orig_bcast; | |
be9df4af | 4679 | int off; |
d4455169 JF |
4680 | |
4681 | /* Reinjected packets coming from act_mirred or similar should | |
4682 | * not get XDP generic processing. | |
4683 | */ | |
2c64605b | 4684 | if (skb_is_redirected(skb)) |
d4455169 JF |
4685 | return XDP_PASS; |
4686 | ||
de8f3a83 DB |
4687 | /* XDP packets must be linear and must have sufficient headroom |
4688 | * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also | |
4689 | * native XDP provides, thus we need to do it here as well. | |
4690 | */ | |
ad1e03b2 | 4691 | if (skb_cloned(skb) || skb_is_nonlinear(skb) || |
de8f3a83 DB |
4692 | skb_headroom(skb) < XDP_PACKET_HEADROOM) { |
4693 | int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); | |
4694 | int troom = skb->tail + skb->data_len - skb->end; | |
4695 | ||
4696 | /* In case we have to go down the path and also linearize, | |
4697 | * then lets do the pskb_expand_head() work just once here. | |
4698 | */ | |
4699 | if (pskb_expand_head(skb, | |
4700 | hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, | |
4701 | troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) | |
4702 | goto do_drop; | |
2d17d8d7 | 4703 | if (skb_linearize(skb)) |
de8f3a83 DB |
4704 | goto do_drop; |
4705 | } | |
d4455169 JF |
4706 | |
4707 | /* The XDP program wants to see the packet starting at the MAC | |
4708 | * header. | |
4709 | */ | |
4710 | mac_len = skb->data - skb_mac_header(skb); | |
be9df4af | 4711 | hard_start = skb->data - skb_headroom(skb); |
a075767b JDB |
4712 | |
4713 | /* SKB "head" area always have tailroom for skb_shared_info */ | |
be9df4af | 4714 | frame_sz = (void *)skb_end_pointer(skb) - hard_start; |
43b5169d | 4715 | frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
a075767b | 4716 | |
be9df4af LB |
4717 | rxqueue = netif_get_rxqueue(skb); |
4718 | xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); | |
4719 | xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, | |
4720 | skb_headlen(skb) + mac_len, true); | |
a075767b | 4721 | |
02671e23 BT |
4722 | orig_data_end = xdp->data_end; |
4723 | orig_data = xdp->data; | |
29724956 JDB |
4724 | eth = (struct ethhdr *)xdp->data; |
4725 | orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); | |
4726 | orig_eth_type = eth->h_proto; | |
d4455169 | 4727 | |
02671e23 | 4728 | act = bpf_prog_run_xdp(xdp_prog, xdp); |
d4455169 | 4729 | |
065af355 | 4730 | /* check if bpf_xdp_adjust_head was used */ |
02671e23 | 4731 | off = xdp->data - orig_data; |
065af355 JDB |
4732 | if (off) { |
4733 | if (off > 0) | |
4734 | __skb_pull(skb, off); | |
4735 | else if (off < 0) | |
4736 | __skb_push(skb, -off); | |
4737 | ||
4738 | skb->mac_header += off; | |
4739 | skb_reset_network_header(skb); | |
4740 | } | |
d4455169 | 4741 | |
a075767b JDB |
4742 | /* check if bpf_xdp_adjust_tail was used */ |
4743 | off = xdp->data_end - orig_data_end; | |
f7613120 | 4744 | if (off != 0) { |
02671e23 | 4745 | skb_set_tail_pointer(skb, xdp->data_end - xdp->data); |
a075767b | 4746 | skb->len += off; /* positive on grow, negative on shrink */ |
f7613120 | 4747 | } |
198d83bb | 4748 | |
29724956 JDB |
4749 | /* check if XDP changed eth hdr such SKB needs update */ |
4750 | eth = (struct ethhdr *)xdp->data; | |
4751 | if ((orig_eth_type != eth->h_proto) || | |
4752 | (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { | |
4753 | __skb_push(skb, ETH_HLEN); | |
4754 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4755 | } | |
4756 | ||
d4455169 | 4757 | switch (act) { |
6103aa96 | 4758 | case XDP_REDIRECT: |
d4455169 JF |
4759 | case XDP_TX: |
4760 | __skb_push(skb, mac_len); | |
de8f3a83 | 4761 | break; |
d4455169 | 4762 | case XDP_PASS: |
02671e23 | 4763 | metalen = xdp->data - xdp->data_meta; |
de8f3a83 DB |
4764 | if (metalen) |
4765 | skb_metadata_set(skb, metalen); | |
d4455169 | 4766 | break; |
d4455169 JF |
4767 | default: |
4768 | bpf_warn_invalid_xdp_action(act); | |
df561f66 | 4769 | fallthrough; |
d4455169 JF |
4770 | case XDP_ABORTED: |
4771 | trace_xdp_exception(skb->dev, xdp_prog, act); | |
df561f66 | 4772 | fallthrough; |
d4455169 JF |
4773 | case XDP_DROP: |
4774 | do_drop: | |
4775 | kfree_skb(skb); | |
4776 | break; | |
4777 | } | |
4778 | ||
4779 | return act; | |
4780 | } | |
4781 | ||
4782 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
4783 | * network taps in order to match in-driver-XDP behavior. | |
4784 | */ | |
7c497478 | 4785 | void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) |
d4455169 JF |
4786 | { |
4787 | struct net_device *dev = skb->dev; | |
4788 | struct netdev_queue *txq; | |
4789 | bool free_skb = true; | |
4790 | int cpu, rc; | |
4791 | ||
4bd97d51 | 4792 | txq = netdev_core_pick_tx(dev, skb, NULL); |
d4455169 JF |
4793 | cpu = smp_processor_id(); |
4794 | HARD_TX_LOCK(dev, txq, cpu); | |
4795 | if (!netif_xmit_stopped(txq)) { | |
4796 | rc = netdev_start_xmit(skb, dev, txq, 0); | |
4797 | if (dev_xmit_complete(rc)) | |
4798 | free_skb = false; | |
4799 | } | |
4800 | HARD_TX_UNLOCK(dev, txq); | |
4801 | if (free_skb) { | |
4802 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
4803 | kfree_skb(skb); | |
4804 | } | |
4805 | } | |
4806 | ||
02786475 | 4807 | static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); |
d4455169 | 4808 | |
7c497478 | 4809 | int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) |
d4455169 | 4810 | { |
d4455169 | 4811 | if (xdp_prog) { |
02671e23 BT |
4812 | struct xdp_buff xdp; |
4813 | u32 act; | |
6103aa96 | 4814 | int err; |
d4455169 | 4815 | |
02671e23 | 4816 | act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); |
d4455169 | 4817 | if (act != XDP_PASS) { |
6103aa96 JF |
4818 | switch (act) { |
4819 | case XDP_REDIRECT: | |
2facaad6 | 4820 | err = xdp_do_generic_redirect(skb->dev, skb, |
02671e23 | 4821 | &xdp, xdp_prog); |
6103aa96 JF |
4822 | if (err) |
4823 | goto out_redir; | |
02671e23 | 4824 | break; |
6103aa96 | 4825 | case XDP_TX: |
d4455169 | 4826 | generic_xdp_tx(skb, xdp_prog); |
6103aa96 JF |
4827 | break; |
4828 | } | |
d4455169 JF |
4829 | return XDP_DROP; |
4830 | } | |
4831 | } | |
4832 | return XDP_PASS; | |
6103aa96 | 4833 | out_redir: |
6103aa96 JF |
4834 | kfree_skb(skb); |
4835 | return XDP_DROP; | |
d4455169 | 4836 | } |
7c497478 | 4837 | EXPORT_SYMBOL_GPL(do_xdp_generic); |
d4455169 | 4838 | |
ae78dbfa | 4839 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 4840 | { |
b0e28f1e | 4841 | int ret; |
1da177e4 | 4842 | |
588f0330 | 4843 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 4844 | |
cf66ba58 | 4845 | trace_netif_rx(skb); |
d4455169 | 4846 | |
df334545 | 4847 | #ifdef CONFIG_RPS |
dc05360f | 4848 | if (static_branch_unlikely(&rps_needed)) { |
fec5e652 | 4849 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
4850 | int cpu; |
4851 | ||
cece1945 | 4852 | preempt_disable(); |
b0e28f1e | 4853 | rcu_read_lock(); |
fec5e652 TH |
4854 | |
4855 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
4856 | if (cpu < 0) |
4857 | cpu = smp_processor_id(); | |
fec5e652 TH |
4858 | |
4859 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4860 | ||
b0e28f1e | 4861 | rcu_read_unlock(); |
cece1945 | 4862 | preempt_enable(); |
adc9300e ED |
4863 | } else |
4864 | #endif | |
fec5e652 TH |
4865 | { |
4866 | unsigned int qtail; | |
f4563a75 | 4867 | |
fec5e652 TH |
4868 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); |
4869 | put_cpu(); | |
4870 | } | |
b0e28f1e | 4871 | return ret; |
1da177e4 | 4872 | } |
ae78dbfa BH |
4873 | |
4874 | /** | |
4875 | * netif_rx - post buffer to the network code | |
4876 | * @skb: buffer to post | |
4877 | * | |
4878 | * This function receives a packet from a device driver and queues it for | |
4879 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
4880 | * may be dropped during processing for congestion control or by the | |
4881 | * protocol layers. | |
4882 | * | |
4883 | * return values: | |
4884 | * NET_RX_SUCCESS (no congestion) | |
4885 | * NET_RX_DROP (packet was dropped) | |
4886 | * | |
4887 | */ | |
4888 | ||
4889 | int netif_rx(struct sk_buff *skb) | |
4890 | { | |
b0e3f1bd GB |
4891 | int ret; |
4892 | ||
ae78dbfa BH |
4893 | trace_netif_rx_entry(skb); |
4894 | ||
b0e3f1bd GB |
4895 | ret = netif_rx_internal(skb); |
4896 | trace_netif_rx_exit(ret); | |
4897 | ||
4898 | return ret; | |
ae78dbfa | 4899 | } |
d1b19dff | 4900 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
4901 | |
4902 | int netif_rx_ni(struct sk_buff *skb) | |
4903 | { | |
4904 | int err; | |
4905 | ||
ae78dbfa BH |
4906 | trace_netif_rx_ni_entry(skb); |
4907 | ||
1da177e4 | 4908 | preempt_disable(); |
ae78dbfa | 4909 | err = netif_rx_internal(skb); |
1da177e4 LT |
4910 | if (local_softirq_pending()) |
4911 | do_softirq(); | |
4912 | preempt_enable(); | |
b0e3f1bd | 4913 | trace_netif_rx_ni_exit(err); |
1da177e4 LT |
4914 | |
4915 | return err; | |
4916 | } | |
1da177e4 LT |
4917 | EXPORT_SYMBOL(netif_rx_ni); |
4918 | ||
c11171a4 SAS |
4919 | int netif_rx_any_context(struct sk_buff *skb) |
4920 | { | |
4921 | /* | |
4922 | * If invoked from contexts which do not invoke bottom half | |
4923 | * processing either at return from interrupt or when softrqs are | |
4924 | * reenabled, use netif_rx_ni() which invokes bottomhalf processing | |
4925 | * directly. | |
4926 | */ | |
4927 | if (in_interrupt()) | |
4928 | return netif_rx(skb); | |
4929 | else | |
4930 | return netif_rx_ni(skb); | |
4931 | } | |
4932 | EXPORT_SYMBOL(netif_rx_any_context); | |
4933 | ||
0766f788 | 4934 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 4935 | { |
903ceff7 | 4936 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
4937 | |
4938 | if (sd->completion_queue) { | |
4939 | struct sk_buff *clist; | |
4940 | ||
4941 | local_irq_disable(); | |
4942 | clist = sd->completion_queue; | |
4943 | sd->completion_queue = NULL; | |
4944 | local_irq_enable(); | |
4945 | ||
4946 | while (clist) { | |
4947 | struct sk_buff *skb = clist; | |
f4563a75 | 4948 | |
1da177e4 LT |
4949 | clist = clist->next; |
4950 | ||
63354797 | 4951 | WARN_ON(refcount_read(&skb->users)); |
e6247027 ED |
4952 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
4953 | trace_consume_skb(skb); | |
4954 | else | |
4955 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
4956 | |
4957 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
4958 | __kfree_skb(skb); | |
4959 | else | |
4960 | __kfree_skb_defer(skb); | |
1da177e4 LT |
4961 | } |
4962 | } | |
4963 | ||
4964 | if (sd->output_queue) { | |
37437bb2 | 4965 | struct Qdisc *head; |
1da177e4 LT |
4966 | |
4967 | local_irq_disable(); | |
4968 | head = sd->output_queue; | |
4969 | sd->output_queue = NULL; | |
a9cbd588 | 4970 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
4971 | local_irq_enable(); |
4972 | ||
4973 | while (head) { | |
37437bb2 | 4974 | struct Qdisc *q = head; |
6b3ba914 | 4975 | spinlock_t *root_lock = NULL; |
37437bb2 | 4976 | |
1da177e4 LT |
4977 | head = head->next_sched; |
4978 | ||
6b3ba914 JF |
4979 | if (!(q->flags & TCQ_F_NOLOCK)) { |
4980 | root_lock = qdisc_lock(q); | |
4981 | spin_lock(root_lock); | |
4982 | } | |
3bcb846c ED |
4983 | /* We need to make sure head->next_sched is read |
4984 | * before clearing __QDISC_STATE_SCHED | |
4985 | */ | |
4986 | smp_mb__before_atomic(); | |
4987 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
4988 | qdisc_run(q); | |
6b3ba914 JF |
4989 | if (root_lock) |
4990 | spin_unlock(root_lock); | |
1da177e4 LT |
4991 | } |
4992 | } | |
f53c7239 SK |
4993 | |
4994 | xfrm_dev_backlog(sd); | |
1da177e4 LT |
4995 | } |
4996 | ||
181402a5 | 4997 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
4998 | /* This hook is defined here for ATM LANE */ |
4999 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
5000 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 5001 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 5002 | #endif |
1da177e4 | 5003 | |
1f211a1b DB |
5004 | static inline struct sk_buff * |
5005 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
9aa1206e | 5006 | struct net_device *orig_dev, bool *another) |
f697c3e8 | 5007 | { |
e7582bab | 5008 | #ifdef CONFIG_NET_CLS_ACT |
46209401 | 5009 | struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); |
d2788d34 | 5010 | struct tcf_result cl_res; |
24824a09 | 5011 | |
c9e99fd0 DB |
5012 | /* If there's at least one ingress present somewhere (so |
5013 | * we get here via enabled static key), remaining devices | |
5014 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 5015 | * out here. |
c9e99fd0 | 5016 | */ |
46209401 | 5017 | if (!miniq) |
4577139b | 5018 | return skb; |
46209401 | 5019 | |
f697c3e8 HX |
5020 | if (*pt_prev) { |
5021 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
5022 | *pt_prev = NULL; | |
1da177e4 LT |
5023 | } |
5024 | ||
3365495c | 5025 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
aadaca9e | 5026 | qdisc_skb_cb(skb)->mru = 0; |
7baf2429 | 5027 | qdisc_skb_cb(skb)->post_ct = false; |
8dc07fdb | 5028 | skb->tc_at_ingress = 1; |
46209401 | 5029 | mini_qdisc_bstats_cpu_update(miniq, skb); |
c9e99fd0 | 5030 | |
7d17c544 PB |
5031 | switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list, |
5032 | &cl_res, false)) { | |
d2788d34 DB |
5033 | case TC_ACT_OK: |
5034 | case TC_ACT_RECLASSIFY: | |
5035 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
5036 | break; | |
5037 | case TC_ACT_SHOT: | |
46209401 | 5038 | mini_qdisc_qstats_cpu_drop(miniq); |
8a3a4c6e ED |
5039 | kfree_skb(skb); |
5040 | return NULL; | |
d2788d34 DB |
5041 | case TC_ACT_STOLEN: |
5042 | case TC_ACT_QUEUED: | |
e25ea21f | 5043 | case TC_ACT_TRAP: |
8a3a4c6e | 5044 | consume_skb(skb); |
d2788d34 | 5045 | return NULL; |
27b29f63 AS |
5046 | case TC_ACT_REDIRECT: |
5047 | /* skb_mac_header check was done by cls/act_bpf, so | |
5048 | * we can safely push the L2 header back before | |
5049 | * redirecting to another netdev | |
5050 | */ | |
5051 | __skb_push(skb, skb->mac_len); | |
9aa1206e DB |
5052 | if (skb_do_redirect(skb) == -EAGAIN) { |
5053 | __skb_pull(skb, skb->mac_len); | |
5054 | *another = true; | |
5055 | break; | |
5056 | } | |
27b29f63 | 5057 | return NULL; |
720f22fe | 5058 | case TC_ACT_CONSUMED: |
cd11b164 | 5059 | return NULL; |
d2788d34 DB |
5060 | default: |
5061 | break; | |
f697c3e8 | 5062 | } |
e7582bab | 5063 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
5064 | return skb; |
5065 | } | |
1da177e4 | 5066 | |
24b27fc4 MB |
5067 | /** |
5068 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
5069 | * @dev: device to check | |
5070 | * | |
5071 | * Check if a receive handler is already registered for a given device. | |
5072 | * Return true if there one. | |
5073 | * | |
5074 | * The caller must hold the rtnl_mutex. | |
5075 | */ | |
5076 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
5077 | { | |
5078 | ASSERT_RTNL(); | |
5079 | return dev && rtnl_dereference(dev->rx_handler); | |
5080 | } | |
5081 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
5082 | ||
ab95bfe0 JP |
5083 | /** |
5084 | * netdev_rx_handler_register - register receive handler | |
5085 | * @dev: device to register a handler for | |
5086 | * @rx_handler: receive handler to register | |
93e2c32b | 5087 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 5088 | * |
e227867f | 5089 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
5090 | * called from __netif_receive_skb. A negative errno code is returned |
5091 | * on a failure. | |
5092 | * | |
5093 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
5094 | * |
5095 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
5096 | */ |
5097 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
5098 | rx_handler_func_t *rx_handler, |
5099 | void *rx_handler_data) | |
ab95bfe0 | 5100 | { |
1b7cd004 | 5101 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
5102 | return -EBUSY; |
5103 | ||
f5426250 PA |
5104 | if (dev->priv_flags & IFF_NO_RX_HANDLER) |
5105 | return -EINVAL; | |
5106 | ||
00cfec37 | 5107 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 5108 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
5109 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
5110 | ||
5111 | return 0; | |
5112 | } | |
5113 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
5114 | ||
5115 | /** | |
5116 | * netdev_rx_handler_unregister - unregister receive handler | |
5117 | * @dev: device to unregister a handler from | |
5118 | * | |
166ec369 | 5119 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
5120 | * |
5121 | * The caller must hold the rtnl_mutex. | |
5122 | */ | |
5123 | void netdev_rx_handler_unregister(struct net_device *dev) | |
5124 | { | |
5125 | ||
5126 | ASSERT_RTNL(); | |
a9b3cd7f | 5127 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
5128 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
5129 | * section has a guarantee to see a non NULL rx_handler_data | |
5130 | * as well. | |
5131 | */ | |
5132 | synchronize_net(); | |
a9b3cd7f | 5133 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
5134 | } |
5135 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
5136 | ||
b4b9e355 MG |
5137 | /* |
5138 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
5139 | * the special handling of PFMEMALLOC skbs. | |
5140 | */ | |
5141 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
5142 | { | |
5143 | switch (skb->protocol) { | |
2b8837ae JP |
5144 | case htons(ETH_P_ARP): |
5145 | case htons(ETH_P_IP): | |
5146 | case htons(ETH_P_IPV6): | |
5147 | case htons(ETH_P_8021Q): | |
5148 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
5149 | return true; |
5150 | default: | |
5151 | return false; | |
5152 | } | |
5153 | } | |
5154 | ||
e687ad60 PN |
5155 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
5156 | int *ret, struct net_device *orig_dev) | |
5157 | { | |
5158 | if (nf_hook_ingress_active(skb)) { | |
2c1e2703 AC |
5159 | int ingress_retval; |
5160 | ||
e687ad60 PN |
5161 | if (*pt_prev) { |
5162 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
5163 | *pt_prev = NULL; | |
5164 | } | |
5165 | ||
2c1e2703 AC |
5166 | rcu_read_lock(); |
5167 | ingress_retval = nf_hook_ingress(skb); | |
5168 | rcu_read_unlock(); | |
5169 | return ingress_retval; | |
e687ad60 PN |
5170 | } |
5171 | return 0; | |
5172 | } | |
e687ad60 | 5173 | |
c0bbbdc3 | 5174 | static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, |
88eb1944 | 5175 | struct packet_type **ppt_prev) |
1da177e4 LT |
5176 | { |
5177 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 5178 | rx_handler_func_t *rx_handler; |
c0bbbdc3 | 5179 | struct sk_buff *skb = *pskb; |
f2ccd8fa | 5180 | struct net_device *orig_dev; |
8a4eb573 | 5181 | bool deliver_exact = false; |
1da177e4 | 5182 | int ret = NET_RX_DROP; |
252e3346 | 5183 | __be16 type; |
1da177e4 | 5184 | |
588f0330 | 5185 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 5186 | |
cf66ba58 | 5187 | trace_netif_receive_skb(skb); |
9b22ea56 | 5188 | |
cc9bd5ce | 5189 | orig_dev = skb->dev; |
8f903c70 | 5190 | |
c1d2bbe1 | 5191 | skb_reset_network_header(skb); |
fda55eca ED |
5192 | if (!skb_transport_header_was_set(skb)) |
5193 | skb_reset_transport_header(skb); | |
0b5c9db1 | 5194 | skb_reset_mac_len(skb); |
1da177e4 LT |
5195 | |
5196 | pt_prev = NULL; | |
5197 | ||
63d8ea7f | 5198 | another_round: |
b6858177 | 5199 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
5200 | |
5201 | __this_cpu_inc(softnet_data.processed); | |
5202 | ||
458bf2f2 SH |
5203 | if (static_branch_unlikely(&generic_xdp_needed_key)) { |
5204 | int ret2; | |
5205 | ||
5206 | preempt_disable(); | |
5207 | ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
5208 | preempt_enable(); | |
5209 | ||
c0bbbdc3 BS |
5210 | if (ret2 != XDP_PASS) { |
5211 | ret = NET_RX_DROP; | |
5212 | goto out; | |
5213 | } | |
458bf2f2 SH |
5214 | skb_reset_mac_len(skb); |
5215 | } | |
5216 | ||
324cefaf | 5217 | if (eth_type_vlan(skb->protocol)) { |
0d5501c1 | 5218 | skb = skb_vlan_untag(skb); |
bcc6d479 | 5219 | if (unlikely(!skb)) |
2c17d27c | 5220 | goto out; |
bcc6d479 JP |
5221 | } |
5222 | ||
e7246e12 WB |
5223 | if (skb_skip_tc_classify(skb)) |
5224 | goto skip_classify; | |
1da177e4 | 5225 | |
9754e293 | 5226 | if (pfmemalloc) |
b4b9e355 MG |
5227 | goto skip_taps; |
5228 | ||
1da177e4 | 5229 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
5230 | if (pt_prev) |
5231 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5232 | pt_prev = ptype; | |
5233 | } | |
5234 | ||
5235 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
5236 | if (pt_prev) | |
5237 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5238 | pt_prev = ptype; | |
1da177e4 LT |
5239 | } |
5240 | ||
b4b9e355 | 5241 | skip_taps: |
1cf51900 | 5242 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 5243 | if (static_branch_unlikely(&ingress_needed_key)) { |
9aa1206e DB |
5244 | bool another = false; |
5245 | ||
5246 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, | |
5247 | &another); | |
5248 | if (another) | |
5249 | goto another_round; | |
4577139b | 5250 | if (!skb) |
2c17d27c | 5251 | goto out; |
e687ad60 PN |
5252 | |
5253 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 5254 | goto out; |
4577139b | 5255 | } |
1cf51900 | 5256 | #endif |
2c64605b | 5257 | skb_reset_redirect(skb); |
e7246e12 | 5258 | skip_classify: |
9754e293 | 5259 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
5260 | goto drop; |
5261 | ||
df8a39de | 5262 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
5263 | if (pt_prev) { |
5264 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5265 | pt_prev = NULL; | |
5266 | } | |
48cc32d3 | 5267 | if (vlan_do_receive(&skb)) |
2425717b JF |
5268 | goto another_round; |
5269 | else if (unlikely(!skb)) | |
2c17d27c | 5270 | goto out; |
2425717b JF |
5271 | } |
5272 | ||
48cc32d3 | 5273 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
5274 | if (rx_handler) { |
5275 | if (pt_prev) { | |
5276 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5277 | pt_prev = NULL; | |
5278 | } | |
8a4eb573 JP |
5279 | switch (rx_handler(&skb)) { |
5280 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 5281 | ret = NET_RX_SUCCESS; |
2c17d27c | 5282 | goto out; |
8a4eb573 | 5283 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 5284 | goto another_round; |
8a4eb573 JP |
5285 | case RX_HANDLER_EXACT: |
5286 | deliver_exact = true; | |
5287 | case RX_HANDLER_PASS: | |
5288 | break; | |
5289 | default: | |
5290 | BUG(); | |
5291 | } | |
ab95bfe0 | 5292 | } |
1da177e4 | 5293 | |
b14a9fc4 | 5294 | if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { |
36b2f61a GV |
5295 | check_vlan_id: |
5296 | if (skb_vlan_tag_get_id(skb)) { | |
5297 | /* Vlan id is non 0 and vlan_do_receive() above couldn't | |
5298 | * find vlan device. | |
5299 | */ | |
d4b812de | 5300 | skb->pkt_type = PACKET_OTHERHOST; |
324cefaf | 5301 | } else if (eth_type_vlan(skb->protocol)) { |
36b2f61a GV |
5302 | /* Outer header is 802.1P with vlan 0, inner header is |
5303 | * 802.1Q or 802.1AD and vlan_do_receive() above could | |
5304 | * not find vlan dev for vlan id 0. | |
5305 | */ | |
5306 | __vlan_hwaccel_clear_tag(skb); | |
5307 | skb = skb_vlan_untag(skb); | |
5308 | if (unlikely(!skb)) | |
5309 | goto out; | |
5310 | if (vlan_do_receive(&skb)) | |
5311 | /* After stripping off 802.1P header with vlan 0 | |
5312 | * vlan dev is found for inner header. | |
5313 | */ | |
5314 | goto another_round; | |
5315 | else if (unlikely(!skb)) | |
5316 | goto out; | |
5317 | else | |
5318 | /* We have stripped outer 802.1P vlan 0 header. | |
5319 | * But could not find vlan dev. | |
5320 | * check again for vlan id to set OTHERHOST. | |
5321 | */ | |
5322 | goto check_vlan_id; | |
5323 | } | |
d4b812de ED |
5324 | /* Note: we might in the future use prio bits |
5325 | * and set skb->priority like in vlan_do_receive() | |
5326 | * For the time being, just ignore Priority Code Point | |
5327 | */ | |
b1817524 | 5328 | __vlan_hwaccel_clear_tag(skb); |
d4b812de | 5329 | } |
48cc32d3 | 5330 | |
7866a621 SN |
5331 | type = skb->protocol; |
5332 | ||
63d8ea7f | 5333 | /* deliver only exact match when indicated */ |
7866a621 SN |
5334 | if (likely(!deliver_exact)) { |
5335 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
5336 | &ptype_base[ntohs(type) & | |
5337 | PTYPE_HASH_MASK]); | |
5338 | } | |
1f3c8804 | 5339 | |
7866a621 SN |
5340 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
5341 | &orig_dev->ptype_specific); | |
5342 | ||
5343 | if (unlikely(skb->dev != orig_dev)) { | |
5344 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
5345 | &skb->dev->ptype_specific); | |
1da177e4 LT |
5346 | } |
5347 | ||
5348 | if (pt_prev) { | |
1f8b977a | 5349 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
0e698bf6 | 5350 | goto drop; |
88eb1944 | 5351 | *ppt_prev = pt_prev; |
1da177e4 | 5352 | } else { |
b4b9e355 | 5353 | drop: |
6e7333d3 JW |
5354 | if (!deliver_exact) |
5355 | atomic_long_inc(&skb->dev->rx_dropped); | |
5356 | else | |
5357 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
5358 | kfree_skb(skb); |
5359 | /* Jamal, now you will not able to escape explaining | |
5360 | * me how you were going to use this. :-) | |
5361 | */ | |
5362 | ret = NET_RX_DROP; | |
5363 | } | |
5364 | ||
2c17d27c | 5365 | out: |
c0bbbdc3 BS |
5366 | /* The invariant here is that if *ppt_prev is not NULL |
5367 | * then skb should also be non-NULL. | |
5368 | * | |
5369 | * Apparently *ppt_prev assignment above holds this invariant due to | |
5370 | * skb dereferencing near it. | |
5371 | */ | |
5372 | *pskb = skb; | |
9754e293 DM |
5373 | return ret; |
5374 | } | |
5375 | ||
88eb1944 EC |
5376 | static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) |
5377 | { | |
5378 | struct net_device *orig_dev = skb->dev; | |
5379 | struct packet_type *pt_prev = NULL; | |
5380 | int ret; | |
5381 | ||
c0bbbdc3 | 5382 | ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); |
88eb1944 | 5383 | if (pt_prev) |
f5737cba PA |
5384 | ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, |
5385 | skb->dev, pt_prev, orig_dev); | |
88eb1944 EC |
5386 | return ret; |
5387 | } | |
5388 | ||
1c601d82 JDB |
5389 | /** |
5390 | * netif_receive_skb_core - special purpose version of netif_receive_skb | |
5391 | * @skb: buffer to process | |
5392 | * | |
5393 | * More direct receive version of netif_receive_skb(). It should | |
5394 | * only be used by callers that have a need to skip RPS and Generic XDP. | |
2de9780f | 5395 | * Caller must also take care of handling if ``(page_is_)pfmemalloc``. |
1c601d82 JDB |
5396 | * |
5397 | * This function may only be called from softirq context and interrupts | |
5398 | * should be enabled. | |
5399 | * | |
5400 | * Return values (usually ignored): | |
5401 | * NET_RX_SUCCESS: no congestion | |
5402 | * NET_RX_DROP: packet was dropped | |
5403 | */ | |
5404 | int netif_receive_skb_core(struct sk_buff *skb) | |
5405 | { | |
5406 | int ret; | |
5407 | ||
5408 | rcu_read_lock(); | |
88eb1944 | 5409 | ret = __netif_receive_skb_one_core(skb, false); |
1c601d82 JDB |
5410 | rcu_read_unlock(); |
5411 | ||
5412 | return ret; | |
5413 | } | |
5414 | EXPORT_SYMBOL(netif_receive_skb_core); | |
5415 | ||
88eb1944 EC |
5416 | static inline void __netif_receive_skb_list_ptype(struct list_head *head, |
5417 | struct packet_type *pt_prev, | |
5418 | struct net_device *orig_dev) | |
4ce0017a EC |
5419 | { |
5420 | struct sk_buff *skb, *next; | |
5421 | ||
88eb1944 EC |
5422 | if (!pt_prev) |
5423 | return; | |
5424 | if (list_empty(head)) | |
5425 | return; | |
17266ee9 | 5426 | if (pt_prev->list_func != NULL) |
fdf71426 PA |
5427 | INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, |
5428 | ip_list_rcv, head, pt_prev, orig_dev); | |
17266ee9 | 5429 | else |
9a5a90d1 AL |
5430 | list_for_each_entry_safe(skb, next, head, list) { |
5431 | skb_list_del_init(skb); | |
fdf71426 | 5432 | pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
9a5a90d1 | 5433 | } |
88eb1944 EC |
5434 | } |
5435 | ||
5436 | static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) | |
5437 | { | |
5438 | /* Fast-path assumptions: | |
5439 | * - There is no RX handler. | |
5440 | * - Only one packet_type matches. | |
5441 | * If either of these fails, we will end up doing some per-packet | |
5442 | * processing in-line, then handling the 'last ptype' for the whole | |
5443 | * sublist. This can't cause out-of-order delivery to any single ptype, | |
5444 | * because the 'last ptype' must be constant across the sublist, and all | |
5445 | * other ptypes are handled per-packet. | |
5446 | */ | |
5447 | /* Current (common) ptype of sublist */ | |
5448 | struct packet_type *pt_curr = NULL; | |
5449 | /* Current (common) orig_dev of sublist */ | |
5450 | struct net_device *od_curr = NULL; | |
5451 | struct list_head sublist; | |
5452 | struct sk_buff *skb, *next; | |
5453 | ||
9af86f93 | 5454 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5455 | list_for_each_entry_safe(skb, next, head, list) { |
5456 | struct net_device *orig_dev = skb->dev; | |
5457 | struct packet_type *pt_prev = NULL; | |
5458 | ||
22f6bbb7 | 5459 | skb_list_del_init(skb); |
c0bbbdc3 | 5460 | __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); |
9af86f93 EC |
5461 | if (!pt_prev) |
5462 | continue; | |
88eb1944 EC |
5463 | if (pt_curr != pt_prev || od_curr != orig_dev) { |
5464 | /* dispatch old sublist */ | |
88eb1944 EC |
5465 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
5466 | /* start new sublist */ | |
9af86f93 | 5467 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5468 | pt_curr = pt_prev; |
5469 | od_curr = orig_dev; | |
5470 | } | |
9af86f93 | 5471 | list_add_tail(&skb->list, &sublist); |
88eb1944 EC |
5472 | } |
5473 | ||
5474 | /* dispatch final sublist */ | |
9af86f93 | 5475 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
4ce0017a EC |
5476 | } |
5477 | ||
9754e293 DM |
5478 | static int __netif_receive_skb(struct sk_buff *skb) |
5479 | { | |
5480 | int ret; | |
5481 | ||
5482 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 5483 | unsigned int noreclaim_flag; |
9754e293 DM |
5484 | |
5485 | /* | |
5486 | * PFMEMALLOC skbs are special, they should | |
5487 | * - be delivered to SOCK_MEMALLOC sockets only | |
5488 | * - stay away from userspace | |
5489 | * - have bounded memory usage | |
5490 | * | |
5491 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
5492 | * context down to all allocation sites. | |
5493 | */ | |
f1083048 | 5494 | noreclaim_flag = memalloc_noreclaim_save(); |
88eb1944 | 5495 | ret = __netif_receive_skb_one_core(skb, true); |
f1083048 | 5496 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 | 5497 | } else |
88eb1944 | 5498 | ret = __netif_receive_skb_one_core(skb, false); |
9754e293 | 5499 | |
1da177e4 LT |
5500 | return ret; |
5501 | } | |
0a9627f2 | 5502 | |
4ce0017a EC |
5503 | static void __netif_receive_skb_list(struct list_head *head) |
5504 | { | |
5505 | unsigned long noreclaim_flag = 0; | |
5506 | struct sk_buff *skb, *next; | |
5507 | bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ | |
5508 | ||
5509 | list_for_each_entry_safe(skb, next, head, list) { | |
5510 | if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { | |
5511 | struct list_head sublist; | |
5512 | ||
5513 | /* Handle the previous sublist */ | |
5514 | list_cut_before(&sublist, head, &skb->list); | |
b9f463d6 EC |
5515 | if (!list_empty(&sublist)) |
5516 | __netif_receive_skb_list_core(&sublist, pfmemalloc); | |
4ce0017a EC |
5517 | pfmemalloc = !pfmemalloc; |
5518 | /* See comments in __netif_receive_skb */ | |
5519 | if (pfmemalloc) | |
5520 | noreclaim_flag = memalloc_noreclaim_save(); | |
5521 | else | |
5522 | memalloc_noreclaim_restore(noreclaim_flag); | |
5523 | } | |
5524 | } | |
5525 | /* Handle the remaining sublist */ | |
b9f463d6 EC |
5526 | if (!list_empty(head)) |
5527 | __netif_receive_skb_list_core(head, pfmemalloc); | |
4ce0017a EC |
5528 | /* Restore pflags */ |
5529 | if (pfmemalloc) | |
5530 | memalloc_noreclaim_restore(noreclaim_flag); | |
5531 | } | |
5532 | ||
f4e63525 | 5533 | static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) |
b5cdae32 | 5534 | { |
58038695 | 5535 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); |
b5cdae32 DM |
5536 | struct bpf_prog *new = xdp->prog; |
5537 | int ret = 0; | |
5538 | ||
fbee97fe DA |
5539 | if (new) { |
5540 | u32 i; | |
5541 | ||
984fe94f YZ |
5542 | mutex_lock(&new->aux->used_maps_mutex); |
5543 | ||
fbee97fe DA |
5544 | /* generic XDP does not work with DEVMAPs that can |
5545 | * have a bpf_prog installed on an entry | |
5546 | */ | |
5547 | for (i = 0; i < new->aux->used_map_cnt; i++) { | |
984fe94f YZ |
5548 | if (dev_map_can_have_prog(new->aux->used_maps[i]) || |
5549 | cpu_map_prog_allowed(new->aux->used_maps[i])) { | |
5550 | mutex_unlock(&new->aux->used_maps_mutex); | |
92164774 | 5551 | return -EINVAL; |
984fe94f | 5552 | } |
fbee97fe | 5553 | } |
984fe94f YZ |
5554 | |
5555 | mutex_unlock(&new->aux->used_maps_mutex); | |
fbee97fe DA |
5556 | } |
5557 | ||
b5cdae32 | 5558 | switch (xdp->command) { |
58038695 | 5559 | case XDP_SETUP_PROG: |
b5cdae32 DM |
5560 | rcu_assign_pointer(dev->xdp_prog, new); |
5561 | if (old) | |
5562 | bpf_prog_put(old); | |
5563 | ||
5564 | if (old && !new) { | |
02786475 | 5565 | static_branch_dec(&generic_xdp_needed_key); |
b5cdae32 | 5566 | } else if (new && !old) { |
02786475 | 5567 | static_branch_inc(&generic_xdp_needed_key); |
b5cdae32 | 5568 | dev_disable_lro(dev); |
56f5aa77 | 5569 | dev_disable_gro_hw(dev); |
b5cdae32 DM |
5570 | } |
5571 | break; | |
b5cdae32 | 5572 | |
b5cdae32 DM |
5573 | default: |
5574 | ret = -EINVAL; | |
5575 | break; | |
5576 | } | |
5577 | ||
5578 | return ret; | |
5579 | } | |
5580 | ||
ae78dbfa | 5581 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 5582 | { |
2c17d27c JA |
5583 | int ret; |
5584 | ||
588f0330 | 5585 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 5586 | |
c1f19b51 RC |
5587 | if (skb_defer_rx_timestamp(skb)) |
5588 | return NET_RX_SUCCESS; | |
5589 | ||
bbbe211c | 5590 | rcu_read_lock(); |
df334545 | 5591 | #ifdef CONFIG_RPS |
dc05360f | 5592 | if (static_branch_unlikely(&rps_needed)) { |
3b098e2d | 5593 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 5594 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 5595 | |
3b098e2d ED |
5596 | if (cpu >= 0) { |
5597 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
5598 | rcu_read_unlock(); | |
adc9300e | 5599 | return ret; |
3b098e2d | 5600 | } |
fec5e652 | 5601 | } |
1e94d72f | 5602 | #endif |
2c17d27c JA |
5603 | ret = __netif_receive_skb(skb); |
5604 | rcu_read_unlock(); | |
5605 | return ret; | |
0a9627f2 | 5606 | } |
ae78dbfa | 5607 | |
7da517a3 EC |
5608 | static void netif_receive_skb_list_internal(struct list_head *head) |
5609 | { | |
7da517a3 | 5610 | struct sk_buff *skb, *next; |
8c057efa | 5611 | struct list_head sublist; |
7da517a3 | 5612 | |
8c057efa | 5613 | INIT_LIST_HEAD(&sublist); |
7da517a3 EC |
5614 | list_for_each_entry_safe(skb, next, head, list) { |
5615 | net_timestamp_check(netdev_tstamp_prequeue, skb); | |
22f6bbb7 | 5616 | skb_list_del_init(skb); |
8c057efa EC |
5617 | if (!skb_defer_rx_timestamp(skb)) |
5618 | list_add_tail(&skb->list, &sublist); | |
7da517a3 | 5619 | } |
8c057efa | 5620 | list_splice_init(&sublist, head); |
7da517a3 | 5621 | |
7da517a3 EC |
5622 | rcu_read_lock(); |
5623 | #ifdef CONFIG_RPS | |
dc05360f | 5624 | if (static_branch_unlikely(&rps_needed)) { |
7da517a3 EC |
5625 | list_for_each_entry_safe(skb, next, head, list) { |
5626 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
5627 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
5628 | ||
5629 | if (cpu >= 0) { | |
8c057efa | 5630 | /* Will be handled, remove from list */ |
22f6bbb7 | 5631 | skb_list_del_init(skb); |
8c057efa | 5632 | enqueue_to_backlog(skb, cpu, &rflow->last_qtail); |
7da517a3 EC |
5633 | } |
5634 | } | |
5635 | } | |
5636 | #endif | |
5637 | __netif_receive_skb_list(head); | |
5638 | rcu_read_unlock(); | |
5639 | } | |
5640 | ||
ae78dbfa BH |
5641 | /** |
5642 | * netif_receive_skb - process receive buffer from network | |
5643 | * @skb: buffer to process | |
5644 | * | |
5645 | * netif_receive_skb() is the main receive data processing function. | |
5646 | * It always succeeds. The buffer may be dropped during processing | |
5647 | * for congestion control or by the protocol layers. | |
5648 | * | |
5649 | * This function may only be called from softirq context and interrupts | |
5650 | * should be enabled. | |
5651 | * | |
5652 | * Return values (usually ignored): | |
5653 | * NET_RX_SUCCESS: no congestion | |
5654 | * NET_RX_DROP: packet was dropped | |
5655 | */ | |
04eb4489 | 5656 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa | 5657 | { |
b0e3f1bd GB |
5658 | int ret; |
5659 | ||
ae78dbfa BH |
5660 | trace_netif_receive_skb_entry(skb); |
5661 | ||
b0e3f1bd GB |
5662 | ret = netif_receive_skb_internal(skb); |
5663 | trace_netif_receive_skb_exit(ret); | |
5664 | ||
5665 | return ret; | |
ae78dbfa | 5666 | } |
04eb4489 | 5667 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 5668 | |
f6ad8c1b EC |
5669 | /** |
5670 | * netif_receive_skb_list - process many receive buffers from network | |
5671 | * @head: list of skbs to process. | |
5672 | * | |
7da517a3 EC |
5673 | * Since return value of netif_receive_skb() is normally ignored, and |
5674 | * wouldn't be meaningful for a list, this function returns void. | |
f6ad8c1b EC |
5675 | * |
5676 | * This function may only be called from softirq context and interrupts | |
5677 | * should be enabled. | |
5678 | */ | |
5679 | void netif_receive_skb_list(struct list_head *head) | |
5680 | { | |
7da517a3 | 5681 | struct sk_buff *skb; |
f6ad8c1b | 5682 | |
b9f463d6 EC |
5683 | if (list_empty(head)) |
5684 | return; | |
b0e3f1bd GB |
5685 | if (trace_netif_receive_skb_list_entry_enabled()) { |
5686 | list_for_each_entry(skb, head, list) | |
5687 | trace_netif_receive_skb_list_entry(skb); | |
5688 | } | |
7da517a3 | 5689 | netif_receive_skb_list_internal(head); |
b0e3f1bd | 5690 | trace_netif_receive_skb_list_exit(0); |
f6ad8c1b EC |
5691 | } |
5692 | EXPORT_SYMBOL(netif_receive_skb_list); | |
5693 | ||
ce1e2a77 | 5694 | static DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
5695 | |
5696 | /* Network device is going away, flush any packets still pending */ | |
5697 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 5698 | { |
6e583ce5 | 5699 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
5700 | struct softnet_data *sd; |
5701 | ||
5702 | local_bh_disable(); | |
5703 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 5704 | |
145dd5f9 | 5705 | local_irq_disable(); |
e36fa2f7 | 5706 | rps_lock(sd); |
6e7676c1 | 5707 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 5708 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 5709 | __skb_unlink(skb, &sd->input_pkt_queue); |
7df5cb75 | 5710 | dev_kfree_skb_irq(skb); |
76cc8b13 | 5711 | input_queue_head_incr(sd); |
6e583ce5 | 5712 | } |
6e7676c1 | 5713 | } |
e36fa2f7 | 5714 | rps_unlock(sd); |
145dd5f9 | 5715 | local_irq_enable(); |
6e7676c1 CG |
5716 | |
5717 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 5718 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
5719 | __skb_unlink(skb, &sd->process_queue); |
5720 | kfree_skb(skb); | |
76cc8b13 | 5721 | input_queue_head_incr(sd); |
6e7676c1 CG |
5722 | } |
5723 | } | |
145dd5f9 PA |
5724 | local_bh_enable(); |
5725 | } | |
5726 | ||
2de79ee2 PA |
5727 | static bool flush_required(int cpu) |
5728 | { | |
5729 | #if IS_ENABLED(CONFIG_RPS) | |
5730 | struct softnet_data *sd = &per_cpu(softnet_data, cpu); | |
5731 | bool do_flush; | |
5732 | ||
5733 | local_irq_disable(); | |
5734 | rps_lock(sd); | |
5735 | ||
5736 | /* as insertion into process_queue happens with the rps lock held, | |
5737 | * process_queue access may race only with dequeue | |
5738 | */ | |
5739 | do_flush = !skb_queue_empty(&sd->input_pkt_queue) || | |
5740 | !skb_queue_empty_lockless(&sd->process_queue); | |
5741 | rps_unlock(sd); | |
5742 | local_irq_enable(); | |
5743 | ||
5744 | return do_flush; | |
5745 | #endif | |
5746 | /* without RPS we can't safely check input_pkt_queue: during a | |
5747 | * concurrent remote skb_queue_splice() we can detect as empty both | |
5748 | * input_pkt_queue and process_queue even if the latter could end-up | |
5749 | * containing a lot of packets. | |
5750 | */ | |
5751 | return true; | |
5752 | } | |
5753 | ||
41852497 | 5754 | static void flush_all_backlogs(void) |
145dd5f9 | 5755 | { |
2de79ee2 | 5756 | static cpumask_t flush_cpus; |
145dd5f9 PA |
5757 | unsigned int cpu; |
5758 | ||
2de79ee2 PA |
5759 | /* since we are under rtnl lock protection we can use static data |
5760 | * for the cpumask and avoid allocating on stack the possibly | |
5761 | * large mask | |
5762 | */ | |
5763 | ASSERT_RTNL(); | |
5764 | ||
145dd5f9 PA |
5765 | get_online_cpus(); |
5766 | ||
2de79ee2 PA |
5767 | cpumask_clear(&flush_cpus); |
5768 | for_each_online_cpu(cpu) { | |
5769 | if (flush_required(cpu)) { | |
5770 | queue_work_on(cpu, system_highpri_wq, | |
5771 | per_cpu_ptr(&flush_works, cpu)); | |
5772 | cpumask_set_cpu(cpu, &flush_cpus); | |
5773 | } | |
5774 | } | |
145dd5f9 | 5775 | |
2de79ee2 | 5776 | /* we can have in flight packet[s] on the cpus we are not flushing, |
0cbe1e57 | 5777 | * synchronize_net() in unregister_netdevice_many() will take care of |
2de79ee2 PA |
5778 | * them |
5779 | */ | |
5780 | for_each_cpu(cpu, &flush_cpus) | |
41852497 | 5781 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 PA |
5782 | |
5783 | put_online_cpus(); | |
6e583ce5 SH |
5784 | } |
5785 | ||
c8079432 MM |
5786 | /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ |
5787 | static void gro_normal_list(struct napi_struct *napi) | |
5788 | { | |
5789 | if (!napi->rx_count) | |
5790 | return; | |
5791 | netif_receive_skb_list_internal(&napi->rx_list); | |
5792 | INIT_LIST_HEAD(&napi->rx_list); | |
5793 | napi->rx_count = 0; | |
5794 | } | |
5795 | ||
5796 | /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, | |
5797 | * pass the whole batch up to the stack. | |
5798 | */ | |
8dc1c444 | 5799 | static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs) |
c8079432 MM |
5800 | { |
5801 | list_add_tail(&skb->list, &napi->rx_list); | |
8dc1c444 ED |
5802 | napi->rx_count += segs; |
5803 | if (napi->rx_count >= gro_normal_batch) | |
c8079432 MM |
5804 | gro_normal_list(napi); |
5805 | } | |
5806 | ||
c8079432 | 5807 | static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 | 5808 | { |
22061d80 | 5809 | struct packet_offload *ptype; |
d565b0a1 | 5810 | __be16 type = skb->protocol; |
22061d80 | 5811 | struct list_head *head = &offload_base; |
d565b0a1 HX |
5812 | int err = -ENOENT; |
5813 | ||
c3c7c254 ED |
5814 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
5815 | ||
fc59f9a3 HX |
5816 | if (NAPI_GRO_CB(skb)->count == 1) { |
5817 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 5818 | goto out; |
fc59f9a3 | 5819 | } |
d565b0a1 HX |
5820 | |
5821 | rcu_read_lock(); | |
5822 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 5823 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
5824 | continue; |
5825 | ||
aaa5d90b PA |
5826 | err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, |
5827 | ipv6_gro_complete, inet_gro_complete, | |
5828 | skb, 0); | |
d565b0a1 HX |
5829 | break; |
5830 | } | |
5831 | rcu_read_unlock(); | |
5832 | ||
5833 | if (err) { | |
5834 | WARN_ON(&ptype->list == head); | |
5835 | kfree_skb(skb); | |
5836 | return NET_RX_SUCCESS; | |
5837 | } | |
5838 | ||
5839 | out: | |
8dc1c444 | 5840 | gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count); |
c8079432 | 5841 | return NET_RX_SUCCESS; |
d565b0a1 HX |
5842 | } |
5843 | ||
6312fe77 | 5844 | static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, |
07d78363 | 5845 | bool flush_old) |
d565b0a1 | 5846 | { |
6312fe77 | 5847 | struct list_head *head = &napi->gro_hash[index].list; |
d4546c25 | 5848 | struct sk_buff *skb, *p; |
2e71a6f8 | 5849 | |
07d78363 | 5850 | list_for_each_entry_safe_reverse(skb, p, head, list) { |
2e71a6f8 ED |
5851 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) |
5852 | return; | |
992cba7e | 5853 | skb_list_del_init(skb); |
c8079432 | 5854 | napi_gro_complete(napi, skb); |
6312fe77 | 5855 | napi->gro_hash[index].count--; |
d565b0a1 | 5856 | } |
d9f37d01 LR |
5857 | |
5858 | if (!napi->gro_hash[index].count) | |
5859 | __clear_bit(index, &napi->gro_bitmask); | |
d565b0a1 | 5860 | } |
07d78363 | 5861 | |
6312fe77 | 5862 | /* napi->gro_hash[].list contains packets ordered by age. |
07d78363 DM |
5863 | * youngest packets at the head of it. |
5864 | * Complete skbs in reverse order to reduce latencies. | |
5865 | */ | |
5866 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
5867 | { | |
42519ede ED |
5868 | unsigned long bitmask = napi->gro_bitmask; |
5869 | unsigned int i, base = ~0U; | |
07d78363 | 5870 | |
42519ede ED |
5871 | while ((i = ffs(bitmask)) != 0) { |
5872 | bitmask >>= i; | |
5873 | base += i; | |
5874 | __napi_gro_flush_chain(napi, base, flush_old); | |
d9f37d01 | 5875 | } |
07d78363 | 5876 | } |
86cac58b | 5877 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 5878 | |
07d78363 DM |
5879 | static struct list_head *gro_list_prepare(struct napi_struct *napi, |
5880 | struct sk_buff *skb) | |
89c5fa33 | 5881 | { |
89c5fa33 | 5882 | unsigned int maclen = skb->dev->hard_header_len; |
0b4cec8c | 5883 | u32 hash = skb_get_hash_raw(skb); |
07d78363 | 5884 | struct list_head *head; |
d4546c25 | 5885 | struct sk_buff *p; |
89c5fa33 | 5886 | |
6312fe77 | 5887 | head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list; |
07d78363 | 5888 | list_for_each_entry(p, head, list) { |
89c5fa33 ED |
5889 | unsigned long diffs; |
5890 | ||
0b4cec8c TH |
5891 | NAPI_GRO_CB(p)->flush = 0; |
5892 | ||
5893 | if (hash != skb_get_hash_raw(p)) { | |
5894 | NAPI_GRO_CB(p)->same_flow = 0; | |
5895 | continue; | |
5896 | } | |
5897 | ||
89c5fa33 | 5898 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
b1817524 MM |
5899 | diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); |
5900 | if (skb_vlan_tag_present(p)) | |
fc5141cb | 5901 | diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); |
ce87fc6c | 5902 | diffs |= skb_metadata_dst_cmp(p, skb); |
de8f3a83 | 5903 | diffs |= skb_metadata_differs(p, skb); |
89c5fa33 ED |
5904 | if (maclen == ETH_HLEN) |
5905 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 5906 | skb_mac_header(skb)); |
89c5fa33 ED |
5907 | else if (!diffs) |
5908 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 5909 | skb_mac_header(skb), |
89c5fa33 ED |
5910 | maclen); |
5911 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 | 5912 | } |
07d78363 DM |
5913 | |
5914 | return head; | |
89c5fa33 ED |
5915 | } |
5916 | ||
299603e8 JC |
5917 | static void skb_gro_reset_offset(struct sk_buff *skb) |
5918 | { | |
5919 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
5920 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
5921 | ||
5922 | NAPI_GRO_CB(skb)->data_offset = 0; | |
5923 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
5924 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
5925 | ||
8aef998d | 5926 | if (!skb_headlen(skb) && pinfo->nr_frags && |
299603e8 JC |
5927 | !PageHighMem(skb_frag_page(frag0))) { |
5928 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
7cfd5fd5 ED |
5929 | NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, |
5930 | skb_frag_size(frag0), | |
5931 | skb->end - skb->tail); | |
89c5fa33 ED |
5932 | } |
5933 | } | |
5934 | ||
a50e233c ED |
5935 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
5936 | { | |
5937 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
5938 | ||
5939 | BUG_ON(skb->end - skb->tail < grow); | |
5940 | ||
5941 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
5942 | ||
5943 | skb->data_len -= grow; | |
5944 | skb->tail += grow; | |
5945 | ||
b54c9d5b | 5946 | skb_frag_off_add(&pinfo->frags[0], grow); |
a50e233c ED |
5947 | skb_frag_size_sub(&pinfo->frags[0], grow); |
5948 | ||
5949 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
5950 | skb_frag_unref(skb, 0); | |
5951 | memmove(pinfo->frags, pinfo->frags + 1, | |
5952 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
5953 | } | |
5954 | } | |
5955 | ||
c8079432 | 5956 | static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) |
07d78363 | 5957 | { |
6312fe77 | 5958 | struct sk_buff *oldest; |
07d78363 | 5959 | |
6312fe77 | 5960 | oldest = list_last_entry(head, struct sk_buff, list); |
07d78363 | 5961 | |
6312fe77 | 5962 | /* We are called with head length >= MAX_GRO_SKBS, so this is |
07d78363 DM |
5963 | * impossible. |
5964 | */ | |
5965 | if (WARN_ON_ONCE(!oldest)) | |
5966 | return; | |
5967 | ||
d9f37d01 LR |
5968 | /* Do not adjust napi->gro_hash[].count, caller is adding a new |
5969 | * SKB to the chain. | |
07d78363 | 5970 | */ |
ece23711 | 5971 | skb_list_del_init(oldest); |
c8079432 | 5972 | napi_gro_complete(napi, oldest); |
07d78363 DM |
5973 | } |
5974 | ||
bb728820 | 5975 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 | 5976 | { |
6312fe77 | 5977 | u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); |
d4546c25 | 5978 | struct list_head *head = &offload_base; |
22061d80 | 5979 | struct packet_offload *ptype; |
d565b0a1 | 5980 | __be16 type = skb->protocol; |
07d78363 | 5981 | struct list_head *gro_head; |
d4546c25 | 5982 | struct sk_buff *pp = NULL; |
5b252f0c | 5983 | enum gro_result ret; |
d4546c25 | 5984 | int same_flow; |
a50e233c | 5985 | int grow; |
d565b0a1 | 5986 | |
b5cdae32 | 5987 | if (netif_elide_gro(skb->dev)) |
d565b0a1 HX |
5988 | goto normal; |
5989 | ||
07d78363 | 5990 | gro_head = gro_list_prepare(napi, skb); |
89c5fa33 | 5991 | |
d565b0a1 HX |
5992 | rcu_read_lock(); |
5993 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 5994 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
5995 | continue; |
5996 | ||
86911732 | 5997 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 5998 | skb_reset_mac_len(skb); |
d565b0a1 | 5999 | NAPI_GRO_CB(skb)->same_flow = 0; |
d61d072e | 6000 | NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); |
5d38a079 | 6001 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 6002 | NAPI_GRO_CB(skb)->encap_mark = 0; |
fcd91dd4 | 6003 | NAPI_GRO_CB(skb)->recursion_counter = 0; |
a0ca153f | 6004 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 6005 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 6006 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 6007 | |
662880f4 TH |
6008 | /* Setup for GRO checksum validation */ |
6009 | switch (skb->ip_summed) { | |
6010 | case CHECKSUM_COMPLETE: | |
6011 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
6012 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
6013 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
6014 | break; | |
6015 | case CHECKSUM_UNNECESSARY: | |
6016 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
6017 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
6018 | break; | |
6019 | default: | |
6020 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
6021 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
6022 | } | |
d565b0a1 | 6023 | |
aaa5d90b PA |
6024 | pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, |
6025 | ipv6_gro_receive, inet_gro_receive, | |
6026 | gro_head, skb); | |
d565b0a1 HX |
6027 | break; |
6028 | } | |
6029 | rcu_read_unlock(); | |
6030 | ||
6031 | if (&ptype->list == head) | |
6032 | goto normal; | |
6033 | ||
45586c70 | 6034 | if (PTR_ERR(pp) == -EINPROGRESS) { |
25393d3f SK |
6035 | ret = GRO_CONSUMED; |
6036 | goto ok; | |
6037 | } | |
6038 | ||
0da2afd5 | 6039 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 6040 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 6041 | |
d565b0a1 | 6042 | if (pp) { |
992cba7e | 6043 | skb_list_del_init(pp); |
c8079432 | 6044 | napi_gro_complete(napi, pp); |
6312fe77 | 6045 | napi->gro_hash[hash].count--; |
d565b0a1 HX |
6046 | } |
6047 | ||
0da2afd5 | 6048 | if (same_flow) |
d565b0a1 HX |
6049 | goto ok; |
6050 | ||
600adc18 | 6051 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 6052 | goto normal; |
d565b0a1 | 6053 | |
6312fe77 | 6054 | if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) { |
c8079432 | 6055 | gro_flush_oldest(napi, gro_head); |
600adc18 | 6056 | } else { |
6312fe77 | 6057 | napi->gro_hash[hash].count++; |
600adc18 | 6058 | } |
d565b0a1 | 6059 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 6060 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 6061 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 6062 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
07d78363 | 6063 | list_add(&skb->list, gro_head); |
5d0d9be8 | 6064 | ret = GRO_HELD; |
d565b0a1 | 6065 | |
ad0f9904 | 6066 | pull: |
a50e233c ED |
6067 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
6068 | if (grow > 0) | |
6069 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 6070 | ok: |
d9f37d01 LR |
6071 | if (napi->gro_hash[hash].count) { |
6072 | if (!test_bit(hash, &napi->gro_bitmask)) | |
6073 | __set_bit(hash, &napi->gro_bitmask); | |
6074 | } else if (test_bit(hash, &napi->gro_bitmask)) { | |
6075 | __clear_bit(hash, &napi->gro_bitmask); | |
6076 | } | |
6077 | ||
5d0d9be8 | 6078 | return ret; |
d565b0a1 HX |
6079 | |
6080 | normal: | |
ad0f9904 HX |
6081 | ret = GRO_NORMAL; |
6082 | goto pull; | |
5d38a079 | 6083 | } |
96e93eab | 6084 | |
bf5a755f JC |
6085 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
6086 | { | |
6087 | struct list_head *offload_head = &offload_base; | |
6088 | struct packet_offload *ptype; | |
6089 | ||
6090 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
6091 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
6092 | continue; | |
6093 | return ptype; | |
6094 | } | |
6095 | return NULL; | |
6096 | } | |
e27a2f83 | 6097 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
6098 | |
6099 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
6100 | { | |
6101 | struct list_head *offload_head = &offload_base; | |
6102 | struct packet_offload *ptype; | |
6103 | ||
6104 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
6105 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
6106 | continue; | |
6107 | return ptype; | |
6108 | } | |
6109 | return NULL; | |
6110 | } | |
e27a2f83 | 6111 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 6112 | |
6570bc79 AL |
6113 | static gro_result_t napi_skb_finish(struct napi_struct *napi, |
6114 | struct sk_buff *skb, | |
6115 | gro_result_t ret) | |
5d38a079 | 6116 | { |
5d0d9be8 HX |
6117 | switch (ret) { |
6118 | case GRO_NORMAL: | |
8dc1c444 | 6119 | gro_normal_one(napi, skb, 1); |
c7c4b3b6 | 6120 | break; |
5d38a079 | 6121 | |
daa86548 | 6122 | case GRO_MERGED_FREE: |
e44699d2 MK |
6123 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
6124 | napi_skb_free_stolen_head(skb); | |
6125 | else | |
9243adfc | 6126 | __kfree_skb_defer(skb); |
daa86548 ED |
6127 | break; |
6128 | ||
5b252f0c BH |
6129 | case GRO_HELD: |
6130 | case GRO_MERGED: | |
25393d3f | 6131 | case GRO_CONSUMED: |
5b252f0c | 6132 | break; |
5d38a079 HX |
6133 | } |
6134 | ||
c7c4b3b6 | 6135 | return ret; |
5d0d9be8 | 6136 | } |
5d0d9be8 | 6137 | |
c7c4b3b6 | 6138 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 6139 | { |
b0e3f1bd GB |
6140 | gro_result_t ret; |
6141 | ||
93f93a44 | 6142 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 6143 | trace_napi_gro_receive_entry(skb); |
86911732 | 6144 | |
a50e233c ED |
6145 | skb_gro_reset_offset(skb); |
6146 | ||
6570bc79 | 6147 | ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); |
b0e3f1bd GB |
6148 | trace_napi_gro_receive_exit(ret); |
6149 | ||
6150 | return ret; | |
d565b0a1 HX |
6151 | } |
6152 | EXPORT_SYMBOL(napi_gro_receive); | |
6153 | ||
d0c2b0d2 | 6154 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 6155 | { |
93a35f59 ED |
6156 | if (unlikely(skb->pfmemalloc)) { |
6157 | consume_skb(skb); | |
6158 | return; | |
6159 | } | |
96e93eab | 6160 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
6161 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
6162 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
b1817524 | 6163 | __vlan_hwaccel_clear_tag(skb); |
66c46d74 | 6164 | skb->dev = napi->dev; |
6d152e23 | 6165 | skb->skb_iif = 0; |
33d9a2c7 ED |
6166 | |
6167 | /* eth_type_trans() assumes pkt_type is PACKET_HOST */ | |
6168 | skb->pkt_type = PACKET_HOST; | |
6169 | ||
c3caf119 JC |
6170 | skb->encapsulation = 0; |
6171 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 6172 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
174e2381 | 6173 | skb_ext_reset(skb); |
96e93eab HX |
6174 | |
6175 | napi->skb = skb; | |
6176 | } | |
96e93eab | 6177 | |
76620aaf | 6178 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 6179 | { |
5d38a079 | 6180 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
6181 | |
6182 | if (!skb) { | |
fd11a83d | 6183 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
6184 | if (skb) { |
6185 | napi->skb = skb; | |
6186 | skb_mark_napi_id(skb, napi); | |
6187 | } | |
80595d59 | 6188 | } |
96e93eab HX |
6189 | return skb; |
6190 | } | |
76620aaf | 6191 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 6192 | |
a50e233c ED |
6193 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
6194 | struct sk_buff *skb, | |
6195 | gro_result_t ret) | |
96e93eab | 6196 | { |
5d0d9be8 HX |
6197 | switch (ret) { |
6198 | case GRO_NORMAL: | |
a50e233c ED |
6199 | case GRO_HELD: |
6200 | __skb_push(skb, ETH_HLEN); | |
6201 | skb->protocol = eth_type_trans(skb, skb->dev); | |
323ebb61 | 6202 | if (ret == GRO_NORMAL) |
8dc1c444 | 6203 | gro_normal_one(napi, skb, 1); |
86911732 | 6204 | break; |
5d38a079 | 6205 | |
e44699d2 MK |
6206 | case GRO_MERGED_FREE: |
6207 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) | |
6208 | napi_skb_free_stolen_head(skb); | |
6209 | else | |
6210 | napi_reuse_skb(napi, skb); | |
6211 | break; | |
6212 | ||
5b252f0c | 6213 | case GRO_MERGED: |
25393d3f | 6214 | case GRO_CONSUMED: |
5b252f0c | 6215 | break; |
5d0d9be8 | 6216 | } |
5d38a079 | 6217 | |
c7c4b3b6 | 6218 | return ret; |
5d38a079 | 6219 | } |
5d0d9be8 | 6220 | |
a50e233c ED |
6221 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
6222 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
6223 | * We copy ethernet header into skb->data to have a common layout. | |
6224 | */ | |
4adb9c4a | 6225 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
6226 | { |
6227 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
6228 | const struct ethhdr *eth; |
6229 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
6230 | |
6231 | napi->skb = NULL; | |
6232 | ||
a50e233c ED |
6233 | skb_reset_mac_header(skb); |
6234 | skb_gro_reset_offset(skb); | |
6235 | ||
a50e233c ED |
6236 | if (unlikely(skb_gro_header_hard(skb, hlen))) { |
6237 | eth = skb_gro_header_slow(skb, hlen, 0); | |
6238 | if (unlikely(!eth)) { | |
4da46ceb AC |
6239 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
6240 | __func__, napi->dev->name); | |
a50e233c ED |
6241 | napi_reuse_skb(napi, skb); |
6242 | return NULL; | |
6243 | } | |
6244 | } else { | |
a4270d67 | 6245 | eth = (const struct ethhdr *)skb->data; |
a50e233c ED |
6246 | gro_pull_from_frag0(skb, hlen); |
6247 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
6248 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 6249 | } |
a50e233c ED |
6250 | __skb_pull(skb, hlen); |
6251 | ||
6252 | /* | |
6253 | * This works because the only protocols we care about don't require | |
6254 | * special handling. | |
6255 | * We'll fix it up properly in napi_frags_finish() | |
6256 | */ | |
6257 | skb->protocol = eth->h_proto; | |
76620aaf | 6258 | |
76620aaf HX |
6259 | return skb; |
6260 | } | |
76620aaf | 6261 | |
c7c4b3b6 | 6262 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 6263 | { |
b0e3f1bd | 6264 | gro_result_t ret; |
76620aaf | 6265 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 | 6266 | |
ae78dbfa BH |
6267 | trace_napi_gro_frags_entry(skb); |
6268 | ||
b0e3f1bd GB |
6269 | ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
6270 | trace_napi_gro_frags_exit(ret); | |
6271 | ||
6272 | return ret; | |
5d0d9be8 | 6273 | } |
5d38a079 HX |
6274 | EXPORT_SYMBOL(napi_gro_frags); |
6275 | ||
573e8fca TH |
6276 | /* Compute the checksum from gro_offset and return the folded value |
6277 | * after adding in any pseudo checksum. | |
6278 | */ | |
6279 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
6280 | { | |
6281 | __wsum wsum; | |
6282 | __sum16 sum; | |
6283 | ||
6284 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
6285 | ||
6286 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
6287 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
14641931 | 6288 | /* See comments in __skb_checksum_complete(). */ |
573e8fca TH |
6289 | if (likely(!sum)) { |
6290 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
6291 | !skb->csum_complete_sw) | |
7fe50ac8 | 6292 | netdev_rx_csum_fault(skb->dev, skb); |
573e8fca TH |
6293 | } |
6294 | ||
6295 | NAPI_GRO_CB(skb)->csum = wsum; | |
6296 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
6297 | ||
6298 | return sum; | |
6299 | } | |
6300 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
6301 | ||
773fc8f6 | 6302 | static void net_rps_send_ipi(struct softnet_data *remsd) |
6303 | { | |
6304 | #ifdef CONFIG_RPS | |
6305 | while (remsd) { | |
6306 | struct softnet_data *next = remsd->rps_ipi_next; | |
6307 | ||
6308 | if (cpu_online(remsd->cpu)) | |
6309 | smp_call_function_single_async(remsd->cpu, &remsd->csd); | |
6310 | remsd = next; | |
6311 | } | |
6312 | #endif | |
6313 | } | |
6314 | ||
e326bed2 | 6315 | /* |
855abcf0 | 6316 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
6317 | * Note: called with local irq disabled, but exits with local irq enabled. |
6318 | */ | |
6319 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
6320 | { | |
6321 | #ifdef CONFIG_RPS | |
6322 | struct softnet_data *remsd = sd->rps_ipi_list; | |
6323 | ||
6324 | if (remsd) { | |
6325 | sd->rps_ipi_list = NULL; | |
6326 | ||
6327 | local_irq_enable(); | |
6328 | ||
6329 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
773fc8f6 | 6330 | net_rps_send_ipi(remsd); |
e326bed2 ED |
6331 | } else |
6332 | #endif | |
6333 | local_irq_enable(); | |
6334 | } | |
6335 | ||
d75b1ade ED |
6336 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
6337 | { | |
6338 | #ifdef CONFIG_RPS | |
6339 | return sd->rps_ipi_list != NULL; | |
6340 | #else | |
6341 | return false; | |
6342 | #endif | |
6343 | } | |
6344 | ||
bea3348e | 6345 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 6346 | { |
eecfd7c4 | 6347 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
6348 | bool again = true; |
6349 | int work = 0; | |
1da177e4 | 6350 | |
e326bed2 ED |
6351 | /* Check if we have pending ipi, its better to send them now, |
6352 | * not waiting net_rx_action() end. | |
6353 | */ | |
d75b1ade | 6354 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
6355 | local_irq_disable(); |
6356 | net_rps_action_and_irq_enable(sd); | |
6357 | } | |
d75b1ade | 6358 | |
3d48b53f | 6359 | napi->weight = dev_rx_weight; |
145dd5f9 | 6360 | while (again) { |
1da177e4 | 6361 | struct sk_buff *skb; |
6e7676c1 CG |
6362 | |
6363 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 6364 | rcu_read_lock(); |
6e7676c1 | 6365 | __netif_receive_skb(skb); |
2c17d27c | 6366 | rcu_read_unlock(); |
76cc8b13 | 6367 | input_queue_head_incr(sd); |
145dd5f9 | 6368 | if (++work >= quota) |
76cc8b13 | 6369 | return work; |
145dd5f9 | 6370 | |
6e7676c1 | 6371 | } |
1da177e4 | 6372 | |
145dd5f9 | 6373 | local_irq_disable(); |
e36fa2f7 | 6374 | rps_lock(sd); |
11ef7a89 | 6375 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
6376 | /* |
6377 | * Inline a custom version of __napi_complete(). | |
6378 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
6379 | * and NAPI_STATE_SCHED is the only possible flag set |
6380 | * on backlog. | |
6381 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
6382 | * and we dont need an smp_mb() memory barrier. |
6383 | */ | |
eecfd7c4 | 6384 | napi->state = 0; |
145dd5f9 PA |
6385 | again = false; |
6386 | } else { | |
6387 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
6388 | &sd->process_queue); | |
bea3348e | 6389 | } |
e36fa2f7 | 6390 | rps_unlock(sd); |
145dd5f9 | 6391 | local_irq_enable(); |
6e7676c1 | 6392 | } |
1da177e4 | 6393 | |
bea3348e SH |
6394 | return work; |
6395 | } | |
1da177e4 | 6396 | |
bea3348e SH |
6397 | /** |
6398 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 6399 | * @n: entry to schedule |
bea3348e | 6400 | * |
bc9ad166 ED |
6401 | * The entry's receive function will be scheduled to run. |
6402 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 6403 | */ |
b5606c2d | 6404 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
6405 | { |
6406 | unsigned long flags; | |
1da177e4 | 6407 | |
bea3348e | 6408 | local_irq_save(flags); |
903ceff7 | 6409 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 6410 | local_irq_restore(flags); |
1da177e4 | 6411 | } |
bea3348e SH |
6412 | EXPORT_SYMBOL(__napi_schedule); |
6413 | ||
39e6c820 ED |
6414 | /** |
6415 | * napi_schedule_prep - check if napi can be scheduled | |
6416 | * @n: napi context | |
6417 | * | |
6418 | * Test if NAPI routine is already running, and if not mark | |
ee1a4c84 | 6419 | * it as running. This is used as a condition variable to |
39e6c820 ED |
6420 | * insure only one NAPI poll instance runs. We also make |
6421 | * sure there is no pending NAPI disable. | |
6422 | */ | |
6423 | bool napi_schedule_prep(struct napi_struct *n) | |
6424 | { | |
6425 | unsigned long val, new; | |
6426 | ||
6427 | do { | |
6428 | val = READ_ONCE(n->state); | |
6429 | if (unlikely(val & NAPIF_STATE_DISABLE)) | |
6430 | return false; | |
6431 | new = val | NAPIF_STATE_SCHED; | |
6432 | ||
6433 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
6434 | * This was suggested by Alexander Duyck, as compiler | |
6435 | * emits better code than : | |
6436 | * if (val & NAPIF_STATE_SCHED) | |
6437 | * new |= NAPIF_STATE_MISSED; | |
6438 | */ | |
6439 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
6440 | NAPIF_STATE_MISSED; | |
6441 | } while (cmpxchg(&n->state, val, new) != val); | |
6442 | ||
6443 | return !(val & NAPIF_STATE_SCHED); | |
6444 | } | |
6445 | EXPORT_SYMBOL(napi_schedule_prep); | |
6446 | ||
bc9ad166 ED |
6447 | /** |
6448 | * __napi_schedule_irqoff - schedule for receive | |
6449 | * @n: entry to schedule | |
6450 | * | |
6451 | * Variant of __napi_schedule() assuming hard irqs are masked | |
6452 | */ | |
6453 | void __napi_schedule_irqoff(struct napi_struct *n) | |
6454 | { | |
6455 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
6456 | } | |
6457 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
6458 | ||
364b6055 | 6459 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 6460 | { |
6f8b12d6 ED |
6461 | unsigned long flags, val, new, timeout = 0; |
6462 | bool ret = true; | |
d565b0a1 HX |
6463 | |
6464 | /* | |
217f6974 ED |
6465 | * 1) Don't let napi dequeue from the cpu poll list |
6466 | * just in case its running on a different cpu. | |
6467 | * 2) If we are busy polling, do nothing here, we have | |
6468 | * the guarantee we will be called later. | |
d565b0a1 | 6469 | */ |
217f6974 ED |
6470 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
6471 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 6472 | return false; |
d565b0a1 | 6473 | |
6f8b12d6 ED |
6474 | if (work_done) { |
6475 | if (n->gro_bitmask) | |
7e417a66 ED |
6476 | timeout = READ_ONCE(n->dev->gro_flush_timeout); |
6477 | n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); | |
6f8b12d6 ED |
6478 | } |
6479 | if (n->defer_hard_irqs_count > 0) { | |
6480 | n->defer_hard_irqs_count--; | |
7e417a66 | 6481 | timeout = READ_ONCE(n->dev->gro_flush_timeout); |
6f8b12d6 ED |
6482 | if (timeout) |
6483 | ret = false; | |
6484 | } | |
6485 | if (n->gro_bitmask) { | |
605108ac PA |
6486 | /* When the NAPI instance uses a timeout and keeps postponing |
6487 | * it, we need to bound somehow the time packets are kept in | |
6488 | * the GRO layer | |
6489 | */ | |
6490 | napi_gro_flush(n, !!timeout); | |
3b47d303 | 6491 | } |
c8079432 MM |
6492 | |
6493 | gro_normal_list(n); | |
6494 | ||
02c1602e | 6495 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
6496 | /* If n->poll_list is not empty, we need to mask irqs */ |
6497 | local_irq_save(flags); | |
02c1602e | 6498 | list_del_init(&n->poll_list); |
d75b1ade ED |
6499 | local_irq_restore(flags); |
6500 | } | |
39e6c820 ED |
6501 | |
6502 | do { | |
6503 | val = READ_ONCE(n->state); | |
6504 | ||
6505 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); | |
6506 | ||
7fd3253a | 6507 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | |
cb038357 | 6508 | NAPIF_STATE_SCHED_THREADED | |
7fd3253a | 6509 | NAPIF_STATE_PREFER_BUSY_POLL); |
39e6c820 ED |
6510 | |
6511 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
6512 | * because we will call napi->poll() one more time. | |
6513 | * This C code was suggested by Alexander Duyck to help gcc. | |
6514 | */ | |
6515 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
6516 | NAPIF_STATE_SCHED; | |
6517 | } while (cmpxchg(&n->state, val, new) != val); | |
6518 | ||
6519 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
6520 | __napi_schedule(n); | |
6521 | return false; | |
6522 | } | |
6523 | ||
6f8b12d6 ED |
6524 | if (timeout) |
6525 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
6526 | HRTIMER_MODE_REL_PINNED); | |
6527 | return ret; | |
d565b0a1 | 6528 | } |
3b47d303 | 6529 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 6530 | |
af12fa6e | 6531 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 6532 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
6533 | { |
6534 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
6535 | struct napi_struct *napi; | |
6536 | ||
6537 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
6538 | if (napi->napi_id == napi_id) | |
6539 | return napi; | |
6540 | ||
6541 | return NULL; | |
6542 | } | |
02d62e86 ED |
6543 | |
6544 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
217f6974 | 6545 | |
7fd3253a | 6546 | static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) |
217f6974 | 6547 | { |
7fd3253a BT |
6548 | if (!skip_schedule) { |
6549 | gro_normal_list(napi); | |
6550 | __napi_schedule(napi); | |
6551 | return; | |
6552 | } | |
217f6974 | 6553 | |
7fd3253a BT |
6554 | if (napi->gro_bitmask) { |
6555 | /* flush too old packets | |
6556 | * If HZ < 1000, flush all packets. | |
6557 | */ | |
6558 | napi_gro_flush(napi, HZ >= 1000); | |
6559 | } | |
217f6974 | 6560 | |
7fd3253a BT |
6561 | gro_normal_list(napi); |
6562 | clear_bit(NAPI_STATE_SCHED, &napi->state); | |
6563 | } | |
6564 | ||
7c951caf BT |
6565 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, |
6566 | u16 budget) | |
217f6974 | 6567 | { |
7fd3253a BT |
6568 | bool skip_schedule = false; |
6569 | unsigned long timeout; | |
217f6974 ED |
6570 | int rc; |
6571 | ||
39e6c820 ED |
6572 | /* Busy polling means there is a high chance device driver hard irq |
6573 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
6574 | * set in napi_schedule_prep(). | |
6575 | * Since we are about to call napi->poll() once more, we can safely | |
6576 | * clear NAPI_STATE_MISSED. | |
6577 | * | |
6578 | * Note: x86 could use a single "lock and ..." instruction | |
6579 | * to perform these two clear_bit() | |
6580 | */ | |
6581 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
6582 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
6583 | ||
6584 | local_bh_disable(); | |
6585 | ||
7fd3253a BT |
6586 | if (prefer_busy_poll) { |
6587 | napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); | |
6588 | timeout = READ_ONCE(napi->dev->gro_flush_timeout); | |
6589 | if (napi->defer_hard_irqs_count && timeout) { | |
6590 | hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); | |
6591 | skip_schedule = true; | |
6592 | } | |
6593 | } | |
6594 | ||
217f6974 ED |
6595 | /* All we really want here is to re-enable device interrupts. |
6596 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
6597 | */ | |
7c951caf | 6598 | rc = napi->poll(napi, budget); |
323ebb61 EC |
6599 | /* We can't gro_normal_list() here, because napi->poll() might have |
6600 | * rearmed the napi (napi_complete_done()) in which case it could | |
6601 | * already be running on another CPU. | |
6602 | */ | |
7c951caf | 6603 | trace_napi_poll(napi, rc, budget); |
217f6974 | 6604 | netpoll_poll_unlock(have_poll_lock); |
7c951caf | 6605 | if (rc == budget) |
7fd3253a | 6606 | __busy_poll_stop(napi, skip_schedule); |
217f6974 | 6607 | local_bh_enable(); |
217f6974 ED |
6608 | } |
6609 | ||
7db6b048 SS |
6610 | void napi_busy_loop(unsigned int napi_id, |
6611 | bool (*loop_end)(void *, unsigned long), | |
7c951caf | 6612 | void *loop_end_arg, bool prefer_busy_poll, u16 budget) |
02d62e86 | 6613 | { |
7db6b048 | 6614 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 6615 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 6616 | void *have_poll_lock = NULL; |
02d62e86 | 6617 | struct napi_struct *napi; |
217f6974 ED |
6618 | |
6619 | restart: | |
217f6974 | 6620 | napi_poll = NULL; |
02d62e86 | 6621 | |
2a028ecb | 6622 | rcu_read_lock(); |
02d62e86 | 6623 | |
545cd5e5 | 6624 | napi = napi_by_id(napi_id); |
02d62e86 ED |
6625 | if (!napi) |
6626 | goto out; | |
6627 | ||
217f6974 ED |
6628 | preempt_disable(); |
6629 | for (;;) { | |
2b5cd0df AD |
6630 | int work = 0; |
6631 | ||
2a028ecb | 6632 | local_bh_disable(); |
217f6974 ED |
6633 | if (!napi_poll) { |
6634 | unsigned long val = READ_ONCE(napi->state); | |
6635 | ||
6636 | /* If multiple threads are competing for this napi, | |
6637 | * we avoid dirtying napi->state as much as we can. | |
6638 | */ | |
6639 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
7fd3253a BT |
6640 | NAPIF_STATE_IN_BUSY_POLL)) { |
6641 | if (prefer_busy_poll) | |
6642 | set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); | |
217f6974 | 6643 | goto count; |
7fd3253a | 6644 | } |
217f6974 ED |
6645 | if (cmpxchg(&napi->state, val, |
6646 | val | NAPIF_STATE_IN_BUSY_POLL | | |
7fd3253a BT |
6647 | NAPIF_STATE_SCHED) != val) { |
6648 | if (prefer_busy_poll) | |
6649 | set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); | |
217f6974 | 6650 | goto count; |
7fd3253a | 6651 | } |
217f6974 ED |
6652 | have_poll_lock = netpoll_poll_lock(napi); |
6653 | napi_poll = napi->poll; | |
6654 | } | |
7c951caf BT |
6655 | work = napi_poll(napi, budget); |
6656 | trace_napi_poll(napi, work, budget); | |
323ebb61 | 6657 | gro_normal_list(napi); |
217f6974 | 6658 | count: |
2b5cd0df | 6659 | if (work > 0) |
7db6b048 | 6660 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 6661 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
2a028ecb | 6662 | local_bh_enable(); |
02d62e86 | 6663 | |
7db6b048 | 6664 | if (!loop_end || loop_end(loop_end_arg, start_time)) |
217f6974 | 6665 | break; |
02d62e86 | 6666 | |
217f6974 ED |
6667 | if (unlikely(need_resched())) { |
6668 | if (napi_poll) | |
7c951caf | 6669 | busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); |
217f6974 ED |
6670 | preempt_enable(); |
6671 | rcu_read_unlock(); | |
6672 | cond_resched(); | |
7db6b048 | 6673 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 6674 | return; |
217f6974 ED |
6675 | goto restart; |
6676 | } | |
6cdf89b1 | 6677 | cpu_relax(); |
217f6974 ED |
6678 | } |
6679 | if (napi_poll) | |
7c951caf | 6680 | busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); |
217f6974 | 6681 | preempt_enable(); |
02d62e86 | 6682 | out: |
2a028ecb | 6683 | rcu_read_unlock(); |
02d62e86 | 6684 | } |
7db6b048 | 6685 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
6686 | |
6687 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 6688 | |
149d6ad8 | 6689 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 6690 | { |
4d092dd2 | 6691 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) |
52bd2d62 | 6692 | return; |
af12fa6e | 6693 | |
52bd2d62 | 6694 | spin_lock(&napi_hash_lock); |
af12fa6e | 6695 | |
545cd5e5 | 6696 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 6697 | do { |
545cd5e5 AD |
6698 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
6699 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
6700 | } while (napi_by_id(napi_gen_id)); |
6701 | napi->napi_id = napi_gen_id; | |
af12fa6e | 6702 | |
52bd2d62 ED |
6703 | hlist_add_head_rcu(&napi->napi_hash_node, |
6704 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 6705 | |
52bd2d62 | 6706 | spin_unlock(&napi_hash_lock); |
af12fa6e | 6707 | } |
af12fa6e ET |
6708 | |
6709 | /* Warning : caller is responsible to make sure rcu grace period | |
6710 | * is respected before freeing memory containing @napi | |
6711 | */ | |
5198d545 | 6712 | static void napi_hash_del(struct napi_struct *napi) |
af12fa6e ET |
6713 | { |
6714 | spin_lock(&napi_hash_lock); | |
6715 | ||
4d092dd2 | 6716 | hlist_del_init_rcu(&napi->napi_hash_node); |
5198d545 | 6717 | |
af12fa6e ET |
6718 | spin_unlock(&napi_hash_lock); |
6719 | } | |
af12fa6e | 6720 | |
3b47d303 ED |
6721 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
6722 | { | |
6723 | struct napi_struct *napi; | |
6724 | ||
6725 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
6726 | |
6727 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
6728 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
6729 | */ | |
6f8b12d6 | 6730 | if (!napi_disable_pending(napi) && |
7fd3253a BT |
6731 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { |
6732 | clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); | |
39e6c820 | 6733 | __napi_schedule_irqoff(napi); |
7fd3253a | 6734 | } |
3b47d303 ED |
6735 | |
6736 | return HRTIMER_NORESTART; | |
6737 | } | |
6738 | ||
7c4ec749 | 6739 | static void init_gro_hash(struct napi_struct *napi) |
d565b0a1 | 6740 | { |
07d78363 DM |
6741 | int i; |
6742 | ||
6312fe77 LR |
6743 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6744 | INIT_LIST_HEAD(&napi->gro_hash[i].list); | |
6745 | napi->gro_hash[i].count = 0; | |
6746 | } | |
7c4ec749 DM |
6747 | napi->gro_bitmask = 0; |
6748 | } | |
6749 | ||
5fdd2f0e WW |
6750 | int dev_set_threaded(struct net_device *dev, bool threaded) |
6751 | { | |
6752 | struct napi_struct *napi; | |
6753 | int err = 0; | |
6754 | ||
6755 | if (dev->threaded == threaded) | |
6756 | return 0; | |
6757 | ||
6758 | if (threaded) { | |
6759 | list_for_each_entry(napi, &dev->napi_list, dev_list) { | |
6760 | if (!napi->thread) { | |
6761 | err = napi_kthread_create(napi); | |
6762 | if (err) { | |
6763 | threaded = false; | |
6764 | break; | |
6765 | } | |
6766 | } | |
6767 | } | |
6768 | } | |
6769 | ||
6770 | dev->threaded = threaded; | |
6771 | ||
6772 | /* Make sure kthread is created before THREADED bit | |
6773 | * is set. | |
6774 | */ | |
6775 | smp_mb__before_atomic(); | |
6776 | ||
6777 | /* Setting/unsetting threaded mode on a napi might not immediately | |
6778 | * take effect, if the current napi instance is actively being | |
6779 | * polled. In this case, the switch between threaded mode and | |
6780 | * softirq mode will happen in the next round of napi_schedule(). | |
6781 | * This should not cause hiccups/stalls to the live traffic. | |
6782 | */ | |
6783 | list_for_each_entry(napi, &dev->napi_list, dev_list) { | |
6784 | if (threaded) | |
6785 | set_bit(NAPI_STATE_THREADED, &napi->state); | |
6786 | else | |
6787 | clear_bit(NAPI_STATE_THREADED, &napi->state); | |
6788 | } | |
6789 | ||
6790 | return err; | |
6791 | } | |
6792 | ||
7c4ec749 DM |
6793 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
6794 | int (*poll)(struct napi_struct *, int), int weight) | |
6795 | { | |
4d092dd2 JK |
6796 | if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) |
6797 | return; | |
6798 | ||
7c4ec749 | 6799 | INIT_LIST_HEAD(&napi->poll_list); |
4d092dd2 | 6800 | INIT_HLIST_NODE(&napi->napi_hash_node); |
7c4ec749 DM |
6801 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
6802 | napi->timer.function = napi_watchdog; | |
6803 | init_gro_hash(napi); | |
5d38a079 | 6804 | napi->skb = NULL; |
323ebb61 EC |
6805 | INIT_LIST_HEAD(&napi->rx_list); |
6806 | napi->rx_count = 0; | |
d565b0a1 | 6807 | napi->poll = poll; |
82dc3c63 | 6808 | if (weight > NAPI_POLL_WEIGHT) |
bf29e9e9 QC |
6809 | netdev_err_once(dev, "%s() called with weight %d\n", __func__, |
6810 | weight); | |
d565b0a1 | 6811 | napi->weight = weight; |
d565b0a1 | 6812 | napi->dev = dev; |
5d38a079 | 6813 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
6814 | napi->poll_owner = -1; |
6815 | #endif | |
6816 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
96e97bc0 JK |
6817 | set_bit(NAPI_STATE_NPSVC, &napi->state); |
6818 | list_add_rcu(&napi->dev_list, &dev->napi_list); | |
93d05d4a | 6819 | napi_hash_add(napi); |
29863d41 WW |
6820 | /* Create kthread for this napi if dev->threaded is set. |
6821 | * Clear dev->threaded if kthread creation failed so that | |
6822 | * threaded mode will not be enabled in napi_enable(). | |
6823 | */ | |
6824 | if (dev->threaded && napi_kthread_create(napi)) | |
6825 | dev->threaded = 0; | |
d565b0a1 HX |
6826 | } |
6827 | EXPORT_SYMBOL(netif_napi_add); | |
6828 | ||
3b47d303 ED |
6829 | void napi_disable(struct napi_struct *n) |
6830 | { | |
6831 | might_sleep(); | |
6832 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
6833 | ||
6834 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
6835 | msleep(1); | |
2d8bff12 NH |
6836 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
6837 | msleep(1); | |
3b47d303 ED |
6838 | |
6839 | hrtimer_cancel(&n->timer); | |
6840 | ||
7fd3253a | 6841 | clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state); |
3b47d303 | 6842 | clear_bit(NAPI_STATE_DISABLE, &n->state); |
29863d41 | 6843 | clear_bit(NAPI_STATE_THREADED, &n->state); |
3b47d303 ED |
6844 | } |
6845 | EXPORT_SYMBOL(napi_disable); | |
6846 | ||
29863d41 WW |
6847 | /** |
6848 | * napi_enable - enable NAPI scheduling | |
6849 | * @n: NAPI context | |
6850 | * | |
6851 | * Resume NAPI from being scheduled on this context. | |
6852 | * Must be paired with napi_disable. | |
6853 | */ | |
6854 | void napi_enable(struct napi_struct *n) | |
6855 | { | |
6856 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
6857 | smp_mb__before_atomic(); | |
6858 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
6859 | clear_bit(NAPI_STATE_NPSVC, &n->state); | |
6860 | if (n->dev->threaded && n->thread) | |
6861 | set_bit(NAPI_STATE_THREADED, &n->state); | |
6862 | } | |
6863 | EXPORT_SYMBOL(napi_enable); | |
6864 | ||
07d78363 | 6865 | static void flush_gro_hash(struct napi_struct *napi) |
d4546c25 | 6866 | { |
07d78363 | 6867 | int i; |
d4546c25 | 6868 | |
07d78363 DM |
6869 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6870 | struct sk_buff *skb, *n; | |
6871 | ||
6312fe77 | 6872 | list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) |
07d78363 | 6873 | kfree_skb(skb); |
6312fe77 | 6874 | napi->gro_hash[i].count = 0; |
07d78363 | 6875 | } |
d4546c25 DM |
6876 | } |
6877 | ||
93d05d4a | 6878 | /* Must be called in process context */ |
5198d545 | 6879 | void __netif_napi_del(struct napi_struct *napi) |
d565b0a1 | 6880 | { |
4d092dd2 JK |
6881 | if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) |
6882 | return; | |
6883 | ||
5198d545 | 6884 | napi_hash_del(napi); |
5251ef82 | 6885 | list_del_rcu(&napi->dev_list); |
76620aaf | 6886 | napi_free_frags(napi); |
d565b0a1 | 6887 | |
07d78363 | 6888 | flush_gro_hash(napi); |
d9f37d01 | 6889 | napi->gro_bitmask = 0; |
29863d41 WW |
6890 | |
6891 | if (napi->thread) { | |
6892 | kthread_stop(napi->thread); | |
6893 | napi->thread = NULL; | |
6894 | } | |
d565b0a1 | 6895 | } |
5198d545 | 6896 | EXPORT_SYMBOL(__netif_napi_del); |
d565b0a1 | 6897 | |
898f8015 | 6898 | static int __napi_poll(struct napi_struct *n, bool *repoll) |
726ce70e | 6899 | { |
726ce70e HX |
6900 | int work, weight; |
6901 | ||
726ce70e HX |
6902 | weight = n->weight; |
6903 | ||
6904 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
6905 | * with netpoll's poll_napi(). Only the entity which | |
6906 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
6907 | * actually make the ->poll() call. Therefore we avoid | |
6908 | * accidentally calling ->poll() when NAPI is not scheduled. | |
6909 | */ | |
6910 | work = 0; | |
6911 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
6912 | work = n->poll(n, weight); | |
1db19db7 | 6913 | trace_napi_poll(n, work, weight); |
726ce70e HX |
6914 | } |
6915 | ||
427d5838 ED |
6916 | if (unlikely(work > weight)) |
6917 | pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n", | |
6918 | n->poll, work, weight); | |
726ce70e HX |
6919 | |
6920 | if (likely(work < weight)) | |
898f8015 | 6921 | return work; |
726ce70e HX |
6922 | |
6923 | /* Drivers must not modify the NAPI state if they | |
6924 | * consume the entire weight. In such cases this code | |
6925 | * still "owns" the NAPI instance and therefore can | |
6926 | * move the instance around on the list at-will. | |
6927 | */ | |
6928 | if (unlikely(napi_disable_pending(n))) { | |
6929 | napi_complete(n); | |
898f8015 | 6930 | return work; |
726ce70e HX |
6931 | } |
6932 | ||
7fd3253a BT |
6933 | /* The NAPI context has more processing work, but busy-polling |
6934 | * is preferred. Exit early. | |
6935 | */ | |
6936 | if (napi_prefer_busy_poll(n)) { | |
6937 | if (napi_complete_done(n, work)) { | |
6938 | /* If timeout is not set, we need to make sure | |
6939 | * that the NAPI is re-scheduled. | |
6940 | */ | |
6941 | napi_schedule(n); | |
6942 | } | |
898f8015 | 6943 | return work; |
7fd3253a BT |
6944 | } |
6945 | ||
d9f37d01 | 6946 | if (n->gro_bitmask) { |
726ce70e HX |
6947 | /* flush too old packets |
6948 | * If HZ < 1000, flush all packets. | |
6949 | */ | |
6950 | napi_gro_flush(n, HZ >= 1000); | |
6951 | } | |
6952 | ||
c8079432 MM |
6953 | gro_normal_list(n); |
6954 | ||
001ce546 HX |
6955 | /* Some drivers may have called napi_schedule |
6956 | * prior to exhausting their budget. | |
6957 | */ | |
6958 | if (unlikely(!list_empty(&n->poll_list))) { | |
6959 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
6960 | n->dev ? n->dev->name : "backlog"); | |
898f8015 | 6961 | return work; |
001ce546 HX |
6962 | } |
6963 | ||
898f8015 FF |
6964 | *repoll = true; |
6965 | ||
6966 | return work; | |
6967 | } | |
6968 | ||
6969 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) | |
6970 | { | |
6971 | bool do_repoll = false; | |
6972 | void *have; | |
6973 | int work; | |
6974 | ||
6975 | list_del_init(&n->poll_list); | |
6976 | ||
6977 | have = netpoll_poll_lock(n); | |
6978 | ||
6979 | work = __napi_poll(n, &do_repoll); | |
6980 | ||
6981 | if (do_repoll) | |
6982 | list_add_tail(&n->poll_list, repoll); | |
726ce70e | 6983 | |
726ce70e HX |
6984 | netpoll_poll_unlock(have); |
6985 | ||
6986 | return work; | |
6987 | } | |
6988 | ||
29863d41 WW |
6989 | static int napi_thread_wait(struct napi_struct *napi) |
6990 | { | |
cb038357 WW |
6991 | bool woken = false; |
6992 | ||
29863d41 WW |
6993 | set_current_state(TASK_INTERRUPTIBLE); |
6994 | ||
27f0ad71 | 6995 | while (!kthread_should_stop()) { |
cb038357 WW |
6996 | /* Testing SCHED_THREADED bit here to make sure the current |
6997 | * kthread owns this napi and could poll on this napi. | |
6998 | * Testing SCHED bit is not enough because SCHED bit might be | |
6999 | * set by some other busy poll thread or by napi_disable(). | |
7000 | */ | |
7001 | if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { | |
29863d41 WW |
7002 | WARN_ON(!list_empty(&napi->poll_list)); |
7003 | __set_current_state(TASK_RUNNING); | |
7004 | return 0; | |
7005 | } | |
7006 | ||
7007 | schedule(); | |
cb038357 WW |
7008 | /* woken being true indicates this thread owns this napi. */ |
7009 | woken = true; | |
29863d41 WW |
7010 | set_current_state(TASK_INTERRUPTIBLE); |
7011 | } | |
7012 | __set_current_state(TASK_RUNNING); | |
27f0ad71 | 7013 | |
29863d41 WW |
7014 | return -1; |
7015 | } | |
7016 | ||
7017 | static int napi_threaded_poll(void *data) | |
7018 | { | |
7019 | struct napi_struct *napi = data; | |
7020 | void *have; | |
7021 | ||
7022 | while (!napi_thread_wait(napi)) { | |
7023 | for (;;) { | |
7024 | bool repoll = false; | |
7025 | ||
7026 | local_bh_disable(); | |
7027 | ||
7028 | have = netpoll_poll_lock(napi); | |
7029 | __napi_poll(napi, &repoll); | |
7030 | netpoll_poll_unlock(have); | |
7031 | ||
29863d41 WW |
7032 | local_bh_enable(); |
7033 | ||
7034 | if (!repoll) | |
7035 | break; | |
7036 | ||
7037 | cond_resched(); | |
7038 | } | |
7039 | } | |
7040 | return 0; | |
7041 | } | |
7042 | ||
0766f788 | 7043 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 7044 | { |
903ceff7 | 7045 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e MW |
7046 | unsigned long time_limit = jiffies + |
7047 | usecs_to_jiffies(netdev_budget_usecs); | |
51b0bded | 7048 | int budget = netdev_budget; |
d75b1ade ED |
7049 | LIST_HEAD(list); |
7050 | LIST_HEAD(repoll); | |
53fb95d3 | 7051 | |
1da177e4 | 7052 | local_irq_disable(); |
d75b1ade ED |
7053 | list_splice_init(&sd->poll_list, &list); |
7054 | local_irq_enable(); | |
1da177e4 | 7055 | |
ceb8d5bf | 7056 | for (;;) { |
bea3348e | 7057 | struct napi_struct *n; |
1da177e4 | 7058 | |
ceb8d5bf HX |
7059 | if (list_empty(&list)) { |
7060 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
fec6e49b | 7061 | return; |
ceb8d5bf HX |
7062 | break; |
7063 | } | |
7064 | ||
6bd373eb HX |
7065 | n = list_first_entry(&list, struct napi_struct, poll_list); |
7066 | budget -= napi_poll(n, &repoll); | |
7067 | ||
d75b1ade | 7068 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
7069 | * Allow this to run for 2 jiffies since which will allow |
7070 | * an average latency of 1.5/HZ. | |
bea3348e | 7071 | */ |
ceb8d5bf HX |
7072 | if (unlikely(budget <= 0 || |
7073 | time_after_eq(jiffies, time_limit))) { | |
7074 | sd->time_squeeze++; | |
7075 | break; | |
7076 | } | |
1da177e4 | 7077 | } |
d75b1ade | 7078 | |
d75b1ade ED |
7079 | local_irq_disable(); |
7080 | ||
7081 | list_splice_tail_init(&sd->poll_list, &list); | |
7082 | list_splice_tail(&repoll, &list); | |
7083 | list_splice(&list, &sd->poll_list); | |
7084 | if (!list_empty(&sd->poll_list)) | |
7085 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
7086 | ||
e326bed2 | 7087 | net_rps_action_and_irq_enable(sd); |
1da177e4 LT |
7088 | } |
7089 | ||
aa9d8560 | 7090 | struct netdev_adjacent { |
9ff162a8 | 7091 | struct net_device *dev; |
5d261913 VF |
7092 | |
7093 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 7094 | bool master; |
5d261913 | 7095 | |
32b6d34f TY |
7096 | /* lookup ignore flag */ |
7097 | bool ignore; | |
7098 | ||
5d261913 VF |
7099 | /* counter for the number of times this device was added to us */ |
7100 | u16 ref_nr; | |
7101 | ||
402dae96 VF |
7102 | /* private field for the users */ |
7103 | void *private; | |
7104 | ||
9ff162a8 JP |
7105 | struct list_head list; |
7106 | struct rcu_head rcu; | |
9ff162a8 JP |
7107 | }; |
7108 | ||
6ea29da1 | 7109 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 7110 | struct list_head *adj_list) |
9ff162a8 | 7111 | { |
5d261913 | 7112 | struct netdev_adjacent *adj; |
5d261913 | 7113 | |
2f268f12 | 7114 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
7115 | if (adj->dev == adj_dev) |
7116 | return adj; | |
9ff162a8 JP |
7117 | } |
7118 | return NULL; | |
7119 | } | |
7120 | ||
eff74233 TY |
7121 | static int ____netdev_has_upper_dev(struct net_device *upper_dev, |
7122 | struct netdev_nested_priv *priv) | |
f1170fd4 | 7123 | { |
eff74233 | 7124 | struct net_device *dev = (struct net_device *)priv->data; |
f1170fd4 DA |
7125 | |
7126 | return upper_dev == dev; | |
7127 | } | |
7128 | ||
9ff162a8 JP |
7129 | /** |
7130 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
7131 | * @dev: device | |
7132 | * @upper_dev: upper device to check | |
7133 | * | |
7134 | * Find out if a device is linked to specified upper device and return true | |
7135 | * in case it is. Note that this checks only immediate upper device, | |
7136 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
7137 | */ | |
7138 | bool netdev_has_upper_dev(struct net_device *dev, | |
7139 | struct net_device *upper_dev) | |
7140 | { | |
eff74233 TY |
7141 | struct netdev_nested_priv priv = { |
7142 | .data = (void *)upper_dev, | |
7143 | }; | |
7144 | ||
9ff162a8 JP |
7145 | ASSERT_RTNL(); |
7146 | ||
32b6d34f | 7147 | return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, |
eff74233 | 7148 | &priv); |
9ff162a8 JP |
7149 | } |
7150 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
7151 | ||
1a3f060c | 7152 | /** |
c1639be9 | 7153 | * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device |
1a3f060c DA |
7154 | * @dev: device |
7155 | * @upper_dev: upper device to check | |
7156 | * | |
7157 | * Find out if a device is linked to specified upper device and return true | |
7158 | * in case it is. Note that this checks the entire upper device chain. | |
7159 | * The caller must hold rcu lock. | |
7160 | */ | |
7161 | ||
1a3f060c DA |
7162 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
7163 | struct net_device *upper_dev) | |
7164 | { | |
eff74233 TY |
7165 | struct netdev_nested_priv priv = { |
7166 | .data = (void *)upper_dev, | |
7167 | }; | |
7168 | ||
32b6d34f | 7169 | return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, |
eff74233 | 7170 | &priv); |
1a3f060c DA |
7171 | } |
7172 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
7173 | ||
9ff162a8 JP |
7174 | /** |
7175 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
7176 | * @dev: device | |
7177 | * | |
7178 | * Find out if a device is linked to an upper device and return true in case | |
7179 | * it is. The caller must hold the RTNL lock. | |
7180 | */ | |
25cc72a3 | 7181 | bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
7182 | { |
7183 | ASSERT_RTNL(); | |
7184 | ||
f1170fd4 | 7185 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 7186 | } |
25cc72a3 | 7187 | EXPORT_SYMBOL(netdev_has_any_upper_dev); |
9ff162a8 JP |
7188 | |
7189 | /** | |
7190 | * netdev_master_upper_dev_get - Get master upper device | |
7191 | * @dev: device | |
7192 | * | |
7193 | * Find a master upper device and return pointer to it or NULL in case | |
7194 | * it's not there. The caller must hold the RTNL lock. | |
7195 | */ | |
7196 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
7197 | { | |
aa9d8560 | 7198 | struct netdev_adjacent *upper; |
9ff162a8 JP |
7199 | |
7200 | ASSERT_RTNL(); | |
7201 | ||
2f268f12 | 7202 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
7203 | return NULL; |
7204 | ||
2f268f12 | 7205 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 7206 | struct netdev_adjacent, list); |
9ff162a8 JP |
7207 | if (likely(upper->master)) |
7208 | return upper->dev; | |
7209 | return NULL; | |
7210 | } | |
7211 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
7212 | ||
32b6d34f TY |
7213 | static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) |
7214 | { | |
7215 | struct netdev_adjacent *upper; | |
7216 | ||
7217 | ASSERT_RTNL(); | |
7218 | ||
7219 | if (list_empty(&dev->adj_list.upper)) | |
7220 | return NULL; | |
7221 | ||
7222 | upper = list_first_entry(&dev->adj_list.upper, | |
7223 | struct netdev_adjacent, list); | |
7224 | if (likely(upper->master) && !upper->ignore) | |
7225 | return upper->dev; | |
7226 | return NULL; | |
7227 | } | |
7228 | ||
0f524a80 DA |
7229 | /** |
7230 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
7231 | * @dev: device | |
7232 | * | |
7233 | * Find out if a device is linked to a lower device and return true in case | |
7234 | * it is. The caller must hold the RTNL lock. | |
7235 | */ | |
7236 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
7237 | { | |
7238 | ASSERT_RTNL(); | |
7239 | ||
7240 | return !list_empty(&dev->adj_list.lower); | |
7241 | } | |
7242 | ||
b6ccba4c VF |
7243 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
7244 | { | |
7245 | struct netdev_adjacent *adj; | |
7246 | ||
7247 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
7248 | ||
7249 | return adj->private; | |
7250 | } | |
7251 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
7252 | ||
44a40855 VY |
7253 | /** |
7254 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
7255 | * @dev: device | |
7256 | * @iter: list_head ** of the current position | |
7257 | * | |
7258 | * Gets the next device from the dev's upper list, starting from iter | |
7259 | * position. The caller must hold RCU read lock. | |
7260 | */ | |
7261 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
7262 | struct list_head **iter) | |
7263 | { | |
7264 | struct netdev_adjacent *upper; | |
7265 | ||
7266 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
7267 | ||
7268 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7269 | ||
7270 | if (&upper->list == &dev->adj_list.upper) | |
7271 | return NULL; | |
7272 | ||
7273 | *iter = &upper->list; | |
7274 | ||
7275 | return upper->dev; | |
7276 | } | |
7277 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
7278 | ||
32b6d34f TY |
7279 | static struct net_device *__netdev_next_upper_dev(struct net_device *dev, |
7280 | struct list_head **iter, | |
7281 | bool *ignore) | |
5343da4c TY |
7282 | { |
7283 | struct netdev_adjacent *upper; | |
7284 | ||
7285 | upper = list_entry((*iter)->next, struct netdev_adjacent, list); | |
7286 | ||
7287 | if (&upper->list == &dev->adj_list.upper) | |
7288 | return NULL; | |
7289 | ||
7290 | *iter = &upper->list; | |
32b6d34f | 7291 | *ignore = upper->ignore; |
5343da4c TY |
7292 | |
7293 | return upper->dev; | |
7294 | } | |
7295 | ||
1a3f060c DA |
7296 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
7297 | struct list_head **iter) | |
7298 | { | |
7299 | struct netdev_adjacent *upper; | |
7300 | ||
7301 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
7302 | ||
7303 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7304 | ||
7305 | if (&upper->list == &dev->adj_list.upper) | |
7306 | return NULL; | |
7307 | ||
7308 | *iter = &upper->list; | |
7309 | ||
7310 | return upper->dev; | |
7311 | } | |
7312 | ||
32b6d34f TY |
7313 | static int __netdev_walk_all_upper_dev(struct net_device *dev, |
7314 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7315 | struct netdev_nested_priv *priv), |
7316 | struct netdev_nested_priv *priv) | |
5343da4c TY |
7317 | { |
7318 | struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7319 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7320 | int ret, cur = 0; | |
32b6d34f | 7321 | bool ignore; |
5343da4c TY |
7322 | |
7323 | now = dev; | |
7324 | iter = &dev->adj_list.upper; | |
7325 | ||
7326 | while (1) { | |
7327 | if (now != dev) { | |
eff74233 | 7328 | ret = fn(now, priv); |
5343da4c TY |
7329 | if (ret) |
7330 | return ret; | |
7331 | } | |
7332 | ||
7333 | next = NULL; | |
7334 | while (1) { | |
32b6d34f | 7335 | udev = __netdev_next_upper_dev(now, &iter, &ignore); |
5343da4c TY |
7336 | if (!udev) |
7337 | break; | |
32b6d34f TY |
7338 | if (ignore) |
7339 | continue; | |
5343da4c TY |
7340 | |
7341 | next = udev; | |
7342 | niter = &udev->adj_list.upper; | |
7343 | dev_stack[cur] = now; | |
7344 | iter_stack[cur++] = iter; | |
7345 | break; | |
7346 | } | |
7347 | ||
7348 | if (!next) { | |
7349 | if (!cur) | |
7350 | return 0; | |
7351 | next = dev_stack[--cur]; | |
7352 | niter = iter_stack[cur]; | |
7353 | } | |
7354 | ||
7355 | now = next; | |
7356 | iter = niter; | |
7357 | } | |
7358 | ||
7359 | return 0; | |
7360 | } | |
7361 | ||
1a3f060c DA |
7362 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, |
7363 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7364 | struct netdev_nested_priv *priv), |
7365 | struct netdev_nested_priv *priv) | |
1a3f060c | 7366 | { |
5343da4c TY |
7367 | struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; |
7368 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7369 | int ret, cur = 0; | |
1a3f060c | 7370 | |
5343da4c TY |
7371 | now = dev; |
7372 | iter = &dev->adj_list.upper; | |
1a3f060c | 7373 | |
5343da4c TY |
7374 | while (1) { |
7375 | if (now != dev) { | |
eff74233 | 7376 | ret = fn(now, priv); |
5343da4c TY |
7377 | if (ret) |
7378 | return ret; | |
7379 | } | |
7380 | ||
7381 | next = NULL; | |
7382 | while (1) { | |
7383 | udev = netdev_next_upper_dev_rcu(now, &iter); | |
7384 | if (!udev) | |
7385 | break; | |
7386 | ||
7387 | next = udev; | |
7388 | niter = &udev->adj_list.upper; | |
7389 | dev_stack[cur] = now; | |
7390 | iter_stack[cur++] = iter; | |
7391 | break; | |
7392 | } | |
7393 | ||
7394 | if (!next) { | |
7395 | if (!cur) | |
7396 | return 0; | |
7397 | next = dev_stack[--cur]; | |
7398 | niter = iter_stack[cur]; | |
7399 | } | |
7400 | ||
7401 | now = next; | |
7402 | iter = niter; | |
1a3f060c DA |
7403 | } |
7404 | ||
7405 | return 0; | |
7406 | } | |
7407 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
7408 | ||
32b6d34f TY |
7409 | static bool __netdev_has_upper_dev(struct net_device *dev, |
7410 | struct net_device *upper_dev) | |
7411 | { | |
eff74233 | 7412 | struct netdev_nested_priv priv = { |
1fc70edb | 7413 | .flags = 0, |
eff74233 TY |
7414 | .data = (void *)upper_dev, |
7415 | }; | |
7416 | ||
32b6d34f TY |
7417 | ASSERT_RTNL(); |
7418 | ||
7419 | return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, | |
eff74233 | 7420 | &priv); |
32b6d34f TY |
7421 | } |
7422 | ||
31088a11 VF |
7423 | /** |
7424 | * netdev_lower_get_next_private - Get the next ->private from the | |
7425 | * lower neighbour list | |
7426 | * @dev: device | |
7427 | * @iter: list_head ** of the current position | |
7428 | * | |
7429 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
7430 | * list, starting from iter position. The caller must hold either hold the | |
7431 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 7432 | * list will remain unchanged. |
31088a11 VF |
7433 | */ |
7434 | void *netdev_lower_get_next_private(struct net_device *dev, | |
7435 | struct list_head **iter) | |
7436 | { | |
7437 | struct netdev_adjacent *lower; | |
7438 | ||
7439 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
7440 | ||
7441 | if (&lower->list == &dev->adj_list.lower) | |
7442 | return NULL; | |
7443 | ||
6859e7df | 7444 | *iter = lower->list.next; |
31088a11 VF |
7445 | |
7446 | return lower->private; | |
7447 | } | |
7448 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
7449 | ||
7450 | /** | |
7451 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
7452 | * lower neighbour list, RCU | |
7453 | * variant | |
7454 | * @dev: device | |
7455 | * @iter: list_head ** of the current position | |
7456 | * | |
7457 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
7458 | * list, starting from iter position. The caller must hold RCU read lock. | |
7459 | */ | |
7460 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
7461 | struct list_head **iter) | |
7462 | { | |
7463 | struct netdev_adjacent *lower; | |
7464 | ||
7465 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
7466 | ||
7467 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7468 | ||
7469 | if (&lower->list == &dev->adj_list.lower) | |
7470 | return NULL; | |
7471 | ||
6859e7df | 7472 | *iter = &lower->list; |
31088a11 VF |
7473 | |
7474 | return lower->private; | |
7475 | } | |
7476 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
7477 | ||
4085ebe8 VY |
7478 | /** |
7479 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
7480 | * list | |
7481 | * @dev: device | |
7482 | * @iter: list_head ** of the current position | |
7483 | * | |
7484 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
7485 | * list, starting from iter position. The caller must hold RTNL lock or | |
7486 | * its own locking that guarantees that the neighbour lower | |
b469139e | 7487 | * list will remain unchanged. |
4085ebe8 VY |
7488 | */ |
7489 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
7490 | { | |
7491 | struct netdev_adjacent *lower; | |
7492 | ||
cfdd28be | 7493 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
7494 | |
7495 | if (&lower->list == &dev->adj_list.lower) | |
7496 | return NULL; | |
7497 | ||
cfdd28be | 7498 | *iter = lower->list.next; |
4085ebe8 VY |
7499 | |
7500 | return lower->dev; | |
7501 | } | |
7502 | EXPORT_SYMBOL(netdev_lower_get_next); | |
7503 | ||
1a3f060c DA |
7504 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
7505 | struct list_head **iter) | |
7506 | { | |
7507 | struct netdev_adjacent *lower; | |
7508 | ||
46b5ab1a | 7509 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
7510 | |
7511 | if (&lower->list == &dev->adj_list.lower) | |
7512 | return NULL; | |
7513 | ||
46b5ab1a | 7514 | *iter = &lower->list; |
1a3f060c DA |
7515 | |
7516 | return lower->dev; | |
7517 | } | |
7518 | ||
32b6d34f TY |
7519 | static struct net_device *__netdev_next_lower_dev(struct net_device *dev, |
7520 | struct list_head **iter, | |
7521 | bool *ignore) | |
7522 | { | |
7523 | struct netdev_adjacent *lower; | |
7524 | ||
7525 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
7526 | ||
7527 | if (&lower->list == &dev->adj_list.lower) | |
7528 | return NULL; | |
7529 | ||
7530 | *iter = &lower->list; | |
7531 | *ignore = lower->ignore; | |
7532 | ||
7533 | return lower->dev; | |
7534 | } | |
7535 | ||
1a3f060c DA |
7536 | int netdev_walk_all_lower_dev(struct net_device *dev, |
7537 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7538 | struct netdev_nested_priv *priv), |
7539 | struct netdev_nested_priv *priv) | |
1a3f060c | 7540 | { |
5343da4c TY |
7541 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; |
7542 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7543 | int ret, cur = 0; | |
1a3f060c | 7544 | |
5343da4c TY |
7545 | now = dev; |
7546 | iter = &dev->adj_list.lower; | |
1a3f060c | 7547 | |
5343da4c TY |
7548 | while (1) { |
7549 | if (now != dev) { | |
eff74233 | 7550 | ret = fn(now, priv); |
5343da4c TY |
7551 | if (ret) |
7552 | return ret; | |
7553 | } | |
7554 | ||
7555 | next = NULL; | |
7556 | while (1) { | |
7557 | ldev = netdev_next_lower_dev(now, &iter); | |
7558 | if (!ldev) | |
7559 | break; | |
7560 | ||
7561 | next = ldev; | |
7562 | niter = &ldev->adj_list.lower; | |
7563 | dev_stack[cur] = now; | |
7564 | iter_stack[cur++] = iter; | |
7565 | break; | |
7566 | } | |
7567 | ||
7568 | if (!next) { | |
7569 | if (!cur) | |
7570 | return 0; | |
7571 | next = dev_stack[--cur]; | |
7572 | niter = iter_stack[cur]; | |
7573 | } | |
7574 | ||
7575 | now = next; | |
7576 | iter = niter; | |
1a3f060c DA |
7577 | } |
7578 | ||
7579 | return 0; | |
7580 | } | |
7581 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
7582 | ||
32b6d34f TY |
7583 | static int __netdev_walk_all_lower_dev(struct net_device *dev, |
7584 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7585 | struct netdev_nested_priv *priv), |
7586 | struct netdev_nested_priv *priv) | |
32b6d34f TY |
7587 | { |
7588 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7589 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7590 | int ret, cur = 0; | |
7591 | bool ignore; | |
7592 | ||
7593 | now = dev; | |
7594 | iter = &dev->adj_list.lower; | |
7595 | ||
7596 | while (1) { | |
7597 | if (now != dev) { | |
eff74233 | 7598 | ret = fn(now, priv); |
32b6d34f TY |
7599 | if (ret) |
7600 | return ret; | |
7601 | } | |
7602 | ||
7603 | next = NULL; | |
7604 | while (1) { | |
7605 | ldev = __netdev_next_lower_dev(now, &iter, &ignore); | |
7606 | if (!ldev) | |
7607 | break; | |
7608 | if (ignore) | |
7609 | continue; | |
7610 | ||
7611 | next = ldev; | |
7612 | niter = &ldev->adj_list.lower; | |
7613 | dev_stack[cur] = now; | |
7614 | iter_stack[cur++] = iter; | |
7615 | break; | |
7616 | } | |
7617 | ||
7618 | if (!next) { | |
7619 | if (!cur) | |
7620 | return 0; | |
7621 | next = dev_stack[--cur]; | |
7622 | niter = iter_stack[cur]; | |
7623 | } | |
7624 | ||
7625 | now = next; | |
7626 | iter = niter; | |
7627 | } | |
7628 | ||
7629 | return 0; | |
7630 | } | |
7631 | ||
7151affe TY |
7632 | struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
7633 | struct list_head **iter) | |
1a3f060c DA |
7634 | { |
7635 | struct netdev_adjacent *lower; | |
7636 | ||
7637 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7638 | if (&lower->list == &dev->adj_list.lower) | |
7639 | return NULL; | |
7640 | ||
7641 | *iter = &lower->list; | |
7642 | ||
7643 | return lower->dev; | |
7644 | } | |
7151affe | 7645 | EXPORT_SYMBOL(netdev_next_lower_dev_rcu); |
1a3f060c | 7646 | |
5343da4c TY |
7647 | static u8 __netdev_upper_depth(struct net_device *dev) |
7648 | { | |
7649 | struct net_device *udev; | |
7650 | struct list_head *iter; | |
7651 | u8 max_depth = 0; | |
32b6d34f | 7652 | bool ignore; |
5343da4c TY |
7653 | |
7654 | for (iter = &dev->adj_list.upper, | |
32b6d34f | 7655 | udev = __netdev_next_upper_dev(dev, &iter, &ignore); |
5343da4c | 7656 | udev; |
32b6d34f TY |
7657 | udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { |
7658 | if (ignore) | |
7659 | continue; | |
5343da4c TY |
7660 | if (max_depth < udev->upper_level) |
7661 | max_depth = udev->upper_level; | |
7662 | } | |
7663 | ||
7664 | return max_depth; | |
7665 | } | |
7666 | ||
7667 | static u8 __netdev_lower_depth(struct net_device *dev) | |
1a3f060c DA |
7668 | { |
7669 | struct net_device *ldev; | |
7670 | struct list_head *iter; | |
5343da4c | 7671 | u8 max_depth = 0; |
32b6d34f | 7672 | bool ignore; |
1a3f060c DA |
7673 | |
7674 | for (iter = &dev->adj_list.lower, | |
32b6d34f | 7675 | ldev = __netdev_next_lower_dev(dev, &iter, &ignore); |
1a3f060c | 7676 | ldev; |
32b6d34f TY |
7677 | ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { |
7678 | if (ignore) | |
7679 | continue; | |
5343da4c TY |
7680 | if (max_depth < ldev->lower_level) |
7681 | max_depth = ldev->lower_level; | |
7682 | } | |
1a3f060c | 7683 | |
5343da4c TY |
7684 | return max_depth; |
7685 | } | |
7686 | ||
eff74233 TY |
7687 | static int __netdev_update_upper_level(struct net_device *dev, |
7688 | struct netdev_nested_priv *__unused) | |
5343da4c TY |
7689 | { |
7690 | dev->upper_level = __netdev_upper_depth(dev) + 1; | |
7691 | return 0; | |
7692 | } | |
7693 | ||
eff74233 | 7694 | static int __netdev_update_lower_level(struct net_device *dev, |
1fc70edb | 7695 | struct netdev_nested_priv *priv) |
5343da4c TY |
7696 | { |
7697 | dev->lower_level = __netdev_lower_depth(dev) + 1; | |
1fc70edb TY |
7698 | |
7699 | #ifdef CONFIG_LOCKDEP | |
7700 | if (!priv) | |
7701 | return 0; | |
7702 | ||
7703 | if (priv->flags & NESTED_SYNC_IMM) | |
7704 | dev->nested_level = dev->lower_level - 1; | |
7705 | if (priv->flags & NESTED_SYNC_TODO) | |
7706 | net_unlink_todo(dev); | |
7707 | #endif | |
5343da4c TY |
7708 | return 0; |
7709 | } | |
7710 | ||
7711 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
7712 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7713 | struct netdev_nested_priv *priv), |
7714 | struct netdev_nested_priv *priv) | |
5343da4c TY |
7715 | { |
7716 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7717 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7718 | int ret, cur = 0; | |
7719 | ||
7720 | now = dev; | |
7721 | iter = &dev->adj_list.lower; | |
7722 | ||
7723 | while (1) { | |
7724 | if (now != dev) { | |
eff74233 | 7725 | ret = fn(now, priv); |
5343da4c TY |
7726 | if (ret) |
7727 | return ret; | |
7728 | } | |
7729 | ||
7730 | next = NULL; | |
7731 | while (1) { | |
7732 | ldev = netdev_next_lower_dev_rcu(now, &iter); | |
7733 | if (!ldev) | |
7734 | break; | |
7735 | ||
7736 | next = ldev; | |
7737 | niter = &ldev->adj_list.lower; | |
7738 | dev_stack[cur] = now; | |
7739 | iter_stack[cur++] = iter; | |
7740 | break; | |
7741 | } | |
7742 | ||
7743 | if (!next) { | |
7744 | if (!cur) | |
7745 | return 0; | |
7746 | next = dev_stack[--cur]; | |
7747 | niter = iter_stack[cur]; | |
7748 | } | |
7749 | ||
7750 | now = next; | |
7751 | iter = niter; | |
1a3f060c DA |
7752 | } |
7753 | ||
7754 | return 0; | |
7755 | } | |
7756 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
7757 | ||
e001bfad | 7758 | /** |
7759 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
7760 | * lower neighbour list, RCU | |
7761 | * variant | |
7762 | * @dev: device | |
7763 | * | |
7764 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
7765 | * list. The caller must hold RCU read lock. | |
7766 | */ | |
7767 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
7768 | { | |
7769 | struct netdev_adjacent *lower; | |
7770 | ||
7771 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
7772 | struct netdev_adjacent, list); | |
7773 | if (lower) | |
7774 | return lower->private; | |
7775 | return NULL; | |
7776 | } | |
7777 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
7778 | ||
9ff162a8 JP |
7779 | /** |
7780 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
7781 | * @dev: device | |
7782 | * | |
7783 | * Find a master upper device and return pointer to it or NULL in case | |
7784 | * it's not there. The caller must hold the RCU read lock. | |
7785 | */ | |
7786 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
7787 | { | |
aa9d8560 | 7788 | struct netdev_adjacent *upper; |
9ff162a8 | 7789 | |
2f268f12 | 7790 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 7791 | struct netdev_adjacent, list); |
9ff162a8 JP |
7792 | if (upper && likely(upper->master)) |
7793 | return upper->dev; | |
7794 | return NULL; | |
7795 | } | |
7796 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
7797 | ||
0a59f3a9 | 7798 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
7799 | struct net_device *adj_dev, |
7800 | struct list_head *dev_list) | |
7801 | { | |
7802 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 7803 | |
3ee32707 VF |
7804 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
7805 | "upper_%s" : "lower_%s", adj_dev->name); | |
7806 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
7807 | linkname); | |
7808 | } | |
0a59f3a9 | 7809 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
7810 | char *name, |
7811 | struct list_head *dev_list) | |
7812 | { | |
7813 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 7814 | |
3ee32707 VF |
7815 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
7816 | "upper_%s" : "lower_%s", name); | |
7817 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
7818 | } | |
7819 | ||
7ce64c79 AF |
7820 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
7821 | struct net_device *adj_dev, | |
7822 | struct list_head *dev_list) | |
7823 | { | |
7824 | return (dev_list == &dev->adj_list.upper || | |
7825 | dev_list == &dev->adj_list.lower) && | |
7826 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
7827 | } | |
3ee32707 | 7828 | |
5d261913 VF |
7829 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
7830 | struct net_device *adj_dev, | |
7863c054 | 7831 | struct list_head *dev_list, |
402dae96 | 7832 | void *private, bool master) |
5d261913 VF |
7833 | { |
7834 | struct netdev_adjacent *adj; | |
842d67a7 | 7835 | int ret; |
5d261913 | 7836 | |
6ea29da1 | 7837 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
7838 | |
7839 | if (adj) { | |
790510d9 | 7840 | adj->ref_nr += 1; |
67b62f98 DA |
7841 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
7842 | dev->name, adj_dev->name, adj->ref_nr); | |
7843 | ||
5d261913 VF |
7844 | return 0; |
7845 | } | |
7846 | ||
7847 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
7848 | if (!adj) | |
7849 | return -ENOMEM; | |
7850 | ||
7851 | adj->dev = adj_dev; | |
7852 | adj->master = master; | |
790510d9 | 7853 | adj->ref_nr = 1; |
402dae96 | 7854 | adj->private = private; |
32b6d34f | 7855 | adj->ignore = false; |
5d261913 | 7856 | dev_hold(adj_dev); |
2f268f12 | 7857 | |
67b62f98 DA |
7858 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
7859 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 7860 | |
7ce64c79 | 7861 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 7862 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
7863 | if (ret) |
7864 | goto free_adj; | |
7865 | } | |
7866 | ||
7863c054 | 7867 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
7868 | if (master) { |
7869 | ret = sysfs_create_link(&(dev->dev.kobj), | |
7870 | &(adj_dev->dev.kobj), "master"); | |
7871 | if (ret) | |
5831d66e | 7872 | goto remove_symlinks; |
842d67a7 | 7873 | |
7863c054 | 7874 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 7875 | } else { |
7863c054 | 7876 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 7877 | } |
5d261913 VF |
7878 | |
7879 | return 0; | |
842d67a7 | 7880 | |
5831d66e | 7881 | remove_symlinks: |
7ce64c79 | 7882 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 7883 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
7884 | free_adj: |
7885 | kfree(adj); | |
974daef7 | 7886 | dev_put(adj_dev); |
842d67a7 VF |
7887 | |
7888 | return ret; | |
5d261913 VF |
7889 | } |
7890 | ||
1d143d9f | 7891 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
7892 | struct net_device *adj_dev, | |
93409033 | 7893 | u16 ref_nr, |
1d143d9f | 7894 | struct list_head *dev_list) |
5d261913 VF |
7895 | { |
7896 | struct netdev_adjacent *adj; | |
7897 | ||
67b62f98 DA |
7898 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
7899 | dev->name, adj_dev->name, ref_nr); | |
7900 | ||
6ea29da1 | 7901 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 7902 | |
2f268f12 | 7903 | if (!adj) { |
67b62f98 | 7904 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 7905 | dev->name, adj_dev->name); |
67b62f98 DA |
7906 | WARN_ON(1); |
7907 | return; | |
2f268f12 | 7908 | } |
5d261913 | 7909 | |
93409033 | 7910 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
7911 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
7912 | dev->name, adj_dev->name, ref_nr, | |
7913 | adj->ref_nr - ref_nr); | |
93409033 | 7914 | adj->ref_nr -= ref_nr; |
5d261913 VF |
7915 | return; |
7916 | } | |
7917 | ||
842d67a7 VF |
7918 | if (adj->master) |
7919 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
7920 | ||
7ce64c79 | 7921 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 7922 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 7923 | |
5d261913 | 7924 | list_del_rcu(&adj->list); |
67b62f98 | 7925 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 7926 | adj_dev->name, dev->name, adj_dev->name); |
5d261913 VF |
7927 | dev_put(adj_dev); |
7928 | kfree_rcu(adj, rcu); | |
7929 | } | |
7930 | ||
1d143d9f | 7931 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
7932 | struct net_device *upper_dev, | |
7933 | struct list_head *up_list, | |
7934 | struct list_head *down_list, | |
7935 | void *private, bool master) | |
5d261913 VF |
7936 | { |
7937 | int ret; | |
7938 | ||
790510d9 | 7939 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 7940 | private, master); |
5d261913 VF |
7941 | if (ret) |
7942 | return ret; | |
7943 | ||
790510d9 | 7944 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 7945 | private, false); |
5d261913 | 7946 | if (ret) { |
790510d9 | 7947 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
7948 | return ret; |
7949 | } | |
7950 | ||
7951 | return 0; | |
7952 | } | |
7953 | ||
1d143d9f | 7954 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
7955 | struct net_device *upper_dev, | |
93409033 | 7956 | u16 ref_nr, |
1d143d9f | 7957 | struct list_head *up_list, |
7958 | struct list_head *down_list) | |
5d261913 | 7959 | { |
93409033 AC |
7960 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
7961 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
7962 | } |
7963 | ||
1d143d9f | 7964 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
7965 | struct net_device *upper_dev, | |
7966 | void *private, bool master) | |
2f268f12 | 7967 | { |
f1170fd4 DA |
7968 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
7969 | &dev->adj_list.upper, | |
7970 | &upper_dev->adj_list.lower, | |
7971 | private, master); | |
5d261913 VF |
7972 | } |
7973 | ||
1d143d9f | 7974 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
7975 | struct net_device *upper_dev) | |
2f268f12 | 7976 | { |
93409033 | 7977 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
7978 | &dev->adj_list.upper, |
7979 | &upper_dev->adj_list.lower); | |
7980 | } | |
5d261913 | 7981 | |
9ff162a8 | 7982 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 7983 | struct net_device *upper_dev, bool master, |
42ab19ee | 7984 | void *upper_priv, void *upper_info, |
1fc70edb | 7985 | struct netdev_nested_priv *priv, |
42ab19ee | 7986 | struct netlink_ext_ack *extack) |
9ff162a8 | 7987 | { |
51d0c047 DA |
7988 | struct netdev_notifier_changeupper_info changeupper_info = { |
7989 | .info = { | |
7990 | .dev = dev, | |
42ab19ee | 7991 | .extack = extack, |
51d0c047 DA |
7992 | }, |
7993 | .upper_dev = upper_dev, | |
7994 | .master = master, | |
7995 | .linking = true, | |
7996 | .upper_info = upper_info, | |
7997 | }; | |
50d629e7 | 7998 | struct net_device *master_dev; |
5d261913 | 7999 | int ret = 0; |
9ff162a8 JP |
8000 | |
8001 | ASSERT_RTNL(); | |
8002 | ||
8003 | if (dev == upper_dev) | |
8004 | return -EBUSY; | |
8005 | ||
8006 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
32b6d34f | 8007 | if (__netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
8008 | return -EBUSY; |
8009 | ||
5343da4c TY |
8010 | if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) |
8011 | return -EMLINK; | |
8012 | ||
50d629e7 | 8013 | if (!master) { |
32b6d34f | 8014 | if (__netdev_has_upper_dev(dev, upper_dev)) |
50d629e7 MM |
8015 | return -EEXIST; |
8016 | } else { | |
32b6d34f | 8017 | master_dev = __netdev_master_upper_dev_get(dev); |
50d629e7 MM |
8018 | if (master_dev) |
8019 | return master_dev == upper_dev ? -EEXIST : -EBUSY; | |
8020 | } | |
9ff162a8 | 8021 | |
51d0c047 | 8022 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
8023 | &changeupper_info.info); |
8024 | ret = notifier_to_errno(ret); | |
8025 | if (ret) | |
8026 | return ret; | |
8027 | ||
6dffb044 | 8028 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 8029 | master); |
5d261913 VF |
8030 | if (ret) |
8031 | return ret; | |
9ff162a8 | 8032 | |
51d0c047 | 8033 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
b03804e7 IS |
8034 | &changeupper_info.info); |
8035 | ret = notifier_to_errno(ret); | |
8036 | if (ret) | |
f1170fd4 | 8037 | goto rollback; |
b03804e7 | 8038 | |
5343da4c | 8039 | __netdev_update_upper_level(dev, NULL); |
32b6d34f | 8040 | __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); |
5343da4c | 8041 | |
1fc70edb | 8042 | __netdev_update_lower_level(upper_dev, priv); |
32b6d34f | 8043 | __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, |
1fc70edb | 8044 | priv); |
5343da4c | 8045 | |
9ff162a8 | 8046 | return 0; |
5d261913 | 8047 | |
f1170fd4 | 8048 | rollback: |
2f268f12 | 8049 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
8050 | |
8051 | return ret; | |
9ff162a8 JP |
8052 | } |
8053 | ||
8054 | /** | |
8055 | * netdev_upper_dev_link - Add a link to the upper device | |
8056 | * @dev: device | |
8057 | * @upper_dev: new upper device | |
7a006d59 | 8058 | * @extack: netlink extended ack |
9ff162a8 JP |
8059 | * |
8060 | * Adds a link to device which is upper to this one. The caller must hold | |
8061 | * the RTNL lock. On a failure a negative errno code is returned. | |
8062 | * On success the reference counts are adjusted and the function | |
8063 | * returns zero. | |
8064 | */ | |
8065 | int netdev_upper_dev_link(struct net_device *dev, | |
42ab19ee DA |
8066 | struct net_device *upper_dev, |
8067 | struct netlink_ext_ack *extack) | |
9ff162a8 | 8068 | { |
1fc70edb TY |
8069 | struct netdev_nested_priv priv = { |
8070 | .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, | |
8071 | .data = NULL, | |
8072 | }; | |
8073 | ||
42ab19ee | 8074 | return __netdev_upper_dev_link(dev, upper_dev, false, |
1fc70edb | 8075 | NULL, NULL, &priv, extack); |
9ff162a8 JP |
8076 | } |
8077 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
8078 | ||
8079 | /** | |
8080 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
8081 | * @dev: device | |
8082 | * @upper_dev: new upper device | |
6dffb044 | 8083 | * @upper_priv: upper device private |
29bf24af | 8084 | * @upper_info: upper info to be passed down via notifier |
7a006d59 | 8085 | * @extack: netlink extended ack |
9ff162a8 JP |
8086 | * |
8087 | * Adds a link to device which is upper to this one. In this case, only | |
8088 | * one master upper device can be linked, although other non-master devices | |
8089 | * might be linked as well. The caller must hold the RTNL lock. | |
8090 | * On a failure a negative errno code is returned. On success the reference | |
8091 | * counts are adjusted and the function returns zero. | |
8092 | */ | |
8093 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 8094 | struct net_device *upper_dev, |
42ab19ee DA |
8095 | void *upper_priv, void *upper_info, |
8096 | struct netlink_ext_ack *extack) | |
9ff162a8 | 8097 | { |
1fc70edb TY |
8098 | struct netdev_nested_priv priv = { |
8099 | .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, | |
8100 | .data = NULL, | |
8101 | }; | |
8102 | ||
29bf24af | 8103 | return __netdev_upper_dev_link(dev, upper_dev, true, |
1fc70edb | 8104 | upper_priv, upper_info, &priv, extack); |
9ff162a8 JP |
8105 | } |
8106 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
8107 | ||
fe8300fd | 8108 | static void __netdev_upper_dev_unlink(struct net_device *dev, |
1fc70edb TY |
8109 | struct net_device *upper_dev, |
8110 | struct netdev_nested_priv *priv) | |
9ff162a8 | 8111 | { |
51d0c047 DA |
8112 | struct netdev_notifier_changeupper_info changeupper_info = { |
8113 | .info = { | |
8114 | .dev = dev, | |
8115 | }, | |
8116 | .upper_dev = upper_dev, | |
8117 | .linking = false, | |
8118 | }; | |
f4563a75 | 8119 | |
9ff162a8 JP |
8120 | ASSERT_RTNL(); |
8121 | ||
0e4ead9d | 8122 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; |
0e4ead9d | 8123 | |
51d0c047 | 8124 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
8125 | &changeupper_info.info); |
8126 | ||
2f268f12 | 8127 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 8128 | |
51d0c047 | 8129 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
0e4ead9d | 8130 | &changeupper_info.info); |
5343da4c TY |
8131 | |
8132 | __netdev_update_upper_level(dev, NULL); | |
32b6d34f | 8133 | __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); |
5343da4c | 8134 | |
1fc70edb | 8135 | __netdev_update_lower_level(upper_dev, priv); |
32b6d34f | 8136 | __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, |
1fc70edb | 8137 | priv); |
9ff162a8 | 8138 | } |
fe8300fd TY |
8139 | |
8140 | /** | |
8141 | * netdev_upper_dev_unlink - Removes a link to upper device | |
8142 | * @dev: device | |
8143 | * @upper_dev: new upper device | |
8144 | * | |
8145 | * Removes a link to device which is upper to this one. The caller must hold | |
8146 | * the RTNL lock. | |
8147 | */ | |
8148 | void netdev_upper_dev_unlink(struct net_device *dev, | |
8149 | struct net_device *upper_dev) | |
8150 | { | |
1fc70edb TY |
8151 | struct netdev_nested_priv priv = { |
8152 | .flags = NESTED_SYNC_TODO, | |
8153 | .data = NULL, | |
8154 | }; | |
8155 | ||
8156 | __netdev_upper_dev_unlink(dev, upper_dev, &priv); | |
9ff162a8 JP |
8157 | } |
8158 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
8159 | ||
32b6d34f TY |
8160 | static void __netdev_adjacent_dev_set(struct net_device *upper_dev, |
8161 | struct net_device *lower_dev, | |
8162 | bool val) | |
8163 | { | |
8164 | struct netdev_adjacent *adj; | |
8165 | ||
8166 | adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); | |
8167 | if (adj) | |
8168 | adj->ignore = val; | |
8169 | ||
8170 | adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); | |
8171 | if (adj) | |
8172 | adj->ignore = val; | |
8173 | } | |
8174 | ||
8175 | static void netdev_adjacent_dev_disable(struct net_device *upper_dev, | |
8176 | struct net_device *lower_dev) | |
8177 | { | |
8178 | __netdev_adjacent_dev_set(upper_dev, lower_dev, true); | |
8179 | } | |
8180 | ||
8181 | static void netdev_adjacent_dev_enable(struct net_device *upper_dev, | |
8182 | struct net_device *lower_dev) | |
8183 | { | |
8184 | __netdev_adjacent_dev_set(upper_dev, lower_dev, false); | |
8185 | } | |
8186 | ||
8187 | int netdev_adjacent_change_prepare(struct net_device *old_dev, | |
8188 | struct net_device *new_dev, | |
8189 | struct net_device *dev, | |
8190 | struct netlink_ext_ack *extack) | |
8191 | { | |
1fc70edb TY |
8192 | struct netdev_nested_priv priv = { |
8193 | .flags = 0, | |
8194 | .data = NULL, | |
8195 | }; | |
32b6d34f TY |
8196 | int err; |
8197 | ||
8198 | if (!new_dev) | |
8199 | return 0; | |
8200 | ||
8201 | if (old_dev && new_dev != old_dev) | |
8202 | netdev_adjacent_dev_disable(dev, old_dev); | |
1fc70edb TY |
8203 | err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, |
8204 | extack); | |
32b6d34f TY |
8205 | if (err) { |
8206 | if (old_dev && new_dev != old_dev) | |
8207 | netdev_adjacent_dev_enable(dev, old_dev); | |
8208 | return err; | |
8209 | } | |
8210 | ||
8211 | return 0; | |
8212 | } | |
8213 | EXPORT_SYMBOL(netdev_adjacent_change_prepare); | |
8214 | ||
8215 | void netdev_adjacent_change_commit(struct net_device *old_dev, | |
8216 | struct net_device *new_dev, | |
8217 | struct net_device *dev) | |
8218 | { | |
1fc70edb TY |
8219 | struct netdev_nested_priv priv = { |
8220 | .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, | |
8221 | .data = NULL, | |
8222 | }; | |
8223 | ||
32b6d34f TY |
8224 | if (!new_dev || !old_dev) |
8225 | return; | |
8226 | ||
8227 | if (new_dev == old_dev) | |
8228 | return; | |
8229 | ||
8230 | netdev_adjacent_dev_enable(dev, old_dev); | |
1fc70edb | 8231 | __netdev_upper_dev_unlink(old_dev, dev, &priv); |
32b6d34f TY |
8232 | } |
8233 | EXPORT_SYMBOL(netdev_adjacent_change_commit); | |
8234 | ||
8235 | void netdev_adjacent_change_abort(struct net_device *old_dev, | |
8236 | struct net_device *new_dev, | |
8237 | struct net_device *dev) | |
8238 | { | |
1fc70edb TY |
8239 | struct netdev_nested_priv priv = { |
8240 | .flags = 0, | |
8241 | .data = NULL, | |
8242 | }; | |
8243 | ||
32b6d34f TY |
8244 | if (!new_dev) |
8245 | return; | |
8246 | ||
8247 | if (old_dev && new_dev != old_dev) | |
8248 | netdev_adjacent_dev_enable(dev, old_dev); | |
8249 | ||
1fc70edb | 8250 | __netdev_upper_dev_unlink(new_dev, dev, &priv); |
32b6d34f TY |
8251 | } |
8252 | EXPORT_SYMBOL(netdev_adjacent_change_abort); | |
8253 | ||
61bd3857 MS |
8254 | /** |
8255 | * netdev_bonding_info_change - Dispatch event about slave change | |
8256 | * @dev: device | |
4a26e453 | 8257 | * @bonding_info: info to dispatch |
61bd3857 MS |
8258 | * |
8259 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
8260 | * The caller must hold the RTNL lock. | |
8261 | */ | |
8262 | void netdev_bonding_info_change(struct net_device *dev, | |
8263 | struct netdev_bonding_info *bonding_info) | |
8264 | { | |
51d0c047 DA |
8265 | struct netdev_notifier_bonding_info info = { |
8266 | .info.dev = dev, | |
8267 | }; | |
61bd3857 MS |
8268 | |
8269 | memcpy(&info.bonding_info, bonding_info, | |
8270 | sizeof(struct netdev_bonding_info)); | |
51d0c047 | 8271 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, |
61bd3857 MS |
8272 | &info.info); |
8273 | } | |
8274 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
8275 | ||
cff9f12b MG |
8276 | /** |
8277 | * netdev_get_xmit_slave - Get the xmit slave of master device | |
8842500d | 8278 | * @dev: device |
cff9f12b MG |
8279 | * @skb: The packet |
8280 | * @all_slaves: assume all the slaves are active | |
8281 | * | |
8282 | * The reference counters are not incremented so the caller must be | |
8283 | * careful with locks. The caller must hold RCU lock. | |
8284 | * %NULL is returned if no slave is found. | |
8285 | */ | |
8286 | ||
8287 | struct net_device *netdev_get_xmit_slave(struct net_device *dev, | |
8288 | struct sk_buff *skb, | |
8289 | bool all_slaves) | |
8290 | { | |
8291 | const struct net_device_ops *ops = dev->netdev_ops; | |
8292 | ||
8293 | if (!ops->ndo_get_xmit_slave) | |
8294 | return NULL; | |
8295 | return ops->ndo_get_xmit_slave(dev, skb, all_slaves); | |
8296 | } | |
8297 | EXPORT_SYMBOL(netdev_get_xmit_slave); | |
8298 | ||
719a402c TT |
8299 | static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, |
8300 | struct sock *sk) | |
8301 | { | |
8302 | const struct net_device_ops *ops = dev->netdev_ops; | |
8303 | ||
8304 | if (!ops->ndo_sk_get_lower_dev) | |
8305 | return NULL; | |
8306 | return ops->ndo_sk_get_lower_dev(dev, sk); | |
8307 | } | |
8308 | ||
8309 | /** | |
8310 | * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket | |
8311 | * @dev: device | |
8312 | * @sk: the socket | |
8313 | * | |
8314 | * %NULL is returned if no lower device is found. | |
8315 | */ | |
8316 | ||
8317 | struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, | |
8318 | struct sock *sk) | |
8319 | { | |
8320 | struct net_device *lower; | |
8321 | ||
8322 | lower = netdev_sk_get_lower_dev(dev, sk); | |
8323 | while (lower) { | |
8324 | dev = lower; | |
8325 | lower = netdev_sk_get_lower_dev(dev, sk); | |
8326 | } | |
8327 | ||
8328 | return dev; | |
8329 | } | |
8330 | EXPORT_SYMBOL(netdev_sk_get_lowest_dev); | |
8331 | ||
2ce1ee17 | 8332 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
8333 | { |
8334 | struct netdev_adjacent *iter; | |
8335 | ||
8336 | struct net *net = dev_net(dev); | |
8337 | ||
8338 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 8339 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
8340 | continue; |
8341 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
8342 | &iter->dev->adj_list.lower); | |
8343 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
8344 | &dev->adj_list.upper); | |
8345 | } | |
8346 | ||
8347 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 8348 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
8349 | continue; |
8350 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
8351 | &iter->dev->adj_list.upper); | |
8352 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
8353 | &dev->adj_list.lower); | |
8354 | } | |
8355 | } | |
8356 | ||
2ce1ee17 | 8357 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
8358 | { |
8359 | struct netdev_adjacent *iter; | |
8360 | ||
8361 | struct net *net = dev_net(dev); | |
8362 | ||
8363 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 8364 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
8365 | continue; |
8366 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
8367 | &iter->dev->adj_list.lower); | |
8368 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
8369 | &dev->adj_list.upper); | |
8370 | } | |
8371 | ||
8372 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 8373 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
8374 | continue; |
8375 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
8376 | &iter->dev->adj_list.upper); | |
8377 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
8378 | &dev->adj_list.lower); | |
8379 | } | |
8380 | } | |
8381 | ||
5bb025fa | 8382 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 8383 | { |
5bb025fa | 8384 | struct netdev_adjacent *iter; |
402dae96 | 8385 | |
4c75431a AF |
8386 | struct net *net = dev_net(dev); |
8387 | ||
5bb025fa | 8388 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 8389 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 8390 | continue; |
5bb025fa VF |
8391 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
8392 | &iter->dev->adj_list.lower); | |
8393 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
8394 | &iter->dev->adj_list.lower); | |
8395 | } | |
402dae96 | 8396 | |
5bb025fa | 8397 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 8398 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 8399 | continue; |
5bb025fa VF |
8400 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
8401 | &iter->dev->adj_list.upper); | |
8402 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
8403 | &iter->dev->adj_list.upper); | |
8404 | } | |
402dae96 | 8405 | } |
402dae96 VF |
8406 | |
8407 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
8408 | struct net_device *lower_dev) | |
8409 | { | |
8410 | struct netdev_adjacent *lower; | |
8411 | ||
8412 | if (!lower_dev) | |
8413 | return NULL; | |
6ea29da1 | 8414 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
8415 | if (!lower) |
8416 | return NULL; | |
8417 | ||
8418 | return lower->private; | |
8419 | } | |
8420 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
8421 | ||
4085ebe8 | 8422 | |
04d48266 | 8423 | /** |
c1639be9 | 8424 | * netdev_lower_state_changed - Dispatch event about lower device state change |
04d48266 JP |
8425 | * @lower_dev: device |
8426 | * @lower_state_info: state to dispatch | |
8427 | * | |
8428 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
8429 | * The caller must hold the RTNL lock. | |
8430 | */ | |
8431 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
8432 | void *lower_state_info) | |
8433 | { | |
51d0c047 DA |
8434 | struct netdev_notifier_changelowerstate_info changelowerstate_info = { |
8435 | .info.dev = lower_dev, | |
8436 | }; | |
04d48266 JP |
8437 | |
8438 | ASSERT_RTNL(); | |
8439 | changelowerstate_info.lower_state_info = lower_state_info; | |
51d0c047 | 8440 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, |
04d48266 JP |
8441 | &changelowerstate_info.info); |
8442 | } | |
8443 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
8444 | ||
b6c40d68 PM |
8445 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
8446 | { | |
d314774c SH |
8447 | const struct net_device_ops *ops = dev->netdev_ops; |
8448 | ||
d2615bf4 | 8449 | if (ops->ndo_change_rx_flags) |
d314774c | 8450 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
8451 | } |
8452 | ||
991fb3f7 | 8453 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 8454 | { |
b536db93 | 8455 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
8456 | kuid_t uid; |
8457 | kgid_t gid; | |
1da177e4 | 8458 | |
24023451 PM |
8459 | ASSERT_RTNL(); |
8460 | ||
dad9b335 WC |
8461 | dev->flags |= IFF_PROMISC; |
8462 | dev->promiscuity += inc; | |
8463 | if (dev->promiscuity == 0) { | |
8464 | /* | |
8465 | * Avoid overflow. | |
8466 | * If inc causes overflow, untouch promisc and return error. | |
8467 | */ | |
8468 | if (inc < 0) | |
8469 | dev->flags &= ~IFF_PROMISC; | |
8470 | else { | |
8471 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
8472 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
8473 | dev->name); | |
dad9b335 WC |
8474 | return -EOVERFLOW; |
8475 | } | |
8476 | } | |
52609c0b | 8477 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
8478 | pr_info("device %s %s promiscuous mode\n", |
8479 | dev->name, | |
8480 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
8481 | if (audit_enabled) { |
8482 | current_uid_gid(&uid, &gid); | |
cdfb6b34 RGB |
8483 | audit_log(audit_context(), GFP_ATOMIC, |
8484 | AUDIT_ANOM_PROMISCUOUS, | |
8485 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
8486 | dev->name, (dev->flags & IFF_PROMISC), | |
8487 | (old_flags & IFF_PROMISC), | |
8488 | from_kuid(&init_user_ns, audit_get_loginuid(current)), | |
8489 | from_kuid(&init_user_ns, uid), | |
8490 | from_kgid(&init_user_ns, gid), | |
8491 | audit_get_sessionid(current)); | |
8192b0c4 | 8492 | } |
24023451 | 8493 | |
b6c40d68 | 8494 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 8495 | } |
991fb3f7 ND |
8496 | if (notify) |
8497 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 8498 | return 0; |
1da177e4 LT |
8499 | } |
8500 | ||
4417da66 PM |
8501 | /** |
8502 | * dev_set_promiscuity - update promiscuity count on a device | |
8503 | * @dev: device | |
8504 | * @inc: modifier | |
8505 | * | |
8506 | * Add or remove promiscuity from a device. While the count in the device | |
8507 | * remains above zero the interface remains promiscuous. Once it hits zero | |
8508 | * the device reverts back to normal filtering operation. A negative inc | |
8509 | * value is used to drop promiscuity on the device. | |
dad9b335 | 8510 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 8511 | */ |
dad9b335 | 8512 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 8513 | { |
b536db93 | 8514 | unsigned int old_flags = dev->flags; |
dad9b335 | 8515 | int err; |
4417da66 | 8516 | |
991fb3f7 | 8517 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 8518 | if (err < 0) |
dad9b335 | 8519 | return err; |
4417da66 PM |
8520 | if (dev->flags != old_flags) |
8521 | dev_set_rx_mode(dev); | |
dad9b335 | 8522 | return err; |
4417da66 | 8523 | } |
d1b19dff | 8524 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 8525 | |
991fb3f7 | 8526 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 8527 | { |
991fb3f7 | 8528 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 8529 | |
24023451 PM |
8530 | ASSERT_RTNL(); |
8531 | ||
1da177e4 | 8532 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
8533 | dev->allmulti += inc; |
8534 | if (dev->allmulti == 0) { | |
8535 | /* | |
8536 | * Avoid overflow. | |
8537 | * If inc causes overflow, untouch allmulti and return error. | |
8538 | */ | |
8539 | if (inc < 0) | |
8540 | dev->flags &= ~IFF_ALLMULTI; | |
8541 | else { | |
8542 | dev->allmulti -= inc; | |
7b6cd1ce JP |
8543 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
8544 | dev->name); | |
dad9b335 WC |
8545 | return -EOVERFLOW; |
8546 | } | |
8547 | } | |
24023451 | 8548 | if (dev->flags ^ old_flags) { |
b6c40d68 | 8549 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 8550 | dev_set_rx_mode(dev); |
991fb3f7 ND |
8551 | if (notify) |
8552 | __dev_notify_flags(dev, old_flags, | |
8553 | dev->gflags ^ old_gflags); | |
24023451 | 8554 | } |
dad9b335 | 8555 | return 0; |
4417da66 | 8556 | } |
991fb3f7 ND |
8557 | |
8558 | /** | |
8559 | * dev_set_allmulti - update allmulti count on a device | |
8560 | * @dev: device | |
8561 | * @inc: modifier | |
8562 | * | |
8563 | * Add or remove reception of all multicast frames to a device. While the | |
8564 | * count in the device remains above zero the interface remains listening | |
8565 | * to all interfaces. Once it hits zero the device reverts back to normal | |
8566 | * filtering operation. A negative @inc value is used to drop the counter | |
8567 | * when releasing a resource needing all multicasts. | |
8568 | * Return 0 if successful or a negative errno code on error. | |
8569 | */ | |
8570 | ||
8571 | int dev_set_allmulti(struct net_device *dev, int inc) | |
8572 | { | |
8573 | return __dev_set_allmulti(dev, inc, true); | |
8574 | } | |
d1b19dff | 8575 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
8576 | |
8577 | /* | |
8578 | * Upload unicast and multicast address lists to device and | |
8579 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 8580 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
8581 | * are present. |
8582 | */ | |
8583 | void __dev_set_rx_mode(struct net_device *dev) | |
8584 | { | |
d314774c SH |
8585 | const struct net_device_ops *ops = dev->netdev_ops; |
8586 | ||
4417da66 PM |
8587 | /* dev_open will call this function so the list will stay sane. */ |
8588 | if (!(dev->flags&IFF_UP)) | |
8589 | return; | |
8590 | ||
8591 | if (!netif_device_present(dev)) | |
40b77c94 | 8592 | return; |
4417da66 | 8593 | |
01789349 | 8594 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
8595 | /* Unicast addresses changes may only happen under the rtnl, |
8596 | * therefore calling __dev_set_promiscuity here is safe. | |
8597 | */ | |
32e7bfc4 | 8598 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 8599 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 8600 | dev->uc_promisc = true; |
32e7bfc4 | 8601 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 8602 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 8603 | dev->uc_promisc = false; |
4417da66 | 8604 | } |
4417da66 | 8605 | } |
01789349 JP |
8606 | |
8607 | if (ops->ndo_set_rx_mode) | |
8608 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
8609 | } |
8610 | ||
8611 | void dev_set_rx_mode(struct net_device *dev) | |
8612 | { | |
b9e40857 | 8613 | netif_addr_lock_bh(dev); |
4417da66 | 8614 | __dev_set_rx_mode(dev); |
b9e40857 | 8615 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
8616 | } |
8617 | ||
f0db275a SH |
8618 | /** |
8619 | * dev_get_flags - get flags reported to userspace | |
8620 | * @dev: device | |
8621 | * | |
8622 | * Get the combination of flag bits exported through APIs to userspace. | |
8623 | */ | |
95c96174 | 8624 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 8625 | { |
95c96174 | 8626 | unsigned int flags; |
1da177e4 LT |
8627 | |
8628 | flags = (dev->flags & ~(IFF_PROMISC | | |
8629 | IFF_ALLMULTI | | |
b00055aa SR |
8630 | IFF_RUNNING | |
8631 | IFF_LOWER_UP | | |
8632 | IFF_DORMANT)) | | |
1da177e4 LT |
8633 | (dev->gflags & (IFF_PROMISC | |
8634 | IFF_ALLMULTI)); | |
8635 | ||
b00055aa SR |
8636 | if (netif_running(dev)) { |
8637 | if (netif_oper_up(dev)) | |
8638 | flags |= IFF_RUNNING; | |
8639 | if (netif_carrier_ok(dev)) | |
8640 | flags |= IFF_LOWER_UP; | |
8641 | if (netif_dormant(dev)) | |
8642 | flags |= IFF_DORMANT; | |
8643 | } | |
1da177e4 LT |
8644 | |
8645 | return flags; | |
8646 | } | |
d1b19dff | 8647 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 8648 | |
6d040321 PM |
8649 | int __dev_change_flags(struct net_device *dev, unsigned int flags, |
8650 | struct netlink_ext_ack *extack) | |
1da177e4 | 8651 | { |
b536db93 | 8652 | unsigned int old_flags = dev->flags; |
bd380811 | 8653 | int ret; |
1da177e4 | 8654 | |
24023451 PM |
8655 | ASSERT_RTNL(); |
8656 | ||
1da177e4 LT |
8657 | /* |
8658 | * Set the flags on our device. | |
8659 | */ | |
8660 | ||
8661 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
8662 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
8663 | IFF_AUTOMEDIA)) | | |
8664 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
8665 | IFF_ALLMULTI)); | |
8666 | ||
8667 | /* | |
8668 | * Load in the correct multicast list now the flags have changed. | |
8669 | */ | |
8670 | ||
b6c40d68 PM |
8671 | if ((old_flags ^ flags) & IFF_MULTICAST) |
8672 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 8673 | |
4417da66 | 8674 | dev_set_rx_mode(dev); |
1da177e4 LT |
8675 | |
8676 | /* | |
8677 | * Have we downed the interface. We handle IFF_UP ourselves | |
8678 | * according to user attempts to set it, rather than blindly | |
8679 | * setting it. | |
8680 | */ | |
8681 | ||
8682 | ret = 0; | |
7051b88a | 8683 | if ((old_flags ^ flags) & IFF_UP) { |
8684 | if (old_flags & IFF_UP) | |
8685 | __dev_close(dev); | |
8686 | else | |
40c900aa | 8687 | ret = __dev_open(dev, extack); |
7051b88a | 8688 | } |
1da177e4 | 8689 | |
1da177e4 | 8690 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 8691 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 8692 | unsigned int old_flags = dev->flags; |
d1b19dff | 8693 | |
1da177e4 | 8694 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
8695 | |
8696 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
8697 | if (dev->flags != old_flags) | |
8698 | dev_set_rx_mode(dev); | |
1da177e4 LT |
8699 | } |
8700 | ||
8701 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 8702 | * is important. Some (broken) drivers set IFF_PROMISC, when |
8703 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
8704 | */ |
8705 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
8706 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
8707 | ||
1da177e4 | 8708 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 8709 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
8710 | } |
8711 | ||
bd380811 PM |
8712 | return ret; |
8713 | } | |
8714 | ||
a528c219 ND |
8715 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
8716 | unsigned int gchanges) | |
bd380811 PM |
8717 | { |
8718 | unsigned int changes = dev->flags ^ old_flags; | |
8719 | ||
a528c219 | 8720 | if (gchanges) |
7f294054 | 8721 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 8722 | |
bd380811 PM |
8723 | if (changes & IFF_UP) { |
8724 | if (dev->flags & IFF_UP) | |
8725 | call_netdevice_notifiers(NETDEV_UP, dev); | |
8726 | else | |
8727 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
8728 | } | |
8729 | ||
8730 | if (dev->flags & IFF_UP && | |
be9efd36 | 8731 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
51d0c047 DA |
8732 | struct netdev_notifier_change_info change_info = { |
8733 | .info = { | |
8734 | .dev = dev, | |
8735 | }, | |
8736 | .flags_changed = changes, | |
8737 | }; | |
be9efd36 | 8738 | |
51d0c047 | 8739 | call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); |
be9efd36 | 8740 | } |
bd380811 PM |
8741 | } |
8742 | ||
8743 | /** | |
8744 | * dev_change_flags - change device settings | |
8745 | * @dev: device | |
8746 | * @flags: device state flags | |
567c5e13 | 8747 | * @extack: netlink extended ack |
bd380811 PM |
8748 | * |
8749 | * Change settings on device based state flags. The flags are | |
8750 | * in the userspace exported format. | |
8751 | */ | |
567c5e13 PM |
8752 | int dev_change_flags(struct net_device *dev, unsigned int flags, |
8753 | struct netlink_ext_ack *extack) | |
bd380811 | 8754 | { |
b536db93 | 8755 | int ret; |
991fb3f7 | 8756 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 | 8757 | |
6d040321 | 8758 | ret = __dev_change_flags(dev, flags, extack); |
bd380811 PM |
8759 | if (ret < 0) |
8760 | return ret; | |
8761 | ||
991fb3f7 | 8762 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 8763 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
8764 | return ret; |
8765 | } | |
d1b19dff | 8766 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 8767 | |
f51048c3 | 8768 | int __dev_set_mtu(struct net_device *dev, int new_mtu) |
2315dc91 VF |
8769 | { |
8770 | const struct net_device_ops *ops = dev->netdev_ops; | |
8771 | ||
8772 | if (ops->ndo_change_mtu) | |
8773 | return ops->ndo_change_mtu(dev, new_mtu); | |
8774 | ||
501a90c9 ED |
8775 | /* Pairs with all the lockless reads of dev->mtu in the stack */ |
8776 | WRITE_ONCE(dev->mtu, new_mtu); | |
2315dc91 VF |
8777 | return 0; |
8778 | } | |
f51048c3 | 8779 | EXPORT_SYMBOL(__dev_set_mtu); |
2315dc91 | 8780 | |
d836f5c6 ED |
8781 | int dev_validate_mtu(struct net_device *dev, int new_mtu, |
8782 | struct netlink_ext_ack *extack) | |
8783 | { | |
8784 | /* MTU must be positive, and in range */ | |
8785 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
8786 | NL_SET_ERR_MSG(extack, "mtu less than device minimum"); | |
8787 | return -EINVAL; | |
8788 | } | |
8789 | ||
8790 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
8791 | NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); | |
8792 | return -EINVAL; | |
8793 | } | |
8794 | return 0; | |
8795 | } | |
8796 | ||
f0db275a | 8797 | /** |
7a4c53be | 8798 | * dev_set_mtu_ext - Change maximum transfer unit |
f0db275a SH |
8799 | * @dev: device |
8800 | * @new_mtu: new transfer unit | |
7a4c53be | 8801 | * @extack: netlink extended ack |
f0db275a SH |
8802 | * |
8803 | * Change the maximum transfer size of the network device. | |
8804 | */ | |
7a4c53be SH |
8805 | int dev_set_mtu_ext(struct net_device *dev, int new_mtu, |
8806 | struct netlink_ext_ack *extack) | |
1da177e4 | 8807 | { |
2315dc91 | 8808 | int err, orig_mtu; |
1da177e4 LT |
8809 | |
8810 | if (new_mtu == dev->mtu) | |
8811 | return 0; | |
8812 | ||
d836f5c6 ED |
8813 | err = dev_validate_mtu(dev, new_mtu, extack); |
8814 | if (err) | |
8815 | return err; | |
1da177e4 LT |
8816 | |
8817 | if (!netif_device_present(dev)) | |
8818 | return -ENODEV; | |
8819 | ||
1d486bfb VF |
8820 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
8821 | err = notifier_to_errno(err); | |
8822 | if (err) | |
8823 | return err; | |
d314774c | 8824 | |
2315dc91 VF |
8825 | orig_mtu = dev->mtu; |
8826 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 8827 | |
2315dc91 | 8828 | if (!err) { |
af7d6cce SD |
8829 | err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
8830 | orig_mtu); | |
2315dc91 VF |
8831 | err = notifier_to_errno(err); |
8832 | if (err) { | |
8833 | /* setting mtu back and notifying everyone again, | |
8834 | * so that they have a chance to revert changes. | |
8835 | */ | |
8836 | __dev_set_mtu(dev, orig_mtu); | |
af7d6cce SD |
8837 | call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
8838 | new_mtu); | |
2315dc91 VF |
8839 | } |
8840 | } | |
1da177e4 LT |
8841 | return err; |
8842 | } | |
7a4c53be SH |
8843 | |
8844 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
8845 | { | |
8846 | struct netlink_ext_ack extack; | |
8847 | int err; | |
8848 | ||
a6bcfc89 | 8849 | memset(&extack, 0, sizeof(extack)); |
7a4c53be | 8850 | err = dev_set_mtu_ext(dev, new_mtu, &extack); |
a6bcfc89 | 8851 | if (err && extack._msg) |
7a4c53be SH |
8852 | net_err_ratelimited("%s: %s\n", dev->name, extack._msg); |
8853 | return err; | |
8854 | } | |
d1b19dff | 8855 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 8856 | |
6a643ddb CW |
8857 | /** |
8858 | * dev_change_tx_queue_len - Change TX queue length of a netdevice | |
8859 | * @dev: device | |
8860 | * @new_len: new tx queue length | |
8861 | */ | |
8862 | int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) | |
8863 | { | |
8864 | unsigned int orig_len = dev->tx_queue_len; | |
8865 | int res; | |
8866 | ||
8867 | if (new_len != (unsigned int)new_len) | |
8868 | return -ERANGE; | |
8869 | ||
8870 | if (new_len != orig_len) { | |
8871 | dev->tx_queue_len = new_len; | |
8872 | res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); | |
8873 | res = notifier_to_errno(res); | |
7effaf06 TT |
8874 | if (res) |
8875 | goto err_rollback; | |
8876 | res = dev_qdisc_change_tx_queue_len(dev); | |
8877 | if (res) | |
8878 | goto err_rollback; | |
6a643ddb CW |
8879 | } |
8880 | ||
8881 | return 0; | |
7effaf06 TT |
8882 | |
8883 | err_rollback: | |
8884 | netdev_err(dev, "refused to change device tx_queue_len\n"); | |
8885 | dev->tx_queue_len = orig_len; | |
8886 | return res; | |
6a643ddb CW |
8887 | } |
8888 | ||
cbda10fa VD |
8889 | /** |
8890 | * dev_set_group - Change group this device belongs to | |
8891 | * @dev: device | |
8892 | * @new_group: group this device should belong to | |
8893 | */ | |
8894 | void dev_set_group(struct net_device *dev, int new_group) | |
8895 | { | |
8896 | dev->group = new_group; | |
8897 | } | |
8898 | EXPORT_SYMBOL(dev_set_group); | |
8899 | ||
d59cdf94 PM |
8900 | /** |
8901 | * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. | |
8902 | * @dev: device | |
8903 | * @addr: new address | |
8904 | * @extack: netlink extended ack | |
8905 | */ | |
8906 | int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, | |
8907 | struct netlink_ext_ack *extack) | |
8908 | { | |
8909 | struct netdev_notifier_pre_changeaddr_info info = { | |
8910 | .info.dev = dev, | |
8911 | .info.extack = extack, | |
8912 | .dev_addr = addr, | |
8913 | }; | |
8914 | int rc; | |
8915 | ||
8916 | rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); | |
8917 | return notifier_to_errno(rc); | |
8918 | } | |
8919 | EXPORT_SYMBOL(dev_pre_changeaddr_notify); | |
8920 | ||
f0db275a SH |
8921 | /** |
8922 | * dev_set_mac_address - Change Media Access Control Address | |
8923 | * @dev: device | |
8924 | * @sa: new address | |
3a37a963 | 8925 | * @extack: netlink extended ack |
f0db275a SH |
8926 | * |
8927 | * Change the hardware (MAC) address of the device | |
8928 | */ | |
3a37a963 PM |
8929 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, |
8930 | struct netlink_ext_ack *extack) | |
1da177e4 | 8931 | { |
d314774c | 8932 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
8933 | int err; |
8934 | ||
d314774c | 8935 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
8936 | return -EOPNOTSUPP; |
8937 | if (sa->sa_family != dev->type) | |
8938 | return -EINVAL; | |
8939 | if (!netif_device_present(dev)) | |
8940 | return -ENODEV; | |
d59cdf94 PM |
8941 | err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); |
8942 | if (err) | |
8943 | return err; | |
d314774c | 8944 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
8945 | if (err) |
8946 | return err; | |
fbdeca2d | 8947 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 8948 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 8949 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 8950 | return 0; |
1da177e4 | 8951 | } |
d1b19dff | 8952 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 8953 | |
3b23a32a CW |
8954 | static DECLARE_RWSEM(dev_addr_sem); |
8955 | ||
8956 | int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, | |
8957 | struct netlink_ext_ack *extack) | |
8958 | { | |
8959 | int ret; | |
8960 | ||
8961 | down_write(&dev_addr_sem); | |
8962 | ret = dev_set_mac_address(dev, sa, extack); | |
8963 | up_write(&dev_addr_sem); | |
8964 | return ret; | |
8965 | } | |
8966 | EXPORT_SYMBOL(dev_set_mac_address_user); | |
8967 | ||
8968 | int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) | |
8969 | { | |
8970 | size_t size = sizeof(sa->sa_data); | |
8971 | struct net_device *dev; | |
8972 | int ret = 0; | |
8973 | ||
8974 | down_read(&dev_addr_sem); | |
8975 | rcu_read_lock(); | |
8976 | ||
8977 | dev = dev_get_by_name_rcu(net, dev_name); | |
8978 | if (!dev) { | |
8979 | ret = -ENODEV; | |
8980 | goto unlock; | |
8981 | } | |
8982 | if (!dev->addr_len) | |
8983 | memset(sa->sa_data, 0, size); | |
8984 | else | |
8985 | memcpy(sa->sa_data, dev->dev_addr, | |
8986 | min_t(size_t, size, dev->addr_len)); | |
8987 | sa->sa_family = dev->type; | |
8988 | ||
8989 | unlock: | |
8990 | rcu_read_unlock(); | |
8991 | up_read(&dev_addr_sem); | |
8992 | return ret; | |
8993 | } | |
8994 | EXPORT_SYMBOL(dev_get_mac_address); | |
8995 | ||
4bf84c35 JP |
8996 | /** |
8997 | * dev_change_carrier - Change device carrier | |
8998 | * @dev: device | |
691b3b7e | 8999 | * @new_carrier: new value |
4bf84c35 JP |
9000 | * |
9001 | * Change device carrier | |
9002 | */ | |
9003 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
9004 | { | |
9005 | const struct net_device_ops *ops = dev->netdev_ops; | |
9006 | ||
9007 | if (!ops->ndo_change_carrier) | |
9008 | return -EOPNOTSUPP; | |
9009 | if (!netif_device_present(dev)) | |
9010 | return -ENODEV; | |
9011 | return ops->ndo_change_carrier(dev, new_carrier); | |
9012 | } | |
9013 | EXPORT_SYMBOL(dev_change_carrier); | |
9014 | ||
66b52b0d JP |
9015 | /** |
9016 | * dev_get_phys_port_id - Get device physical port ID | |
9017 | * @dev: device | |
9018 | * @ppid: port ID | |
9019 | * | |
9020 | * Get device physical port ID | |
9021 | */ | |
9022 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 9023 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
9024 | { |
9025 | const struct net_device_ops *ops = dev->netdev_ops; | |
9026 | ||
9027 | if (!ops->ndo_get_phys_port_id) | |
9028 | return -EOPNOTSUPP; | |
9029 | return ops->ndo_get_phys_port_id(dev, ppid); | |
9030 | } | |
9031 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
9032 | ||
db24a904 DA |
9033 | /** |
9034 | * dev_get_phys_port_name - Get device physical port name | |
9035 | * @dev: device | |
9036 | * @name: port name | |
ed49e650 | 9037 | * @len: limit of bytes to copy to name |
db24a904 DA |
9038 | * |
9039 | * Get device physical port name | |
9040 | */ | |
9041 | int dev_get_phys_port_name(struct net_device *dev, | |
9042 | char *name, size_t len) | |
9043 | { | |
9044 | const struct net_device_ops *ops = dev->netdev_ops; | |
af3836df | 9045 | int err; |
db24a904 | 9046 | |
af3836df JP |
9047 | if (ops->ndo_get_phys_port_name) { |
9048 | err = ops->ndo_get_phys_port_name(dev, name, len); | |
9049 | if (err != -EOPNOTSUPP) | |
9050 | return err; | |
9051 | } | |
9052 | return devlink_compat_phys_port_name_get(dev, name, len); | |
db24a904 DA |
9053 | } |
9054 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
9055 | ||
d6abc596 FF |
9056 | /** |
9057 | * dev_get_port_parent_id - Get the device's port parent identifier | |
9058 | * @dev: network device | |
9059 | * @ppid: pointer to a storage for the port's parent identifier | |
9060 | * @recurse: allow/disallow recursion to lower devices | |
9061 | * | |
9062 | * Get the devices's port parent identifier | |
9063 | */ | |
9064 | int dev_get_port_parent_id(struct net_device *dev, | |
9065 | struct netdev_phys_item_id *ppid, | |
9066 | bool recurse) | |
9067 | { | |
9068 | const struct net_device_ops *ops = dev->netdev_ops; | |
9069 | struct netdev_phys_item_id first = { }; | |
9070 | struct net_device *lower_dev; | |
9071 | struct list_head *iter; | |
7e1146e8 JP |
9072 | int err; |
9073 | ||
9074 | if (ops->ndo_get_port_parent_id) { | |
9075 | err = ops->ndo_get_port_parent_id(dev, ppid); | |
9076 | if (err != -EOPNOTSUPP) | |
9077 | return err; | |
9078 | } | |
d6abc596 | 9079 | |
7e1146e8 JP |
9080 | err = devlink_compat_switch_id_get(dev, ppid); |
9081 | if (!err || err != -EOPNOTSUPP) | |
9082 | return err; | |
d6abc596 FF |
9083 | |
9084 | if (!recurse) | |
7e1146e8 | 9085 | return -EOPNOTSUPP; |
d6abc596 FF |
9086 | |
9087 | netdev_for_each_lower_dev(dev, lower_dev, iter) { | |
9088 | err = dev_get_port_parent_id(lower_dev, ppid, recurse); | |
9089 | if (err) | |
9090 | break; | |
9091 | if (!first.id_len) | |
9092 | first = *ppid; | |
9093 | else if (memcmp(&first, ppid, sizeof(*ppid))) | |
e1b9efe6 | 9094 | return -EOPNOTSUPP; |
d6abc596 FF |
9095 | } |
9096 | ||
9097 | return err; | |
9098 | } | |
9099 | EXPORT_SYMBOL(dev_get_port_parent_id); | |
9100 | ||
9101 | /** | |
9102 | * netdev_port_same_parent_id - Indicate if two network devices have | |
9103 | * the same port parent identifier | |
9104 | * @a: first network device | |
9105 | * @b: second network device | |
9106 | */ | |
9107 | bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) | |
9108 | { | |
9109 | struct netdev_phys_item_id a_id = { }; | |
9110 | struct netdev_phys_item_id b_id = { }; | |
9111 | ||
9112 | if (dev_get_port_parent_id(a, &a_id, true) || | |
9113 | dev_get_port_parent_id(b, &b_id, true)) | |
9114 | return false; | |
9115 | ||
9116 | return netdev_phys_item_id_same(&a_id, &b_id); | |
9117 | } | |
9118 | EXPORT_SYMBOL(netdev_port_same_parent_id); | |
9119 | ||
d746d707 AK |
9120 | /** |
9121 | * dev_change_proto_down - update protocol port state information | |
9122 | * @dev: device | |
9123 | * @proto_down: new value | |
9124 | * | |
9125 | * This info can be used by switch drivers to set the phys state of the | |
9126 | * port. | |
9127 | */ | |
9128 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
9129 | { | |
9130 | const struct net_device_ops *ops = dev->netdev_ops; | |
9131 | ||
9132 | if (!ops->ndo_change_proto_down) | |
9133 | return -EOPNOTSUPP; | |
9134 | if (!netif_device_present(dev)) | |
9135 | return -ENODEV; | |
9136 | return ops->ndo_change_proto_down(dev, proto_down); | |
9137 | } | |
9138 | EXPORT_SYMBOL(dev_change_proto_down); | |
9139 | ||
b5899679 AR |
9140 | /** |
9141 | * dev_change_proto_down_generic - generic implementation for | |
9142 | * ndo_change_proto_down that sets carrier according to | |
9143 | * proto_down. | |
9144 | * | |
9145 | * @dev: device | |
9146 | * @proto_down: new value | |
9147 | */ | |
9148 | int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) | |
9149 | { | |
9150 | if (proto_down) | |
9151 | netif_carrier_off(dev); | |
9152 | else | |
9153 | netif_carrier_on(dev); | |
9154 | dev->proto_down = proto_down; | |
9155 | return 0; | |
9156 | } | |
9157 | EXPORT_SYMBOL(dev_change_proto_down_generic); | |
9158 | ||
829eb208 RP |
9159 | /** |
9160 | * dev_change_proto_down_reason - proto down reason | |
9161 | * | |
9162 | * @dev: device | |
9163 | * @mask: proto down mask | |
9164 | * @value: proto down value | |
9165 | */ | |
9166 | void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, | |
9167 | u32 value) | |
9168 | { | |
9169 | int b; | |
9170 | ||
9171 | if (!mask) { | |
9172 | dev->proto_down_reason = value; | |
9173 | } else { | |
9174 | for_each_set_bit(b, &mask, 32) { | |
9175 | if (value & (1 << b)) | |
9176 | dev->proto_down_reason |= BIT(b); | |
9177 | else | |
9178 | dev->proto_down_reason &= ~BIT(b); | |
9179 | } | |
9180 | } | |
9181 | } | |
9182 | EXPORT_SYMBOL(dev_change_proto_down_reason); | |
9183 | ||
aa8d3a71 AN |
9184 | struct bpf_xdp_link { |
9185 | struct bpf_link link; | |
9186 | struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ | |
9187 | int flags; | |
9188 | }; | |
9189 | ||
c8a36f19 | 9190 | static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) |
d67b9cd2 | 9191 | { |
7f0a8382 AN |
9192 | if (flags & XDP_FLAGS_HW_MODE) |
9193 | return XDP_MODE_HW; | |
9194 | if (flags & XDP_FLAGS_DRV_MODE) | |
9195 | return XDP_MODE_DRV; | |
c8a36f19 AN |
9196 | if (flags & XDP_FLAGS_SKB_MODE) |
9197 | return XDP_MODE_SKB; | |
9198 | return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; | |
7f0a8382 | 9199 | } |
d67b9cd2 | 9200 | |
7f0a8382 AN |
9201 | static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) |
9202 | { | |
9203 | switch (mode) { | |
9204 | case XDP_MODE_SKB: | |
9205 | return generic_xdp_install; | |
9206 | case XDP_MODE_DRV: | |
9207 | case XDP_MODE_HW: | |
9208 | return dev->netdev_ops->ndo_bpf; | |
9209 | default: | |
9210 | return NULL; | |
5d867245 | 9211 | } |
7f0a8382 | 9212 | } |
118b4aa2 | 9213 | |
aa8d3a71 AN |
9214 | static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, |
9215 | enum bpf_xdp_mode mode) | |
9216 | { | |
9217 | return dev->xdp_state[mode].link; | |
9218 | } | |
9219 | ||
7f0a8382 AN |
9220 | static struct bpf_prog *dev_xdp_prog(struct net_device *dev, |
9221 | enum bpf_xdp_mode mode) | |
9222 | { | |
aa8d3a71 AN |
9223 | struct bpf_xdp_link *link = dev_xdp_link(dev, mode); |
9224 | ||
9225 | if (link) | |
9226 | return link->link.prog; | |
7f0a8382 AN |
9227 | return dev->xdp_state[mode].prog; |
9228 | } | |
9229 | ||
998f1729 THJ |
9230 | static u8 dev_xdp_prog_count(struct net_device *dev) |
9231 | { | |
9232 | u8 count = 0; | |
9233 | int i; | |
9234 | ||
9235 | for (i = 0; i < __MAX_XDP_MODE; i++) | |
9236 | if (dev->xdp_state[i].prog || dev->xdp_state[i].link) | |
9237 | count++; | |
9238 | return count; | |
9239 | } | |
9240 | ||
7f0a8382 AN |
9241 | u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) |
9242 | { | |
9243 | struct bpf_prog *prog = dev_xdp_prog(dev, mode); | |
118b4aa2 | 9244 | |
7f0a8382 AN |
9245 | return prog ? prog->aux->id : 0; |
9246 | } | |
58038695 | 9247 | |
aa8d3a71 AN |
9248 | static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, |
9249 | struct bpf_xdp_link *link) | |
9250 | { | |
9251 | dev->xdp_state[mode].link = link; | |
9252 | dev->xdp_state[mode].prog = NULL; | |
d67b9cd2 DB |
9253 | } |
9254 | ||
7f0a8382 AN |
9255 | static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, |
9256 | struct bpf_prog *prog) | |
9257 | { | |
aa8d3a71 | 9258 | dev->xdp_state[mode].link = NULL; |
7f0a8382 | 9259 | dev->xdp_state[mode].prog = prog; |
d67b9cd2 DB |
9260 | } |
9261 | ||
7f0a8382 AN |
9262 | static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, |
9263 | bpf_op_t bpf_op, struct netlink_ext_ack *extack, | |
9264 | u32 flags, struct bpf_prog *prog) | |
d67b9cd2 | 9265 | { |
f4e63525 | 9266 | struct netdev_bpf xdp; |
7e6897f9 BT |
9267 | int err; |
9268 | ||
d67b9cd2 | 9269 | memset(&xdp, 0, sizeof(xdp)); |
7f0a8382 | 9270 | xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; |
d67b9cd2 | 9271 | xdp.extack = extack; |
32d60277 | 9272 | xdp.flags = flags; |
d67b9cd2 DB |
9273 | xdp.prog = prog; |
9274 | ||
7f0a8382 AN |
9275 | /* Drivers assume refcnt is already incremented (i.e, prog pointer is |
9276 | * "moved" into driver), so they don't increment it on their own, but | |
9277 | * they do decrement refcnt when program is detached or replaced. | |
9278 | * Given net_device also owns link/prog, we need to bump refcnt here | |
9279 | * to prevent drivers from underflowing it. | |
9280 | */ | |
9281 | if (prog) | |
9282 | bpf_prog_inc(prog); | |
7e6897f9 | 9283 | err = bpf_op(dev, &xdp); |
7f0a8382 AN |
9284 | if (err) { |
9285 | if (prog) | |
9286 | bpf_prog_put(prog); | |
9287 | return err; | |
9288 | } | |
7e6897f9 | 9289 | |
7f0a8382 AN |
9290 | if (mode != XDP_MODE_HW) |
9291 | bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); | |
7e6897f9 | 9292 | |
7f0a8382 | 9293 | return 0; |
d67b9cd2 DB |
9294 | } |
9295 | ||
bd0b2e7f JK |
9296 | static void dev_xdp_uninstall(struct net_device *dev) |
9297 | { | |
aa8d3a71 | 9298 | struct bpf_xdp_link *link; |
7f0a8382 AN |
9299 | struct bpf_prog *prog; |
9300 | enum bpf_xdp_mode mode; | |
9301 | bpf_op_t bpf_op; | |
bd0b2e7f | 9302 | |
7f0a8382 | 9303 | ASSERT_RTNL(); |
bd0b2e7f | 9304 | |
7f0a8382 AN |
9305 | for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { |
9306 | prog = dev_xdp_prog(dev, mode); | |
9307 | if (!prog) | |
9308 | continue; | |
bd0b2e7f | 9309 | |
7f0a8382 AN |
9310 | bpf_op = dev_xdp_bpf_op(dev, mode); |
9311 | if (!bpf_op) | |
9312 | continue; | |
bd0b2e7f | 9313 | |
7f0a8382 AN |
9314 | WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); |
9315 | ||
aa8d3a71 AN |
9316 | /* auto-detach link from net device */ |
9317 | link = dev_xdp_link(dev, mode); | |
9318 | if (link) | |
9319 | link->dev = NULL; | |
9320 | else | |
9321 | bpf_prog_put(prog); | |
9322 | ||
9323 | dev_xdp_set_link(dev, mode, NULL); | |
7f0a8382 | 9324 | } |
bd0b2e7f JK |
9325 | } |
9326 | ||
d4baa936 | 9327 | static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, |
aa8d3a71 AN |
9328 | struct bpf_xdp_link *link, struct bpf_prog *new_prog, |
9329 | struct bpf_prog *old_prog, u32 flags) | |
a7862b45 | 9330 | { |
998f1729 | 9331 | unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); |
d4baa936 AN |
9332 | struct bpf_prog *cur_prog; |
9333 | enum bpf_xdp_mode mode; | |
7f0a8382 | 9334 | bpf_op_t bpf_op; |
a7862b45 BB |
9335 | int err; |
9336 | ||
85de8576 DB |
9337 | ASSERT_RTNL(); |
9338 | ||
aa8d3a71 AN |
9339 | /* either link or prog attachment, never both */ |
9340 | if (link && (new_prog || old_prog)) | |
9341 | return -EINVAL; | |
9342 | /* link supports only XDP mode flags */ | |
9343 | if (link && (flags & ~XDP_FLAGS_MODES)) { | |
9344 | NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); | |
9345 | return -EINVAL; | |
9346 | } | |
998f1729 THJ |
9347 | /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ |
9348 | if (num_modes > 1) { | |
d4baa936 AN |
9349 | NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); |
9350 | return -EINVAL; | |
9351 | } | |
998f1729 THJ |
9352 | /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ |
9353 | if (!num_modes && dev_xdp_prog_count(dev) > 1) { | |
9354 | NL_SET_ERR_MSG(extack, | |
9355 | "More than one program loaded, unset mode is ambiguous"); | |
9356 | return -EINVAL; | |
9357 | } | |
d4baa936 AN |
9358 | /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ |
9359 | if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { | |
9360 | NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); | |
9361 | return -EINVAL; | |
01dde20c | 9362 | } |
a25717d2 | 9363 | |
c8a36f19 | 9364 | mode = dev_xdp_mode(dev, flags); |
aa8d3a71 AN |
9365 | /* can't replace attached link */ |
9366 | if (dev_xdp_link(dev, mode)) { | |
9367 | NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); | |
9368 | return -EBUSY; | |
01dde20c | 9369 | } |
c14a9f63 | 9370 | |
d4baa936 | 9371 | cur_prog = dev_xdp_prog(dev, mode); |
aa8d3a71 AN |
9372 | /* can't replace attached prog with link */ |
9373 | if (link && cur_prog) { | |
9374 | NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); | |
9375 | return -EBUSY; | |
9376 | } | |
d4baa936 AN |
9377 | if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { |
9378 | NL_SET_ERR_MSG(extack, "Active program does not match expected"); | |
9379 | return -EEXIST; | |
92234c8f | 9380 | } |
c14a9f63 | 9381 | |
aa8d3a71 AN |
9382 | /* put effective new program into new_prog */ |
9383 | if (link) | |
9384 | new_prog = link->link.prog; | |
85de8576 | 9385 | |
d4baa936 AN |
9386 | if (new_prog) { |
9387 | bool offload = mode == XDP_MODE_HW; | |
7f0a8382 AN |
9388 | enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB |
9389 | ? XDP_MODE_DRV : XDP_MODE_SKB; | |
441a3303 | 9390 | |
068d9d1e AN |
9391 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { |
9392 | NL_SET_ERR_MSG(extack, "XDP program already attached"); | |
9393 | return -EBUSY; | |
9394 | } | |
d4baa936 | 9395 | if (!offload && dev_xdp_prog(dev, other_mode)) { |
7f0a8382 | 9396 | NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); |
d67b9cd2 | 9397 | return -EEXIST; |
01dde20c | 9398 | } |
d4baa936 | 9399 | if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { |
7f0a8382 | 9400 | NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); |
441a3303 JK |
9401 | return -EINVAL; |
9402 | } | |
d4baa936 | 9403 | if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { |
fbee97fe | 9404 | NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); |
fbee97fe DA |
9405 | return -EINVAL; |
9406 | } | |
d4baa936 AN |
9407 | if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { |
9408 | NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); | |
92164774 LB |
9409 | return -EINVAL; |
9410 | } | |
d4baa936 | 9411 | } |
92164774 | 9412 | |
d4baa936 AN |
9413 | /* don't call drivers if the effective program didn't change */ |
9414 | if (new_prog != cur_prog) { | |
9415 | bpf_op = dev_xdp_bpf_op(dev, mode); | |
9416 | if (!bpf_op) { | |
9417 | NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); | |
9418 | return -EOPNOTSUPP; | |
c14a9f63 | 9419 | } |
a7862b45 | 9420 | |
d4baa936 AN |
9421 | err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); |
9422 | if (err) | |
9423 | return err; | |
7f0a8382 | 9424 | } |
d4baa936 | 9425 | |
aa8d3a71 AN |
9426 | if (link) |
9427 | dev_xdp_set_link(dev, mode, link); | |
9428 | else | |
9429 | dev_xdp_set_prog(dev, mode, new_prog); | |
d4baa936 AN |
9430 | if (cur_prog) |
9431 | bpf_prog_put(cur_prog); | |
a7862b45 | 9432 | |
7f0a8382 | 9433 | return 0; |
a7862b45 | 9434 | } |
a7862b45 | 9435 | |
aa8d3a71 AN |
9436 | static int dev_xdp_attach_link(struct net_device *dev, |
9437 | struct netlink_ext_ack *extack, | |
9438 | struct bpf_xdp_link *link) | |
9439 | { | |
9440 | return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); | |
9441 | } | |
9442 | ||
9443 | static int dev_xdp_detach_link(struct net_device *dev, | |
9444 | struct netlink_ext_ack *extack, | |
9445 | struct bpf_xdp_link *link) | |
9446 | { | |
9447 | enum bpf_xdp_mode mode; | |
9448 | bpf_op_t bpf_op; | |
9449 | ||
9450 | ASSERT_RTNL(); | |
9451 | ||
c8a36f19 | 9452 | mode = dev_xdp_mode(dev, link->flags); |
aa8d3a71 AN |
9453 | if (dev_xdp_link(dev, mode) != link) |
9454 | return -EINVAL; | |
9455 | ||
9456 | bpf_op = dev_xdp_bpf_op(dev, mode); | |
9457 | WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); | |
9458 | dev_xdp_set_link(dev, mode, NULL); | |
9459 | return 0; | |
9460 | } | |
9461 | ||
9462 | static void bpf_xdp_link_release(struct bpf_link *link) | |
9463 | { | |
9464 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9465 | ||
9466 | rtnl_lock(); | |
9467 | ||
9468 | /* if racing with net_device's tear down, xdp_link->dev might be | |
9469 | * already NULL, in which case link was already auto-detached | |
9470 | */ | |
73b11c2a | 9471 | if (xdp_link->dev) { |
aa8d3a71 | 9472 | WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); |
73b11c2a AN |
9473 | xdp_link->dev = NULL; |
9474 | } | |
aa8d3a71 AN |
9475 | |
9476 | rtnl_unlock(); | |
9477 | } | |
9478 | ||
73b11c2a AN |
9479 | static int bpf_xdp_link_detach(struct bpf_link *link) |
9480 | { | |
9481 | bpf_xdp_link_release(link); | |
9482 | return 0; | |
9483 | } | |
9484 | ||
aa8d3a71 AN |
9485 | static void bpf_xdp_link_dealloc(struct bpf_link *link) |
9486 | { | |
9487 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9488 | ||
9489 | kfree(xdp_link); | |
9490 | } | |
9491 | ||
c1931c97 AN |
9492 | static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, |
9493 | struct seq_file *seq) | |
9494 | { | |
9495 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9496 | u32 ifindex = 0; | |
9497 | ||
9498 | rtnl_lock(); | |
9499 | if (xdp_link->dev) | |
9500 | ifindex = xdp_link->dev->ifindex; | |
9501 | rtnl_unlock(); | |
9502 | ||
9503 | seq_printf(seq, "ifindex:\t%u\n", ifindex); | |
9504 | } | |
9505 | ||
9506 | static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, | |
9507 | struct bpf_link_info *info) | |
9508 | { | |
9509 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9510 | u32 ifindex = 0; | |
9511 | ||
9512 | rtnl_lock(); | |
9513 | if (xdp_link->dev) | |
9514 | ifindex = xdp_link->dev->ifindex; | |
9515 | rtnl_unlock(); | |
9516 | ||
9517 | info->xdp.ifindex = ifindex; | |
9518 | return 0; | |
9519 | } | |
9520 | ||
026a4c28 AN |
9521 | static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, |
9522 | struct bpf_prog *old_prog) | |
9523 | { | |
9524 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9525 | enum bpf_xdp_mode mode; | |
9526 | bpf_op_t bpf_op; | |
9527 | int err = 0; | |
9528 | ||
9529 | rtnl_lock(); | |
9530 | ||
9531 | /* link might have been auto-released already, so fail */ | |
9532 | if (!xdp_link->dev) { | |
9533 | err = -ENOLINK; | |
9534 | goto out_unlock; | |
9535 | } | |
9536 | ||
9537 | if (old_prog && link->prog != old_prog) { | |
9538 | err = -EPERM; | |
9539 | goto out_unlock; | |
9540 | } | |
9541 | old_prog = link->prog; | |
9542 | if (old_prog == new_prog) { | |
9543 | /* no-op, don't disturb drivers */ | |
9544 | bpf_prog_put(new_prog); | |
9545 | goto out_unlock; | |
9546 | } | |
9547 | ||
c8a36f19 | 9548 | mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); |
026a4c28 AN |
9549 | bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); |
9550 | err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, | |
9551 | xdp_link->flags, new_prog); | |
9552 | if (err) | |
9553 | goto out_unlock; | |
9554 | ||
9555 | old_prog = xchg(&link->prog, new_prog); | |
9556 | bpf_prog_put(old_prog); | |
9557 | ||
9558 | out_unlock: | |
9559 | rtnl_unlock(); | |
9560 | return err; | |
9561 | } | |
9562 | ||
aa8d3a71 AN |
9563 | static const struct bpf_link_ops bpf_xdp_link_lops = { |
9564 | .release = bpf_xdp_link_release, | |
9565 | .dealloc = bpf_xdp_link_dealloc, | |
73b11c2a | 9566 | .detach = bpf_xdp_link_detach, |
c1931c97 AN |
9567 | .show_fdinfo = bpf_xdp_link_show_fdinfo, |
9568 | .fill_link_info = bpf_xdp_link_fill_link_info, | |
026a4c28 | 9569 | .update_prog = bpf_xdp_link_update, |
aa8d3a71 AN |
9570 | }; |
9571 | ||
9572 | int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) | |
9573 | { | |
9574 | struct net *net = current->nsproxy->net_ns; | |
9575 | struct bpf_link_primer link_primer; | |
9576 | struct bpf_xdp_link *link; | |
9577 | struct net_device *dev; | |
9578 | int err, fd; | |
9579 | ||
9580 | dev = dev_get_by_index(net, attr->link_create.target_ifindex); | |
9581 | if (!dev) | |
9582 | return -EINVAL; | |
9583 | ||
9584 | link = kzalloc(sizeof(*link), GFP_USER); | |
9585 | if (!link) { | |
9586 | err = -ENOMEM; | |
9587 | goto out_put_dev; | |
9588 | } | |
9589 | ||
9590 | bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); | |
9591 | link->dev = dev; | |
9592 | link->flags = attr->link_create.flags; | |
9593 | ||
9594 | err = bpf_link_prime(&link->link, &link_primer); | |
9595 | if (err) { | |
9596 | kfree(link); | |
9597 | goto out_put_dev; | |
9598 | } | |
9599 | ||
9600 | rtnl_lock(); | |
9601 | err = dev_xdp_attach_link(dev, NULL, link); | |
9602 | rtnl_unlock(); | |
9603 | ||
9604 | if (err) { | |
9605 | bpf_link_cleanup(&link_primer); | |
9606 | goto out_put_dev; | |
9607 | } | |
9608 | ||
9609 | fd = bpf_link_settle(&link_primer); | |
9610 | /* link itself doesn't hold dev's refcnt to not complicate shutdown */ | |
9611 | dev_put(dev); | |
9612 | return fd; | |
9613 | ||
9614 | out_put_dev: | |
9615 | dev_put(dev); | |
9616 | return err; | |
9617 | } | |
9618 | ||
d4baa936 AN |
9619 | /** |
9620 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
9621 | * @dev: device | |
9622 | * @extack: netlink extended ack | |
9623 | * @fd: new program fd or negative value to clear | |
9624 | * @expected_fd: old program fd that userspace expects to replace or clear | |
9625 | * @flags: xdp-related flags | |
9626 | * | |
9627 | * Set or clear a bpf program for a device | |
9628 | */ | |
9629 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, | |
9630 | int fd, int expected_fd, u32 flags) | |
9631 | { | |
c8a36f19 | 9632 | enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); |
d4baa936 AN |
9633 | struct bpf_prog *new_prog = NULL, *old_prog = NULL; |
9634 | int err; | |
9635 | ||
9636 | ASSERT_RTNL(); | |
9637 | ||
9638 | if (fd >= 0) { | |
9639 | new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, | |
9640 | mode != XDP_MODE_SKB); | |
9641 | if (IS_ERR(new_prog)) | |
9642 | return PTR_ERR(new_prog); | |
9643 | } | |
9644 | ||
9645 | if (expected_fd >= 0) { | |
9646 | old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, | |
9647 | mode != XDP_MODE_SKB); | |
9648 | if (IS_ERR(old_prog)) { | |
9649 | err = PTR_ERR(old_prog); | |
9650 | old_prog = NULL; | |
9651 | goto err_out; | |
c14a9f63 | 9652 | } |
a7862b45 BB |
9653 | } |
9654 | ||
aa8d3a71 | 9655 | err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); |
a7862b45 | 9656 | |
d4baa936 AN |
9657 | err_out: |
9658 | if (err && new_prog) | |
9659 | bpf_prog_put(new_prog); | |
9660 | if (old_prog) | |
9661 | bpf_prog_put(old_prog); | |
a7862b45 BB |
9662 | return err; |
9663 | } | |
a7862b45 | 9664 | |
1da177e4 LT |
9665 | /** |
9666 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 9667 | * @net: the applicable net namespace |
1da177e4 LT |
9668 | * |
9669 | * Returns a suitable unique value for a new device interface | |
9670 | * number. The caller must hold the rtnl semaphore or the | |
9671 | * dev_base_lock to be sure it remains unique. | |
9672 | */ | |
881d966b | 9673 | static int dev_new_index(struct net *net) |
1da177e4 | 9674 | { |
aa79e66e | 9675 | int ifindex = net->ifindex; |
f4563a75 | 9676 | |
1da177e4 LT |
9677 | for (;;) { |
9678 | if (++ifindex <= 0) | |
9679 | ifindex = 1; | |
881d966b | 9680 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 9681 | return net->ifindex = ifindex; |
1da177e4 LT |
9682 | } |
9683 | } | |
9684 | ||
1da177e4 | 9685 | /* Delayed registration/unregisteration */ |
3b5b34fd | 9686 | static LIST_HEAD(net_todo_list); |
200b916f | 9687 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 9688 | |
6f05f629 | 9689 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 9690 | { |
1da177e4 | 9691 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 9692 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
9693 | } |
9694 | ||
fd867d51 JW |
9695 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
9696 | struct net_device *upper, netdev_features_t features) | |
9697 | { | |
9698 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
9699 | netdev_features_t feature; | |
5ba3f7d6 | 9700 | int feature_bit; |
fd867d51 | 9701 | |
3b89ea9c | 9702 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 9703 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
9704 | if (!(upper->wanted_features & feature) |
9705 | && (features & feature)) { | |
9706 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
9707 | &feature, upper->name); | |
9708 | features &= ~feature; | |
9709 | } | |
9710 | } | |
9711 | ||
9712 | return features; | |
9713 | } | |
9714 | ||
9715 | static void netdev_sync_lower_features(struct net_device *upper, | |
9716 | struct net_device *lower, netdev_features_t features) | |
9717 | { | |
9718 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
9719 | netdev_features_t feature; | |
5ba3f7d6 | 9720 | int feature_bit; |
fd867d51 | 9721 | |
3b89ea9c | 9722 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 9723 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
9724 | if (!(features & feature) && (lower->features & feature)) { |
9725 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
9726 | &feature, lower->name); | |
9727 | lower->wanted_features &= ~feature; | |
dd912306 | 9728 | __netdev_update_features(lower); |
fd867d51 JW |
9729 | |
9730 | if (unlikely(lower->features & feature)) | |
9731 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
9732 | &feature, lower->name); | |
dd912306 CW |
9733 | else |
9734 | netdev_features_change(lower); | |
fd867d51 JW |
9735 | } |
9736 | } | |
9737 | } | |
9738 | ||
c8f44aff MM |
9739 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
9740 | netdev_features_t features) | |
b63365a2 | 9741 | { |
57422dc5 MM |
9742 | /* Fix illegal checksum combinations */ |
9743 | if ((features & NETIF_F_HW_CSUM) && | |
9744 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 9745 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
9746 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
9747 | } | |
9748 | ||
b63365a2 | 9749 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 9750 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 9751 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 9752 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
9753 | } |
9754 | ||
ec5f0615 PS |
9755 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
9756 | !(features & NETIF_F_IP_CSUM)) { | |
9757 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
9758 | features &= ~NETIF_F_TSO; | |
9759 | features &= ~NETIF_F_TSO_ECN; | |
9760 | } | |
9761 | ||
9762 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
9763 | !(features & NETIF_F_IPV6_CSUM)) { | |
9764 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
9765 | features &= ~NETIF_F_TSO6; | |
9766 | } | |
9767 | ||
b1dc497b AD |
9768 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
9769 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
9770 | features &= ~NETIF_F_TSO_MANGLEID; | |
9771 | ||
31d8b9e0 BH |
9772 | /* TSO ECN requires that TSO is present as well. */ |
9773 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
9774 | features &= ~NETIF_F_TSO_ECN; | |
9775 | ||
212b573f MM |
9776 | /* Software GSO depends on SG. */ |
9777 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 9778 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
9779 | features &= ~NETIF_F_GSO; |
9780 | } | |
9781 | ||
802ab55a AD |
9782 | /* GSO partial features require GSO partial be set */ |
9783 | if ((features & dev->gso_partial_features) && | |
9784 | !(features & NETIF_F_GSO_PARTIAL)) { | |
9785 | netdev_dbg(dev, | |
9786 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
9787 | features &= ~dev->gso_partial_features; | |
9788 | } | |
9789 | ||
fb1f5f79 MC |
9790 | if (!(features & NETIF_F_RXCSUM)) { |
9791 | /* NETIF_F_GRO_HW implies doing RXCSUM since every packet | |
9792 | * successfully merged by hardware must also have the | |
9793 | * checksum verified by hardware. If the user does not | |
9794 | * want to enable RXCSUM, logically, we should disable GRO_HW. | |
9795 | */ | |
9796 | if (features & NETIF_F_GRO_HW) { | |
9797 | netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); | |
9798 | features &= ~NETIF_F_GRO_HW; | |
9799 | } | |
9800 | } | |
9801 | ||
de8d5ab2 GP |
9802 | /* LRO/HW-GRO features cannot be combined with RX-FCS */ |
9803 | if (features & NETIF_F_RXFCS) { | |
9804 | if (features & NETIF_F_LRO) { | |
9805 | netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); | |
9806 | features &= ~NETIF_F_LRO; | |
9807 | } | |
9808 | ||
9809 | if (features & NETIF_F_GRO_HW) { | |
9810 | netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); | |
9811 | features &= ~NETIF_F_GRO_HW; | |
9812 | } | |
e6c6a929 GP |
9813 | } |
9814 | ||
25537d71 TT |
9815 | if (features & NETIF_F_HW_TLS_TX) { |
9816 | bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == | |
9817 | (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
9818 | bool hw_csum = features & NETIF_F_HW_CSUM; | |
9819 | ||
9820 | if (!ip_csum && !hw_csum) { | |
9821 | netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); | |
9822 | features &= ~NETIF_F_HW_TLS_TX; | |
9823 | } | |
ae0b04b2 TT |
9824 | } |
9825 | ||
a3eb4e9d TT |
9826 | if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { |
9827 | netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); | |
9828 | features &= ~NETIF_F_HW_TLS_RX; | |
9829 | } | |
9830 | ||
b63365a2 HX |
9831 | return features; |
9832 | } | |
b63365a2 | 9833 | |
6cb6a27c | 9834 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 9835 | { |
fd867d51 | 9836 | struct net_device *upper, *lower; |
c8f44aff | 9837 | netdev_features_t features; |
fd867d51 | 9838 | struct list_head *iter; |
e7868a85 | 9839 | int err = -1; |
5455c699 | 9840 | |
87267485 MM |
9841 | ASSERT_RTNL(); |
9842 | ||
5455c699 MM |
9843 | features = netdev_get_wanted_features(dev); |
9844 | ||
9845 | if (dev->netdev_ops->ndo_fix_features) | |
9846 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
9847 | ||
9848 | /* driver might be less strict about feature dependencies */ | |
9849 | features = netdev_fix_features(dev, features); | |
9850 | ||
4250b75b | 9851 | /* some features can't be enabled if they're off on an upper device */ |
fd867d51 JW |
9852 | netdev_for_each_upper_dev_rcu(dev, upper, iter) |
9853 | features = netdev_sync_upper_features(dev, upper, features); | |
9854 | ||
5455c699 | 9855 | if (dev->features == features) |
e7868a85 | 9856 | goto sync_lower; |
5455c699 | 9857 | |
c8f44aff MM |
9858 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
9859 | &dev->features, &features); | |
5455c699 MM |
9860 | |
9861 | if (dev->netdev_ops->ndo_set_features) | |
9862 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
9863 | else |
9864 | err = 0; | |
5455c699 | 9865 | |
6cb6a27c | 9866 | if (unlikely(err < 0)) { |
5455c699 | 9867 | netdev_err(dev, |
c8f44aff MM |
9868 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
9869 | err, &features, &dev->features); | |
17b85d29 NA |
9870 | /* return non-0 since some features might have changed and |
9871 | * it's better to fire a spurious notification than miss it | |
9872 | */ | |
9873 | return -1; | |
6cb6a27c MM |
9874 | } |
9875 | ||
e7868a85 | 9876 | sync_lower: |
fd867d51 JW |
9877 | /* some features must be disabled on lower devices when disabled |
9878 | * on an upper device (think: bonding master or bridge) | |
9879 | */ | |
9880 | netdev_for_each_lower_dev(dev, lower, iter) | |
9881 | netdev_sync_lower_features(dev, lower, features); | |
9882 | ||
ae847f40 SD |
9883 | if (!err) { |
9884 | netdev_features_t diff = features ^ dev->features; | |
9885 | ||
9886 | if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
9887 | /* udp_tunnel_{get,drop}_rx_info both need | |
9888 | * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the | |
9889 | * device, or they won't do anything. | |
9890 | * Thus we need to update dev->features | |
9891 | * *before* calling udp_tunnel_get_rx_info, | |
9892 | * but *after* calling udp_tunnel_drop_rx_info. | |
9893 | */ | |
9894 | if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
9895 | dev->features = features; | |
9896 | udp_tunnel_get_rx_info(dev); | |
9897 | } else { | |
9898 | udp_tunnel_drop_rx_info(dev); | |
9899 | } | |
9900 | } | |
9901 | ||
9daae9bd GP |
9902 | if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { |
9903 | if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { | |
9904 | dev->features = features; | |
9905 | err |= vlan_get_rx_ctag_filter_info(dev); | |
9906 | } else { | |
9907 | vlan_drop_rx_ctag_filter_info(dev); | |
9908 | } | |
9909 | } | |
9910 | ||
9911 | if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { | |
9912 | if (features & NETIF_F_HW_VLAN_STAG_FILTER) { | |
9913 | dev->features = features; | |
9914 | err |= vlan_get_rx_stag_filter_info(dev); | |
9915 | } else { | |
9916 | vlan_drop_rx_stag_filter_info(dev); | |
9917 | } | |
9918 | } | |
9919 | ||
6cb6a27c | 9920 | dev->features = features; |
ae847f40 | 9921 | } |
6cb6a27c | 9922 | |
e7868a85 | 9923 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
9924 | } |
9925 | ||
afe12cc8 MM |
9926 | /** |
9927 | * netdev_update_features - recalculate device features | |
9928 | * @dev: the device to check | |
9929 | * | |
9930 | * Recalculate dev->features set and send notifications if it | |
9931 | * has changed. Should be called after driver or hardware dependent | |
9932 | * conditions might have changed that influence the features. | |
9933 | */ | |
6cb6a27c MM |
9934 | void netdev_update_features(struct net_device *dev) |
9935 | { | |
9936 | if (__netdev_update_features(dev)) | |
9937 | netdev_features_change(dev); | |
5455c699 MM |
9938 | } |
9939 | EXPORT_SYMBOL(netdev_update_features); | |
9940 | ||
afe12cc8 MM |
9941 | /** |
9942 | * netdev_change_features - recalculate device features | |
9943 | * @dev: the device to check | |
9944 | * | |
9945 | * Recalculate dev->features set and send notifications even | |
9946 | * if they have not changed. Should be called instead of | |
9947 | * netdev_update_features() if also dev->vlan_features might | |
9948 | * have changed to allow the changes to be propagated to stacked | |
9949 | * VLAN devices. | |
9950 | */ | |
9951 | void netdev_change_features(struct net_device *dev) | |
9952 | { | |
9953 | __netdev_update_features(dev); | |
9954 | netdev_features_change(dev); | |
9955 | } | |
9956 | EXPORT_SYMBOL(netdev_change_features); | |
9957 | ||
fc4a7489 PM |
9958 | /** |
9959 | * netif_stacked_transfer_operstate - transfer operstate | |
9960 | * @rootdev: the root or lower level device to transfer state from | |
9961 | * @dev: the device to transfer operstate to | |
9962 | * | |
9963 | * Transfer operational state from root to device. This is normally | |
9964 | * called when a stacking relationship exists between the root | |
9965 | * device and the device(a leaf device). | |
9966 | */ | |
9967 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
9968 | struct net_device *dev) | |
9969 | { | |
9970 | if (rootdev->operstate == IF_OPER_DORMANT) | |
9971 | netif_dormant_on(dev); | |
9972 | else | |
9973 | netif_dormant_off(dev); | |
9974 | ||
eec517cd AL |
9975 | if (rootdev->operstate == IF_OPER_TESTING) |
9976 | netif_testing_on(dev); | |
9977 | else | |
9978 | netif_testing_off(dev); | |
9979 | ||
0575c86b ZS |
9980 | if (netif_carrier_ok(rootdev)) |
9981 | netif_carrier_on(dev); | |
9982 | else | |
9983 | netif_carrier_off(dev); | |
fc4a7489 PM |
9984 | } |
9985 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
9986 | ||
1b4bf461 ED |
9987 | static int netif_alloc_rx_queues(struct net_device *dev) |
9988 | { | |
1b4bf461 | 9989 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 9990 | struct netdev_rx_queue *rx; |
10595902 | 9991 | size_t sz = count * sizeof(*rx); |
e817f856 | 9992 | int err = 0; |
1b4bf461 | 9993 | |
bd25fa7b | 9994 | BUG_ON(count < 1); |
1b4bf461 | 9995 | |
dcda9b04 | 9996 | rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
9997 | if (!rx) |
9998 | return -ENOMEM; | |
9999 | ||
bd25fa7b TH |
10000 | dev->_rx = rx; |
10001 | ||
e817f856 | 10002 | for (i = 0; i < count; i++) { |
fe822240 | 10003 | rx[i].dev = dev; |
e817f856 JDB |
10004 | |
10005 | /* XDP RX-queue setup */ | |
b02e5a0e | 10006 | err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); |
e817f856 JDB |
10007 | if (err < 0) |
10008 | goto err_rxq_info; | |
10009 | } | |
1b4bf461 | 10010 | return 0; |
e817f856 JDB |
10011 | |
10012 | err_rxq_info: | |
10013 | /* Rollback successful reg's and free other resources */ | |
10014 | while (i--) | |
10015 | xdp_rxq_info_unreg(&rx[i].xdp_rxq); | |
141b52a9 | 10016 | kvfree(dev->_rx); |
e817f856 JDB |
10017 | dev->_rx = NULL; |
10018 | return err; | |
10019 | } | |
10020 | ||
10021 | static void netif_free_rx_queues(struct net_device *dev) | |
10022 | { | |
10023 | unsigned int i, count = dev->num_rx_queues; | |
e817f856 JDB |
10024 | |
10025 | /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ | |
10026 | if (!dev->_rx) | |
10027 | return; | |
10028 | ||
e817f856 | 10029 | for (i = 0; i < count; i++) |
82aaff2f JK |
10030 | xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); |
10031 | ||
10032 | kvfree(dev->_rx); | |
1b4bf461 ED |
10033 | } |
10034 | ||
aa942104 CG |
10035 | static void netdev_init_one_queue(struct net_device *dev, |
10036 | struct netdev_queue *queue, void *_unused) | |
10037 | { | |
10038 | /* Initialize queue lock */ | |
10039 | spin_lock_init(&queue->_xmit_lock); | |
1a33e10e | 10040 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); |
aa942104 | 10041 | queue->xmit_lock_owner = -1; |
b236da69 | 10042 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 10043 | queue->dev = dev; |
114cf580 TH |
10044 | #ifdef CONFIG_BQL |
10045 | dql_init(&queue->dql, HZ); | |
10046 | #endif | |
aa942104 CG |
10047 | } |
10048 | ||
60877a32 ED |
10049 | static void netif_free_tx_queues(struct net_device *dev) |
10050 | { | |
4cb28970 | 10051 | kvfree(dev->_tx); |
60877a32 ED |
10052 | } |
10053 | ||
e6484930 TH |
10054 | static int netif_alloc_netdev_queues(struct net_device *dev) |
10055 | { | |
10056 | unsigned int count = dev->num_tx_queues; | |
10057 | struct netdev_queue *tx; | |
60877a32 | 10058 | size_t sz = count * sizeof(*tx); |
e6484930 | 10059 | |
d339727c ED |
10060 | if (count < 1 || count > 0xffff) |
10061 | return -EINVAL; | |
62b5942a | 10062 | |
dcda9b04 | 10063 | tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
10064 | if (!tx) |
10065 | return -ENOMEM; | |
10066 | ||
e6484930 | 10067 | dev->_tx = tx; |
1d24eb48 | 10068 | |
e6484930 TH |
10069 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
10070 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
10071 | |
10072 | return 0; | |
e6484930 TH |
10073 | } |
10074 | ||
a2029240 DV |
10075 | void netif_tx_stop_all_queues(struct net_device *dev) |
10076 | { | |
10077 | unsigned int i; | |
10078 | ||
10079 | for (i = 0; i < dev->num_tx_queues; i++) { | |
10080 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 10081 | |
a2029240 DV |
10082 | netif_tx_stop_queue(txq); |
10083 | } | |
10084 | } | |
10085 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
10086 | ||
1da177e4 LT |
10087 | /** |
10088 | * register_netdevice - register a network device | |
10089 | * @dev: device to register | |
10090 | * | |
10091 | * Take a completed network device structure and add it to the kernel | |
10092 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
10093 | * chain. 0 is returned on success. A negative errno code is returned | |
10094 | * on a failure to set up the device, or if the name is a duplicate. | |
10095 | * | |
10096 | * Callers must hold the rtnl semaphore. You may want | |
10097 | * register_netdev() instead of this. | |
10098 | * | |
10099 | * BUGS: | |
10100 | * The locking appears insufficient to guarantee two parallel registers | |
10101 | * will not get the same name. | |
10102 | */ | |
10103 | ||
10104 | int register_netdevice(struct net_device *dev) | |
10105 | { | |
1da177e4 | 10106 | int ret; |
d314774c | 10107 | struct net *net = dev_net(dev); |
1da177e4 | 10108 | |
e283de3a FF |
10109 | BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < |
10110 | NETDEV_FEATURE_COUNT); | |
1da177e4 LT |
10111 | BUG_ON(dev_boot_phase); |
10112 | ASSERT_RTNL(); | |
10113 | ||
b17a7c17 SH |
10114 | might_sleep(); |
10115 | ||
1da177e4 LT |
10116 | /* When net_device's are persistent, this will be fatal. */ |
10117 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 10118 | BUG_ON(!net); |
1da177e4 | 10119 | |
9000edb7 JK |
10120 | ret = ethtool_check_ops(dev->ethtool_ops); |
10121 | if (ret) | |
10122 | return ret; | |
10123 | ||
f1f28aa3 | 10124 | spin_lock_init(&dev->addr_list_lock); |
845e0ebb | 10125 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 10126 | |
828de4f6 | 10127 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
10128 | if (ret < 0) |
10129 | goto out; | |
10130 | ||
9077f052 | 10131 | ret = -ENOMEM; |
ff927412 JP |
10132 | dev->name_node = netdev_name_node_head_alloc(dev); |
10133 | if (!dev->name_node) | |
10134 | goto out; | |
10135 | ||
1da177e4 | 10136 | /* Init, if this function is available */ |
d314774c SH |
10137 | if (dev->netdev_ops->ndo_init) { |
10138 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
10139 | if (ret) { |
10140 | if (ret > 0) | |
10141 | ret = -EIO; | |
42c17fa6 | 10142 | goto err_free_name; |
1da177e4 LT |
10143 | } |
10144 | } | |
4ec93edb | 10145 | |
f646968f PM |
10146 | if (((dev->hw_features | dev->features) & |
10147 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
10148 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
10149 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
10150 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
10151 | ret = -EINVAL; | |
10152 | goto err_uninit; | |
10153 | } | |
10154 | ||
9c7dafbf PE |
10155 | ret = -EBUSY; |
10156 | if (!dev->ifindex) | |
10157 | dev->ifindex = dev_new_index(net); | |
10158 | else if (__dev_get_by_index(net, dev->ifindex)) | |
10159 | goto err_uninit; | |
10160 | ||
5455c699 MM |
10161 | /* Transfer changeable features to wanted_features and enable |
10162 | * software offloads (GSO and GRO). | |
10163 | */ | |
1a3c998f | 10164 | dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); |
14d1232f | 10165 | dev->features |= NETIF_F_SOFT_FEATURES; |
d764a122 | 10166 | |
876c4384 | 10167 | if (dev->udp_tunnel_nic_info) { |
d764a122 SD |
10168 | dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; |
10169 | dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
10170 | } | |
10171 | ||
14d1232f | 10172 | dev->wanted_features = dev->features & dev->hw_features; |
1da177e4 | 10173 | |
cbc53e08 | 10174 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 10175 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 10176 | |
7f348a60 AD |
10177 | /* If IPv4 TCP segmentation offload is supported we should also |
10178 | * allow the device to enable segmenting the frame with the option | |
10179 | * of ignoring a static IP ID value. This doesn't enable the | |
10180 | * feature itself but allows the user to enable it later. | |
10181 | */ | |
cbc53e08 AD |
10182 | if (dev->hw_features & NETIF_F_TSO) |
10183 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
10184 | if (dev->vlan_features & NETIF_F_TSO) |
10185 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
10186 | if (dev->mpls_features & NETIF_F_TSO) | |
10187 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
10188 | if (dev->hw_enc_features & NETIF_F_TSO) | |
10189 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 10190 | |
1180e7d6 | 10191 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 10192 | */ |
1180e7d6 | 10193 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 10194 | |
ee579677 PS |
10195 | /* Make NETIF_F_SG inheritable to tunnel devices. |
10196 | */ | |
802ab55a | 10197 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 10198 | |
0d89d203 SH |
10199 | /* Make NETIF_F_SG inheritable to MPLS. |
10200 | */ | |
10201 | dev->mpls_features |= NETIF_F_SG; | |
10202 | ||
7ffbe3fd JB |
10203 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
10204 | ret = notifier_to_errno(ret); | |
10205 | if (ret) | |
10206 | goto err_uninit; | |
10207 | ||
8b41d188 | 10208 | ret = netdev_register_kobject(dev); |
cb626bf5 JH |
10209 | if (ret) { |
10210 | dev->reg_state = NETREG_UNREGISTERED; | |
7ce1b0ed | 10211 | goto err_uninit; |
cb626bf5 | 10212 | } |
b17a7c17 SH |
10213 | dev->reg_state = NETREG_REGISTERED; |
10214 | ||
6cb6a27c | 10215 | __netdev_update_features(dev); |
8e9b59b2 | 10216 | |
1da177e4 LT |
10217 | /* |
10218 | * Default initial state at registry is that the | |
10219 | * device is present. | |
10220 | */ | |
10221 | ||
10222 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
10223 | ||
8f4cccbb BH |
10224 | linkwatch_init_dev(dev); |
10225 | ||
1da177e4 | 10226 | dev_init_scheduler(dev); |
1da177e4 | 10227 | dev_hold(dev); |
ce286d32 | 10228 | list_netdevice(dev); |
7bf23575 | 10229 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 10230 | |
948b337e JP |
10231 | /* If the device has permanent device address, driver should |
10232 | * set dev_addr and also addr_assign_type should be set to | |
10233 | * NET_ADDR_PERM (default value). | |
10234 | */ | |
10235 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
10236 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
10237 | ||
1da177e4 | 10238 | /* Notify protocols, that a new device appeared. */ |
056925ab | 10239 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 10240 | ret = notifier_to_errno(ret); |
93ee31f1 | 10241 | if (ret) { |
766b0515 JK |
10242 | /* Expect explicit free_netdev() on failure */ |
10243 | dev->needs_free_netdev = false; | |
037e56bd | 10244 | unregister_netdevice_queue(dev, NULL); |
766b0515 | 10245 | goto out; |
93ee31f1 | 10246 | } |
d90a909e EB |
10247 | /* |
10248 | * Prevent userspace races by waiting until the network | |
10249 | * device is fully setup before sending notifications. | |
10250 | */ | |
a2835763 PM |
10251 | if (!dev->rtnl_link_ops || |
10252 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 10253 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
10254 | |
10255 | out: | |
10256 | return ret; | |
7ce1b0ed HX |
10257 | |
10258 | err_uninit: | |
d314774c SH |
10259 | if (dev->netdev_ops->ndo_uninit) |
10260 | dev->netdev_ops->ndo_uninit(dev); | |
cf124db5 DM |
10261 | if (dev->priv_destructor) |
10262 | dev->priv_destructor(dev); | |
42c17fa6 DC |
10263 | err_free_name: |
10264 | netdev_name_node_free(dev->name_node); | |
7ce1b0ed | 10265 | goto out; |
1da177e4 | 10266 | } |
d1b19dff | 10267 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 10268 | |
937f1ba5 BH |
10269 | /** |
10270 | * init_dummy_netdev - init a dummy network device for NAPI | |
10271 | * @dev: device to init | |
10272 | * | |
10273 | * This takes a network device structure and initialize the minimum | |
10274 | * amount of fields so it can be used to schedule NAPI polls without | |
10275 | * registering a full blown interface. This is to be used by drivers | |
10276 | * that need to tie several hardware interfaces to a single NAPI | |
10277 | * poll scheduler due to HW limitations. | |
10278 | */ | |
10279 | int init_dummy_netdev(struct net_device *dev) | |
10280 | { | |
10281 | /* Clear everything. Note we don't initialize spinlocks | |
10282 | * are they aren't supposed to be taken by any of the | |
10283 | * NAPI code and this dummy netdev is supposed to be | |
10284 | * only ever used for NAPI polls | |
10285 | */ | |
10286 | memset(dev, 0, sizeof(struct net_device)); | |
10287 | ||
10288 | /* make sure we BUG if trying to hit standard | |
10289 | * register/unregister code path | |
10290 | */ | |
10291 | dev->reg_state = NETREG_DUMMY; | |
10292 | ||
937f1ba5 BH |
10293 | /* NAPI wants this */ |
10294 | INIT_LIST_HEAD(&dev->napi_list); | |
10295 | ||
10296 | /* a dummy interface is started by default */ | |
10297 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
10298 | set_bit(__LINK_STATE_START, &dev->state); | |
10299 | ||
35edfdc7 JE |
10300 | /* napi_busy_loop stats accounting wants this */ |
10301 | dev_net_set(dev, &init_net); | |
10302 | ||
29b4433d ED |
10303 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
10304 | * because users of this 'device' dont need to change | |
10305 | * its refcount. | |
10306 | */ | |
10307 | ||
937f1ba5 BH |
10308 | return 0; |
10309 | } | |
10310 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
10311 | ||
10312 | ||
1da177e4 LT |
10313 | /** |
10314 | * register_netdev - register a network device | |
10315 | * @dev: device to register | |
10316 | * | |
10317 | * Take a completed network device structure and add it to the kernel | |
10318 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
10319 | * chain. 0 is returned on success. A negative errno code is returned | |
10320 | * on a failure to set up the device, or if the name is a duplicate. | |
10321 | * | |
38b4da38 | 10322 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
10323 | * and expands the device name if you passed a format string to |
10324 | * alloc_netdev. | |
10325 | */ | |
10326 | int register_netdev(struct net_device *dev) | |
10327 | { | |
10328 | int err; | |
10329 | ||
b0f3debc KT |
10330 | if (rtnl_lock_killable()) |
10331 | return -EINTR; | |
1da177e4 | 10332 | err = register_netdevice(dev); |
1da177e4 LT |
10333 | rtnl_unlock(); |
10334 | return err; | |
10335 | } | |
10336 | EXPORT_SYMBOL(register_netdev); | |
10337 | ||
29b4433d ED |
10338 | int netdev_refcnt_read(const struct net_device *dev) |
10339 | { | |
10340 | int i, refcnt = 0; | |
10341 | ||
10342 | for_each_possible_cpu(i) | |
10343 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
10344 | return refcnt; | |
10345 | } | |
10346 | EXPORT_SYMBOL(netdev_refcnt_read); | |
10347 | ||
de2b541b MCC |
10348 | #define WAIT_REFS_MIN_MSECS 1 |
10349 | #define WAIT_REFS_MAX_MSECS 250 | |
2c53040f | 10350 | /** |
1da177e4 | 10351 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 10352 | * @dev: target net_device |
1da177e4 LT |
10353 | * |
10354 | * This is called when unregistering network devices. | |
10355 | * | |
10356 | * Any protocol or device that holds a reference should register | |
10357 | * for netdevice notification, and cleanup and put back the | |
10358 | * reference if they receive an UNREGISTER event. | |
10359 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 10360 | * call dev_put. |
1da177e4 LT |
10361 | */ |
10362 | static void netdev_wait_allrefs(struct net_device *dev) | |
10363 | { | |
10364 | unsigned long rebroadcast_time, warning_time; | |
0e4be9e5 | 10365 | int wait = 0, refcnt; |
1da177e4 | 10366 | |
e014debe ED |
10367 | linkwatch_forget_dev(dev); |
10368 | ||
1da177e4 | 10369 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
10370 | refcnt = netdev_refcnt_read(dev); |
10371 | ||
10372 | while (refcnt != 0) { | |
1da177e4 | 10373 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 10374 | rtnl_lock(); |
1da177e4 LT |
10375 | |
10376 | /* Rebroadcast unregister notification */ | |
056925ab | 10377 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 10378 | |
748e2d93 | 10379 | __rtnl_unlock(); |
0115e8e3 | 10380 | rcu_barrier(); |
748e2d93 ED |
10381 | rtnl_lock(); |
10382 | ||
1da177e4 LT |
10383 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
10384 | &dev->state)) { | |
10385 | /* We must not have linkwatch events | |
10386 | * pending on unregister. If this | |
10387 | * happens, we simply run the queue | |
10388 | * unscheduled, resulting in a noop | |
10389 | * for this device. | |
10390 | */ | |
10391 | linkwatch_run_queue(); | |
10392 | } | |
10393 | ||
6756ae4b | 10394 | __rtnl_unlock(); |
1da177e4 LT |
10395 | |
10396 | rebroadcast_time = jiffies; | |
10397 | } | |
10398 | ||
0e4be9e5 FR |
10399 | if (!wait) { |
10400 | rcu_barrier(); | |
10401 | wait = WAIT_REFS_MIN_MSECS; | |
10402 | } else { | |
10403 | msleep(wait); | |
10404 | wait = min(wait << 1, WAIT_REFS_MAX_MSECS); | |
10405 | } | |
1da177e4 | 10406 | |
29b4433d ED |
10407 | refcnt = netdev_refcnt_read(dev); |
10408 | ||
d7c04b05 | 10409 | if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
10410 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
10411 | dev->name, refcnt); | |
1da177e4 LT |
10412 | warning_time = jiffies; |
10413 | } | |
10414 | } | |
10415 | } | |
10416 | ||
10417 | /* The sequence is: | |
10418 | * | |
10419 | * rtnl_lock(); | |
10420 | * ... | |
10421 | * register_netdevice(x1); | |
10422 | * register_netdevice(x2); | |
10423 | * ... | |
10424 | * unregister_netdevice(y1); | |
10425 | * unregister_netdevice(y2); | |
10426 | * ... | |
10427 | * rtnl_unlock(); | |
10428 | * free_netdev(y1); | |
10429 | * free_netdev(y2); | |
10430 | * | |
58ec3b4d | 10431 | * We are invoked by rtnl_unlock(). |
1da177e4 | 10432 | * This allows us to deal with problems: |
b17a7c17 | 10433 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
10434 | * without deadlocking with linkwatch via keventd. |
10435 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
10436 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
10437 | * |
10438 | * We must not return until all unregister events added during | |
10439 | * the interval the lock was held have been completed. | |
1da177e4 | 10440 | */ |
1da177e4 LT |
10441 | void netdev_run_todo(void) |
10442 | { | |
626ab0e6 | 10443 | struct list_head list; |
1fc70edb TY |
10444 | #ifdef CONFIG_LOCKDEP |
10445 | struct list_head unlink_list; | |
10446 | ||
10447 | list_replace_init(&net_unlink_list, &unlink_list); | |
10448 | ||
10449 | while (!list_empty(&unlink_list)) { | |
10450 | struct net_device *dev = list_first_entry(&unlink_list, | |
10451 | struct net_device, | |
10452 | unlink_list); | |
0e8b8d6a | 10453 | list_del_init(&dev->unlink_list); |
1fc70edb TY |
10454 | dev->nested_level = dev->lower_level - 1; |
10455 | } | |
10456 | #endif | |
1da177e4 | 10457 | |
1da177e4 | 10458 | /* Snapshot list, allow later requests */ |
626ab0e6 | 10459 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
10460 | |
10461 | __rtnl_unlock(); | |
626ab0e6 | 10462 | |
0115e8e3 ED |
10463 | |
10464 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
10465 | if (!list_empty(&list)) |
10466 | rcu_barrier(); | |
10467 | ||
1da177e4 LT |
10468 | while (!list_empty(&list)) { |
10469 | struct net_device *dev | |
e5e26d75 | 10470 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
10471 | list_del(&dev->todo_list); |
10472 | ||
b17a7c17 | 10473 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 10474 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
10475 | dev->name, dev->reg_state); |
10476 | dump_stack(); | |
10477 | continue; | |
10478 | } | |
1da177e4 | 10479 | |
b17a7c17 | 10480 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 10481 | |
b17a7c17 | 10482 | netdev_wait_allrefs(dev); |
1da177e4 | 10483 | |
b17a7c17 | 10484 | /* paranoia */ |
29b4433d | 10485 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
10486 | BUG_ON(!list_empty(&dev->ptype_all)); |
10487 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
10488 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
10489 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
330c7272 | 10490 | #if IS_ENABLED(CONFIG_DECNET) |
547b792c | 10491 | WARN_ON(dev->dn_ptr); |
330c7272 | 10492 | #endif |
cf124db5 DM |
10493 | if (dev->priv_destructor) |
10494 | dev->priv_destructor(dev); | |
10495 | if (dev->needs_free_netdev) | |
10496 | free_netdev(dev); | |
9093bbb2 | 10497 | |
50624c93 EB |
10498 | /* Report a network device has been unregistered */ |
10499 | rtnl_lock(); | |
10500 | dev_net(dev)->dev_unreg_count--; | |
10501 | __rtnl_unlock(); | |
10502 | wake_up(&netdev_unregistering_wq); | |
10503 | ||
9093bbb2 SH |
10504 | /* Free network device */ |
10505 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 10506 | } |
1da177e4 LT |
10507 | } |
10508 | ||
9256645a JW |
10509 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
10510 | * all the same fields in the same order as net_device_stats, with only | |
10511 | * the type differing, but rtnl_link_stats64 may have additional fields | |
10512 | * at the end for newer counters. | |
3cfde79c | 10513 | */ |
77a1abf5 ED |
10514 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
10515 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
10516 | { |
10517 | #if BITS_PER_LONG == 64 | |
9256645a | 10518 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
9af9959e | 10519 | memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); |
9256645a JW |
10520 | /* zero out counters that only exist in rtnl_link_stats64 */ |
10521 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
10522 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 10523 | #else |
9256645a | 10524 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
10525 | const unsigned long *src = (const unsigned long *)netdev_stats; |
10526 | u64 *dst = (u64 *)stats64; | |
10527 | ||
9256645a | 10528 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
10529 | for (i = 0; i < n; i++) |
10530 | dst[i] = src[i]; | |
9256645a JW |
10531 | /* zero out counters that only exist in rtnl_link_stats64 */ |
10532 | memset((char *)stats64 + n * sizeof(u64), 0, | |
10533 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
10534 | #endif |
10535 | } | |
77a1abf5 | 10536 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 10537 | |
eeda3fd6 SH |
10538 | /** |
10539 | * dev_get_stats - get network device statistics | |
10540 | * @dev: device to get statistics from | |
28172739 | 10541 | * @storage: place to store stats |
eeda3fd6 | 10542 | * |
d7753516 BH |
10543 | * Get network statistics from device. Return @storage. |
10544 | * The device driver may provide its own method by setting | |
10545 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
10546 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 10547 | */ |
d7753516 BH |
10548 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
10549 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 10550 | { |
eeda3fd6 SH |
10551 | const struct net_device_ops *ops = dev->netdev_ops; |
10552 | ||
28172739 ED |
10553 | if (ops->ndo_get_stats64) { |
10554 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
10555 | ops->ndo_get_stats64(dev, storage); |
10556 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 10557 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
10558 | } else { |
10559 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 10560 | } |
6f64ec74 ED |
10561 | storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); |
10562 | storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); | |
10563 | storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); | |
28172739 | 10564 | return storage; |
c45d286e | 10565 | } |
eeda3fd6 | 10566 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 10567 | |
44fa32f0 HK |
10568 | /** |
10569 | * dev_fetch_sw_netstats - get per-cpu network device statistics | |
10570 | * @s: place to store stats | |
10571 | * @netstats: per-cpu network stats to read from | |
10572 | * | |
10573 | * Read per-cpu network statistics and populate the related fields in @s. | |
10574 | */ | |
10575 | void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, | |
10576 | const struct pcpu_sw_netstats __percpu *netstats) | |
10577 | { | |
10578 | int cpu; | |
10579 | ||
10580 | for_each_possible_cpu(cpu) { | |
10581 | const struct pcpu_sw_netstats *stats; | |
10582 | struct pcpu_sw_netstats tmp; | |
10583 | unsigned int start; | |
10584 | ||
10585 | stats = per_cpu_ptr(netstats, cpu); | |
10586 | do { | |
10587 | start = u64_stats_fetch_begin_irq(&stats->syncp); | |
10588 | tmp.rx_packets = stats->rx_packets; | |
10589 | tmp.rx_bytes = stats->rx_bytes; | |
10590 | tmp.tx_packets = stats->tx_packets; | |
10591 | tmp.tx_bytes = stats->tx_bytes; | |
10592 | } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); | |
10593 | ||
10594 | s->rx_packets += tmp.rx_packets; | |
10595 | s->rx_bytes += tmp.rx_bytes; | |
10596 | s->tx_packets += tmp.tx_packets; | |
10597 | s->tx_bytes += tmp.tx_bytes; | |
10598 | } | |
10599 | } | |
10600 | EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); | |
10601 | ||
a1839426 HK |
10602 | /** |
10603 | * dev_get_tstats64 - ndo_get_stats64 implementation | |
10604 | * @dev: device to get statistics from | |
10605 | * @s: place to store stats | |
10606 | * | |
10607 | * Populate @s from dev->stats and dev->tstats. Can be used as | |
10608 | * ndo_get_stats64() callback. | |
10609 | */ | |
10610 | void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) | |
10611 | { | |
10612 | netdev_stats_to_stats64(s, &dev->stats); | |
10613 | dev_fetch_sw_netstats(s, dev->tstats); | |
10614 | } | |
10615 | EXPORT_SYMBOL_GPL(dev_get_tstats64); | |
10616 | ||
24824a09 | 10617 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 10618 | { |
24824a09 | 10619 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 10620 | |
24824a09 ED |
10621 | #ifdef CONFIG_NET_CLS_ACT |
10622 | if (queue) | |
10623 | return queue; | |
10624 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
10625 | if (!queue) | |
10626 | return NULL; | |
10627 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 10628 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
10629 | queue->qdisc_sleeping = &noop_qdisc; |
10630 | rcu_assign_pointer(dev->ingress_queue, queue); | |
10631 | #endif | |
10632 | return queue; | |
bb949fbd DM |
10633 | } |
10634 | ||
2c60db03 ED |
10635 | static const struct ethtool_ops default_ethtool_ops; |
10636 | ||
d07d7507 SG |
10637 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
10638 | const struct ethtool_ops *ops) | |
10639 | { | |
10640 | if (dev->ethtool_ops == &default_ethtool_ops) | |
10641 | dev->ethtool_ops = ops; | |
10642 | } | |
10643 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
10644 | ||
74d332c1 ED |
10645 | void netdev_freemem(struct net_device *dev) |
10646 | { | |
10647 | char *addr = (char *)dev - dev->padded; | |
10648 | ||
4cb28970 | 10649 | kvfree(addr); |
74d332c1 ED |
10650 | } |
10651 | ||
1da177e4 | 10652 | /** |
722c9a0c | 10653 | * alloc_netdev_mqs - allocate network device |
10654 | * @sizeof_priv: size of private data to allocate space for | |
10655 | * @name: device name format string | |
10656 | * @name_assign_type: origin of device name | |
10657 | * @setup: callback to initialize device | |
10658 | * @txqs: the number of TX subqueues to allocate | |
10659 | * @rxqs: the number of RX subqueues to allocate | |
10660 | * | |
10661 | * Allocates a struct net_device with private data area for driver use | |
10662 | * and performs basic initialization. Also allocates subqueue structs | |
10663 | * for each queue on the device. | |
1da177e4 | 10664 | */ |
36909ea4 | 10665 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 10666 | unsigned char name_assign_type, |
36909ea4 TH |
10667 | void (*setup)(struct net_device *), |
10668 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 10669 | { |
1da177e4 | 10670 | struct net_device *dev; |
52a59bd5 | 10671 | unsigned int alloc_size; |
1ce8e7b5 | 10672 | struct net_device *p; |
1da177e4 | 10673 | |
b6fe17d6 SH |
10674 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
10675 | ||
36909ea4 | 10676 | if (txqs < 1) { |
7b6cd1ce | 10677 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
10678 | return NULL; |
10679 | } | |
10680 | ||
36909ea4 | 10681 | if (rxqs < 1) { |
7b6cd1ce | 10682 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
10683 | return NULL; |
10684 | } | |
36909ea4 | 10685 | |
fd2ea0a7 | 10686 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
10687 | if (sizeof_priv) { |
10688 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 10689 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
10690 | alloc_size += sizeof_priv; |
10691 | } | |
10692 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 10693 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 10694 | |
dcda9b04 | 10695 | p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
62b5942a | 10696 | if (!p) |
1da177e4 | 10697 | return NULL; |
1da177e4 | 10698 | |
1ce8e7b5 | 10699 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 10700 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 10701 | |
29b4433d ED |
10702 | dev->pcpu_refcnt = alloc_percpu(int); |
10703 | if (!dev->pcpu_refcnt) | |
74d332c1 | 10704 | goto free_dev; |
ab9c73cc | 10705 | |
ab9c73cc | 10706 | if (dev_addr_init(dev)) |
29b4433d | 10707 | goto free_pcpu; |
ab9c73cc | 10708 | |
22bedad3 | 10709 | dev_mc_init(dev); |
a748ee24 | 10710 | dev_uc_init(dev); |
ccffad25 | 10711 | |
c346dca1 | 10712 | dev_net_set(dev, &init_net); |
1da177e4 | 10713 | |
8d3bdbd5 | 10714 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 10715 | dev->gso_max_segs = GSO_MAX_SEGS; |
5343da4c TY |
10716 | dev->upper_level = 1; |
10717 | dev->lower_level = 1; | |
1fc70edb TY |
10718 | #ifdef CONFIG_LOCKDEP |
10719 | dev->nested_level = 0; | |
10720 | INIT_LIST_HEAD(&dev->unlink_list); | |
10721 | #endif | |
8d3bdbd5 | 10722 | |
8d3bdbd5 DM |
10723 | INIT_LIST_HEAD(&dev->napi_list); |
10724 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 10725 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 10726 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
10727 | INIT_LIST_HEAD(&dev->adj_list.upper); |
10728 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
10729 | INIT_LIST_HEAD(&dev->ptype_all); |
10730 | INIT_LIST_HEAD(&dev->ptype_specific); | |
93642e14 | 10731 | INIT_LIST_HEAD(&dev->net_notifier_list); |
59cc1f61 JK |
10732 | #ifdef CONFIG_NET_SCHED |
10733 | hash_init(dev->qdisc_hash); | |
10734 | #endif | |
02875878 | 10735 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
10736 | setup(dev); |
10737 | ||
a813104d | 10738 | if (!dev->tx_queue_len) { |
f84bb1ea | 10739 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 10740 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 10741 | } |
906470c1 | 10742 | |
36909ea4 TH |
10743 | dev->num_tx_queues = txqs; |
10744 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 10745 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 10746 | goto free_all; |
e8a0464c | 10747 | |
36909ea4 TH |
10748 | dev->num_rx_queues = rxqs; |
10749 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 10750 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 10751 | goto free_all; |
0a9627f2 | 10752 | |
1da177e4 | 10753 | strcpy(dev->name, name); |
c835a677 | 10754 | dev->name_assign_type = name_assign_type; |
cbda10fa | 10755 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
10756 | if (!dev->ethtool_ops) |
10757 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 | 10758 | |
357b6cc5 | 10759 | nf_hook_ingress_init(dev); |
e687ad60 | 10760 | |
1da177e4 | 10761 | return dev; |
ab9c73cc | 10762 | |
8d3bdbd5 DM |
10763 | free_all: |
10764 | free_netdev(dev); | |
10765 | return NULL; | |
10766 | ||
29b4433d ED |
10767 | free_pcpu: |
10768 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
10769 | free_dev: |
10770 | netdev_freemem(dev); | |
ab9c73cc | 10771 | return NULL; |
1da177e4 | 10772 | } |
36909ea4 | 10773 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
10774 | |
10775 | /** | |
722c9a0c | 10776 | * free_netdev - free network device |
10777 | * @dev: device | |
1da177e4 | 10778 | * |
722c9a0c | 10779 | * This function does the last stage of destroying an allocated device |
10780 | * interface. The reference to the device object is released. If this | |
10781 | * is the last reference then it will be freed.Must be called in process | |
10782 | * context. | |
1da177e4 LT |
10783 | */ |
10784 | void free_netdev(struct net_device *dev) | |
10785 | { | |
d565b0a1 HX |
10786 | struct napi_struct *p, *n; |
10787 | ||
93d05d4a | 10788 | might_sleep(); |
c269a24c JK |
10789 | |
10790 | /* When called immediately after register_netdevice() failed the unwind | |
10791 | * handling may still be dismantling the device. Handle that case by | |
10792 | * deferring the free. | |
10793 | */ | |
10794 | if (dev->reg_state == NETREG_UNREGISTERING) { | |
10795 | ASSERT_RTNL(); | |
10796 | dev->needs_free_netdev = true; | |
10797 | return; | |
10798 | } | |
10799 | ||
60877a32 | 10800 | netif_free_tx_queues(dev); |
e817f856 | 10801 | netif_free_rx_queues(dev); |
e8a0464c | 10802 | |
33d480ce | 10803 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 10804 | |
f001fde5 JP |
10805 | /* Flush device addresses */ |
10806 | dev_addr_flush(dev); | |
10807 | ||
d565b0a1 HX |
10808 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
10809 | netif_napi_del(p); | |
10810 | ||
29b4433d ED |
10811 | free_percpu(dev->pcpu_refcnt); |
10812 | dev->pcpu_refcnt = NULL; | |
75ccae62 THJ |
10813 | free_percpu(dev->xdp_bulkq); |
10814 | dev->xdp_bulkq = NULL; | |
29b4433d | 10815 | |
3041a069 | 10816 | /* Compatibility with error handling in drivers */ |
1da177e4 | 10817 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 10818 | netdev_freemem(dev); |
1da177e4 LT |
10819 | return; |
10820 | } | |
10821 | ||
10822 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
10823 | dev->reg_state = NETREG_RELEASED; | |
10824 | ||
43cb76d9 GKH |
10825 | /* will free via device release */ |
10826 | put_device(&dev->dev); | |
1da177e4 | 10827 | } |
d1b19dff | 10828 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 10829 | |
f0db275a SH |
10830 | /** |
10831 | * synchronize_net - Synchronize with packet receive processing | |
10832 | * | |
10833 | * Wait for packets currently being received to be done. | |
10834 | * Does not block later packets from starting. | |
10835 | */ | |
4ec93edb | 10836 | void synchronize_net(void) |
1da177e4 LT |
10837 | { |
10838 | might_sleep(); | |
be3fc413 ED |
10839 | if (rtnl_is_locked()) |
10840 | synchronize_rcu_expedited(); | |
10841 | else | |
10842 | synchronize_rcu(); | |
1da177e4 | 10843 | } |
d1b19dff | 10844 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
10845 | |
10846 | /** | |
44a0873d | 10847 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 10848 | * @dev: device |
44a0873d | 10849 | * @head: list |
6ebfbc06 | 10850 | * |
1da177e4 | 10851 | * This function shuts down a device interface and removes it |
d59b54b1 | 10852 | * from the kernel tables. |
44a0873d | 10853 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
10854 | * |
10855 | * Callers must hold the rtnl semaphore. You may want | |
10856 | * unregister_netdev() instead of this. | |
10857 | */ | |
10858 | ||
44a0873d | 10859 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 10860 | { |
a6620712 HX |
10861 | ASSERT_RTNL(); |
10862 | ||
44a0873d | 10863 | if (head) { |
9fdce099 | 10864 | list_move_tail(&dev->unreg_list, head); |
44a0873d | 10865 | } else { |
037e56bd JK |
10866 | LIST_HEAD(single); |
10867 | ||
10868 | list_add(&dev->unreg_list, &single); | |
0cbe1e57 | 10869 | unregister_netdevice_many(&single); |
44a0873d | 10870 | } |
1da177e4 | 10871 | } |
44a0873d | 10872 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 10873 | |
9b5e383c ED |
10874 | /** |
10875 | * unregister_netdevice_many - unregister many devices | |
10876 | * @head: list of devices | |
87757a91 ED |
10877 | * |
10878 | * Note: As most callers use a stack allocated list_head, | |
10879 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
10880 | */ |
10881 | void unregister_netdevice_many(struct list_head *head) | |
bcfe2f1a JK |
10882 | { |
10883 | struct net_device *dev, *tmp; | |
10884 | LIST_HEAD(close_head); | |
10885 | ||
10886 | BUG_ON(dev_boot_phase); | |
10887 | ASSERT_RTNL(); | |
10888 | ||
0cbe1e57 JK |
10889 | if (list_empty(head)) |
10890 | return; | |
10891 | ||
bcfe2f1a JK |
10892 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
10893 | /* Some devices call without registering | |
10894 | * for initialization unwind. Remove those | |
10895 | * devices and proceed with the remaining. | |
10896 | */ | |
10897 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
10898 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", | |
10899 | dev->name, dev); | |
10900 | ||
10901 | WARN_ON(1); | |
10902 | list_del(&dev->unreg_list); | |
10903 | continue; | |
10904 | } | |
10905 | dev->dismantle = true; | |
10906 | BUG_ON(dev->reg_state != NETREG_REGISTERED); | |
10907 | } | |
10908 | ||
10909 | /* If device is running, close it first. */ | |
10910 | list_for_each_entry(dev, head, unreg_list) | |
10911 | list_add_tail(&dev->close_list, &close_head); | |
10912 | dev_close_many(&close_head, true); | |
10913 | ||
10914 | list_for_each_entry(dev, head, unreg_list) { | |
10915 | /* And unlink it from device chain. */ | |
10916 | unlist_netdevice(dev); | |
10917 | ||
10918 | dev->reg_state = NETREG_UNREGISTERING; | |
10919 | } | |
10920 | flush_all_backlogs(); | |
10921 | ||
10922 | synchronize_net(); | |
10923 | ||
10924 | list_for_each_entry(dev, head, unreg_list) { | |
10925 | struct sk_buff *skb = NULL; | |
10926 | ||
10927 | /* Shutdown queueing discipline. */ | |
10928 | dev_shutdown(dev); | |
10929 | ||
10930 | dev_xdp_uninstall(dev); | |
10931 | ||
10932 | /* Notify protocols, that we are about to destroy | |
10933 | * this device. They should clean all the things. | |
10934 | */ | |
10935 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
10936 | ||
10937 | if (!dev->rtnl_link_ops || | |
10938 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
10939 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, | |
10940 | GFP_KERNEL, NULL, 0); | |
10941 | ||
10942 | /* | |
10943 | * Flush the unicast and multicast chains | |
10944 | */ | |
10945 | dev_uc_flush(dev); | |
10946 | dev_mc_flush(dev); | |
10947 | ||
10948 | netdev_name_node_alt_flush(dev); | |
10949 | netdev_name_node_free(dev->name_node); | |
10950 | ||
10951 | if (dev->netdev_ops->ndo_uninit) | |
10952 | dev->netdev_ops->ndo_uninit(dev); | |
10953 | ||
10954 | if (skb) | |
10955 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
10956 | ||
10957 | /* Notifier chain MUST detach us all upper devices. */ | |
10958 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
10959 | WARN_ON(netdev_has_any_lower_dev(dev)); | |
10960 | ||
10961 | /* Remove entries from kobject tree */ | |
10962 | netdev_unregister_kobject(dev); | |
10963 | #ifdef CONFIG_XPS | |
10964 | /* Remove XPS queueing entries */ | |
10965 | netif_reset_xps_queues_gt(dev, 0); | |
10966 | #endif | |
10967 | } | |
10968 | ||
10969 | synchronize_net(); | |
10970 | ||
10971 | list_for_each_entry(dev, head, unreg_list) { | |
10972 | dev_put(dev); | |
10973 | net_set_todo(dev); | |
10974 | } | |
0cbe1e57 JK |
10975 | |
10976 | list_del(head); | |
bcfe2f1a | 10977 | } |
0cbe1e57 | 10978 | EXPORT_SYMBOL(unregister_netdevice_many); |
bcfe2f1a | 10979 | |
1da177e4 LT |
10980 | /** |
10981 | * unregister_netdev - remove device from the kernel | |
10982 | * @dev: device | |
10983 | * | |
10984 | * This function shuts down a device interface and removes it | |
d59b54b1 | 10985 | * from the kernel tables. |
1da177e4 LT |
10986 | * |
10987 | * This is just a wrapper for unregister_netdevice that takes | |
10988 | * the rtnl semaphore. In general you want to use this and not | |
10989 | * unregister_netdevice. | |
10990 | */ | |
10991 | void unregister_netdev(struct net_device *dev) | |
10992 | { | |
10993 | rtnl_lock(); | |
10994 | unregister_netdevice(dev); | |
10995 | rtnl_unlock(); | |
10996 | } | |
1da177e4 LT |
10997 | EXPORT_SYMBOL(unregister_netdev); |
10998 | ||
ce286d32 EB |
10999 | /** |
11000 | * dev_change_net_namespace - move device to different nethost namespace | |
11001 | * @dev: device | |
11002 | * @net: network namespace | |
11003 | * @pat: If not NULL name pattern to try if the current device name | |
11004 | * is already taken in the destination network namespace. | |
11005 | * | |
11006 | * This function shuts down a device interface and moves it | |
11007 | * to a new network namespace. On success 0 is returned, on | |
11008 | * a failure a netagive errno code is returned. | |
11009 | * | |
11010 | * Callers must hold the rtnl semaphore. | |
11011 | */ | |
11012 | ||
11013 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
11014 | { | |
ef6a4c88 | 11015 | struct net *net_old = dev_net(dev); |
38e01b30 | 11016 | int err, new_nsid, new_ifindex; |
ce286d32 EB |
11017 | |
11018 | ASSERT_RTNL(); | |
11019 | ||
11020 | /* Don't allow namespace local devices to be moved. */ | |
11021 | err = -EINVAL; | |
11022 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
11023 | goto out; | |
11024 | ||
11025 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
11026 | if (dev->reg_state != NETREG_REGISTERED) |
11027 | goto out; | |
11028 | ||
11029 | /* Get out if there is nothing todo */ | |
11030 | err = 0; | |
ef6a4c88 | 11031 | if (net_eq(net_old, net)) |
ce286d32 EB |
11032 | goto out; |
11033 | ||
11034 | /* Pick the destination device name, and ensure | |
11035 | * we can use it in the destination network namespace. | |
11036 | */ | |
11037 | err = -EEXIST; | |
d9031024 | 11038 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
11039 | /* We get here if we can't use the current device name */ |
11040 | if (!pat) | |
11041 | goto out; | |
7892bd08 LR |
11042 | err = dev_get_valid_name(net, dev, pat); |
11043 | if (err < 0) | |
ce286d32 EB |
11044 | goto out; |
11045 | } | |
11046 | ||
11047 | /* | |
11048 | * And now a mini version of register_netdevice unregister_netdevice. | |
11049 | */ | |
11050 | ||
11051 | /* If device is running close it first. */ | |
9b772652 | 11052 | dev_close(dev); |
ce286d32 EB |
11053 | |
11054 | /* And unlink it from device chain */ | |
ce286d32 EB |
11055 | unlist_netdevice(dev); |
11056 | ||
11057 | synchronize_net(); | |
11058 | ||
11059 | /* Shutdown queueing discipline. */ | |
11060 | dev_shutdown(dev); | |
11061 | ||
11062 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 11063 | * this device. They should clean all the things. |
11064 | * | |
11065 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
11066 | * This is wanted because this way 8021q and macvlan know | |
11067 | * the device is just moving and can keep their slaves up. | |
11068 | */ | |
ce286d32 | 11069 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 | 11070 | rcu_barrier(); |
38e01b30 | 11071 | |
d4e4fdf9 | 11072 | new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); |
38e01b30 ND |
11073 | /* If there is an ifindex conflict assign a new one */ |
11074 | if (__dev_get_by_index(net, dev->ifindex)) | |
11075 | new_ifindex = dev_new_index(net); | |
11076 | else | |
11077 | new_ifindex = dev->ifindex; | |
11078 | ||
11079 | rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, | |
11080 | new_ifindex); | |
ce286d32 EB |
11081 | |
11082 | /* | |
11083 | * Flush the unicast and multicast chains | |
11084 | */ | |
a748ee24 | 11085 | dev_uc_flush(dev); |
22bedad3 | 11086 | dev_mc_flush(dev); |
ce286d32 | 11087 | |
4e66ae2e SH |
11088 | /* Send a netdev-removed uevent to the old namespace */ |
11089 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 11090 | netdev_adjacent_del_links(dev); |
4e66ae2e | 11091 | |
93642e14 JP |
11092 | /* Move per-net netdevice notifiers that are following the netdevice */ |
11093 | move_netdevice_notifiers_dev_net(dev, net); | |
11094 | ||
ce286d32 | 11095 | /* Actually switch the network namespace */ |
c346dca1 | 11096 | dev_net_set(dev, net); |
38e01b30 | 11097 | dev->ifindex = new_ifindex; |
ce286d32 | 11098 | |
4e66ae2e SH |
11099 | /* Send a netdev-add uevent to the new namespace */ |
11100 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 11101 | netdev_adjacent_add_links(dev); |
4e66ae2e | 11102 | |
8b41d188 | 11103 | /* Fixup kobjects */ |
a1b3f594 | 11104 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 11105 | WARN_ON(err); |
ce286d32 | 11106 | |
ef6a4c88 CB |
11107 | /* Adapt owner in case owning user namespace of target network |
11108 | * namespace is different from the original one. | |
11109 | */ | |
11110 | err = netdev_change_owner(dev, net_old, net); | |
11111 | WARN_ON(err); | |
11112 | ||
ce286d32 EB |
11113 | /* Add the device back in the hashes */ |
11114 | list_netdevice(dev); | |
11115 | ||
11116 | /* Notify protocols, that a new device appeared. */ | |
11117 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
11118 | ||
d90a909e EB |
11119 | /* |
11120 | * Prevent userspace races by waiting until the network | |
11121 | * device is fully setup before sending notifications. | |
11122 | */ | |
7f294054 | 11123 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 11124 | |
ce286d32 EB |
11125 | synchronize_net(); |
11126 | err = 0; | |
11127 | out: | |
11128 | return err; | |
11129 | } | |
463d0183 | 11130 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 11131 | |
f0bf90de | 11132 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
11133 | { |
11134 | struct sk_buff **list_skb; | |
1da177e4 | 11135 | struct sk_buff *skb; |
f0bf90de | 11136 | unsigned int cpu; |
97d8b6e3 | 11137 | struct softnet_data *sd, *oldsd, *remsd = NULL; |
1da177e4 | 11138 | |
1da177e4 LT |
11139 | local_irq_disable(); |
11140 | cpu = smp_processor_id(); | |
11141 | sd = &per_cpu(softnet_data, cpu); | |
11142 | oldsd = &per_cpu(softnet_data, oldcpu); | |
11143 | ||
11144 | /* Find end of our completion_queue. */ | |
11145 | list_skb = &sd->completion_queue; | |
11146 | while (*list_skb) | |
11147 | list_skb = &(*list_skb)->next; | |
11148 | /* Append completion queue from offline CPU. */ | |
11149 | *list_skb = oldsd->completion_queue; | |
11150 | oldsd->completion_queue = NULL; | |
11151 | ||
1da177e4 | 11152 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
11153 | if (oldsd->output_queue) { |
11154 | *sd->output_queue_tailp = oldsd->output_queue; | |
11155 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
11156 | oldsd->output_queue = NULL; | |
11157 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
11158 | } | |
ac64da0b ED |
11159 | /* Append NAPI poll list from offline CPU, with one exception : |
11160 | * process_backlog() must be called by cpu owning percpu backlog. | |
11161 | * We properly handle process_queue & input_pkt_queue later. | |
11162 | */ | |
11163 | while (!list_empty(&oldsd->poll_list)) { | |
11164 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
11165 | struct napi_struct, | |
11166 | poll_list); | |
11167 | ||
11168 | list_del_init(&napi->poll_list); | |
11169 | if (napi->poll == process_backlog) | |
11170 | napi->state = 0; | |
11171 | else | |
11172 | ____napi_schedule(sd, napi); | |
264524d5 | 11173 | } |
1da177e4 LT |
11174 | |
11175 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
11176 | local_irq_enable(); | |
11177 | ||
773fc8f6 | 11178 | #ifdef CONFIG_RPS |
11179 | remsd = oldsd->rps_ipi_list; | |
11180 | oldsd->rps_ipi_list = NULL; | |
11181 | #endif | |
11182 | /* send out pending IPI's on offline CPU */ | |
11183 | net_rps_send_ipi(remsd); | |
11184 | ||
1da177e4 | 11185 | /* Process offline CPU's input_pkt_queue */ |
76cc8b13 | 11186 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 11187 | netif_rx_ni(skb); |
76cc8b13 | 11188 | input_queue_head_incr(oldsd); |
fec5e652 | 11189 | } |
ac64da0b | 11190 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 11191 | netif_rx_ni(skb); |
76cc8b13 TH |
11192 | input_queue_head_incr(oldsd); |
11193 | } | |
1da177e4 | 11194 | |
f0bf90de | 11195 | return 0; |
1da177e4 | 11196 | } |
1da177e4 | 11197 | |
7f353bf2 | 11198 | /** |
b63365a2 HX |
11199 | * netdev_increment_features - increment feature set by one |
11200 | * @all: current feature set | |
11201 | * @one: new feature set | |
11202 | * @mask: mask feature set | |
7f353bf2 HX |
11203 | * |
11204 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
11205 | * @one to the master device with current feature set @all. Will not |
11206 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 11207 | */ |
c8f44aff MM |
11208 | netdev_features_t netdev_increment_features(netdev_features_t all, |
11209 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 11210 | { |
c8cd0989 | 11211 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 11212 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 11213 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 11214 | |
a188222b | 11215 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 11216 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 11217 | |
1742f183 | 11218 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
11219 | if (all & NETIF_F_HW_CSUM) |
11220 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
11221 | |
11222 | return all; | |
11223 | } | |
b63365a2 | 11224 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 11225 | |
430f03cd | 11226 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
11227 | { |
11228 | int i; | |
11229 | struct hlist_head *hash; | |
11230 | ||
6da2ec56 | 11231 | hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); |
30d97d35 PE |
11232 | if (hash != NULL) |
11233 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
11234 | INIT_HLIST_HEAD(&hash[i]); | |
11235 | ||
11236 | return hash; | |
11237 | } | |
11238 | ||
881d966b | 11239 | /* Initialize per network namespace state */ |
4665079c | 11240 | static int __net_init netdev_init(struct net *net) |
881d966b | 11241 | { |
d9f37d01 | 11242 | BUILD_BUG_ON(GRO_HASH_BUCKETS > |
c593642c | 11243 | 8 * sizeof_field(struct napi_struct, gro_bitmask)); |
d9f37d01 | 11244 | |
734b6541 RM |
11245 | if (net != &init_net) |
11246 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 11247 | |
30d97d35 PE |
11248 | net->dev_name_head = netdev_create_hash(); |
11249 | if (net->dev_name_head == NULL) | |
11250 | goto err_name; | |
881d966b | 11251 | |
30d97d35 PE |
11252 | net->dev_index_head = netdev_create_hash(); |
11253 | if (net->dev_index_head == NULL) | |
11254 | goto err_idx; | |
881d966b | 11255 | |
a30c7b42 JP |
11256 | RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); |
11257 | ||
881d966b | 11258 | return 0; |
30d97d35 PE |
11259 | |
11260 | err_idx: | |
11261 | kfree(net->dev_name_head); | |
11262 | err_name: | |
11263 | return -ENOMEM; | |
881d966b EB |
11264 | } |
11265 | ||
f0db275a SH |
11266 | /** |
11267 | * netdev_drivername - network driver for the device | |
11268 | * @dev: network device | |
f0db275a SH |
11269 | * |
11270 | * Determine network driver for device. | |
11271 | */ | |
3019de12 | 11272 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 11273 | { |
cf04a4c7 SH |
11274 | const struct device_driver *driver; |
11275 | const struct device *parent; | |
3019de12 | 11276 | const char *empty = ""; |
6579e57b AV |
11277 | |
11278 | parent = dev->dev.parent; | |
6579e57b | 11279 | if (!parent) |
3019de12 | 11280 | return empty; |
6579e57b AV |
11281 | |
11282 | driver = parent->driver; | |
11283 | if (driver && driver->name) | |
3019de12 DM |
11284 | return driver->name; |
11285 | return empty; | |
6579e57b AV |
11286 | } |
11287 | ||
6ea754eb JP |
11288 | static void __netdev_printk(const char *level, const struct net_device *dev, |
11289 | struct va_format *vaf) | |
256df2f3 | 11290 | { |
b004ff49 | 11291 | if (dev && dev->dev.parent) { |
6ea754eb JP |
11292 | dev_printk_emit(level[1] - '0', |
11293 | dev->dev.parent, | |
11294 | "%s %s %s%s: %pV", | |
11295 | dev_driver_string(dev->dev.parent), | |
11296 | dev_name(dev->dev.parent), | |
11297 | netdev_name(dev), netdev_reg_state(dev), | |
11298 | vaf); | |
b004ff49 | 11299 | } else if (dev) { |
6ea754eb JP |
11300 | printk("%s%s%s: %pV", |
11301 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 11302 | } else { |
6ea754eb | 11303 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 11304 | } |
256df2f3 JP |
11305 | } |
11306 | ||
6ea754eb JP |
11307 | void netdev_printk(const char *level, const struct net_device *dev, |
11308 | const char *format, ...) | |
256df2f3 JP |
11309 | { |
11310 | struct va_format vaf; | |
11311 | va_list args; | |
256df2f3 JP |
11312 | |
11313 | va_start(args, format); | |
11314 | ||
11315 | vaf.fmt = format; | |
11316 | vaf.va = &args; | |
11317 | ||
6ea754eb | 11318 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 11319 | |
256df2f3 | 11320 | va_end(args); |
256df2f3 JP |
11321 | } |
11322 | EXPORT_SYMBOL(netdev_printk); | |
11323 | ||
11324 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 11325 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 11326 | { \ |
256df2f3 JP |
11327 | struct va_format vaf; \ |
11328 | va_list args; \ | |
11329 | \ | |
11330 | va_start(args, fmt); \ | |
11331 | \ | |
11332 | vaf.fmt = fmt; \ | |
11333 | vaf.va = &args; \ | |
11334 | \ | |
6ea754eb | 11335 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 11336 | \ |
256df2f3 | 11337 | va_end(args); \ |
256df2f3 JP |
11338 | } \ |
11339 | EXPORT_SYMBOL(func); | |
11340 | ||
11341 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
11342 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
11343 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
11344 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
11345 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
11346 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
11347 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
11348 | ||
4665079c | 11349 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
11350 | { |
11351 | kfree(net->dev_name_head); | |
11352 | kfree(net->dev_index_head); | |
ee21b18b VA |
11353 | if (net != &init_net) |
11354 | WARN_ON_ONCE(!list_empty(&net->dev_base_head)); | |
881d966b EB |
11355 | } |
11356 | ||
022cbae6 | 11357 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
11358 | .init = netdev_init, |
11359 | .exit = netdev_exit, | |
11360 | }; | |
11361 | ||
4665079c | 11362 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 11363 | { |
e008b5fc | 11364 | struct net_device *dev, *aux; |
ce286d32 | 11365 | /* |
e008b5fc | 11366 | * Push all migratable network devices back to the |
ce286d32 EB |
11367 | * initial network namespace |
11368 | */ | |
11369 | rtnl_lock(); | |
e008b5fc | 11370 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 11371 | int err; |
aca51397 | 11372 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
11373 | |
11374 | /* Ignore unmoveable devices (i.e. loopback) */ | |
11375 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
11376 | continue; | |
11377 | ||
e008b5fc | 11378 | /* Leave virtual devices for the generic cleanup */ |
3a5ca857 | 11379 | if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) |
e008b5fc | 11380 | continue; |
d0c082ce | 11381 | |
25985edc | 11382 | /* Push remaining network devices to init_net */ |
aca51397 | 11383 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
55b40dbf JP |
11384 | if (__dev_get_by_name(&init_net, fb_name)) |
11385 | snprintf(fb_name, IFNAMSIZ, "dev%%d"); | |
aca51397 | 11386 | err = dev_change_net_namespace(dev, &init_net, fb_name); |
ce286d32 | 11387 | if (err) { |
7b6cd1ce JP |
11388 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
11389 | __func__, dev->name, err); | |
aca51397 | 11390 | BUG(); |
ce286d32 EB |
11391 | } |
11392 | } | |
11393 | rtnl_unlock(); | |
11394 | } | |
11395 | ||
50624c93 EB |
11396 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
11397 | { | |
11398 | /* Return with the rtnl_lock held when there are no network | |
11399 | * devices unregistering in any network namespace in net_list. | |
11400 | */ | |
11401 | struct net *net; | |
11402 | bool unregistering; | |
ff960a73 | 11403 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 11404 | |
ff960a73 | 11405 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 11406 | for (;;) { |
50624c93 EB |
11407 | unregistering = false; |
11408 | rtnl_lock(); | |
11409 | list_for_each_entry(net, net_list, exit_list) { | |
11410 | if (net->dev_unreg_count > 0) { | |
11411 | unregistering = true; | |
11412 | break; | |
11413 | } | |
11414 | } | |
11415 | if (!unregistering) | |
11416 | break; | |
11417 | __rtnl_unlock(); | |
ff960a73 PZ |
11418 | |
11419 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 11420 | } |
ff960a73 | 11421 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
11422 | } |
11423 | ||
04dc7f6b EB |
11424 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
11425 | { | |
11426 | /* At exit all network devices most be removed from a network | |
b595076a | 11427 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
11428 | * Do this across as many network namespaces as possible to |
11429 | * improve batching efficiency. | |
11430 | */ | |
11431 | struct net_device *dev; | |
11432 | struct net *net; | |
11433 | LIST_HEAD(dev_kill_list); | |
11434 | ||
50624c93 EB |
11435 | /* To prevent network device cleanup code from dereferencing |
11436 | * loopback devices or network devices that have been freed | |
11437 | * wait here for all pending unregistrations to complete, | |
11438 | * before unregistring the loopback device and allowing the | |
11439 | * network namespace be freed. | |
11440 | * | |
11441 | * The netdev todo list containing all network devices | |
11442 | * unregistrations that happen in default_device_exit_batch | |
11443 | * will run in the rtnl_unlock() at the end of | |
11444 | * default_device_exit_batch. | |
11445 | */ | |
11446 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
11447 | list_for_each_entry(net, net_list, exit_list) { |
11448 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 11449 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
11450 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
11451 | else | |
11452 | unregister_netdevice_queue(dev, &dev_kill_list); | |
11453 | } | |
11454 | } | |
11455 | unregister_netdevice_many(&dev_kill_list); | |
11456 | rtnl_unlock(); | |
11457 | } | |
11458 | ||
022cbae6 | 11459 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 11460 | .exit = default_device_exit, |
04dc7f6b | 11461 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
11462 | }; |
11463 | ||
1da177e4 LT |
11464 | /* |
11465 | * Initialize the DEV module. At boot time this walks the device list and | |
11466 | * unhooks any devices that fail to initialise (normally hardware not | |
11467 | * present) and leaves us with a valid list of present and active devices. | |
11468 | * | |
11469 | */ | |
11470 | ||
11471 | /* | |
11472 | * This is called single threaded during boot, so no need | |
11473 | * to take the rtnl semaphore. | |
11474 | */ | |
11475 | static int __init net_dev_init(void) | |
11476 | { | |
11477 | int i, rc = -ENOMEM; | |
11478 | ||
11479 | BUG_ON(!dev_boot_phase); | |
11480 | ||
1da177e4 LT |
11481 | if (dev_proc_init()) |
11482 | goto out; | |
11483 | ||
8b41d188 | 11484 | if (netdev_kobject_init()) |
1da177e4 LT |
11485 | goto out; |
11486 | ||
11487 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 11488 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
11489 | INIT_LIST_HEAD(&ptype_base[i]); |
11490 | ||
62532da9 VY |
11491 | INIT_LIST_HEAD(&offload_base); |
11492 | ||
881d966b EB |
11493 | if (register_pernet_subsys(&netdev_net_ops)) |
11494 | goto out; | |
1da177e4 LT |
11495 | |
11496 | /* | |
11497 | * Initialise the packet receive queues. | |
11498 | */ | |
11499 | ||
6f912042 | 11500 | for_each_possible_cpu(i) { |
41852497 | 11501 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 11502 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 11503 | |
41852497 ED |
11504 | INIT_WORK(flush, flush_backlog); |
11505 | ||
e36fa2f7 | 11506 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 11507 | skb_queue_head_init(&sd->process_queue); |
f53c7239 SK |
11508 | #ifdef CONFIG_XFRM_OFFLOAD |
11509 | skb_queue_head_init(&sd->xfrm_backlog); | |
11510 | #endif | |
e36fa2f7 | 11511 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 11512 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 11513 | #ifdef CONFIG_RPS |
545b8c8d | 11514 | INIT_CSD(&sd->csd, rps_trigger_softirq, sd); |
e36fa2f7 | 11515 | sd->cpu = i; |
1e94d72f | 11516 | #endif |
0a9627f2 | 11517 | |
7c4ec749 | 11518 | init_gro_hash(&sd->backlog); |
e36fa2f7 ED |
11519 | sd->backlog.poll = process_backlog; |
11520 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
11521 | } |
11522 | ||
1da177e4 LT |
11523 | dev_boot_phase = 0; |
11524 | ||
505d4f73 EB |
11525 | /* The loopback device is special if any other network devices |
11526 | * is present in a network namespace the loopback device must | |
11527 | * be present. Since we now dynamically allocate and free the | |
11528 | * loopback device ensure this invariant is maintained by | |
11529 | * keeping the loopback device as the first device on the | |
11530 | * list of network devices. Ensuring the loopback devices | |
11531 | * is the first device that appears and the last network device | |
11532 | * that disappears. | |
11533 | */ | |
11534 | if (register_pernet_device(&loopback_net_ops)) | |
11535 | goto out; | |
11536 | ||
11537 | if (register_pernet_device(&default_device_ops)) | |
11538 | goto out; | |
11539 | ||
962cf36c CM |
11540 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
11541 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 11542 | |
f0bf90de SAS |
11543 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
11544 | NULL, dev_cpu_dead); | |
11545 | WARN_ON(rc < 0); | |
1da177e4 LT |
11546 | rc = 0; |
11547 | out: | |
11548 | return rc; | |
11549 | } | |
11550 | ||
11551 | subsys_initcall(net_dev_init); |