]>
Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 | 2 | /* |
722c9a0c | 3 | * NET3 Protocol independent device support routines. |
1da177e4 | 4 | * |
1da177e4 | 5 | * Derived from the non IP parts of dev.c 1.0.19 |
722c9a0c | 6 | * Authors: Ross Biro |
1da177e4 LT |
7 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
8 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
9 | * | |
10 | * Additional Authors: | |
11 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
12 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
13 | * David Hinds <dahinds@users.sourceforge.net> | |
14 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
15 | * Adam Sulmicki <adam@cfar.umd.edu> | |
16 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
17 | * | |
18 | * Changes: | |
19 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
722c9a0c | 20 | * to 2 if register_netdev gets called |
21 | * before net_dev_init & also removed a | |
22 | * few lines of code in the process. | |
1da177e4 LT |
23 | * Alan Cox : device private ioctl copies fields back. |
24 | * Alan Cox : Transmit queue code does relevant | |
25 | * stunts to keep the queue safe. | |
26 | * Alan Cox : Fixed double lock. | |
27 | * Alan Cox : Fixed promisc NULL pointer trap | |
28 | * ???????? : Support the full private ioctl range | |
29 | * Alan Cox : Moved ioctl permission check into | |
30 | * drivers | |
31 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
32 | * Alan Cox : 100 backlog just doesn't cut it when | |
33 | * you start doing multicast video 8) | |
34 | * Alan Cox : Rewrote net_bh and list manager. | |
722c9a0c | 35 | * Alan Cox : Fix ETH_P_ALL echoback lengths. |
1da177e4 LT |
36 | * Alan Cox : Took out transmit every packet pass |
37 | * Saved a few bytes in the ioctl handler | |
38 | * Alan Cox : Network driver sets packet type before | |
39 | * calling netif_rx. Saves a function | |
40 | * call a packet. | |
41 | * Alan Cox : Hashed net_bh() | |
42 | * Richard Kooijman: Timestamp fixes. | |
43 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
44 | * Alan Cox : Device lock protection. | |
722c9a0c | 45 | * Alan Cox : Fixed nasty side effect of device close |
1da177e4 LT |
46 | * changes. |
47 | * Rudi Cilibrasi : Pass the right thing to | |
48 | * set_mac_address() | |
49 | * Dave Miller : 32bit quantity for the device lock to | |
50 | * make it work out on a Sparc. | |
51 | * Bjorn Ekwall : Added KERNELD hack. | |
52 | * Alan Cox : Cleaned up the backlog initialise. | |
53 | * Craig Metz : SIOCGIFCONF fix if space for under | |
54 | * 1 device. | |
55 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
56 | * is no device open function. | |
57 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
58 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
59 | * Cyrus Durgin : Cleaned for KMOD | |
60 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
61 | * A network device unload needs to purge | |
62 | * the backlog queue. | |
63 | * Paul Rusty Russell : SIOCSIFNAME | |
64 | * Pekka Riikonen : Netdev boot-time settings code | |
65 | * Andrew Morton : Make unregister_netdevice wait | |
722c9a0c | 66 | * indefinitely on dev->refcnt |
67 | * J Hadi Salim : - Backlog queue sampling | |
1da177e4 LT |
68 | * - netif_rx() feedback |
69 | */ | |
70 | ||
7c0f6ba6 | 71 | #include <linux/uaccess.h> |
1da177e4 | 72 | #include <linux/bitops.h> |
4fc268d2 | 73 | #include <linux/capability.h> |
1da177e4 LT |
74 | #include <linux/cpu.h> |
75 | #include <linux/types.h> | |
76 | #include <linux/kernel.h> | |
08e9897d | 77 | #include <linux/hash.h> |
5a0e3ad6 | 78 | #include <linux/slab.h> |
1da177e4 | 79 | #include <linux/sched.h> |
f1083048 | 80 | #include <linux/sched/mm.h> |
4a3e2f71 | 81 | #include <linux/mutex.h> |
1da177e4 LT |
82 | #include <linux/string.h> |
83 | #include <linux/mm.h> | |
84 | #include <linux/socket.h> | |
85 | #include <linux/sockios.h> | |
86 | #include <linux/errno.h> | |
87 | #include <linux/interrupt.h> | |
88 | #include <linux/if_ether.h> | |
89 | #include <linux/netdevice.h> | |
90 | #include <linux/etherdevice.h> | |
0187bdfb | 91 | #include <linux/ethtool.h> |
1da177e4 | 92 | #include <linux/skbuff.h> |
a7862b45 | 93 | #include <linux/bpf.h> |
b5cdae32 | 94 | #include <linux/bpf_trace.h> |
457c4cbc | 95 | #include <net/net_namespace.h> |
1da177e4 | 96 | #include <net/sock.h> |
02d62e86 | 97 | #include <net/busy_poll.h> |
1da177e4 | 98 | #include <linux/rtnetlink.h> |
1da177e4 | 99 | #include <linux/stat.h> |
1da177e4 | 100 | #include <net/dst.h> |
fc4099f1 | 101 | #include <net/dst_metadata.h> |
1da177e4 | 102 | #include <net/pkt_sched.h> |
87d83093 | 103 | #include <net/pkt_cls.h> |
1da177e4 | 104 | #include <net/checksum.h> |
44540960 | 105 | #include <net/xfrm.h> |
1da177e4 LT |
106 | #include <linux/highmem.h> |
107 | #include <linux/init.h> | |
1da177e4 | 108 | #include <linux/module.h> |
1da177e4 LT |
109 | #include <linux/netpoll.h> |
110 | #include <linux/rcupdate.h> | |
111 | #include <linux/delay.h> | |
1da177e4 | 112 | #include <net/iw_handler.h> |
1da177e4 | 113 | #include <asm/current.h> |
5bdb9886 | 114 | #include <linux/audit.h> |
db217334 | 115 | #include <linux/dmaengine.h> |
f6a78bfc | 116 | #include <linux/err.h> |
c7fa9d18 | 117 | #include <linux/ctype.h> |
723e98b7 | 118 | #include <linux/if_arp.h> |
6de329e2 | 119 | #include <linux/if_vlan.h> |
8f0f2223 | 120 | #include <linux/ip.h> |
ad55dcaf | 121 | #include <net/ip.h> |
25cd9ba0 | 122 | #include <net/mpls.h> |
8f0f2223 DM |
123 | #include <linux/ipv6.h> |
124 | #include <linux/in.h> | |
b6b2fed1 DM |
125 | #include <linux/jhash.h> |
126 | #include <linux/random.h> | |
9cbc1cb8 | 127 | #include <trace/events/napi.h> |
cf66ba58 | 128 | #include <trace/events/net.h> |
07dc22e7 | 129 | #include <trace/events/skb.h> |
caeda9b9 | 130 | #include <linux/inetdevice.h> |
c445477d | 131 | #include <linux/cpu_rmap.h> |
c5905afb | 132 | #include <linux/static_key.h> |
af12fa6e | 133 | #include <linux/hashtable.h> |
60877a32 | 134 | #include <linux/vmalloc.h> |
529d0489 | 135 | #include <linux/if_macvlan.h> |
e7fd2885 | 136 | #include <linux/errqueue.h> |
3b47d303 | 137 | #include <linux/hrtimer.h> |
e687ad60 | 138 | #include <linux/netfilter_ingress.h> |
40e4e713 | 139 | #include <linux/crash_dump.h> |
b72b5bf6 | 140 | #include <linux/sctp.h> |
ae847f40 | 141 | #include <net/udp_tunnel.h> |
6621dd29 | 142 | #include <linux/net_namespace.h> |
aaa5d90b | 143 | #include <linux/indirect_call_wrapper.h> |
af3836df | 144 | #include <net/devlink.h> |
1da177e4 | 145 | |
342709ef PE |
146 | #include "net-sysfs.h" |
147 | ||
d565b0a1 | 148 | #define MAX_GRO_SKBS 8 |
5343da4c | 149 | #define MAX_NEST_DEV 8 |
d565b0a1 | 150 | |
5d38a079 HX |
151 | /* This should be increased if a protocol with a bigger head is added. */ |
152 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
153 | ||
1da177e4 | 154 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 155 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
156 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
157 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 158 | static struct list_head offload_base __read_mostly; |
1da177e4 | 159 | |
ae78dbfa | 160 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 | 161 | static int call_netdevice_notifiers_info(unsigned long val, |
54951194 | 162 | struct netdev_notifier_info *info); |
26372605 PM |
163 | static int call_netdevice_notifiers_extack(unsigned long val, |
164 | struct net_device *dev, | |
165 | struct netlink_ext_ack *extack); | |
90b602f8 | 166 | static struct napi_struct *napi_by_id(unsigned int napi_id); |
ae78dbfa | 167 | |
1da177e4 | 168 | /* |
7562f876 | 169 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
170 | * semaphore. |
171 | * | |
c6d14c84 | 172 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
173 | * |
174 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 175 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
176 | * actual updates. This allows pure readers to access the list even |
177 | * while a writer is preparing to update it. | |
178 | * | |
179 | * To put it another way, dev_base_lock is held for writing only to | |
180 | * protect against pure readers; the rtnl semaphore provides the | |
181 | * protection against other writers. | |
182 | * | |
183 | * See, for example usages, register_netdevice() and | |
184 | * unregister_netdevice(), which must be called with the rtnl | |
185 | * semaphore held. | |
186 | */ | |
1da177e4 | 187 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
188 | EXPORT_SYMBOL(dev_base_lock); |
189 | ||
6c557001 FW |
190 | static DEFINE_MUTEX(ifalias_mutex); |
191 | ||
af12fa6e ET |
192 | /* protects napi_hash addition/deletion and napi_gen_id */ |
193 | static DEFINE_SPINLOCK(napi_hash_lock); | |
194 | ||
52bd2d62 | 195 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 196 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 197 | |
18afa4b0 | 198 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 199 | |
4e985ada TG |
200 | static inline void dev_base_seq_inc(struct net *net) |
201 | { | |
643aa9cb | 202 | while (++net->dev_base_seq == 0) |
203 | ; | |
4e985ada TG |
204 | } |
205 | ||
881d966b | 206 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 207 | { |
8387ff25 | 208 | unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); |
95c96174 | 209 | |
08e9897d | 210 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
211 | } |
212 | ||
881d966b | 213 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 214 | { |
7c28bd0b | 215 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
216 | } |
217 | ||
e36fa2f7 | 218 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
219 | { |
220 | #ifdef CONFIG_RPS | |
e36fa2f7 | 221 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
222 | #endif |
223 | } | |
224 | ||
e36fa2f7 | 225 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
226 | { |
227 | #ifdef CONFIG_RPS | |
e36fa2f7 | 228 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
229 | #endif |
230 | } | |
231 | ||
ff927412 JP |
232 | static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, |
233 | const char *name) | |
234 | { | |
235 | struct netdev_name_node *name_node; | |
236 | ||
237 | name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); | |
238 | if (!name_node) | |
239 | return NULL; | |
240 | INIT_HLIST_NODE(&name_node->hlist); | |
241 | name_node->dev = dev; | |
242 | name_node->name = name; | |
243 | return name_node; | |
244 | } | |
245 | ||
246 | static struct netdev_name_node * | |
247 | netdev_name_node_head_alloc(struct net_device *dev) | |
248 | { | |
36fbf1e5 JP |
249 | struct netdev_name_node *name_node; |
250 | ||
251 | name_node = netdev_name_node_alloc(dev, dev->name); | |
252 | if (!name_node) | |
253 | return NULL; | |
254 | INIT_LIST_HEAD(&name_node->list); | |
255 | return name_node; | |
ff927412 JP |
256 | } |
257 | ||
258 | static void netdev_name_node_free(struct netdev_name_node *name_node) | |
259 | { | |
260 | kfree(name_node); | |
261 | } | |
262 | ||
263 | static void netdev_name_node_add(struct net *net, | |
264 | struct netdev_name_node *name_node) | |
265 | { | |
266 | hlist_add_head_rcu(&name_node->hlist, | |
267 | dev_name_hash(net, name_node->name)); | |
268 | } | |
269 | ||
270 | static void netdev_name_node_del(struct netdev_name_node *name_node) | |
271 | { | |
272 | hlist_del_rcu(&name_node->hlist); | |
273 | } | |
274 | ||
275 | static struct netdev_name_node *netdev_name_node_lookup(struct net *net, | |
276 | const char *name) | |
277 | { | |
278 | struct hlist_head *head = dev_name_hash(net, name); | |
279 | struct netdev_name_node *name_node; | |
280 | ||
281 | hlist_for_each_entry(name_node, head, hlist) | |
282 | if (!strcmp(name_node->name, name)) | |
283 | return name_node; | |
284 | return NULL; | |
285 | } | |
286 | ||
287 | static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, | |
288 | const char *name) | |
289 | { | |
290 | struct hlist_head *head = dev_name_hash(net, name); | |
291 | struct netdev_name_node *name_node; | |
292 | ||
293 | hlist_for_each_entry_rcu(name_node, head, hlist) | |
294 | if (!strcmp(name_node->name, name)) | |
295 | return name_node; | |
296 | return NULL; | |
297 | } | |
298 | ||
36fbf1e5 JP |
299 | int netdev_name_node_alt_create(struct net_device *dev, const char *name) |
300 | { | |
301 | struct netdev_name_node *name_node; | |
302 | struct net *net = dev_net(dev); | |
303 | ||
304 | name_node = netdev_name_node_lookup(net, name); | |
305 | if (name_node) | |
306 | return -EEXIST; | |
307 | name_node = netdev_name_node_alloc(dev, name); | |
308 | if (!name_node) | |
309 | return -ENOMEM; | |
310 | netdev_name_node_add(net, name_node); | |
311 | /* The node that holds dev->name acts as a head of per-device list. */ | |
312 | list_add_tail(&name_node->list, &dev->name_node->list); | |
313 | ||
314 | return 0; | |
315 | } | |
316 | EXPORT_SYMBOL(netdev_name_node_alt_create); | |
317 | ||
318 | static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) | |
319 | { | |
320 | list_del(&name_node->list); | |
321 | netdev_name_node_del(name_node); | |
322 | kfree(name_node->name); | |
323 | netdev_name_node_free(name_node); | |
324 | } | |
325 | ||
326 | int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) | |
327 | { | |
328 | struct netdev_name_node *name_node; | |
329 | struct net *net = dev_net(dev); | |
330 | ||
331 | name_node = netdev_name_node_lookup(net, name); | |
332 | if (!name_node) | |
333 | return -ENOENT; | |
334 | __netdev_name_node_alt_destroy(name_node); | |
335 | ||
336 | return 0; | |
337 | } | |
338 | EXPORT_SYMBOL(netdev_name_node_alt_destroy); | |
339 | ||
340 | static void netdev_name_node_alt_flush(struct net_device *dev) | |
341 | { | |
342 | struct netdev_name_node *name_node, *tmp; | |
343 | ||
344 | list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) | |
345 | __netdev_name_node_alt_destroy(name_node); | |
346 | } | |
347 | ||
ce286d32 | 348 | /* Device list insertion */ |
53759be9 | 349 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 350 | { |
c346dca1 | 351 | struct net *net = dev_net(dev); |
ce286d32 EB |
352 | |
353 | ASSERT_RTNL(); | |
354 | ||
355 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 356 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
ff927412 | 357 | netdev_name_node_add(net, dev->name_node); |
fb699dfd ED |
358 | hlist_add_head_rcu(&dev->index_hlist, |
359 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 360 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
361 | |
362 | dev_base_seq_inc(net); | |
ce286d32 EB |
363 | } |
364 | ||
fb699dfd ED |
365 | /* Device list removal |
366 | * caller must respect a RCU grace period before freeing/reusing dev | |
367 | */ | |
ce286d32 EB |
368 | static void unlist_netdevice(struct net_device *dev) |
369 | { | |
370 | ASSERT_RTNL(); | |
371 | ||
372 | /* Unlink dev from the device chain */ | |
373 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 374 | list_del_rcu(&dev->dev_list); |
ff927412 | 375 | netdev_name_node_del(dev->name_node); |
fb699dfd | 376 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 377 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
378 | |
379 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
380 | } |
381 | ||
1da177e4 LT |
382 | /* |
383 | * Our notifier list | |
384 | */ | |
385 | ||
f07d5b94 | 386 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
387 | |
388 | /* | |
389 | * Device drivers call our routines to queue packets here. We empty the | |
390 | * queue in the local softnet handler. | |
391 | */ | |
bea3348e | 392 | |
9958da05 | 393 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 394 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 395 | |
1da177e4 | 396 | /******************************************************************************* |
eb13da1a | 397 | * |
398 | * Protocol management and registration routines | |
399 | * | |
400 | *******************************************************************************/ | |
1da177e4 | 401 | |
1da177e4 | 402 | |
1da177e4 LT |
403 | /* |
404 | * Add a protocol ID to the list. Now that the input handler is | |
405 | * smarter we can dispense with all the messy stuff that used to be | |
406 | * here. | |
407 | * | |
408 | * BEWARE!!! Protocol handlers, mangling input packets, | |
409 | * MUST BE last in hash buckets and checking protocol handlers | |
410 | * MUST start from promiscuous ptype_all chain in net_bh. | |
411 | * It is true now, do not change it. | |
412 | * Explanation follows: if protocol handler, mangling packet, will | |
413 | * be the first on list, it is not able to sense, that packet | |
414 | * is cloned and should be copied-on-write, so that it will | |
415 | * change it and subsequent readers will get broken packet. | |
416 | * --ANK (980803) | |
417 | */ | |
418 | ||
c07b68e8 ED |
419 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
420 | { | |
421 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 422 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 423 | else |
7866a621 SN |
424 | return pt->dev ? &pt->dev->ptype_specific : |
425 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
426 | } |
427 | ||
1da177e4 LT |
428 | /** |
429 | * dev_add_pack - add packet handler | |
430 | * @pt: packet type declaration | |
431 | * | |
432 | * Add a protocol handler to the networking stack. The passed &packet_type | |
433 | * is linked into kernel lists and may not be freed until it has been | |
434 | * removed from the kernel lists. | |
435 | * | |
4ec93edb | 436 | * This call does not sleep therefore it can not |
1da177e4 LT |
437 | * guarantee all CPU's that are in middle of receiving packets |
438 | * will see the new packet type (until the next received packet). | |
439 | */ | |
440 | ||
441 | void dev_add_pack(struct packet_type *pt) | |
442 | { | |
c07b68e8 | 443 | struct list_head *head = ptype_head(pt); |
1da177e4 | 444 | |
c07b68e8 ED |
445 | spin_lock(&ptype_lock); |
446 | list_add_rcu(&pt->list, head); | |
447 | spin_unlock(&ptype_lock); | |
1da177e4 | 448 | } |
d1b19dff | 449 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 450 | |
1da177e4 LT |
451 | /** |
452 | * __dev_remove_pack - remove packet handler | |
453 | * @pt: packet type declaration | |
454 | * | |
455 | * Remove a protocol handler that was previously added to the kernel | |
456 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
457 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 458 | * returns. |
1da177e4 LT |
459 | * |
460 | * The packet type might still be in use by receivers | |
461 | * and must not be freed until after all the CPU's have gone | |
462 | * through a quiescent state. | |
463 | */ | |
464 | void __dev_remove_pack(struct packet_type *pt) | |
465 | { | |
c07b68e8 | 466 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
467 | struct packet_type *pt1; |
468 | ||
c07b68e8 | 469 | spin_lock(&ptype_lock); |
1da177e4 LT |
470 | |
471 | list_for_each_entry(pt1, head, list) { | |
472 | if (pt == pt1) { | |
473 | list_del_rcu(&pt->list); | |
474 | goto out; | |
475 | } | |
476 | } | |
477 | ||
7b6cd1ce | 478 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 479 | out: |
c07b68e8 | 480 | spin_unlock(&ptype_lock); |
1da177e4 | 481 | } |
d1b19dff ED |
482 | EXPORT_SYMBOL(__dev_remove_pack); |
483 | ||
1da177e4 LT |
484 | /** |
485 | * dev_remove_pack - remove packet handler | |
486 | * @pt: packet type declaration | |
487 | * | |
488 | * Remove a protocol handler that was previously added to the kernel | |
489 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
490 | * from the kernel lists and can be freed or reused once this function | |
491 | * returns. | |
492 | * | |
493 | * This call sleeps to guarantee that no CPU is looking at the packet | |
494 | * type after return. | |
495 | */ | |
496 | void dev_remove_pack(struct packet_type *pt) | |
497 | { | |
498 | __dev_remove_pack(pt); | |
4ec93edb | 499 | |
1da177e4 LT |
500 | synchronize_net(); |
501 | } | |
d1b19dff | 502 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 503 | |
62532da9 VY |
504 | |
505 | /** | |
506 | * dev_add_offload - register offload handlers | |
507 | * @po: protocol offload declaration | |
508 | * | |
509 | * Add protocol offload handlers to the networking stack. The passed | |
510 | * &proto_offload is linked into kernel lists and may not be freed until | |
511 | * it has been removed from the kernel lists. | |
512 | * | |
513 | * This call does not sleep therefore it can not | |
514 | * guarantee all CPU's that are in middle of receiving packets | |
515 | * will see the new offload handlers (until the next received packet). | |
516 | */ | |
517 | void dev_add_offload(struct packet_offload *po) | |
518 | { | |
bdef7de4 | 519 | struct packet_offload *elem; |
62532da9 VY |
520 | |
521 | spin_lock(&offload_lock); | |
bdef7de4 DM |
522 | list_for_each_entry(elem, &offload_base, list) { |
523 | if (po->priority < elem->priority) | |
524 | break; | |
525 | } | |
526 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
527 | spin_unlock(&offload_lock); |
528 | } | |
529 | EXPORT_SYMBOL(dev_add_offload); | |
530 | ||
531 | /** | |
532 | * __dev_remove_offload - remove offload handler | |
533 | * @po: packet offload declaration | |
534 | * | |
535 | * Remove a protocol offload handler that was previously added to the | |
536 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
537 | * is removed from the kernel lists and can be freed or reused once this | |
538 | * function returns. | |
539 | * | |
540 | * The packet type might still be in use by receivers | |
541 | * and must not be freed until after all the CPU's have gone | |
542 | * through a quiescent state. | |
543 | */ | |
1d143d9f | 544 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
545 | { |
546 | struct list_head *head = &offload_base; | |
547 | struct packet_offload *po1; | |
548 | ||
c53aa505 | 549 | spin_lock(&offload_lock); |
62532da9 VY |
550 | |
551 | list_for_each_entry(po1, head, list) { | |
552 | if (po == po1) { | |
553 | list_del_rcu(&po->list); | |
554 | goto out; | |
555 | } | |
556 | } | |
557 | ||
558 | pr_warn("dev_remove_offload: %p not found\n", po); | |
559 | out: | |
c53aa505 | 560 | spin_unlock(&offload_lock); |
62532da9 | 561 | } |
62532da9 VY |
562 | |
563 | /** | |
564 | * dev_remove_offload - remove packet offload handler | |
565 | * @po: packet offload declaration | |
566 | * | |
567 | * Remove a packet offload handler that was previously added to the kernel | |
568 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
569 | * removed from the kernel lists and can be freed or reused once this | |
570 | * function returns. | |
571 | * | |
572 | * This call sleeps to guarantee that no CPU is looking at the packet | |
573 | * type after return. | |
574 | */ | |
575 | void dev_remove_offload(struct packet_offload *po) | |
576 | { | |
577 | __dev_remove_offload(po); | |
578 | ||
579 | synchronize_net(); | |
580 | } | |
581 | EXPORT_SYMBOL(dev_remove_offload); | |
582 | ||
1da177e4 | 583 | /****************************************************************************** |
eb13da1a | 584 | * |
585 | * Device Boot-time Settings Routines | |
586 | * | |
587 | ******************************************************************************/ | |
1da177e4 LT |
588 | |
589 | /* Boot time configuration table */ | |
590 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
591 | ||
592 | /** | |
593 | * netdev_boot_setup_add - add new setup entry | |
594 | * @name: name of the device | |
595 | * @map: configured settings for the device | |
596 | * | |
597 | * Adds new setup entry to the dev_boot_setup list. The function | |
598 | * returns 0 on error and 1 on success. This is a generic routine to | |
599 | * all netdevices. | |
600 | */ | |
601 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
602 | { | |
603 | struct netdev_boot_setup *s; | |
604 | int i; | |
605 | ||
606 | s = dev_boot_setup; | |
607 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
608 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
609 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 610 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
611 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
612 | break; | |
613 | } | |
614 | } | |
615 | ||
616 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
617 | } | |
618 | ||
619 | /** | |
722c9a0c | 620 | * netdev_boot_setup_check - check boot time settings |
621 | * @dev: the netdevice | |
1da177e4 | 622 | * |
722c9a0c | 623 | * Check boot time settings for the device. |
624 | * The found settings are set for the device to be used | |
625 | * later in the device probing. | |
626 | * Returns 0 if no settings found, 1 if they are. | |
1da177e4 LT |
627 | */ |
628 | int netdev_boot_setup_check(struct net_device *dev) | |
629 | { | |
630 | struct netdev_boot_setup *s = dev_boot_setup; | |
631 | int i; | |
632 | ||
633 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
634 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 635 | !strcmp(dev->name, s[i].name)) { |
722c9a0c | 636 | dev->irq = s[i].map.irq; |
637 | dev->base_addr = s[i].map.base_addr; | |
638 | dev->mem_start = s[i].map.mem_start; | |
639 | dev->mem_end = s[i].map.mem_end; | |
1da177e4 LT |
640 | return 1; |
641 | } | |
642 | } | |
643 | return 0; | |
644 | } | |
d1b19dff | 645 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
646 | |
647 | ||
648 | /** | |
722c9a0c | 649 | * netdev_boot_base - get address from boot time settings |
650 | * @prefix: prefix for network device | |
651 | * @unit: id for network device | |
652 | * | |
653 | * Check boot time settings for the base address of device. | |
654 | * The found settings are set for the device to be used | |
655 | * later in the device probing. | |
656 | * Returns 0 if no settings found. | |
1da177e4 LT |
657 | */ |
658 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
659 | { | |
660 | const struct netdev_boot_setup *s = dev_boot_setup; | |
661 | char name[IFNAMSIZ]; | |
662 | int i; | |
663 | ||
664 | sprintf(name, "%s%d", prefix, unit); | |
665 | ||
666 | /* | |
667 | * If device already registered then return base of 1 | |
668 | * to indicate not to probe for this interface | |
669 | */ | |
881d966b | 670 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
671 | return 1; |
672 | ||
673 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
674 | if (!strcmp(name, s[i].name)) | |
675 | return s[i].map.base_addr; | |
676 | return 0; | |
677 | } | |
678 | ||
679 | /* | |
680 | * Saves at boot time configured settings for any netdevice. | |
681 | */ | |
682 | int __init netdev_boot_setup(char *str) | |
683 | { | |
684 | int ints[5]; | |
685 | struct ifmap map; | |
686 | ||
687 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
688 | if (!str || !*str) | |
689 | return 0; | |
690 | ||
691 | /* Save settings */ | |
692 | memset(&map, 0, sizeof(map)); | |
693 | if (ints[0] > 0) | |
694 | map.irq = ints[1]; | |
695 | if (ints[0] > 1) | |
696 | map.base_addr = ints[2]; | |
697 | if (ints[0] > 2) | |
698 | map.mem_start = ints[3]; | |
699 | if (ints[0] > 3) | |
700 | map.mem_end = ints[4]; | |
701 | ||
702 | /* Add new entry to the list */ | |
703 | return netdev_boot_setup_add(str, &map); | |
704 | } | |
705 | ||
706 | __setup("netdev=", netdev_boot_setup); | |
707 | ||
708 | /******************************************************************************* | |
eb13da1a | 709 | * |
710 | * Device Interface Subroutines | |
711 | * | |
712 | *******************************************************************************/ | |
1da177e4 | 713 | |
a54acb3a ND |
714 | /** |
715 | * dev_get_iflink - get 'iflink' value of a interface | |
716 | * @dev: targeted interface | |
717 | * | |
718 | * Indicates the ifindex the interface is linked to. | |
719 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
720 | */ | |
721 | ||
722 | int dev_get_iflink(const struct net_device *dev) | |
723 | { | |
724 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
725 | return dev->netdev_ops->ndo_get_iflink(dev); | |
726 | ||
7a66bbc9 | 727 | return dev->ifindex; |
a54acb3a ND |
728 | } |
729 | EXPORT_SYMBOL(dev_get_iflink); | |
730 | ||
fc4099f1 PS |
731 | /** |
732 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
733 | * @dev: targeted interface | |
734 | * @skb: The packet. | |
735 | * | |
736 | * For better visibility of tunnel traffic OVS needs to retrieve | |
737 | * egress tunnel information for a packet. Following API allows | |
738 | * user to get this info. | |
739 | */ | |
740 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
741 | { | |
742 | struct ip_tunnel_info *info; | |
743 | ||
744 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
745 | return -EINVAL; | |
746 | ||
747 | info = skb_tunnel_info_unclone(skb); | |
748 | if (!info) | |
749 | return -ENOMEM; | |
750 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
751 | return -EINVAL; | |
752 | ||
753 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
754 | } | |
755 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
756 | ||
1da177e4 LT |
757 | /** |
758 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 759 | * @net: the applicable net namespace |
1da177e4 LT |
760 | * @name: name to find |
761 | * | |
762 | * Find an interface by name. Must be called under RTNL semaphore | |
763 | * or @dev_base_lock. If the name is found a pointer to the device | |
764 | * is returned. If the name is not found then %NULL is returned. The | |
765 | * reference counters are not incremented so the caller must be | |
766 | * careful with locks. | |
767 | */ | |
768 | ||
881d966b | 769 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 770 | { |
ff927412 | 771 | struct netdev_name_node *node_name; |
1da177e4 | 772 | |
ff927412 JP |
773 | node_name = netdev_name_node_lookup(net, name); |
774 | return node_name ? node_name->dev : NULL; | |
1da177e4 | 775 | } |
d1b19dff | 776 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 777 | |
72c9528b | 778 | /** |
722c9a0c | 779 | * dev_get_by_name_rcu - find a device by its name |
780 | * @net: the applicable net namespace | |
781 | * @name: name to find | |
782 | * | |
783 | * Find an interface by name. | |
784 | * If the name is found a pointer to the device is returned. | |
785 | * If the name is not found then %NULL is returned. | |
786 | * The reference counters are not incremented so the caller must be | |
787 | * careful with locks. The caller must hold RCU lock. | |
72c9528b ED |
788 | */ |
789 | ||
790 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
791 | { | |
ff927412 | 792 | struct netdev_name_node *node_name; |
72c9528b | 793 | |
ff927412 JP |
794 | node_name = netdev_name_node_lookup_rcu(net, name); |
795 | return node_name ? node_name->dev : NULL; | |
72c9528b ED |
796 | } |
797 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
798 | ||
1da177e4 LT |
799 | /** |
800 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 801 | * @net: the applicable net namespace |
1da177e4 LT |
802 | * @name: name to find |
803 | * | |
804 | * Find an interface by name. This can be called from any | |
805 | * context and does its own locking. The returned handle has | |
806 | * the usage count incremented and the caller must use dev_put() to | |
807 | * release it when it is no longer needed. %NULL is returned if no | |
808 | * matching device is found. | |
809 | */ | |
810 | ||
881d966b | 811 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
812 | { |
813 | struct net_device *dev; | |
814 | ||
72c9528b ED |
815 | rcu_read_lock(); |
816 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
817 | if (dev) |
818 | dev_hold(dev); | |
72c9528b | 819 | rcu_read_unlock(); |
1da177e4 LT |
820 | return dev; |
821 | } | |
d1b19dff | 822 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
823 | |
824 | /** | |
825 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 826 | * @net: the applicable net namespace |
1da177e4 LT |
827 | * @ifindex: index of device |
828 | * | |
829 | * Search for an interface by index. Returns %NULL if the device | |
830 | * is not found or a pointer to the device. The device has not | |
831 | * had its reference counter increased so the caller must be careful | |
832 | * about locking. The caller must hold either the RTNL semaphore | |
833 | * or @dev_base_lock. | |
834 | */ | |
835 | ||
881d966b | 836 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 837 | { |
0bd8d536 ED |
838 | struct net_device *dev; |
839 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 840 | |
b67bfe0d | 841 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
842 | if (dev->ifindex == ifindex) |
843 | return dev; | |
0bd8d536 | 844 | |
1da177e4 LT |
845 | return NULL; |
846 | } | |
d1b19dff | 847 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 848 | |
fb699dfd ED |
849 | /** |
850 | * dev_get_by_index_rcu - find a device by its ifindex | |
851 | * @net: the applicable net namespace | |
852 | * @ifindex: index of device | |
853 | * | |
854 | * Search for an interface by index. Returns %NULL if the device | |
855 | * is not found or a pointer to the device. The device has not | |
856 | * had its reference counter increased so the caller must be careful | |
857 | * about locking. The caller must hold RCU lock. | |
858 | */ | |
859 | ||
860 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
861 | { | |
fb699dfd ED |
862 | struct net_device *dev; |
863 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
864 | ||
b67bfe0d | 865 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
866 | if (dev->ifindex == ifindex) |
867 | return dev; | |
868 | ||
869 | return NULL; | |
870 | } | |
871 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
872 | ||
1da177e4 LT |
873 | |
874 | /** | |
875 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 876 | * @net: the applicable net namespace |
1da177e4 LT |
877 | * @ifindex: index of device |
878 | * | |
879 | * Search for an interface by index. Returns NULL if the device | |
880 | * is not found or a pointer to the device. The device returned has | |
881 | * had a reference added and the pointer is safe until the user calls | |
882 | * dev_put to indicate they have finished with it. | |
883 | */ | |
884 | ||
881d966b | 885 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
886 | { |
887 | struct net_device *dev; | |
888 | ||
fb699dfd ED |
889 | rcu_read_lock(); |
890 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
891 | if (dev) |
892 | dev_hold(dev); | |
fb699dfd | 893 | rcu_read_unlock(); |
1da177e4 LT |
894 | return dev; |
895 | } | |
d1b19dff | 896 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 897 | |
90b602f8 ML |
898 | /** |
899 | * dev_get_by_napi_id - find a device by napi_id | |
900 | * @napi_id: ID of the NAPI struct | |
901 | * | |
902 | * Search for an interface by NAPI ID. Returns %NULL if the device | |
903 | * is not found or a pointer to the device. The device has not had | |
904 | * its reference counter increased so the caller must be careful | |
905 | * about locking. The caller must hold RCU lock. | |
906 | */ | |
907 | ||
908 | struct net_device *dev_get_by_napi_id(unsigned int napi_id) | |
909 | { | |
910 | struct napi_struct *napi; | |
911 | ||
912 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
913 | ||
914 | if (napi_id < MIN_NAPI_ID) | |
915 | return NULL; | |
916 | ||
917 | napi = napi_by_id(napi_id); | |
918 | ||
919 | return napi ? napi->dev : NULL; | |
920 | } | |
921 | EXPORT_SYMBOL(dev_get_by_napi_id); | |
922 | ||
5dbe7c17 NS |
923 | /** |
924 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
925 | * @net: network namespace | |
926 | * @name: a pointer to the buffer where the name will be stored. | |
927 | * @ifindex: the ifindex of the interface to get the name from. | |
928 | * | |
929 | * The use of raw_seqcount_begin() and cond_resched() before | |
930 | * retrying is required as we want to give the writers a chance | |
931 | * to complete when CONFIG_PREEMPT is not set. | |
932 | */ | |
933 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
934 | { | |
935 | struct net_device *dev; | |
936 | unsigned int seq; | |
937 | ||
938 | retry: | |
939 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
940 | rcu_read_lock(); | |
941 | dev = dev_get_by_index_rcu(net, ifindex); | |
942 | if (!dev) { | |
943 | rcu_read_unlock(); | |
944 | return -ENODEV; | |
945 | } | |
946 | ||
947 | strcpy(name, dev->name); | |
948 | rcu_read_unlock(); | |
949 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
950 | cond_resched(); | |
951 | goto retry; | |
952 | } | |
953 | ||
954 | return 0; | |
955 | } | |
956 | ||
1da177e4 | 957 | /** |
941666c2 | 958 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 959 | * @net: the applicable net namespace |
1da177e4 LT |
960 | * @type: media type of device |
961 | * @ha: hardware address | |
962 | * | |
963 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
964 | * is not found or a pointer to the device. |
965 | * The caller must hold RCU or RTNL. | |
941666c2 | 966 | * The returned device has not had its ref count increased |
1da177e4 LT |
967 | * and the caller must therefore be careful about locking |
968 | * | |
1da177e4 LT |
969 | */ |
970 | ||
941666c2 ED |
971 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
972 | const char *ha) | |
1da177e4 LT |
973 | { |
974 | struct net_device *dev; | |
975 | ||
941666c2 | 976 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
977 | if (dev->type == type && |
978 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
979 | return dev; |
980 | ||
981 | return NULL; | |
1da177e4 | 982 | } |
941666c2 | 983 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 984 | |
881d966b | 985 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
986 | { |
987 | struct net_device *dev; | |
988 | ||
4e9cac2b | 989 | ASSERT_RTNL(); |
881d966b | 990 | for_each_netdev(net, dev) |
4e9cac2b | 991 | if (dev->type == type) |
7562f876 PE |
992 | return dev; |
993 | ||
994 | return NULL; | |
4e9cac2b | 995 | } |
4e9cac2b PM |
996 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
997 | ||
881d966b | 998 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 999 | { |
99fe3c39 | 1000 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 1001 | |
99fe3c39 ED |
1002 | rcu_read_lock(); |
1003 | for_each_netdev_rcu(net, dev) | |
1004 | if (dev->type == type) { | |
1005 | dev_hold(dev); | |
1006 | ret = dev; | |
1007 | break; | |
1008 | } | |
1009 | rcu_read_unlock(); | |
1010 | return ret; | |
1da177e4 | 1011 | } |
1da177e4 LT |
1012 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
1013 | ||
1014 | /** | |
6c555490 | 1015 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 1016 | * @net: the applicable net namespace |
1da177e4 LT |
1017 | * @if_flags: IFF_* values |
1018 | * @mask: bitmask of bits in if_flags to check | |
1019 | * | |
1020 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 1021 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 1022 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
1023 | */ |
1024 | ||
6c555490 WC |
1025 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
1026 | unsigned short mask) | |
1da177e4 | 1027 | { |
7562f876 | 1028 | struct net_device *dev, *ret; |
1da177e4 | 1029 | |
6c555490 WC |
1030 | ASSERT_RTNL(); |
1031 | ||
7562f876 | 1032 | ret = NULL; |
6c555490 | 1033 | for_each_netdev(net, dev) { |
1da177e4 | 1034 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 1035 | ret = dev; |
1da177e4 LT |
1036 | break; |
1037 | } | |
1038 | } | |
7562f876 | 1039 | return ret; |
1da177e4 | 1040 | } |
6c555490 | 1041 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
1042 | |
1043 | /** | |
1044 | * dev_valid_name - check if name is okay for network device | |
1045 | * @name: name string | |
1046 | * | |
1047 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
1048 | * to allow sysfs to work. We also disallow any kind of |
1049 | * whitespace. | |
1da177e4 | 1050 | */ |
95f050bf | 1051 | bool dev_valid_name(const char *name) |
1da177e4 | 1052 | { |
c7fa9d18 | 1053 | if (*name == '\0') |
95f050bf | 1054 | return false; |
a9d48205 | 1055 | if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) |
95f050bf | 1056 | return false; |
c7fa9d18 | 1057 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1058 | return false; |
c7fa9d18 DM |
1059 | |
1060 | while (*name) { | |
a4176a93 | 1061 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1062 | return false; |
c7fa9d18 DM |
1063 | name++; |
1064 | } | |
95f050bf | 1065 | return true; |
1da177e4 | 1066 | } |
d1b19dff | 1067 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1068 | |
1069 | /** | |
b267b179 EB |
1070 | * __dev_alloc_name - allocate a name for a device |
1071 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1072 | * @name: name format string |
b267b179 | 1073 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1074 | * |
1075 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1076 | * id. It scans list of devices to build up a free map, then chooses |
1077 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1078 | * while allocating the name and adding the device in order to avoid | |
1079 | * duplicates. | |
1080 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1081 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1082 | */ |
1083 | ||
b267b179 | 1084 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1085 | { |
1086 | int i = 0; | |
1da177e4 LT |
1087 | const char *p; |
1088 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1089 | unsigned long *inuse; |
1da177e4 LT |
1090 | struct net_device *d; |
1091 | ||
93809105 RV |
1092 | if (!dev_valid_name(name)) |
1093 | return -EINVAL; | |
1094 | ||
51f299dd | 1095 | p = strchr(name, '%'); |
1da177e4 LT |
1096 | if (p) { |
1097 | /* | |
1098 | * Verify the string as this thing may have come from | |
1099 | * the user. There must be either one "%d" and no other "%" | |
1100 | * characters. | |
1101 | */ | |
1102 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1103 | return -EINVAL; | |
1104 | ||
1105 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1106 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1107 | if (!inuse) |
1108 | return -ENOMEM; | |
1109 | ||
881d966b | 1110 | for_each_netdev(net, d) { |
1da177e4 LT |
1111 | if (!sscanf(d->name, name, &i)) |
1112 | continue; | |
1113 | if (i < 0 || i >= max_netdevices) | |
1114 | continue; | |
1115 | ||
1116 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1117 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1118 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1119 | set_bit(i, inuse); | |
1120 | } | |
1121 | ||
1122 | i = find_first_zero_bit(inuse, max_netdevices); | |
1123 | free_page((unsigned long) inuse); | |
1124 | } | |
1125 | ||
6224abda | 1126 | snprintf(buf, IFNAMSIZ, name, i); |
b267b179 | 1127 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1128 | return i; |
1da177e4 LT |
1129 | |
1130 | /* It is possible to run out of possible slots | |
1131 | * when the name is long and there isn't enough space left | |
1132 | * for the digits, or if all bits are used. | |
1133 | */ | |
029b6d14 | 1134 | return -ENFILE; |
1da177e4 LT |
1135 | } |
1136 | ||
2c88b855 RV |
1137 | static int dev_alloc_name_ns(struct net *net, |
1138 | struct net_device *dev, | |
1139 | const char *name) | |
1140 | { | |
1141 | char buf[IFNAMSIZ]; | |
1142 | int ret; | |
1143 | ||
c46d7642 | 1144 | BUG_ON(!net); |
2c88b855 RV |
1145 | ret = __dev_alloc_name(net, name, buf); |
1146 | if (ret >= 0) | |
1147 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1148 | return ret; | |
1da177e4 LT |
1149 | } |
1150 | ||
b267b179 EB |
1151 | /** |
1152 | * dev_alloc_name - allocate a name for a device | |
1153 | * @dev: device | |
1154 | * @name: name format string | |
1155 | * | |
1156 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1157 | * id. It scans list of devices to build up a free map, then chooses | |
1158 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1159 | * while allocating the name and adding the device in order to avoid | |
1160 | * duplicates. | |
1161 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1162 | * Returns the number of the unit assigned or a negative errno code. | |
1163 | */ | |
1164 | ||
1165 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1166 | { | |
c46d7642 | 1167 | return dev_alloc_name_ns(dev_net(dev), dev, name); |
b267b179 | 1168 | } |
d1b19dff | 1169 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1170 | |
bacb7e18 ED |
1171 | static int dev_get_valid_name(struct net *net, struct net_device *dev, |
1172 | const char *name) | |
828de4f6 | 1173 | { |
55a5ec9b DM |
1174 | BUG_ON(!net); |
1175 | ||
1176 | if (!dev_valid_name(name)) | |
1177 | return -EINVAL; | |
1178 | ||
1179 | if (strchr(name, '%')) | |
1180 | return dev_alloc_name_ns(net, dev, name); | |
1181 | else if (__dev_get_by_name(net, name)) | |
1182 | return -EEXIST; | |
1183 | else if (dev->name != name) | |
1184 | strlcpy(dev->name, name, IFNAMSIZ); | |
1185 | ||
1186 | return 0; | |
d9031024 | 1187 | } |
1da177e4 LT |
1188 | |
1189 | /** | |
1190 | * dev_change_name - change name of a device | |
1191 | * @dev: device | |
1192 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1193 | * | |
1194 | * Change name of a device, can pass format strings "eth%d". | |
1195 | * for wildcarding. | |
1196 | */ | |
cf04a4c7 | 1197 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1198 | { |
238fa362 | 1199 | unsigned char old_assign_type; |
fcc5a03a | 1200 | char oldname[IFNAMSIZ]; |
1da177e4 | 1201 | int err = 0; |
fcc5a03a | 1202 | int ret; |
881d966b | 1203 | struct net *net; |
1da177e4 LT |
1204 | |
1205 | ASSERT_RTNL(); | |
c346dca1 | 1206 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1207 | |
c346dca1 | 1208 | net = dev_net(dev); |
8065a779 SWL |
1209 | |
1210 | /* Some auto-enslaved devices e.g. failover slaves are | |
1211 | * special, as userspace might rename the device after | |
1212 | * the interface had been brought up and running since | |
1213 | * the point kernel initiated auto-enslavement. Allow | |
1214 | * live name change even when these slave devices are | |
1215 | * up and running. | |
1216 | * | |
1217 | * Typically, users of these auto-enslaving devices | |
1218 | * don't actually care about slave name change, as | |
1219 | * they are supposed to operate on master interface | |
1220 | * directly. | |
1221 | */ | |
1222 | if (dev->flags & IFF_UP && | |
1223 | likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) | |
1da177e4 LT |
1224 | return -EBUSY; |
1225 | ||
30e6c9fa | 1226 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1227 | |
1228 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1229 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1230 | return 0; |
c91f6df2 | 1231 | } |
c8d90dca | 1232 | |
fcc5a03a HX |
1233 | memcpy(oldname, dev->name, IFNAMSIZ); |
1234 | ||
828de4f6 | 1235 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1236 | if (err < 0) { |
30e6c9fa | 1237 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1238 | return err; |
c91f6df2 | 1239 | } |
1da177e4 | 1240 | |
6fe82a39 VF |
1241 | if (oldname[0] && !strchr(oldname, '%')) |
1242 | netdev_info(dev, "renamed from %s\n", oldname); | |
1243 | ||
238fa362 TG |
1244 | old_assign_type = dev->name_assign_type; |
1245 | dev->name_assign_type = NET_NAME_RENAMED; | |
1246 | ||
fcc5a03a | 1247 | rollback: |
a1b3f594 EB |
1248 | ret = device_rename(&dev->dev, dev->name); |
1249 | if (ret) { | |
1250 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1251 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1252 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1253 | return ret; |
dcc99773 | 1254 | } |
7f988eab | 1255 | |
30e6c9fa | 1256 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1257 | |
5bb025fa VF |
1258 | netdev_adjacent_rename_links(dev, oldname); |
1259 | ||
7f988eab | 1260 | write_lock_bh(&dev_base_lock); |
ff927412 | 1261 | netdev_name_node_del(dev->name_node); |
72c9528b ED |
1262 | write_unlock_bh(&dev_base_lock); |
1263 | ||
1264 | synchronize_rcu(); | |
1265 | ||
1266 | write_lock_bh(&dev_base_lock); | |
ff927412 | 1267 | netdev_name_node_add(net, dev->name_node); |
7f988eab HX |
1268 | write_unlock_bh(&dev_base_lock); |
1269 | ||
056925ab | 1270 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1271 | ret = notifier_to_errno(ret); |
1272 | ||
1273 | if (ret) { | |
91e9c07b ED |
1274 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1275 | if (err >= 0) { | |
fcc5a03a | 1276 | err = ret; |
30e6c9fa | 1277 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1278 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1279 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1280 | dev->name_assign_type = old_assign_type; |
1281 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1282 | goto rollback; |
91e9c07b | 1283 | } else { |
7b6cd1ce | 1284 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1285 | dev->name, ret); |
fcc5a03a HX |
1286 | } |
1287 | } | |
1da177e4 LT |
1288 | |
1289 | return err; | |
1290 | } | |
1291 | ||
0b815a1a SH |
1292 | /** |
1293 | * dev_set_alias - change ifalias of a device | |
1294 | * @dev: device | |
1295 | * @alias: name up to IFALIASZ | |
f0db275a | 1296 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1297 | * |
1298 | * Set ifalias for a device, | |
1299 | */ | |
1300 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1301 | { | |
6c557001 | 1302 | struct dev_ifalias *new_alias = NULL; |
0b815a1a SH |
1303 | |
1304 | if (len >= IFALIASZ) | |
1305 | return -EINVAL; | |
1306 | ||
6c557001 FW |
1307 | if (len) { |
1308 | new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); | |
1309 | if (!new_alias) | |
1310 | return -ENOMEM; | |
1311 | ||
1312 | memcpy(new_alias->ifalias, alias, len); | |
1313 | new_alias->ifalias[len] = 0; | |
96ca4a2c OH |
1314 | } |
1315 | ||
6c557001 FW |
1316 | mutex_lock(&ifalias_mutex); |
1317 | rcu_swap_protected(dev->ifalias, new_alias, | |
1318 | mutex_is_locked(&ifalias_mutex)); | |
1319 | mutex_unlock(&ifalias_mutex); | |
1320 | ||
1321 | if (new_alias) | |
1322 | kfree_rcu(new_alias, rcuhead); | |
0b815a1a | 1323 | |
0b815a1a SH |
1324 | return len; |
1325 | } | |
0fe554a4 | 1326 | EXPORT_SYMBOL(dev_set_alias); |
0b815a1a | 1327 | |
6c557001 FW |
1328 | /** |
1329 | * dev_get_alias - get ifalias of a device | |
1330 | * @dev: device | |
20e88320 | 1331 | * @name: buffer to store name of ifalias |
6c557001 FW |
1332 | * @len: size of buffer |
1333 | * | |
1334 | * get ifalias for a device. Caller must make sure dev cannot go | |
1335 | * away, e.g. rcu read lock or own a reference count to device. | |
1336 | */ | |
1337 | int dev_get_alias(const struct net_device *dev, char *name, size_t len) | |
1338 | { | |
1339 | const struct dev_ifalias *alias; | |
1340 | int ret = 0; | |
1341 | ||
1342 | rcu_read_lock(); | |
1343 | alias = rcu_dereference(dev->ifalias); | |
1344 | if (alias) | |
1345 | ret = snprintf(name, len, "%s", alias->ifalias); | |
1346 | rcu_read_unlock(); | |
1347 | ||
1348 | return ret; | |
1349 | } | |
0b815a1a | 1350 | |
d8a33ac4 | 1351 | /** |
3041a069 | 1352 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1353 | * @dev: device to cause notification |
1354 | * | |
1355 | * Called to indicate a device has changed features. | |
1356 | */ | |
1357 | void netdev_features_change(struct net_device *dev) | |
1358 | { | |
056925ab | 1359 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1360 | } |
1361 | EXPORT_SYMBOL(netdev_features_change); | |
1362 | ||
1da177e4 LT |
1363 | /** |
1364 | * netdev_state_change - device changes state | |
1365 | * @dev: device to cause notification | |
1366 | * | |
1367 | * Called to indicate a device has changed state. This function calls | |
1368 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1369 | * to the routing socket. | |
1370 | */ | |
1371 | void netdev_state_change(struct net_device *dev) | |
1372 | { | |
1373 | if (dev->flags & IFF_UP) { | |
51d0c047 DA |
1374 | struct netdev_notifier_change_info change_info = { |
1375 | .info.dev = dev, | |
1376 | }; | |
54951194 | 1377 | |
51d0c047 | 1378 | call_netdevice_notifiers_info(NETDEV_CHANGE, |
54951194 | 1379 | &change_info.info); |
7f294054 | 1380 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1381 | } |
1382 | } | |
d1b19dff | 1383 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1384 | |
ee89bab1 | 1385 | /** |
722c9a0c | 1386 | * netdev_notify_peers - notify network peers about existence of @dev |
1387 | * @dev: network device | |
ee89bab1 AW |
1388 | * |
1389 | * Generate traffic such that interested network peers are aware of | |
1390 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1391 | * a device wants to inform the rest of the network about some sort of | |
1392 | * reconfiguration such as a failover event or virtual machine | |
1393 | * migration. | |
1394 | */ | |
1395 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1396 | { |
ee89bab1 AW |
1397 | rtnl_lock(); |
1398 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
37c343b4 | 1399 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); |
ee89bab1 | 1400 | rtnl_unlock(); |
c1da4ac7 | 1401 | } |
ee89bab1 | 1402 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1403 | |
40c900aa | 1404 | static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
1da177e4 | 1405 | { |
d314774c | 1406 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1407 | int ret; |
1da177e4 | 1408 | |
e46b66bc BH |
1409 | ASSERT_RTNL(); |
1410 | ||
1da177e4 LT |
1411 | if (!netif_device_present(dev)) |
1412 | return -ENODEV; | |
1413 | ||
ca99ca14 NH |
1414 | /* Block netpoll from trying to do any rx path servicing. |
1415 | * If we don't do this there is a chance ndo_poll_controller | |
1416 | * or ndo_poll may be running while we open the device | |
1417 | */ | |
66b5552f | 1418 | netpoll_poll_disable(dev); |
ca99ca14 | 1419 | |
40c900aa | 1420 | ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); |
3b8bcfd5 JB |
1421 | ret = notifier_to_errno(ret); |
1422 | if (ret) | |
1423 | return ret; | |
1424 | ||
1da177e4 | 1425 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1426 | |
d314774c SH |
1427 | if (ops->ndo_validate_addr) |
1428 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1429 | |
d314774c SH |
1430 | if (!ret && ops->ndo_open) |
1431 | ret = ops->ndo_open(dev); | |
1da177e4 | 1432 | |
66b5552f | 1433 | netpoll_poll_enable(dev); |
ca99ca14 | 1434 | |
bada339b JG |
1435 | if (ret) |
1436 | clear_bit(__LINK_STATE_START, &dev->state); | |
1437 | else { | |
1da177e4 | 1438 | dev->flags |= IFF_UP; |
4417da66 | 1439 | dev_set_rx_mode(dev); |
1da177e4 | 1440 | dev_activate(dev); |
7bf23575 | 1441 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1442 | } |
bada339b | 1443 | |
1da177e4 LT |
1444 | return ret; |
1445 | } | |
1446 | ||
1447 | /** | |
bd380811 | 1448 | * dev_open - prepare an interface for use. |
00f54e68 PM |
1449 | * @dev: device to open |
1450 | * @extack: netlink extended ack | |
1da177e4 | 1451 | * |
bd380811 PM |
1452 | * Takes a device from down to up state. The device's private open |
1453 | * function is invoked and then the multicast lists are loaded. Finally | |
1454 | * the device is moved into the up state and a %NETDEV_UP message is | |
1455 | * sent to the netdev notifier chain. | |
1456 | * | |
1457 | * Calling this function on an active interface is a nop. On a failure | |
1458 | * a negative errno code is returned. | |
1da177e4 | 1459 | */ |
00f54e68 | 1460 | int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
bd380811 PM |
1461 | { |
1462 | int ret; | |
1463 | ||
bd380811 PM |
1464 | if (dev->flags & IFF_UP) |
1465 | return 0; | |
1466 | ||
40c900aa | 1467 | ret = __dev_open(dev, extack); |
bd380811 PM |
1468 | if (ret < 0) |
1469 | return ret; | |
1470 | ||
7f294054 | 1471 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1472 | call_netdevice_notifiers(NETDEV_UP, dev); |
1473 | ||
1474 | return ret; | |
1475 | } | |
1476 | EXPORT_SYMBOL(dev_open); | |
1477 | ||
7051b88a | 1478 | static void __dev_close_many(struct list_head *head) |
1da177e4 | 1479 | { |
44345724 | 1480 | struct net_device *dev; |
e46b66bc | 1481 | |
bd380811 | 1482 | ASSERT_RTNL(); |
9d5010db DM |
1483 | might_sleep(); |
1484 | ||
5cde2829 | 1485 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1486 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1487 | netpoll_poll_disable(dev); |
3f4df206 | 1488 | |
44345724 | 1489 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1490 | |
44345724 | 1491 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1492 | |
44345724 OP |
1493 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1494 | * can be even on different cpu. So just clear netif_running(). | |
1495 | * | |
1496 | * dev->stop() will invoke napi_disable() on all of it's | |
1497 | * napi_struct instances on this device. | |
1498 | */ | |
4e857c58 | 1499 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1500 | } |
1da177e4 | 1501 | |
44345724 | 1502 | dev_deactivate_many(head); |
d8b2a4d2 | 1503 | |
5cde2829 | 1504 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1505 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1506 | |
44345724 OP |
1507 | /* |
1508 | * Call the device specific close. This cannot fail. | |
1509 | * Only if device is UP | |
1510 | * | |
1511 | * We allow it to be called even after a DETACH hot-plug | |
1512 | * event. | |
1513 | */ | |
1514 | if (ops->ndo_stop) | |
1515 | ops->ndo_stop(dev); | |
1516 | ||
44345724 | 1517 | dev->flags &= ~IFF_UP; |
66b5552f | 1518 | netpoll_poll_enable(dev); |
44345724 | 1519 | } |
44345724 OP |
1520 | } |
1521 | ||
7051b88a | 1522 | static void __dev_close(struct net_device *dev) |
44345724 OP |
1523 | { |
1524 | LIST_HEAD(single); | |
1525 | ||
5cde2829 | 1526 | list_add(&dev->close_list, &single); |
7051b88a | 1527 | __dev_close_many(&single); |
f87e6f47 | 1528 | list_del(&single); |
44345724 OP |
1529 | } |
1530 | ||
7051b88a | 1531 | void dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1532 | { |
1533 | struct net_device *dev, *tmp; | |
1da177e4 | 1534 | |
5cde2829 EB |
1535 | /* Remove the devices that don't need to be closed */ |
1536 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1537 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1538 | list_del_init(&dev->close_list); |
44345724 OP |
1539 | |
1540 | __dev_close_many(head); | |
1da177e4 | 1541 | |
5cde2829 | 1542 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1543 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1544 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1545 | if (unlink) |
1546 | list_del_init(&dev->close_list); | |
44345724 | 1547 | } |
bd380811 | 1548 | } |
99c4a26a | 1549 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1550 | |
1551 | /** | |
1552 | * dev_close - shutdown an interface. | |
1553 | * @dev: device to shutdown | |
1554 | * | |
1555 | * This function moves an active device into down state. A | |
1556 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1557 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1558 | * chain. | |
1559 | */ | |
7051b88a | 1560 | void dev_close(struct net_device *dev) |
bd380811 | 1561 | { |
e14a5993 ED |
1562 | if (dev->flags & IFF_UP) { |
1563 | LIST_HEAD(single); | |
1da177e4 | 1564 | |
5cde2829 | 1565 | list_add(&dev->close_list, &single); |
99c4a26a | 1566 | dev_close_many(&single, true); |
e14a5993 ED |
1567 | list_del(&single); |
1568 | } | |
1da177e4 | 1569 | } |
d1b19dff | 1570 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1571 | |
1572 | ||
0187bdfb BH |
1573 | /** |
1574 | * dev_disable_lro - disable Large Receive Offload on a device | |
1575 | * @dev: device | |
1576 | * | |
1577 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1578 | * called under RTNL. This is needed if received packets may be | |
1579 | * forwarded to another interface. | |
1580 | */ | |
1581 | void dev_disable_lro(struct net_device *dev) | |
1582 | { | |
fbe168ba MK |
1583 | struct net_device *lower_dev; |
1584 | struct list_head *iter; | |
529d0489 | 1585 | |
bc5787c6 MM |
1586 | dev->wanted_features &= ~NETIF_F_LRO; |
1587 | netdev_update_features(dev); | |
27660515 | 1588 | |
22d5969f MM |
1589 | if (unlikely(dev->features & NETIF_F_LRO)) |
1590 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1591 | |
1592 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1593 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1594 | } |
1595 | EXPORT_SYMBOL(dev_disable_lro); | |
1596 | ||
56f5aa77 MC |
1597 | /** |
1598 | * dev_disable_gro_hw - disable HW Generic Receive Offload on a device | |
1599 | * @dev: device | |
1600 | * | |
1601 | * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be | |
1602 | * called under RTNL. This is needed if Generic XDP is installed on | |
1603 | * the device. | |
1604 | */ | |
1605 | static void dev_disable_gro_hw(struct net_device *dev) | |
1606 | { | |
1607 | dev->wanted_features &= ~NETIF_F_GRO_HW; | |
1608 | netdev_update_features(dev); | |
1609 | ||
1610 | if (unlikely(dev->features & NETIF_F_GRO_HW)) | |
1611 | netdev_WARN(dev, "failed to disable GRO_HW!\n"); | |
1612 | } | |
1613 | ||
ede2762d KT |
1614 | const char *netdev_cmd_to_name(enum netdev_cmd cmd) |
1615 | { | |
1616 | #define N(val) \ | |
1617 | case NETDEV_##val: \ | |
1618 | return "NETDEV_" __stringify(val); | |
1619 | switch (cmd) { | |
1620 | N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) | |
1621 | N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) | |
1622 | N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) | |
1623 | N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) | |
1624 | N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) | |
1625 | N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) | |
1626 | N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) | |
9daae9bd GP |
1627 | N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) |
1628 | N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) | |
1570415f | 1629 | N(PRE_CHANGEADDR) |
3f5ecd8a | 1630 | } |
ede2762d KT |
1631 | #undef N |
1632 | return "UNKNOWN_NETDEV_EVENT"; | |
1633 | } | |
1634 | EXPORT_SYMBOL_GPL(netdev_cmd_to_name); | |
1635 | ||
351638e7 JP |
1636 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1637 | struct net_device *dev) | |
1638 | { | |
51d0c047 DA |
1639 | struct netdev_notifier_info info = { |
1640 | .dev = dev, | |
1641 | }; | |
351638e7 | 1642 | |
351638e7 JP |
1643 | return nb->notifier_call(nb, val, &info); |
1644 | } | |
0187bdfb | 1645 | |
afa0df59 JP |
1646 | static int call_netdevice_register_notifiers(struct notifier_block *nb, |
1647 | struct net_device *dev) | |
1648 | { | |
1649 | int err; | |
1650 | ||
1651 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); | |
1652 | err = notifier_to_errno(err); | |
1653 | if (err) | |
1654 | return err; | |
1655 | ||
1656 | if (!(dev->flags & IFF_UP)) | |
1657 | return 0; | |
1658 | ||
1659 | call_netdevice_notifier(nb, NETDEV_UP, dev); | |
1660 | return 0; | |
1661 | } | |
1662 | ||
1663 | static void call_netdevice_unregister_notifiers(struct notifier_block *nb, | |
1664 | struct net_device *dev) | |
1665 | { | |
1666 | if (dev->flags & IFF_UP) { | |
1667 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, | |
1668 | dev); | |
1669 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
1670 | } | |
1671 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); | |
1672 | } | |
1673 | ||
1674 | static int call_netdevice_register_net_notifiers(struct notifier_block *nb, | |
1675 | struct net *net) | |
1676 | { | |
1677 | struct net_device *dev; | |
1678 | int err; | |
1679 | ||
1680 | for_each_netdev(net, dev) { | |
1681 | err = call_netdevice_register_notifiers(nb, dev); | |
1682 | if (err) | |
1683 | goto rollback; | |
1684 | } | |
1685 | return 0; | |
1686 | ||
1687 | rollback: | |
1688 | for_each_netdev_continue_reverse(net, dev) | |
1689 | call_netdevice_unregister_notifiers(nb, dev); | |
1690 | return err; | |
1691 | } | |
1692 | ||
1693 | static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, | |
1694 | struct net *net) | |
1695 | { | |
1696 | struct net_device *dev; | |
1697 | ||
1698 | for_each_netdev(net, dev) | |
1699 | call_netdevice_unregister_notifiers(nb, dev); | |
1700 | } | |
1701 | ||
881d966b EB |
1702 | static int dev_boot_phase = 1; |
1703 | ||
1da177e4 | 1704 | /** |
722c9a0c | 1705 | * register_netdevice_notifier - register a network notifier block |
1706 | * @nb: notifier | |
1da177e4 | 1707 | * |
722c9a0c | 1708 | * Register a notifier to be called when network device events occur. |
1709 | * The notifier passed is linked into the kernel structures and must | |
1710 | * not be reused until it has been unregistered. A negative errno code | |
1711 | * is returned on a failure. | |
1da177e4 | 1712 | * |
722c9a0c | 1713 | * When registered all registration and up events are replayed |
1714 | * to the new notifier to allow device to have a race free | |
1715 | * view of the network device list. | |
1da177e4 LT |
1716 | */ |
1717 | ||
1718 | int register_netdevice_notifier(struct notifier_block *nb) | |
1719 | { | |
881d966b | 1720 | struct net *net; |
1da177e4 LT |
1721 | int err; |
1722 | ||
328fbe74 KT |
1723 | /* Close race with setup_net() and cleanup_net() */ |
1724 | down_write(&pernet_ops_rwsem); | |
1da177e4 | 1725 | rtnl_lock(); |
f07d5b94 | 1726 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1727 | if (err) |
1728 | goto unlock; | |
881d966b EB |
1729 | if (dev_boot_phase) |
1730 | goto unlock; | |
1731 | for_each_net(net) { | |
afa0df59 JP |
1732 | err = call_netdevice_register_net_notifiers(nb, net); |
1733 | if (err) | |
1734 | goto rollback; | |
1da177e4 | 1735 | } |
fcc5a03a HX |
1736 | |
1737 | unlock: | |
1da177e4 | 1738 | rtnl_unlock(); |
328fbe74 | 1739 | up_write(&pernet_ops_rwsem); |
1da177e4 | 1740 | return err; |
fcc5a03a HX |
1741 | |
1742 | rollback: | |
afa0df59 JP |
1743 | for_each_net_continue_reverse(net) |
1744 | call_netdevice_unregister_net_notifiers(nb, net); | |
c67625a1 PE |
1745 | |
1746 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1747 | goto unlock; |
1da177e4 | 1748 | } |
d1b19dff | 1749 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1750 | |
1751 | /** | |
722c9a0c | 1752 | * unregister_netdevice_notifier - unregister a network notifier block |
1753 | * @nb: notifier | |
1da177e4 | 1754 | * |
722c9a0c | 1755 | * Unregister a notifier previously registered by |
1756 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1757 | * kernel structures and may then be reused. A negative errno code | |
1758 | * is returned on a failure. | |
7d3d43da | 1759 | * |
722c9a0c | 1760 | * After unregistering unregister and down device events are synthesized |
1761 | * for all devices on the device list to the removed notifier to remove | |
1762 | * the need for special case cleanup code. | |
1da177e4 LT |
1763 | */ |
1764 | ||
1765 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1766 | { | |
7d3d43da EB |
1767 | struct net_device *dev; |
1768 | struct net *net; | |
9f514950 HX |
1769 | int err; |
1770 | ||
328fbe74 KT |
1771 | /* Close race with setup_net() and cleanup_net() */ |
1772 | down_write(&pernet_ops_rwsem); | |
9f514950 | 1773 | rtnl_lock(); |
f07d5b94 | 1774 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1775 | if (err) |
1776 | goto unlock; | |
1777 | ||
1778 | for_each_net(net) { | |
1779 | for_each_netdev(net, dev) { | |
1780 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1781 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1782 | dev); | |
1783 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1784 | } |
351638e7 | 1785 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1786 | } |
1787 | } | |
1788 | unlock: | |
9f514950 | 1789 | rtnl_unlock(); |
328fbe74 | 1790 | up_write(&pernet_ops_rwsem); |
9f514950 | 1791 | return err; |
1da177e4 | 1792 | } |
d1b19dff | 1793 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1794 | |
a30c7b42 JP |
1795 | /** |
1796 | * register_netdevice_notifier_net - register a per-netns network notifier block | |
1797 | * @net: network namespace | |
1798 | * @nb: notifier | |
1799 | * | |
1800 | * Register a notifier to be called when network device events occur. | |
1801 | * The notifier passed is linked into the kernel structures and must | |
1802 | * not be reused until it has been unregistered. A negative errno code | |
1803 | * is returned on a failure. | |
1804 | * | |
1805 | * When registered all registration and up events are replayed | |
1806 | * to the new notifier to allow device to have a race free | |
1807 | * view of the network device list. | |
1808 | */ | |
1809 | ||
1810 | int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) | |
1811 | { | |
1812 | int err; | |
1813 | ||
1814 | rtnl_lock(); | |
1815 | err = raw_notifier_chain_register(&net->netdev_chain, nb); | |
1816 | if (err) | |
1817 | goto unlock; | |
1818 | if (dev_boot_phase) | |
1819 | goto unlock; | |
1820 | ||
1821 | err = call_netdevice_register_net_notifiers(nb, net); | |
1822 | if (err) | |
1823 | goto chain_unregister; | |
1824 | ||
1825 | unlock: | |
1826 | rtnl_unlock(); | |
1827 | return err; | |
1828 | ||
1829 | chain_unregister: | |
1830 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
1831 | goto unlock; | |
1832 | } | |
1833 | EXPORT_SYMBOL(register_netdevice_notifier_net); | |
1834 | ||
1835 | /** | |
1836 | * unregister_netdevice_notifier_net - unregister a per-netns | |
1837 | * network notifier block | |
1838 | * @net: network namespace | |
1839 | * @nb: notifier | |
1840 | * | |
1841 | * Unregister a notifier previously registered by | |
1842 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1843 | * kernel structures and may then be reused. A negative errno code | |
1844 | * is returned on a failure. | |
1845 | * | |
1846 | * After unregistering unregister and down device events are synthesized | |
1847 | * for all devices on the device list to the removed notifier to remove | |
1848 | * the need for special case cleanup code. | |
1849 | */ | |
1850 | ||
1851 | int unregister_netdevice_notifier_net(struct net *net, | |
1852 | struct notifier_block *nb) | |
1853 | { | |
1854 | int err; | |
1855 | ||
1856 | rtnl_lock(); | |
1857 | err = raw_notifier_chain_unregister(&net->netdev_chain, nb); | |
1858 | if (err) | |
1859 | goto unlock; | |
1860 | ||
1861 | call_netdevice_unregister_net_notifiers(nb, net); | |
1862 | ||
1863 | unlock: | |
1864 | rtnl_unlock(); | |
1865 | return err; | |
1866 | } | |
1867 | EXPORT_SYMBOL(unregister_netdevice_notifier_net); | |
1868 | ||
351638e7 JP |
1869 | /** |
1870 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1871 | * @val: value passed unmodified to notifier function | |
351638e7 JP |
1872 | * @info: notifier information data |
1873 | * | |
1874 | * Call all network notifier blocks. Parameters and return value | |
1875 | * are as for raw_notifier_call_chain(). | |
1876 | */ | |
1877 | ||
1d143d9f | 1878 | static int call_netdevice_notifiers_info(unsigned long val, |
1d143d9f | 1879 | struct netdev_notifier_info *info) |
351638e7 | 1880 | { |
a30c7b42 JP |
1881 | struct net *net = dev_net(info->dev); |
1882 | int ret; | |
1883 | ||
351638e7 | 1884 | ASSERT_RTNL(); |
a30c7b42 JP |
1885 | |
1886 | /* Run per-netns notifier block chain first, then run the global one. | |
1887 | * Hopefully, one day, the global one is going to be removed after | |
1888 | * all notifier block registrators get converted to be per-netns. | |
1889 | */ | |
1890 | ret = raw_notifier_call_chain(&net->netdev_chain, val, info); | |
1891 | if (ret & NOTIFY_STOP_MASK) | |
1892 | return ret; | |
351638e7 JP |
1893 | return raw_notifier_call_chain(&netdev_chain, val, info); |
1894 | } | |
351638e7 | 1895 | |
26372605 PM |
1896 | static int call_netdevice_notifiers_extack(unsigned long val, |
1897 | struct net_device *dev, | |
1898 | struct netlink_ext_ack *extack) | |
1899 | { | |
1900 | struct netdev_notifier_info info = { | |
1901 | .dev = dev, | |
1902 | .extack = extack, | |
1903 | }; | |
1904 | ||
1905 | return call_netdevice_notifiers_info(val, &info); | |
1906 | } | |
1907 | ||
1da177e4 LT |
1908 | /** |
1909 | * call_netdevice_notifiers - call all network notifier blocks | |
1910 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1911 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1912 | * |
1913 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1914 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1915 | */ |
1916 | ||
ad7379d4 | 1917 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1918 | { |
26372605 | 1919 | return call_netdevice_notifiers_extack(val, dev, NULL); |
1da177e4 | 1920 | } |
edf947f1 | 1921 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1922 | |
af7d6cce SD |
1923 | /** |
1924 | * call_netdevice_notifiers_mtu - call all network notifier blocks | |
1925 | * @val: value passed unmodified to notifier function | |
1926 | * @dev: net_device pointer passed unmodified to notifier function | |
1927 | * @arg: additional u32 argument passed to the notifier function | |
1928 | * | |
1929 | * Call all network notifier blocks. Parameters and return value | |
1930 | * are as for raw_notifier_call_chain(). | |
1931 | */ | |
1932 | static int call_netdevice_notifiers_mtu(unsigned long val, | |
1933 | struct net_device *dev, u32 arg) | |
1934 | { | |
1935 | struct netdev_notifier_info_ext info = { | |
1936 | .info.dev = dev, | |
1937 | .ext.mtu = arg, | |
1938 | }; | |
1939 | ||
1940 | BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); | |
1941 | ||
1942 | return call_netdevice_notifiers_info(val, &info.info); | |
1943 | } | |
1944 | ||
1cf51900 | 1945 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 1946 | static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); |
4577139b DB |
1947 | |
1948 | void net_inc_ingress_queue(void) | |
1949 | { | |
aabf6772 | 1950 | static_branch_inc(&ingress_needed_key); |
4577139b DB |
1951 | } |
1952 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
1953 | ||
1954 | void net_dec_ingress_queue(void) | |
1955 | { | |
aabf6772 | 1956 | static_branch_dec(&ingress_needed_key); |
4577139b DB |
1957 | } |
1958 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
1959 | #endif | |
1960 | ||
1f211a1b | 1961 | #ifdef CONFIG_NET_EGRESS |
aabf6772 | 1962 | static DEFINE_STATIC_KEY_FALSE(egress_needed_key); |
1f211a1b DB |
1963 | |
1964 | void net_inc_egress_queue(void) | |
1965 | { | |
aabf6772 | 1966 | static_branch_inc(&egress_needed_key); |
1f211a1b DB |
1967 | } |
1968 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
1969 | ||
1970 | void net_dec_egress_queue(void) | |
1971 | { | |
aabf6772 | 1972 | static_branch_dec(&egress_needed_key); |
1f211a1b DB |
1973 | } |
1974 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
1975 | #endif | |
1976 | ||
39e83922 | 1977 | static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); |
e9666d10 | 1978 | #ifdef CONFIG_JUMP_LABEL |
b90e5794 | 1979 | static atomic_t netstamp_needed_deferred; |
13baa00a | 1980 | static atomic_t netstamp_wanted; |
5fa8bbda | 1981 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 1982 | { |
b90e5794 | 1983 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 1984 | int wanted; |
b90e5794 | 1985 | |
13baa00a ED |
1986 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
1987 | if (wanted > 0) | |
39e83922 | 1988 | static_branch_enable(&netstamp_needed_key); |
13baa00a | 1989 | else |
39e83922 | 1990 | static_branch_disable(&netstamp_needed_key); |
5fa8bbda ED |
1991 | } |
1992 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 1993 | #endif |
5fa8bbda ED |
1994 | |
1995 | void net_enable_timestamp(void) | |
1996 | { | |
e9666d10 | 1997 | #ifdef CONFIG_JUMP_LABEL |
13baa00a ED |
1998 | int wanted; |
1999 | ||
2000 | while (1) { | |
2001 | wanted = atomic_read(&netstamp_wanted); | |
2002 | if (wanted <= 0) | |
2003 | break; | |
2004 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) | |
2005 | return; | |
2006 | } | |
2007 | atomic_inc(&netstamp_needed_deferred); | |
2008 | schedule_work(&netstamp_work); | |
2009 | #else | |
39e83922 | 2010 | static_branch_inc(&netstamp_needed_key); |
13baa00a | 2011 | #endif |
1da177e4 | 2012 | } |
d1b19dff | 2013 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
2014 | |
2015 | void net_disable_timestamp(void) | |
2016 | { | |
e9666d10 | 2017 | #ifdef CONFIG_JUMP_LABEL |
13baa00a ED |
2018 | int wanted; |
2019 | ||
2020 | while (1) { | |
2021 | wanted = atomic_read(&netstamp_wanted); | |
2022 | if (wanted <= 1) | |
2023 | break; | |
2024 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) | |
2025 | return; | |
2026 | } | |
2027 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
2028 | schedule_work(&netstamp_work); |
2029 | #else | |
39e83922 | 2030 | static_branch_dec(&netstamp_needed_key); |
5fa8bbda | 2031 | #endif |
1da177e4 | 2032 | } |
d1b19dff | 2033 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 2034 | |
3b098e2d | 2035 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 2036 | { |
2456e855 | 2037 | skb->tstamp = 0; |
39e83922 | 2038 | if (static_branch_unlikely(&netstamp_needed_key)) |
a61bbcf2 | 2039 | __net_timestamp(skb); |
1da177e4 LT |
2040 | } |
2041 | ||
39e83922 DB |
2042 | #define net_timestamp_check(COND, SKB) \ |
2043 | if (static_branch_unlikely(&netstamp_needed_key)) { \ | |
2044 | if ((COND) && !(SKB)->tstamp) \ | |
2045 | __net_timestamp(SKB); \ | |
2046 | } \ | |
3b098e2d | 2047 | |
f4b05d27 | 2048 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 DL |
2049 | { |
2050 | unsigned int len; | |
2051 | ||
2052 | if (!(dev->flags & IFF_UP)) | |
2053 | return false; | |
2054 | ||
2055 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
2056 | if (skb->len <= len) | |
2057 | return true; | |
2058 | ||
2059 | /* if TSO is enabled, we don't care about the length as the packet | |
2060 | * could be forwarded without being segmented before | |
2061 | */ | |
2062 | if (skb_is_gso(skb)) | |
2063 | return true; | |
2064 | ||
2065 | return false; | |
2066 | } | |
1ee481fb | 2067 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 2068 | |
a0265d28 HX |
2069 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
2070 | { | |
4e3264d2 | 2071 | int ret = ____dev_forward_skb(dev, skb); |
a0265d28 | 2072 | |
4e3264d2 MKL |
2073 | if (likely(!ret)) { |
2074 | skb->protocol = eth_type_trans(skb, dev); | |
2075 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
2076 | } | |
a0265d28 | 2077 | |
4e3264d2 | 2078 | return ret; |
a0265d28 HX |
2079 | } |
2080 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
2081 | ||
44540960 AB |
2082 | /** |
2083 | * dev_forward_skb - loopback an skb to another netif | |
2084 | * | |
2085 | * @dev: destination network device | |
2086 | * @skb: buffer to forward | |
2087 | * | |
2088 | * return values: | |
2089 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 2090 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
2091 | * |
2092 | * dev_forward_skb can be used for injecting an skb from the | |
2093 | * start_xmit function of one device into the receive queue | |
2094 | * of another device. | |
2095 | * | |
2096 | * The receiving device may be in another namespace, so | |
2097 | * we have to clear all information in the skb that could | |
2098 | * impact namespace isolation. | |
2099 | */ | |
2100 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
2101 | { | |
a0265d28 | 2102 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
2103 | } |
2104 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
2105 | ||
71d9dec2 CG |
2106 | static inline int deliver_skb(struct sk_buff *skb, |
2107 | struct packet_type *pt_prev, | |
2108 | struct net_device *orig_dev) | |
2109 | { | |
1f8b977a | 2110 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
1080e512 | 2111 | return -ENOMEM; |
63354797 | 2112 | refcount_inc(&skb->users); |
71d9dec2 CG |
2113 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
2114 | } | |
2115 | ||
7866a621 SN |
2116 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
2117 | struct packet_type **pt, | |
fbcb2170 JP |
2118 | struct net_device *orig_dev, |
2119 | __be16 type, | |
7866a621 SN |
2120 | struct list_head *ptype_list) |
2121 | { | |
2122 | struct packet_type *ptype, *pt_prev = *pt; | |
2123 | ||
2124 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
2125 | if (ptype->type != type) | |
2126 | continue; | |
2127 | if (pt_prev) | |
fbcb2170 | 2128 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
2129 | pt_prev = ptype; |
2130 | } | |
2131 | *pt = pt_prev; | |
2132 | } | |
2133 | ||
c0de08d0 EL |
2134 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
2135 | { | |
a3d744e9 | 2136 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
2137 | return false; |
2138 | ||
2139 | if (ptype->id_match) | |
2140 | return ptype->id_match(ptype, skb->sk); | |
2141 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
2142 | return true; | |
2143 | ||
2144 | return false; | |
2145 | } | |
2146 | ||
9f9a742d MR |
2147 | /** |
2148 | * dev_nit_active - return true if any network interface taps are in use | |
2149 | * | |
2150 | * @dev: network device to check for the presence of taps | |
2151 | */ | |
2152 | bool dev_nit_active(struct net_device *dev) | |
2153 | { | |
2154 | return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); | |
2155 | } | |
2156 | EXPORT_SYMBOL_GPL(dev_nit_active); | |
2157 | ||
1da177e4 LT |
2158 | /* |
2159 | * Support routine. Sends outgoing frames to any network | |
2160 | * taps currently in use. | |
2161 | */ | |
2162 | ||
74b20582 | 2163 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
2164 | { |
2165 | struct packet_type *ptype; | |
71d9dec2 CG |
2166 | struct sk_buff *skb2 = NULL; |
2167 | struct packet_type *pt_prev = NULL; | |
7866a621 | 2168 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 2169 | |
1da177e4 | 2170 | rcu_read_lock(); |
7866a621 SN |
2171 | again: |
2172 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
fa788d98 VW |
2173 | if (ptype->ignore_outgoing) |
2174 | continue; | |
2175 | ||
1da177e4 LT |
2176 | /* Never send packets back to the socket |
2177 | * they originated from - MvS (miquels@drinkel.ow.org) | |
2178 | */ | |
7866a621 SN |
2179 | if (skb_loop_sk(ptype, skb)) |
2180 | continue; | |
71d9dec2 | 2181 | |
7866a621 SN |
2182 | if (pt_prev) { |
2183 | deliver_skb(skb2, pt_prev, skb->dev); | |
2184 | pt_prev = ptype; | |
2185 | continue; | |
2186 | } | |
1da177e4 | 2187 | |
7866a621 SN |
2188 | /* need to clone skb, done only once */ |
2189 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
2190 | if (!skb2) | |
2191 | goto out_unlock; | |
70978182 | 2192 | |
7866a621 | 2193 | net_timestamp_set(skb2); |
1da177e4 | 2194 | |
7866a621 SN |
2195 | /* skb->nh should be correctly |
2196 | * set by sender, so that the second statement is | |
2197 | * just protection against buggy protocols. | |
2198 | */ | |
2199 | skb_reset_mac_header(skb2); | |
2200 | ||
2201 | if (skb_network_header(skb2) < skb2->data || | |
2202 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
2203 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
2204 | ntohs(skb2->protocol), | |
2205 | dev->name); | |
2206 | skb_reset_network_header(skb2); | |
1da177e4 | 2207 | } |
7866a621 SN |
2208 | |
2209 | skb2->transport_header = skb2->network_header; | |
2210 | skb2->pkt_type = PACKET_OUTGOING; | |
2211 | pt_prev = ptype; | |
2212 | } | |
2213 | ||
2214 | if (ptype_list == &ptype_all) { | |
2215 | ptype_list = &dev->ptype_all; | |
2216 | goto again; | |
1da177e4 | 2217 | } |
7866a621 | 2218 | out_unlock: |
581fe0ea WB |
2219 | if (pt_prev) { |
2220 | if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) | |
2221 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
2222 | else | |
2223 | kfree_skb(skb2); | |
2224 | } | |
1da177e4 LT |
2225 | rcu_read_unlock(); |
2226 | } | |
74b20582 | 2227 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 2228 | |
2c53040f BH |
2229 | /** |
2230 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
2231 | * @dev: Network device |
2232 | * @txq: number of queues available | |
2233 | * | |
2234 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
2235 | * valid. To resolve this verify the tc mapping remains valid and if | |
2236 | * not NULL the mapping. With no priorities mapping to this | |
2237 | * offset/count pair it will no longer be used. In the worst case TC0 | |
2238 | * is invalid nothing can be done so disable priority mappings. If is | |
2239 | * expected that drivers will fix this mapping if they can before | |
2240 | * calling netif_set_real_num_tx_queues. | |
2241 | */ | |
bb134d22 | 2242 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
2243 | { |
2244 | int i; | |
2245 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2246 | ||
2247 | /* If TC0 is invalidated disable TC mapping */ | |
2248 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 2249 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
2250 | dev->num_tc = 0; |
2251 | return; | |
2252 | } | |
2253 | ||
2254 | /* Invalidated prio to tc mappings set to TC0 */ | |
2255 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
2256 | int q = netdev_get_prio_tc_map(dev, i); | |
2257 | ||
2258 | tc = &dev->tc_to_txq[q]; | |
2259 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
2260 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
2261 | i, q); | |
4f57c087 JF |
2262 | netdev_set_prio_tc_map(dev, i, 0); |
2263 | } | |
2264 | } | |
2265 | } | |
2266 | ||
8d059b0f AD |
2267 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
2268 | { | |
2269 | if (dev->num_tc) { | |
2270 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2271 | int i; | |
2272 | ||
ffcfe25b | 2273 | /* walk through the TCs and see if it falls into any of them */ |
8d059b0f AD |
2274 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { |
2275 | if ((txq - tc->offset) < tc->count) | |
2276 | return i; | |
2277 | } | |
2278 | ||
ffcfe25b | 2279 | /* didn't find it, just return -1 to indicate no match */ |
8d059b0f AD |
2280 | return -1; |
2281 | } | |
2282 | ||
2283 | return 0; | |
2284 | } | |
8a5f2166 | 2285 | EXPORT_SYMBOL(netdev_txq_to_tc); |
8d059b0f | 2286 | |
537c00de | 2287 | #ifdef CONFIG_XPS |
04157469 AN |
2288 | struct static_key xps_needed __read_mostly; |
2289 | EXPORT_SYMBOL(xps_needed); | |
2290 | struct static_key xps_rxqs_needed __read_mostly; | |
2291 | EXPORT_SYMBOL(xps_rxqs_needed); | |
537c00de AD |
2292 | static DEFINE_MUTEX(xps_map_mutex); |
2293 | #define xmap_dereference(P) \ | |
2294 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
2295 | ||
6234f874 AD |
2296 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
2297 | int tci, u16 index) | |
537c00de | 2298 | { |
10cdc3f3 AD |
2299 | struct xps_map *map = NULL; |
2300 | int pos; | |
537c00de | 2301 | |
10cdc3f3 | 2302 | if (dev_maps) |
80d19669 | 2303 | map = xmap_dereference(dev_maps->attr_map[tci]); |
6234f874 AD |
2304 | if (!map) |
2305 | return false; | |
537c00de | 2306 | |
6234f874 AD |
2307 | for (pos = map->len; pos--;) { |
2308 | if (map->queues[pos] != index) | |
2309 | continue; | |
2310 | ||
2311 | if (map->len > 1) { | |
2312 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2313 | break; |
537c00de | 2314 | } |
6234f874 | 2315 | |
80d19669 | 2316 | RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); |
6234f874 AD |
2317 | kfree_rcu(map, rcu); |
2318 | return false; | |
537c00de AD |
2319 | } |
2320 | ||
6234f874 | 2321 | return true; |
10cdc3f3 AD |
2322 | } |
2323 | ||
6234f874 AD |
2324 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2325 | struct xps_dev_maps *dev_maps, | |
2326 | int cpu, u16 offset, u16 count) | |
2327 | { | |
184c449f AD |
2328 | int num_tc = dev->num_tc ? : 1; |
2329 | bool active = false; | |
2330 | int tci; | |
6234f874 | 2331 | |
184c449f AD |
2332 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2333 | int i, j; | |
2334 | ||
2335 | for (i = count, j = offset; i--; j++) { | |
6358d49a | 2336 | if (!remove_xps_queue(dev_maps, tci, j)) |
184c449f AD |
2337 | break; |
2338 | } | |
2339 | ||
2340 | active |= i < 0; | |
6234f874 AD |
2341 | } |
2342 | ||
184c449f | 2343 | return active; |
6234f874 AD |
2344 | } |
2345 | ||
867d0ad4 SD |
2346 | static void reset_xps_maps(struct net_device *dev, |
2347 | struct xps_dev_maps *dev_maps, | |
2348 | bool is_rxqs_map) | |
2349 | { | |
2350 | if (is_rxqs_map) { | |
2351 | static_key_slow_dec_cpuslocked(&xps_rxqs_needed); | |
2352 | RCU_INIT_POINTER(dev->xps_rxqs_map, NULL); | |
2353 | } else { | |
2354 | RCU_INIT_POINTER(dev->xps_cpus_map, NULL); | |
2355 | } | |
2356 | static_key_slow_dec_cpuslocked(&xps_needed); | |
2357 | kfree_rcu(dev_maps, rcu); | |
2358 | } | |
2359 | ||
80d19669 AN |
2360 | static void clean_xps_maps(struct net_device *dev, const unsigned long *mask, |
2361 | struct xps_dev_maps *dev_maps, unsigned int nr_ids, | |
2362 | u16 offset, u16 count, bool is_rxqs_map) | |
2363 | { | |
2364 | bool active = false; | |
2365 | int i, j; | |
2366 | ||
2367 | for (j = -1; j = netif_attrmask_next(j, mask, nr_ids), | |
2368 | j < nr_ids;) | |
2369 | active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, | |
2370 | count); | |
867d0ad4 SD |
2371 | if (!active) |
2372 | reset_xps_maps(dev, dev_maps, is_rxqs_map); | |
80d19669 | 2373 | |
f28c020f SD |
2374 | if (!is_rxqs_map) { |
2375 | for (i = offset + (count - 1); count--; i--) { | |
2376 | netdev_queue_numa_node_write( | |
2377 | netdev_get_tx_queue(dev, i), | |
2378 | NUMA_NO_NODE); | |
80d19669 | 2379 | } |
80d19669 AN |
2380 | } |
2381 | } | |
2382 | ||
6234f874 AD |
2383 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, |
2384 | u16 count) | |
10cdc3f3 | 2385 | { |
80d19669 | 2386 | const unsigned long *possible_mask = NULL; |
10cdc3f3 | 2387 | struct xps_dev_maps *dev_maps; |
80d19669 | 2388 | unsigned int nr_ids; |
10cdc3f3 | 2389 | |
04157469 AN |
2390 | if (!static_key_false(&xps_needed)) |
2391 | return; | |
10cdc3f3 | 2392 | |
4d99f660 | 2393 | cpus_read_lock(); |
04157469 | 2394 | mutex_lock(&xps_map_mutex); |
10cdc3f3 | 2395 | |
04157469 AN |
2396 | if (static_key_false(&xps_rxqs_needed)) { |
2397 | dev_maps = xmap_dereference(dev->xps_rxqs_map); | |
2398 | if (dev_maps) { | |
2399 | nr_ids = dev->num_rx_queues; | |
2400 | clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, | |
2401 | offset, count, true); | |
2402 | } | |
537c00de AD |
2403 | } |
2404 | ||
80d19669 AN |
2405 | dev_maps = xmap_dereference(dev->xps_cpus_map); |
2406 | if (!dev_maps) | |
2407 | goto out_no_maps; | |
2408 | ||
2409 | if (num_possible_cpus() > 1) | |
2410 | possible_mask = cpumask_bits(cpu_possible_mask); | |
2411 | nr_ids = nr_cpu_ids; | |
2412 | clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count, | |
2413 | false); | |
024e9679 | 2414 | |
537c00de AD |
2415 | out_no_maps: |
2416 | mutex_unlock(&xps_map_mutex); | |
4d99f660 | 2417 | cpus_read_unlock(); |
537c00de AD |
2418 | } |
2419 | ||
6234f874 AD |
2420 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2421 | { | |
2422 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2423 | } | |
2424 | ||
80d19669 AN |
2425 | static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, |
2426 | u16 index, bool is_rxqs_map) | |
01c5f864 AD |
2427 | { |
2428 | struct xps_map *new_map; | |
2429 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2430 | int i, pos; | |
2431 | ||
2432 | for (pos = 0; map && pos < map->len; pos++) { | |
2433 | if (map->queues[pos] != index) | |
2434 | continue; | |
2435 | return map; | |
2436 | } | |
2437 | ||
80d19669 | 2438 | /* Need to add tx-queue to this CPU's/rx-queue's existing map */ |
01c5f864 AD |
2439 | if (map) { |
2440 | if (pos < map->alloc_len) | |
2441 | return map; | |
2442 | ||
2443 | alloc_len = map->alloc_len * 2; | |
2444 | } | |
2445 | ||
80d19669 AN |
2446 | /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's |
2447 | * map | |
2448 | */ | |
2449 | if (is_rxqs_map) | |
2450 | new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); | |
2451 | else | |
2452 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2453 | cpu_to_node(attr_index)); | |
01c5f864 AD |
2454 | if (!new_map) |
2455 | return NULL; | |
2456 | ||
2457 | for (i = 0; i < pos; i++) | |
2458 | new_map->queues[i] = map->queues[i]; | |
2459 | new_map->alloc_len = alloc_len; | |
2460 | new_map->len = pos; | |
2461 | ||
2462 | return new_map; | |
2463 | } | |
2464 | ||
4d99f660 | 2465 | /* Must be called under cpus_read_lock */ |
80d19669 AN |
2466 | int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, |
2467 | u16 index, bool is_rxqs_map) | |
537c00de | 2468 | { |
80d19669 | 2469 | const unsigned long *online_mask = NULL, *possible_mask = NULL; |
01c5f864 | 2470 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
80d19669 | 2471 | int i, j, tci, numa_node_id = -2; |
184c449f | 2472 | int maps_sz, num_tc = 1, tc = 0; |
537c00de | 2473 | struct xps_map *map, *new_map; |
01c5f864 | 2474 | bool active = false; |
80d19669 | 2475 | unsigned int nr_ids; |
537c00de | 2476 | |
184c449f | 2477 | if (dev->num_tc) { |
ffcfe25b | 2478 | /* Do not allow XPS on subordinate device directly */ |
184c449f | 2479 | num_tc = dev->num_tc; |
ffcfe25b AD |
2480 | if (num_tc < 0) |
2481 | return -EINVAL; | |
2482 | ||
2483 | /* If queue belongs to subordinate dev use its map */ | |
2484 | dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; | |
2485 | ||
184c449f AD |
2486 | tc = netdev_txq_to_tc(dev, index); |
2487 | if (tc < 0) | |
2488 | return -EINVAL; | |
2489 | } | |
2490 | ||
537c00de | 2491 | mutex_lock(&xps_map_mutex); |
80d19669 AN |
2492 | if (is_rxqs_map) { |
2493 | maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); | |
2494 | dev_maps = xmap_dereference(dev->xps_rxqs_map); | |
2495 | nr_ids = dev->num_rx_queues; | |
2496 | } else { | |
2497 | maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); | |
2498 | if (num_possible_cpus() > 1) { | |
2499 | online_mask = cpumask_bits(cpu_online_mask); | |
2500 | possible_mask = cpumask_bits(cpu_possible_mask); | |
2501 | } | |
2502 | dev_maps = xmap_dereference(dev->xps_cpus_map); | |
2503 | nr_ids = nr_cpu_ids; | |
2504 | } | |
537c00de | 2505 | |
80d19669 AN |
2506 | if (maps_sz < L1_CACHE_BYTES) |
2507 | maps_sz = L1_CACHE_BYTES; | |
537c00de | 2508 | |
01c5f864 | 2509 | /* allocate memory for queue storage */ |
80d19669 AN |
2510 | for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), |
2511 | j < nr_ids;) { | |
01c5f864 AD |
2512 | if (!new_dev_maps) |
2513 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2514 | if (!new_dev_maps) { |
2515 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2516 | return -ENOMEM; |
2bb60cb9 | 2517 | } |
01c5f864 | 2518 | |
80d19669 AN |
2519 | tci = j * num_tc + tc; |
2520 | map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) : | |
01c5f864 AD |
2521 | NULL; |
2522 | ||
80d19669 | 2523 | map = expand_xps_map(map, j, index, is_rxqs_map); |
01c5f864 AD |
2524 | if (!map) |
2525 | goto error; | |
2526 | ||
80d19669 | 2527 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); |
01c5f864 AD |
2528 | } |
2529 | ||
2530 | if (!new_dev_maps) | |
2531 | goto out_no_new_maps; | |
2532 | ||
867d0ad4 SD |
2533 | if (!dev_maps) { |
2534 | /* Increment static keys at most once per type */ | |
2535 | static_key_slow_inc_cpuslocked(&xps_needed); | |
2536 | if (is_rxqs_map) | |
2537 | static_key_slow_inc_cpuslocked(&xps_rxqs_needed); | |
2538 | } | |
04157469 | 2539 | |
80d19669 AN |
2540 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2541 | j < nr_ids;) { | |
184c449f | 2542 | /* copy maps belonging to foreign traffic classes */ |
80d19669 | 2543 | for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) { |
184c449f | 2544 | /* fill in the new device map from the old device map */ |
80d19669 AN |
2545 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2546 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
184c449f AD |
2547 | } |
2548 | ||
2549 | /* We need to explicitly update tci as prevous loop | |
2550 | * could break out early if dev_maps is NULL. | |
2551 | */ | |
80d19669 | 2552 | tci = j * num_tc + tc; |
184c449f | 2553 | |
80d19669 AN |
2554 | if (netif_attr_test_mask(j, mask, nr_ids) && |
2555 | netif_attr_test_online(j, online_mask, nr_ids)) { | |
2556 | /* add tx-queue to CPU/rx-queue maps */ | |
01c5f864 AD |
2557 | int pos = 0; |
2558 | ||
80d19669 | 2559 | map = xmap_dereference(new_dev_maps->attr_map[tci]); |
01c5f864 AD |
2560 | while ((pos < map->len) && (map->queues[pos] != index)) |
2561 | pos++; | |
2562 | ||
2563 | if (pos == map->len) | |
2564 | map->queues[map->len++] = index; | |
537c00de | 2565 | #ifdef CONFIG_NUMA |
80d19669 AN |
2566 | if (!is_rxqs_map) { |
2567 | if (numa_node_id == -2) | |
2568 | numa_node_id = cpu_to_node(j); | |
2569 | else if (numa_node_id != cpu_to_node(j)) | |
2570 | numa_node_id = -1; | |
2571 | } | |
537c00de | 2572 | #endif |
01c5f864 AD |
2573 | } else if (dev_maps) { |
2574 | /* fill in the new device map from the old device map */ | |
80d19669 AN |
2575 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2576 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
537c00de | 2577 | } |
01c5f864 | 2578 | |
184c449f AD |
2579 | /* copy maps belonging to foreign traffic classes */ |
2580 | for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { | |
2581 | /* fill in the new device map from the old device map */ | |
80d19669 AN |
2582 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2583 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
184c449f | 2584 | } |
537c00de AD |
2585 | } |
2586 | ||
80d19669 AN |
2587 | if (is_rxqs_map) |
2588 | rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps); | |
2589 | else | |
2590 | rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps); | |
01c5f864 | 2591 | |
537c00de | 2592 | /* Cleanup old maps */ |
184c449f AD |
2593 | if (!dev_maps) |
2594 | goto out_no_old_maps; | |
2595 | ||
80d19669 AN |
2596 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2597 | j < nr_ids;) { | |
2598 | for (i = num_tc, tci = j * num_tc; i--; tci++) { | |
2599 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
2600 | map = xmap_dereference(dev_maps->attr_map[tci]); | |
01c5f864 AD |
2601 | if (map && map != new_map) |
2602 | kfree_rcu(map, rcu); | |
2603 | } | |
537c00de AD |
2604 | } |
2605 | ||
184c449f AD |
2606 | kfree_rcu(dev_maps, rcu); |
2607 | ||
2608 | out_no_old_maps: | |
01c5f864 AD |
2609 | dev_maps = new_dev_maps; |
2610 | active = true; | |
537c00de | 2611 | |
01c5f864 | 2612 | out_no_new_maps: |
80d19669 AN |
2613 | if (!is_rxqs_map) { |
2614 | /* update Tx queue numa node */ | |
2615 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), | |
2616 | (numa_node_id >= 0) ? | |
2617 | numa_node_id : NUMA_NO_NODE); | |
2618 | } | |
537c00de | 2619 | |
01c5f864 AD |
2620 | if (!dev_maps) |
2621 | goto out_no_maps; | |
2622 | ||
80d19669 AN |
2623 | /* removes tx-queue from unused CPUs/rx-queues */ |
2624 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), | |
2625 | j < nr_ids;) { | |
2626 | for (i = tc, tci = j * num_tc; i--; tci++) | |
184c449f | 2627 | active |= remove_xps_queue(dev_maps, tci, index); |
80d19669 AN |
2628 | if (!netif_attr_test_mask(j, mask, nr_ids) || |
2629 | !netif_attr_test_online(j, online_mask, nr_ids)) | |
184c449f AD |
2630 | active |= remove_xps_queue(dev_maps, tci, index); |
2631 | for (i = num_tc - tc, tci++; --i; tci++) | |
2632 | active |= remove_xps_queue(dev_maps, tci, index); | |
01c5f864 AD |
2633 | } |
2634 | ||
2635 | /* free map if not active */ | |
867d0ad4 SD |
2636 | if (!active) |
2637 | reset_xps_maps(dev, dev_maps, is_rxqs_map); | |
01c5f864 AD |
2638 | |
2639 | out_no_maps: | |
537c00de AD |
2640 | mutex_unlock(&xps_map_mutex); |
2641 | ||
2642 | return 0; | |
2643 | error: | |
01c5f864 | 2644 | /* remove any maps that we added */ |
80d19669 AN |
2645 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2646 | j < nr_ids;) { | |
2647 | for (i = num_tc, tci = j * num_tc; i--; tci++) { | |
2648 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
184c449f | 2649 | map = dev_maps ? |
80d19669 | 2650 | xmap_dereference(dev_maps->attr_map[tci]) : |
184c449f AD |
2651 | NULL; |
2652 | if (new_map && new_map != map) | |
2653 | kfree(new_map); | |
2654 | } | |
01c5f864 AD |
2655 | } |
2656 | ||
537c00de AD |
2657 | mutex_unlock(&xps_map_mutex); |
2658 | ||
537c00de AD |
2659 | kfree(new_dev_maps); |
2660 | return -ENOMEM; | |
2661 | } | |
4d99f660 | 2662 | EXPORT_SYMBOL_GPL(__netif_set_xps_queue); |
80d19669 AN |
2663 | |
2664 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, | |
2665 | u16 index) | |
2666 | { | |
4d99f660 AV |
2667 | int ret; |
2668 | ||
2669 | cpus_read_lock(); | |
2670 | ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false); | |
2671 | cpus_read_unlock(); | |
2672 | ||
2673 | return ret; | |
80d19669 | 2674 | } |
537c00de AD |
2675 | EXPORT_SYMBOL(netif_set_xps_queue); |
2676 | ||
2677 | #endif | |
ffcfe25b AD |
2678 | static void netdev_unbind_all_sb_channels(struct net_device *dev) |
2679 | { | |
2680 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2681 | ||
2682 | /* Unbind any subordinate channels */ | |
2683 | while (txq-- != &dev->_tx[0]) { | |
2684 | if (txq->sb_dev) | |
2685 | netdev_unbind_sb_channel(dev, txq->sb_dev); | |
2686 | } | |
2687 | } | |
2688 | ||
9cf1f6a8 AD |
2689 | void netdev_reset_tc(struct net_device *dev) |
2690 | { | |
6234f874 AD |
2691 | #ifdef CONFIG_XPS |
2692 | netif_reset_xps_queues_gt(dev, 0); | |
2693 | #endif | |
ffcfe25b AD |
2694 | netdev_unbind_all_sb_channels(dev); |
2695 | ||
2696 | /* Reset TC configuration of device */ | |
9cf1f6a8 AD |
2697 | dev->num_tc = 0; |
2698 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2699 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2700 | } | |
2701 | EXPORT_SYMBOL(netdev_reset_tc); | |
2702 | ||
2703 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2704 | { | |
2705 | if (tc >= dev->num_tc) | |
2706 | return -EINVAL; | |
2707 | ||
6234f874 AD |
2708 | #ifdef CONFIG_XPS |
2709 | netif_reset_xps_queues(dev, offset, count); | |
2710 | #endif | |
9cf1f6a8 AD |
2711 | dev->tc_to_txq[tc].count = count; |
2712 | dev->tc_to_txq[tc].offset = offset; | |
2713 | return 0; | |
2714 | } | |
2715 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2716 | ||
2717 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2718 | { | |
2719 | if (num_tc > TC_MAX_QUEUE) | |
2720 | return -EINVAL; | |
2721 | ||
6234f874 AD |
2722 | #ifdef CONFIG_XPS |
2723 | netif_reset_xps_queues_gt(dev, 0); | |
2724 | #endif | |
ffcfe25b AD |
2725 | netdev_unbind_all_sb_channels(dev); |
2726 | ||
9cf1f6a8 AD |
2727 | dev->num_tc = num_tc; |
2728 | return 0; | |
2729 | } | |
2730 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2731 | ||
ffcfe25b AD |
2732 | void netdev_unbind_sb_channel(struct net_device *dev, |
2733 | struct net_device *sb_dev) | |
2734 | { | |
2735 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2736 | ||
2737 | #ifdef CONFIG_XPS | |
2738 | netif_reset_xps_queues_gt(sb_dev, 0); | |
2739 | #endif | |
2740 | memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); | |
2741 | memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); | |
2742 | ||
2743 | while (txq-- != &dev->_tx[0]) { | |
2744 | if (txq->sb_dev == sb_dev) | |
2745 | txq->sb_dev = NULL; | |
2746 | } | |
2747 | } | |
2748 | EXPORT_SYMBOL(netdev_unbind_sb_channel); | |
2749 | ||
2750 | int netdev_bind_sb_channel_queue(struct net_device *dev, | |
2751 | struct net_device *sb_dev, | |
2752 | u8 tc, u16 count, u16 offset) | |
2753 | { | |
2754 | /* Make certain the sb_dev and dev are already configured */ | |
2755 | if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) | |
2756 | return -EINVAL; | |
2757 | ||
2758 | /* We cannot hand out queues we don't have */ | |
2759 | if ((offset + count) > dev->real_num_tx_queues) | |
2760 | return -EINVAL; | |
2761 | ||
2762 | /* Record the mapping */ | |
2763 | sb_dev->tc_to_txq[tc].count = count; | |
2764 | sb_dev->tc_to_txq[tc].offset = offset; | |
2765 | ||
2766 | /* Provide a way for Tx queue to find the tc_to_txq map or | |
2767 | * XPS map for itself. | |
2768 | */ | |
2769 | while (count--) | |
2770 | netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; | |
2771 | ||
2772 | return 0; | |
2773 | } | |
2774 | EXPORT_SYMBOL(netdev_bind_sb_channel_queue); | |
2775 | ||
2776 | int netdev_set_sb_channel(struct net_device *dev, u16 channel) | |
2777 | { | |
2778 | /* Do not use a multiqueue device to represent a subordinate channel */ | |
2779 | if (netif_is_multiqueue(dev)) | |
2780 | return -ENODEV; | |
2781 | ||
2782 | /* We allow channels 1 - 32767 to be used for subordinate channels. | |
2783 | * Channel 0 is meant to be "native" mode and used only to represent | |
2784 | * the main root device. We allow writing 0 to reset the device back | |
2785 | * to normal mode after being used as a subordinate channel. | |
2786 | */ | |
2787 | if (channel > S16_MAX) | |
2788 | return -EINVAL; | |
2789 | ||
2790 | dev->num_tc = -channel; | |
2791 | ||
2792 | return 0; | |
2793 | } | |
2794 | EXPORT_SYMBOL(netdev_set_sb_channel); | |
2795 | ||
f0796d5c JF |
2796 | /* |
2797 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
3a053b1a | 2798 | * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. |
f0796d5c | 2799 | */ |
e6484930 | 2800 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2801 | { |
ac5b7019 | 2802 | bool disabling; |
1d24eb48 TH |
2803 | int rc; |
2804 | ||
ac5b7019 JK |
2805 | disabling = txq < dev->real_num_tx_queues; |
2806 | ||
e6484930 TH |
2807 | if (txq < 1 || txq > dev->num_tx_queues) |
2808 | return -EINVAL; | |
f0796d5c | 2809 | |
5c56580b BH |
2810 | if (dev->reg_state == NETREG_REGISTERED || |
2811 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2812 | ASSERT_RTNL(); |
2813 | ||
1d24eb48 TH |
2814 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2815 | txq); | |
bf264145 TH |
2816 | if (rc) |
2817 | return rc; | |
2818 | ||
4f57c087 JF |
2819 | if (dev->num_tc) |
2820 | netif_setup_tc(dev, txq); | |
2821 | ||
ac5b7019 JK |
2822 | dev->real_num_tx_queues = txq; |
2823 | ||
2824 | if (disabling) { | |
2825 | synchronize_net(); | |
e6484930 | 2826 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2827 | #ifdef CONFIG_XPS |
2828 | netif_reset_xps_queues_gt(dev, txq); | |
2829 | #endif | |
2830 | } | |
ac5b7019 JK |
2831 | } else { |
2832 | dev->real_num_tx_queues = txq; | |
f0796d5c | 2833 | } |
e6484930 | 2834 | |
e6484930 | 2835 | return 0; |
f0796d5c JF |
2836 | } |
2837 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2838 | |
a953be53 | 2839 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2840 | /** |
2841 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2842 | * @dev: Network device | |
2843 | * @rxq: Actual number of RX queues | |
2844 | * | |
2845 | * This must be called either with the rtnl_lock held or before | |
2846 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2847 | * negative error code. If called before registration, it always |
2848 | * succeeds. | |
62fe0b40 BH |
2849 | */ |
2850 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2851 | { | |
2852 | int rc; | |
2853 | ||
bd25fa7b TH |
2854 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2855 | return -EINVAL; | |
2856 | ||
62fe0b40 BH |
2857 | if (dev->reg_state == NETREG_REGISTERED) { |
2858 | ASSERT_RTNL(); | |
2859 | ||
62fe0b40 BH |
2860 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2861 | rxq); | |
2862 | if (rc) | |
2863 | return rc; | |
62fe0b40 BH |
2864 | } |
2865 | ||
2866 | dev->real_num_rx_queues = rxq; | |
2867 | return 0; | |
2868 | } | |
2869 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2870 | #endif | |
2871 | ||
2c53040f BH |
2872 | /** |
2873 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2874 | * |
2875 | * This routine should set an upper limit on the number of RSS queues | |
2876 | * used by default by multiqueue devices. | |
2877 | */ | |
a55b138b | 2878 | int netif_get_num_default_rss_queues(void) |
16917b87 | 2879 | { |
40e4e713 HS |
2880 | return is_kdump_kernel() ? |
2881 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
2882 | } |
2883 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2884 | ||
3bcb846c | 2885 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 2886 | { |
def82a1d JP |
2887 | struct softnet_data *sd; |
2888 | unsigned long flags; | |
56079431 | 2889 | |
def82a1d | 2890 | local_irq_save(flags); |
903ceff7 | 2891 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2892 | q->next_sched = NULL; |
2893 | *sd->output_queue_tailp = q; | |
2894 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2895 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2896 | local_irq_restore(flags); | |
2897 | } | |
2898 | ||
2899 | void __netif_schedule(struct Qdisc *q) | |
2900 | { | |
2901 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2902 | __netif_reschedule(q); | |
56079431 DV |
2903 | } |
2904 | EXPORT_SYMBOL(__netif_schedule); | |
2905 | ||
e6247027 ED |
2906 | struct dev_kfree_skb_cb { |
2907 | enum skb_free_reason reason; | |
2908 | }; | |
2909 | ||
2910 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2911 | { |
e6247027 ED |
2912 | return (struct dev_kfree_skb_cb *)skb->cb; |
2913 | } | |
2914 | ||
46e5da40 JF |
2915 | void netif_schedule_queue(struct netdev_queue *txq) |
2916 | { | |
2917 | rcu_read_lock(); | |
5be5515a | 2918 | if (!netif_xmit_stopped(txq)) { |
46e5da40 JF |
2919 | struct Qdisc *q = rcu_dereference(txq->qdisc); |
2920 | ||
2921 | __netif_schedule(q); | |
2922 | } | |
2923 | rcu_read_unlock(); | |
2924 | } | |
2925 | EXPORT_SYMBOL(netif_schedule_queue); | |
2926 | ||
46e5da40 JF |
2927 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
2928 | { | |
2929 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2930 | struct Qdisc *q; | |
2931 | ||
2932 | rcu_read_lock(); | |
2933 | q = rcu_dereference(dev_queue->qdisc); | |
2934 | __netif_schedule(q); | |
2935 | rcu_read_unlock(); | |
2936 | } | |
2937 | } | |
2938 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2939 | ||
e6247027 | 2940 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2941 | { |
e6247027 | 2942 | unsigned long flags; |
56079431 | 2943 | |
9899886d MJ |
2944 | if (unlikely(!skb)) |
2945 | return; | |
2946 | ||
63354797 | 2947 | if (likely(refcount_read(&skb->users) == 1)) { |
e6247027 | 2948 | smp_rmb(); |
63354797 RE |
2949 | refcount_set(&skb->users, 0); |
2950 | } else if (likely(!refcount_dec_and_test(&skb->users))) { | |
e6247027 | 2951 | return; |
bea3348e | 2952 | } |
e6247027 ED |
2953 | get_kfree_skb_cb(skb)->reason = reason; |
2954 | local_irq_save(flags); | |
2955 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2956 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2957 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2958 | local_irq_restore(flags); | |
56079431 | 2959 | } |
e6247027 | 2960 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2961 | |
e6247027 | 2962 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2963 | { |
2964 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2965 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2966 | else |
2967 | dev_kfree_skb(skb); | |
2968 | } | |
e6247027 | 2969 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2970 | |
2971 | ||
bea3348e SH |
2972 | /** |
2973 | * netif_device_detach - mark device as removed | |
2974 | * @dev: network device | |
2975 | * | |
2976 | * Mark device as removed from system and therefore no longer available. | |
2977 | */ | |
56079431 DV |
2978 | void netif_device_detach(struct net_device *dev) |
2979 | { | |
2980 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2981 | netif_running(dev)) { | |
d543103a | 2982 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2983 | } |
2984 | } | |
2985 | EXPORT_SYMBOL(netif_device_detach); | |
2986 | ||
bea3348e SH |
2987 | /** |
2988 | * netif_device_attach - mark device as attached | |
2989 | * @dev: network device | |
2990 | * | |
2991 | * Mark device as attached from system and restart if needed. | |
2992 | */ | |
56079431 DV |
2993 | void netif_device_attach(struct net_device *dev) |
2994 | { | |
2995 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2996 | netif_running(dev)) { | |
d543103a | 2997 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2998 | __netdev_watchdog_up(dev); |
56079431 DV |
2999 | } |
3000 | } | |
3001 | EXPORT_SYMBOL(netif_device_attach); | |
3002 | ||
5605c762 JP |
3003 | /* |
3004 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
3005 | * to be used as a distribution range. | |
3006 | */ | |
eadec877 AD |
3007 | static u16 skb_tx_hash(const struct net_device *dev, |
3008 | const struct net_device *sb_dev, | |
3009 | struct sk_buff *skb) | |
5605c762 JP |
3010 | { |
3011 | u32 hash; | |
3012 | u16 qoffset = 0; | |
1b837d48 | 3013 | u16 qcount = dev->real_num_tx_queues; |
5605c762 | 3014 | |
eadec877 AD |
3015 | if (dev->num_tc) { |
3016 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
3017 | ||
3018 | qoffset = sb_dev->tc_to_txq[tc].offset; | |
3019 | qcount = sb_dev->tc_to_txq[tc].count; | |
3020 | } | |
3021 | ||
5605c762 JP |
3022 | if (skb_rx_queue_recorded(skb)) { |
3023 | hash = skb_get_rx_queue(skb); | |
1b837d48 AD |
3024 | while (unlikely(hash >= qcount)) |
3025 | hash -= qcount; | |
eadec877 | 3026 | return hash + qoffset; |
5605c762 JP |
3027 | } |
3028 | ||
3029 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
3030 | } | |
5605c762 | 3031 | |
36c92474 BH |
3032 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
3033 | { | |
84d15ae5 | 3034 | static const netdev_features_t null_features; |
36c92474 | 3035 | struct net_device *dev = skb->dev; |
88ad4175 | 3036 | const char *name = ""; |
36c92474 | 3037 | |
c846ad9b BG |
3038 | if (!net_ratelimit()) |
3039 | return; | |
3040 | ||
88ad4175 BM |
3041 | if (dev) { |
3042 | if (dev->dev.parent) | |
3043 | name = dev_driver_string(dev->dev.parent); | |
3044 | else | |
3045 | name = netdev_name(dev); | |
3046 | } | |
6413139d WB |
3047 | skb_dump(KERN_WARNING, skb, false); |
3048 | WARN(1, "%s: caps=(%pNF, %pNF)\n", | |
88ad4175 | 3049 | name, dev ? &dev->features : &null_features, |
6413139d | 3050 | skb->sk ? &skb->sk->sk_route_caps : &null_features); |
36c92474 BH |
3051 | } |
3052 | ||
1da177e4 LT |
3053 | /* |
3054 | * Invalidate hardware checksum when packet is to be mangled, and | |
3055 | * complete checksum manually on outgoing path. | |
3056 | */ | |
84fa7933 | 3057 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 3058 | { |
d3bc23e7 | 3059 | __wsum csum; |
663ead3b | 3060 | int ret = 0, offset; |
1da177e4 | 3061 | |
84fa7933 | 3062 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
3063 | goto out_set_summed; |
3064 | ||
3065 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
3066 | skb_warn_bad_offload(skb); |
3067 | return -EINVAL; | |
1da177e4 LT |
3068 | } |
3069 | ||
cef401de ED |
3070 | /* Before computing a checksum, we should make sure no frag could |
3071 | * be modified by an external entity : checksum could be wrong. | |
3072 | */ | |
3073 | if (skb_has_shared_frag(skb)) { | |
3074 | ret = __skb_linearize(skb); | |
3075 | if (ret) | |
3076 | goto out; | |
3077 | } | |
3078 | ||
55508d60 | 3079 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
3080 | BUG_ON(offset >= skb_headlen(skb)); |
3081 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
3082 | ||
3083 | offset += skb->csum_offset; | |
3084 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
3085 | ||
8211fbfa HK |
3086 | ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); |
3087 | if (ret) | |
3088 | goto out; | |
1da177e4 | 3089 | |
4f2e4ad5 | 3090 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 3091 | out_set_summed: |
1da177e4 | 3092 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 3093 | out: |
1da177e4 LT |
3094 | return ret; |
3095 | } | |
d1b19dff | 3096 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 3097 | |
b72b5bf6 DC |
3098 | int skb_crc32c_csum_help(struct sk_buff *skb) |
3099 | { | |
3100 | __le32 crc32c_csum; | |
3101 | int ret = 0, offset, start; | |
3102 | ||
3103 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
3104 | goto out; | |
3105 | ||
3106 | if (unlikely(skb_is_gso(skb))) | |
3107 | goto out; | |
3108 | ||
3109 | /* Before computing a checksum, we should make sure no frag could | |
3110 | * be modified by an external entity : checksum could be wrong. | |
3111 | */ | |
3112 | if (unlikely(skb_has_shared_frag(skb))) { | |
3113 | ret = __skb_linearize(skb); | |
3114 | if (ret) | |
3115 | goto out; | |
3116 | } | |
3117 | start = skb_checksum_start_offset(skb); | |
3118 | offset = start + offsetof(struct sctphdr, checksum); | |
3119 | if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { | |
3120 | ret = -EINVAL; | |
3121 | goto out; | |
3122 | } | |
8211fbfa HK |
3123 | |
3124 | ret = skb_ensure_writable(skb, offset + sizeof(__le32)); | |
3125 | if (ret) | |
3126 | goto out; | |
3127 | ||
b72b5bf6 DC |
3128 | crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, |
3129 | skb->len - start, ~(__u32)0, | |
3130 | crc32c_csum_stub)); | |
3131 | *(__le32 *)(skb->data + offset) = crc32c_csum; | |
3132 | skb->ip_summed = CHECKSUM_NONE; | |
dba00306 | 3133 | skb->csum_not_inet = 0; |
b72b5bf6 DC |
3134 | out: |
3135 | return ret; | |
3136 | } | |
3137 | ||
53d6471c | 3138 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 3139 | { |
252e3346 | 3140 | __be16 type = skb->protocol; |
f6a78bfc | 3141 | |
19acc327 PS |
3142 | /* Tunnel gso handlers can set protocol to ethernet. */ |
3143 | if (type == htons(ETH_P_TEB)) { | |
3144 | struct ethhdr *eth; | |
3145 | ||
3146 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
3147 | return 0; | |
3148 | ||
1dfe82eb | 3149 | eth = (struct ethhdr *)skb->data; |
19acc327 PS |
3150 | type = eth->h_proto; |
3151 | } | |
3152 | ||
d4bcef3f | 3153 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
3154 | } |
3155 | ||
3156 | /** | |
3157 | * skb_mac_gso_segment - mac layer segmentation handler. | |
3158 | * @skb: buffer to segment | |
3159 | * @features: features for the output path (see dev->features) | |
3160 | */ | |
3161 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
3162 | netdev_features_t features) | |
3163 | { | |
3164 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
3165 | struct packet_offload *ptype; | |
53d6471c VY |
3166 | int vlan_depth = skb->mac_len; |
3167 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
3168 | |
3169 | if (unlikely(!type)) | |
3170 | return ERR_PTR(-EINVAL); | |
3171 | ||
53d6471c | 3172 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
3173 | |
3174 | rcu_read_lock(); | |
22061d80 | 3175 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 3176 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 3177 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
3178 | break; |
3179 | } | |
3180 | } | |
3181 | rcu_read_unlock(); | |
3182 | ||
98e399f8 | 3183 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 3184 | |
f6a78bfc HX |
3185 | return segs; |
3186 | } | |
05e8ef4a PS |
3187 | EXPORT_SYMBOL(skb_mac_gso_segment); |
3188 | ||
3189 | ||
3190 | /* openvswitch calls this on rx path, so we need a different check. | |
3191 | */ | |
3192 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
3193 | { | |
3194 | if (tx_path) | |
0c19f846 WB |
3195 | return skb->ip_summed != CHECKSUM_PARTIAL && |
3196 | skb->ip_summed != CHECKSUM_UNNECESSARY; | |
6e7bc478 ED |
3197 | |
3198 | return skb->ip_summed == CHECKSUM_NONE; | |
05e8ef4a PS |
3199 | } |
3200 | ||
3201 | /** | |
3202 | * __skb_gso_segment - Perform segmentation on skb. | |
3203 | * @skb: buffer to segment | |
3204 | * @features: features for the output path (see dev->features) | |
3205 | * @tx_path: whether it is called in TX path | |
3206 | * | |
3207 | * This function segments the given skb and returns a list of segments. | |
3208 | * | |
3209 | * It may return NULL if the skb requires no segmentation. This is | |
3210 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 KK |
3211 | * |
3212 | * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. | |
05e8ef4a PS |
3213 | */ |
3214 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
3215 | netdev_features_t features, bool tx_path) | |
3216 | { | |
b2504a5d ED |
3217 | struct sk_buff *segs; |
3218 | ||
05e8ef4a PS |
3219 | if (unlikely(skb_needs_check(skb, tx_path))) { |
3220 | int err; | |
3221 | ||
b2504a5d | 3222 | /* We're going to init ->check field in TCP or UDP header */ |
a40e0a66 | 3223 | err = skb_cow_head(skb, 0); |
3224 | if (err < 0) | |
05e8ef4a PS |
3225 | return ERR_PTR(err); |
3226 | } | |
3227 | ||
802ab55a AD |
3228 | /* Only report GSO partial support if it will enable us to |
3229 | * support segmentation on this frame without needing additional | |
3230 | * work. | |
3231 | */ | |
3232 | if (features & NETIF_F_GSO_PARTIAL) { | |
3233 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
3234 | struct net_device *dev = skb->dev; | |
3235 | ||
3236 | partial_features |= dev->features & dev->gso_partial_features; | |
3237 | if (!skb_gso_ok(skb, features | partial_features)) | |
3238 | features &= ~NETIF_F_GSO_PARTIAL; | |
3239 | } | |
3240 | ||
9207f9d4 KK |
3241 | BUILD_BUG_ON(SKB_SGO_CB_OFFSET + |
3242 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); | |
3243 | ||
68c33163 | 3244 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
3245 | SKB_GSO_CB(skb)->encap_level = 0; |
3246 | ||
05e8ef4a PS |
3247 | skb_reset_mac_header(skb); |
3248 | skb_reset_mac_len(skb); | |
3249 | ||
b2504a5d ED |
3250 | segs = skb_mac_gso_segment(skb, features); |
3251 | ||
8d74e9f8 | 3252 | if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) |
b2504a5d ED |
3253 | skb_warn_bad_offload(skb); |
3254 | ||
3255 | return segs; | |
05e8ef4a | 3256 | } |
12b0004d | 3257 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 3258 | |
fb286bb2 HX |
3259 | /* Take action when hardware reception checksum errors are detected. */ |
3260 | #ifdef CONFIG_BUG | |
7fe50ac8 | 3261 | void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) |
fb286bb2 HX |
3262 | { |
3263 | if (net_ratelimit()) { | |
7b6cd1ce | 3264 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
6413139d | 3265 | skb_dump(KERN_ERR, skb, true); |
fb286bb2 HX |
3266 | dump_stack(); |
3267 | } | |
3268 | } | |
3269 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
3270 | #endif | |
3271 | ||
ab74cfeb | 3272 | /* XXX: check that highmem exists at all on the given machine. */ |
c1e756bf | 3273 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 3274 | { |
3d3a8533 | 3275 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 3276 | int i; |
f4563a75 | 3277 | |
5acbbd42 | 3278 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
3279 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
3280 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 3281 | |
ea2ab693 | 3282 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 3283 | return 1; |
ea2ab693 | 3284 | } |
5acbbd42 | 3285 | } |
3d3a8533 | 3286 | #endif |
1da177e4 LT |
3287 | return 0; |
3288 | } | |
1da177e4 | 3289 | |
3b392ddb SH |
3290 | /* If MPLS offload request, verify we are testing hardware MPLS features |
3291 | * instead of standard features for the netdev. | |
3292 | */ | |
d0edc7bf | 3293 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
3294 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
3295 | netdev_features_t features, | |
3296 | __be16 type) | |
3297 | { | |
25cd9ba0 | 3298 | if (eth_p_mpls(type)) |
3b392ddb SH |
3299 | features &= skb->dev->mpls_features; |
3300 | ||
3301 | return features; | |
3302 | } | |
3303 | #else | |
3304 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
3305 | netdev_features_t features, | |
3306 | __be16 type) | |
3307 | { | |
3308 | return features; | |
3309 | } | |
3310 | #endif | |
3311 | ||
c8f44aff | 3312 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 3313 | netdev_features_t features) |
f01a5236 | 3314 | { |
53d6471c | 3315 | int tmp; |
3b392ddb SH |
3316 | __be16 type; |
3317 | ||
3318 | type = skb_network_protocol(skb, &tmp); | |
3319 | features = net_mpls_features(skb, features, type); | |
53d6471c | 3320 | |
c0d680e5 | 3321 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 3322 | !can_checksum_protocol(features, type)) { |
996e8021 | 3323 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 3324 | } |
7be2c82c ED |
3325 | if (illegal_highdma(skb->dev, skb)) |
3326 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
3327 | |
3328 | return features; | |
3329 | } | |
3330 | ||
e38f3025 TM |
3331 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
3332 | struct net_device *dev, | |
3333 | netdev_features_t features) | |
3334 | { | |
3335 | return features; | |
3336 | } | |
3337 | EXPORT_SYMBOL(passthru_features_check); | |
3338 | ||
7ce23672 | 3339 | static netdev_features_t dflt_features_check(struct sk_buff *skb, |
8cb65d00 TM |
3340 | struct net_device *dev, |
3341 | netdev_features_t features) | |
3342 | { | |
3343 | return vlan_features_check(skb, features); | |
3344 | } | |
3345 | ||
cbc53e08 AD |
3346 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
3347 | struct net_device *dev, | |
3348 | netdev_features_t features) | |
3349 | { | |
3350 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
3351 | ||
3352 | if (gso_segs > dev->gso_max_segs) | |
3353 | return features & ~NETIF_F_GSO_MASK; | |
3354 | ||
802ab55a AD |
3355 | /* Support for GSO partial features requires software |
3356 | * intervention before we can actually process the packets | |
3357 | * so we need to strip support for any partial features now | |
3358 | * and we can pull them back in after we have partially | |
3359 | * segmented the frame. | |
3360 | */ | |
3361 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
3362 | features &= ~dev->gso_partial_features; | |
3363 | ||
3364 | /* Make sure to clear the IPv4 ID mangling feature if the | |
3365 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
3366 | */ |
3367 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
3368 | struct iphdr *iph = skb->encapsulation ? | |
3369 | inner_ip_hdr(skb) : ip_hdr(skb); | |
3370 | ||
3371 | if (!(iph->frag_off & htons(IP_DF))) | |
3372 | features &= ~NETIF_F_TSO_MANGLEID; | |
3373 | } | |
3374 | ||
3375 | return features; | |
3376 | } | |
3377 | ||
c1e756bf | 3378 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 3379 | { |
5f35227e | 3380 | struct net_device *dev = skb->dev; |
fcbeb976 | 3381 | netdev_features_t features = dev->features; |
58e998c6 | 3382 | |
cbc53e08 AD |
3383 | if (skb_is_gso(skb)) |
3384 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 3385 | |
5f35227e JG |
3386 | /* If encapsulation offload request, verify we are testing |
3387 | * hardware encapsulation features instead of standard | |
3388 | * features for the netdev | |
3389 | */ | |
3390 | if (skb->encapsulation) | |
3391 | features &= dev->hw_enc_features; | |
3392 | ||
f5a7fb88 TM |
3393 | if (skb_vlan_tagged(skb)) |
3394 | features = netdev_intersect_features(features, | |
3395 | dev->vlan_features | | |
3396 | NETIF_F_HW_VLAN_CTAG_TX | | |
3397 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 3398 | |
5f35227e JG |
3399 | if (dev->netdev_ops->ndo_features_check) |
3400 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
3401 | features); | |
8cb65d00 TM |
3402 | else |
3403 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 3404 | |
c1e756bf | 3405 | return harmonize_features(skb, features); |
58e998c6 | 3406 | } |
c1e756bf | 3407 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 3408 | |
2ea25513 | 3409 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 3410 | struct netdev_queue *txq, bool more) |
f6a78bfc | 3411 | { |
2ea25513 DM |
3412 | unsigned int len; |
3413 | int rc; | |
00829823 | 3414 | |
9f9a742d | 3415 | if (dev_nit_active(dev)) |
2ea25513 | 3416 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 3417 | |
2ea25513 DM |
3418 | len = skb->len; |
3419 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 3420 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 3421 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 3422 | |
2ea25513 DM |
3423 | return rc; |
3424 | } | |
7b9c6090 | 3425 | |
8dcda22a DM |
3426 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
3427 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
3428 | { |
3429 | struct sk_buff *skb = first; | |
3430 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 3431 | |
7f2e870f DM |
3432 | while (skb) { |
3433 | struct sk_buff *next = skb->next; | |
fc70fb64 | 3434 | |
a8305bff | 3435 | skb_mark_not_on_list(skb); |
95f6b3dd | 3436 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
3437 | if (unlikely(!dev_xmit_complete(rc))) { |
3438 | skb->next = next; | |
3439 | goto out; | |
3440 | } | |
6afff0ca | 3441 | |
7f2e870f | 3442 | skb = next; |
fe60faa5 | 3443 | if (netif_tx_queue_stopped(txq) && skb) { |
7f2e870f DM |
3444 | rc = NETDEV_TX_BUSY; |
3445 | break; | |
9ccb8975 | 3446 | } |
7f2e870f | 3447 | } |
9ccb8975 | 3448 | |
7f2e870f DM |
3449 | out: |
3450 | *ret = rc; | |
3451 | return skb; | |
3452 | } | |
b40863c6 | 3453 | |
1ff0dc94 ED |
3454 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
3455 | netdev_features_t features) | |
f6a78bfc | 3456 | { |
df8a39de | 3457 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
3458 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
3459 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
3460 | return skb; |
3461 | } | |
f6a78bfc | 3462 | |
43c26a1a DC |
3463 | int skb_csum_hwoffload_help(struct sk_buff *skb, |
3464 | const netdev_features_t features) | |
3465 | { | |
3466 | if (unlikely(skb->csum_not_inet)) | |
3467 | return !!(features & NETIF_F_SCTP_CRC) ? 0 : | |
3468 | skb_crc32c_csum_help(skb); | |
3469 | ||
3470 | return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb); | |
3471 | } | |
3472 | EXPORT_SYMBOL(skb_csum_hwoffload_help); | |
3473 | ||
f53c7239 | 3474 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) |
eae3f88e DM |
3475 | { |
3476 | netdev_features_t features; | |
f6a78bfc | 3477 | |
eae3f88e DM |
3478 | features = netif_skb_features(skb); |
3479 | skb = validate_xmit_vlan(skb, features); | |
3480 | if (unlikely(!skb)) | |
3481 | goto out_null; | |
7b9c6090 | 3482 | |
ebf4e808 IL |
3483 | skb = sk_validate_xmit_skb(skb, dev); |
3484 | if (unlikely(!skb)) | |
3485 | goto out_null; | |
3486 | ||
8b86a61d | 3487 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
3488 | struct sk_buff *segs; |
3489 | ||
3490 | segs = skb_gso_segment(skb, features); | |
cecda693 | 3491 | if (IS_ERR(segs)) { |
af6dabc9 | 3492 | goto out_kfree_skb; |
cecda693 JW |
3493 | } else if (segs) { |
3494 | consume_skb(skb); | |
3495 | skb = segs; | |
f6a78bfc | 3496 | } |
eae3f88e DM |
3497 | } else { |
3498 | if (skb_needs_linearize(skb, features) && | |
3499 | __skb_linearize(skb)) | |
3500 | goto out_kfree_skb; | |
4ec93edb | 3501 | |
eae3f88e DM |
3502 | /* If packet is not checksummed and device does not |
3503 | * support checksumming for this protocol, complete | |
3504 | * checksumming here. | |
3505 | */ | |
3506 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
3507 | if (skb->encapsulation) | |
3508 | skb_set_inner_transport_header(skb, | |
3509 | skb_checksum_start_offset(skb)); | |
3510 | else | |
3511 | skb_set_transport_header(skb, | |
3512 | skb_checksum_start_offset(skb)); | |
43c26a1a | 3513 | if (skb_csum_hwoffload_help(skb, features)) |
eae3f88e | 3514 | goto out_kfree_skb; |
7b9c6090 | 3515 | } |
0c772159 | 3516 | } |
7b9c6090 | 3517 | |
f53c7239 | 3518 | skb = validate_xmit_xfrm(skb, features, again); |
3dca3f38 | 3519 | |
eae3f88e | 3520 | return skb; |
fc70fb64 | 3521 | |
f6a78bfc HX |
3522 | out_kfree_skb: |
3523 | kfree_skb(skb); | |
eae3f88e | 3524 | out_null: |
d21fd63e | 3525 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3526 | return NULL; |
3527 | } | |
6afff0ca | 3528 | |
f53c7239 | 3529 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) |
55a93b3e ED |
3530 | { |
3531 | struct sk_buff *next, *head = NULL, *tail; | |
3532 | ||
bec3cfdc | 3533 | for (; skb != NULL; skb = next) { |
55a93b3e | 3534 | next = skb->next; |
a8305bff | 3535 | skb_mark_not_on_list(skb); |
bec3cfdc ED |
3536 | |
3537 | /* in case skb wont be segmented, point to itself */ | |
3538 | skb->prev = skb; | |
3539 | ||
f53c7239 | 3540 | skb = validate_xmit_skb(skb, dev, again); |
bec3cfdc ED |
3541 | if (!skb) |
3542 | continue; | |
55a93b3e | 3543 | |
bec3cfdc ED |
3544 | if (!head) |
3545 | head = skb; | |
3546 | else | |
3547 | tail->next = skb; | |
3548 | /* If skb was segmented, skb->prev points to | |
3549 | * the last segment. If not, it still contains skb. | |
3550 | */ | |
3551 | tail = skb->prev; | |
55a93b3e ED |
3552 | } |
3553 | return head; | |
f6a78bfc | 3554 | } |
104ba78c | 3555 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3556 | |
1def9238 ED |
3557 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3558 | { | |
3559 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3560 | ||
3561 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3562 | ||
3563 | /* To get more precise estimation of bytes sent on wire, | |
3564 | * we add to pkt_len the headers size of all segments | |
3565 | */ | |
a0dce875 | 3566 | if (shinfo->gso_size && skb_transport_header_was_set(skb)) { |
757b8b1d | 3567 | unsigned int hdr_len; |
15e5a030 | 3568 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3569 | |
757b8b1d ED |
3570 | /* mac layer + network layer */ |
3571 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3572 | ||
3573 | /* + transport layer */ | |
7c68d1a6 ED |
3574 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { |
3575 | const struct tcphdr *th; | |
3576 | struct tcphdr _tcphdr; | |
3577 | ||
3578 | th = skb_header_pointer(skb, skb_transport_offset(skb), | |
3579 | sizeof(_tcphdr), &_tcphdr); | |
3580 | if (likely(th)) | |
3581 | hdr_len += __tcp_hdrlen(th); | |
3582 | } else { | |
3583 | struct udphdr _udphdr; | |
3584 | ||
3585 | if (skb_header_pointer(skb, skb_transport_offset(skb), | |
3586 | sizeof(_udphdr), &_udphdr)) | |
3587 | hdr_len += sizeof(struct udphdr); | |
3588 | } | |
15e5a030 JW |
3589 | |
3590 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3591 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3592 | shinfo->gso_size); | |
3593 | ||
3594 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3595 | } |
3596 | } | |
3597 | ||
bbd8a0d3 KK |
3598 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3599 | struct net_device *dev, | |
3600 | struct netdev_queue *txq) | |
3601 | { | |
3602 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3603 | struct sk_buff *to_free = NULL; |
a2da570d | 3604 | bool contended; |
bbd8a0d3 KK |
3605 | int rc; |
3606 | ||
a2da570d | 3607 | qdisc_calculate_pkt_len(skb, q); |
6b3ba914 JF |
3608 | |
3609 | if (q->flags & TCQ_F_NOLOCK) { | |
90b2be27 | 3610 | if ((q->flags & TCQ_F_CAN_BYPASS) && READ_ONCE(q->empty) && |
d518d2ed PA |
3611 | qdisc_run_begin(q)) { |
3612 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, | |
3613 | &q->state))) { | |
3614 | __qdisc_drop(skb, &to_free); | |
3615 | rc = NET_XMIT_DROP; | |
3616 | goto end_run; | |
3617 | } | |
ba27b4cd PA |
3618 | qdisc_bstats_cpu_update(q, skb); |
3619 | ||
d518d2ed | 3620 | rc = NET_XMIT_SUCCESS; |
ba27b4cd PA |
3621 | if (sch_direct_xmit(skb, q, dev, txq, NULL, true)) |
3622 | __qdisc_run(q); | |
3623 | ||
d518d2ed | 3624 | end_run: |
ba27b4cd | 3625 | qdisc_run_end(q); |
6b3ba914 JF |
3626 | } else { |
3627 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; | |
32f7b44d | 3628 | qdisc_run(q); |
6b3ba914 JF |
3629 | } |
3630 | ||
3631 | if (unlikely(to_free)) | |
3632 | kfree_skb_list(to_free); | |
3633 | return rc; | |
3634 | } | |
3635 | ||
79640a4c ED |
3636 | /* |
3637 | * Heuristic to force contended enqueues to serialize on a | |
3638 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3639 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3640 | * often and dequeue packets faster. |
79640a4c | 3641 | */ |
a2da570d | 3642 | contended = qdisc_is_running(q); |
79640a4c ED |
3643 | if (unlikely(contended)) |
3644 | spin_lock(&q->busylock); | |
3645 | ||
bbd8a0d3 KK |
3646 | spin_lock(root_lock); |
3647 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3648 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3649 | rc = NET_XMIT_DROP; |
3650 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3651 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3652 | /* |
3653 | * This is a work-conserving queue; there are no old skbs | |
3654 | * waiting to be sent out; and the qdisc is not running - | |
3655 | * xmit the skb directly. | |
3656 | */ | |
bfe0d029 | 3657 | |
bfe0d029 ED |
3658 | qdisc_bstats_update(q, skb); |
3659 | ||
55a93b3e | 3660 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3661 | if (unlikely(contended)) { |
3662 | spin_unlock(&q->busylock); | |
3663 | contended = false; | |
3664 | } | |
bbd8a0d3 | 3665 | __qdisc_run(q); |
6c148184 | 3666 | } |
bbd8a0d3 | 3667 | |
6c148184 | 3668 | qdisc_run_end(q); |
bbd8a0d3 KK |
3669 | rc = NET_XMIT_SUCCESS; |
3670 | } else { | |
520ac30f | 3671 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3672 | if (qdisc_run_begin(q)) { |
3673 | if (unlikely(contended)) { | |
3674 | spin_unlock(&q->busylock); | |
3675 | contended = false; | |
3676 | } | |
3677 | __qdisc_run(q); | |
6c148184 | 3678 | qdisc_run_end(q); |
79640a4c | 3679 | } |
bbd8a0d3 KK |
3680 | } |
3681 | spin_unlock(root_lock); | |
520ac30f ED |
3682 | if (unlikely(to_free)) |
3683 | kfree_skb_list(to_free); | |
79640a4c ED |
3684 | if (unlikely(contended)) |
3685 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3686 | return rc; |
3687 | } | |
3688 | ||
86f8515f | 3689 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3690 | static void skb_update_prio(struct sk_buff *skb) |
3691 | { | |
4dcb31d4 ED |
3692 | const struct netprio_map *map; |
3693 | const struct sock *sk; | |
3694 | unsigned int prioidx; | |
5bc1421e | 3695 | |
4dcb31d4 ED |
3696 | if (skb->priority) |
3697 | return; | |
3698 | map = rcu_dereference_bh(skb->dev->priomap); | |
3699 | if (!map) | |
3700 | return; | |
3701 | sk = skb_to_full_sk(skb); | |
3702 | if (!sk) | |
3703 | return; | |
91c68ce2 | 3704 | |
4dcb31d4 ED |
3705 | prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); |
3706 | ||
3707 | if (prioidx < map->priomap_len) | |
3708 | skb->priority = map->priomap[prioidx]; | |
5bc1421e NH |
3709 | } |
3710 | #else | |
3711 | #define skb_update_prio(skb) | |
3712 | #endif | |
3713 | ||
95603e22 MM |
3714 | /** |
3715 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3716 | * @net: network namespace this loopback is happening in |
3717 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3718 | * @skb: buffer to transmit |
3719 | */ | |
0c4b51f0 | 3720 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3721 | { |
3722 | skb_reset_mac_header(skb); | |
3723 | __skb_pull(skb, skb_network_offset(skb)); | |
3724 | skb->pkt_type = PACKET_LOOPBACK; | |
3725 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3726 | WARN_ON(!skb_dst(skb)); | |
3727 | skb_dst_force(skb); | |
3728 | netif_rx_ni(skb); | |
3729 | return 0; | |
3730 | } | |
3731 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3732 | ||
1f211a1b DB |
3733 | #ifdef CONFIG_NET_EGRESS |
3734 | static struct sk_buff * | |
3735 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3736 | { | |
46209401 | 3737 | struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); |
1f211a1b DB |
3738 | struct tcf_result cl_res; |
3739 | ||
46209401 | 3740 | if (!miniq) |
1f211a1b DB |
3741 | return skb; |
3742 | ||
8dc07fdb | 3743 | /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ |
46209401 | 3744 | mini_qdisc_bstats_cpu_update(miniq, skb); |
1f211a1b | 3745 | |
46209401 | 3746 | switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { |
1f211a1b DB |
3747 | case TC_ACT_OK: |
3748 | case TC_ACT_RECLASSIFY: | |
3749 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3750 | break; | |
3751 | case TC_ACT_SHOT: | |
46209401 | 3752 | mini_qdisc_qstats_cpu_drop(miniq); |
1f211a1b | 3753 | *ret = NET_XMIT_DROP; |
7e2c3aea DB |
3754 | kfree_skb(skb); |
3755 | return NULL; | |
1f211a1b DB |
3756 | case TC_ACT_STOLEN: |
3757 | case TC_ACT_QUEUED: | |
e25ea21f | 3758 | case TC_ACT_TRAP: |
1f211a1b | 3759 | *ret = NET_XMIT_SUCCESS; |
7e2c3aea | 3760 | consume_skb(skb); |
1f211a1b DB |
3761 | return NULL; |
3762 | case TC_ACT_REDIRECT: | |
3763 | /* No need to push/pop skb's mac_header here on egress! */ | |
3764 | skb_do_redirect(skb); | |
3765 | *ret = NET_XMIT_SUCCESS; | |
3766 | return NULL; | |
3767 | default: | |
3768 | break; | |
3769 | } | |
3770 | ||
3771 | return skb; | |
3772 | } | |
3773 | #endif /* CONFIG_NET_EGRESS */ | |
3774 | ||
fc9bab24 AN |
3775 | #ifdef CONFIG_XPS |
3776 | static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, | |
3777 | struct xps_dev_maps *dev_maps, unsigned int tci) | |
3778 | { | |
3779 | struct xps_map *map; | |
3780 | int queue_index = -1; | |
3781 | ||
3782 | if (dev->num_tc) { | |
3783 | tci *= dev->num_tc; | |
3784 | tci += netdev_get_prio_tc_map(dev, skb->priority); | |
3785 | } | |
3786 | ||
3787 | map = rcu_dereference(dev_maps->attr_map[tci]); | |
3788 | if (map) { | |
3789 | if (map->len == 1) | |
3790 | queue_index = map->queues[0]; | |
3791 | else | |
3792 | queue_index = map->queues[reciprocal_scale( | |
3793 | skb_get_hash(skb), map->len)]; | |
3794 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3795 | queue_index = -1; | |
3796 | } | |
3797 | return queue_index; | |
3798 | } | |
3799 | #endif | |
3800 | ||
eadec877 AD |
3801 | static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, |
3802 | struct sk_buff *skb) | |
638b2a69 JP |
3803 | { |
3804 | #ifdef CONFIG_XPS | |
3805 | struct xps_dev_maps *dev_maps; | |
fc9bab24 | 3806 | struct sock *sk = skb->sk; |
638b2a69 JP |
3807 | int queue_index = -1; |
3808 | ||
04157469 AN |
3809 | if (!static_key_false(&xps_needed)) |
3810 | return -1; | |
3811 | ||
638b2a69 | 3812 | rcu_read_lock(); |
fc9bab24 AN |
3813 | if (!static_key_false(&xps_rxqs_needed)) |
3814 | goto get_cpus_map; | |
3815 | ||
eadec877 | 3816 | dev_maps = rcu_dereference(sb_dev->xps_rxqs_map); |
638b2a69 | 3817 | if (dev_maps) { |
fc9bab24 | 3818 | int tci = sk_rx_queue_get(sk); |
184c449f | 3819 | |
fc9bab24 AN |
3820 | if (tci >= 0 && tci < dev->num_rx_queues) |
3821 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
3822 | tci); | |
3823 | } | |
184c449f | 3824 | |
fc9bab24 AN |
3825 | get_cpus_map: |
3826 | if (queue_index < 0) { | |
eadec877 | 3827 | dev_maps = rcu_dereference(sb_dev->xps_cpus_map); |
fc9bab24 AN |
3828 | if (dev_maps) { |
3829 | unsigned int tci = skb->sender_cpu - 1; | |
3830 | ||
3831 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
3832 | tci); | |
638b2a69 JP |
3833 | } |
3834 | } | |
3835 | rcu_read_unlock(); | |
3836 | ||
3837 | return queue_index; | |
3838 | #else | |
3839 | return -1; | |
3840 | #endif | |
3841 | } | |
3842 | ||
a4ea8a3d | 3843 | u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, |
a350ecce | 3844 | struct net_device *sb_dev) |
a4ea8a3d AD |
3845 | { |
3846 | return 0; | |
3847 | } | |
3848 | EXPORT_SYMBOL(dev_pick_tx_zero); | |
3849 | ||
3850 | u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, | |
a350ecce | 3851 | struct net_device *sb_dev) |
a4ea8a3d AD |
3852 | { |
3853 | return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; | |
3854 | } | |
3855 | EXPORT_SYMBOL(dev_pick_tx_cpu_id); | |
3856 | ||
b71b5837 PA |
3857 | u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, |
3858 | struct net_device *sb_dev) | |
638b2a69 JP |
3859 | { |
3860 | struct sock *sk = skb->sk; | |
3861 | int queue_index = sk_tx_queue_get(sk); | |
3862 | ||
eadec877 AD |
3863 | sb_dev = sb_dev ? : dev; |
3864 | ||
638b2a69 JP |
3865 | if (queue_index < 0 || skb->ooo_okay || |
3866 | queue_index >= dev->real_num_tx_queues) { | |
eadec877 | 3867 | int new_index = get_xps_queue(dev, sb_dev, skb); |
f4563a75 | 3868 | |
638b2a69 | 3869 | if (new_index < 0) |
eadec877 | 3870 | new_index = skb_tx_hash(dev, sb_dev, skb); |
638b2a69 JP |
3871 | |
3872 | if (queue_index != new_index && sk && | |
004a5d01 | 3873 | sk_fullsock(sk) && |
638b2a69 JP |
3874 | rcu_access_pointer(sk->sk_dst_cache)) |
3875 | sk_tx_queue_set(sk, new_index); | |
3876 | ||
3877 | queue_index = new_index; | |
3878 | } | |
3879 | ||
3880 | return queue_index; | |
3881 | } | |
b71b5837 | 3882 | EXPORT_SYMBOL(netdev_pick_tx); |
638b2a69 | 3883 | |
4bd97d51 PA |
3884 | struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, |
3885 | struct sk_buff *skb, | |
3886 | struct net_device *sb_dev) | |
638b2a69 JP |
3887 | { |
3888 | int queue_index = 0; | |
3889 | ||
3890 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
3891 | u32 sender_cpu = skb->sender_cpu - 1; |
3892 | ||
3893 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
3894 | skb->sender_cpu = raw_smp_processor_id() + 1; |
3895 | #endif | |
3896 | ||
3897 | if (dev->real_num_tx_queues != 1) { | |
3898 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 3899 | |
638b2a69 | 3900 | if (ops->ndo_select_queue) |
a350ecce | 3901 | queue_index = ops->ndo_select_queue(dev, skb, sb_dev); |
638b2a69 | 3902 | else |
4bd97d51 | 3903 | queue_index = netdev_pick_tx(dev, skb, sb_dev); |
638b2a69 | 3904 | |
d584527c | 3905 | queue_index = netdev_cap_txqueue(dev, queue_index); |
638b2a69 JP |
3906 | } |
3907 | ||
3908 | skb_set_queue_mapping(skb, queue_index); | |
3909 | return netdev_get_tx_queue(dev, queue_index); | |
3910 | } | |
3911 | ||
d29f749e | 3912 | /** |
9d08dd3d | 3913 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3914 | * @skb: buffer to transmit |
eadec877 | 3915 | * @sb_dev: suboordinate device used for L2 forwarding offload |
d29f749e DJ |
3916 | * |
3917 | * Queue a buffer for transmission to a network device. The caller must | |
3918 | * have set the device and priority and built the buffer before calling | |
3919 | * this function. The function can be called from an interrupt. | |
3920 | * | |
3921 | * A negative errno code is returned on a failure. A success does not | |
3922 | * guarantee the frame will be transmitted as it may be dropped due | |
3923 | * to congestion or traffic shaping. | |
3924 | * | |
3925 | * ----------------------------------------------------------------------------------- | |
3926 | * I notice this method can also return errors from the queue disciplines, | |
3927 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3928 | * be positive. | |
3929 | * | |
3930 | * Regardless of the return value, the skb is consumed, so it is currently | |
3931 | * difficult to retry a send to this method. (You can bump the ref count | |
3932 | * before sending to hold a reference for retry if you are careful.) | |
3933 | * | |
3934 | * When calling this method, interrupts MUST be enabled. This is because | |
3935 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3936 | * --BLG | |
3937 | */ | |
eadec877 | 3938 | static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) |
1da177e4 LT |
3939 | { |
3940 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3941 | struct netdev_queue *txq; |
1da177e4 LT |
3942 | struct Qdisc *q; |
3943 | int rc = -ENOMEM; | |
f53c7239 | 3944 | bool again = false; |
1da177e4 | 3945 | |
6d1ccff6 ED |
3946 | skb_reset_mac_header(skb); |
3947 | ||
e7fd2885 WB |
3948 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3949 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3950 | ||
4ec93edb YH |
3951 | /* Disable soft irqs for various locks below. Also |
3952 | * stops preemption for RCU. | |
1da177e4 | 3953 | */ |
4ec93edb | 3954 | rcu_read_lock_bh(); |
1da177e4 | 3955 | |
5bc1421e NH |
3956 | skb_update_prio(skb); |
3957 | ||
1f211a1b DB |
3958 | qdisc_pkt_len_init(skb); |
3959 | #ifdef CONFIG_NET_CLS_ACT | |
8dc07fdb | 3960 | skb->tc_at_ingress = 0; |
1f211a1b | 3961 | # ifdef CONFIG_NET_EGRESS |
aabf6772 | 3962 | if (static_branch_unlikely(&egress_needed_key)) { |
1f211a1b DB |
3963 | skb = sch_handle_egress(skb, &rc, dev); |
3964 | if (!skb) | |
3965 | goto out; | |
3966 | } | |
3967 | # endif | |
3968 | #endif | |
02875878 ED |
3969 | /* If device/qdisc don't need skb->dst, release it right now while |
3970 | * its hot in this cpu cache. | |
3971 | */ | |
3972 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
3973 | skb_dst_drop(skb); | |
3974 | else | |
3975 | skb_dst_force(skb); | |
3976 | ||
4bd97d51 | 3977 | txq = netdev_core_pick_tx(dev, skb, sb_dev); |
a898def2 | 3978 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 3979 | |
cf66ba58 | 3980 | trace_net_dev_queue(skb); |
1da177e4 | 3981 | if (q->enqueue) { |
bbd8a0d3 | 3982 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 3983 | goto out; |
1da177e4 LT |
3984 | } |
3985 | ||
3986 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 3987 | * loopback, all the sorts of tunnels... |
1da177e4 | 3988 | |
eb13da1a | 3989 | * Really, it is unlikely that netif_tx_lock protection is necessary |
3990 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
3991 | * counters.) | |
3992 | * However, it is possible, that they rely on protection | |
3993 | * made by us here. | |
1da177e4 | 3994 | |
eb13da1a | 3995 | * Check this and shot the lock. It is not prone from deadlocks. |
3996 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
3997 | */ |
3998 | if (dev->flags & IFF_UP) { | |
3999 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
4000 | ||
c773e847 | 4001 | if (txq->xmit_lock_owner != cpu) { |
97cdcf37 | 4002 | if (dev_xmit_recursion()) |
745e20f1 ED |
4003 | goto recursion_alert; |
4004 | ||
f53c7239 | 4005 | skb = validate_xmit_skb(skb, dev, &again); |
1f59533f | 4006 | if (!skb) |
d21fd63e | 4007 | goto out; |
1f59533f | 4008 | |
c773e847 | 4009 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 4010 | |
73466498 | 4011 | if (!netif_xmit_stopped(txq)) { |
97cdcf37 | 4012 | dev_xmit_recursion_inc(); |
ce93718f | 4013 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
97cdcf37 | 4014 | dev_xmit_recursion_dec(); |
572a9d7b | 4015 | if (dev_xmit_complete(rc)) { |
c773e847 | 4016 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
4017 | goto out; |
4018 | } | |
4019 | } | |
c773e847 | 4020 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
4021 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
4022 | dev->name); | |
1da177e4 LT |
4023 | } else { |
4024 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
4025 | * unfortunately |
4026 | */ | |
4027 | recursion_alert: | |
e87cc472 JP |
4028 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
4029 | dev->name); | |
1da177e4 LT |
4030 | } |
4031 | } | |
4032 | ||
4033 | rc = -ENETDOWN; | |
d4828d85 | 4034 | rcu_read_unlock_bh(); |
1da177e4 | 4035 | |
015f0688 | 4036 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 4037 | kfree_skb_list(skb); |
1da177e4 LT |
4038 | return rc; |
4039 | out: | |
d4828d85 | 4040 | rcu_read_unlock_bh(); |
1da177e4 LT |
4041 | return rc; |
4042 | } | |
f663dd9a | 4043 | |
2b4aa3ce | 4044 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
4045 | { |
4046 | return __dev_queue_xmit(skb, NULL); | |
4047 | } | |
2b4aa3ce | 4048 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 4049 | |
eadec877 | 4050 | int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) |
f663dd9a | 4051 | { |
eadec877 | 4052 | return __dev_queue_xmit(skb, sb_dev); |
f663dd9a JW |
4053 | } |
4054 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
4055 | ||
865b03f2 MK |
4056 | int dev_direct_xmit(struct sk_buff *skb, u16 queue_id) |
4057 | { | |
4058 | struct net_device *dev = skb->dev; | |
4059 | struct sk_buff *orig_skb = skb; | |
4060 | struct netdev_queue *txq; | |
4061 | int ret = NETDEV_TX_BUSY; | |
4062 | bool again = false; | |
4063 | ||
4064 | if (unlikely(!netif_running(dev) || | |
4065 | !netif_carrier_ok(dev))) | |
4066 | goto drop; | |
4067 | ||
4068 | skb = validate_xmit_skb_list(skb, dev, &again); | |
4069 | if (skb != orig_skb) | |
4070 | goto drop; | |
4071 | ||
4072 | skb_set_queue_mapping(skb, queue_id); | |
4073 | txq = skb_get_tx_queue(dev, skb); | |
4074 | ||
4075 | local_bh_disable(); | |
4076 | ||
4077 | HARD_TX_LOCK(dev, txq, smp_processor_id()); | |
4078 | if (!netif_xmit_frozen_or_drv_stopped(txq)) | |
4079 | ret = netdev_start_xmit(skb, dev, txq, false); | |
4080 | HARD_TX_UNLOCK(dev, txq); | |
4081 | ||
4082 | local_bh_enable(); | |
4083 | ||
4084 | if (!dev_xmit_complete(ret)) | |
4085 | kfree_skb(skb); | |
4086 | ||
4087 | return ret; | |
4088 | drop: | |
4089 | atomic_long_inc(&dev->tx_dropped); | |
4090 | kfree_skb_list(skb); | |
4091 | return NET_XMIT_DROP; | |
4092 | } | |
4093 | EXPORT_SYMBOL(dev_direct_xmit); | |
1da177e4 | 4094 | |
eb13da1a | 4095 | /************************************************************************* |
4096 | * Receiver routines | |
4097 | *************************************************************************/ | |
1da177e4 | 4098 | |
6b2bedc3 | 4099 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
4100 | EXPORT_SYMBOL(netdev_max_backlog); |
4101 | ||
3b098e2d | 4102 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 | 4103 | int netdev_budget __read_mostly = 300; |
7acf8a1e | 4104 | unsigned int __read_mostly netdev_budget_usecs = 2000; |
3d48b53f MT |
4105 | int weight_p __read_mostly = 64; /* old backlog weight */ |
4106 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
4107 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
4108 | int dev_rx_weight __read_mostly = 64; | |
4109 | int dev_tx_weight __read_mostly = 64; | |
323ebb61 EC |
4110 | /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ |
4111 | int gro_normal_batch __read_mostly = 8; | |
1da177e4 | 4112 | |
eecfd7c4 ED |
4113 | /* Called with irq disabled */ |
4114 | static inline void ____napi_schedule(struct softnet_data *sd, | |
4115 | struct napi_struct *napi) | |
4116 | { | |
4117 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
4118 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4119 | } | |
4120 | ||
bfb564e7 KK |
4121 | #ifdef CONFIG_RPS |
4122 | ||
4123 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 4124 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 4125 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
4126 | u32 rps_cpu_mask __read_mostly; |
4127 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 4128 | |
dc05360f | 4129 | struct static_key_false rps_needed __read_mostly; |
3df97ba8 | 4130 | EXPORT_SYMBOL(rps_needed); |
dc05360f | 4131 | struct static_key_false rfs_needed __read_mostly; |
13bfff25 | 4132 | EXPORT_SYMBOL(rfs_needed); |
adc9300e | 4133 | |
c445477d BH |
4134 | static struct rps_dev_flow * |
4135 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4136 | struct rps_dev_flow *rflow, u16 next_cpu) | |
4137 | { | |
a31196b0 | 4138 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
4139 | #ifdef CONFIG_RFS_ACCEL |
4140 | struct netdev_rx_queue *rxqueue; | |
4141 | struct rps_dev_flow_table *flow_table; | |
4142 | struct rps_dev_flow *old_rflow; | |
4143 | u32 flow_id; | |
4144 | u16 rxq_index; | |
4145 | int rc; | |
4146 | ||
4147 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
4148 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
4149 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
4150 | goto out; |
4151 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
4152 | if (rxq_index == skb_get_rx_queue(skb)) | |
4153 | goto out; | |
4154 | ||
4155 | rxqueue = dev->_rx + rxq_index; | |
4156 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4157 | if (!flow_table) | |
4158 | goto out; | |
61b905da | 4159 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
4160 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
4161 | rxq_index, flow_id); | |
4162 | if (rc < 0) | |
4163 | goto out; | |
4164 | old_rflow = rflow; | |
4165 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
4166 | rflow->filter = rc; |
4167 | if (old_rflow->filter == rflow->filter) | |
4168 | old_rflow->filter = RPS_NO_FILTER; | |
4169 | out: | |
4170 | #endif | |
4171 | rflow->last_qtail = | |
09994d1b | 4172 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
4173 | } |
4174 | ||
09994d1b | 4175 | rflow->cpu = next_cpu; |
c445477d BH |
4176 | return rflow; |
4177 | } | |
4178 | ||
bfb564e7 KK |
4179 | /* |
4180 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
4181 | * CPU from the RPS map of the receiving queue for a given skb. | |
4182 | * rcu_read_lock must be held on entry. | |
4183 | */ | |
4184 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4185 | struct rps_dev_flow **rflowp) | |
4186 | { | |
567e4b79 ED |
4187 | const struct rps_sock_flow_table *sock_flow_table; |
4188 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 4189 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 4190 | struct rps_map *map; |
bfb564e7 | 4191 | int cpu = -1; |
567e4b79 | 4192 | u32 tcpu; |
61b905da | 4193 | u32 hash; |
bfb564e7 KK |
4194 | |
4195 | if (skb_rx_queue_recorded(skb)) { | |
4196 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 4197 | |
62fe0b40 BH |
4198 | if (unlikely(index >= dev->real_num_rx_queues)) { |
4199 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4200 | "%s received packet on queue %u, but number " | |
4201 | "of RX queues is %u\n", | |
4202 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
4203 | goto done; |
4204 | } | |
567e4b79 ED |
4205 | rxqueue += index; |
4206 | } | |
bfb564e7 | 4207 | |
567e4b79 ED |
4208 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
4209 | ||
4210 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 4211 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 4212 | if (!flow_table && !map) |
bfb564e7 KK |
4213 | goto done; |
4214 | ||
2d47b459 | 4215 | skb_reset_network_header(skb); |
61b905da TH |
4216 | hash = skb_get_hash(skb); |
4217 | if (!hash) | |
bfb564e7 KK |
4218 | goto done; |
4219 | ||
fec5e652 TH |
4220 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
4221 | if (flow_table && sock_flow_table) { | |
fec5e652 | 4222 | struct rps_dev_flow *rflow; |
567e4b79 ED |
4223 | u32 next_cpu; |
4224 | u32 ident; | |
4225 | ||
4226 | /* First check into global flow table if there is a match */ | |
4227 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
4228 | if ((ident ^ hash) & ~rps_cpu_mask) | |
4229 | goto try_rps; | |
fec5e652 | 4230 | |
567e4b79 ED |
4231 | next_cpu = ident & rps_cpu_mask; |
4232 | ||
4233 | /* OK, now we know there is a match, | |
4234 | * we can look at the local (per receive queue) flow table | |
4235 | */ | |
61b905da | 4236 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
4237 | tcpu = rflow->cpu; |
4238 | ||
fec5e652 TH |
4239 | /* |
4240 | * If the desired CPU (where last recvmsg was done) is | |
4241 | * different from current CPU (one in the rx-queue flow | |
4242 | * table entry), switch if one of the following holds: | |
a31196b0 | 4243 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
4244 | * - Current CPU is offline. |
4245 | * - The current CPU's queue tail has advanced beyond the | |
4246 | * last packet that was enqueued using this table entry. | |
4247 | * This guarantees that all previous packets for the flow | |
4248 | * have been dequeued, thus preserving in order delivery. | |
4249 | */ | |
4250 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 4251 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 4252 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
4253 | rflow->last_qtail)) >= 0)) { |
4254 | tcpu = next_cpu; | |
c445477d | 4255 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 4256 | } |
c445477d | 4257 | |
a31196b0 | 4258 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
4259 | *rflowp = rflow; |
4260 | cpu = tcpu; | |
4261 | goto done; | |
4262 | } | |
4263 | } | |
4264 | ||
567e4b79 ED |
4265 | try_rps: |
4266 | ||
0a9627f2 | 4267 | if (map) { |
8fc54f68 | 4268 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
4269 | if (cpu_online(tcpu)) { |
4270 | cpu = tcpu; | |
4271 | goto done; | |
4272 | } | |
4273 | } | |
4274 | ||
4275 | done: | |
0a9627f2 TH |
4276 | return cpu; |
4277 | } | |
4278 | ||
c445477d BH |
4279 | #ifdef CONFIG_RFS_ACCEL |
4280 | ||
4281 | /** | |
4282 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
4283 | * @dev: Device on which the filter was set | |
4284 | * @rxq_index: RX queue index | |
4285 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
4286 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
4287 | * | |
4288 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
4289 | * this function for each installed filter and remove the filters for | |
4290 | * which it returns %true. | |
4291 | */ | |
4292 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
4293 | u32 flow_id, u16 filter_id) | |
4294 | { | |
4295 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
4296 | struct rps_dev_flow_table *flow_table; | |
4297 | struct rps_dev_flow *rflow; | |
4298 | bool expire = true; | |
a31196b0 | 4299 | unsigned int cpu; |
c445477d BH |
4300 | |
4301 | rcu_read_lock(); | |
4302 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4303 | if (flow_table && flow_id <= flow_table->mask) { | |
4304 | rflow = &flow_table->flows[flow_id]; | |
6aa7de05 | 4305 | cpu = READ_ONCE(rflow->cpu); |
a31196b0 | 4306 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
4307 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
4308 | rflow->last_qtail) < | |
4309 | (int)(10 * flow_table->mask))) | |
4310 | expire = false; | |
4311 | } | |
4312 | rcu_read_unlock(); | |
4313 | return expire; | |
4314 | } | |
4315 | EXPORT_SYMBOL(rps_may_expire_flow); | |
4316 | ||
4317 | #endif /* CONFIG_RFS_ACCEL */ | |
4318 | ||
0a9627f2 | 4319 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 4320 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 4321 | { |
e36fa2f7 ED |
4322 | struct softnet_data *sd = data; |
4323 | ||
eecfd7c4 | 4324 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 4325 | sd->received_rps++; |
0a9627f2 | 4326 | } |
e36fa2f7 | 4327 | |
fec5e652 | 4328 | #endif /* CONFIG_RPS */ |
0a9627f2 | 4329 | |
e36fa2f7 ED |
4330 | /* |
4331 | * Check if this softnet_data structure is another cpu one | |
4332 | * If yes, queue it to our IPI list and return 1 | |
4333 | * If no, return 0 | |
4334 | */ | |
4335 | static int rps_ipi_queued(struct softnet_data *sd) | |
4336 | { | |
4337 | #ifdef CONFIG_RPS | |
903ceff7 | 4338 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
4339 | |
4340 | if (sd != mysd) { | |
4341 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
4342 | mysd->rps_ipi_list = sd; | |
4343 | ||
4344 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4345 | return 1; | |
4346 | } | |
4347 | #endif /* CONFIG_RPS */ | |
4348 | return 0; | |
4349 | } | |
4350 | ||
99bbc707 WB |
4351 | #ifdef CONFIG_NET_FLOW_LIMIT |
4352 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
4353 | #endif | |
4354 | ||
4355 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
4356 | { | |
4357 | #ifdef CONFIG_NET_FLOW_LIMIT | |
4358 | struct sd_flow_limit *fl; | |
4359 | struct softnet_data *sd; | |
4360 | unsigned int old_flow, new_flow; | |
4361 | ||
4362 | if (qlen < (netdev_max_backlog >> 1)) | |
4363 | return false; | |
4364 | ||
903ceff7 | 4365 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
4366 | |
4367 | rcu_read_lock(); | |
4368 | fl = rcu_dereference(sd->flow_limit); | |
4369 | if (fl) { | |
3958afa1 | 4370 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
4371 | old_flow = fl->history[fl->history_head]; |
4372 | fl->history[fl->history_head] = new_flow; | |
4373 | ||
4374 | fl->history_head++; | |
4375 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
4376 | ||
4377 | if (likely(fl->buckets[old_flow])) | |
4378 | fl->buckets[old_flow]--; | |
4379 | ||
4380 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
4381 | fl->count++; | |
4382 | rcu_read_unlock(); | |
4383 | return true; | |
4384 | } | |
4385 | } | |
4386 | rcu_read_unlock(); | |
4387 | #endif | |
4388 | return false; | |
4389 | } | |
4390 | ||
0a9627f2 TH |
4391 | /* |
4392 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
4393 | * queue (may be a remote CPU queue). | |
4394 | */ | |
fec5e652 TH |
4395 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
4396 | unsigned int *qtail) | |
0a9627f2 | 4397 | { |
e36fa2f7 | 4398 | struct softnet_data *sd; |
0a9627f2 | 4399 | unsigned long flags; |
99bbc707 | 4400 | unsigned int qlen; |
0a9627f2 | 4401 | |
e36fa2f7 | 4402 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
4403 | |
4404 | local_irq_save(flags); | |
0a9627f2 | 4405 | |
e36fa2f7 | 4406 | rps_lock(sd); |
e9e4dd32 JA |
4407 | if (!netif_running(skb->dev)) |
4408 | goto drop; | |
99bbc707 WB |
4409 | qlen = skb_queue_len(&sd->input_pkt_queue); |
4410 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 4411 | if (qlen) { |
0a9627f2 | 4412 | enqueue: |
e36fa2f7 | 4413 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 4414 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 4415 | rps_unlock(sd); |
152102c7 | 4416 | local_irq_restore(flags); |
0a9627f2 TH |
4417 | return NET_RX_SUCCESS; |
4418 | } | |
4419 | ||
ebda37c2 ED |
4420 | /* Schedule NAPI for backlog device |
4421 | * We can use non atomic operation since we own the queue lock | |
4422 | */ | |
4423 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 4424 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 4425 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
4426 | } |
4427 | goto enqueue; | |
4428 | } | |
4429 | ||
e9e4dd32 | 4430 | drop: |
dee42870 | 4431 | sd->dropped++; |
e36fa2f7 | 4432 | rps_unlock(sd); |
0a9627f2 | 4433 | |
0a9627f2 TH |
4434 | local_irq_restore(flags); |
4435 | ||
caf586e5 | 4436 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
4437 | kfree_skb(skb); |
4438 | return NET_RX_DROP; | |
4439 | } | |
1da177e4 | 4440 | |
e817f856 JDB |
4441 | static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) |
4442 | { | |
4443 | struct net_device *dev = skb->dev; | |
4444 | struct netdev_rx_queue *rxqueue; | |
4445 | ||
4446 | rxqueue = dev->_rx; | |
4447 | ||
4448 | if (skb_rx_queue_recorded(skb)) { | |
4449 | u16 index = skb_get_rx_queue(skb); | |
4450 | ||
4451 | if (unlikely(index >= dev->real_num_rx_queues)) { | |
4452 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4453 | "%s received packet on queue %u, but number " | |
4454 | "of RX queues is %u\n", | |
4455 | dev->name, index, dev->real_num_rx_queues); | |
4456 | ||
4457 | return rxqueue; /* Return first rxqueue */ | |
4458 | } | |
4459 | rxqueue += index; | |
4460 | } | |
4461 | return rxqueue; | |
4462 | } | |
4463 | ||
d4455169 | 4464 | static u32 netif_receive_generic_xdp(struct sk_buff *skb, |
02671e23 | 4465 | struct xdp_buff *xdp, |
d4455169 JF |
4466 | struct bpf_prog *xdp_prog) |
4467 | { | |
e817f856 | 4468 | struct netdev_rx_queue *rxqueue; |
198d83bb | 4469 | void *orig_data, *orig_data_end; |
de8f3a83 | 4470 | u32 metalen, act = XDP_DROP; |
29724956 JDB |
4471 | __be16 orig_eth_type; |
4472 | struct ethhdr *eth; | |
4473 | bool orig_bcast; | |
d4455169 JF |
4474 | int hlen, off; |
4475 | u32 mac_len; | |
4476 | ||
4477 | /* Reinjected packets coming from act_mirred or similar should | |
4478 | * not get XDP generic processing. | |
4479 | */ | |
cd11b164 | 4480 | if (skb_cloned(skb) || skb_is_tc_redirected(skb)) |
d4455169 JF |
4481 | return XDP_PASS; |
4482 | ||
de8f3a83 DB |
4483 | /* XDP packets must be linear and must have sufficient headroom |
4484 | * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also | |
4485 | * native XDP provides, thus we need to do it here as well. | |
4486 | */ | |
4487 | if (skb_is_nonlinear(skb) || | |
4488 | skb_headroom(skb) < XDP_PACKET_HEADROOM) { | |
4489 | int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); | |
4490 | int troom = skb->tail + skb->data_len - skb->end; | |
4491 | ||
4492 | /* In case we have to go down the path and also linearize, | |
4493 | * then lets do the pskb_expand_head() work just once here. | |
4494 | */ | |
4495 | if (pskb_expand_head(skb, | |
4496 | hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, | |
4497 | troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) | |
4498 | goto do_drop; | |
2d17d8d7 | 4499 | if (skb_linearize(skb)) |
de8f3a83 DB |
4500 | goto do_drop; |
4501 | } | |
d4455169 JF |
4502 | |
4503 | /* The XDP program wants to see the packet starting at the MAC | |
4504 | * header. | |
4505 | */ | |
4506 | mac_len = skb->data - skb_mac_header(skb); | |
4507 | hlen = skb_headlen(skb) + mac_len; | |
02671e23 BT |
4508 | xdp->data = skb->data - mac_len; |
4509 | xdp->data_meta = xdp->data; | |
4510 | xdp->data_end = xdp->data + hlen; | |
4511 | xdp->data_hard_start = skb->data - skb_headroom(skb); | |
4512 | orig_data_end = xdp->data_end; | |
4513 | orig_data = xdp->data; | |
29724956 JDB |
4514 | eth = (struct ethhdr *)xdp->data; |
4515 | orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); | |
4516 | orig_eth_type = eth->h_proto; | |
d4455169 | 4517 | |
e817f856 | 4518 | rxqueue = netif_get_rxqueue(skb); |
02671e23 | 4519 | xdp->rxq = &rxqueue->xdp_rxq; |
e817f856 | 4520 | |
02671e23 | 4521 | act = bpf_prog_run_xdp(xdp_prog, xdp); |
d4455169 | 4522 | |
065af355 | 4523 | /* check if bpf_xdp_adjust_head was used */ |
02671e23 | 4524 | off = xdp->data - orig_data; |
065af355 JDB |
4525 | if (off) { |
4526 | if (off > 0) | |
4527 | __skb_pull(skb, off); | |
4528 | else if (off < 0) | |
4529 | __skb_push(skb, -off); | |
4530 | ||
4531 | skb->mac_header += off; | |
4532 | skb_reset_network_header(skb); | |
4533 | } | |
d4455169 | 4534 | |
198d83bb NS |
4535 | /* check if bpf_xdp_adjust_tail was used. it can only "shrink" |
4536 | * pckt. | |
4537 | */ | |
02671e23 | 4538 | off = orig_data_end - xdp->data_end; |
f7613120 | 4539 | if (off != 0) { |
02671e23 | 4540 | skb_set_tail_pointer(skb, xdp->data_end - xdp->data); |
f7613120 | 4541 | skb->len -= off; |
02671e23 | 4542 | |
f7613120 | 4543 | } |
198d83bb | 4544 | |
29724956 JDB |
4545 | /* check if XDP changed eth hdr such SKB needs update */ |
4546 | eth = (struct ethhdr *)xdp->data; | |
4547 | if ((orig_eth_type != eth->h_proto) || | |
4548 | (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { | |
4549 | __skb_push(skb, ETH_HLEN); | |
4550 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4551 | } | |
4552 | ||
d4455169 | 4553 | switch (act) { |
6103aa96 | 4554 | case XDP_REDIRECT: |
d4455169 JF |
4555 | case XDP_TX: |
4556 | __skb_push(skb, mac_len); | |
de8f3a83 | 4557 | break; |
d4455169 | 4558 | case XDP_PASS: |
02671e23 | 4559 | metalen = xdp->data - xdp->data_meta; |
de8f3a83 DB |
4560 | if (metalen) |
4561 | skb_metadata_set(skb, metalen); | |
d4455169 | 4562 | break; |
d4455169 JF |
4563 | default: |
4564 | bpf_warn_invalid_xdp_action(act); | |
4565 | /* fall through */ | |
4566 | case XDP_ABORTED: | |
4567 | trace_xdp_exception(skb->dev, xdp_prog, act); | |
4568 | /* fall through */ | |
4569 | case XDP_DROP: | |
4570 | do_drop: | |
4571 | kfree_skb(skb); | |
4572 | break; | |
4573 | } | |
4574 | ||
4575 | return act; | |
4576 | } | |
4577 | ||
4578 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
4579 | * network taps in order to match in-driver-XDP behavior. | |
4580 | */ | |
7c497478 | 4581 | void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) |
d4455169 JF |
4582 | { |
4583 | struct net_device *dev = skb->dev; | |
4584 | struct netdev_queue *txq; | |
4585 | bool free_skb = true; | |
4586 | int cpu, rc; | |
4587 | ||
4bd97d51 | 4588 | txq = netdev_core_pick_tx(dev, skb, NULL); |
d4455169 JF |
4589 | cpu = smp_processor_id(); |
4590 | HARD_TX_LOCK(dev, txq, cpu); | |
4591 | if (!netif_xmit_stopped(txq)) { | |
4592 | rc = netdev_start_xmit(skb, dev, txq, 0); | |
4593 | if (dev_xmit_complete(rc)) | |
4594 | free_skb = false; | |
4595 | } | |
4596 | HARD_TX_UNLOCK(dev, txq); | |
4597 | if (free_skb) { | |
4598 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
4599 | kfree_skb(skb); | |
4600 | } | |
4601 | } | |
7c497478 | 4602 | EXPORT_SYMBOL_GPL(generic_xdp_tx); |
d4455169 | 4603 | |
02786475 | 4604 | static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); |
d4455169 | 4605 | |
7c497478 | 4606 | int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) |
d4455169 | 4607 | { |
d4455169 | 4608 | if (xdp_prog) { |
02671e23 BT |
4609 | struct xdp_buff xdp; |
4610 | u32 act; | |
6103aa96 | 4611 | int err; |
d4455169 | 4612 | |
02671e23 | 4613 | act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); |
d4455169 | 4614 | if (act != XDP_PASS) { |
6103aa96 JF |
4615 | switch (act) { |
4616 | case XDP_REDIRECT: | |
2facaad6 | 4617 | err = xdp_do_generic_redirect(skb->dev, skb, |
02671e23 | 4618 | &xdp, xdp_prog); |
6103aa96 JF |
4619 | if (err) |
4620 | goto out_redir; | |
02671e23 | 4621 | break; |
6103aa96 | 4622 | case XDP_TX: |
d4455169 | 4623 | generic_xdp_tx(skb, xdp_prog); |
6103aa96 JF |
4624 | break; |
4625 | } | |
d4455169 JF |
4626 | return XDP_DROP; |
4627 | } | |
4628 | } | |
4629 | return XDP_PASS; | |
6103aa96 | 4630 | out_redir: |
6103aa96 JF |
4631 | kfree_skb(skb); |
4632 | return XDP_DROP; | |
d4455169 | 4633 | } |
7c497478 | 4634 | EXPORT_SYMBOL_GPL(do_xdp_generic); |
d4455169 | 4635 | |
ae78dbfa | 4636 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 4637 | { |
b0e28f1e | 4638 | int ret; |
1da177e4 | 4639 | |
588f0330 | 4640 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 4641 | |
cf66ba58 | 4642 | trace_netif_rx(skb); |
d4455169 | 4643 | |
df334545 | 4644 | #ifdef CONFIG_RPS |
dc05360f | 4645 | if (static_branch_unlikely(&rps_needed)) { |
fec5e652 | 4646 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
4647 | int cpu; |
4648 | ||
cece1945 | 4649 | preempt_disable(); |
b0e28f1e | 4650 | rcu_read_lock(); |
fec5e652 TH |
4651 | |
4652 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
4653 | if (cpu < 0) |
4654 | cpu = smp_processor_id(); | |
fec5e652 TH |
4655 | |
4656 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4657 | ||
b0e28f1e | 4658 | rcu_read_unlock(); |
cece1945 | 4659 | preempt_enable(); |
adc9300e ED |
4660 | } else |
4661 | #endif | |
fec5e652 TH |
4662 | { |
4663 | unsigned int qtail; | |
f4563a75 | 4664 | |
fec5e652 TH |
4665 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); |
4666 | put_cpu(); | |
4667 | } | |
b0e28f1e | 4668 | return ret; |
1da177e4 | 4669 | } |
ae78dbfa BH |
4670 | |
4671 | /** | |
4672 | * netif_rx - post buffer to the network code | |
4673 | * @skb: buffer to post | |
4674 | * | |
4675 | * This function receives a packet from a device driver and queues it for | |
4676 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
4677 | * may be dropped during processing for congestion control or by the | |
4678 | * protocol layers. | |
4679 | * | |
4680 | * return values: | |
4681 | * NET_RX_SUCCESS (no congestion) | |
4682 | * NET_RX_DROP (packet was dropped) | |
4683 | * | |
4684 | */ | |
4685 | ||
4686 | int netif_rx(struct sk_buff *skb) | |
4687 | { | |
b0e3f1bd GB |
4688 | int ret; |
4689 | ||
ae78dbfa BH |
4690 | trace_netif_rx_entry(skb); |
4691 | ||
b0e3f1bd GB |
4692 | ret = netif_rx_internal(skb); |
4693 | trace_netif_rx_exit(ret); | |
4694 | ||
4695 | return ret; | |
ae78dbfa | 4696 | } |
d1b19dff | 4697 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
4698 | |
4699 | int netif_rx_ni(struct sk_buff *skb) | |
4700 | { | |
4701 | int err; | |
4702 | ||
ae78dbfa BH |
4703 | trace_netif_rx_ni_entry(skb); |
4704 | ||
1da177e4 | 4705 | preempt_disable(); |
ae78dbfa | 4706 | err = netif_rx_internal(skb); |
1da177e4 LT |
4707 | if (local_softirq_pending()) |
4708 | do_softirq(); | |
4709 | preempt_enable(); | |
b0e3f1bd | 4710 | trace_netif_rx_ni_exit(err); |
1da177e4 LT |
4711 | |
4712 | return err; | |
4713 | } | |
1da177e4 LT |
4714 | EXPORT_SYMBOL(netif_rx_ni); |
4715 | ||
0766f788 | 4716 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 4717 | { |
903ceff7 | 4718 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
4719 | |
4720 | if (sd->completion_queue) { | |
4721 | struct sk_buff *clist; | |
4722 | ||
4723 | local_irq_disable(); | |
4724 | clist = sd->completion_queue; | |
4725 | sd->completion_queue = NULL; | |
4726 | local_irq_enable(); | |
4727 | ||
4728 | while (clist) { | |
4729 | struct sk_buff *skb = clist; | |
f4563a75 | 4730 | |
1da177e4 LT |
4731 | clist = clist->next; |
4732 | ||
63354797 | 4733 | WARN_ON(refcount_read(&skb->users)); |
e6247027 ED |
4734 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
4735 | trace_consume_skb(skb); | |
4736 | else | |
4737 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
4738 | |
4739 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
4740 | __kfree_skb(skb); | |
4741 | else | |
4742 | __kfree_skb_defer(skb); | |
1da177e4 | 4743 | } |
15fad714 JDB |
4744 | |
4745 | __kfree_skb_flush(); | |
1da177e4 LT |
4746 | } |
4747 | ||
4748 | if (sd->output_queue) { | |
37437bb2 | 4749 | struct Qdisc *head; |
1da177e4 LT |
4750 | |
4751 | local_irq_disable(); | |
4752 | head = sd->output_queue; | |
4753 | sd->output_queue = NULL; | |
a9cbd588 | 4754 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
4755 | local_irq_enable(); |
4756 | ||
4757 | while (head) { | |
37437bb2 | 4758 | struct Qdisc *q = head; |
6b3ba914 | 4759 | spinlock_t *root_lock = NULL; |
37437bb2 | 4760 | |
1da177e4 LT |
4761 | head = head->next_sched; |
4762 | ||
6b3ba914 JF |
4763 | if (!(q->flags & TCQ_F_NOLOCK)) { |
4764 | root_lock = qdisc_lock(q); | |
4765 | spin_lock(root_lock); | |
4766 | } | |
3bcb846c ED |
4767 | /* We need to make sure head->next_sched is read |
4768 | * before clearing __QDISC_STATE_SCHED | |
4769 | */ | |
4770 | smp_mb__before_atomic(); | |
4771 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
4772 | qdisc_run(q); | |
6b3ba914 JF |
4773 | if (root_lock) |
4774 | spin_unlock(root_lock); | |
1da177e4 LT |
4775 | } |
4776 | } | |
f53c7239 SK |
4777 | |
4778 | xfrm_dev_backlog(sd); | |
1da177e4 LT |
4779 | } |
4780 | ||
181402a5 | 4781 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
4782 | /* This hook is defined here for ATM LANE */ |
4783 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
4784 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 4785 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 4786 | #endif |
1da177e4 | 4787 | |
1f211a1b DB |
4788 | static inline struct sk_buff * |
4789 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
4790 | struct net_device *orig_dev) | |
f697c3e8 | 4791 | { |
e7582bab | 4792 | #ifdef CONFIG_NET_CLS_ACT |
46209401 | 4793 | struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); |
d2788d34 | 4794 | struct tcf_result cl_res; |
24824a09 | 4795 | |
c9e99fd0 DB |
4796 | /* If there's at least one ingress present somewhere (so |
4797 | * we get here via enabled static key), remaining devices | |
4798 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 4799 | * out here. |
c9e99fd0 | 4800 | */ |
46209401 | 4801 | if (!miniq) |
4577139b | 4802 | return skb; |
46209401 | 4803 | |
f697c3e8 HX |
4804 | if (*pt_prev) { |
4805 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4806 | *pt_prev = NULL; | |
1da177e4 LT |
4807 | } |
4808 | ||
3365495c | 4809 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
8dc07fdb | 4810 | skb->tc_at_ingress = 1; |
46209401 | 4811 | mini_qdisc_bstats_cpu_update(miniq, skb); |
c9e99fd0 | 4812 | |
46209401 | 4813 | switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { |
d2788d34 DB |
4814 | case TC_ACT_OK: |
4815 | case TC_ACT_RECLASSIFY: | |
4816 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
4817 | break; | |
4818 | case TC_ACT_SHOT: | |
46209401 | 4819 | mini_qdisc_qstats_cpu_drop(miniq); |
8a3a4c6e ED |
4820 | kfree_skb(skb); |
4821 | return NULL; | |
d2788d34 DB |
4822 | case TC_ACT_STOLEN: |
4823 | case TC_ACT_QUEUED: | |
e25ea21f | 4824 | case TC_ACT_TRAP: |
8a3a4c6e | 4825 | consume_skb(skb); |
d2788d34 | 4826 | return NULL; |
27b29f63 AS |
4827 | case TC_ACT_REDIRECT: |
4828 | /* skb_mac_header check was done by cls/act_bpf, so | |
4829 | * we can safely push the L2 header back before | |
4830 | * redirecting to another netdev | |
4831 | */ | |
4832 | __skb_push(skb, skb->mac_len); | |
4833 | skb_do_redirect(skb); | |
4834 | return NULL; | |
720f22fe | 4835 | case TC_ACT_CONSUMED: |
cd11b164 | 4836 | return NULL; |
d2788d34 DB |
4837 | default: |
4838 | break; | |
f697c3e8 | 4839 | } |
e7582bab | 4840 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
4841 | return skb; |
4842 | } | |
1da177e4 | 4843 | |
24b27fc4 MB |
4844 | /** |
4845 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
4846 | * @dev: device to check | |
4847 | * | |
4848 | * Check if a receive handler is already registered for a given device. | |
4849 | * Return true if there one. | |
4850 | * | |
4851 | * The caller must hold the rtnl_mutex. | |
4852 | */ | |
4853 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
4854 | { | |
4855 | ASSERT_RTNL(); | |
4856 | return dev && rtnl_dereference(dev->rx_handler); | |
4857 | } | |
4858 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
4859 | ||
ab95bfe0 JP |
4860 | /** |
4861 | * netdev_rx_handler_register - register receive handler | |
4862 | * @dev: device to register a handler for | |
4863 | * @rx_handler: receive handler to register | |
93e2c32b | 4864 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 4865 | * |
e227867f | 4866 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
4867 | * called from __netif_receive_skb. A negative errno code is returned |
4868 | * on a failure. | |
4869 | * | |
4870 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
4871 | * |
4872 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
4873 | */ |
4874 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
4875 | rx_handler_func_t *rx_handler, |
4876 | void *rx_handler_data) | |
ab95bfe0 | 4877 | { |
1b7cd004 | 4878 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
4879 | return -EBUSY; |
4880 | ||
f5426250 PA |
4881 | if (dev->priv_flags & IFF_NO_RX_HANDLER) |
4882 | return -EINVAL; | |
4883 | ||
00cfec37 | 4884 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 4885 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
4886 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
4887 | ||
4888 | return 0; | |
4889 | } | |
4890 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
4891 | ||
4892 | /** | |
4893 | * netdev_rx_handler_unregister - unregister receive handler | |
4894 | * @dev: device to unregister a handler from | |
4895 | * | |
166ec369 | 4896 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
4897 | * |
4898 | * The caller must hold the rtnl_mutex. | |
4899 | */ | |
4900 | void netdev_rx_handler_unregister(struct net_device *dev) | |
4901 | { | |
4902 | ||
4903 | ASSERT_RTNL(); | |
a9b3cd7f | 4904 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
4905 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
4906 | * section has a guarantee to see a non NULL rx_handler_data | |
4907 | * as well. | |
4908 | */ | |
4909 | synchronize_net(); | |
a9b3cd7f | 4910 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
4911 | } |
4912 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
4913 | ||
b4b9e355 MG |
4914 | /* |
4915 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
4916 | * the special handling of PFMEMALLOC skbs. | |
4917 | */ | |
4918 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
4919 | { | |
4920 | switch (skb->protocol) { | |
2b8837ae JP |
4921 | case htons(ETH_P_ARP): |
4922 | case htons(ETH_P_IP): | |
4923 | case htons(ETH_P_IPV6): | |
4924 | case htons(ETH_P_8021Q): | |
4925 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
4926 | return true; |
4927 | default: | |
4928 | return false; | |
4929 | } | |
4930 | } | |
4931 | ||
e687ad60 PN |
4932 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
4933 | int *ret, struct net_device *orig_dev) | |
4934 | { | |
e7582bab | 4935 | #ifdef CONFIG_NETFILTER_INGRESS |
e687ad60 | 4936 | if (nf_hook_ingress_active(skb)) { |
2c1e2703 AC |
4937 | int ingress_retval; |
4938 | ||
e687ad60 PN |
4939 | if (*pt_prev) { |
4940 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4941 | *pt_prev = NULL; | |
4942 | } | |
4943 | ||
2c1e2703 AC |
4944 | rcu_read_lock(); |
4945 | ingress_retval = nf_hook_ingress(skb); | |
4946 | rcu_read_unlock(); | |
4947 | return ingress_retval; | |
e687ad60 | 4948 | } |
e7582bab | 4949 | #endif /* CONFIG_NETFILTER_INGRESS */ |
e687ad60 PN |
4950 | return 0; |
4951 | } | |
e687ad60 | 4952 | |
88eb1944 EC |
4953 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, |
4954 | struct packet_type **ppt_prev) | |
1da177e4 LT |
4955 | { |
4956 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 4957 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 4958 | struct net_device *orig_dev; |
8a4eb573 | 4959 | bool deliver_exact = false; |
1da177e4 | 4960 | int ret = NET_RX_DROP; |
252e3346 | 4961 | __be16 type; |
1da177e4 | 4962 | |
588f0330 | 4963 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 4964 | |
cf66ba58 | 4965 | trace_netif_receive_skb(skb); |
9b22ea56 | 4966 | |
cc9bd5ce | 4967 | orig_dev = skb->dev; |
8f903c70 | 4968 | |
c1d2bbe1 | 4969 | skb_reset_network_header(skb); |
fda55eca ED |
4970 | if (!skb_transport_header_was_set(skb)) |
4971 | skb_reset_transport_header(skb); | |
0b5c9db1 | 4972 | skb_reset_mac_len(skb); |
1da177e4 LT |
4973 | |
4974 | pt_prev = NULL; | |
4975 | ||
63d8ea7f | 4976 | another_round: |
b6858177 | 4977 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
4978 | |
4979 | __this_cpu_inc(softnet_data.processed); | |
4980 | ||
458bf2f2 SH |
4981 | if (static_branch_unlikely(&generic_xdp_needed_key)) { |
4982 | int ret2; | |
4983 | ||
4984 | preempt_disable(); | |
4985 | ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
4986 | preempt_enable(); | |
4987 | ||
4988 | if (ret2 != XDP_PASS) | |
4989 | return NET_RX_DROP; | |
4990 | skb_reset_mac_len(skb); | |
4991 | } | |
4992 | ||
8ad227ff PM |
4993 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
4994 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 4995 | skb = skb_vlan_untag(skb); |
bcc6d479 | 4996 | if (unlikely(!skb)) |
2c17d27c | 4997 | goto out; |
bcc6d479 JP |
4998 | } |
4999 | ||
e7246e12 WB |
5000 | if (skb_skip_tc_classify(skb)) |
5001 | goto skip_classify; | |
1da177e4 | 5002 | |
9754e293 | 5003 | if (pfmemalloc) |
b4b9e355 MG |
5004 | goto skip_taps; |
5005 | ||
1da177e4 | 5006 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
5007 | if (pt_prev) |
5008 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5009 | pt_prev = ptype; | |
5010 | } | |
5011 | ||
5012 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
5013 | if (pt_prev) | |
5014 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5015 | pt_prev = ptype; | |
1da177e4 LT |
5016 | } |
5017 | ||
b4b9e355 | 5018 | skip_taps: |
1cf51900 | 5019 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 5020 | if (static_branch_unlikely(&ingress_needed_key)) { |
1f211a1b | 5021 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); |
4577139b | 5022 | if (!skb) |
2c17d27c | 5023 | goto out; |
e687ad60 PN |
5024 | |
5025 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 5026 | goto out; |
4577139b | 5027 | } |
1cf51900 | 5028 | #endif |
a5135bcf | 5029 | skb_reset_tc(skb); |
e7246e12 | 5030 | skip_classify: |
9754e293 | 5031 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
5032 | goto drop; |
5033 | ||
df8a39de | 5034 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
5035 | if (pt_prev) { |
5036 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5037 | pt_prev = NULL; | |
5038 | } | |
48cc32d3 | 5039 | if (vlan_do_receive(&skb)) |
2425717b JF |
5040 | goto another_round; |
5041 | else if (unlikely(!skb)) | |
2c17d27c | 5042 | goto out; |
2425717b JF |
5043 | } |
5044 | ||
48cc32d3 | 5045 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
5046 | if (rx_handler) { |
5047 | if (pt_prev) { | |
5048 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5049 | pt_prev = NULL; | |
5050 | } | |
8a4eb573 JP |
5051 | switch (rx_handler(&skb)) { |
5052 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 5053 | ret = NET_RX_SUCCESS; |
2c17d27c | 5054 | goto out; |
8a4eb573 | 5055 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 5056 | goto another_round; |
8a4eb573 JP |
5057 | case RX_HANDLER_EXACT: |
5058 | deliver_exact = true; | |
5059 | case RX_HANDLER_PASS: | |
5060 | break; | |
5061 | default: | |
5062 | BUG(); | |
5063 | } | |
ab95bfe0 | 5064 | } |
1da177e4 | 5065 | |
df8a39de | 5066 | if (unlikely(skb_vlan_tag_present(skb))) { |
36b2f61a GV |
5067 | check_vlan_id: |
5068 | if (skb_vlan_tag_get_id(skb)) { | |
5069 | /* Vlan id is non 0 and vlan_do_receive() above couldn't | |
5070 | * find vlan device. | |
5071 | */ | |
d4b812de | 5072 | skb->pkt_type = PACKET_OTHERHOST; |
36b2f61a GV |
5073 | } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
5074 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
5075 | /* Outer header is 802.1P with vlan 0, inner header is | |
5076 | * 802.1Q or 802.1AD and vlan_do_receive() above could | |
5077 | * not find vlan dev for vlan id 0. | |
5078 | */ | |
5079 | __vlan_hwaccel_clear_tag(skb); | |
5080 | skb = skb_vlan_untag(skb); | |
5081 | if (unlikely(!skb)) | |
5082 | goto out; | |
5083 | if (vlan_do_receive(&skb)) | |
5084 | /* After stripping off 802.1P header with vlan 0 | |
5085 | * vlan dev is found for inner header. | |
5086 | */ | |
5087 | goto another_round; | |
5088 | else if (unlikely(!skb)) | |
5089 | goto out; | |
5090 | else | |
5091 | /* We have stripped outer 802.1P vlan 0 header. | |
5092 | * But could not find vlan dev. | |
5093 | * check again for vlan id to set OTHERHOST. | |
5094 | */ | |
5095 | goto check_vlan_id; | |
5096 | } | |
d4b812de ED |
5097 | /* Note: we might in the future use prio bits |
5098 | * and set skb->priority like in vlan_do_receive() | |
5099 | * For the time being, just ignore Priority Code Point | |
5100 | */ | |
b1817524 | 5101 | __vlan_hwaccel_clear_tag(skb); |
d4b812de | 5102 | } |
48cc32d3 | 5103 | |
7866a621 SN |
5104 | type = skb->protocol; |
5105 | ||
63d8ea7f | 5106 | /* deliver only exact match when indicated */ |
7866a621 SN |
5107 | if (likely(!deliver_exact)) { |
5108 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
5109 | &ptype_base[ntohs(type) & | |
5110 | PTYPE_HASH_MASK]); | |
5111 | } | |
1f3c8804 | 5112 | |
7866a621 SN |
5113 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
5114 | &orig_dev->ptype_specific); | |
5115 | ||
5116 | if (unlikely(skb->dev != orig_dev)) { | |
5117 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
5118 | &skb->dev->ptype_specific); | |
1da177e4 LT |
5119 | } |
5120 | ||
5121 | if (pt_prev) { | |
1f8b977a | 5122 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
0e698bf6 | 5123 | goto drop; |
88eb1944 | 5124 | *ppt_prev = pt_prev; |
1da177e4 | 5125 | } else { |
b4b9e355 | 5126 | drop: |
6e7333d3 JW |
5127 | if (!deliver_exact) |
5128 | atomic_long_inc(&skb->dev->rx_dropped); | |
5129 | else | |
5130 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
5131 | kfree_skb(skb); |
5132 | /* Jamal, now you will not able to escape explaining | |
5133 | * me how you were going to use this. :-) | |
5134 | */ | |
5135 | ret = NET_RX_DROP; | |
5136 | } | |
5137 | ||
2c17d27c | 5138 | out: |
9754e293 DM |
5139 | return ret; |
5140 | } | |
5141 | ||
88eb1944 EC |
5142 | static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) |
5143 | { | |
5144 | struct net_device *orig_dev = skb->dev; | |
5145 | struct packet_type *pt_prev = NULL; | |
5146 | int ret; | |
5147 | ||
5148 | ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev); | |
5149 | if (pt_prev) | |
f5737cba PA |
5150 | ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, |
5151 | skb->dev, pt_prev, orig_dev); | |
88eb1944 EC |
5152 | return ret; |
5153 | } | |
5154 | ||
1c601d82 JDB |
5155 | /** |
5156 | * netif_receive_skb_core - special purpose version of netif_receive_skb | |
5157 | * @skb: buffer to process | |
5158 | * | |
5159 | * More direct receive version of netif_receive_skb(). It should | |
5160 | * only be used by callers that have a need to skip RPS and Generic XDP. | |
5161 | * Caller must also take care of handling if (page_is_)pfmemalloc. | |
5162 | * | |
5163 | * This function may only be called from softirq context and interrupts | |
5164 | * should be enabled. | |
5165 | * | |
5166 | * Return values (usually ignored): | |
5167 | * NET_RX_SUCCESS: no congestion | |
5168 | * NET_RX_DROP: packet was dropped | |
5169 | */ | |
5170 | int netif_receive_skb_core(struct sk_buff *skb) | |
5171 | { | |
5172 | int ret; | |
5173 | ||
5174 | rcu_read_lock(); | |
88eb1944 | 5175 | ret = __netif_receive_skb_one_core(skb, false); |
1c601d82 JDB |
5176 | rcu_read_unlock(); |
5177 | ||
5178 | return ret; | |
5179 | } | |
5180 | EXPORT_SYMBOL(netif_receive_skb_core); | |
5181 | ||
88eb1944 EC |
5182 | static inline void __netif_receive_skb_list_ptype(struct list_head *head, |
5183 | struct packet_type *pt_prev, | |
5184 | struct net_device *orig_dev) | |
4ce0017a EC |
5185 | { |
5186 | struct sk_buff *skb, *next; | |
5187 | ||
88eb1944 EC |
5188 | if (!pt_prev) |
5189 | return; | |
5190 | if (list_empty(head)) | |
5191 | return; | |
17266ee9 | 5192 | if (pt_prev->list_func != NULL) |
fdf71426 PA |
5193 | INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, |
5194 | ip_list_rcv, head, pt_prev, orig_dev); | |
17266ee9 | 5195 | else |
9a5a90d1 AL |
5196 | list_for_each_entry_safe(skb, next, head, list) { |
5197 | skb_list_del_init(skb); | |
fdf71426 | 5198 | pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
9a5a90d1 | 5199 | } |
88eb1944 EC |
5200 | } |
5201 | ||
5202 | static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) | |
5203 | { | |
5204 | /* Fast-path assumptions: | |
5205 | * - There is no RX handler. | |
5206 | * - Only one packet_type matches. | |
5207 | * If either of these fails, we will end up doing some per-packet | |
5208 | * processing in-line, then handling the 'last ptype' for the whole | |
5209 | * sublist. This can't cause out-of-order delivery to any single ptype, | |
5210 | * because the 'last ptype' must be constant across the sublist, and all | |
5211 | * other ptypes are handled per-packet. | |
5212 | */ | |
5213 | /* Current (common) ptype of sublist */ | |
5214 | struct packet_type *pt_curr = NULL; | |
5215 | /* Current (common) orig_dev of sublist */ | |
5216 | struct net_device *od_curr = NULL; | |
5217 | struct list_head sublist; | |
5218 | struct sk_buff *skb, *next; | |
5219 | ||
9af86f93 | 5220 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5221 | list_for_each_entry_safe(skb, next, head, list) { |
5222 | struct net_device *orig_dev = skb->dev; | |
5223 | struct packet_type *pt_prev = NULL; | |
5224 | ||
22f6bbb7 | 5225 | skb_list_del_init(skb); |
88eb1944 | 5226 | __netif_receive_skb_core(skb, pfmemalloc, &pt_prev); |
9af86f93 EC |
5227 | if (!pt_prev) |
5228 | continue; | |
88eb1944 EC |
5229 | if (pt_curr != pt_prev || od_curr != orig_dev) { |
5230 | /* dispatch old sublist */ | |
88eb1944 EC |
5231 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
5232 | /* start new sublist */ | |
9af86f93 | 5233 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5234 | pt_curr = pt_prev; |
5235 | od_curr = orig_dev; | |
5236 | } | |
9af86f93 | 5237 | list_add_tail(&skb->list, &sublist); |
88eb1944 EC |
5238 | } |
5239 | ||
5240 | /* dispatch final sublist */ | |
9af86f93 | 5241 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
4ce0017a EC |
5242 | } |
5243 | ||
9754e293 DM |
5244 | static int __netif_receive_skb(struct sk_buff *skb) |
5245 | { | |
5246 | int ret; | |
5247 | ||
5248 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 5249 | unsigned int noreclaim_flag; |
9754e293 DM |
5250 | |
5251 | /* | |
5252 | * PFMEMALLOC skbs are special, they should | |
5253 | * - be delivered to SOCK_MEMALLOC sockets only | |
5254 | * - stay away from userspace | |
5255 | * - have bounded memory usage | |
5256 | * | |
5257 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
5258 | * context down to all allocation sites. | |
5259 | */ | |
f1083048 | 5260 | noreclaim_flag = memalloc_noreclaim_save(); |
88eb1944 | 5261 | ret = __netif_receive_skb_one_core(skb, true); |
f1083048 | 5262 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 | 5263 | } else |
88eb1944 | 5264 | ret = __netif_receive_skb_one_core(skb, false); |
9754e293 | 5265 | |
1da177e4 LT |
5266 | return ret; |
5267 | } | |
0a9627f2 | 5268 | |
4ce0017a EC |
5269 | static void __netif_receive_skb_list(struct list_head *head) |
5270 | { | |
5271 | unsigned long noreclaim_flag = 0; | |
5272 | struct sk_buff *skb, *next; | |
5273 | bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ | |
5274 | ||
5275 | list_for_each_entry_safe(skb, next, head, list) { | |
5276 | if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { | |
5277 | struct list_head sublist; | |
5278 | ||
5279 | /* Handle the previous sublist */ | |
5280 | list_cut_before(&sublist, head, &skb->list); | |
b9f463d6 EC |
5281 | if (!list_empty(&sublist)) |
5282 | __netif_receive_skb_list_core(&sublist, pfmemalloc); | |
4ce0017a EC |
5283 | pfmemalloc = !pfmemalloc; |
5284 | /* See comments in __netif_receive_skb */ | |
5285 | if (pfmemalloc) | |
5286 | noreclaim_flag = memalloc_noreclaim_save(); | |
5287 | else | |
5288 | memalloc_noreclaim_restore(noreclaim_flag); | |
5289 | } | |
5290 | } | |
5291 | /* Handle the remaining sublist */ | |
b9f463d6 EC |
5292 | if (!list_empty(head)) |
5293 | __netif_receive_skb_list_core(head, pfmemalloc); | |
4ce0017a EC |
5294 | /* Restore pflags */ |
5295 | if (pfmemalloc) | |
5296 | memalloc_noreclaim_restore(noreclaim_flag); | |
5297 | } | |
5298 | ||
f4e63525 | 5299 | static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) |
b5cdae32 | 5300 | { |
58038695 | 5301 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); |
b5cdae32 DM |
5302 | struct bpf_prog *new = xdp->prog; |
5303 | int ret = 0; | |
5304 | ||
5305 | switch (xdp->command) { | |
58038695 | 5306 | case XDP_SETUP_PROG: |
b5cdae32 DM |
5307 | rcu_assign_pointer(dev->xdp_prog, new); |
5308 | if (old) | |
5309 | bpf_prog_put(old); | |
5310 | ||
5311 | if (old && !new) { | |
02786475 | 5312 | static_branch_dec(&generic_xdp_needed_key); |
b5cdae32 | 5313 | } else if (new && !old) { |
02786475 | 5314 | static_branch_inc(&generic_xdp_needed_key); |
b5cdae32 | 5315 | dev_disable_lro(dev); |
56f5aa77 | 5316 | dev_disable_gro_hw(dev); |
b5cdae32 DM |
5317 | } |
5318 | break; | |
b5cdae32 DM |
5319 | |
5320 | case XDP_QUERY_PROG: | |
58038695 | 5321 | xdp->prog_id = old ? old->aux->id : 0; |
b5cdae32 DM |
5322 | break; |
5323 | ||
5324 | default: | |
5325 | ret = -EINVAL; | |
5326 | break; | |
5327 | } | |
5328 | ||
5329 | return ret; | |
5330 | } | |
5331 | ||
ae78dbfa | 5332 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 5333 | { |
2c17d27c JA |
5334 | int ret; |
5335 | ||
588f0330 | 5336 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 5337 | |
c1f19b51 RC |
5338 | if (skb_defer_rx_timestamp(skb)) |
5339 | return NET_RX_SUCCESS; | |
5340 | ||
bbbe211c | 5341 | rcu_read_lock(); |
df334545 | 5342 | #ifdef CONFIG_RPS |
dc05360f | 5343 | if (static_branch_unlikely(&rps_needed)) { |
3b098e2d | 5344 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 5345 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 5346 | |
3b098e2d ED |
5347 | if (cpu >= 0) { |
5348 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
5349 | rcu_read_unlock(); | |
adc9300e | 5350 | return ret; |
3b098e2d | 5351 | } |
fec5e652 | 5352 | } |
1e94d72f | 5353 | #endif |
2c17d27c JA |
5354 | ret = __netif_receive_skb(skb); |
5355 | rcu_read_unlock(); | |
5356 | return ret; | |
0a9627f2 | 5357 | } |
ae78dbfa | 5358 | |
7da517a3 EC |
5359 | static void netif_receive_skb_list_internal(struct list_head *head) |
5360 | { | |
7da517a3 | 5361 | struct sk_buff *skb, *next; |
8c057efa | 5362 | struct list_head sublist; |
7da517a3 | 5363 | |
8c057efa | 5364 | INIT_LIST_HEAD(&sublist); |
7da517a3 EC |
5365 | list_for_each_entry_safe(skb, next, head, list) { |
5366 | net_timestamp_check(netdev_tstamp_prequeue, skb); | |
22f6bbb7 | 5367 | skb_list_del_init(skb); |
8c057efa EC |
5368 | if (!skb_defer_rx_timestamp(skb)) |
5369 | list_add_tail(&skb->list, &sublist); | |
7da517a3 | 5370 | } |
8c057efa | 5371 | list_splice_init(&sublist, head); |
7da517a3 | 5372 | |
7da517a3 EC |
5373 | rcu_read_lock(); |
5374 | #ifdef CONFIG_RPS | |
dc05360f | 5375 | if (static_branch_unlikely(&rps_needed)) { |
7da517a3 EC |
5376 | list_for_each_entry_safe(skb, next, head, list) { |
5377 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
5378 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
5379 | ||
5380 | if (cpu >= 0) { | |
8c057efa | 5381 | /* Will be handled, remove from list */ |
22f6bbb7 | 5382 | skb_list_del_init(skb); |
8c057efa | 5383 | enqueue_to_backlog(skb, cpu, &rflow->last_qtail); |
7da517a3 EC |
5384 | } |
5385 | } | |
5386 | } | |
5387 | #endif | |
5388 | __netif_receive_skb_list(head); | |
5389 | rcu_read_unlock(); | |
5390 | } | |
5391 | ||
ae78dbfa BH |
5392 | /** |
5393 | * netif_receive_skb - process receive buffer from network | |
5394 | * @skb: buffer to process | |
5395 | * | |
5396 | * netif_receive_skb() is the main receive data processing function. | |
5397 | * It always succeeds. The buffer may be dropped during processing | |
5398 | * for congestion control or by the protocol layers. | |
5399 | * | |
5400 | * This function may only be called from softirq context and interrupts | |
5401 | * should be enabled. | |
5402 | * | |
5403 | * Return values (usually ignored): | |
5404 | * NET_RX_SUCCESS: no congestion | |
5405 | * NET_RX_DROP: packet was dropped | |
5406 | */ | |
04eb4489 | 5407 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa | 5408 | { |
b0e3f1bd GB |
5409 | int ret; |
5410 | ||
ae78dbfa BH |
5411 | trace_netif_receive_skb_entry(skb); |
5412 | ||
b0e3f1bd GB |
5413 | ret = netif_receive_skb_internal(skb); |
5414 | trace_netif_receive_skb_exit(ret); | |
5415 | ||
5416 | return ret; | |
ae78dbfa | 5417 | } |
04eb4489 | 5418 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 5419 | |
f6ad8c1b EC |
5420 | /** |
5421 | * netif_receive_skb_list - process many receive buffers from network | |
5422 | * @head: list of skbs to process. | |
5423 | * | |
7da517a3 EC |
5424 | * Since return value of netif_receive_skb() is normally ignored, and |
5425 | * wouldn't be meaningful for a list, this function returns void. | |
f6ad8c1b EC |
5426 | * |
5427 | * This function may only be called from softirq context and interrupts | |
5428 | * should be enabled. | |
5429 | */ | |
5430 | void netif_receive_skb_list(struct list_head *head) | |
5431 | { | |
7da517a3 | 5432 | struct sk_buff *skb; |
f6ad8c1b | 5433 | |
b9f463d6 EC |
5434 | if (list_empty(head)) |
5435 | return; | |
b0e3f1bd GB |
5436 | if (trace_netif_receive_skb_list_entry_enabled()) { |
5437 | list_for_each_entry(skb, head, list) | |
5438 | trace_netif_receive_skb_list_entry(skb); | |
5439 | } | |
7da517a3 | 5440 | netif_receive_skb_list_internal(head); |
b0e3f1bd | 5441 | trace_netif_receive_skb_list_exit(0); |
f6ad8c1b EC |
5442 | } |
5443 | EXPORT_SYMBOL(netif_receive_skb_list); | |
5444 | ||
41852497 | 5445 | DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
5446 | |
5447 | /* Network device is going away, flush any packets still pending */ | |
5448 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 5449 | { |
6e583ce5 | 5450 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
5451 | struct softnet_data *sd; |
5452 | ||
5453 | local_bh_disable(); | |
5454 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 5455 | |
145dd5f9 | 5456 | local_irq_disable(); |
e36fa2f7 | 5457 | rps_lock(sd); |
6e7676c1 | 5458 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 5459 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 5460 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 5461 | kfree_skb(skb); |
76cc8b13 | 5462 | input_queue_head_incr(sd); |
6e583ce5 | 5463 | } |
6e7676c1 | 5464 | } |
e36fa2f7 | 5465 | rps_unlock(sd); |
145dd5f9 | 5466 | local_irq_enable(); |
6e7676c1 CG |
5467 | |
5468 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 5469 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
5470 | __skb_unlink(skb, &sd->process_queue); |
5471 | kfree_skb(skb); | |
76cc8b13 | 5472 | input_queue_head_incr(sd); |
6e7676c1 CG |
5473 | } |
5474 | } | |
145dd5f9 PA |
5475 | local_bh_enable(); |
5476 | } | |
5477 | ||
41852497 | 5478 | static void flush_all_backlogs(void) |
145dd5f9 PA |
5479 | { |
5480 | unsigned int cpu; | |
5481 | ||
5482 | get_online_cpus(); | |
5483 | ||
41852497 ED |
5484 | for_each_online_cpu(cpu) |
5485 | queue_work_on(cpu, system_highpri_wq, | |
5486 | per_cpu_ptr(&flush_works, cpu)); | |
145dd5f9 PA |
5487 | |
5488 | for_each_online_cpu(cpu) | |
41852497 | 5489 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 PA |
5490 | |
5491 | put_online_cpus(); | |
6e583ce5 SH |
5492 | } |
5493 | ||
aaa5d90b PA |
5494 | INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int)); |
5495 | INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int)); | |
d565b0a1 HX |
5496 | static int napi_gro_complete(struct sk_buff *skb) |
5497 | { | |
22061d80 | 5498 | struct packet_offload *ptype; |
d565b0a1 | 5499 | __be16 type = skb->protocol; |
22061d80 | 5500 | struct list_head *head = &offload_base; |
d565b0a1 HX |
5501 | int err = -ENOENT; |
5502 | ||
c3c7c254 ED |
5503 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
5504 | ||
fc59f9a3 HX |
5505 | if (NAPI_GRO_CB(skb)->count == 1) { |
5506 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 5507 | goto out; |
fc59f9a3 | 5508 | } |
d565b0a1 HX |
5509 | |
5510 | rcu_read_lock(); | |
5511 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 5512 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
5513 | continue; |
5514 | ||
aaa5d90b PA |
5515 | err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, |
5516 | ipv6_gro_complete, inet_gro_complete, | |
5517 | skb, 0); | |
d565b0a1 HX |
5518 | break; |
5519 | } | |
5520 | rcu_read_unlock(); | |
5521 | ||
5522 | if (err) { | |
5523 | WARN_ON(&ptype->list == head); | |
5524 | kfree_skb(skb); | |
5525 | return NET_RX_SUCCESS; | |
5526 | } | |
5527 | ||
5528 | out: | |
ae78dbfa | 5529 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
5530 | } |
5531 | ||
6312fe77 | 5532 | static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, |
07d78363 | 5533 | bool flush_old) |
d565b0a1 | 5534 | { |
6312fe77 | 5535 | struct list_head *head = &napi->gro_hash[index].list; |
d4546c25 | 5536 | struct sk_buff *skb, *p; |
2e71a6f8 | 5537 | |
07d78363 | 5538 | list_for_each_entry_safe_reverse(skb, p, head, list) { |
2e71a6f8 ED |
5539 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) |
5540 | return; | |
992cba7e | 5541 | skb_list_del_init(skb); |
d565b0a1 | 5542 | napi_gro_complete(skb); |
6312fe77 | 5543 | napi->gro_hash[index].count--; |
d565b0a1 | 5544 | } |
d9f37d01 LR |
5545 | |
5546 | if (!napi->gro_hash[index].count) | |
5547 | __clear_bit(index, &napi->gro_bitmask); | |
d565b0a1 | 5548 | } |
07d78363 | 5549 | |
6312fe77 | 5550 | /* napi->gro_hash[].list contains packets ordered by age. |
07d78363 DM |
5551 | * youngest packets at the head of it. |
5552 | * Complete skbs in reverse order to reduce latencies. | |
5553 | */ | |
5554 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
5555 | { | |
42519ede ED |
5556 | unsigned long bitmask = napi->gro_bitmask; |
5557 | unsigned int i, base = ~0U; | |
07d78363 | 5558 | |
42519ede ED |
5559 | while ((i = ffs(bitmask)) != 0) { |
5560 | bitmask >>= i; | |
5561 | base += i; | |
5562 | __napi_gro_flush_chain(napi, base, flush_old); | |
d9f37d01 | 5563 | } |
07d78363 | 5564 | } |
86cac58b | 5565 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 5566 | |
07d78363 DM |
5567 | static struct list_head *gro_list_prepare(struct napi_struct *napi, |
5568 | struct sk_buff *skb) | |
89c5fa33 | 5569 | { |
89c5fa33 | 5570 | unsigned int maclen = skb->dev->hard_header_len; |
0b4cec8c | 5571 | u32 hash = skb_get_hash_raw(skb); |
07d78363 | 5572 | struct list_head *head; |
d4546c25 | 5573 | struct sk_buff *p; |
89c5fa33 | 5574 | |
6312fe77 | 5575 | head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list; |
07d78363 | 5576 | list_for_each_entry(p, head, list) { |
89c5fa33 ED |
5577 | unsigned long diffs; |
5578 | ||
0b4cec8c TH |
5579 | NAPI_GRO_CB(p)->flush = 0; |
5580 | ||
5581 | if (hash != skb_get_hash_raw(p)) { | |
5582 | NAPI_GRO_CB(p)->same_flow = 0; | |
5583 | continue; | |
5584 | } | |
5585 | ||
89c5fa33 | 5586 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
b1817524 MM |
5587 | diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); |
5588 | if (skb_vlan_tag_present(p)) | |
fc5141cb | 5589 | diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); |
ce87fc6c | 5590 | diffs |= skb_metadata_dst_cmp(p, skb); |
de8f3a83 | 5591 | diffs |= skb_metadata_differs(p, skb); |
89c5fa33 ED |
5592 | if (maclen == ETH_HLEN) |
5593 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 5594 | skb_mac_header(skb)); |
89c5fa33 ED |
5595 | else if (!diffs) |
5596 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 5597 | skb_mac_header(skb), |
89c5fa33 ED |
5598 | maclen); |
5599 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 | 5600 | } |
07d78363 DM |
5601 | |
5602 | return head; | |
89c5fa33 ED |
5603 | } |
5604 | ||
299603e8 JC |
5605 | static void skb_gro_reset_offset(struct sk_buff *skb) |
5606 | { | |
5607 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
5608 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
5609 | ||
5610 | NAPI_GRO_CB(skb)->data_offset = 0; | |
5611 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
5612 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
5613 | ||
8aef998d | 5614 | if (!skb_headlen(skb) && pinfo->nr_frags && |
299603e8 JC |
5615 | !PageHighMem(skb_frag_page(frag0))) { |
5616 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
7cfd5fd5 ED |
5617 | NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, |
5618 | skb_frag_size(frag0), | |
5619 | skb->end - skb->tail); | |
89c5fa33 ED |
5620 | } |
5621 | } | |
5622 | ||
a50e233c ED |
5623 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
5624 | { | |
5625 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
5626 | ||
5627 | BUG_ON(skb->end - skb->tail < grow); | |
5628 | ||
5629 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
5630 | ||
5631 | skb->data_len -= grow; | |
5632 | skb->tail += grow; | |
5633 | ||
b54c9d5b | 5634 | skb_frag_off_add(&pinfo->frags[0], grow); |
a50e233c ED |
5635 | skb_frag_size_sub(&pinfo->frags[0], grow); |
5636 | ||
5637 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
5638 | skb_frag_unref(skb, 0); | |
5639 | memmove(pinfo->frags, pinfo->frags + 1, | |
5640 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
5641 | } | |
5642 | } | |
5643 | ||
6312fe77 | 5644 | static void gro_flush_oldest(struct list_head *head) |
07d78363 | 5645 | { |
6312fe77 | 5646 | struct sk_buff *oldest; |
07d78363 | 5647 | |
6312fe77 | 5648 | oldest = list_last_entry(head, struct sk_buff, list); |
07d78363 | 5649 | |
6312fe77 | 5650 | /* We are called with head length >= MAX_GRO_SKBS, so this is |
07d78363 DM |
5651 | * impossible. |
5652 | */ | |
5653 | if (WARN_ON_ONCE(!oldest)) | |
5654 | return; | |
5655 | ||
d9f37d01 LR |
5656 | /* Do not adjust napi->gro_hash[].count, caller is adding a new |
5657 | * SKB to the chain. | |
07d78363 | 5658 | */ |
ece23711 | 5659 | skb_list_del_init(oldest); |
07d78363 DM |
5660 | napi_gro_complete(oldest); |
5661 | } | |
5662 | ||
aaa5d90b PA |
5663 | INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *, |
5664 | struct sk_buff *)); | |
5665 | INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *, | |
5666 | struct sk_buff *)); | |
bb728820 | 5667 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 | 5668 | { |
6312fe77 | 5669 | u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); |
d4546c25 | 5670 | struct list_head *head = &offload_base; |
22061d80 | 5671 | struct packet_offload *ptype; |
d565b0a1 | 5672 | __be16 type = skb->protocol; |
07d78363 | 5673 | struct list_head *gro_head; |
d4546c25 | 5674 | struct sk_buff *pp = NULL; |
5b252f0c | 5675 | enum gro_result ret; |
d4546c25 | 5676 | int same_flow; |
a50e233c | 5677 | int grow; |
d565b0a1 | 5678 | |
b5cdae32 | 5679 | if (netif_elide_gro(skb->dev)) |
d565b0a1 HX |
5680 | goto normal; |
5681 | ||
07d78363 | 5682 | gro_head = gro_list_prepare(napi, skb); |
89c5fa33 | 5683 | |
d565b0a1 HX |
5684 | rcu_read_lock(); |
5685 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 5686 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
5687 | continue; |
5688 | ||
86911732 | 5689 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 5690 | skb_reset_mac_len(skb); |
d565b0a1 | 5691 | NAPI_GRO_CB(skb)->same_flow = 0; |
d61d072e | 5692 | NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); |
5d38a079 | 5693 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 5694 | NAPI_GRO_CB(skb)->encap_mark = 0; |
fcd91dd4 | 5695 | NAPI_GRO_CB(skb)->recursion_counter = 0; |
a0ca153f | 5696 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 5697 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 5698 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 5699 | |
662880f4 TH |
5700 | /* Setup for GRO checksum validation */ |
5701 | switch (skb->ip_summed) { | |
5702 | case CHECKSUM_COMPLETE: | |
5703 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
5704 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
5705 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
5706 | break; | |
5707 | case CHECKSUM_UNNECESSARY: | |
5708 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
5709 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
5710 | break; | |
5711 | default: | |
5712 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
5713 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
5714 | } | |
d565b0a1 | 5715 | |
aaa5d90b PA |
5716 | pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, |
5717 | ipv6_gro_receive, inet_gro_receive, | |
5718 | gro_head, skb); | |
d565b0a1 HX |
5719 | break; |
5720 | } | |
5721 | rcu_read_unlock(); | |
5722 | ||
5723 | if (&ptype->list == head) | |
5724 | goto normal; | |
5725 | ||
25393d3f SK |
5726 | if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) { |
5727 | ret = GRO_CONSUMED; | |
5728 | goto ok; | |
5729 | } | |
5730 | ||
0da2afd5 | 5731 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 5732 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 5733 | |
d565b0a1 | 5734 | if (pp) { |
992cba7e | 5735 | skb_list_del_init(pp); |
d4546c25 | 5736 | napi_gro_complete(pp); |
6312fe77 | 5737 | napi->gro_hash[hash].count--; |
d565b0a1 HX |
5738 | } |
5739 | ||
0da2afd5 | 5740 | if (same_flow) |
d565b0a1 HX |
5741 | goto ok; |
5742 | ||
600adc18 | 5743 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 5744 | goto normal; |
d565b0a1 | 5745 | |
6312fe77 LR |
5746 | if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) { |
5747 | gro_flush_oldest(gro_head); | |
600adc18 | 5748 | } else { |
6312fe77 | 5749 | napi->gro_hash[hash].count++; |
600adc18 | 5750 | } |
d565b0a1 | 5751 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 5752 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 5753 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 5754 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
07d78363 | 5755 | list_add(&skb->list, gro_head); |
5d0d9be8 | 5756 | ret = GRO_HELD; |
d565b0a1 | 5757 | |
ad0f9904 | 5758 | pull: |
a50e233c ED |
5759 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
5760 | if (grow > 0) | |
5761 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 5762 | ok: |
d9f37d01 LR |
5763 | if (napi->gro_hash[hash].count) { |
5764 | if (!test_bit(hash, &napi->gro_bitmask)) | |
5765 | __set_bit(hash, &napi->gro_bitmask); | |
5766 | } else if (test_bit(hash, &napi->gro_bitmask)) { | |
5767 | __clear_bit(hash, &napi->gro_bitmask); | |
5768 | } | |
5769 | ||
5d0d9be8 | 5770 | return ret; |
d565b0a1 HX |
5771 | |
5772 | normal: | |
ad0f9904 HX |
5773 | ret = GRO_NORMAL; |
5774 | goto pull; | |
5d38a079 | 5775 | } |
96e93eab | 5776 | |
bf5a755f JC |
5777 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
5778 | { | |
5779 | struct list_head *offload_head = &offload_base; | |
5780 | struct packet_offload *ptype; | |
5781 | ||
5782 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
5783 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
5784 | continue; | |
5785 | return ptype; | |
5786 | } | |
5787 | return NULL; | |
5788 | } | |
e27a2f83 | 5789 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
5790 | |
5791 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
5792 | { | |
5793 | struct list_head *offload_head = &offload_base; | |
5794 | struct packet_offload *ptype; | |
5795 | ||
5796 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
5797 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
5798 | continue; | |
5799 | return ptype; | |
5800 | } | |
5801 | return NULL; | |
5802 | } | |
e27a2f83 | 5803 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 5804 | |
6570bc79 AL |
5805 | /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ |
5806 | static void gro_normal_list(struct napi_struct *napi) | |
5807 | { | |
5808 | if (!napi->rx_count) | |
5809 | return; | |
5810 | netif_receive_skb_list_internal(&napi->rx_list); | |
5811 | INIT_LIST_HEAD(&napi->rx_list); | |
5812 | napi->rx_count = 0; | |
5813 | } | |
5814 | ||
5815 | /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, | |
5816 | * pass the whole batch up to the stack. | |
5817 | */ | |
5818 | static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb) | |
5819 | { | |
5820 | list_add_tail(&skb->list, &napi->rx_list); | |
5821 | if (++napi->rx_count >= gro_normal_batch) | |
5822 | gro_normal_list(napi); | |
5823 | } | |
5824 | ||
e44699d2 MK |
5825 | static void napi_skb_free_stolen_head(struct sk_buff *skb) |
5826 | { | |
5827 | skb_dst_drop(skb); | |
174e2381 | 5828 | skb_ext_put(skb); |
e44699d2 MK |
5829 | kmem_cache_free(skbuff_head_cache, skb); |
5830 | } | |
5831 | ||
6570bc79 AL |
5832 | static gro_result_t napi_skb_finish(struct napi_struct *napi, |
5833 | struct sk_buff *skb, | |
5834 | gro_result_t ret) | |
5d38a079 | 5835 | { |
5d0d9be8 HX |
5836 | switch (ret) { |
5837 | case GRO_NORMAL: | |
6570bc79 | 5838 | gro_normal_one(napi, skb); |
c7c4b3b6 | 5839 | break; |
5d38a079 | 5840 | |
5d0d9be8 | 5841 | case GRO_DROP: |
5d38a079 HX |
5842 | kfree_skb(skb); |
5843 | break; | |
5b252f0c | 5844 | |
daa86548 | 5845 | case GRO_MERGED_FREE: |
e44699d2 MK |
5846 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
5847 | napi_skb_free_stolen_head(skb); | |
5848 | else | |
d7e8883c | 5849 | __kfree_skb(skb); |
daa86548 ED |
5850 | break; |
5851 | ||
5b252f0c BH |
5852 | case GRO_HELD: |
5853 | case GRO_MERGED: | |
25393d3f | 5854 | case GRO_CONSUMED: |
5b252f0c | 5855 | break; |
5d38a079 HX |
5856 | } |
5857 | ||
c7c4b3b6 | 5858 | return ret; |
5d0d9be8 | 5859 | } |
5d0d9be8 | 5860 | |
c7c4b3b6 | 5861 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 5862 | { |
b0e3f1bd GB |
5863 | gro_result_t ret; |
5864 | ||
93f93a44 | 5865 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 5866 | trace_napi_gro_receive_entry(skb); |
86911732 | 5867 | |
a50e233c ED |
5868 | skb_gro_reset_offset(skb); |
5869 | ||
6570bc79 | 5870 | ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); |
b0e3f1bd GB |
5871 | trace_napi_gro_receive_exit(ret); |
5872 | ||
5873 | return ret; | |
d565b0a1 HX |
5874 | } |
5875 | EXPORT_SYMBOL(napi_gro_receive); | |
5876 | ||
d0c2b0d2 | 5877 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 5878 | { |
93a35f59 ED |
5879 | if (unlikely(skb->pfmemalloc)) { |
5880 | consume_skb(skb); | |
5881 | return; | |
5882 | } | |
96e93eab | 5883 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
5884 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
5885 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
b1817524 | 5886 | __vlan_hwaccel_clear_tag(skb); |
66c46d74 | 5887 | skb->dev = napi->dev; |
6d152e23 | 5888 | skb->skb_iif = 0; |
33d9a2c7 ED |
5889 | |
5890 | /* eth_type_trans() assumes pkt_type is PACKET_HOST */ | |
5891 | skb->pkt_type = PACKET_HOST; | |
5892 | ||
c3caf119 JC |
5893 | skb->encapsulation = 0; |
5894 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 5895 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
174e2381 | 5896 | skb_ext_reset(skb); |
96e93eab HX |
5897 | |
5898 | napi->skb = skb; | |
5899 | } | |
96e93eab | 5900 | |
76620aaf | 5901 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 5902 | { |
5d38a079 | 5903 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
5904 | |
5905 | if (!skb) { | |
fd11a83d | 5906 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
5907 | if (skb) { |
5908 | napi->skb = skb; | |
5909 | skb_mark_napi_id(skb, napi); | |
5910 | } | |
80595d59 | 5911 | } |
96e93eab HX |
5912 | return skb; |
5913 | } | |
76620aaf | 5914 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 5915 | |
a50e233c ED |
5916 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
5917 | struct sk_buff *skb, | |
5918 | gro_result_t ret) | |
96e93eab | 5919 | { |
5d0d9be8 HX |
5920 | switch (ret) { |
5921 | case GRO_NORMAL: | |
a50e233c ED |
5922 | case GRO_HELD: |
5923 | __skb_push(skb, ETH_HLEN); | |
5924 | skb->protocol = eth_type_trans(skb, skb->dev); | |
323ebb61 EC |
5925 | if (ret == GRO_NORMAL) |
5926 | gro_normal_one(napi, skb); | |
86911732 | 5927 | break; |
5d38a079 | 5928 | |
5d0d9be8 | 5929 | case GRO_DROP: |
5d0d9be8 HX |
5930 | napi_reuse_skb(napi, skb); |
5931 | break; | |
5b252f0c | 5932 | |
e44699d2 MK |
5933 | case GRO_MERGED_FREE: |
5934 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) | |
5935 | napi_skb_free_stolen_head(skb); | |
5936 | else | |
5937 | napi_reuse_skb(napi, skb); | |
5938 | break; | |
5939 | ||
5b252f0c | 5940 | case GRO_MERGED: |
25393d3f | 5941 | case GRO_CONSUMED: |
5b252f0c | 5942 | break; |
5d0d9be8 | 5943 | } |
5d38a079 | 5944 | |
c7c4b3b6 | 5945 | return ret; |
5d38a079 | 5946 | } |
5d0d9be8 | 5947 | |
a50e233c ED |
5948 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
5949 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
5950 | * We copy ethernet header into skb->data to have a common layout. | |
5951 | */ | |
4adb9c4a | 5952 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
5953 | { |
5954 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
5955 | const struct ethhdr *eth; |
5956 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
5957 | |
5958 | napi->skb = NULL; | |
5959 | ||
a50e233c ED |
5960 | skb_reset_mac_header(skb); |
5961 | skb_gro_reset_offset(skb); | |
5962 | ||
a50e233c ED |
5963 | if (unlikely(skb_gro_header_hard(skb, hlen))) { |
5964 | eth = skb_gro_header_slow(skb, hlen, 0); | |
5965 | if (unlikely(!eth)) { | |
4da46ceb AC |
5966 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
5967 | __func__, napi->dev->name); | |
a50e233c ED |
5968 | napi_reuse_skb(napi, skb); |
5969 | return NULL; | |
5970 | } | |
5971 | } else { | |
a4270d67 | 5972 | eth = (const struct ethhdr *)skb->data; |
a50e233c ED |
5973 | gro_pull_from_frag0(skb, hlen); |
5974 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
5975 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 5976 | } |
a50e233c ED |
5977 | __skb_pull(skb, hlen); |
5978 | ||
5979 | /* | |
5980 | * This works because the only protocols we care about don't require | |
5981 | * special handling. | |
5982 | * We'll fix it up properly in napi_frags_finish() | |
5983 | */ | |
5984 | skb->protocol = eth->h_proto; | |
76620aaf | 5985 | |
76620aaf HX |
5986 | return skb; |
5987 | } | |
76620aaf | 5988 | |
c7c4b3b6 | 5989 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 5990 | { |
b0e3f1bd | 5991 | gro_result_t ret; |
76620aaf | 5992 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
5993 | |
5994 | if (!skb) | |
c7c4b3b6 | 5995 | return GRO_DROP; |
5d0d9be8 | 5996 | |
ae78dbfa BH |
5997 | trace_napi_gro_frags_entry(skb); |
5998 | ||
b0e3f1bd GB |
5999 | ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
6000 | trace_napi_gro_frags_exit(ret); | |
6001 | ||
6002 | return ret; | |
5d0d9be8 | 6003 | } |
5d38a079 HX |
6004 | EXPORT_SYMBOL(napi_gro_frags); |
6005 | ||
573e8fca TH |
6006 | /* Compute the checksum from gro_offset and return the folded value |
6007 | * after adding in any pseudo checksum. | |
6008 | */ | |
6009 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
6010 | { | |
6011 | __wsum wsum; | |
6012 | __sum16 sum; | |
6013 | ||
6014 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
6015 | ||
6016 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
6017 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
14641931 | 6018 | /* See comments in __skb_checksum_complete(). */ |
573e8fca TH |
6019 | if (likely(!sum)) { |
6020 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
6021 | !skb->csum_complete_sw) | |
7fe50ac8 | 6022 | netdev_rx_csum_fault(skb->dev, skb); |
573e8fca TH |
6023 | } |
6024 | ||
6025 | NAPI_GRO_CB(skb)->csum = wsum; | |
6026 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
6027 | ||
6028 | return sum; | |
6029 | } | |
6030 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
6031 | ||
773fc8f6 | 6032 | static void net_rps_send_ipi(struct softnet_data *remsd) |
6033 | { | |
6034 | #ifdef CONFIG_RPS | |
6035 | while (remsd) { | |
6036 | struct softnet_data *next = remsd->rps_ipi_next; | |
6037 | ||
6038 | if (cpu_online(remsd->cpu)) | |
6039 | smp_call_function_single_async(remsd->cpu, &remsd->csd); | |
6040 | remsd = next; | |
6041 | } | |
6042 | #endif | |
6043 | } | |
6044 | ||
e326bed2 | 6045 | /* |
855abcf0 | 6046 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
6047 | * Note: called with local irq disabled, but exits with local irq enabled. |
6048 | */ | |
6049 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
6050 | { | |
6051 | #ifdef CONFIG_RPS | |
6052 | struct softnet_data *remsd = sd->rps_ipi_list; | |
6053 | ||
6054 | if (remsd) { | |
6055 | sd->rps_ipi_list = NULL; | |
6056 | ||
6057 | local_irq_enable(); | |
6058 | ||
6059 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
773fc8f6 | 6060 | net_rps_send_ipi(remsd); |
e326bed2 ED |
6061 | } else |
6062 | #endif | |
6063 | local_irq_enable(); | |
6064 | } | |
6065 | ||
d75b1ade ED |
6066 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
6067 | { | |
6068 | #ifdef CONFIG_RPS | |
6069 | return sd->rps_ipi_list != NULL; | |
6070 | #else | |
6071 | return false; | |
6072 | #endif | |
6073 | } | |
6074 | ||
bea3348e | 6075 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 6076 | { |
eecfd7c4 | 6077 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
6078 | bool again = true; |
6079 | int work = 0; | |
1da177e4 | 6080 | |
e326bed2 ED |
6081 | /* Check if we have pending ipi, its better to send them now, |
6082 | * not waiting net_rx_action() end. | |
6083 | */ | |
d75b1ade | 6084 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
6085 | local_irq_disable(); |
6086 | net_rps_action_and_irq_enable(sd); | |
6087 | } | |
d75b1ade | 6088 | |
3d48b53f | 6089 | napi->weight = dev_rx_weight; |
145dd5f9 | 6090 | while (again) { |
1da177e4 | 6091 | struct sk_buff *skb; |
6e7676c1 CG |
6092 | |
6093 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 6094 | rcu_read_lock(); |
6e7676c1 | 6095 | __netif_receive_skb(skb); |
2c17d27c | 6096 | rcu_read_unlock(); |
76cc8b13 | 6097 | input_queue_head_incr(sd); |
145dd5f9 | 6098 | if (++work >= quota) |
76cc8b13 | 6099 | return work; |
145dd5f9 | 6100 | |
6e7676c1 | 6101 | } |
1da177e4 | 6102 | |
145dd5f9 | 6103 | local_irq_disable(); |
e36fa2f7 | 6104 | rps_lock(sd); |
11ef7a89 | 6105 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
6106 | /* |
6107 | * Inline a custom version of __napi_complete(). | |
6108 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
6109 | * and NAPI_STATE_SCHED is the only possible flag set |
6110 | * on backlog. | |
6111 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
6112 | * and we dont need an smp_mb() memory barrier. |
6113 | */ | |
eecfd7c4 | 6114 | napi->state = 0; |
145dd5f9 PA |
6115 | again = false; |
6116 | } else { | |
6117 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
6118 | &sd->process_queue); | |
bea3348e | 6119 | } |
e36fa2f7 | 6120 | rps_unlock(sd); |
145dd5f9 | 6121 | local_irq_enable(); |
6e7676c1 | 6122 | } |
1da177e4 | 6123 | |
bea3348e SH |
6124 | return work; |
6125 | } | |
1da177e4 | 6126 | |
bea3348e SH |
6127 | /** |
6128 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 6129 | * @n: entry to schedule |
bea3348e | 6130 | * |
bc9ad166 ED |
6131 | * The entry's receive function will be scheduled to run. |
6132 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 6133 | */ |
b5606c2d | 6134 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
6135 | { |
6136 | unsigned long flags; | |
1da177e4 | 6137 | |
bea3348e | 6138 | local_irq_save(flags); |
903ceff7 | 6139 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 6140 | local_irq_restore(flags); |
1da177e4 | 6141 | } |
bea3348e SH |
6142 | EXPORT_SYMBOL(__napi_schedule); |
6143 | ||
39e6c820 ED |
6144 | /** |
6145 | * napi_schedule_prep - check if napi can be scheduled | |
6146 | * @n: napi context | |
6147 | * | |
6148 | * Test if NAPI routine is already running, and if not mark | |
6149 | * it as running. This is used as a condition variable | |
6150 | * insure only one NAPI poll instance runs. We also make | |
6151 | * sure there is no pending NAPI disable. | |
6152 | */ | |
6153 | bool napi_schedule_prep(struct napi_struct *n) | |
6154 | { | |
6155 | unsigned long val, new; | |
6156 | ||
6157 | do { | |
6158 | val = READ_ONCE(n->state); | |
6159 | if (unlikely(val & NAPIF_STATE_DISABLE)) | |
6160 | return false; | |
6161 | new = val | NAPIF_STATE_SCHED; | |
6162 | ||
6163 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
6164 | * This was suggested by Alexander Duyck, as compiler | |
6165 | * emits better code than : | |
6166 | * if (val & NAPIF_STATE_SCHED) | |
6167 | * new |= NAPIF_STATE_MISSED; | |
6168 | */ | |
6169 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
6170 | NAPIF_STATE_MISSED; | |
6171 | } while (cmpxchg(&n->state, val, new) != val); | |
6172 | ||
6173 | return !(val & NAPIF_STATE_SCHED); | |
6174 | } | |
6175 | EXPORT_SYMBOL(napi_schedule_prep); | |
6176 | ||
bc9ad166 ED |
6177 | /** |
6178 | * __napi_schedule_irqoff - schedule for receive | |
6179 | * @n: entry to schedule | |
6180 | * | |
6181 | * Variant of __napi_schedule() assuming hard irqs are masked | |
6182 | */ | |
6183 | void __napi_schedule_irqoff(struct napi_struct *n) | |
6184 | { | |
6185 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
6186 | } | |
6187 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
6188 | ||
364b6055 | 6189 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 6190 | { |
39e6c820 | 6191 | unsigned long flags, val, new; |
d565b0a1 HX |
6192 | |
6193 | /* | |
217f6974 ED |
6194 | * 1) Don't let napi dequeue from the cpu poll list |
6195 | * just in case its running on a different cpu. | |
6196 | * 2) If we are busy polling, do nothing here, we have | |
6197 | * the guarantee we will be called later. | |
d565b0a1 | 6198 | */ |
217f6974 ED |
6199 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
6200 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 6201 | return false; |
d565b0a1 | 6202 | |
323ebb61 EC |
6203 | gro_normal_list(n); |
6204 | ||
d9f37d01 | 6205 | if (n->gro_bitmask) { |
3b47d303 | 6206 | unsigned long timeout = 0; |
d75b1ade | 6207 | |
3b47d303 ED |
6208 | if (work_done) |
6209 | timeout = n->dev->gro_flush_timeout; | |
6210 | ||
605108ac PA |
6211 | /* When the NAPI instance uses a timeout and keeps postponing |
6212 | * it, we need to bound somehow the time packets are kept in | |
6213 | * the GRO layer | |
6214 | */ | |
6215 | napi_gro_flush(n, !!timeout); | |
3b47d303 ED |
6216 | if (timeout) |
6217 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
6218 | HRTIMER_MODE_REL_PINNED); | |
3b47d303 | 6219 | } |
02c1602e | 6220 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
6221 | /* If n->poll_list is not empty, we need to mask irqs */ |
6222 | local_irq_save(flags); | |
02c1602e | 6223 | list_del_init(&n->poll_list); |
d75b1ade ED |
6224 | local_irq_restore(flags); |
6225 | } | |
39e6c820 ED |
6226 | |
6227 | do { | |
6228 | val = READ_ONCE(n->state); | |
6229 | ||
6230 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); | |
6231 | ||
6232 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); | |
6233 | ||
6234 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
6235 | * because we will call napi->poll() one more time. | |
6236 | * This C code was suggested by Alexander Duyck to help gcc. | |
6237 | */ | |
6238 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
6239 | NAPIF_STATE_SCHED; | |
6240 | } while (cmpxchg(&n->state, val, new) != val); | |
6241 | ||
6242 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
6243 | __napi_schedule(n); | |
6244 | return false; | |
6245 | } | |
6246 | ||
364b6055 | 6247 | return true; |
d565b0a1 | 6248 | } |
3b47d303 | 6249 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 6250 | |
af12fa6e | 6251 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 6252 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
6253 | { |
6254 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
6255 | struct napi_struct *napi; | |
6256 | ||
6257 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
6258 | if (napi->napi_id == napi_id) | |
6259 | return napi; | |
6260 | ||
6261 | return NULL; | |
6262 | } | |
02d62e86 ED |
6263 | |
6264 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
217f6974 | 6265 | |
ce6aea93 | 6266 | #define BUSY_POLL_BUDGET 8 |
217f6974 ED |
6267 | |
6268 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) | |
6269 | { | |
6270 | int rc; | |
6271 | ||
39e6c820 ED |
6272 | /* Busy polling means there is a high chance device driver hard irq |
6273 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
6274 | * set in napi_schedule_prep(). | |
6275 | * Since we are about to call napi->poll() once more, we can safely | |
6276 | * clear NAPI_STATE_MISSED. | |
6277 | * | |
6278 | * Note: x86 could use a single "lock and ..." instruction | |
6279 | * to perform these two clear_bit() | |
6280 | */ | |
6281 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
6282 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
6283 | ||
6284 | local_bh_disable(); | |
6285 | ||
6286 | /* All we really want here is to re-enable device interrupts. | |
6287 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
6288 | */ | |
6289 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
323ebb61 EC |
6290 | /* We can't gro_normal_list() here, because napi->poll() might have |
6291 | * rearmed the napi (napi_complete_done()) in which case it could | |
6292 | * already be running on another CPU. | |
6293 | */ | |
1e22391e | 6294 | trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); |
217f6974 | 6295 | netpoll_poll_unlock(have_poll_lock); |
323ebb61 EC |
6296 | if (rc == BUSY_POLL_BUDGET) { |
6297 | /* As the whole budget was spent, we still own the napi so can | |
6298 | * safely handle the rx_list. | |
6299 | */ | |
6300 | gro_normal_list(napi); | |
217f6974 | 6301 | __napi_schedule(napi); |
323ebb61 | 6302 | } |
217f6974 | 6303 | local_bh_enable(); |
217f6974 ED |
6304 | } |
6305 | ||
7db6b048 SS |
6306 | void napi_busy_loop(unsigned int napi_id, |
6307 | bool (*loop_end)(void *, unsigned long), | |
6308 | void *loop_end_arg) | |
02d62e86 | 6309 | { |
7db6b048 | 6310 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 6311 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 6312 | void *have_poll_lock = NULL; |
02d62e86 | 6313 | struct napi_struct *napi; |
217f6974 ED |
6314 | |
6315 | restart: | |
217f6974 | 6316 | napi_poll = NULL; |
02d62e86 | 6317 | |
2a028ecb | 6318 | rcu_read_lock(); |
02d62e86 | 6319 | |
545cd5e5 | 6320 | napi = napi_by_id(napi_id); |
02d62e86 ED |
6321 | if (!napi) |
6322 | goto out; | |
6323 | ||
217f6974 ED |
6324 | preempt_disable(); |
6325 | for (;;) { | |
2b5cd0df AD |
6326 | int work = 0; |
6327 | ||
2a028ecb | 6328 | local_bh_disable(); |
217f6974 ED |
6329 | if (!napi_poll) { |
6330 | unsigned long val = READ_ONCE(napi->state); | |
6331 | ||
6332 | /* If multiple threads are competing for this napi, | |
6333 | * we avoid dirtying napi->state as much as we can. | |
6334 | */ | |
6335 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
6336 | NAPIF_STATE_IN_BUSY_POLL)) | |
6337 | goto count; | |
6338 | if (cmpxchg(&napi->state, val, | |
6339 | val | NAPIF_STATE_IN_BUSY_POLL | | |
6340 | NAPIF_STATE_SCHED) != val) | |
6341 | goto count; | |
6342 | have_poll_lock = netpoll_poll_lock(napi); | |
6343 | napi_poll = napi->poll; | |
6344 | } | |
2b5cd0df AD |
6345 | work = napi_poll(napi, BUSY_POLL_BUDGET); |
6346 | trace_napi_poll(napi, work, BUSY_POLL_BUDGET); | |
323ebb61 | 6347 | gro_normal_list(napi); |
217f6974 | 6348 | count: |
2b5cd0df | 6349 | if (work > 0) |
7db6b048 | 6350 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 6351 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
2a028ecb | 6352 | local_bh_enable(); |
02d62e86 | 6353 | |
7db6b048 | 6354 | if (!loop_end || loop_end(loop_end_arg, start_time)) |
217f6974 | 6355 | break; |
02d62e86 | 6356 | |
217f6974 ED |
6357 | if (unlikely(need_resched())) { |
6358 | if (napi_poll) | |
6359 | busy_poll_stop(napi, have_poll_lock); | |
6360 | preempt_enable(); | |
6361 | rcu_read_unlock(); | |
6362 | cond_resched(); | |
7db6b048 | 6363 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 6364 | return; |
217f6974 ED |
6365 | goto restart; |
6366 | } | |
6cdf89b1 | 6367 | cpu_relax(); |
217f6974 ED |
6368 | } |
6369 | if (napi_poll) | |
6370 | busy_poll_stop(napi, have_poll_lock); | |
6371 | preempt_enable(); | |
02d62e86 | 6372 | out: |
2a028ecb | 6373 | rcu_read_unlock(); |
02d62e86 | 6374 | } |
7db6b048 | 6375 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
6376 | |
6377 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 6378 | |
149d6ad8 | 6379 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 6380 | { |
d64b5e85 ED |
6381 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
6382 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 6383 | return; |
af12fa6e | 6384 | |
52bd2d62 | 6385 | spin_lock(&napi_hash_lock); |
af12fa6e | 6386 | |
545cd5e5 | 6387 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 6388 | do { |
545cd5e5 AD |
6389 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
6390 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
6391 | } while (napi_by_id(napi_gen_id)); |
6392 | napi->napi_id = napi_gen_id; | |
af12fa6e | 6393 | |
52bd2d62 ED |
6394 | hlist_add_head_rcu(&napi->napi_hash_node, |
6395 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 6396 | |
52bd2d62 | 6397 | spin_unlock(&napi_hash_lock); |
af12fa6e | 6398 | } |
af12fa6e ET |
6399 | |
6400 | /* Warning : caller is responsible to make sure rcu grace period | |
6401 | * is respected before freeing memory containing @napi | |
6402 | */ | |
34cbe27e | 6403 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 6404 | { |
34cbe27e ED |
6405 | bool rcu_sync_needed = false; |
6406 | ||
af12fa6e ET |
6407 | spin_lock(&napi_hash_lock); |
6408 | ||
34cbe27e ED |
6409 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
6410 | rcu_sync_needed = true; | |
af12fa6e | 6411 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 6412 | } |
af12fa6e | 6413 | spin_unlock(&napi_hash_lock); |
34cbe27e | 6414 | return rcu_sync_needed; |
af12fa6e ET |
6415 | } |
6416 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
6417 | ||
3b47d303 ED |
6418 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
6419 | { | |
6420 | struct napi_struct *napi; | |
6421 | ||
6422 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
6423 | |
6424 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
6425 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
6426 | */ | |
d9f37d01 | 6427 | if (napi->gro_bitmask && !napi_disable_pending(napi) && |
39e6c820 ED |
6428 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) |
6429 | __napi_schedule_irqoff(napi); | |
3b47d303 ED |
6430 | |
6431 | return HRTIMER_NORESTART; | |
6432 | } | |
6433 | ||
7c4ec749 | 6434 | static void init_gro_hash(struct napi_struct *napi) |
d565b0a1 | 6435 | { |
07d78363 DM |
6436 | int i; |
6437 | ||
6312fe77 LR |
6438 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6439 | INIT_LIST_HEAD(&napi->gro_hash[i].list); | |
6440 | napi->gro_hash[i].count = 0; | |
6441 | } | |
7c4ec749 DM |
6442 | napi->gro_bitmask = 0; |
6443 | } | |
6444 | ||
6445 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
6446 | int (*poll)(struct napi_struct *, int), int weight) | |
6447 | { | |
6448 | INIT_LIST_HEAD(&napi->poll_list); | |
6449 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); | |
6450 | napi->timer.function = napi_watchdog; | |
6451 | init_gro_hash(napi); | |
5d38a079 | 6452 | napi->skb = NULL; |
323ebb61 EC |
6453 | INIT_LIST_HEAD(&napi->rx_list); |
6454 | napi->rx_count = 0; | |
d565b0a1 | 6455 | napi->poll = poll; |
82dc3c63 | 6456 | if (weight > NAPI_POLL_WEIGHT) |
bf29e9e9 QC |
6457 | netdev_err_once(dev, "%s() called with weight %d\n", __func__, |
6458 | weight); | |
d565b0a1 HX |
6459 | napi->weight = weight; |
6460 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 6461 | napi->dev = dev; |
5d38a079 | 6462 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
6463 | napi->poll_owner = -1; |
6464 | #endif | |
6465 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 6466 | napi_hash_add(napi); |
d565b0a1 HX |
6467 | } |
6468 | EXPORT_SYMBOL(netif_napi_add); | |
6469 | ||
3b47d303 ED |
6470 | void napi_disable(struct napi_struct *n) |
6471 | { | |
6472 | might_sleep(); | |
6473 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
6474 | ||
6475 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
6476 | msleep(1); | |
2d8bff12 NH |
6477 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
6478 | msleep(1); | |
3b47d303 ED |
6479 | |
6480 | hrtimer_cancel(&n->timer); | |
6481 | ||
6482 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
6483 | } | |
6484 | EXPORT_SYMBOL(napi_disable); | |
6485 | ||
07d78363 | 6486 | static void flush_gro_hash(struct napi_struct *napi) |
d4546c25 | 6487 | { |
07d78363 | 6488 | int i; |
d4546c25 | 6489 | |
07d78363 DM |
6490 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6491 | struct sk_buff *skb, *n; | |
6492 | ||
6312fe77 | 6493 | list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) |
07d78363 | 6494 | kfree_skb(skb); |
6312fe77 | 6495 | napi->gro_hash[i].count = 0; |
07d78363 | 6496 | } |
d4546c25 DM |
6497 | } |
6498 | ||
93d05d4a | 6499 | /* Must be called in process context */ |
d565b0a1 HX |
6500 | void netif_napi_del(struct napi_struct *napi) |
6501 | { | |
93d05d4a ED |
6502 | might_sleep(); |
6503 | if (napi_hash_del(napi)) | |
6504 | synchronize_net(); | |
d7b06636 | 6505 | list_del_init(&napi->dev_list); |
76620aaf | 6506 | napi_free_frags(napi); |
d565b0a1 | 6507 | |
07d78363 | 6508 | flush_gro_hash(napi); |
d9f37d01 | 6509 | napi->gro_bitmask = 0; |
d565b0a1 HX |
6510 | } |
6511 | EXPORT_SYMBOL(netif_napi_del); | |
6512 | ||
726ce70e HX |
6513 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
6514 | { | |
6515 | void *have; | |
6516 | int work, weight; | |
6517 | ||
6518 | list_del_init(&n->poll_list); | |
6519 | ||
6520 | have = netpoll_poll_lock(n); | |
6521 | ||
6522 | weight = n->weight; | |
6523 | ||
6524 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
6525 | * with netpoll's poll_napi(). Only the entity which | |
6526 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
6527 | * actually make the ->poll() call. Therefore we avoid | |
6528 | * accidentally calling ->poll() when NAPI is not scheduled. | |
6529 | */ | |
6530 | work = 0; | |
6531 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
6532 | work = n->poll(n, weight); | |
1db19db7 | 6533 | trace_napi_poll(n, work, weight); |
726ce70e HX |
6534 | } |
6535 | ||
6536 | WARN_ON_ONCE(work > weight); | |
6537 | ||
6538 | if (likely(work < weight)) | |
6539 | goto out_unlock; | |
6540 | ||
6541 | /* Drivers must not modify the NAPI state if they | |
6542 | * consume the entire weight. In such cases this code | |
6543 | * still "owns" the NAPI instance and therefore can | |
6544 | * move the instance around on the list at-will. | |
6545 | */ | |
6546 | if (unlikely(napi_disable_pending(n))) { | |
6547 | napi_complete(n); | |
6548 | goto out_unlock; | |
6549 | } | |
6550 | ||
323ebb61 EC |
6551 | gro_normal_list(n); |
6552 | ||
d9f37d01 | 6553 | if (n->gro_bitmask) { |
726ce70e HX |
6554 | /* flush too old packets |
6555 | * If HZ < 1000, flush all packets. | |
6556 | */ | |
6557 | napi_gro_flush(n, HZ >= 1000); | |
6558 | } | |
6559 | ||
001ce546 HX |
6560 | /* Some drivers may have called napi_schedule |
6561 | * prior to exhausting their budget. | |
6562 | */ | |
6563 | if (unlikely(!list_empty(&n->poll_list))) { | |
6564 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
6565 | n->dev ? n->dev->name : "backlog"); | |
6566 | goto out_unlock; | |
6567 | } | |
6568 | ||
726ce70e HX |
6569 | list_add_tail(&n->poll_list, repoll); |
6570 | ||
6571 | out_unlock: | |
6572 | netpoll_poll_unlock(have); | |
6573 | ||
6574 | return work; | |
6575 | } | |
6576 | ||
0766f788 | 6577 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 6578 | { |
903ceff7 | 6579 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e MW |
6580 | unsigned long time_limit = jiffies + |
6581 | usecs_to_jiffies(netdev_budget_usecs); | |
51b0bded | 6582 | int budget = netdev_budget; |
d75b1ade ED |
6583 | LIST_HEAD(list); |
6584 | LIST_HEAD(repoll); | |
53fb95d3 | 6585 | |
1da177e4 | 6586 | local_irq_disable(); |
d75b1ade ED |
6587 | list_splice_init(&sd->poll_list, &list); |
6588 | local_irq_enable(); | |
1da177e4 | 6589 | |
ceb8d5bf | 6590 | for (;;) { |
bea3348e | 6591 | struct napi_struct *n; |
1da177e4 | 6592 | |
ceb8d5bf HX |
6593 | if (list_empty(&list)) { |
6594 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
f52dffe0 | 6595 | goto out; |
ceb8d5bf HX |
6596 | break; |
6597 | } | |
6598 | ||
6bd373eb HX |
6599 | n = list_first_entry(&list, struct napi_struct, poll_list); |
6600 | budget -= napi_poll(n, &repoll); | |
6601 | ||
d75b1ade | 6602 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
6603 | * Allow this to run for 2 jiffies since which will allow |
6604 | * an average latency of 1.5/HZ. | |
bea3348e | 6605 | */ |
ceb8d5bf HX |
6606 | if (unlikely(budget <= 0 || |
6607 | time_after_eq(jiffies, time_limit))) { | |
6608 | sd->time_squeeze++; | |
6609 | break; | |
6610 | } | |
1da177e4 | 6611 | } |
d75b1ade | 6612 | |
d75b1ade ED |
6613 | local_irq_disable(); |
6614 | ||
6615 | list_splice_tail_init(&sd->poll_list, &list); | |
6616 | list_splice_tail(&repoll, &list); | |
6617 | list_splice(&list, &sd->poll_list); | |
6618 | if (!list_empty(&sd->poll_list)) | |
6619 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6620 | ||
e326bed2 | 6621 | net_rps_action_and_irq_enable(sd); |
f52dffe0 ED |
6622 | out: |
6623 | __kfree_skb_flush(); | |
1da177e4 LT |
6624 | } |
6625 | ||
aa9d8560 | 6626 | struct netdev_adjacent { |
9ff162a8 | 6627 | struct net_device *dev; |
5d261913 VF |
6628 | |
6629 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 6630 | bool master; |
5d261913 | 6631 | |
32b6d34f TY |
6632 | /* lookup ignore flag */ |
6633 | bool ignore; | |
6634 | ||
5d261913 VF |
6635 | /* counter for the number of times this device was added to us */ |
6636 | u16 ref_nr; | |
6637 | ||
402dae96 VF |
6638 | /* private field for the users */ |
6639 | void *private; | |
6640 | ||
9ff162a8 JP |
6641 | struct list_head list; |
6642 | struct rcu_head rcu; | |
9ff162a8 JP |
6643 | }; |
6644 | ||
6ea29da1 | 6645 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 6646 | struct list_head *adj_list) |
9ff162a8 | 6647 | { |
5d261913 | 6648 | struct netdev_adjacent *adj; |
5d261913 | 6649 | |
2f268f12 | 6650 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
6651 | if (adj->dev == adj_dev) |
6652 | return adj; | |
9ff162a8 JP |
6653 | } |
6654 | return NULL; | |
6655 | } | |
6656 | ||
32b6d34f | 6657 | static int ____netdev_has_upper_dev(struct net_device *upper_dev, void *data) |
f1170fd4 DA |
6658 | { |
6659 | struct net_device *dev = data; | |
6660 | ||
6661 | return upper_dev == dev; | |
6662 | } | |
6663 | ||
9ff162a8 JP |
6664 | /** |
6665 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
6666 | * @dev: device | |
6667 | * @upper_dev: upper device to check | |
6668 | * | |
6669 | * Find out if a device is linked to specified upper device and return true | |
6670 | * in case it is. Note that this checks only immediate upper device, | |
6671 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
6672 | */ | |
6673 | bool netdev_has_upper_dev(struct net_device *dev, | |
6674 | struct net_device *upper_dev) | |
6675 | { | |
6676 | ASSERT_RTNL(); | |
6677 | ||
32b6d34f | 6678 | return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, |
f1170fd4 | 6679 | upper_dev); |
9ff162a8 JP |
6680 | } |
6681 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
6682 | ||
1a3f060c DA |
6683 | /** |
6684 | * netdev_has_upper_dev_all - Check if device is linked to an upper device | |
6685 | * @dev: device | |
6686 | * @upper_dev: upper device to check | |
6687 | * | |
6688 | * Find out if a device is linked to specified upper device and return true | |
6689 | * in case it is. Note that this checks the entire upper device chain. | |
6690 | * The caller must hold rcu lock. | |
6691 | */ | |
6692 | ||
1a3f060c DA |
6693 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
6694 | struct net_device *upper_dev) | |
6695 | { | |
32b6d34f | 6696 | return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, |
1a3f060c DA |
6697 | upper_dev); |
6698 | } | |
6699 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
6700 | ||
9ff162a8 JP |
6701 | /** |
6702 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
6703 | * @dev: device | |
6704 | * | |
6705 | * Find out if a device is linked to an upper device and return true in case | |
6706 | * it is. The caller must hold the RTNL lock. | |
6707 | */ | |
25cc72a3 | 6708 | bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
6709 | { |
6710 | ASSERT_RTNL(); | |
6711 | ||
f1170fd4 | 6712 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 6713 | } |
25cc72a3 | 6714 | EXPORT_SYMBOL(netdev_has_any_upper_dev); |
9ff162a8 JP |
6715 | |
6716 | /** | |
6717 | * netdev_master_upper_dev_get - Get master upper device | |
6718 | * @dev: device | |
6719 | * | |
6720 | * Find a master upper device and return pointer to it or NULL in case | |
6721 | * it's not there. The caller must hold the RTNL lock. | |
6722 | */ | |
6723 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
6724 | { | |
aa9d8560 | 6725 | struct netdev_adjacent *upper; |
9ff162a8 JP |
6726 | |
6727 | ASSERT_RTNL(); | |
6728 | ||
2f268f12 | 6729 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
6730 | return NULL; |
6731 | ||
2f268f12 | 6732 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 6733 | struct netdev_adjacent, list); |
9ff162a8 JP |
6734 | if (likely(upper->master)) |
6735 | return upper->dev; | |
6736 | return NULL; | |
6737 | } | |
6738 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
6739 | ||
32b6d34f TY |
6740 | static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) |
6741 | { | |
6742 | struct netdev_adjacent *upper; | |
6743 | ||
6744 | ASSERT_RTNL(); | |
6745 | ||
6746 | if (list_empty(&dev->adj_list.upper)) | |
6747 | return NULL; | |
6748 | ||
6749 | upper = list_first_entry(&dev->adj_list.upper, | |
6750 | struct netdev_adjacent, list); | |
6751 | if (likely(upper->master) && !upper->ignore) | |
6752 | return upper->dev; | |
6753 | return NULL; | |
6754 | } | |
6755 | ||
0f524a80 DA |
6756 | /** |
6757 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
6758 | * @dev: device | |
6759 | * | |
6760 | * Find out if a device is linked to a lower device and return true in case | |
6761 | * it is. The caller must hold the RTNL lock. | |
6762 | */ | |
6763 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
6764 | { | |
6765 | ASSERT_RTNL(); | |
6766 | ||
6767 | return !list_empty(&dev->adj_list.lower); | |
6768 | } | |
6769 | ||
b6ccba4c VF |
6770 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
6771 | { | |
6772 | struct netdev_adjacent *adj; | |
6773 | ||
6774 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
6775 | ||
6776 | return adj->private; | |
6777 | } | |
6778 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
6779 | ||
44a40855 VY |
6780 | /** |
6781 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
6782 | * @dev: device | |
6783 | * @iter: list_head ** of the current position | |
6784 | * | |
6785 | * Gets the next device from the dev's upper list, starting from iter | |
6786 | * position. The caller must hold RCU read lock. | |
6787 | */ | |
6788 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
6789 | struct list_head **iter) | |
6790 | { | |
6791 | struct netdev_adjacent *upper; | |
6792 | ||
6793 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
6794 | ||
6795 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6796 | ||
6797 | if (&upper->list == &dev->adj_list.upper) | |
6798 | return NULL; | |
6799 | ||
6800 | *iter = &upper->list; | |
6801 | ||
6802 | return upper->dev; | |
6803 | } | |
6804 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
6805 | ||
32b6d34f TY |
6806 | static struct net_device *__netdev_next_upper_dev(struct net_device *dev, |
6807 | struct list_head **iter, | |
6808 | bool *ignore) | |
5343da4c TY |
6809 | { |
6810 | struct netdev_adjacent *upper; | |
6811 | ||
6812 | upper = list_entry((*iter)->next, struct netdev_adjacent, list); | |
6813 | ||
6814 | if (&upper->list == &dev->adj_list.upper) | |
6815 | return NULL; | |
6816 | ||
6817 | *iter = &upper->list; | |
32b6d34f | 6818 | *ignore = upper->ignore; |
5343da4c TY |
6819 | |
6820 | return upper->dev; | |
6821 | } | |
6822 | ||
1a3f060c DA |
6823 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
6824 | struct list_head **iter) | |
6825 | { | |
6826 | struct netdev_adjacent *upper; | |
6827 | ||
6828 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
6829 | ||
6830 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6831 | ||
6832 | if (&upper->list == &dev->adj_list.upper) | |
6833 | return NULL; | |
6834 | ||
6835 | *iter = &upper->list; | |
6836 | ||
6837 | return upper->dev; | |
6838 | } | |
6839 | ||
32b6d34f TY |
6840 | static int __netdev_walk_all_upper_dev(struct net_device *dev, |
6841 | int (*fn)(struct net_device *dev, | |
6842 | void *data), | |
6843 | void *data) | |
5343da4c TY |
6844 | { |
6845 | struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
6846 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
6847 | int ret, cur = 0; | |
32b6d34f | 6848 | bool ignore; |
5343da4c TY |
6849 | |
6850 | now = dev; | |
6851 | iter = &dev->adj_list.upper; | |
6852 | ||
6853 | while (1) { | |
6854 | if (now != dev) { | |
6855 | ret = fn(now, data); | |
6856 | if (ret) | |
6857 | return ret; | |
6858 | } | |
6859 | ||
6860 | next = NULL; | |
6861 | while (1) { | |
32b6d34f | 6862 | udev = __netdev_next_upper_dev(now, &iter, &ignore); |
5343da4c TY |
6863 | if (!udev) |
6864 | break; | |
32b6d34f TY |
6865 | if (ignore) |
6866 | continue; | |
5343da4c TY |
6867 | |
6868 | next = udev; | |
6869 | niter = &udev->adj_list.upper; | |
6870 | dev_stack[cur] = now; | |
6871 | iter_stack[cur++] = iter; | |
6872 | break; | |
6873 | } | |
6874 | ||
6875 | if (!next) { | |
6876 | if (!cur) | |
6877 | return 0; | |
6878 | next = dev_stack[--cur]; | |
6879 | niter = iter_stack[cur]; | |
6880 | } | |
6881 | ||
6882 | now = next; | |
6883 | iter = niter; | |
6884 | } | |
6885 | ||
6886 | return 0; | |
6887 | } | |
6888 | ||
1a3f060c DA |
6889 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, |
6890 | int (*fn)(struct net_device *dev, | |
6891 | void *data), | |
6892 | void *data) | |
6893 | { | |
5343da4c TY |
6894 | struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; |
6895 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
6896 | int ret, cur = 0; | |
1a3f060c | 6897 | |
5343da4c TY |
6898 | now = dev; |
6899 | iter = &dev->adj_list.upper; | |
1a3f060c | 6900 | |
5343da4c TY |
6901 | while (1) { |
6902 | if (now != dev) { | |
6903 | ret = fn(now, data); | |
6904 | if (ret) | |
6905 | return ret; | |
6906 | } | |
6907 | ||
6908 | next = NULL; | |
6909 | while (1) { | |
6910 | udev = netdev_next_upper_dev_rcu(now, &iter); | |
6911 | if (!udev) | |
6912 | break; | |
6913 | ||
6914 | next = udev; | |
6915 | niter = &udev->adj_list.upper; | |
6916 | dev_stack[cur] = now; | |
6917 | iter_stack[cur++] = iter; | |
6918 | break; | |
6919 | } | |
6920 | ||
6921 | if (!next) { | |
6922 | if (!cur) | |
6923 | return 0; | |
6924 | next = dev_stack[--cur]; | |
6925 | niter = iter_stack[cur]; | |
6926 | } | |
6927 | ||
6928 | now = next; | |
6929 | iter = niter; | |
1a3f060c DA |
6930 | } |
6931 | ||
6932 | return 0; | |
6933 | } | |
6934 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
6935 | ||
32b6d34f TY |
6936 | static bool __netdev_has_upper_dev(struct net_device *dev, |
6937 | struct net_device *upper_dev) | |
6938 | { | |
6939 | ASSERT_RTNL(); | |
6940 | ||
6941 | return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, | |
6942 | upper_dev); | |
6943 | } | |
6944 | ||
31088a11 VF |
6945 | /** |
6946 | * netdev_lower_get_next_private - Get the next ->private from the | |
6947 | * lower neighbour list | |
6948 | * @dev: device | |
6949 | * @iter: list_head ** of the current position | |
6950 | * | |
6951 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
6952 | * list, starting from iter position. The caller must hold either hold the | |
6953 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 6954 | * list will remain unchanged. |
31088a11 VF |
6955 | */ |
6956 | void *netdev_lower_get_next_private(struct net_device *dev, | |
6957 | struct list_head **iter) | |
6958 | { | |
6959 | struct netdev_adjacent *lower; | |
6960 | ||
6961 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
6962 | ||
6963 | if (&lower->list == &dev->adj_list.lower) | |
6964 | return NULL; | |
6965 | ||
6859e7df | 6966 | *iter = lower->list.next; |
31088a11 VF |
6967 | |
6968 | return lower->private; | |
6969 | } | |
6970 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
6971 | ||
6972 | /** | |
6973 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
6974 | * lower neighbour list, RCU | |
6975 | * variant | |
6976 | * @dev: device | |
6977 | * @iter: list_head ** of the current position | |
6978 | * | |
6979 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
6980 | * list, starting from iter position. The caller must hold RCU read lock. | |
6981 | */ | |
6982 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
6983 | struct list_head **iter) | |
6984 | { | |
6985 | struct netdev_adjacent *lower; | |
6986 | ||
6987 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
6988 | ||
6989 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6990 | ||
6991 | if (&lower->list == &dev->adj_list.lower) | |
6992 | return NULL; | |
6993 | ||
6859e7df | 6994 | *iter = &lower->list; |
31088a11 VF |
6995 | |
6996 | return lower->private; | |
6997 | } | |
6998 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
6999 | ||
4085ebe8 VY |
7000 | /** |
7001 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
7002 | * list | |
7003 | * @dev: device | |
7004 | * @iter: list_head ** of the current position | |
7005 | * | |
7006 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
7007 | * list, starting from iter position. The caller must hold RTNL lock or | |
7008 | * its own locking that guarantees that the neighbour lower | |
b469139e | 7009 | * list will remain unchanged. |
4085ebe8 VY |
7010 | */ |
7011 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
7012 | { | |
7013 | struct netdev_adjacent *lower; | |
7014 | ||
cfdd28be | 7015 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
7016 | |
7017 | if (&lower->list == &dev->adj_list.lower) | |
7018 | return NULL; | |
7019 | ||
cfdd28be | 7020 | *iter = lower->list.next; |
4085ebe8 VY |
7021 | |
7022 | return lower->dev; | |
7023 | } | |
7024 | EXPORT_SYMBOL(netdev_lower_get_next); | |
7025 | ||
1a3f060c DA |
7026 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
7027 | struct list_head **iter) | |
7028 | { | |
7029 | struct netdev_adjacent *lower; | |
7030 | ||
46b5ab1a | 7031 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
7032 | |
7033 | if (&lower->list == &dev->adj_list.lower) | |
7034 | return NULL; | |
7035 | ||
46b5ab1a | 7036 | *iter = &lower->list; |
1a3f060c DA |
7037 | |
7038 | return lower->dev; | |
7039 | } | |
7040 | ||
32b6d34f TY |
7041 | static struct net_device *__netdev_next_lower_dev(struct net_device *dev, |
7042 | struct list_head **iter, | |
7043 | bool *ignore) | |
7044 | { | |
7045 | struct netdev_adjacent *lower; | |
7046 | ||
7047 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
7048 | ||
7049 | if (&lower->list == &dev->adj_list.lower) | |
7050 | return NULL; | |
7051 | ||
7052 | *iter = &lower->list; | |
7053 | *ignore = lower->ignore; | |
7054 | ||
7055 | return lower->dev; | |
7056 | } | |
7057 | ||
1a3f060c DA |
7058 | int netdev_walk_all_lower_dev(struct net_device *dev, |
7059 | int (*fn)(struct net_device *dev, | |
7060 | void *data), | |
7061 | void *data) | |
7062 | { | |
5343da4c TY |
7063 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; |
7064 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7065 | int ret, cur = 0; | |
1a3f060c | 7066 | |
5343da4c TY |
7067 | now = dev; |
7068 | iter = &dev->adj_list.lower; | |
1a3f060c | 7069 | |
5343da4c TY |
7070 | while (1) { |
7071 | if (now != dev) { | |
7072 | ret = fn(now, data); | |
7073 | if (ret) | |
7074 | return ret; | |
7075 | } | |
7076 | ||
7077 | next = NULL; | |
7078 | while (1) { | |
7079 | ldev = netdev_next_lower_dev(now, &iter); | |
7080 | if (!ldev) | |
7081 | break; | |
7082 | ||
7083 | next = ldev; | |
7084 | niter = &ldev->adj_list.lower; | |
7085 | dev_stack[cur] = now; | |
7086 | iter_stack[cur++] = iter; | |
7087 | break; | |
7088 | } | |
7089 | ||
7090 | if (!next) { | |
7091 | if (!cur) | |
7092 | return 0; | |
7093 | next = dev_stack[--cur]; | |
7094 | niter = iter_stack[cur]; | |
7095 | } | |
7096 | ||
7097 | now = next; | |
7098 | iter = niter; | |
1a3f060c DA |
7099 | } |
7100 | ||
7101 | return 0; | |
7102 | } | |
7103 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
7104 | ||
32b6d34f TY |
7105 | static int __netdev_walk_all_lower_dev(struct net_device *dev, |
7106 | int (*fn)(struct net_device *dev, | |
7107 | void *data), | |
7108 | void *data) | |
7109 | { | |
7110 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7111 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7112 | int ret, cur = 0; | |
7113 | bool ignore; | |
7114 | ||
7115 | now = dev; | |
7116 | iter = &dev->adj_list.lower; | |
7117 | ||
7118 | while (1) { | |
7119 | if (now != dev) { | |
7120 | ret = fn(now, data); | |
7121 | if (ret) | |
7122 | return ret; | |
7123 | } | |
7124 | ||
7125 | next = NULL; | |
7126 | while (1) { | |
7127 | ldev = __netdev_next_lower_dev(now, &iter, &ignore); | |
7128 | if (!ldev) | |
7129 | break; | |
7130 | if (ignore) | |
7131 | continue; | |
7132 | ||
7133 | next = ldev; | |
7134 | niter = &ldev->adj_list.lower; | |
7135 | dev_stack[cur] = now; | |
7136 | iter_stack[cur++] = iter; | |
7137 | break; | |
7138 | } | |
7139 | ||
7140 | if (!next) { | |
7141 | if (!cur) | |
7142 | return 0; | |
7143 | next = dev_stack[--cur]; | |
7144 | niter = iter_stack[cur]; | |
7145 | } | |
7146 | ||
7147 | now = next; | |
7148 | iter = niter; | |
7149 | } | |
7150 | ||
7151 | return 0; | |
7152 | } | |
7153 | ||
1a3f060c DA |
7154 | static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
7155 | struct list_head **iter) | |
7156 | { | |
7157 | struct netdev_adjacent *lower; | |
7158 | ||
7159 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7160 | if (&lower->list == &dev->adj_list.lower) | |
7161 | return NULL; | |
7162 | ||
7163 | *iter = &lower->list; | |
7164 | ||
7165 | return lower->dev; | |
7166 | } | |
7167 | ||
5343da4c TY |
7168 | static u8 __netdev_upper_depth(struct net_device *dev) |
7169 | { | |
7170 | struct net_device *udev; | |
7171 | struct list_head *iter; | |
7172 | u8 max_depth = 0; | |
32b6d34f | 7173 | bool ignore; |
5343da4c TY |
7174 | |
7175 | for (iter = &dev->adj_list.upper, | |
32b6d34f | 7176 | udev = __netdev_next_upper_dev(dev, &iter, &ignore); |
5343da4c | 7177 | udev; |
32b6d34f TY |
7178 | udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { |
7179 | if (ignore) | |
7180 | continue; | |
5343da4c TY |
7181 | if (max_depth < udev->upper_level) |
7182 | max_depth = udev->upper_level; | |
7183 | } | |
7184 | ||
7185 | return max_depth; | |
7186 | } | |
7187 | ||
7188 | static u8 __netdev_lower_depth(struct net_device *dev) | |
1a3f060c DA |
7189 | { |
7190 | struct net_device *ldev; | |
7191 | struct list_head *iter; | |
5343da4c | 7192 | u8 max_depth = 0; |
32b6d34f | 7193 | bool ignore; |
1a3f060c DA |
7194 | |
7195 | for (iter = &dev->adj_list.lower, | |
32b6d34f | 7196 | ldev = __netdev_next_lower_dev(dev, &iter, &ignore); |
1a3f060c | 7197 | ldev; |
32b6d34f TY |
7198 | ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { |
7199 | if (ignore) | |
7200 | continue; | |
5343da4c TY |
7201 | if (max_depth < ldev->lower_level) |
7202 | max_depth = ldev->lower_level; | |
7203 | } | |
1a3f060c | 7204 | |
5343da4c TY |
7205 | return max_depth; |
7206 | } | |
7207 | ||
7208 | static int __netdev_update_upper_level(struct net_device *dev, void *data) | |
7209 | { | |
7210 | dev->upper_level = __netdev_upper_depth(dev) + 1; | |
7211 | return 0; | |
7212 | } | |
7213 | ||
7214 | static int __netdev_update_lower_level(struct net_device *dev, void *data) | |
7215 | { | |
7216 | dev->lower_level = __netdev_lower_depth(dev) + 1; | |
7217 | return 0; | |
7218 | } | |
7219 | ||
7220 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
7221 | int (*fn)(struct net_device *dev, | |
7222 | void *data), | |
7223 | void *data) | |
7224 | { | |
7225 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7226 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7227 | int ret, cur = 0; | |
7228 | ||
7229 | now = dev; | |
7230 | iter = &dev->adj_list.lower; | |
7231 | ||
7232 | while (1) { | |
7233 | if (now != dev) { | |
7234 | ret = fn(now, data); | |
7235 | if (ret) | |
7236 | return ret; | |
7237 | } | |
7238 | ||
7239 | next = NULL; | |
7240 | while (1) { | |
7241 | ldev = netdev_next_lower_dev_rcu(now, &iter); | |
7242 | if (!ldev) | |
7243 | break; | |
7244 | ||
7245 | next = ldev; | |
7246 | niter = &ldev->adj_list.lower; | |
7247 | dev_stack[cur] = now; | |
7248 | iter_stack[cur++] = iter; | |
7249 | break; | |
7250 | } | |
7251 | ||
7252 | if (!next) { | |
7253 | if (!cur) | |
7254 | return 0; | |
7255 | next = dev_stack[--cur]; | |
7256 | niter = iter_stack[cur]; | |
7257 | } | |
7258 | ||
7259 | now = next; | |
7260 | iter = niter; | |
1a3f060c DA |
7261 | } |
7262 | ||
7263 | return 0; | |
7264 | } | |
7265 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
7266 | ||
e001bfad | 7267 | /** |
7268 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
7269 | * lower neighbour list, RCU | |
7270 | * variant | |
7271 | * @dev: device | |
7272 | * | |
7273 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
7274 | * list. The caller must hold RCU read lock. | |
7275 | */ | |
7276 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
7277 | { | |
7278 | struct netdev_adjacent *lower; | |
7279 | ||
7280 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
7281 | struct netdev_adjacent, list); | |
7282 | if (lower) | |
7283 | return lower->private; | |
7284 | return NULL; | |
7285 | } | |
7286 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
7287 | ||
9ff162a8 JP |
7288 | /** |
7289 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
7290 | * @dev: device | |
7291 | * | |
7292 | * Find a master upper device and return pointer to it or NULL in case | |
7293 | * it's not there. The caller must hold the RCU read lock. | |
7294 | */ | |
7295 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
7296 | { | |
aa9d8560 | 7297 | struct netdev_adjacent *upper; |
9ff162a8 | 7298 | |
2f268f12 | 7299 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 7300 | struct netdev_adjacent, list); |
9ff162a8 JP |
7301 | if (upper && likely(upper->master)) |
7302 | return upper->dev; | |
7303 | return NULL; | |
7304 | } | |
7305 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
7306 | ||
0a59f3a9 | 7307 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
7308 | struct net_device *adj_dev, |
7309 | struct list_head *dev_list) | |
7310 | { | |
7311 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 7312 | |
3ee32707 VF |
7313 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
7314 | "upper_%s" : "lower_%s", adj_dev->name); | |
7315 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
7316 | linkname); | |
7317 | } | |
0a59f3a9 | 7318 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
7319 | char *name, |
7320 | struct list_head *dev_list) | |
7321 | { | |
7322 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 7323 | |
3ee32707 VF |
7324 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
7325 | "upper_%s" : "lower_%s", name); | |
7326 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
7327 | } | |
7328 | ||
7ce64c79 AF |
7329 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
7330 | struct net_device *adj_dev, | |
7331 | struct list_head *dev_list) | |
7332 | { | |
7333 | return (dev_list == &dev->adj_list.upper || | |
7334 | dev_list == &dev->adj_list.lower) && | |
7335 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
7336 | } | |
3ee32707 | 7337 | |
5d261913 VF |
7338 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
7339 | struct net_device *adj_dev, | |
7863c054 | 7340 | struct list_head *dev_list, |
402dae96 | 7341 | void *private, bool master) |
5d261913 VF |
7342 | { |
7343 | struct netdev_adjacent *adj; | |
842d67a7 | 7344 | int ret; |
5d261913 | 7345 | |
6ea29da1 | 7346 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
7347 | |
7348 | if (adj) { | |
790510d9 | 7349 | adj->ref_nr += 1; |
67b62f98 DA |
7350 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
7351 | dev->name, adj_dev->name, adj->ref_nr); | |
7352 | ||
5d261913 VF |
7353 | return 0; |
7354 | } | |
7355 | ||
7356 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
7357 | if (!adj) | |
7358 | return -ENOMEM; | |
7359 | ||
7360 | adj->dev = adj_dev; | |
7361 | adj->master = master; | |
790510d9 | 7362 | adj->ref_nr = 1; |
402dae96 | 7363 | adj->private = private; |
32b6d34f | 7364 | adj->ignore = false; |
5d261913 | 7365 | dev_hold(adj_dev); |
2f268f12 | 7366 | |
67b62f98 DA |
7367 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
7368 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 7369 | |
7ce64c79 | 7370 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 7371 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
7372 | if (ret) |
7373 | goto free_adj; | |
7374 | } | |
7375 | ||
7863c054 | 7376 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
7377 | if (master) { |
7378 | ret = sysfs_create_link(&(dev->dev.kobj), | |
7379 | &(adj_dev->dev.kobj), "master"); | |
7380 | if (ret) | |
5831d66e | 7381 | goto remove_symlinks; |
842d67a7 | 7382 | |
7863c054 | 7383 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 7384 | } else { |
7863c054 | 7385 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 7386 | } |
5d261913 VF |
7387 | |
7388 | return 0; | |
842d67a7 | 7389 | |
5831d66e | 7390 | remove_symlinks: |
7ce64c79 | 7391 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 7392 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
7393 | free_adj: |
7394 | kfree(adj); | |
974daef7 | 7395 | dev_put(adj_dev); |
842d67a7 VF |
7396 | |
7397 | return ret; | |
5d261913 VF |
7398 | } |
7399 | ||
1d143d9f | 7400 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
7401 | struct net_device *adj_dev, | |
93409033 | 7402 | u16 ref_nr, |
1d143d9f | 7403 | struct list_head *dev_list) |
5d261913 VF |
7404 | { |
7405 | struct netdev_adjacent *adj; | |
7406 | ||
67b62f98 DA |
7407 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
7408 | dev->name, adj_dev->name, ref_nr); | |
7409 | ||
6ea29da1 | 7410 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 7411 | |
2f268f12 | 7412 | if (!adj) { |
67b62f98 | 7413 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 7414 | dev->name, adj_dev->name); |
67b62f98 DA |
7415 | WARN_ON(1); |
7416 | return; | |
2f268f12 | 7417 | } |
5d261913 | 7418 | |
93409033 | 7419 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
7420 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
7421 | dev->name, adj_dev->name, ref_nr, | |
7422 | adj->ref_nr - ref_nr); | |
93409033 | 7423 | adj->ref_nr -= ref_nr; |
5d261913 VF |
7424 | return; |
7425 | } | |
7426 | ||
842d67a7 VF |
7427 | if (adj->master) |
7428 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
7429 | ||
7ce64c79 | 7430 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 7431 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 7432 | |
5d261913 | 7433 | list_del_rcu(&adj->list); |
67b62f98 | 7434 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 7435 | adj_dev->name, dev->name, adj_dev->name); |
5d261913 VF |
7436 | dev_put(adj_dev); |
7437 | kfree_rcu(adj, rcu); | |
7438 | } | |
7439 | ||
1d143d9f | 7440 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
7441 | struct net_device *upper_dev, | |
7442 | struct list_head *up_list, | |
7443 | struct list_head *down_list, | |
7444 | void *private, bool master) | |
5d261913 VF |
7445 | { |
7446 | int ret; | |
7447 | ||
790510d9 | 7448 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 7449 | private, master); |
5d261913 VF |
7450 | if (ret) |
7451 | return ret; | |
7452 | ||
790510d9 | 7453 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 7454 | private, false); |
5d261913 | 7455 | if (ret) { |
790510d9 | 7456 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
7457 | return ret; |
7458 | } | |
7459 | ||
7460 | return 0; | |
7461 | } | |
7462 | ||
1d143d9f | 7463 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
7464 | struct net_device *upper_dev, | |
93409033 | 7465 | u16 ref_nr, |
1d143d9f | 7466 | struct list_head *up_list, |
7467 | struct list_head *down_list) | |
5d261913 | 7468 | { |
93409033 AC |
7469 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
7470 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
7471 | } |
7472 | ||
1d143d9f | 7473 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
7474 | struct net_device *upper_dev, | |
7475 | void *private, bool master) | |
2f268f12 | 7476 | { |
f1170fd4 DA |
7477 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
7478 | &dev->adj_list.upper, | |
7479 | &upper_dev->adj_list.lower, | |
7480 | private, master); | |
5d261913 VF |
7481 | } |
7482 | ||
1d143d9f | 7483 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
7484 | struct net_device *upper_dev) | |
2f268f12 | 7485 | { |
93409033 | 7486 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
7487 | &dev->adj_list.upper, |
7488 | &upper_dev->adj_list.lower); | |
7489 | } | |
5d261913 | 7490 | |
9ff162a8 | 7491 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 7492 | struct net_device *upper_dev, bool master, |
42ab19ee DA |
7493 | void *upper_priv, void *upper_info, |
7494 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7495 | { |
51d0c047 DA |
7496 | struct netdev_notifier_changeupper_info changeupper_info = { |
7497 | .info = { | |
7498 | .dev = dev, | |
42ab19ee | 7499 | .extack = extack, |
51d0c047 DA |
7500 | }, |
7501 | .upper_dev = upper_dev, | |
7502 | .master = master, | |
7503 | .linking = true, | |
7504 | .upper_info = upper_info, | |
7505 | }; | |
50d629e7 | 7506 | struct net_device *master_dev; |
5d261913 | 7507 | int ret = 0; |
9ff162a8 JP |
7508 | |
7509 | ASSERT_RTNL(); | |
7510 | ||
7511 | if (dev == upper_dev) | |
7512 | return -EBUSY; | |
7513 | ||
7514 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
32b6d34f | 7515 | if (__netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
7516 | return -EBUSY; |
7517 | ||
5343da4c TY |
7518 | if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) |
7519 | return -EMLINK; | |
7520 | ||
50d629e7 | 7521 | if (!master) { |
32b6d34f | 7522 | if (__netdev_has_upper_dev(dev, upper_dev)) |
50d629e7 MM |
7523 | return -EEXIST; |
7524 | } else { | |
32b6d34f | 7525 | master_dev = __netdev_master_upper_dev_get(dev); |
50d629e7 MM |
7526 | if (master_dev) |
7527 | return master_dev == upper_dev ? -EEXIST : -EBUSY; | |
7528 | } | |
9ff162a8 | 7529 | |
51d0c047 | 7530 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
7531 | &changeupper_info.info); |
7532 | ret = notifier_to_errno(ret); | |
7533 | if (ret) | |
7534 | return ret; | |
7535 | ||
6dffb044 | 7536 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 7537 | master); |
5d261913 VF |
7538 | if (ret) |
7539 | return ret; | |
9ff162a8 | 7540 | |
51d0c047 | 7541 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
b03804e7 IS |
7542 | &changeupper_info.info); |
7543 | ret = notifier_to_errno(ret); | |
7544 | if (ret) | |
f1170fd4 | 7545 | goto rollback; |
b03804e7 | 7546 | |
5343da4c | 7547 | __netdev_update_upper_level(dev, NULL); |
32b6d34f | 7548 | __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); |
5343da4c TY |
7549 | |
7550 | __netdev_update_lower_level(upper_dev, NULL); | |
32b6d34f TY |
7551 | __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, |
7552 | NULL); | |
5343da4c | 7553 | |
9ff162a8 | 7554 | return 0; |
5d261913 | 7555 | |
f1170fd4 | 7556 | rollback: |
2f268f12 | 7557 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
7558 | |
7559 | return ret; | |
9ff162a8 JP |
7560 | } |
7561 | ||
7562 | /** | |
7563 | * netdev_upper_dev_link - Add a link to the upper device | |
7564 | * @dev: device | |
7565 | * @upper_dev: new upper device | |
7a006d59 | 7566 | * @extack: netlink extended ack |
9ff162a8 JP |
7567 | * |
7568 | * Adds a link to device which is upper to this one. The caller must hold | |
7569 | * the RTNL lock. On a failure a negative errno code is returned. | |
7570 | * On success the reference counts are adjusted and the function | |
7571 | * returns zero. | |
7572 | */ | |
7573 | int netdev_upper_dev_link(struct net_device *dev, | |
42ab19ee DA |
7574 | struct net_device *upper_dev, |
7575 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7576 | { |
42ab19ee DA |
7577 | return __netdev_upper_dev_link(dev, upper_dev, false, |
7578 | NULL, NULL, extack); | |
9ff162a8 JP |
7579 | } |
7580 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
7581 | ||
7582 | /** | |
7583 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
7584 | * @dev: device | |
7585 | * @upper_dev: new upper device | |
6dffb044 | 7586 | * @upper_priv: upper device private |
29bf24af | 7587 | * @upper_info: upper info to be passed down via notifier |
7a006d59 | 7588 | * @extack: netlink extended ack |
9ff162a8 JP |
7589 | * |
7590 | * Adds a link to device which is upper to this one. In this case, only | |
7591 | * one master upper device can be linked, although other non-master devices | |
7592 | * might be linked as well. The caller must hold the RTNL lock. | |
7593 | * On a failure a negative errno code is returned. On success the reference | |
7594 | * counts are adjusted and the function returns zero. | |
7595 | */ | |
7596 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 7597 | struct net_device *upper_dev, |
42ab19ee DA |
7598 | void *upper_priv, void *upper_info, |
7599 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7600 | { |
29bf24af | 7601 | return __netdev_upper_dev_link(dev, upper_dev, true, |
42ab19ee | 7602 | upper_priv, upper_info, extack); |
9ff162a8 JP |
7603 | } |
7604 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
7605 | ||
7606 | /** | |
7607 | * netdev_upper_dev_unlink - Removes a link to upper device | |
7608 | * @dev: device | |
7609 | * @upper_dev: new upper device | |
7610 | * | |
7611 | * Removes a link to device which is upper to this one. The caller must hold | |
7612 | * the RTNL lock. | |
7613 | */ | |
7614 | void netdev_upper_dev_unlink(struct net_device *dev, | |
7615 | struct net_device *upper_dev) | |
7616 | { | |
51d0c047 DA |
7617 | struct netdev_notifier_changeupper_info changeupper_info = { |
7618 | .info = { | |
7619 | .dev = dev, | |
7620 | }, | |
7621 | .upper_dev = upper_dev, | |
7622 | .linking = false, | |
7623 | }; | |
f4563a75 | 7624 | |
9ff162a8 JP |
7625 | ASSERT_RTNL(); |
7626 | ||
0e4ead9d | 7627 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; |
0e4ead9d | 7628 | |
51d0c047 | 7629 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
7630 | &changeupper_info.info); |
7631 | ||
2f268f12 | 7632 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 7633 | |
51d0c047 | 7634 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
0e4ead9d | 7635 | &changeupper_info.info); |
5343da4c TY |
7636 | |
7637 | __netdev_update_upper_level(dev, NULL); | |
32b6d34f | 7638 | __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); |
5343da4c TY |
7639 | |
7640 | __netdev_update_lower_level(upper_dev, NULL); | |
32b6d34f TY |
7641 | __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, |
7642 | NULL); | |
9ff162a8 JP |
7643 | } |
7644 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
7645 | ||
32b6d34f TY |
7646 | static void __netdev_adjacent_dev_set(struct net_device *upper_dev, |
7647 | struct net_device *lower_dev, | |
7648 | bool val) | |
7649 | { | |
7650 | struct netdev_adjacent *adj; | |
7651 | ||
7652 | adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); | |
7653 | if (adj) | |
7654 | adj->ignore = val; | |
7655 | ||
7656 | adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); | |
7657 | if (adj) | |
7658 | adj->ignore = val; | |
7659 | } | |
7660 | ||
7661 | static void netdev_adjacent_dev_disable(struct net_device *upper_dev, | |
7662 | struct net_device *lower_dev) | |
7663 | { | |
7664 | __netdev_adjacent_dev_set(upper_dev, lower_dev, true); | |
7665 | } | |
7666 | ||
7667 | static void netdev_adjacent_dev_enable(struct net_device *upper_dev, | |
7668 | struct net_device *lower_dev) | |
7669 | { | |
7670 | __netdev_adjacent_dev_set(upper_dev, lower_dev, false); | |
7671 | } | |
7672 | ||
7673 | int netdev_adjacent_change_prepare(struct net_device *old_dev, | |
7674 | struct net_device *new_dev, | |
7675 | struct net_device *dev, | |
7676 | struct netlink_ext_ack *extack) | |
7677 | { | |
7678 | int err; | |
7679 | ||
7680 | if (!new_dev) | |
7681 | return 0; | |
7682 | ||
7683 | if (old_dev && new_dev != old_dev) | |
7684 | netdev_adjacent_dev_disable(dev, old_dev); | |
7685 | ||
7686 | err = netdev_upper_dev_link(new_dev, dev, extack); | |
7687 | if (err) { | |
7688 | if (old_dev && new_dev != old_dev) | |
7689 | netdev_adjacent_dev_enable(dev, old_dev); | |
7690 | return err; | |
7691 | } | |
7692 | ||
7693 | return 0; | |
7694 | } | |
7695 | EXPORT_SYMBOL(netdev_adjacent_change_prepare); | |
7696 | ||
7697 | void netdev_adjacent_change_commit(struct net_device *old_dev, | |
7698 | struct net_device *new_dev, | |
7699 | struct net_device *dev) | |
7700 | { | |
7701 | if (!new_dev || !old_dev) | |
7702 | return; | |
7703 | ||
7704 | if (new_dev == old_dev) | |
7705 | return; | |
7706 | ||
7707 | netdev_adjacent_dev_enable(dev, old_dev); | |
7708 | netdev_upper_dev_unlink(old_dev, dev); | |
7709 | } | |
7710 | EXPORT_SYMBOL(netdev_adjacent_change_commit); | |
7711 | ||
7712 | void netdev_adjacent_change_abort(struct net_device *old_dev, | |
7713 | struct net_device *new_dev, | |
7714 | struct net_device *dev) | |
7715 | { | |
7716 | if (!new_dev) | |
7717 | return; | |
7718 | ||
7719 | if (old_dev && new_dev != old_dev) | |
7720 | netdev_adjacent_dev_enable(dev, old_dev); | |
7721 | ||
7722 | netdev_upper_dev_unlink(new_dev, dev); | |
7723 | } | |
7724 | EXPORT_SYMBOL(netdev_adjacent_change_abort); | |
7725 | ||
61bd3857 MS |
7726 | /** |
7727 | * netdev_bonding_info_change - Dispatch event about slave change | |
7728 | * @dev: device | |
4a26e453 | 7729 | * @bonding_info: info to dispatch |
61bd3857 MS |
7730 | * |
7731 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
7732 | * The caller must hold the RTNL lock. | |
7733 | */ | |
7734 | void netdev_bonding_info_change(struct net_device *dev, | |
7735 | struct netdev_bonding_info *bonding_info) | |
7736 | { | |
51d0c047 DA |
7737 | struct netdev_notifier_bonding_info info = { |
7738 | .info.dev = dev, | |
7739 | }; | |
61bd3857 MS |
7740 | |
7741 | memcpy(&info.bonding_info, bonding_info, | |
7742 | sizeof(struct netdev_bonding_info)); | |
51d0c047 | 7743 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, |
61bd3857 MS |
7744 | &info.info); |
7745 | } | |
7746 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
7747 | ||
2ce1ee17 | 7748 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
7749 | { |
7750 | struct netdev_adjacent *iter; | |
7751 | ||
7752 | struct net *net = dev_net(dev); | |
7753 | ||
7754 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 7755 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7756 | continue; |
7757 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7758 | &iter->dev->adj_list.lower); | |
7759 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
7760 | &dev->adj_list.upper); | |
7761 | } | |
7762 | ||
7763 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 7764 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7765 | continue; |
7766 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7767 | &iter->dev->adj_list.upper); | |
7768 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
7769 | &dev->adj_list.lower); | |
7770 | } | |
7771 | } | |
7772 | ||
2ce1ee17 | 7773 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
7774 | { |
7775 | struct netdev_adjacent *iter; | |
7776 | ||
7777 | struct net *net = dev_net(dev); | |
7778 | ||
7779 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 7780 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7781 | continue; |
7782 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
7783 | &iter->dev->adj_list.lower); | |
7784 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
7785 | &dev->adj_list.upper); | |
7786 | } | |
7787 | ||
7788 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 7789 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7790 | continue; |
7791 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
7792 | &iter->dev->adj_list.upper); | |
7793 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
7794 | &dev->adj_list.lower); | |
7795 | } | |
7796 | } | |
7797 | ||
5bb025fa | 7798 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 7799 | { |
5bb025fa | 7800 | struct netdev_adjacent *iter; |
402dae96 | 7801 | |
4c75431a AF |
7802 | struct net *net = dev_net(dev); |
7803 | ||
5bb025fa | 7804 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 7805 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 7806 | continue; |
5bb025fa VF |
7807 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
7808 | &iter->dev->adj_list.lower); | |
7809 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7810 | &iter->dev->adj_list.lower); | |
7811 | } | |
402dae96 | 7812 | |
5bb025fa | 7813 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 7814 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 7815 | continue; |
5bb025fa VF |
7816 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
7817 | &iter->dev->adj_list.upper); | |
7818 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7819 | &iter->dev->adj_list.upper); | |
7820 | } | |
402dae96 | 7821 | } |
402dae96 VF |
7822 | |
7823 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
7824 | struct net_device *lower_dev) | |
7825 | { | |
7826 | struct netdev_adjacent *lower; | |
7827 | ||
7828 | if (!lower_dev) | |
7829 | return NULL; | |
6ea29da1 | 7830 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
7831 | if (!lower) |
7832 | return NULL; | |
7833 | ||
7834 | return lower->private; | |
7835 | } | |
7836 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
7837 | ||
4085ebe8 | 7838 | |
04d48266 JP |
7839 | /** |
7840 | * netdev_lower_change - Dispatch event about lower device state change | |
7841 | * @lower_dev: device | |
7842 | * @lower_state_info: state to dispatch | |
7843 | * | |
7844 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
7845 | * The caller must hold the RTNL lock. | |
7846 | */ | |
7847 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
7848 | void *lower_state_info) | |
7849 | { | |
51d0c047 DA |
7850 | struct netdev_notifier_changelowerstate_info changelowerstate_info = { |
7851 | .info.dev = lower_dev, | |
7852 | }; | |
04d48266 JP |
7853 | |
7854 | ASSERT_RTNL(); | |
7855 | changelowerstate_info.lower_state_info = lower_state_info; | |
51d0c047 | 7856 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, |
04d48266 JP |
7857 | &changelowerstate_info.info); |
7858 | } | |
7859 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
7860 | ||
b6c40d68 PM |
7861 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
7862 | { | |
d314774c SH |
7863 | const struct net_device_ops *ops = dev->netdev_ops; |
7864 | ||
d2615bf4 | 7865 | if (ops->ndo_change_rx_flags) |
d314774c | 7866 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
7867 | } |
7868 | ||
991fb3f7 | 7869 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 7870 | { |
b536db93 | 7871 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
7872 | kuid_t uid; |
7873 | kgid_t gid; | |
1da177e4 | 7874 | |
24023451 PM |
7875 | ASSERT_RTNL(); |
7876 | ||
dad9b335 WC |
7877 | dev->flags |= IFF_PROMISC; |
7878 | dev->promiscuity += inc; | |
7879 | if (dev->promiscuity == 0) { | |
7880 | /* | |
7881 | * Avoid overflow. | |
7882 | * If inc causes overflow, untouch promisc and return error. | |
7883 | */ | |
7884 | if (inc < 0) | |
7885 | dev->flags &= ~IFF_PROMISC; | |
7886 | else { | |
7887 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
7888 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
7889 | dev->name); | |
dad9b335 WC |
7890 | return -EOVERFLOW; |
7891 | } | |
7892 | } | |
52609c0b | 7893 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
7894 | pr_info("device %s %s promiscuous mode\n", |
7895 | dev->name, | |
7896 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
7897 | if (audit_enabled) { |
7898 | current_uid_gid(&uid, &gid); | |
cdfb6b34 RGB |
7899 | audit_log(audit_context(), GFP_ATOMIC, |
7900 | AUDIT_ANOM_PROMISCUOUS, | |
7901 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
7902 | dev->name, (dev->flags & IFF_PROMISC), | |
7903 | (old_flags & IFF_PROMISC), | |
7904 | from_kuid(&init_user_ns, audit_get_loginuid(current)), | |
7905 | from_kuid(&init_user_ns, uid), | |
7906 | from_kgid(&init_user_ns, gid), | |
7907 | audit_get_sessionid(current)); | |
8192b0c4 | 7908 | } |
24023451 | 7909 | |
b6c40d68 | 7910 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 7911 | } |
991fb3f7 ND |
7912 | if (notify) |
7913 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 7914 | return 0; |
1da177e4 LT |
7915 | } |
7916 | ||
4417da66 PM |
7917 | /** |
7918 | * dev_set_promiscuity - update promiscuity count on a device | |
7919 | * @dev: device | |
7920 | * @inc: modifier | |
7921 | * | |
7922 | * Add or remove promiscuity from a device. While the count in the device | |
7923 | * remains above zero the interface remains promiscuous. Once it hits zero | |
7924 | * the device reverts back to normal filtering operation. A negative inc | |
7925 | * value is used to drop promiscuity on the device. | |
dad9b335 | 7926 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 7927 | */ |
dad9b335 | 7928 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 7929 | { |
b536db93 | 7930 | unsigned int old_flags = dev->flags; |
dad9b335 | 7931 | int err; |
4417da66 | 7932 | |
991fb3f7 | 7933 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 7934 | if (err < 0) |
dad9b335 | 7935 | return err; |
4417da66 PM |
7936 | if (dev->flags != old_flags) |
7937 | dev_set_rx_mode(dev); | |
dad9b335 | 7938 | return err; |
4417da66 | 7939 | } |
d1b19dff | 7940 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 7941 | |
991fb3f7 | 7942 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 7943 | { |
991fb3f7 | 7944 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 7945 | |
24023451 PM |
7946 | ASSERT_RTNL(); |
7947 | ||
1da177e4 | 7948 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
7949 | dev->allmulti += inc; |
7950 | if (dev->allmulti == 0) { | |
7951 | /* | |
7952 | * Avoid overflow. | |
7953 | * If inc causes overflow, untouch allmulti and return error. | |
7954 | */ | |
7955 | if (inc < 0) | |
7956 | dev->flags &= ~IFF_ALLMULTI; | |
7957 | else { | |
7958 | dev->allmulti -= inc; | |
7b6cd1ce JP |
7959 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
7960 | dev->name); | |
dad9b335 WC |
7961 | return -EOVERFLOW; |
7962 | } | |
7963 | } | |
24023451 | 7964 | if (dev->flags ^ old_flags) { |
b6c40d68 | 7965 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 7966 | dev_set_rx_mode(dev); |
991fb3f7 ND |
7967 | if (notify) |
7968 | __dev_notify_flags(dev, old_flags, | |
7969 | dev->gflags ^ old_gflags); | |
24023451 | 7970 | } |
dad9b335 | 7971 | return 0; |
4417da66 | 7972 | } |
991fb3f7 ND |
7973 | |
7974 | /** | |
7975 | * dev_set_allmulti - update allmulti count on a device | |
7976 | * @dev: device | |
7977 | * @inc: modifier | |
7978 | * | |
7979 | * Add or remove reception of all multicast frames to a device. While the | |
7980 | * count in the device remains above zero the interface remains listening | |
7981 | * to all interfaces. Once it hits zero the device reverts back to normal | |
7982 | * filtering operation. A negative @inc value is used to drop the counter | |
7983 | * when releasing a resource needing all multicasts. | |
7984 | * Return 0 if successful or a negative errno code on error. | |
7985 | */ | |
7986 | ||
7987 | int dev_set_allmulti(struct net_device *dev, int inc) | |
7988 | { | |
7989 | return __dev_set_allmulti(dev, inc, true); | |
7990 | } | |
d1b19dff | 7991 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
7992 | |
7993 | /* | |
7994 | * Upload unicast and multicast address lists to device and | |
7995 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 7996 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
7997 | * are present. |
7998 | */ | |
7999 | void __dev_set_rx_mode(struct net_device *dev) | |
8000 | { | |
d314774c SH |
8001 | const struct net_device_ops *ops = dev->netdev_ops; |
8002 | ||
4417da66 PM |
8003 | /* dev_open will call this function so the list will stay sane. */ |
8004 | if (!(dev->flags&IFF_UP)) | |
8005 | return; | |
8006 | ||
8007 | if (!netif_device_present(dev)) | |
40b77c94 | 8008 | return; |
4417da66 | 8009 | |
01789349 | 8010 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
8011 | /* Unicast addresses changes may only happen under the rtnl, |
8012 | * therefore calling __dev_set_promiscuity here is safe. | |
8013 | */ | |
32e7bfc4 | 8014 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 8015 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 8016 | dev->uc_promisc = true; |
32e7bfc4 | 8017 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 8018 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 8019 | dev->uc_promisc = false; |
4417da66 | 8020 | } |
4417da66 | 8021 | } |
01789349 JP |
8022 | |
8023 | if (ops->ndo_set_rx_mode) | |
8024 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
8025 | } |
8026 | ||
8027 | void dev_set_rx_mode(struct net_device *dev) | |
8028 | { | |
b9e40857 | 8029 | netif_addr_lock_bh(dev); |
4417da66 | 8030 | __dev_set_rx_mode(dev); |
b9e40857 | 8031 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
8032 | } |
8033 | ||
f0db275a SH |
8034 | /** |
8035 | * dev_get_flags - get flags reported to userspace | |
8036 | * @dev: device | |
8037 | * | |
8038 | * Get the combination of flag bits exported through APIs to userspace. | |
8039 | */ | |
95c96174 | 8040 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 8041 | { |
95c96174 | 8042 | unsigned int flags; |
1da177e4 LT |
8043 | |
8044 | flags = (dev->flags & ~(IFF_PROMISC | | |
8045 | IFF_ALLMULTI | | |
b00055aa SR |
8046 | IFF_RUNNING | |
8047 | IFF_LOWER_UP | | |
8048 | IFF_DORMANT)) | | |
1da177e4 LT |
8049 | (dev->gflags & (IFF_PROMISC | |
8050 | IFF_ALLMULTI)); | |
8051 | ||
b00055aa SR |
8052 | if (netif_running(dev)) { |
8053 | if (netif_oper_up(dev)) | |
8054 | flags |= IFF_RUNNING; | |
8055 | if (netif_carrier_ok(dev)) | |
8056 | flags |= IFF_LOWER_UP; | |
8057 | if (netif_dormant(dev)) | |
8058 | flags |= IFF_DORMANT; | |
8059 | } | |
1da177e4 LT |
8060 | |
8061 | return flags; | |
8062 | } | |
d1b19dff | 8063 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 8064 | |
6d040321 PM |
8065 | int __dev_change_flags(struct net_device *dev, unsigned int flags, |
8066 | struct netlink_ext_ack *extack) | |
1da177e4 | 8067 | { |
b536db93 | 8068 | unsigned int old_flags = dev->flags; |
bd380811 | 8069 | int ret; |
1da177e4 | 8070 | |
24023451 PM |
8071 | ASSERT_RTNL(); |
8072 | ||
1da177e4 LT |
8073 | /* |
8074 | * Set the flags on our device. | |
8075 | */ | |
8076 | ||
8077 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
8078 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
8079 | IFF_AUTOMEDIA)) | | |
8080 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
8081 | IFF_ALLMULTI)); | |
8082 | ||
8083 | /* | |
8084 | * Load in the correct multicast list now the flags have changed. | |
8085 | */ | |
8086 | ||
b6c40d68 PM |
8087 | if ((old_flags ^ flags) & IFF_MULTICAST) |
8088 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 8089 | |
4417da66 | 8090 | dev_set_rx_mode(dev); |
1da177e4 LT |
8091 | |
8092 | /* | |
8093 | * Have we downed the interface. We handle IFF_UP ourselves | |
8094 | * according to user attempts to set it, rather than blindly | |
8095 | * setting it. | |
8096 | */ | |
8097 | ||
8098 | ret = 0; | |
7051b88a | 8099 | if ((old_flags ^ flags) & IFF_UP) { |
8100 | if (old_flags & IFF_UP) | |
8101 | __dev_close(dev); | |
8102 | else | |
40c900aa | 8103 | ret = __dev_open(dev, extack); |
7051b88a | 8104 | } |
1da177e4 | 8105 | |
1da177e4 | 8106 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 8107 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 8108 | unsigned int old_flags = dev->flags; |
d1b19dff | 8109 | |
1da177e4 | 8110 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
8111 | |
8112 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
8113 | if (dev->flags != old_flags) | |
8114 | dev_set_rx_mode(dev); | |
1da177e4 LT |
8115 | } |
8116 | ||
8117 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 8118 | * is important. Some (broken) drivers set IFF_PROMISC, when |
8119 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
8120 | */ |
8121 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
8122 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
8123 | ||
1da177e4 | 8124 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 8125 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
8126 | } |
8127 | ||
bd380811 PM |
8128 | return ret; |
8129 | } | |
8130 | ||
a528c219 ND |
8131 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
8132 | unsigned int gchanges) | |
bd380811 PM |
8133 | { |
8134 | unsigned int changes = dev->flags ^ old_flags; | |
8135 | ||
a528c219 | 8136 | if (gchanges) |
7f294054 | 8137 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 8138 | |
bd380811 PM |
8139 | if (changes & IFF_UP) { |
8140 | if (dev->flags & IFF_UP) | |
8141 | call_netdevice_notifiers(NETDEV_UP, dev); | |
8142 | else | |
8143 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
8144 | } | |
8145 | ||
8146 | if (dev->flags & IFF_UP && | |
be9efd36 | 8147 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
51d0c047 DA |
8148 | struct netdev_notifier_change_info change_info = { |
8149 | .info = { | |
8150 | .dev = dev, | |
8151 | }, | |
8152 | .flags_changed = changes, | |
8153 | }; | |
be9efd36 | 8154 | |
51d0c047 | 8155 | call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); |
be9efd36 | 8156 | } |
bd380811 PM |
8157 | } |
8158 | ||
8159 | /** | |
8160 | * dev_change_flags - change device settings | |
8161 | * @dev: device | |
8162 | * @flags: device state flags | |
567c5e13 | 8163 | * @extack: netlink extended ack |
bd380811 PM |
8164 | * |
8165 | * Change settings on device based state flags. The flags are | |
8166 | * in the userspace exported format. | |
8167 | */ | |
567c5e13 PM |
8168 | int dev_change_flags(struct net_device *dev, unsigned int flags, |
8169 | struct netlink_ext_ack *extack) | |
bd380811 | 8170 | { |
b536db93 | 8171 | int ret; |
991fb3f7 | 8172 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 | 8173 | |
6d040321 | 8174 | ret = __dev_change_flags(dev, flags, extack); |
bd380811 PM |
8175 | if (ret < 0) |
8176 | return ret; | |
8177 | ||
991fb3f7 | 8178 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 8179 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
8180 | return ret; |
8181 | } | |
d1b19dff | 8182 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 8183 | |
f51048c3 | 8184 | int __dev_set_mtu(struct net_device *dev, int new_mtu) |
2315dc91 VF |
8185 | { |
8186 | const struct net_device_ops *ops = dev->netdev_ops; | |
8187 | ||
8188 | if (ops->ndo_change_mtu) | |
8189 | return ops->ndo_change_mtu(dev, new_mtu); | |
8190 | ||
8191 | dev->mtu = new_mtu; | |
8192 | return 0; | |
8193 | } | |
f51048c3 | 8194 | EXPORT_SYMBOL(__dev_set_mtu); |
2315dc91 | 8195 | |
f0db275a | 8196 | /** |
7a4c53be | 8197 | * dev_set_mtu_ext - Change maximum transfer unit |
f0db275a SH |
8198 | * @dev: device |
8199 | * @new_mtu: new transfer unit | |
7a4c53be | 8200 | * @extack: netlink extended ack |
f0db275a SH |
8201 | * |
8202 | * Change the maximum transfer size of the network device. | |
8203 | */ | |
7a4c53be SH |
8204 | int dev_set_mtu_ext(struct net_device *dev, int new_mtu, |
8205 | struct netlink_ext_ack *extack) | |
1da177e4 | 8206 | { |
2315dc91 | 8207 | int err, orig_mtu; |
1da177e4 LT |
8208 | |
8209 | if (new_mtu == dev->mtu) | |
8210 | return 0; | |
8211 | ||
61e84623 JW |
8212 | /* MTU must be positive, and in range */ |
8213 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
7a4c53be | 8214 | NL_SET_ERR_MSG(extack, "mtu less than device minimum"); |
1da177e4 | 8215 | return -EINVAL; |
61e84623 JW |
8216 | } |
8217 | ||
8218 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
7a4c53be | 8219 | NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); |
61e84623 JW |
8220 | return -EINVAL; |
8221 | } | |
1da177e4 LT |
8222 | |
8223 | if (!netif_device_present(dev)) | |
8224 | return -ENODEV; | |
8225 | ||
1d486bfb VF |
8226 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
8227 | err = notifier_to_errno(err); | |
8228 | if (err) | |
8229 | return err; | |
d314774c | 8230 | |
2315dc91 VF |
8231 | orig_mtu = dev->mtu; |
8232 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 8233 | |
2315dc91 | 8234 | if (!err) { |
af7d6cce SD |
8235 | err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
8236 | orig_mtu); | |
2315dc91 VF |
8237 | err = notifier_to_errno(err); |
8238 | if (err) { | |
8239 | /* setting mtu back and notifying everyone again, | |
8240 | * so that they have a chance to revert changes. | |
8241 | */ | |
8242 | __dev_set_mtu(dev, orig_mtu); | |
af7d6cce SD |
8243 | call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
8244 | new_mtu); | |
2315dc91 VF |
8245 | } |
8246 | } | |
1da177e4 LT |
8247 | return err; |
8248 | } | |
7a4c53be SH |
8249 | |
8250 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
8251 | { | |
8252 | struct netlink_ext_ack extack; | |
8253 | int err; | |
8254 | ||
a6bcfc89 | 8255 | memset(&extack, 0, sizeof(extack)); |
7a4c53be | 8256 | err = dev_set_mtu_ext(dev, new_mtu, &extack); |
a6bcfc89 | 8257 | if (err && extack._msg) |
7a4c53be SH |
8258 | net_err_ratelimited("%s: %s\n", dev->name, extack._msg); |
8259 | return err; | |
8260 | } | |
d1b19dff | 8261 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 8262 | |
6a643ddb CW |
8263 | /** |
8264 | * dev_change_tx_queue_len - Change TX queue length of a netdevice | |
8265 | * @dev: device | |
8266 | * @new_len: new tx queue length | |
8267 | */ | |
8268 | int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) | |
8269 | { | |
8270 | unsigned int orig_len = dev->tx_queue_len; | |
8271 | int res; | |
8272 | ||
8273 | if (new_len != (unsigned int)new_len) | |
8274 | return -ERANGE; | |
8275 | ||
8276 | if (new_len != orig_len) { | |
8277 | dev->tx_queue_len = new_len; | |
8278 | res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); | |
8279 | res = notifier_to_errno(res); | |
7effaf06 TT |
8280 | if (res) |
8281 | goto err_rollback; | |
8282 | res = dev_qdisc_change_tx_queue_len(dev); | |
8283 | if (res) | |
8284 | goto err_rollback; | |
6a643ddb CW |
8285 | } |
8286 | ||
8287 | return 0; | |
7effaf06 TT |
8288 | |
8289 | err_rollback: | |
8290 | netdev_err(dev, "refused to change device tx_queue_len\n"); | |
8291 | dev->tx_queue_len = orig_len; | |
8292 | return res; | |
6a643ddb CW |
8293 | } |
8294 | ||
cbda10fa VD |
8295 | /** |
8296 | * dev_set_group - Change group this device belongs to | |
8297 | * @dev: device | |
8298 | * @new_group: group this device should belong to | |
8299 | */ | |
8300 | void dev_set_group(struct net_device *dev, int new_group) | |
8301 | { | |
8302 | dev->group = new_group; | |
8303 | } | |
8304 | EXPORT_SYMBOL(dev_set_group); | |
8305 | ||
d59cdf94 PM |
8306 | /** |
8307 | * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. | |
8308 | * @dev: device | |
8309 | * @addr: new address | |
8310 | * @extack: netlink extended ack | |
8311 | */ | |
8312 | int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, | |
8313 | struct netlink_ext_ack *extack) | |
8314 | { | |
8315 | struct netdev_notifier_pre_changeaddr_info info = { | |
8316 | .info.dev = dev, | |
8317 | .info.extack = extack, | |
8318 | .dev_addr = addr, | |
8319 | }; | |
8320 | int rc; | |
8321 | ||
8322 | rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); | |
8323 | return notifier_to_errno(rc); | |
8324 | } | |
8325 | EXPORT_SYMBOL(dev_pre_changeaddr_notify); | |
8326 | ||
f0db275a SH |
8327 | /** |
8328 | * dev_set_mac_address - Change Media Access Control Address | |
8329 | * @dev: device | |
8330 | * @sa: new address | |
3a37a963 | 8331 | * @extack: netlink extended ack |
f0db275a SH |
8332 | * |
8333 | * Change the hardware (MAC) address of the device | |
8334 | */ | |
3a37a963 PM |
8335 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, |
8336 | struct netlink_ext_ack *extack) | |
1da177e4 | 8337 | { |
d314774c | 8338 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
8339 | int err; |
8340 | ||
d314774c | 8341 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
8342 | return -EOPNOTSUPP; |
8343 | if (sa->sa_family != dev->type) | |
8344 | return -EINVAL; | |
8345 | if (!netif_device_present(dev)) | |
8346 | return -ENODEV; | |
d59cdf94 PM |
8347 | err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); |
8348 | if (err) | |
8349 | return err; | |
d314774c | 8350 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
8351 | if (err) |
8352 | return err; | |
fbdeca2d | 8353 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 8354 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 8355 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 8356 | return 0; |
1da177e4 | 8357 | } |
d1b19dff | 8358 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 8359 | |
4bf84c35 JP |
8360 | /** |
8361 | * dev_change_carrier - Change device carrier | |
8362 | * @dev: device | |
691b3b7e | 8363 | * @new_carrier: new value |
4bf84c35 JP |
8364 | * |
8365 | * Change device carrier | |
8366 | */ | |
8367 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
8368 | { | |
8369 | const struct net_device_ops *ops = dev->netdev_ops; | |
8370 | ||
8371 | if (!ops->ndo_change_carrier) | |
8372 | return -EOPNOTSUPP; | |
8373 | if (!netif_device_present(dev)) | |
8374 | return -ENODEV; | |
8375 | return ops->ndo_change_carrier(dev, new_carrier); | |
8376 | } | |
8377 | EXPORT_SYMBOL(dev_change_carrier); | |
8378 | ||
66b52b0d JP |
8379 | /** |
8380 | * dev_get_phys_port_id - Get device physical port ID | |
8381 | * @dev: device | |
8382 | * @ppid: port ID | |
8383 | * | |
8384 | * Get device physical port ID | |
8385 | */ | |
8386 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 8387 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
8388 | { |
8389 | const struct net_device_ops *ops = dev->netdev_ops; | |
8390 | ||
8391 | if (!ops->ndo_get_phys_port_id) | |
8392 | return -EOPNOTSUPP; | |
8393 | return ops->ndo_get_phys_port_id(dev, ppid); | |
8394 | } | |
8395 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
8396 | ||
db24a904 DA |
8397 | /** |
8398 | * dev_get_phys_port_name - Get device physical port name | |
8399 | * @dev: device | |
8400 | * @name: port name | |
ed49e650 | 8401 | * @len: limit of bytes to copy to name |
db24a904 DA |
8402 | * |
8403 | * Get device physical port name | |
8404 | */ | |
8405 | int dev_get_phys_port_name(struct net_device *dev, | |
8406 | char *name, size_t len) | |
8407 | { | |
8408 | const struct net_device_ops *ops = dev->netdev_ops; | |
af3836df | 8409 | int err; |
db24a904 | 8410 | |
af3836df JP |
8411 | if (ops->ndo_get_phys_port_name) { |
8412 | err = ops->ndo_get_phys_port_name(dev, name, len); | |
8413 | if (err != -EOPNOTSUPP) | |
8414 | return err; | |
8415 | } | |
8416 | return devlink_compat_phys_port_name_get(dev, name, len); | |
db24a904 DA |
8417 | } |
8418 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
8419 | ||
d6abc596 FF |
8420 | /** |
8421 | * dev_get_port_parent_id - Get the device's port parent identifier | |
8422 | * @dev: network device | |
8423 | * @ppid: pointer to a storage for the port's parent identifier | |
8424 | * @recurse: allow/disallow recursion to lower devices | |
8425 | * | |
8426 | * Get the devices's port parent identifier | |
8427 | */ | |
8428 | int dev_get_port_parent_id(struct net_device *dev, | |
8429 | struct netdev_phys_item_id *ppid, | |
8430 | bool recurse) | |
8431 | { | |
8432 | const struct net_device_ops *ops = dev->netdev_ops; | |
8433 | struct netdev_phys_item_id first = { }; | |
8434 | struct net_device *lower_dev; | |
8435 | struct list_head *iter; | |
7e1146e8 JP |
8436 | int err; |
8437 | ||
8438 | if (ops->ndo_get_port_parent_id) { | |
8439 | err = ops->ndo_get_port_parent_id(dev, ppid); | |
8440 | if (err != -EOPNOTSUPP) | |
8441 | return err; | |
8442 | } | |
d6abc596 | 8443 | |
7e1146e8 JP |
8444 | err = devlink_compat_switch_id_get(dev, ppid); |
8445 | if (!err || err != -EOPNOTSUPP) | |
8446 | return err; | |
d6abc596 FF |
8447 | |
8448 | if (!recurse) | |
7e1146e8 | 8449 | return -EOPNOTSUPP; |
d6abc596 FF |
8450 | |
8451 | netdev_for_each_lower_dev(dev, lower_dev, iter) { | |
8452 | err = dev_get_port_parent_id(lower_dev, ppid, recurse); | |
8453 | if (err) | |
8454 | break; | |
8455 | if (!first.id_len) | |
8456 | first = *ppid; | |
8457 | else if (memcmp(&first, ppid, sizeof(*ppid))) | |
8458 | return -ENODATA; | |
8459 | } | |
8460 | ||
8461 | return err; | |
8462 | } | |
8463 | EXPORT_SYMBOL(dev_get_port_parent_id); | |
8464 | ||
8465 | /** | |
8466 | * netdev_port_same_parent_id - Indicate if two network devices have | |
8467 | * the same port parent identifier | |
8468 | * @a: first network device | |
8469 | * @b: second network device | |
8470 | */ | |
8471 | bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) | |
8472 | { | |
8473 | struct netdev_phys_item_id a_id = { }; | |
8474 | struct netdev_phys_item_id b_id = { }; | |
8475 | ||
8476 | if (dev_get_port_parent_id(a, &a_id, true) || | |
8477 | dev_get_port_parent_id(b, &b_id, true)) | |
8478 | return false; | |
8479 | ||
8480 | return netdev_phys_item_id_same(&a_id, &b_id); | |
8481 | } | |
8482 | EXPORT_SYMBOL(netdev_port_same_parent_id); | |
8483 | ||
d746d707 AK |
8484 | /** |
8485 | * dev_change_proto_down - update protocol port state information | |
8486 | * @dev: device | |
8487 | * @proto_down: new value | |
8488 | * | |
8489 | * This info can be used by switch drivers to set the phys state of the | |
8490 | * port. | |
8491 | */ | |
8492 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
8493 | { | |
8494 | const struct net_device_ops *ops = dev->netdev_ops; | |
8495 | ||
8496 | if (!ops->ndo_change_proto_down) | |
8497 | return -EOPNOTSUPP; | |
8498 | if (!netif_device_present(dev)) | |
8499 | return -ENODEV; | |
8500 | return ops->ndo_change_proto_down(dev, proto_down); | |
8501 | } | |
8502 | EXPORT_SYMBOL(dev_change_proto_down); | |
8503 | ||
b5899679 AR |
8504 | /** |
8505 | * dev_change_proto_down_generic - generic implementation for | |
8506 | * ndo_change_proto_down that sets carrier according to | |
8507 | * proto_down. | |
8508 | * | |
8509 | * @dev: device | |
8510 | * @proto_down: new value | |
8511 | */ | |
8512 | int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) | |
8513 | { | |
8514 | if (proto_down) | |
8515 | netif_carrier_off(dev); | |
8516 | else | |
8517 | netif_carrier_on(dev); | |
8518 | dev->proto_down = proto_down; | |
8519 | return 0; | |
8520 | } | |
8521 | EXPORT_SYMBOL(dev_change_proto_down_generic); | |
8522 | ||
a25717d2 JK |
8523 | u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op, |
8524 | enum bpf_netdev_command cmd) | |
d67b9cd2 | 8525 | { |
a25717d2 | 8526 | struct netdev_bpf xdp; |
d67b9cd2 | 8527 | |
a25717d2 JK |
8528 | if (!bpf_op) |
8529 | return 0; | |
118b4aa2 | 8530 | |
a25717d2 JK |
8531 | memset(&xdp, 0, sizeof(xdp)); |
8532 | xdp.command = cmd; | |
118b4aa2 | 8533 | |
a25717d2 JK |
8534 | /* Query must always succeed. */ |
8535 | WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG); | |
58038695 | 8536 | |
6b867589 | 8537 | return xdp.prog_id; |
d67b9cd2 DB |
8538 | } |
8539 | ||
f4e63525 | 8540 | static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op, |
32d60277 | 8541 | struct netlink_ext_ack *extack, u32 flags, |
d67b9cd2 DB |
8542 | struct bpf_prog *prog) |
8543 | { | |
f4e63525 | 8544 | struct netdev_bpf xdp; |
d67b9cd2 DB |
8545 | |
8546 | memset(&xdp, 0, sizeof(xdp)); | |
ee5d032f JK |
8547 | if (flags & XDP_FLAGS_HW_MODE) |
8548 | xdp.command = XDP_SETUP_PROG_HW; | |
8549 | else | |
8550 | xdp.command = XDP_SETUP_PROG; | |
d67b9cd2 | 8551 | xdp.extack = extack; |
32d60277 | 8552 | xdp.flags = flags; |
d67b9cd2 DB |
8553 | xdp.prog = prog; |
8554 | ||
f4e63525 | 8555 | return bpf_op(dev, &xdp); |
d67b9cd2 DB |
8556 | } |
8557 | ||
bd0b2e7f JK |
8558 | static void dev_xdp_uninstall(struct net_device *dev) |
8559 | { | |
8560 | struct netdev_bpf xdp; | |
8561 | bpf_op_t ndo_bpf; | |
8562 | ||
8563 | /* Remove generic XDP */ | |
8564 | WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL)); | |
8565 | ||
8566 | /* Remove from the driver */ | |
8567 | ndo_bpf = dev->netdev_ops->ndo_bpf; | |
8568 | if (!ndo_bpf) | |
8569 | return; | |
8570 | ||
a25717d2 JK |
8571 | memset(&xdp, 0, sizeof(xdp)); |
8572 | xdp.command = XDP_QUERY_PROG; | |
8573 | WARN_ON(ndo_bpf(dev, &xdp)); | |
8574 | if (xdp.prog_id) | |
8575 | WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, | |
8576 | NULL)); | |
bd0b2e7f | 8577 | |
a25717d2 JK |
8578 | /* Remove HW offload */ |
8579 | memset(&xdp, 0, sizeof(xdp)); | |
8580 | xdp.command = XDP_QUERY_PROG_HW; | |
8581 | if (!ndo_bpf(dev, &xdp) && xdp.prog_id) | |
8582 | WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, | |
8583 | NULL)); | |
bd0b2e7f JK |
8584 | } |
8585 | ||
a7862b45 BB |
8586 | /** |
8587 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
8588 | * @dev: device | |
b5d60989 | 8589 | * @extack: netlink extended ack |
a7862b45 | 8590 | * @fd: new program fd or negative value to clear |
85de8576 | 8591 | * @flags: xdp-related flags |
a7862b45 BB |
8592 | * |
8593 | * Set or clear a bpf program for a device | |
8594 | */ | |
ddf9f970 JK |
8595 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, |
8596 | int fd, u32 flags) | |
a7862b45 BB |
8597 | { |
8598 | const struct net_device_ops *ops = dev->netdev_ops; | |
a25717d2 | 8599 | enum bpf_netdev_command query; |
a7862b45 | 8600 | struct bpf_prog *prog = NULL; |
f4e63525 | 8601 | bpf_op_t bpf_op, bpf_chk; |
9ee963d6 | 8602 | bool offload; |
a7862b45 BB |
8603 | int err; |
8604 | ||
85de8576 DB |
8605 | ASSERT_RTNL(); |
8606 | ||
9ee963d6 JK |
8607 | offload = flags & XDP_FLAGS_HW_MODE; |
8608 | query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG; | |
a25717d2 | 8609 | |
f4e63525 | 8610 | bpf_op = bpf_chk = ops->ndo_bpf; |
01dde20c MF |
8611 | if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) { |
8612 | NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode"); | |
0489df9a | 8613 | return -EOPNOTSUPP; |
01dde20c | 8614 | } |
f4e63525 JK |
8615 | if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE)) |
8616 | bpf_op = generic_xdp_install; | |
8617 | if (bpf_op == bpf_chk) | |
8618 | bpf_chk = generic_xdp_install; | |
b5cdae32 | 8619 | |
a7862b45 | 8620 | if (fd >= 0) { |
c14a9f63 MM |
8621 | u32 prog_id; |
8622 | ||
9ee963d6 | 8623 | if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) { |
01dde20c | 8624 | NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time"); |
d67b9cd2 | 8625 | return -EEXIST; |
01dde20c | 8626 | } |
c14a9f63 MM |
8627 | |
8628 | prog_id = __dev_xdp_query(dev, bpf_op, query); | |
8629 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && prog_id) { | |
01dde20c | 8630 | NL_SET_ERR_MSG(extack, "XDP program already attached"); |
d67b9cd2 | 8631 | return -EBUSY; |
01dde20c | 8632 | } |
85de8576 | 8633 | |
288b3de5 JK |
8634 | prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, |
8635 | bpf_op == ops->ndo_bpf); | |
a7862b45 BB |
8636 | if (IS_ERR(prog)) |
8637 | return PTR_ERR(prog); | |
441a3303 | 8638 | |
9ee963d6 | 8639 | if (!offload && bpf_prog_is_dev_bound(prog->aux)) { |
441a3303 JK |
8640 | NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported"); |
8641 | bpf_prog_put(prog); | |
8642 | return -EINVAL; | |
8643 | } | |
c14a9f63 | 8644 | |
aefc3e72 JK |
8645 | /* prog->aux->id may be 0 for orphaned device-bound progs */ |
8646 | if (prog->aux->id && prog->aux->id == prog_id) { | |
c14a9f63 MM |
8647 | bpf_prog_put(prog); |
8648 | return 0; | |
8649 | } | |
8650 | } else { | |
8651 | if (!__dev_xdp_query(dev, bpf_op, query)) | |
8652 | return 0; | |
a7862b45 BB |
8653 | } |
8654 | ||
f4e63525 | 8655 | err = dev_xdp_install(dev, bpf_op, extack, flags, prog); |
a7862b45 BB |
8656 | if (err < 0 && prog) |
8657 | bpf_prog_put(prog); | |
8658 | ||
8659 | return err; | |
8660 | } | |
a7862b45 | 8661 | |
1da177e4 LT |
8662 | /** |
8663 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 8664 | * @net: the applicable net namespace |
1da177e4 LT |
8665 | * |
8666 | * Returns a suitable unique value for a new device interface | |
8667 | * number. The caller must hold the rtnl semaphore or the | |
8668 | * dev_base_lock to be sure it remains unique. | |
8669 | */ | |
881d966b | 8670 | static int dev_new_index(struct net *net) |
1da177e4 | 8671 | { |
aa79e66e | 8672 | int ifindex = net->ifindex; |
f4563a75 | 8673 | |
1da177e4 LT |
8674 | for (;;) { |
8675 | if (++ifindex <= 0) | |
8676 | ifindex = 1; | |
881d966b | 8677 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 8678 | return net->ifindex = ifindex; |
1da177e4 LT |
8679 | } |
8680 | } | |
8681 | ||
1da177e4 | 8682 | /* Delayed registration/unregisteration */ |
3b5b34fd | 8683 | static LIST_HEAD(net_todo_list); |
200b916f | 8684 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 8685 | |
6f05f629 | 8686 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 8687 | { |
1da177e4 | 8688 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 8689 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
8690 | } |
8691 | ||
9b5e383c | 8692 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 8693 | { |
e93737b0 | 8694 | struct net_device *dev, *tmp; |
5cde2829 | 8695 | LIST_HEAD(close_head); |
9b5e383c | 8696 | |
93ee31f1 DL |
8697 | BUG_ON(dev_boot_phase); |
8698 | ASSERT_RTNL(); | |
8699 | ||
e93737b0 | 8700 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 8701 | /* Some devices call without registering |
e93737b0 KK |
8702 | * for initialization unwind. Remove those |
8703 | * devices and proceed with the remaining. | |
9b5e383c ED |
8704 | */ |
8705 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
8706 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
8707 | dev->name, dev); | |
93ee31f1 | 8708 | |
9b5e383c | 8709 | WARN_ON(1); |
e93737b0 KK |
8710 | list_del(&dev->unreg_list); |
8711 | continue; | |
9b5e383c | 8712 | } |
449f4544 | 8713 | dev->dismantle = true; |
9b5e383c | 8714 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 8715 | } |
93ee31f1 | 8716 | |
44345724 | 8717 | /* If device is running, close it first. */ |
5cde2829 EB |
8718 | list_for_each_entry(dev, head, unreg_list) |
8719 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 8720 | dev_close_many(&close_head, true); |
93ee31f1 | 8721 | |
44345724 | 8722 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
8723 | /* And unlink it from device chain. */ |
8724 | unlist_netdevice(dev); | |
93ee31f1 | 8725 | |
9b5e383c ED |
8726 | dev->reg_state = NETREG_UNREGISTERING; |
8727 | } | |
41852497 | 8728 | flush_all_backlogs(); |
93ee31f1 DL |
8729 | |
8730 | synchronize_net(); | |
8731 | ||
9b5e383c | 8732 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
8733 | struct sk_buff *skb = NULL; |
8734 | ||
9b5e383c ED |
8735 | /* Shutdown queueing discipline. */ |
8736 | dev_shutdown(dev); | |
93ee31f1 | 8737 | |
bd0b2e7f | 8738 | dev_xdp_uninstall(dev); |
93ee31f1 | 8739 | |
9b5e383c | 8740 | /* Notify protocols, that we are about to destroy |
eb13da1a | 8741 | * this device. They should clean all the things. |
8742 | */ | |
9b5e383c | 8743 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
93ee31f1 | 8744 | |
395eea6c MB |
8745 | if (!dev->rtnl_link_ops || |
8746 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
3d3ea5af | 8747 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, |
38e01b30 | 8748 | GFP_KERNEL, NULL, 0); |
395eea6c | 8749 | |
9b5e383c ED |
8750 | /* |
8751 | * Flush the unicast and multicast chains | |
8752 | */ | |
a748ee24 | 8753 | dev_uc_flush(dev); |
22bedad3 | 8754 | dev_mc_flush(dev); |
93ee31f1 | 8755 | |
36fbf1e5 | 8756 | netdev_name_node_alt_flush(dev); |
ff927412 JP |
8757 | netdev_name_node_free(dev->name_node); |
8758 | ||
9b5e383c ED |
8759 | if (dev->netdev_ops->ndo_uninit) |
8760 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 8761 | |
395eea6c MB |
8762 | if (skb) |
8763 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 8764 | |
9ff162a8 JP |
8765 | /* Notifier chain MUST detach us all upper devices. */ |
8766 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
0f524a80 | 8767 | WARN_ON(netdev_has_any_lower_dev(dev)); |
93ee31f1 | 8768 | |
9b5e383c ED |
8769 | /* Remove entries from kobject tree */ |
8770 | netdev_unregister_kobject(dev); | |
024e9679 AD |
8771 | #ifdef CONFIG_XPS |
8772 | /* Remove XPS queueing entries */ | |
8773 | netif_reset_xps_queues_gt(dev, 0); | |
8774 | #endif | |
9b5e383c | 8775 | } |
93ee31f1 | 8776 | |
850a545b | 8777 | synchronize_net(); |
395264d5 | 8778 | |
a5ee1551 | 8779 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
8780 | dev_put(dev); |
8781 | } | |
8782 | ||
8783 | static void rollback_registered(struct net_device *dev) | |
8784 | { | |
8785 | LIST_HEAD(single); | |
8786 | ||
8787 | list_add(&dev->unreg_list, &single); | |
8788 | rollback_registered_many(&single); | |
ceaaec98 | 8789 | list_del(&single); |
93ee31f1 DL |
8790 | } |
8791 | ||
fd867d51 JW |
8792 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
8793 | struct net_device *upper, netdev_features_t features) | |
8794 | { | |
8795 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
8796 | netdev_features_t feature; | |
5ba3f7d6 | 8797 | int feature_bit; |
fd867d51 | 8798 | |
3b89ea9c | 8799 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 8800 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
8801 | if (!(upper->wanted_features & feature) |
8802 | && (features & feature)) { | |
8803 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
8804 | &feature, upper->name); | |
8805 | features &= ~feature; | |
8806 | } | |
8807 | } | |
8808 | ||
8809 | return features; | |
8810 | } | |
8811 | ||
8812 | static void netdev_sync_lower_features(struct net_device *upper, | |
8813 | struct net_device *lower, netdev_features_t features) | |
8814 | { | |
8815 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
8816 | netdev_features_t feature; | |
5ba3f7d6 | 8817 | int feature_bit; |
fd867d51 | 8818 | |
3b89ea9c | 8819 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 8820 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
8821 | if (!(features & feature) && (lower->features & feature)) { |
8822 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
8823 | &feature, lower->name); | |
8824 | lower->wanted_features &= ~feature; | |
8825 | netdev_update_features(lower); | |
8826 | ||
8827 | if (unlikely(lower->features & feature)) | |
8828 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
8829 | &feature, lower->name); | |
8830 | } | |
8831 | } | |
8832 | } | |
8833 | ||
c8f44aff MM |
8834 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
8835 | netdev_features_t features) | |
b63365a2 | 8836 | { |
57422dc5 MM |
8837 | /* Fix illegal checksum combinations */ |
8838 | if ((features & NETIF_F_HW_CSUM) && | |
8839 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 8840 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
8841 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
8842 | } | |
8843 | ||
b63365a2 | 8844 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 8845 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 8846 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 8847 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
8848 | } |
8849 | ||
ec5f0615 PS |
8850 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
8851 | !(features & NETIF_F_IP_CSUM)) { | |
8852 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
8853 | features &= ~NETIF_F_TSO; | |
8854 | features &= ~NETIF_F_TSO_ECN; | |
8855 | } | |
8856 | ||
8857 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
8858 | !(features & NETIF_F_IPV6_CSUM)) { | |
8859 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
8860 | features &= ~NETIF_F_TSO6; | |
8861 | } | |
8862 | ||
b1dc497b AD |
8863 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
8864 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
8865 | features &= ~NETIF_F_TSO_MANGLEID; | |
8866 | ||
31d8b9e0 BH |
8867 | /* TSO ECN requires that TSO is present as well. */ |
8868 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
8869 | features &= ~NETIF_F_TSO_ECN; | |
8870 | ||
212b573f MM |
8871 | /* Software GSO depends on SG. */ |
8872 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 8873 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
8874 | features &= ~NETIF_F_GSO; |
8875 | } | |
8876 | ||
802ab55a AD |
8877 | /* GSO partial features require GSO partial be set */ |
8878 | if ((features & dev->gso_partial_features) && | |
8879 | !(features & NETIF_F_GSO_PARTIAL)) { | |
8880 | netdev_dbg(dev, | |
8881 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
8882 | features &= ~dev->gso_partial_features; | |
8883 | } | |
8884 | ||
fb1f5f79 MC |
8885 | if (!(features & NETIF_F_RXCSUM)) { |
8886 | /* NETIF_F_GRO_HW implies doing RXCSUM since every packet | |
8887 | * successfully merged by hardware must also have the | |
8888 | * checksum verified by hardware. If the user does not | |
8889 | * want to enable RXCSUM, logically, we should disable GRO_HW. | |
8890 | */ | |
8891 | if (features & NETIF_F_GRO_HW) { | |
8892 | netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); | |
8893 | features &= ~NETIF_F_GRO_HW; | |
8894 | } | |
8895 | } | |
8896 | ||
de8d5ab2 GP |
8897 | /* LRO/HW-GRO features cannot be combined with RX-FCS */ |
8898 | if (features & NETIF_F_RXFCS) { | |
8899 | if (features & NETIF_F_LRO) { | |
8900 | netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); | |
8901 | features &= ~NETIF_F_LRO; | |
8902 | } | |
8903 | ||
8904 | if (features & NETIF_F_GRO_HW) { | |
8905 | netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); | |
8906 | features &= ~NETIF_F_GRO_HW; | |
8907 | } | |
e6c6a929 GP |
8908 | } |
8909 | ||
b63365a2 HX |
8910 | return features; |
8911 | } | |
b63365a2 | 8912 | |
6cb6a27c | 8913 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 8914 | { |
fd867d51 | 8915 | struct net_device *upper, *lower; |
c8f44aff | 8916 | netdev_features_t features; |
fd867d51 | 8917 | struct list_head *iter; |
e7868a85 | 8918 | int err = -1; |
5455c699 | 8919 | |
87267485 MM |
8920 | ASSERT_RTNL(); |
8921 | ||
5455c699 MM |
8922 | features = netdev_get_wanted_features(dev); |
8923 | ||
8924 | if (dev->netdev_ops->ndo_fix_features) | |
8925 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
8926 | ||
8927 | /* driver might be less strict about feature dependencies */ | |
8928 | features = netdev_fix_features(dev, features); | |
8929 | ||
fd867d51 JW |
8930 | /* some features can't be enabled if they're off an an upper device */ |
8931 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
8932 | features = netdev_sync_upper_features(dev, upper, features); | |
8933 | ||
5455c699 | 8934 | if (dev->features == features) |
e7868a85 | 8935 | goto sync_lower; |
5455c699 | 8936 | |
c8f44aff MM |
8937 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
8938 | &dev->features, &features); | |
5455c699 MM |
8939 | |
8940 | if (dev->netdev_ops->ndo_set_features) | |
8941 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
8942 | else |
8943 | err = 0; | |
5455c699 | 8944 | |
6cb6a27c | 8945 | if (unlikely(err < 0)) { |
5455c699 | 8946 | netdev_err(dev, |
c8f44aff MM |
8947 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
8948 | err, &features, &dev->features); | |
17b85d29 NA |
8949 | /* return non-0 since some features might have changed and |
8950 | * it's better to fire a spurious notification than miss it | |
8951 | */ | |
8952 | return -1; | |
6cb6a27c MM |
8953 | } |
8954 | ||
e7868a85 | 8955 | sync_lower: |
fd867d51 JW |
8956 | /* some features must be disabled on lower devices when disabled |
8957 | * on an upper device (think: bonding master or bridge) | |
8958 | */ | |
8959 | netdev_for_each_lower_dev(dev, lower, iter) | |
8960 | netdev_sync_lower_features(dev, lower, features); | |
8961 | ||
ae847f40 SD |
8962 | if (!err) { |
8963 | netdev_features_t diff = features ^ dev->features; | |
8964 | ||
8965 | if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
8966 | /* udp_tunnel_{get,drop}_rx_info both need | |
8967 | * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the | |
8968 | * device, or they won't do anything. | |
8969 | * Thus we need to update dev->features | |
8970 | * *before* calling udp_tunnel_get_rx_info, | |
8971 | * but *after* calling udp_tunnel_drop_rx_info. | |
8972 | */ | |
8973 | if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
8974 | dev->features = features; | |
8975 | udp_tunnel_get_rx_info(dev); | |
8976 | } else { | |
8977 | udp_tunnel_drop_rx_info(dev); | |
8978 | } | |
8979 | } | |
8980 | ||
9daae9bd GP |
8981 | if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { |
8982 | if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { | |
8983 | dev->features = features; | |
8984 | err |= vlan_get_rx_ctag_filter_info(dev); | |
8985 | } else { | |
8986 | vlan_drop_rx_ctag_filter_info(dev); | |
8987 | } | |
8988 | } | |
8989 | ||
8990 | if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { | |
8991 | if (features & NETIF_F_HW_VLAN_STAG_FILTER) { | |
8992 | dev->features = features; | |
8993 | err |= vlan_get_rx_stag_filter_info(dev); | |
8994 | } else { | |
8995 | vlan_drop_rx_stag_filter_info(dev); | |
8996 | } | |
8997 | } | |
8998 | ||
6cb6a27c | 8999 | dev->features = features; |
ae847f40 | 9000 | } |
6cb6a27c | 9001 | |
e7868a85 | 9002 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
9003 | } |
9004 | ||
afe12cc8 MM |
9005 | /** |
9006 | * netdev_update_features - recalculate device features | |
9007 | * @dev: the device to check | |
9008 | * | |
9009 | * Recalculate dev->features set and send notifications if it | |
9010 | * has changed. Should be called after driver or hardware dependent | |
9011 | * conditions might have changed that influence the features. | |
9012 | */ | |
6cb6a27c MM |
9013 | void netdev_update_features(struct net_device *dev) |
9014 | { | |
9015 | if (__netdev_update_features(dev)) | |
9016 | netdev_features_change(dev); | |
5455c699 MM |
9017 | } |
9018 | EXPORT_SYMBOL(netdev_update_features); | |
9019 | ||
afe12cc8 MM |
9020 | /** |
9021 | * netdev_change_features - recalculate device features | |
9022 | * @dev: the device to check | |
9023 | * | |
9024 | * Recalculate dev->features set and send notifications even | |
9025 | * if they have not changed. Should be called instead of | |
9026 | * netdev_update_features() if also dev->vlan_features might | |
9027 | * have changed to allow the changes to be propagated to stacked | |
9028 | * VLAN devices. | |
9029 | */ | |
9030 | void netdev_change_features(struct net_device *dev) | |
9031 | { | |
9032 | __netdev_update_features(dev); | |
9033 | netdev_features_change(dev); | |
9034 | } | |
9035 | EXPORT_SYMBOL(netdev_change_features); | |
9036 | ||
fc4a7489 PM |
9037 | /** |
9038 | * netif_stacked_transfer_operstate - transfer operstate | |
9039 | * @rootdev: the root or lower level device to transfer state from | |
9040 | * @dev: the device to transfer operstate to | |
9041 | * | |
9042 | * Transfer operational state from root to device. This is normally | |
9043 | * called when a stacking relationship exists between the root | |
9044 | * device and the device(a leaf device). | |
9045 | */ | |
9046 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
9047 | struct net_device *dev) | |
9048 | { | |
9049 | if (rootdev->operstate == IF_OPER_DORMANT) | |
9050 | netif_dormant_on(dev); | |
9051 | else | |
9052 | netif_dormant_off(dev); | |
9053 | ||
0575c86b ZS |
9054 | if (netif_carrier_ok(rootdev)) |
9055 | netif_carrier_on(dev); | |
9056 | else | |
9057 | netif_carrier_off(dev); | |
fc4a7489 PM |
9058 | } |
9059 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
9060 | ||
1b4bf461 ED |
9061 | static int netif_alloc_rx_queues(struct net_device *dev) |
9062 | { | |
1b4bf461 | 9063 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 9064 | struct netdev_rx_queue *rx; |
10595902 | 9065 | size_t sz = count * sizeof(*rx); |
e817f856 | 9066 | int err = 0; |
1b4bf461 | 9067 | |
bd25fa7b | 9068 | BUG_ON(count < 1); |
1b4bf461 | 9069 | |
dcda9b04 | 9070 | rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
9071 | if (!rx) |
9072 | return -ENOMEM; | |
9073 | ||
bd25fa7b TH |
9074 | dev->_rx = rx; |
9075 | ||
e817f856 | 9076 | for (i = 0; i < count; i++) { |
fe822240 | 9077 | rx[i].dev = dev; |
e817f856 JDB |
9078 | |
9079 | /* XDP RX-queue setup */ | |
9080 | err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i); | |
9081 | if (err < 0) | |
9082 | goto err_rxq_info; | |
9083 | } | |
1b4bf461 | 9084 | return 0; |
e817f856 JDB |
9085 | |
9086 | err_rxq_info: | |
9087 | /* Rollback successful reg's and free other resources */ | |
9088 | while (i--) | |
9089 | xdp_rxq_info_unreg(&rx[i].xdp_rxq); | |
141b52a9 | 9090 | kvfree(dev->_rx); |
e817f856 JDB |
9091 | dev->_rx = NULL; |
9092 | return err; | |
9093 | } | |
9094 | ||
9095 | static void netif_free_rx_queues(struct net_device *dev) | |
9096 | { | |
9097 | unsigned int i, count = dev->num_rx_queues; | |
e817f856 JDB |
9098 | |
9099 | /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ | |
9100 | if (!dev->_rx) | |
9101 | return; | |
9102 | ||
e817f856 | 9103 | for (i = 0; i < count; i++) |
82aaff2f JK |
9104 | xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); |
9105 | ||
9106 | kvfree(dev->_rx); | |
1b4bf461 ED |
9107 | } |
9108 | ||
aa942104 CG |
9109 | static void netdev_init_one_queue(struct net_device *dev, |
9110 | struct netdev_queue *queue, void *_unused) | |
9111 | { | |
9112 | /* Initialize queue lock */ | |
9113 | spin_lock_init(&queue->_xmit_lock); | |
ab92d68f | 9114 | lockdep_set_class(&queue->_xmit_lock, &dev->qdisc_xmit_lock_key); |
aa942104 | 9115 | queue->xmit_lock_owner = -1; |
b236da69 | 9116 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 9117 | queue->dev = dev; |
114cf580 TH |
9118 | #ifdef CONFIG_BQL |
9119 | dql_init(&queue->dql, HZ); | |
9120 | #endif | |
aa942104 CG |
9121 | } |
9122 | ||
60877a32 ED |
9123 | static void netif_free_tx_queues(struct net_device *dev) |
9124 | { | |
4cb28970 | 9125 | kvfree(dev->_tx); |
60877a32 ED |
9126 | } |
9127 | ||
e6484930 TH |
9128 | static int netif_alloc_netdev_queues(struct net_device *dev) |
9129 | { | |
9130 | unsigned int count = dev->num_tx_queues; | |
9131 | struct netdev_queue *tx; | |
60877a32 | 9132 | size_t sz = count * sizeof(*tx); |
e6484930 | 9133 | |
d339727c ED |
9134 | if (count < 1 || count > 0xffff) |
9135 | return -EINVAL; | |
62b5942a | 9136 | |
dcda9b04 | 9137 | tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
9138 | if (!tx) |
9139 | return -ENOMEM; | |
9140 | ||
e6484930 | 9141 | dev->_tx = tx; |
1d24eb48 | 9142 | |
e6484930 TH |
9143 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
9144 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
9145 | |
9146 | return 0; | |
e6484930 TH |
9147 | } |
9148 | ||
a2029240 DV |
9149 | void netif_tx_stop_all_queues(struct net_device *dev) |
9150 | { | |
9151 | unsigned int i; | |
9152 | ||
9153 | for (i = 0; i < dev->num_tx_queues; i++) { | |
9154 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 9155 | |
a2029240 DV |
9156 | netif_tx_stop_queue(txq); |
9157 | } | |
9158 | } | |
9159 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
9160 | ||
ab92d68f TY |
9161 | static void netdev_register_lockdep_key(struct net_device *dev) |
9162 | { | |
9163 | lockdep_register_key(&dev->qdisc_tx_busylock_key); | |
9164 | lockdep_register_key(&dev->qdisc_running_key); | |
9165 | lockdep_register_key(&dev->qdisc_xmit_lock_key); | |
9166 | lockdep_register_key(&dev->addr_list_lock_key); | |
9167 | } | |
9168 | ||
9169 | static void netdev_unregister_lockdep_key(struct net_device *dev) | |
9170 | { | |
9171 | lockdep_unregister_key(&dev->qdisc_tx_busylock_key); | |
9172 | lockdep_unregister_key(&dev->qdisc_running_key); | |
9173 | lockdep_unregister_key(&dev->qdisc_xmit_lock_key); | |
9174 | lockdep_unregister_key(&dev->addr_list_lock_key); | |
9175 | } | |
9176 | ||
9177 | void netdev_update_lockdep_key(struct net_device *dev) | |
9178 | { | |
9179 | struct netdev_queue *queue; | |
9180 | int i; | |
9181 | ||
9182 | lockdep_unregister_key(&dev->qdisc_xmit_lock_key); | |
9183 | lockdep_unregister_key(&dev->addr_list_lock_key); | |
9184 | ||
9185 | lockdep_register_key(&dev->qdisc_xmit_lock_key); | |
9186 | lockdep_register_key(&dev->addr_list_lock_key); | |
9187 | ||
9188 | lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key); | |
9189 | for (i = 0; i < dev->num_tx_queues; i++) { | |
9190 | queue = netdev_get_tx_queue(dev, i); | |
9191 | ||
9192 | lockdep_set_class(&queue->_xmit_lock, | |
9193 | &dev->qdisc_xmit_lock_key); | |
9194 | } | |
9195 | } | |
9196 | EXPORT_SYMBOL(netdev_update_lockdep_key); | |
9197 | ||
1da177e4 LT |
9198 | /** |
9199 | * register_netdevice - register a network device | |
9200 | * @dev: device to register | |
9201 | * | |
9202 | * Take a completed network device structure and add it to the kernel | |
9203 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
9204 | * chain. 0 is returned on success. A negative errno code is returned | |
9205 | * on a failure to set up the device, or if the name is a duplicate. | |
9206 | * | |
9207 | * Callers must hold the rtnl semaphore. You may want | |
9208 | * register_netdev() instead of this. | |
9209 | * | |
9210 | * BUGS: | |
9211 | * The locking appears insufficient to guarantee two parallel registers | |
9212 | * will not get the same name. | |
9213 | */ | |
9214 | ||
9215 | int register_netdevice(struct net_device *dev) | |
9216 | { | |
1da177e4 | 9217 | int ret; |
d314774c | 9218 | struct net *net = dev_net(dev); |
1da177e4 | 9219 | |
e283de3a FF |
9220 | BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < |
9221 | NETDEV_FEATURE_COUNT); | |
1da177e4 LT |
9222 | BUG_ON(dev_boot_phase); |
9223 | ASSERT_RTNL(); | |
9224 | ||
b17a7c17 SH |
9225 | might_sleep(); |
9226 | ||
1da177e4 LT |
9227 | /* When net_device's are persistent, this will be fatal. */ |
9228 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 9229 | BUG_ON(!net); |
1da177e4 | 9230 | |
f1f28aa3 | 9231 | spin_lock_init(&dev->addr_list_lock); |
ab92d68f | 9232 | lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key); |
1da177e4 | 9233 | |
828de4f6 | 9234 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
9235 | if (ret < 0) |
9236 | goto out; | |
9237 | ||
9077f052 | 9238 | ret = -ENOMEM; |
ff927412 JP |
9239 | dev->name_node = netdev_name_node_head_alloc(dev); |
9240 | if (!dev->name_node) | |
9241 | goto out; | |
9242 | ||
1da177e4 | 9243 | /* Init, if this function is available */ |
d314774c SH |
9244 | if (dev->netdev_ops->ndo_init) { |
9245 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
9246 | if (ret) { |
9247 | if (ret > 0) | |
9248 | ret = -EIO; | |
90833aa4 | 9249 | goto out; |
1da177e4 LT |
9250 | } |
9251 | } | |
4ec93edb | 9252 | |
f646968f PM |
9253 | if (((dev->hw_features | dev->features) & |
9254 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
9255 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
9256 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
9257 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
9258 | ret = -EINVAL; | |
9259 | goto err_uninit; | |
9260 | } | |
9261 | ||
9c7dafbf PE |
9262 | ret = -EBUSY; |
9263 | if (!dev->ifindex) | |
9264 | dev->ifindex = dev_new_index(net); | |
9265 | else if (__dev_get_by_index(net, dev->ifindex)) | |
9266 | goto err_uninit; | |
9267 | ||
5455c699 MM |
9268 | /* Transfer changeable features to wanted_features and enable |
9269 | * software offloads (GSO and GRO). | |
9270 | */ | |
9271 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f | 9272 | dev->features |= NETIF_F_SOFT_FEATURES; |
d764a122 SD |
9273 | |
9274 | if (dev->netdev_ops->ndo_udp_tunnel_add) { | |
9275 | dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
9276 | dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
9277 | } | |
9278 | ||
14d1232f | 9279 | dev->wanted_features = dev->features & dev->hw_features; |
1da177e4 | 9280 | |
cbc53e08 | 9281 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 9282 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 9283 | |
7f348a60 AD |
9284 | /* If IPv4 TCP segmentation offload is supported we should also |
9285 | * allow the device to enable segmenting the frame with the option | |
9286 | * of ignoring a static IP ID value. This doesn't enable the | |
9287 | * feature itself but allows the user to enable it later. | |
9288 | */ | |
cbc53e08 AD |
9289 | if (dev->hw_features & NETIF_F_TSO) |
9290 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
9291 | if (dev->vlan_features & NETIF_F_TSO) |
9292 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
9293 | if (dev->mpls_features & NETIF_F_TSO) | |
9294 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
9295 | if (dev->hw_enc_features & NETIF_F_TSO) | |
9296 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 9297 | |
1180e7d6 | 9298 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 9299 | */ |
1180e7d6 | 9300 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 9301 | |
ee579677 PS |
9302 | /* Make NETIF_F_SG inheritable to tunnel devices. |
9303 | */ | |
802ab55a | 9304 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 9305 | |
0d89d203 SH |
9306 | /* Make NETIF_F_SG inheritable to MPLS. |
9307 | */ | |
9308 | dev->mpls_features |= NETIF_F_SG; | |
9309 | ||
7ffbe3fd JB |
9310 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
9311 | ret = notifier_to_errno(ret); | |
9312 | if (ret) | |
9313 | goto err_uninit; | |
9314 | ||
8b41d188 | 9315 | ret = netdev_register_kobject(dev); |
b17a7c17 | 9316 | if (ret) |
7ce1b0ed | 9317 | goto err_uninit; |
b17a7c17 SH |
9318 | dev->reg_state = NETREG_REGISTERED; |
9319 | ||
6cb6a27c | 9320 | __netdev_update_features(dev); |
8e9b59b2 | 9321 | |
1da177e4 LT |
9322 | /* |
9323 | * Default initial state at registry is that the | |
9324 | * device is present. | |
9325 | */ | |
9326 | ||
9327 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
9328 | ||
8f4cccbb BH |
9329 | linkwatch_init_dev(dev); |
9330 | ||
1da177e4 | 9331 | dev_init_scheduler(dev); |
1da177e4 | 9332 | dev_hold(dev); |
ce286d32 | 9333 | list_netdevice(dev); |
7bf23575 | 9334 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 9335 | |
948b337e JP |
9336 | /* If the device has permanent device address, driver should |
9337 | * set dev_addr and also addr_assign_type should be set to | |
9338 | * NET_ADDR_PERM (default value). | |
9339 | */ | |
9340 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
9341 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
9342 | ||
1da177e4 | 9343 | /* Notify protocols, that a new device appeared. */ |
056925ab | 9344 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 9345 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
9346 | if (ret) { |
9347 | rollback_registered(dev); | |
10cc514f SAK |
9348 | rcu_barrier(); |
9349 | ||
93ee31f1 DL |
9350 | dev->reg_state = NETREG_UNREGISTERED; |
9351 | } | |
d90a909e EB |
9352 | /* |
9353 | * Prevent userspace races by waiting until the network | |
9354 | * device is fully setup before sending notifications. | |
9355 | */ | |
a2835763 PM |
9356 | if (!dev->rtnl_link_ops || |
9357 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 9358 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
9359 | |
9360 | out: | |
9361 | return ret; | |
7ce1b0ed HX |
9362 | |
9363 | err_uninit: | |
ff927412 JP |
9364 | if (dev->name_node) |
9365 | netdev_name_node_free(dev->name_node); | |
d314774c SH |
9366 | if (dev->netdev_ops->ndo_uninit) |
9367 | dev->netdev_ops->ndo_uninit(dev); | |
cf124db5 DM |
9368 | if (dev->priv_destructor) |
9369 | dev->priv_destructor(dev); | |
7ce1b0ed | 9370 | goto out; |
1da177e4 | 9371 | } |
d1b19dff | 9372 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 9373 | |
937f1ba5 BH |
9374 | /** |
9375 | * init_dummy_netdev - init a dummy network device for NAPI | |
9376 | * @dev: device to init | |
9377 | * | |
9378 | * This takes a network device structure and initialize the minimum | |
9379 | * amount of fields so it can be used to schedule NAPI polls without | |
9380 | * registering a full blown interface. This is to be used by drivers | |
9381 | * that need to tie several hardware interfaces to a single NAPI | |
9382 | * poll scheduler due to HW limitations. | |
9383 | */ | |
9384 | int init_dummy_netdev(struct net_device *dev) | |
9385 | { | |
9386 | /* Clear everything. Note we don't initialize spinlocks | |
9387 | * are they aren't supposed to be taken by any of the | |
9388 | * NAPI code and this dummy netdev is supposed to be | |
9389 | * only ever used for NAPI polls | |
9390 | */ | |
9391 | memset(dev, 0, sizeof(struct net_device)); | |
9392 | ||
9393 | /* make sure we BUG if trying to hit standard | |
9394 | * register/unregister code path | |
9395 | */ | |
9396 | dev->reg_state = NETREG_DUMMY; | |
9397 | ||
937f1ba5 BH |
9398 | /* NAPI wants this */ |
9399 | INIT_LIST_HEAD(&dev->napi_list); | |
9400 | ||
9401 | /* a dummy interface is started by default */ | |
9402 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
9403 | set_bit(__LINK_STATE_START, &dev->state); | |
9404 | ||
35edfdc7 JE |
9405 | /* napi_busy_loop stats accounting wants this */ |
9406 | dev_net_set(dev, &init_net); | |
9407 | ||
29b4433d ED |
9408 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
9409 | * because users of this 'device' dont need to change | |
9410 | * its refcount. | |
9411 | */ | |
9412 | ||
937f1ba5 BH |
9413 | return 0; |
9414 | } | |
9415 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
9416 | ||
9417 | ||
1da177e4 LT |
9418 | /** |
9419 | * register_netdev - register a network device | |
9420 | * @dev: device to register | |
9421 | * | |
9422 | * Take a completed network device structure and add it to the kernel | |
9423 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
9424 | * chain. 0 is returned on success. A negative errno code is returned | |
9425 | * on a failure to set up the device, or if the name is a duplicate. | |
9426 | * | |
38b4da38 | 9427 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
9428 | * and expands the device name if you passed a format string to |
9429 | * alloc_netdev. | |
9430 | */ | |
9431 | int register_netdev(struct net_device *dev) | |
9432 | { | |
9433 | int err; | |
9434 | ||
b0f3debc KT |
9435 | if (rtnl_lock_killable()) |
9436 | return -EINTR; | |
1da177e4 | 9437 | err = register_netdevice(dev); |
1da177e4 LT |
9438 | rtnl_unlock(); |
9439 | return err; | |
9440 | } | |
9441 | EXPORT_SYMBOL(register_netdev); | |
9442 | ||
29b4433d ED |
9443 | int netdev_refcnt_read(const struct net_device *dev) |
9444 | { | |
9445 | int i, refcnt = 0; | |
9446 | ||
9447 | for_each_possible_cpu(i) | |
9448 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
9449 | return refcnt; | |
9450 | } | |
9451 | EXPORT_SYMBOL(netdev_refcnt_read); | |
9452 | ||
2c53040f | 9453 | /** |
1da177e4 | 9454 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 9455 | * @dev: target net_device |
1da177e4 LT |
9456 | * |
9457 | * This is called when unregistering network devices. | |
9458 | * | |
9459 | * Any protocol or device that holds a reference should register | |
9460 | * for netdevice notification, and cleanup and put back the | |
9461 | * reference if they receive an UNREGISTER event. | |
9462 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 9463 | * call dev_put. |
1da177e4 LT |
9464 | */ |
9465 | static void netdev_wait_allrefs(struct net_device *dev) | |
9466 | { | |
9467 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 9468 | int refcnt; |
1da177e4 | 9469 | |
e014debe ED |
9470 | linkwatch_forget_dev(dev); |
9471 | ||
1da177e4 | 9472 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
9473 | refcnt = netdev_refcnt_read(dev); |
9474 | ||
9475 | while (refcnt != 0) { | |
1da177e4 | 9476 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 9477 | rtnl_lock(); |
1da177e4 LT |
9478 | |
9479 | /* Rebroadcast unregister notification */ | |
056925ab | 9480 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 9481 | |
748e2d93 | 9482 | __rtnl_unlock(); |
0115e8e3 | 9483 | rcu_barrier(); |
748e2d93 ED |
9484 | rtnl_lock(); |
9485 | ||
1da177e4 LT |
9486 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
9487 | &dev->state)) { | |
9488 | /* We must not have linkwatch events | |
9489 | * pending on unregister. If this | |
9490 | * happens, we simply run the queue | |
9491 | * unscheduled, resulting in a noop | |
9492 | * for this device. | |
9493 | */ | |
9494 | linkwatch_run_queue(); | |
9495 | } | |
9496 | ||
6756ae4b | 9497 | __rtnl_unlock(); |
1da177e4 LT |
9498 | |
9499 | rebroadcast_time = jiffies; | |
9500 | } | |
9501 | ||
9502 | msleep(250); | |
9503 | ||
29b4433d ED |
9504 | refcnt = netdev_refcnt_read(dev); |
9505 | ||
d7c04b05 | 9506 | if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
9507 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
9508 | dev->name, refcnt); | |
1da177e4 LT |
9509 | warning_time = jiffies; |
9510 | } | |
9511 | } | |
9512 | } | |
9513 | ||
9514 | /* The sequence is: | |
9515 | * | |
9516 | * rtnl_lock(); | |
9517 | * ... | |
9518 | * register_netdevice(x1); | |
9519 | * register_netdevice(x2); | |
9520 | * ... | |
9521 | * unregister_netdevice(y1); | |
9522 | * unregister_netdevice(y2); | |
9523 | * ... | |
9524 | * rtnl_unlock(); | |
9525 | * free_netdev(y1); | |
9526 | * free_netdev(y2); | |
9527 | * | |
58ec3b4d | 9528 | * We are invoked by rtnl_unlock(). |
1da177e4 | 9529 | * This allows us to deal with problems: |
b17a7c17 | 9530 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
9531 | * without deadlocking with linkwatch via keventd. |
9532 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
9533 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
9534 | * |
9535 | * We must not return until all unregister events added during | |
9536 | * the interval the lock was held have been completed. | |
1da177e4 | 9537 | */ |
1da177e4 LT |
9538 | void netdev_run_todo(void) |
9539 | { | |
626ab0e6 | 9540 | struct list_head list; |
1da177e4 | 9541 | |
1da177e4 | 9542 | /* Snapshot list, allow later requests */ |
626ab0e6 | 9543 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
9544 | |
9545 | __rtnl_unlock(); | |
626ab0e6 | 9546 | |
0115e8e3 ED |
9547 | |
9548 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
9549 | if (!list_empty(&list)) |
9550 | rcu_barrier(); | |
9551 | ||
1da177e4 LT |
9552 | while (!list_empty(&list)) { |
9553 | struct net_device *dev | |
e5e26d75 | 9554 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
9555 | list_del(&dev->todo_list); |
9556 | ||
b17a7c17 | 9557 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 9558 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
9559 | dev->name, dev->reg_state); |
9560 | dump_stack(); | |
9561 | continue; | |
9562 | } | |
1da177e4 | 9563 | |
b17a7c17 | 9564 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 9565 | |
b17a7c17 | 9566 | netdev_wait_allrefs(dev); |
1da177e4 | 9567 | |
b17a7c17 | 9568 | /* paranoia */ |
29b4433d | 9569 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
9570 | BUG_ON(!list_empty(&dev->ptype_all)); |
9571 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
9572 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
9573 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
330c7272 | 9574 | #if IS_ENABLED(CONFIG_DECNET) |
547b792c | 9575 | WARN_ON(dev->dn_ptr); |
330c7272 | 9576 | #endif |
cf124db5 DM |
9577 | if (dev->priv_destructor) |
9578 | dev->priv_destructor(dev); | |
9579 | if (dev->needs_free_netdev) | |
9580 | free_netdev(dev); | |
9093bbb2 | 9581 | |
50624c93 EB |
9582 | /* Report a network device has been unregistered */ |
9583 | rtnl_lock(); | |
9584 | dev_net(dev)->dev_unreg_count--; | |
9585 | __rtnl_unlock(); | |
9586 | wake_up(&netdev_unregistering_wq); | |
9587 | ||
9093bbb2 SH |
9588 | /* Free network device */ |
9589 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 9590 | } |
1da177e4 LT |
9591 | } |
9592 | ||
9256645a JW |
9593 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
9594 | * all the same fields in the same order as net_device_stats, with only | |
9595 | * the type differing, but rtnl_link_stats64 may have additional fields | |
9596 | * at the end for newer counters. | |
3cfde79c | 9597 | */ |
77a1abf5 ED |
9598 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
9599 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
9600 | { |
9601 | #if BITS_PER_LONG == 64 | |
9256645a | 9602 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
9af9959e | 9603 | memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); |
9256645a JW |
9604 | /* zero out counters that only exist in rtnl_link_stats64 */ |
9605 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
9606 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 9607 | #else |
9256645a | 9608 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
9609 | const unsigned long *src = (const unsigned long *)netdev_stats; |
9610 | u64 *dst = (u64 *)stats64; | |
9611 | ||
9256645a | 9612 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
9613 | for (i = 0; i < n; i++) |
9614 | dst[i] = src[i]; | |
9256645a JW |
9615 | /* zero out counters that only exist in rtnl_link_stats64 */ |
9616 | memset((char *)stats64 + n * sizeof(u64), 0, | |
9617 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
9618 | #endif |
9619 | } | |
77a1abf5 | 9620 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 9621 | |
eeda3fd6 SH |
9622 | /** |
9623 | * dev_get_stats - get network device statistics | |
9624 | * @dev: device to get statistics from | |
28172739 | 9625 | * @storage: place to store stats |
eeda3fd6 | 9626 | * |
d7753516 BH |
9627 | * Get network statistics from device. Return @storage. |
9628 | * The device driver may provide its own method by setting | |
9629 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
9630 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 9631 | */ |
d7753516 BH |
9632 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
9633 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 9634 | { |
eeda3fd6 SH |
9635 | const struct net_device_ops *ops = dev->netdev_ops; |
9636 | ||
28172739 ED |
9637 | if (ops->ndo_get_stats64) { |
9638 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
9639 | ops->ndo_get_stats64(dev, storage); |
9640 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 9641 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
9642 | } else { |
9643 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 9644 | } |
6f64ec74 ED |
9645 | storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); |
9646 | storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); | |
9647 | storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); | |
28172739 | 9648 | return storage; |
c45d286e | 9649 | } |
eeda3fd6 | 9650 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 9651 | |
24824a09 | 9652 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 9653 | { |
24824a09 | 9654 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 9655 | |
24824a09 ED |
9656 | #ifdef CONFIG_NET_CLS_ACT |
9657 | if (queue) | |
9658 | return queue; | |
9659 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
9660 | if (!queue) | |
9661 | return NULL; | |
9662 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 9663 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
9664 | queue->qdisc_sleeping = &noop_qdisc; |
9665 | rcu_assign_pointer(dev->ingress_queue, queue); | |
9666 | #endif | |
9667 | return queue; | |
bb949fbd DM |
9668 | } |
9669 | ||
2c60db03 ED |
9670 | static const struct ethtool_ops default_ethtool_ops; |
9671 | ||
d07d7507 SG |
9672 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
9673 | const struct ethtool_ops *ops) | |
9674 | { | |
9675 | if (dev->ethtool_ops == &default_ethtool_ops) | |
9676 | dev->ethtool_ops = ops; | |
9677 | } | |
9678 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
9679 | ||
74d332c1 ED |
9680 | void netdev_freemem(struct net_device *dev) |
9681 | { | |
9682 | char *addr = (char *)dev - dev->padded; | |
9683 | ||
4cb28970 | 9684 | kvfree(addr); |
74d332c1 ED |
9685 | } |
9686 | ||
1da177e4 | 9687 | /** |
722c9a0c | 9688 | * alloc_netdev_mqs - allocate network device |
9689 | * @sizeof_priv: size of private data to allocate space for | |
9690 | * @name: device name format string | |
9691 | * @name_assign_type: origin of device name | |
9692 | * @setup: callback to initialize device | |
9693 | * @txqs: the number of TX subqueues to allocate | |
9694 | * @rxqs: the number of RX subqueues to allocate | |
9695 | * | |
9696 | * Allocates a struct net_device with private data area for driver use | |
9697 | * and performs basic initialization. Also allocates subqueue structs | |
9698 | * for each queue on the device. | |
1da177e4 | 9699 | */ |
36909ea4 | 9700 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 9701 | unsigned char name_assign_type, |
36909ea4 TH |
9702 | void (*setup)(struct net_device *), |
9703 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 9704 | { |
1da177e4 | 9705 | struct net_device *dev; |
52a59bd5 | 9706 | unsigned int alloc_size; |
1ce8e7b5 | 9707 | struct net_device *p; |
1da177e4 | 9708 | |
b6fe17d6 SH |
9709 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
9710 | ||
36909ea4 | 9711 | if (txqs < 1) { |
7b6cd1ce | 9712 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
9713 | return NULL; |
9714 | } | |
9715 | ||
36909ea4 | 9716 | if (rxqs < 1) { |
7b6cd1ce | 9717 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
9718 | return NULL; |
9719 | } | |
36909ea4 | 9720 | |
fd2ea0a7 | 9721 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
9722 | if (sizeof_priv) { |
9723 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 9724 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
9725 | alloc_size += sizeof_priv; |
9726 | } | |
9727 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 9728 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 9729 | |
dcda9b04 | 9730 | p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
62b5942a | 9731 | if (!p) |
1da177e4 | 9732 | return NULL; |
1da177e4 | 9733 | |
1ce8e7b5 | 9734 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 9735 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 9736 | |
29b4433d ED |
9737 | dev->pcpu_refcnt = alloc_percpu(int); |
9738 | if (!dev->pcpu_refcnt) | |
74d332c1 | 9739 | goto free_dev; |
ab9c73cc | 9740 | |
ab9c73cc | 9741 | if (dev_addr_init(dev)) |
29b4433d | 9742 | goto free_pcpu; |
ab9c73cc | 9743 | |
22bedad3 | 9744 | dev_mc_init(dev); |
a748ee24 | 9745 | dev_uc_init(dev); |
ccffad25 | 9746 | |
c346dca1 | 9747 | dev_net_set(dev, &init_net); |
1da177e4 | 9748 | |
ab92d68f TY |
9749 | netdev_register_lockdep_key(dev); |
9750 | ||
8d3bdbd5 | 9751 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 9752 | dev->gso_max_segs = GSO_MAX_SEGS; |
5343da4c TY |
9753 | dev->upper_level = 1; |
9754 | dev->lower_level = 1; | |
8d3bdbd5 | 9755 | |
8d3bdbd5 DM |
9756 | INIT_LIST_HEAD(&dev->napi_list); |
9757 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 9758 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 9759 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
9760 | INIT_LIST_HEAD(&dev->adj_list.upper); |
9761 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
9762 | INIT_LIST_HEAD(&dev->ptype_all); |
9763 | INIT_LIST_HEAD(&dev->ptype_specific); | |
59cc1f61 JK |
9764 | #ifdef CONFIG_NET_SCHED |
9765 | hash_init(dev->qdisc_hash); | |
9766 | #endif | |
02875878 | 9767 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
9768 | setup(dev); |
9769 | ||
a813104d | 9770 | if (!dev->tx_queue_len) { |
f84bb1ea | 9771 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 9772 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 9773 | } |
906470c1 | 9774 | |
36909ea4 TH |
9775 | dev->num_tx_queues = txqs; |
9776 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 9777 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 9778 | goto free_all; |
e8a0464c | 9779 | |
36909ea4 TH |
9780 | dev->num_rx_queues = rxqs; |
9781 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 9782 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 9783 | goto free_all; |
0a9627f2 | 9784 | |
1da177e4 | 9785 | strcpy(dev->name, name); |
c835a677 | 9786 | dev->name_assign_type = name_assign_type; |
cbda10fa | 9787 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
9788 | if (!dev->ethtool_ops) |
9789 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 PN |
9790 | |
9791 | nf_hook_ingress_init(dev); | |
9792 | ||
1da177e4 | 9793 | return dev; |
ab9c73cc | 9794 | |
8d3bdbd5 DM |
9795 | free_all: |
9796 | free_netdev(dev); | |
9797 | return NULL; | |
9798 | ||
29b4433d ED |
9799 | free_pcpu: |
9800 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
9801 | free_dev: |
9802 | netdev_freemem(dev); | |
ab9c73cc | 9803 | return NULL; |
1da177e4 | 9804 | } |
36909ea4 | 9805 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
9806 | |
9807 | /** | |
722c9a0c | 9808 | * free_netdev - free network device |
9809 | * @dev: device | |
1da177e4 | 9810 | * |
722c9a0c | 9811 | * This function does the last stage of destroying an allocated device |
9812 | * interface. The reference to the device object is released. If this | |
9813 | * is the last reference then it will be freed.Must be called in process | |
9814 | * context. | |
1da177e4 LT |
9815 | */ |
9816 | void free_netdev(struct net_device *dev) | |
9817 | { | |
d565b0a1 HX |
9818 | struct napi_struct *p, *n; |
9819 | ||
93d05d4a | 9820 | might_sleep(); |
60877a32 | 9821 | netif_free_tx_queues(dev); |
e817f856 | 9822 | netif_free_rx_queues(dev); |
e8a0464c | 9823 | |
33d480ce | 9824 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 9825 | |
f001fde5 JP |
9826 | /* Flush device addresses */ |
9827 | dev_addr_flush(dev); | |
9828 | ||
d565b0a1 HX |
9829 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
9830 | netif_napi_del(p); | |
9831 | ||
29b4433d ED |
9832 | free_percpu(dev->pcpu_refcnt); |
9833 | dev->pcpu_refcnt = NULL; | |
9834 | ||
ab92d68f TY |
9835 | netdev_unregister_lockdep_key(dev); |
9836 | ||
3041a069 | 9837 | /* Compatibility with error handling in drivers */ |
1da177e4 | 9838 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 9839 | netdev_freemem(dev); |
1da177e4 LT |
9840 | return; |
9841 | } | |
9842 | ||
9843 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
9844 | dev->reg_state = NETREG_RELEASED; | |
9845 | ||
43cb76d9 GKH |
9846 | /* will free via device release */ |
9847 | put_device(&dev->dev); | |
1da177e4 | 9848 | } |
d1b19dff | 9849 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 9850 | |
f0db275a SH |
9851 | /** |
9852 | * synchronize_net - Synchronize with packet receive processing | |
9853 | * | |
9854 | * Wait for packets currently being received to be done. | |
9855 | * Does not block later packets from starting. | |
9856 | */ | |
4ec93edb | 9857 | void synchronize_net(void) |
1da177e4 LT |
9858 | { |
9859 | might_sleep(); | |
be3fc413 ED |
9860 | if (rtnl_is_locked()) |
9861 | synchronize_rcu_expedited(); | |
9862 | else | |
9863 | synchronize_rcu(); | |
1da177e4 | 9864 | } |
d1b19dff | 9865 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
9866 | |
9867 | /** | |
44a0873d | 9868 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 9869 | * @dev: device |
44a0873d | 9870 | * @head: list |
6ebfbc06 | 9871 | * |
1da177e4 | 9872 | * This function shuts down a device interface and removes it |
d59b54b1 | 9873 | * from the kernel tables. |
44a0873d | 9874 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
9875 | * |
9876 | * Callers must hold the rtnl semaphore. You may want | |
9877 | * unregister_netdev() instead of this. | |
9878 | */ | |
9879 | ||
44a0873d | 9880 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 9881 | { |
a6620712 HX |
9882 | ASSERT_RTNL(); |
9883 | ||
44a0873d | 9884 | if (head) { |
9fdce099 | 9885 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
9886 | } else { |
9887 | rollback_registered(dev); | |
9888 | /* Finish processing unregister after unlock */ | |
9889 | net_set_todo(dev); | |
9890 | } | |
1da177e4 | 9891 | } |
44a0873d | 9892 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 9893 | |
9b5e383c ED |
9894 | /** |
9895 | * unregister_netdevice_many - unregister many devices | |
9896 | * @head: list of devices | |
87757a91 ED |
9897 | * |
9898 | * Note: As most callers use a stack allocated list_head, | |
9899 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
9900 | */ |
9901 | void unregister_netdevice_many(struct list_head *head) | |
9902 | { | |
9903 | struct net_device *dev; | |
9904 | ||
9905 | if (!list_empty(head)) { | |
9906 | rollback_registered_many(head); | |
9907 | list_for_each_entry(dev, head, unreg_list) | |
9908 | net_set_todo(dev); | |
87757a91 | 9909 | list_del(head); |
9b5e383c ED |
9910 | } |
9911 | } | |
63c8099d | 9912 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 9913 | |
1da177e4 LT |
9914 | /** |
9915 | * unregister_netdev - remove device from the kernel | |
9916 | * @dev: device | |
9917 | * | |
9918 | * This function shuts down a device interface and removes it | |
d59b54b1 | 9919 | * from the kernel tables. |
1da177e4 LT |
9920 | * |
9921 | * This is just a wrapper for unregister_netdevice that takes | |
9922 | * the rtnl semaphore. In general you want to use this and not | |
9923 | * unregister_netdevice. | |
9924 | */ | |
9925 | void unregister_netdev(struct net_device *dev) | |
9926 | { | |
9927 | rtnl_lock(); | |
9928 | unregister_netdevice(dev); | |
9929 | rtnl_unlock(); | |
9930 | } | |
1da177e4 LT |
9931 | EXPORT_SYMBOL(unregister_netdev); |
9932 | ||
ce286d32 EB |
9933 | /** |
9934 | * dev_change_net_namespace - move device to different nethost namespace | |
9935 | * @dev: device | |
9936 | * @net: network namespace | |
9937 | * @pat: If not NULL name pattern to try if the current device name | |
9938 | * is already taken in the destination network namespace. | |
9939 | * | |
9940 | * This function shuts down a device interface and moves it | |
9941 | * to a new network namespace. On success 0 is returned, on | |
9942 | * a failure a netagive errno code is returned. | |
9943 | * | |
9944 | * Callers must hold the rtnl semaphore. | |
9945 | */ | |
9946 | ||
9947 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
9948 | { | |
38e01b30 | 9949 | int err, new_nsid, new_ifindex; |
ce286d32 EB |
9950 | |
9951 | ASSERT_RTNL(); | |
9952 | ||
9953 | /* Don't allow namespace local devices to be moved. */ | |
9954 | err = -EINVAL; | |
9955 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
9956 | goto out; | |
9957 | ||
9958 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
9959 | if (dev->reg_state != NETREG_REGISTERED) |
9960 | goto out; | |
9961 | ||
9962 | /* Get out if there is nothing todo */ | |
9963 | err = 0; | |
878628fb | 9964 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
9965 | goto out; |
9966 | ||
9967 | /* Pick the destination device name, and ensure | |
9968 | * we can use it in the destination network namespace. | |
9969 | */ | |
9970 | err = -EEXIST; | |
d9031024 | 9971 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
9972 | /* We get here if we can't use the current device name */ |
9973 | if (!pat) | |
9974 | goto out; | |
7892bd08 LR |
9975 | err = dev_get_valid_name(net, dev, pat); |
9976 | if (err < 0) | |
ce286d32 EB |
9977 | goto out; |
9978 | } | |
9979 | ||
9980 | /* | |
9981 | * And now a mini version of register_netdevice unregister_netdevice. | |
9982 | */ | |
9983 | ||
9984 | /* If device is running close it first. */ | |
9b772652 | 9985 | dev_close(dev); |
ce286d32 EB |
9986 | |
9987 | /* And unlink it from device chain */ | |
ce286d32 EB |
9988 | unlist_netdevice(dev); |
9989 | ||
9990 | synchronize_net(); | |
9991 | ||
9992 | /* Shutdown queueing discipline. */ | |
9993 | dev_shutdown(dev); | |
9994 | ||
9995 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 9996 | * this device. They should clean all the things. |
9997 | * | |
9998 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
9999 | * This is wanted because this way 8021q and macvlan know | |
10000 | * the device is just moving and can keep their slaves up. | |
10001 | */ | |
ce286d32 | 10002 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 | 10003 | rcu_barrier(); |
38e01b30 | 10004 | |
d4e4fdf9 | 10005 | new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); |
38e01b30 ND |
10006 | /* If there is an ifindex conflict assign a new one */ |
10007 | if (__dev_get_by_index(net, dev->ifindex)) | |
10008 | new_ifindex = dev_new_index(net); | |
10009 | else | |
10010 | new_ifindex = dev->ifindex; | |
10011 | ||
10012 | rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, | |
10013 | new_ifindex); | |
ce286d32 EB |
10014 | |
10015 | /* | |
10016 | * Flush the unicast and multicast chains | |
10017 | */ | |
a748ee24 | 10018 | dev_uc_flush(dev); |
22bedad3 | 10019 | dev_mc_flush(dev); |
ce286d32 | 10020 | |
4e66ae2e SH |
10021 | /* Send a netdev-removed uevent to the old namespace */ |
10022 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 10023 | netdev_adjacent_del_links(dev); |
4e66ae2e | 10024 | |
ce286d32 | 10025 | /* Actually switch the network namespace */ |
c346dca1 | 10026 | dev_net_set(dev, net); |
38e01b30 | 10027 | dev->ifindex = new_ifindex; |
ce286d32 | 10028 | |
4e66ae2e SH |
10029 | /* Send a netdev-add uevent to the new namespace */ |
10030 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 10031 | netdev_adjacent_add_links(dev); |
4e66ae2e | 10032 | |
8b41d188 | 10033 | /* Fixup kobjects */ |
a1b3f594 | 10034 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 10035 | WARN_ON(err); |
ce286d32 EB |
10036 | |
10037 | /* Add the device back in the hashes */ | |
10038 | list_netdevice(dev); | |
10039 | ||
10040 | /* Notify protocols, that a new device appeared. */ | |
10041 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
10042 | ||
d90a909e EB |
10043 | /* |
10044 | * Prevent userspace races by waiting until the network | |
10045 | * device is fully setup before sending notifications. | |
10046 | */ | |
7f294054 | 10047 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 10048 | |
ce286d32 EB |
10049 | synchronize_net(); |
10050 | err = 0; | |
10051 | out: | |
10052 | return err; | |
10053 | } | |
463d0183 | 10054 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 10055 | |
f0bf90de | 10056 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
10057 | { |
10058 | struct sk_buff **list_skb; | |
1da177e4 | 10059 | struct sk_buff *skb; |
f0bf90de | 10060 | unsigned int cpu; |
97d8b6e3 | 10061 | struct softnet_data *sd, *oldsd, *remsd = NULL; |
1da177e4 | 10062 | |
1da177e4 LT |
10063 | local_irq_disable(); |
10064 | cpu = smp_processor_id(); | |
10065 | sd = &per_cpu(softnet_data, cpu); | |
10066 | oldsd = &per_cpu(softnet_data, oldcpu); | |
10067 | ||
10068 | /* Find end of our completion_queue. */ | |
10069 | list_skb = &sd->completion_queue; | |
10070 | while (*list_skb) | |
10071 | list_skb = &(*list_skb)->next; | |
10072 | /* Append completion queue from offline CPU. */ | |
10073 | *list_skb = oldsd->completion_queue; | |
10074 | oldsd->completion_queue = NULL; | |
10075 | ||
1da177e4 | 10076 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
10077 | if (oldsd->output_queue) { |
10078 | *sd->output_queue_tailp = oldsd->output_queue; | |
10079 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
10080 | oldsd->output_queue = NULL; | |
10081 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
10082 | } | |
ac64da0b ED |
10083 | /* Append NAPI poll list from offline CPU, with one exception : |
10084 | * process_backlog() must be called by cpu owning percpu backlog. | |
10085 | * We properly handle process_queue & input_pkt_queue later. | |
10086 | */ | |
10087 | while (!list_empty(&oldsd->poll_list)) { | |
10088 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
10089 | struct napi_struct, | |
10090 | poll_list); | |
10091 | ||
10092 | list_del_init(&napi->poll_list); | |
10093 | if (napi->poll == process_backlog) | |
10094 | napi->state = 0; | |
10095 | else | |
10096 | ____napi_schedule(sd, napi); | |
264524d5 | 10097 | } |
1da177e4 LT |
10098 | |
10099 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
10100 | local_irq_enable(); | |
10101 | ||
773fc8f6 | 10102 | #ifdef CONFIG_RPS |
10103 | remsd = oldsd->rps_ipi_list; | |
10104 | oldsd->rps_ipi_list = NULL; | |
10105 | #endif | |
10106 | /* send out pending IPI's on offline CPU */ | |
10107 | net_rps_send_ipi(remsd); | |
10108 | ||
1da177e4 | 10109 | /* Process offline CPU's input_pkt_queue */ |
76cc8b13 | 10110 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 10111 | netif_rx_ni(skb); |
76cc8b13 | 10112 | input_queue_head_incr(oldsd); |
fec5e652 | 10113 | } |
ac64da0b | 10114 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 10115 | netif_rx_ni(skb); |
76cc8b13 TH |
10116 | input_queue_head_incr(oldsd); |
10117 | } | |
1da177e4 | 10118 | |
f0bf90de | 10119 | return 0; |
1da177e4 | 10120 | } |
1da177e4 | 10121 | |
7f353bf2 | 10122 | /** |
b63365a2 HX |
10123 | * netdev_increment_features - increment feature set by one |
10124 | * @all: current feature set | |
10125 | * @one: new feature set | |
10126 | * @mask: mask feature set | |
7f353bf2 HX |
10127 | * |
10128 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
10129 | * @one to the master device with current feature set @all. Will not |
10130 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 10131 | */ |
c8f44aff MM |
10132 | netdev_features_t netdev_increment_features(netdev_features_t all, |
10133 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 10134 | { |
c8cd0989 | 10135 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 10136 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 10137 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 10138 | |
a188222b | 10139 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 10140 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 10141 | |
1742f183 | 10142 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
10143 | if (all & NETIF_F_HW_CSUM) |
10144 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
10145 | |
10146 | return all; | |
10147 | } | |
b63365a2 | 10148 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 10149 | |
430f03cd | 10150 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
10151 | { |
10152 | int i; | |
10153 | struct hlist_head *hash; | |
10154 | ||
6da2ec56 | 10155 | hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); |
30d97d35 PE |
10156 | if (hash != NULL) |
10157 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
10158 | INIT_HLIST_HEAD(&hash[i]); | |
10159 | ||
10160 | return hash; | |
10161 | } | |
10162 | ||
881d966b | 10163 | /* Initialize per network namespace state */ |
4665079c | 10164 | static int __net_init netdev_init(struct net *net) |
881d966b | 10165 | { |
d9f37d01 | 10166 | BUILD_BUG_ON(GRO_HASH_BUCKETS > |
ccdb5171 | 10167 | 8 * FIELD_SIZEOF(struct napi_struct, gro_bitmask)); |
d9f37d01 | 10168 | |
734b6541 RM |
10169 | if (net != &init_net) |
10170 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 10171 | |
30d97d35 PE |
10172 | net->dev_name_head = netdev_create_hash(); |
10173 | if (net->dev_name_head == NULL) | |
10174 | goto err_name; | |
881d966b | 10175 | |
30d97d35 PE |
10176 | net->dev_index_head = netdev_create_hash(); |
10177 | if (net->dev_index_head == NULL) | |
10178 | goto err_idx; | |
881d966b | 10179 | |
a30c7b42 JP |
10180 | RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); |
10181 | ||
881d966b | 10182 | return 0; |
30d97d35 PE |
10183 | |
10184 | err_idx: | |
10185 | kfree(net->dev_name_head); | |
10186 | err_name: | |
10187 | return -ENOMEM; | |
881d966b EB |
10188 | } |
10189 | ||
f0db275a SH |
10190 | /** |
10191 | * netdev_drivername - network driver for the device | |
10192 | * @dev: network device | |
f0db275a SH |
10193 | * |
10194 | * Determine network driver for device. | |
10195 | */ | |
3019de12 | 10196 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 10197 | { |
cf04a4c7 SH |
10198 | const struct device_driver *driver; |
10199 | const struct device *parent; | |
3019de12 | 10200 | const char *empty = ""; |
6579e57b AV |
10201 | |
10202 | parent = dev->dev.parent; | |
6579e57b | 10203 | if (!parent) |
3019de12 | 10204 | return empty; |
6579e57b AV |
10205 | |
10206 | driver = parent->driver; | |
10207 | if (driver && driver->name) | |
3019de12 DM |
10208 | return driver->name; |
10209 | return empty; | |
6579e57b AV |
10210 | } |
10211 | ||
6ea754eb JP |
10212 | static void __netdev_printk(const char *level, const struct net_device *dev, |
10213 | struct va_format *vaf) | |
256df2f3 | 10214 | { |
b004ff49 | 10215 | if (dev && dev->dev.parent) { |
6ea754eb JP |
10216 | dev_printk_emit(level[1] - '0', |
10217 | dev->dev.parent, | |
10218 | "%s %s %s%s: %pV", | |
10219 | dev_driver_string(dev->dev.parent), | |
10220 | dev_name(dev->dev.parent), | |
10221 | netdev_name(dev), netdev_reg_state(dev), | |
10222 | vaf); | |
b004ff49 | 10223 | } else if (dev) { |
6ea754eb JP |
10224 | printk("%s%s%s: %pV", |
10225 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 10226 | } else { |
6ea754eb | 10227 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 10228 | } |
256df2f3 JP |
10229 | } |
10230 | ||
6ea754eb JP |
10231 | void netdev_printk(const char *level, const struct net_device *dev, |
10232 | const char *format, ...) | |
256df2f3 JP |
10233 | { |
10234 | struct va_format vaf; | |
10235 | va_list args; | |
256df2f3 JP |
10236 | |
10237 | va_start(args, format); | |
10238 | ||
10239 | vaf.fmt = format; | |
10240 | vaf.va = &args; | |
10241 | ||
6ea754eb | 10242 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 10243 | |
256df2f3 | 10244 | va_end(args); |
256df2f3 JP |
10245 | } |
10246 | EXPORT_SYMBOL(netdev_printk); | |
10247 | ||
10248 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 10249 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 10250 | { \ |
256df2f3 JP |
10251 | struct va_format vaf; \ |
10252 | va_list args; \ | |
10253 | \ | |
10254 | va_start(args, fmt); \ | |
10255 | \ | |
10256 | vaf.fmt = fmt; \ | |
10257 | vaf.va = &args; \ | |
10258 | \ | |
6ea754eb | 10259 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 10260 | \ |
256df2f3 | 10261 | va_end(args); \ |
256df2f3 JP |
10262 | } \ |
10263 | EXPORT_SYMBOL(func); | |
10264 | ||
10265 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
10266 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
10267 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
10268 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
10269 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
10270 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
10271 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
10272 | ||
4665079c | 10273 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
10274 | { |
10275 | kfree(net->dev_name_head); | |
10276 | kfree(net->dev_index_head); | |
ee21b18b VA |
10277 | if (net != &init_net) |
10278 | WARN_ON_ONCE(!list_empty(&net->dev_base_head)); | |
881d966b EB |
10279 | } |
10280 | ||
022cbae6 | 10281 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
10282 | .init = netdev_init, |
10283 | .exit = netdev_exit, | |
10284 | }; | |
10285 | ||
4665079c | 10286 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 10287 | { |
e008b5fc | 10288 | struct net_device *dev, *aux; |
ce286d32 | 10289 | /* |
e008b5fc | 10290 | * Push all migratable network devices back to the |
ce286d32 EB |
10291 | * initial network namespace |
10292 | */ | |
10293 | rtnl_lock(); | |
e008b5fc | 10294 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 10295 | int err; |
aca51397 | 10296 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
10297 | |
10298 | /* Ignore unmoveable devices (i.e. loopback) */ | |
10299 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
10300 | continue; | |
10301 | ||
e008b5fc EB |
10302 | /* Leave virtual devices for the generic cleanup */ |
10303 | if (dev->rtnl_link_ops) | |
10304 | continue; | |
d0c082ce | 10305 | |
25985edc | 10306 | /* Push remaining network devices to init_net */ |
aca51397 | 10307 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
55b40dbf JP |
10308 | if (__dev_get_by_name(&init_net, fb_name)) |
10309 | snprintf(fb_name, IFNAMSIZ, "dev%%d"); | |
aca51397 | 10310 | err = dev_change_net_namespace(dev, &init_net, fb_name); |
ce286d32 | 10311 | if (err) { |
7b6cd1ce JP |
10312 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
10313 | __func__, dev->name, err); | |
aca51397 | 10314 | BUG(); |
ce286d32 EB |
10315 | } |
10316 | } | |
10317 | rtnl_unlock(); | |
10318 | } | |
10319 | ||
50624c93 EB |
10320 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
10321 | { | |
10322 | /* Return with the rtnl_lock held when there are no network | |
10323 | * devices unregistering in any network namespace in net_list. | |
10324 | */ | |
10325 | struct net *net; | |
10326 | bool unregistering; | |
ff960a73 | 10327 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 10328 | |
ff960a73 | 10329 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 10330 | for (;;) { |
50624c93 EB |
10331 | unregistering = false; |
10332 | rtnl_lock(); | |
10333 | list_for_each_entry(net, net_list, exit_list) { | |
10334 | if (net->dev_unreg_count > 0) { | |
10335 | unregistering = true; | |
10336 | break; | |
10337 | } | |
10338 | } | |
10339 | if (!unregistering) | |
10340 | break; | |
10341 | __rtnl_unlock(); | |
ff960a73 PZ |
10342 | |
10343 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 10344 | } |
ff960a73 | 10345 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
10346 | } |
10347 | ||
04dc7f6b EB |
10348 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
10349 | { | |
10350 | /* At exit all network devices most be removed from a network | |
b595076a | 10351 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
10352 | * Do this across as many network namespaces as possible to |
10353 | * improve batching efficiency. | |
10354 | */ | |
10355 | struct net_device *dev; | |
10356 | struct net *net; | |
10357 | LIST_HEAD(dev_kill_list); | |
10358 | ||
50624c93 EB |
10359 | /* To prevent network device cleanup code from dereferencing |
10360 | * loopback devices or network devices that have been freed | |
10361 | * wait here for all pending unregistrations to complete, | |
10362 | * before unregistring the loopback device and allowing the | |
10363 | * network namespace be freed. | |
10364 | * | |
10365 | * The netdev todo list containing all network devices | |
10366 | * unregistrations that happen in default_device_exit_batch | |
10367 | * will run in the rtnl_unlock() at the end of | |
10368 | * default_device_exit_batch. | |
10369 | */ | |
10370 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
10371 | list_for_each_entry(net, net_list, exit_list) { |
10372 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 10373 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
10374 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
10375 | else | |
10376 | unregister_netdevice_queue(dev, &dev_kill_list); | |
10377 | } | |
10378 | } | |
10379 | unregister_netdevice_many(&dev_kill_list); | |
10380 | rtnl_unlock(); | |
10381 | } | |
10382 | ||
022cbae6 | 10383 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 10384 | .exit = default_device_exit, |
04dc7f6b | 10385 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
10386 | }; |
10387 | ||
1da177e4 LT |
10388 | /* |
10389 | * Initialize the DEV module. At boot time this walks the device list and | |
10390 | * unhooks any devices that fail to initialise (normally hardware not | |
10391 | * present) and leaves us with a valid list of present and active devices. | |
10392 | * | |
10393 | */ | |
10394 | ||
10395 | /* | |
10396 | * This is called single threaded during boot, so no need | |
10397 | * to take the rtnl semaphore. | |
10398 | */ | |
10399 | static int __init net_dev_init(void) | |
10400 | { | |
10401 | int i, rc = -ENOMEM; | |
10402 | ||
10403 | BUG_ON(!dev_boot_phase); | |
10404 | ||
1da177e4 LT |
10405 | if (dev_proc_init()) |
10406 | goto out; | |
10407 | ||
8b41d188 | 10408 | if (netdev_kobject_init()) |
1da177e4 LT |
10409 | goto out; |
10410 | ||
10411 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 10412 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
10413 | INIT_LIST_HEAD(&ptype_base[i]); |
10414 | ||
62532da9 VY |
10415 | INIT_LIST_HEAD(&offload_base); |
10416 | ||
881d966b EB |
10417 | if (register_pernet_subsys(&netdev_net_ops)) |
10418 | goto out; | |
1da177e4 LT |
10419 | |
10420 | /* | |
10421 | * Initialise the packet receive queues. | |
10422 | */ | |
10423 | ||
6f912042 | 10424 | for_each_possible_cpu(i) { |
41852497 | 10425 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 10426 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 10427 | |
41852497 ED |
10428 | INIT_WORK(flush, flush_backlog); |
10429 | ||
e36fa2f7 | 10430 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 10431 | skb_queue_head_init(&sd->process_queue); |
f53c7239 SK |
10432 | #ifdef CONFIG_XFRM_OFFLOAD |
10433 | skb_queue_head_init(&sd->xfrm_backlog); | |
10434 | #endif | |
e36fa2f7 | 10435 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 10436 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 10437 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
10438 | sd->csd.func = rps_trigger_softirq; |
10439 | sd->csd.info = sd; | |
e36fa2f7 | 10440 | sd->cpu = i; |
1e94d72f | 10441 | #endif |
0a9627f2 | 10442 | |
7c4ec749 | 10443 | init_gro_hash(&sd->backlog); |
e36fa2f7 ED |
10444 | sd->backlog.poll = process_backlog; |
10445 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
10446 | } |
10447 | ||
1da177e4 LT |
10448 | dev_boot_phase = 0; |
10449 | ||
505d4f73 EB |
10450 | /* The loopback device is special if any other network devices |
10451 | * is present in a network namespace the loopback device must | |
10452 | * be present. Since we now dynamically allocate and free the | |
10453 | * loopback device ensure this invariant is maintained by | |
10454 | * keeping the loopback device as the first device on the | |
10455 | * list of network devices. Ensuring the loopback devices | |
10456 | * is the first device that appears and the last network device | |
10457 | * that disappears. | |
10458 | */ | |
10459 | if (register_pernet_device(&loopback_net_ops)) | |
10460 | goto out; | |
10461 | ||
10462 | if (register_pernet_device(&default_device_ops)) | |
10463 | goto out; | |
10464 | ||
962cf36c CM |
10465 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
10466 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 10467 | |
f0bf90de SAS |
10468 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
10469 | NULL, dev_cpu_dead); | |
10470 | WARN_ON(rc < 0); | |
1da177e4 LT |
10471 | rc = 0; |
10472 | out: | |
10473 | return rc; | |
10474 | } | |
10475 | ||
10476 | subsys_initcall(net_dev_init); |