Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 | 2 | /* |
722c9a0c | 3 | * NET3 Protocol independent device support routines. |
1da177e4 | 4 | * |
1da177e4 | 5 | * Derived from the non IP parts of dev.c 1.0.19 |
722c9a0c | 6 | * Authors: Ross Biro |
1da177e4 LT |
7 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
8 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
9 | * | |
10 | * Additional Authors: | |
11 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
12 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
13 | * David Hinds <dahinds@users.sourceforge.net> | |
14 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
15 | * Adam Sulmicki <adam@cfar.umd.edu> | |
16 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
17 | * | |
18 | * Changes: | |
19 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
722c9a0c | 20 | * to 2 if register_netdev gets called |
21 | * before net_dev_init & also removed a | |
22 | * few lines of code in the process. | |
1da177e4 LT |
23 | * Alan Cox : device private ioctl copies fields back. |
24 | * Alan Cox : Transmit queue code does relevant | |
25 | * stunts to keep the queue safe. | |
26 | * Alan Cox : Fixed double lock. | |
27 | * Alan Cox : Fixed promisc NULL pointer trap | |
28 | * ???????? : Support the full private ioctl range | |
29 | * Alan Cox : Moved ioctl permission check into | |
30 | * drivers | |
31 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
32 | * Alan Cox : 100 backlog just doesn't cut it when | |
33 | * you start doing multicast video 8) | |
34 | * Alan Cox : Rewrote net_bh and list manager. | |
722c9a0c | 35 | * Alan Cox : Fix ETH_P_ALL echoback lengths. |
1da177e4 LT |
36 | * Alan Cox : Took out transmit every packet pass |
37 | * Saved a few bytes in the ioctl handler | |
38 | * Alan Cox : Network driver sets packet type before | |
39 | * calling netif_rx. Saves a function | |
40 | * call a packet. | |
41 | * Alan Cox : Hashed net_bh() | |
42 | * Richard Kooijman: Timestamp fixes. | |
43 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
44 | * Alan Cox : Device lock protection. | |
722c9a0c | 45 | * Alan Cox : Fixed nasty side effect of device close |
1da177e4 LT |
46 | * changes. |
47 | * Rudi Cilibrasi : Pass the right thing to | |
48 | * set_mac_address() | |
49 | * Dave Miller : 32bit quantity for the device lock to | |
50 | * make it work out on a Sparc. | |
51 | * Bjorn Ekwall : Added KERNELD hack. | |
52 | * Alan Cox : Cleaned up the backlog initialise. | |
53 | * Craig Metz : SIOCGIFCONF fix if space for under | |
54 | * 1 device. | |
55 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
56 | * is no device open function. | |
57 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
58 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
59 | * Cyrus Durgin : Cleaned for KMOD | |
60 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
61 | * A network device unload needs to purge | |
62 | * the backlog queue. | |
63 | * Paul Rusty Russell : SIOCSIFNAME | |
64 | * Pekka Riikonen : Netdev boot-time settings code | |
65 | * Andrew Morton : Make unregister_netdevice wait | |
722c9a0c | 66 | * indefinitely on dev->refcnt |
67 | * J Hadi Salim : - Backlog queue sampling | |
1da177e4 LT |
68 | * - netif_rx() feedback |
69 | */ | |
70 | ||
7c0f6ba6 | 71 | #include <linux/uaccess.h> |
1da177e4 | 72 | #include <linux/bitops.h> |
4fc268d2 | 73 | #include <linux/capability.h> |
1da177e4 LT |
74 | #include <linux/cpu.h> |
75 | #include <linux/types.h> | |
76 | #include <linux/kernel.h> | |
08e9897d | 77 | #include <linux/hash.h> |
5a0e3ad6 | 78 | #include <linux/slab.h> |
1da177e4 | 79 | #include <linux/sched.h> |
f1083048 | 80 | #include <linux/sched/mm.h> |
4a3e2f71 | 81 | #include <linux/mutex.h> |
11d6011c | 82 | #include <linux/rwsem.h> |
1da177e4 LT |
83 | #include <linux/string.h> |
84 | #include <linux/mm.h> | |
85 | #include <linux/socket.h> | |
86 | #include <linux/sockios.h> | |
87 | #include <linux/errno.h> | |
88 | #include <linux/interrupt.h> | |
89 | #include <linux/if_ether.h> | |
90 | #include <linux/netdevice.h> | |
91 | #include <linux/etherdevice.h> | |
0187bdfb | 92 | #include <linux/ethtool.h> |
1da177e4 | 93 | #include <linux/skbuff.h> |
a7862b45 | 94 | #include <linux/bpf.h> |
b5cdae32 | 95 | #include <linux/bpf_trace.h> |
457c4cbc | 96 | #include <net/net_namespace.h> |
1da177e4 | 97 | #include <net/sock.h> |
02d62e86 | 98 | #include <net/busy_poll.h> |
1da177e4 | 99 | #include <linux/rtnetlink.h> |
1da177e4 | 100 | #include <linux/stat.h> |
1da177e4 | 101 | #include <net/dst.h> |
fc4099f1 | 102 | #include <net/dst_metadata.h> |
1da177e4 | 103 | #include <net/pkt_sched.h> |
87d83093 | 104 | #include <net/pkt_cls.h> |
1da177e4 | 105 | #include <net/checksum.h> |
44540960 | 106 | #include <net/xfrm.h> |
1da177e4 LT |
107 | #include <linux/highmem.h> |
108 | #include <linux/init.h> | |
1da177e4 | 109 | #include <linux/module.h> |
1da177e4 LT |
110 | #include <linux/netpoll.h> |
111 | #include <linux/rcupdate.h> | |
112 | #include <linux/delay.h> | |
1da177e4 | 113 | #include <net/iw_handler.h> |
1da177e4 | 114 | #include <asm/current.h> |
5bdb9886 | 115 | #include <linux/audit.h> |
db217334 | 116 | #include <linux/dmaengine.h> |
f6a78bfc | 117 | #include <linux/err.h> |
c7fa9d18 | 118 | #include <linux/ctype.h> |
723e98b7 | 119 | #include <linux/if_arp.h> |
6de329e2 | 120 | #include <linux/if_vlan.h> |
8f0f2223 | 121 | #include <linux/ip.h> |
ad55dcaf | 122 | #include <net/ip.h> |
25cd9ba0 | 123 | #include <net/mpls.h> |
8f0f2223 DM |
124 | #include <linux/ipv6.h> |
125 | #include <linux/in.h> | |
b6b2fed1 DM |
126 | #include <linux/jhash.h> |
127 | #include <linux/random.h> | |
9cbc1cb8 | 128 | #include <trace/events/napi.h> |
cf66ba58 | 129 | #include <trace/events/net.h> |
07dc22e7 | 130 | #include <trace/events/skb.h> |
caeda9b9 | 131 | #include <linux/inetdevice.h> |
c445477d | 132 | #include <linux/cpu_rmap.h> |
c5905afb | 133 | #include <linux/static_key.h> |
af12fa6e | 134 | #include <linux/hashtable.h> |
60877a32 | 135 | #include <linux/vmalloc.h> |
529d0489 | 136 | #include <linux/if_macvlan.h> |
e7fd2885 | 137 | #include <linux/errqueue.h> |
3b47d303 | 138 | #include <linux/hrtimer.h> |
357b6cc5 | 139 | #include <linux/netfilter_ingress.h> |
40e4e713 | 140 | #include <linux/crash_dump.h> |
b72b5bf6 | 141 | #include <linux/sctp.h> |
ae847f40 | 142 | #include <net/udp_tunnel.h> |
6621dd29 | 143 | #include <linux/net_namespace.h> |
aaa5d90b | 144 | #include <linux/indirect_call_wrapper.h> |
af3836df | 145 | #include <net/devlink.h> |
1da177e4 | 146 | |
342709ef PE |
147 | #include "net-sysfs.h" |
148 | ||
d565b0a1 HX |
149 | #define MAX_GRO_SKBS 8 |
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 | |
11d6011c | 198 | static DECLARE_RWSEM(devnet_rename_sem); |
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; | |
e08ad805 ED |
334 | /* lookup might have found our primary name or a name belonging |
335 | * to another device. | |
336 | */ | |
337 | if (name_node == dev->name_node || name_node->dev != dev) | |
338 | return -EINVAL; | |
339 | ||
36fbf1e5 JP |
340 | __netdev_name_node_alt_destroy(name_node); |
341 | ||
342 | return 0; | |
343 | } | |
344 | EXPORT_SYMBOL(netdev_name_node_alt_destroy); | |
345 | ||
346 | static void netdev_name_node_alt_flush(struct net_device *dev) | |
347 | { | |
348 | struct netdev_name_node *name_node, *tmp; | |
349 | ||
350 | list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) | |
351 | __netdev_name_node_alt_destroy(name_node); | |
352 | } | |
353 | ||
ce286d32 | 354 | /* Device list insertion */ |
53759be9 | 355 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 356 | { |
c346dca1 | 357 | struct net *net = dev_net(dev); |
ce286d32 EB |
358 | |
359 | ASSERT_RTNL(); | |
360 | ||
361 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 362 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
ff927412 | 363 | netdev_name_node_add(net, dev->name_node); |
fb699dfd ED |
364 | hlist_add_head_rcu(&dev->index_hlist, |
365 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 366 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
367 | |
368 | dev_base_seq_inc(net); | |
ce286d32 EB |
369 | } |
370 | ||
fb699dfd ED |
371 | /* Device list removal |
372 | * caller must respect a RCU grace period before freeing/reusing dev | |
373 | */ | |
ce286d32 EB |
374 | static void unlist_netdevice(struct net_device *dev) |
375 | { | |
376 | ASSERT_RTNL(); | |
377 | ||
378 | /* Unlink dev from the device chain */ | |
379 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 380 | list_del_rcu(&dev->dev_list); |
ff927412 | 381 | netdev_name_node_del(dev->name_node); |
fb699dfd | 382 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 383 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
384 | |
385 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
386 | } |
387 | ||
1da177e4 LT |
388 | /* |
389 | * Our notifier list | |
390 | */ | |
391 | ||
f07d5b94 | 392 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
393 | |
394 | /* | |
395 | * Device drivers call our routines to queue packets here. We empty the | |
396 | * queue in the local softnet handler. | |
397 | */ | |
bea3348e | 398 | |
9958da05 | 399 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 400 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 401 | |
1a33e10e CW |
402 | #ifdef CONFIG_LOCKDEP |
403 | /* | |
404 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class | |
405 | * according to dev->type | |
406 | */ | |
407 | static const unsigned short netdev_lock_type[] = { | |
408 | ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
409 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
410 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
411 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
412 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
413 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
414 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
415 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
416 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
417 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
418 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
419 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
420 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, | |
421 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
422 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
423 | ||
424 | static const char *const netdev_lock_name[] = { | |
425 | "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
426 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
427 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
428 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
429 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
430 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
431 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
432 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
433 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
434 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
435 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
436 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
437 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", | |
438 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
439 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
440 | ||
441 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
845e0ebb | 442 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
1a33e10e CW |
443 | |
444 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
445 | { | |
446 | int i; | |
447 | ||
448 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
449 | if (netdev_lock_type[i] == dev_type) | |
450 | return i; | |
451 | /* the last key is used by default */ | |
452 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
453 | } | |
454 | ||
455 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, | |
456 | unsigned short dev_type) | |
457 | { | |
458 | int i; | |
459 | ||
460 | i = netdev_lock_pos(dev_type); | |
461 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
462 | netdev_lock_name[i]); | |
463 | } | |
845e0ebb CW |
464 | |
465 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
466 | { | |
467 | int i; | |
468 | ||
469 | i = netdev_lock_pos(dev->type); | |
470 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
471 | &netdev_addr_lock_key[i], | |
472 | netdev_lock_name[i]); | |
473 | } | |
1a33e10e CW |
474 | #else |
475 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, | |
476 | unsigned short dev_type) | |
477 | { | |
478 | } | |
845e0ebb CW |
479 | |
480 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
481 | { | |
482 | } | |
1a33e10e CW |
483 | #endif |
484 | ||
1da177e4 | 485 | /******************************************************************************* |
eb13da1a | 486 | * |
487 | * Protocol management and registration routines | |
488 | * | |
489 | *******************************************************************************/ | |
1da177e4 | 490 | |
1da177e4 | 491 | |
1da177e4 LT |
492 | /* |
493 | * Add a protocol ID to the list. Now that the input handler is | |
494 | * smarter we can dispense with all the messy stuff that used to be | |
495 | * here. | |
496 | * | |
497 | * BEWARE!!! Protocol handlers, mangling input packets, | |
498 | * MUST BE last in hash buckets and checking protocol handlers | |
499 | * MUST start from promiscuous ptype_all chain in net_bh. | |
500 | * It is true now, do not change it. | |
501 | * Explanation follows: if protocol handler, mangling packet, will | |
502 | * be the first on list, it is not able to sense, that packet | |
503 | * is cloned and should be copied-on-write, so that it will | |
504 | * change it and subsequent readers will get broken packet. | |
505 | * --ANK (980803) | |
506 | */ | |
507 | ||
c07b68e8 ED |
508 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
509 | { | |
510 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 511 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 512 | else |
7866a621 SN |
513 | return pt->dev ? &pt->dev->ptype_specific : |
514 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
515 | } |
516 | ||
1da177e4 LT |
517 | /** |
518 | * dev_add_pack - add packet handler | |
519 | * @pt: packet type declaration | |
520 | * | |
521 | * Add a protocol handler to the networking stack. The passed &packet_type | |
522 | * is linked into kernel lists and may not be freed until it has been | |
523 | * removed from the kernel lists. | |
524 | * | |
4ec93edb | 525 | * This call does not sleep therefore it can not |
1da177e4 LT |
526 | * guarantee all CPU's that are in middle of receiving packets |
527 | * will see the new packet type (until the next received packet). | |
528 | */ | |
529 | ||
530 | void dev_add_pack(struct packet_type *pt) | |
531 | { | |
c07b68e8 | 532 | struct list_head *head = ptype_head(pt); |
1da177e4 | 533 | |
c07b68e8 ED |
534 | spin_lock(&ptype_lock); |
535 | list_add_rcu(&pt->list, head); | |
536 | spin_unlock(&ptype_lock); | |
1da177e4 | 537 | } |
d1b19dff | 538 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 539 | |
1da177e4 LT |
540 | /** |
541 | * __dev_remove_pack - remove packet handler | |
542 | * @pt: packet type declaration | |
543 | * | |
544 | * Remove a protocol handler that was previously added to the kernel | |
545 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
546 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 547 | * returns. |
1da177e4 LT |
548 | * |
549 | * The packet type might still be in use by receivers | |
550 | * and must not be freed until after all the CPU's have gone | |
551 | * through a quiescent state. | |
552 | */ | |
553 | void __dev_remove_pack(struct packet_type *pt) | |
554 | { | |
c07b68e8 | 555 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
556 | struct packet_type *pt1; |
557 | ||
c07b68e8 | 558 | spin_lock(&ptype_lock); |
1da177e4 LT |
559 | |
560 | list_for_each_entry(pt1, head, list) { | |
561 | if (pt == pt1) { | |
562 | list_del_rcu(&pt->list); | |
563 | goto out; | |
564 | } | |
565 | } | |
566 | ||
7b6cd1ce | 567 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 568 | out: |
c07b68e8 | 569 | spin_unlock(&ptype_lock); |
1da177e4 | 570 | } |
d1b19dff ED |
571 | EXPORT_SYMBOL(__dev_remove_pack); |
572 | ||
1da177e4 LT |
573 | /** |
574 | * dev_remove_pack - remove packet handler | |
575 | * @pt: packet type declaration | |
576 | * | |
577 | * Remove a protocol handler that was previously added to the kernel | |
578 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
579 | * from the kernel lists and can be freed or reused once this function | |
580 | * returns. | |
581 | * | |
582 | * This call sleeps to guarantee that no CPU is looking at the packet | |
583 | * type after return. | |
584 | */ | |
585 | void dev_remove_pack(struct packet_type *pt) | |
586 | { | |
587 | __dev_remove_pack(pt); | |
4ec93edb | 588 | |
1da177e4 LT |
589 | synchronize_net(); |
590 | } | |
d1b19dff | 591 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 592 | |
62532da9 VY |
593 | |
594 | /** | |
595 | * dev_add_offload - register offload handlers | |
596 | * @po: protocol offload declaration | |
597 | * | |
598 | * Add protocol offload handlers to the networking stack. The passed | |
599 | * &proto_offload is linked into kernel lists and may not be freed until | |
600 | * it has been removed from the kernel lists. | |
601 | * | |
602 | * This call does not sleep therefore it can not | |
603 | * guarantee all CPU's that are in middle of receiving packets | |
604 | * will see the new offload handlers (until the next received packet). | |
605 | */ | |
606 | void dev_add_offload(struct packet_offload *po) | |
607 | { | |
bdef7de4 | 608 | struct packet_offload *elem; |
62532da9 VY |
609 | |
610 | spin_lock(&offload_lock); | |
bdef7de4 DM |
611 | list_for_each_entry(elem, &offload_base, list) { |
612 | if (po->priority < elem->priority) | |
613 | break; | |
614 | } | |
615 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
616 | spin_unlock(&offload_lock); |
617 | } | |
618 | EXPORT_SYMBOL(dev_add_offload); | |
619 | ||
620 | /** | |
621 | * __dev_remove_offload - remove offload handler | |
622 | * @po: packet offload declaration | |
623 | * | |
624 | * Remove a protocol offload handler that was previously added to the | |
625 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
626 | * is removed from the kernel lists and can be freed or reused once this | |
627 | * function returns. | |
628 | * | |
629 | * The packet type might still be in use by receivers | |
630 | * and must not be freed until after all the CPU's have gone | |
631 | * through a quiescent state. | |
632 | */ | |
1d143d9f | 633 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
634 | { |
635 | struct list_head *head = &offload_base; | |
636 | struct packet_offload *po1; | |
637 | ||
c53aa505 | 638 | spin_lock(&offload_lock); |
62532da9 VY |
639 | |
640 | list_for_each_entry(po1, head, list) { | |
641 | if (po == po1) { | |
642 | list_del_rcu(&po->list); | |
643 | goto out; | |
644 | } | |
645 | } | |
646 | ||
647 | pr_warn("dev_remove_offload: %p not found\n", po); | |
648 | out: | |
c53aa505 | 649 | spin_unlock(&offload_lock); |
62532da9 | 650 | } |
62532da9 VY |
651 | |
652 | /** | |
653 | * dev_remove_offload - remove packet offload handler | |
654 | * @po: packet offload declaration | |
655 | * | |
656 | * Remove a packet offload handler that was previously added to the kernel | |
657 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
658 | * removed from the kernel lists and can be freed or reused once this | |
659 | * function returns. | |
660 | * | |
661 | * This call sleeps to guarantee that no CPU is looking at the packet | |
662 | * type after return. | |
663 | */ | |
664 | void dev_remove_offload(struct packet_offload *po) | |
665 | { | |
666 | __dev_remove_offload(po); | |
667 | ||
668 | synchronize_net(); | |
669 | } | |
670 | EXPORT_SYMBOL(dev_remove_offload); | |
671 | ||
1da177e4 | 672 | /****************************************************************************** |
eb13da1a | 673 | * |
674 | * Device Boot-time Settings Routines | |
675 | * | |
676 | ******************************************************************************/ | |
1da177e4 LT |
677 | |
678 | /* Boot time configuration table */ | |
679 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
680 | ||
681 | /** | |
682 | * netdev_boot_setup_add - add new setup entry | |
683 | * @name: name of the device | |
684 | * @map: configured settings for the device | |
685 | * | |
686 | * Adds new setup entry to the dev_boot_setup list. The function | |
687 | * returns 0 on error and 1 on success. This is a generic routine to | |
688 | * all netdevices. | |
689 | */ | |
690 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
691 | { | |
692 | struct netdev_boot_setup *s; | |
693 | int i; | |
694 | ||
695 | s = dev_boot_setup; | |
696 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
697 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
698 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 699 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
700 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
701 | break; | |
702 | } | |
703 | } | |
704 | ||
705 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
706 | } | |
707 | ||
708 | /** | |
722c9a0c | 709 | * netdev_boot_setup_check - check boot time settings |
710 | * @dev: the netdevice | |
1da177e4 | 711 | * |
722c9a0c | 712 | * Check boot time settings for the device. |
713 | * The found settings are set for the device to be used | |
714 | * later in the device probing. | |
715 | * Returns 0 if no settings found, 1 if they are. | |
1da177e4 LT |
716 | */ |
717 | int netdev_boot_setup_check(struct net_device *dev) | |
718 | { | |
719 | struct netdev_boot_setup *s = dev_boot_setup; | |
720 | int i; | |
721 | ||
722 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
723 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 724 | !strcmp(dev->name, s[i].name)) { |
722c9a0c | 725 | dev->irq = s[i].map.irq; |
726 | dev->base_addr = s[i].map.base_addr; | |
727 | dev->mem_start = s[i].map.mem_start; | |
728 | dev->mem_end = s[i].map.mem_end; | |
1da177e4 LT |
729 | return 1; |
730 | } | |
731 | } | |
732 | return 0; | |
733 | } | |
d1b19dff | 734 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
735 | |
736 | ||
737 | /** | |
722c9a0c | 738 | * netdev_boot_base - get address from boot time settings |
739 | * @prefix: prefix for network device | |
740 | * @unit: id for network device | |
741 | * | |
742 | * Check boot time settings for the base address of device. | |
743 | * The found settings are set for the device to be used | |
744 | * later in the device probing. | |
745 | * Returns 0 if no settings found. | |
1da177e4 LT |
746 | */ |
747 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
748 | { | |
749 | const struct netdev_boot_setup *s = dev_boot_setup; | |
750 | char name[IFNAMSIZ]; | |
751 | int i; | |
752 | ||
753 | sprintf(name, "%s%d", prefix, unit); | |
754 | ||
755 | /* | |
756 | * If device already registered then return base of 1 | |
757 | * to indicate not to probe for this interface | |
758 | */ | |
881d966b | 759 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
760 | return 1; |
761 | ||
762 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
763 | if (!strcmp(name, s[i].name)) | |
764 | return s[i].map.base_addr; | |
765 | return 0; | |
766 | } | |
767 | ||
768 | /* | |
769 | * Saves at boot time configured settings for any netdevice. | |
770 | */ | |
771 | int __init netdev_boot_setup(char *str) | |
772 | { | |
773 | int ints[5]; | |
774 | struct ifmap map; | |
775 | ||
776 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
777 | if (!str || !*str) | |
778 | return 0; | |
779 | ||
780 | /* Save settings */ | |
781 | memset(&map, 0, sizeof(map)); | |
782 | if (ints[0] > 0) | |
783 | map.irq = ints[1]; | |
784 | if (ints[0] > 1) | |
785 | map.base_addr = ints[2]; | |
786 | if (ints[0] > 2) | |
787 | map.mem_start = ints[3]; | |
788 | if (ints[0] > 3) | |
789 | map.mem_end = ints[4]; | |
790 | ||
791 | /* Add new entry to the list */ | |
792 | return netdev_boot_setup_add(str, &map); | |
793 | } | |
794 | ||
795 | __setup("netdev=", netdev_boot_setup); | |
796 | ||
797 | /******************************************************************************* | |
eb13da1a | 798 | * |
799 | * Device Interface Subroutines | |
800 | * | |
801 | *******************************************************************************/ | |
1da177e4 | 802 | |
a54acb3a ND |
803 | /** |
804 | * dev_get_iflink - get 'iflink' value of a interface | |
805 | * @dev: targeted interface | |
806 | * | |
807 | * Indicates the ifindex the interface is linked to. | |
808 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
809 | */ | |
810 | ||
811 | int dev_get_iflink(const struct net_device *dev) | |
812 | { | |
813 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
814 | return dev->netdev_ops->ndo_get_iflink(dev); | |
815 | ||
7a66bbc9 | 816 | return dev->ifindex; |
a54acb3a ND |
817 | } |
818 | EXPORT_SYMBOL(dev_get_iflink); | |
819 | ||
fc4099f1 PS |
820 | /** |
821 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
822 | * @dev: targeted interface | |
823 | * @skb: The packet. | |
824 | * | |
825 | * For better visibility of tunnel traffic OVS needs to retrieve | |
826 | * egress tunnel information for a packet. Following API allows | |
827 | * user to get this info. | |
828 | */ | |
829 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
830 | { | |
831 | struct ip_tunnel_info *info; | |
832 | ||
833 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
834 | return -EINVAL; | |
835 | ||
836 | info = skb_tunnel_info_unclone(skb); | |
837 | if (!info) | |
838 | return -ENOMEM; | |
839 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
840 | return -EINVAL; | |
841 | ||
842 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
843 | } | |
844 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
845 | ||
1da177e4 LT |
846 | /** |
847 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 848 | * @net: the applicable net namespace |
1da177e4 LT |
849 | * @name: name to find |
850 | * | |
851 | * Find an interface by name. Must be called under RTNL semaphore | |
852 | * or @dev_base_lock. If the name is found a pointer to the device | |
853 | * is returned. If the name is not found then %NULL is returned. The | |
854 | * reference counters are not incremented so the caller must be | |
855 | * careful with locks. | |
856 | */ | |
857 | ||
881d966b | 858 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 859 | { |
ff927412 | 860 | struct netdev_name_node *node_name; |
1da177e4 | 861 | |
ff927412 JP |
862 | node_name = netdev_name_node_lookup(net, name); |
863 | return node_name ? node_name->dev : NULL; | |
1da177e4 | 864 | } |
d1b19dff | 865 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 866 | |
72c9528b | 867 | /** |
722c9a0c | 868 | * dev_get_by_name_rcu - find a device by its name |
869 | * @net: the applicable net namespace | |
870 | * @name: name to find | |
871 | * | |
872 | * Find an interface by name. | |
873 | * If the name is found a pointer to the device is returned. | |
874 | * If the name is not found then %NULL is returned. | |
875 | * The reference counters are not incremented so the caller must be | |
876 | * careful with locks. The caller must hold RCU lock. | |
72c9528b ED |
877 | */ |
878 | ||
879 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
880 | { | |
ff927412 | 881 | struct netdev_name_node *node_name; |
72c9528b | 882 | |
ff927412 JP |
883 | node_name = netdev_name_node_lookup_rcu(net, name); |
884 | return node_name ? node_name->dev : NULL; | |
72c9528b ED |
885 | } |
886 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
887 | ||
1da177e4 LT |
888 | /** |
889 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 890 | * @net: the applicable net namespace |
1da177e4 LT |
891 | * @name: name to find |
892 | * | |
893 | * Find an interface by name. This can be called from any | |
894 | * context and does its own locking. The returned handle has | |
895 | * the usage count incremented and the caller must use dev_put() to | |
896 | * release it when it is no longer needed. %NULL is returned if no | |
897 | * matching device is found. | |
898 | */ | |
899 | ||
881d966b | 900 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
901 | { |
902 | struct net_device *dev; | |
903 | ||
72c9528b ED |
904 | rcu_read_lock(); |
905 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
906 | if (dev) |
907 | dev_hold(dev); | |
72c9528b | 908 | rcu_read_unlock(); |
1da177e4 LT |
909 | return dev; |
910 | } | |
d1b19dff | 911 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
912 | |
913 | /** | |
914 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 915 | * @net: the applicable net namespace |
1da177e4 LT |
916 | * @ifindex: index of device |
917 | * | |
918 | * Search for an interface by index. Returns %NULL if the device | |
919 | * is not found or a pointer to the device. The device has not | |
920 | * had its reference counter increased so the caller must be careful | |
921 | * about locking. The caller must hold either the RTNL semaphore | |
922 | * or @dev_base_lock. | |
923 | */ | |
924 | ||
881d966b | 925 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 926 | { |
0bd8d536 ED |
927 | struct net_device *dev; |
928 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 929 | |
b67bfe0d | 930 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
931 | if (dev->ifindex == ifindex) |
932 | return dev; | |
0bd8d536 | 933 | |
1da177e4 LT |
934 | return NULL; |
935 | } | |
d1b19dff | 936 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 937 | |
fb699dfd ED |
938 | /** |
939 | * dev_get_by_index_rcu - find a device by its ifindex | |
940 | * @net: the applicable net namespace | |
941 | * @ifindex: index of device | |
942 | * | |
943 | * Search for an interface by index. Returns %NULL if the device | |
944 | * is not found or a pointer to the device. The device has not | |
945 | * had its reference counter increased so the caller must be careful | |
946 | * about locking. The caller must hold RCU lock. | |
947 | */ | |
948 | ||
949 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
950 | { | |
fb699dfd ED |
951 | struct net_device *dev; |
952 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
953 | ||
b67bfe0d | 954 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
955 | if (dev->ifindex == ifindex) |
956 | return dev; | |
957 | ||
958 | return NULL; | |
959 | } | |
960 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
961 | ||
1da177e4 LT |
962 | |
963 | /** | |
964 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 965 | * @net: the applicable net namespace |
1da177e4 LT |
966 | * @ifindex: index of device |
967 | * | |
968 | * Search for an interface by index. Returns NULL if the device | |
969 | * is not found or a pointer to the device. The device returned has | |
970 | * had a reference added and the pointer is safe until the user calls | |
971 | * dev_put to indicate they have finished with it. | |
972 | */ | |
973 | ||
881d966b | 974 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
975 | { |
976 | struct net_device *dev; | |
977 | ||
fb699dfd ED |
978 | rcu_read_lock(); |
979 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
980 | if (dev) |
981 | dev_hold(dev); | |
fb699dfd | 982 | rcu_read_unlock(); |
1da177e4 LT |
983 | return dev; |
984 | } | |
d1b19dff | 985 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 986 | |
90b602f8 ML |
987 | /** |
988 | * dev_get_by_napi_id - find a device by napi_id | |
989 | * @napi_id: ID of the NAPI struct | |
990 | * | |
991 | * Search for an interface by NAPI ID. Returns %NULL if the device | |
992 | * is not found or a pointer to the device. The device has not had | |
993 | * its reference counter increased so the caller must be careful | |
994 | * about locking. The caller must hold RCU lock. | |
995 | */ | |
996 | ||
997 | struct net_device *dev_get_by_napi_id(unsigned int napi_id) | |
998 | { | |
999 | struct napi_struct *napi; | |
1000 | ||
1001 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
1002 | ||
1003 | if (napi_id < MIN_NAPI_ID) | |
1004 | return NULL; | |
1005 | ||
1006 | napi = napi_by_id(napi_id); | |
1007 | ||
1008 | return napi ? napi->dev : NULL; | |
1009 | } | |
1010 | EXPORT_SYMBOL(dev_get_by_napi_id); | |
1011 | ||
5dbe7c17 NS |
1012 | /** |
1013 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
1014 | * @net: network namespace | |
1015 | * @name: a pointer to the buffer where the name will be stored. | |
1016 | * @ifindex: the ifindex of the interface to get the name from. | |
5dbe7c17 NS |
1017 | */ |
1018 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
1019 | { | |
1020 | struct net_device *dev; | |
11d6011c | 1021 | int ret; |
5dbe7c17 | 1022 | |
11d6011c | 1023 | down_read(&devnet_rename_sem); |
5dbe7c17 | 1024 | rcu_read_lock(); |
11d6011c | 1025 | |
5dbe7c17 NS |
1026 | dev = dev_get_by_index_rcu(net, ifindex); |
1027 | if (!dev) { | |
11d6011c AD |
1028 | ret = -ENODEV; |
1029 | goto out; | |
5dbe7c17 NS |
1030 | } |
1031 | ||
1032 | strcpy(name, dev->name); | |
5dbe7c17 | 1033 | |
11d6011c AD |
1034 | ret = 0; |
1035 | out: | |
1036 | rcu_read_unlock(); | |
1037 | up_read(&devnet_rename_sem); | |
1038 | return ret; | |
5dbe7c17 NS |
1039 | } |
1040 | ||
1da177e4 | 1041 | /** |
941666c2 | 1042 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 1043 | * @net: the applicable net namespace |
1da177e4 LT |
1044 | * @type: media type of device |
1045 | * @ha: hardware address | |
1046 | * | |
1047 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
1048 | * is not found or a pointer to the device. |
1049 | * The caller must hold RCU or RTNL. | |
941666c2 | 1050 | * The returned device has not had its ref count increased |
1da177e4 LT |
1051 | * and the caller must therefore be careful about locking |
1052 | * | |
1da177e4 LT |
1053 | */ |
1054 | ||
941666c2 ED |
1055 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
1056 | const char *ha) | |
1da177e4 LT |
1057 | { |
1058 | struct net_device *dev; | |
1059 | ||
941666c2 | 1060 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
1061 | if (dev->type == type && |
1062 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
1063 | return dev; |
1064 | ||
1065 | return NULL; | |
1da177e4 | 1066 | } |
941666c2 | 1067 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 1068 | |
881d966b | 1069 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
1070 | { |
1071 | struct net_device *dev; | |
1072 | ||
4e9cac2b | 1073 | ASSERT_RTNL(); |
881d966b | 1074 | for_each_netdev(net, dev) |
4e9cac2b | 1075 | if (dev->type == type) |
7562f876 PE |
1076 | return dev; |
1077 | ||
1078 | return NULL; | |
4e9cac2b | 1079 | } |
4e9cac2b PM |
1080 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
1081 | ||
881d966b | 1082 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 1083 | { |
99fe3c39 | 1084 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 1085 | |
99fe3c39 ED |
1086 | rcu_read_lock(); |
1087 | for_each_netdev_rcu(net, dev) | |
1088 | if (dev->type == type) { | |
1089 | dev_hold(dev); | |
1090 | ret = dev; | |
1091 | break; | |
1092 | } | |
1093 | rcu_read_unlock(); | |
1094 | return ret; | |
1da177e4 | 1095 | } |
1da177e4 LT |
1096 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
1097 | ||
1098 | /** | |
6c555490 | 1099 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 1100 | * @net: the applicable net namespace |
1da177e4 LT |
1101 | * @if_flags: IFF_* values |
1102 | * @mask: bitmask of bits in if_flags to check | |
1103 | * | |
1104 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 1105 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 1106 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
1107 | */ |
1108 | ||
6c555490 WC |
1109 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
1110 | unsigned short mask) | |
1da177e4 | 1111 | { |
7562f876 | 1112 | struct net_device *dev, *ret; |
1da177e4 | 1113 | |
6c555490 WC |
1114 | ASSERT_RTNL(); |
1115 | ||
7562f876 | 1116 | ret = NULL; |
6c555490 | 1117 | for_each_netdev(net, dev) { |
1da177e4 | 1118 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 1119 | ret = dev; |
1da177e4 LT |
1120 | break; |
1121 | } | |
1122 | } | |
7562f876 | 1123 | return ret; |
1da177e4 | 1124 | } |
6c555490 | 1125 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
1126 | |
1127 | /** | |
1128 | * dev_valid_name - check if name is okay for network device | |
1129 | * @name: name string | |
1130 | * | |
1131 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
1132 | * to allow sysfs to work. We also disallow any kind of |
1133 | * whitespace. | |
1da177e4 | 1134 | */ |
95f050bf | 1135 | bool dev_valid_name(const char *name) |
1da177e4 | 1136 | { |
c7fa9d18 | 1137 | if (*name == '\0') |
95f050bf | 1138 | return false; |
a9d48205 | 1139 | if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) |
95f050bf | 1140 | return false; |
c7fa9d18 | 1141 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1142 | return false; |
c7fa9d18 DM |
1143 | |
1144 | while (*name) { | |
a4176a93 | 1145 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1146 | return false; |
c7fa9d18 DM |
1147 | name++; |
1148 | } | |
95f050bf | 1149 | return true; |
1da177e4 | 1150 | } |
d1b19dff | 1151 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1152 | |
1153 | /** | |
b267b179 EB |
1154 | * __dev_alloc_name - allocate a name for a device |
1155 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1156 | * @name: name format string |
b267b179 | 1157 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1158 | * |
1159 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1160 | * id. It scans list of devices to build up a free map, then chooses |
1161 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1162 | * while allocating the name and adding the device in order to avoid | |
1163 | * duplicates. | |
1164 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1165 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1166 | */ |
1167 | ||
b267b179 | 1168 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1169 | { |
1170 | int i = 0; | |
1da177e4 LT |
1171 | const char *p; |
1172 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1173 | unsigned long *inuse; |
1da177e4 LT |
1174 | struct net_device *d; |
1175 | ||
93809105 RV |
1176 | if (!dev_valid_name(name)) |
1177 | return -EINVAL; | |
1178 | ||
51f299dd | 1179 | p = strchr(name, '%'); |
1da177e4 LT |
1180 | if (p) { |
1181 | /* | |
1182 | * Verify the string as this thing may have come from | |
1183 | * the user. There must be either one "%d" and no other "%" | |
1184 | * characters. | |
1185 | */ | |
1186 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1187 | return -EINVAL; | |
1188 | ||
1189 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1190 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1191 | if (!inuse) |
1192 | return -ENOMEM; | |
1193 | ||
881d966b | 1194 | for_each_netdev(net, d) { |
1da177e4 LT |
1195 | if (!sscanf(d->name, name, &i)) |
1196 | continue; | |
1197 | if (i < 0 || i >= max_netdevices) | |
1198 | continue; | |
1199 | ||
1200 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1201 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1202 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1203 | set_bit(i, inuse); | |
1204 | } | |
1205 | ||
1206 | i = find_first_zero_bit(inuse, max_netdevices); | |
1207 | free_page((unsigned long) inuse); | |
1208 | } | |
1209 | ||
6224abda | 1210 | snprintf(buf, IFNAMSIZ, name, i); |
b267b179 | 1211 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1212 | return i; |
1da177e4 LT |
1213 | |
1214 | /* It is possible to run out of possible slots | |
1215 | * when the name is long and there isn't enough space left | |
1216 | * for the digits, or if all bits are used. | |
1217 | */ | |
029b6d14 | 1218 | return -ENFILE; |
1da177e4 LT |
1219 | } |
1220 | ||
2c88b855 RV |
1221 | static int dev_alloc_name_ns(struct net *net, |
1222 | struct net_device *dev, | |
1223 | const char *name) | |
1224 | { | |
1225 | char buf[IFNAMSIZ]; | |
1226 | int ret; | |
1227 | ||
c46d7642 | 1228 | BUG_ON(!net); |
2c88b855 RV |
1229 | ret = __dev_alloc_name(net, name, buf); |
1230 | if (ret >= 0) | |
1231 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1232 | return ret; | |
1da177e4 LT |
1233 | } |
1234 | ||
b267b179 EB |
1235 | /** |
1236 | * dev_alloc_name - allocate a name for a device | |
1237 | * @dev: device | |
1238 | * @name: name format string | |
1239 | * | |
1240 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1241 | * id. It scans list of devices to build up a free map, then chooses | |
1242 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1243 | * while allocating the name and adding the device in order to avoid | |
1244 | * duplicates. | |
1245 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1246 | * Returns the number of the unit assigned or a negative errno code. | |
1247 | */ | |
1248 | ||
1249 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1250 | { | |
c46d7642 | 1251 | return dev_alloc_name_ns(dev_net(dev), dev, name); |
b267b179 | 1252 | } |
d1b19dff | 1253 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1254 | |
bacb7e18 ED |
1255 | static int dev_get_valid_name(struct net *net, struct net_device *dev, |
1256 | const char *name) | |
828de4f6 | 1257 | { |
55a5ec9b DM |
1258 | BUG_ON(!net); |
1259 | ||
1260 | if (!dev_valid_name(name)) | |
1261 | return -EINVAL; | |
1262 | ||
1263 | if (strchr(name, '%')) | |
1264 | return dev_alloc_name_ns(net, dev, name); | |
1265 | else if (__dev_get_by_name(net, name)) | |
1266 | return -EEXIST; | |
1267 | else if (dev->name != name) | |
1268 | strlcpy(dev->name, name, IFNAMSIZ); | |
1269 | ||
1270 | return 0; | |
d9031024 | 1271 | } |
1da177e4 LT |
1272 | |
1273 | /** | |
1274 | * dev_change_name - change name of a device | |
1275 | * @dev: device | |
1276 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1277 | * | |
1278 | * Change name of a device, can pass format strings "eth%d". | |
1279 | * for wildcarding. | |
1280 | */ | |
cf04a4c7 | 1281 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1282 | { |
238fa362 | 1283 | unsigned char old_assign_type; |
fcc5a03a | 1284 | char oldname[IFNAMSIZ]; |
1da177e4 | 1285 | int err = 0; |
fcc5a03a | 1286 | int ret; |
881d966b | 1287 | struct net *net; |
1da177e4 LT |
1288 | |
1289 | ASSERT_RTNL(); | |
c346dca1 | 1290 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1291 | |
c346dca1 | 1292 | net = dev_net(dev); |
8065a779 SWL |
1293 | |
1294 | /* Some auto-enslaved devices e.g. failover slaves are | |
1295 | * special, as userspace might rename the device after | |
1296 | * the interface had been brought up and running since | |
1297 | * the point kernel initiated auto-enslavement. Allow | |
1298 | * live name change even when these slave devices are | |
1299 | * up and running. | |
1300 | * | |
1301 | * Typically, users of these auto-enslaving devices | |
1302 | * don't actually care about slave name change, as | |
1303 | * they are supposed to operate on master interface | |
1304 | * directly. | |
1305 | */ | |
1306 | if (dev->flags & IFF_UP && | |
1307 | likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) | |
1da177e4 LT |
1308 | return -EBUSY; |
1309 | ||
11d6011c | 1310 | down_write(&devnet_rename_sem); |
c91f6df2 BH |
1311 | |
1312 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
11d6011c | 1313 | up_write(&devnet_rename_sem); |
c8d90dca | 1314 | return 0; |
c91f6df2 | 1315 | } |
c8d90dca | 1316 | |
fcc5a03a HX |
1317 | memcpy(oldname, dev->name, IFNAMSIZ); |
1318 | ||
828de4f6 | 1319 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1320 | if (err < 0) { |
11d6011c | 1321 | up_write(&devnet_rename_sem); |
d9031024 | 1322 | return err; |
c91f6df2 | 1323 | } |
1da177e4 | 1324 | |
6fe82a39 VF |
1325 | if (oldname[0] && !strchr(oldname, '%')) |
1326 | netdev_info(dev, "renamed from %s\n", oldname); | |
1327 | ||
238fa362 TG |
1328 | old_assign_type = dev->name_assign_type; |
1329 | dev->name_assign_type = NET_NAME_RENAMED; | |
1330 | ||
fcc5a03a | 1331 | rollback: |
a1b3f594 EB |
1332 | ret = device_rename(&dev->dev, dev->name); |
1333 | if (ret) { | |
1334 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1335 | dev->name_assign_type = old_assign_type; |
11d6011c | 1336 | up_write(&devnet_rename_sem); |
a1b3f594 | 1337 | return ret; |
dcc99773 | 1338 | } |
7f988eab | 1339 | |
11d6011c | 1340 | up_write(&devnet_rename_sem); |
c91f6df2 | 1341 | |
5bb025fa VF |
1342 | netdev_adjacent_rename_links(dev, oldname); |
1343 | ||
7f988eab | 1344 | write_lock_bh(&dev_base_lock); |
ff927412 | 1345 | netdev_name_node_del(dev->name_node); |
72c9528b ED |
1346 | write_unlock_bh(&dev_base_lock); |
1347 | ||
1348 | synchronize_rcu(); | |
1349 | ||
1350 | write_lock_bh(&dev_base_lock); | |
ff927412 | 1351 | netdev_name_node_add(net, dev->name_node); |
7f988eab HX |
1352 | write_unlock_bh(&dev_base_lock); |
1353 | ||
056925ab | 1354 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1355 | ret = notifier_to_errno(ret); |
1356 | ||
1357 | if (ret) { | |
91e9c07b ED |
1358 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1359 | if (err >= 0) { | |
fcc5a03a | 1360 | err = ret; |
11d6011c | 1361 | down_write(&devnet_rename_sem); |
fcc5a03a | 1362 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1363 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1364 | dev->name_assign_type = old_assign_type; |
1365 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1366 | goto rollback; |
91e9c07b | 1367 | } else { |
7b6cd1ce | 1368 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1369 | dev->name, ret); |
fcc5a03a HX |
1370 | } |
1371 | } | |
1da177e4 LT |
1372 | |
1373 | return err; | |
1374 | } | |
1375 | ||
0b815a1a SH |
1376 | /** |
1377 | * dev_set_alias - change ifalias of a device | |
1378 | * @dev: device | |
1379 | * @alias: name up to IFALIASZ | |
f0db275a | 1380 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1381 | * |
1382 | * Set ifalias for a device, | |
1383 | */ | |
1384 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1385 | { | |
6c557001 | 1386 | struct dev_ifalias *new_alias = NULL; |
0b815a1a SH |
1387 | |
1388 | if (len >= IFALIASZ) | |
1389 | return -EINVAL; | |
1390 | ||
6c557001 FW |
1391 | if (len) { |
1392 | new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); | |
1393 | if (!new_alias) | |
1394 | return -ENOMEM; | |
1395 | ||
1396 | memcpy(new_alias->ifalias, alias, len); | |
1397 | new_alias->ifalias[len] = 0; | |
96ca4a2c OH |
1398 | } |
1399 | ||
6c557001 | 1400 | mutex_lock(&ifalias_mutex); |
e3f0d761 PM |
1401 | new_alias = rcu_replace_pointer(dev->ifalias, new_alias, |
1402 | mutex_is_locked(&ifalias_mutex)); | |
6c557001 FW |
1403 | mutex_unlock(&ifalias_mutex); |
1404 | ||
1405 | if (new_alias) | |
1406 | kfree_rcu(new_alias, rcuhead); | |
0b815a1a | 1407 | |
0b815a1a SH |
1408 | return len; |
1409 | } | |
0fe554a4 | 1410 | EXPORT_SYMBOL(dev_set_alias); |
0b815a1a | 1411 | |
6c557001 FW |
1412 | /** |
1413 | * dev_get_alias - get ifalias of a device | |
1414 | * @dev: device | |
20e88320 | 1415 | * @name: buffer to store name of ifalias |
6c557001 FW |
1416 | * @len: size of buffer |
1417 | * | |
1418 | * get ifalias for a device. Caller must make sure dev cannot go | |
1419 | * away, e.g. rcu read lock or own a reference count to device. | |
1420 | */ | |
1421 | int dev_get_alias(const struct net_device *dev, char *name, size_t len) | |
1422 | { | |
1423 | const struct dev_ifalias *alias; | |
1424 | int ret = 0; | |
1425 | ||
1426 | rcu_read_lock(); | |
1427 | alias = rcu_dereference(dev->ifalias); | |
1428 | if (alias) | |
1429 | ret = snprintf(name, len, "%s", alias->ifalias); | |
1430 | rcu_read_unlock(); | |
1431 | ||
1432 | return ret; | |
1433 | } | |
0b815a1a | 1434 | |
d8a33ac4 | 1435 | /** |
3041a069 | 1436 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1437 | * @dev: device to cause notification |
1438 | * | |
1439 | * Called to indicate a device has changed features. | |
1440 | */ | |
1441 | void netdev_features_change(struct net_device *dev) | |
1442 | { | |
056925ab | 1443 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1444 | } |
1445 | EXPORT_SYMBOL(netdev_features_change); | |
1446 | ||
1da177e4 LT |
1447 | /** |
1448 | * netdev_state_change - device changes state | |
1449 | * @dev: device to cause notification | |
1450 | * | |
1451 | * Called to indicate a device has changed state. This function calls | |
1452 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1453 | * to the routing socket. | |
1454 | */ | |
1455 | void netdev_state_change(struct net_device *dev) | |
1456 | { | |
1457 | if (dev->flags & IFF_UP) { | |
51d0c047 DA |
1458 | struct netdev_notifier_change_info change_info = { |
1459 | .info.dev = dev, | |
1460 | }; | |
54951194 | 1461 | |
51d0c047 | 1462 | call_netdevice_notifiers_info(NETDEV_CHANGE, |
54951194 | 1463 | &change_info.info); |
7f294054 | 1464 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1465 | } |
1466 | } | |
d1b19dff | 1467 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1468 | |
ee89bab1 | 1469 | /** |
722c9a0c | 1470 | * netdev_notify_peers - notify network peers about existence of @dev |
1471 | * @dev: network device | |
ee89bab1 AW |
1472 | * |
1473 | * Generate traffic such that interested network peers are aware of | |
1474 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1475 | * a device wants to inform the rest of the network about some sort of | |
1476 | * reconfiguration such as a failover event or virtual machine | |
1477 | * migration. | |
1478 | */ | |
1479 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1480 | { |
ee89bab1 AW |
1481 | rtnl_lock(); |
1482 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
37c343b4 | 1483 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); |
ee89bab1 | 1484 | rtnl_unlock(); |
c1da4ac7 | 1485 | } |
ee89bab1 | 1486 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1487 | |
40c900aa | 1488 | static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
1da177e4 | 1489 | { |
d314774c | 1490 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1491 | int ret; |
1da177e4 | 1492 | |
e46b66bc BH |
1493 | ASSERT_RTNL(); |
1494 | ||
1da177e4 LT |
1495 | if (!netif_device_present(dev)) |
1496 | return -ENODEV; | |
1497 | ||
ca99ca14 NH |
1498 | /* Block netpoll from trying to do any rx path servicing. |
1499 | * If we don't do this there is a chance ndo_poll_controller | |
1500 | * or ndo_poll may be running while we open the device | |
1501 | */ | |
66b5552f | 1502 | netpoll_poll_disable(dev); |
ca99ca14 | 1503 | |
40c900aa | 1504 | ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); |
3b8bcfd5 JB |
1505 | ret = notifier_to_errno(ret); |
1506 | if (ret) | |
1507 | return ret; | |
1508 | ||
1da177e4 | 1509 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1510 | |
d314774c SH |
1511 | if (ops->ndo_validate_addr) |
1512 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1513 | |
d314774c SH |
1514 | if (!ret && ops->ndo_open) |
1515 | ret = ops->ndo_open(dev); | |
1da177e4 | 1516 | |
66b5552f | 1517 | netpoll_poll_enable(dev); |
ca99ca14 | 1518 | |
bada339b JG |
1519 | if (ret) |
1520 | clear_bit(__LINK_STATE_START, &dev->state); | |
1521 | else { | |
1da177e4 | 1522 | dev->flags |= IFF_UP; |
4417da66 | 1523 | dev_set_rx_mode(dev); |
1da177e4 | 1524 | dev_activate(dev); |
7bf23575 | 1525 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1526 | } |
bada339b | 1527 | |
1da177e4 LT |
1528 | return ret; |
1529 | } | |
1530 | ||
1531 | /** | |
bd380811 | 1532 | * dev_open - prepare an interface for use. |
00f54e68 PM |
1533 | * @dev: device to open |
1534 | * @extack: netlink extended ack | |
1da177e4 | 1535 | * |
bd380811 PM |
1536 | * Takes a device from down to up state. The device's private open |
1537 | * function is invoked and then the multicast lists are loaded. Finally | |
1538 | * the device is moved into the up state and a %NETDEV_UP message is | |
1539 | * sent to the netdev notifier chain. | |
1540 | * | |
1541 | * Calling this function on an active interface is a nop. On a failure | |
1542 | * a negative errno code is returned. | |
1da177e4 | 1543 | */ |
00f54e68 | 1544 | int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
bd380811 PM |
1545 | { |
1546 | int ret; | |
1547 | ||
bd380811 PM |
1548 | if (dev->flags & IFF_UP) |
1549 | return 0; | |
1550 | ||
40c900aa | 1551 | ret = __dev_open(dev, extack); |
bd380811 PM |
1552 | if (ret < 0) |
1553 | return ret; | |
1554 | ||
7f294054 | 1555 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1556 | call_netdevice_notifiers(NETDEV_UP, dev); |
1557 | ||
1558 | return ret; | |
1559 | } | |
1560 | EXPORT_SYMBOL(dev_open); | |
1561 | ||
7051b88a | 1562 | static void __dev_close_many(struct list_head *head) |
1da177e4 | 1563 | { |
44345724 | 1564 | struct net_device *dev; |
e46b66bc | 1565 | |
bd380811 | 1566 | ASSERT_RTNL(); |
9d5010db DM |
1567 | might_sleep(); |
1568 | ||
5cde2829 | 1569 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1570 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1571 | netpoll_poll_disable(dev); |
3f4df206 | 1572 | |
44345724 | 1573 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1574 | |
44345724 | 1575 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1576 | |
44345724 OP |
1577 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1578 | * can be even on different cpu. So just clear netif_running(). | |
1579 | * | |
1580 | * dev->stop() will invoke napi_disable() on all of it's | |
1581 | * napi_struct instances on this device. | |
1582 | */ | |
4e857c58 | 1583 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1584 | } |
1da177e4 | 1585 | |
44345724 | 1586 | dev_deactivate_many(head); |
d8b2a4d2 | 1587 | |
5cde2829 | 1588 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1589 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1590 | |
44345724 OP |
1591 | /* |
1592 | * Call the device specific close. This cannot fail. | |
1593 | * Only if device is UP | |
1594 | * | |
1595 | * We allow it to be called even after a DETACH hot-plug | |
1596 | * event. | |
1597 | */ | |
1598 | if (ops->ndo_stop) | |
1599 | ops->ndo_stop(dev); | |
1600 | ||
44345724 | 1601 | dev->flags &= ~IFF_UP; |
66b5552f | 1602 | netpoll_poll_enable(dev); |
44345724 | 1603 | } |
44345724 OP |
1604 | } |
1605 | ||
7051b88a | 1606 | static void __dev_close(struct net_device *dev) |
44345724 OP |
1607 | { |
1608 | LIST_HEAD(single); | |
1609 | ||
5cde2829 | 1610 | list_add(&dev->close_list, &single); |
7051b88a | 1611 | __dev_close_many(&single); |
f87e6f47 | 1612 | list_del(&single); |
44345724 OP |
1613 | } |
1614 | ||
7051b88a | 1615 | void dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1616 | { |
1617 | struct net_device *dev, *tmp; | |
1da177e4 | 1618 | |
5cde2829 EB |
1619 | /* Remove the devices that don't need to be closed */ |
1620 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1621 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1622 | list_del_init(&dev->close_list); |
44345724 OP |
1623 | |
1624 | __dev_close_many(head); | |
1da177e4 | 1625 | |
5cde2829 | 1626 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1627 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1628 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1629 | if (unlink) |
1630 | list_del_init(&dev->close_list); | |
44345724 | 1631 | } |
bd380811 | 1632 | } |
99c4a26a | 1633 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1634 | |
1635 | /** | |
1636 | * dev_close - shutdown an interface. | |
1637 | * @dev: device to shutdown | |
1638 | * | |
1639 | * This function moves an active device into down state. A | |
1640 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1641 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1642 | * chain. | |
1643 | */ | |
7051b88a | 1644 | void dev_close(struct net_device *dev) |
bd380811 | 1645 | { |
e14a5993 ED |
1646 | if (dev->flags & IFF_UP) { |
1647 | LIST_HEAD(single); | |
1da177e4 | 1648 | |
5cde2829 | 1649 | list_add(&dev->close_list, &single); |
99c4a26a | 1650 | dev_close_many(&single, true); |
e14a5993 ED |
1651 | list_del(&single); |
1652 | } | |
1da177e4 | 1653 | } |
d1b19dff | 1654 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1655 | |
1656 | ||
0187bdfb BH |
1657 | /** |
1658 | * dev_disable_lro - disable Large Receive Offload on a device | |
1659 | * @dev: device | |
1660 | * | |
1661 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1662 | * called under RTNL. This is needed if received packets may be | |
1663 | * forwarded to another interface. | |
1664 | */ | |
1665 | void dev_disable_lro(struct net_device *dev) | |
1666 | { | |
fbe168ba MK |
1667 | struct net_device *lower_dev; |
1668 | struct list_head *iter; | |
529d0489 | 1669 | |
bc5787c6 MM |
1670 | dev->wanted_features &= ~NETIF_F_LRO; |
1671 | netdev_update_features(dev); | |
27660515 | 1672 | |
22d5969f MM |
1673 | if (unlikely(dev->features & NETIF_F_LRO)) |
1674 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1675 | |
1676 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1677 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1678 | } |
1679 | EXPORT_SYMBOL(dev_disable_lro); | |
1680 | ||
56f5aa77 MC |
1681 | /** |
1682 | * dev_disable_gro_hw - disable HW Generic Receive Offload on a device | |
1683 | * @dev: device | |
1684 | * | |
1685 | * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be | |
1686 | * called under RTNL. This is needed if Generic XDP is installed on | |
1687 | * the device. | |
1688 | */ | |
1689 | static void dev_disable_gro_hw(struct net_device *dev) | |
1690 | { | |
1691 | dev->wanted_features &= ~NETIF_F_GRO_HW; | |
1692 | netdev_update_features(dev); | |
1693 | ||
1694 | if (unlikely(dev->features & NETIF_F_GRO_HW)) | |
1695 | netdev_WARN(dev, "failed to disable GRO_HW!\n"); | |
1696 | } | |
1697 | ||
ede2762d KT |
1698 | const char *netdev_cmd_to_name(enum netdev_cmd cmd) |
1699 | { | |
1700 | #define N(val) \ | |
1701 | case NETDEV_##val: \ | |
1702 | return "NETDEV_" __stringify(val); | |
1703 | switch (cmd) { | |
1704 | N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) | |
1705 | N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) | |
1706 | N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) | |
1707 | N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) | |
1708 | N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) | |
1709 | N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) | |
1710 | N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) | |
9daae9bd GP |
1711 | N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) |
1712 | N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) | |
1570415f | 1713 | N(PRE_CHANGEADDR) |
3f5ecd8a | 1714 | } |
ede2762d KT |
1715 | #undef N |
1716 | return "UNKNOWN_NETDEV_EVENT"; | |
1717 | } | |
1718 | EXPORT_SYMBOL_GPL(netdev_cmd_to_name); | |
1719 | ||
351638e7 JP |
1720 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1721 | struct net_device *dev) | |
1722 | { | |
51d0c047 DA |
1723 | struct netdev_notifier_info info = { |
1724 | .dev = dev, | |
1725 | }; | |
351638e7 | 1726 | |
351638e7 JP |
1727 | return nb->notifier_call(nb, val, &info); |
1728 | } | |
0187bdfb | 1729 | |
afa0df59 JP |
1730 | static int call_netdevice_register_notifiers(struct notifier_block *nb, |
1731 | struct net_device *dev) | |
1732 | { | |
1733 | int err; | |
1734 | ||
1735 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); | |
1736 | err = notifier_to_errno(err); | |
1737 | if (err) | |
1738 | return err; | |
1739 | ||
1740 | if (!(dev->flags & IFF_UP)) | |
1741 | return 0; | |
1742 | ||
1743 | call_netdevice_notifier(nb, NETDEV_UP, dev); | |
1744 | return 0; | |
1745 | } | |
1746 | ||
1747 | static void call_netdevice_unregister_notifiers(struct notifier_block *nb, | |
1748 | struct net_device *dev) | |
1749 | { | |
1750 | if (dev->flags & IFF_UP) { | |
1751 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, | |
1752 | dev); | |
1753 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
1754 | } | |
1755 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); | |
1756 | } | |
1757 | ||
1758 | static int call_netdevice_register_net_notifiers(struct notifier_block *nb, | |
1759 | struct net *net) | |
1760 | { | |
1761 | struct net_device *dev; | |
1762 | int err; | |
1763 | ||
1764 | for_each_netdev(net, dev) { | |
1765 | err = call_netdevice_register_notifiers(nb, dev); | |
1766 | if (err) | |
1767 | goto rollback; | |
1768 | } | |
1769 | return 0; | |
1770 | ||
1771 | rollback: | |
1772 | for_each_netdev_continue_reverse(net, dev) | |
1773 | call_netdevice_unregister_notifiers(nb, dev); | |
1774 | return err; | |
1775 | } | |
1776 | ||
1777 | static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, | |
1778 | struct net *net) | |
1779 | { | |
1780 | struct net_device *dev; | |
1781 | ||
1782 | for_each_netdev(net, dev) | |
1783 | call_netdevice_unregister_notifiers(nb, dev); | |
1784 | } | |
1785 | ||
881d966b EB |
1786 | static int dev_boot_phase = 1; |
1787 | ||
1da177e4 | 1788 | /** |
722c9a0c | 1789 | * register_netdevice_notifier - register a network notifier block |
1790 | * @nb: notifier | |
1da177e4 | 1791 | * |
722c9a0c | 1792 | * Register a notifier to be called when network device events occur. |
1793 | * The notifier passed is linked into the kernel structures and must | |
1794 | * not be reused until it has been unregistered. A negative errno code | |
1795 | * is returned on a failure. | |
1da177e4 | 1796 | * |
722c9a0c | 1797 | * When registered all registration and up events are replayed |
1798 | * to the new notifier to allow device to have a race free | |
1799 | * view of the network device list. | |
1da177e4 LT |
1800 | */ |
1801 | ||
1802 | int register_netdevice_notifier(struct notifier_block *nb) | |
1803 | { | |
881d966b | 1804 | struct net *net; |
1da177e4 LT |
1805 | int err; |
1806 | ||
328fbe74 KT |
1807 | /* Close race with setup_net() and cleanup_net() */ |
1808 | down_write(&pernet_ops_rwsem); | |
1da177e4 | 1809 | rtnl_lock(); |
f07d5b94 | 1810 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1811 | if (err) |
1812 | goto unlock; | |
881d966b EB |
1813 | if (dev_boot_phase) |
1814 | goto unlock; | |
1815 | for_each_net(net) { | |
afa0df59 JP |
1816 | err = call_netdevice_register_net_notifiers(nb, net); |
1817 | if (err) | |
1818 | goto rollback; | |
1da177e4 | 1819 | } |
fcc5a03a HX |
1820 | |
1821 | unlock: | |
1da177e4 | 1822 | rtnl_unlock(); |
328fbe74 | 1823 | up_write(&pernet_ops_rwsem); |
1da177e4 | 1824 | return err; |
fcc5a03a HX |
1825 | |
1826 | rollback: | |
afa0df59 JP |
1827 | for_each_net_continue_reverse(net) |
1828 | call_netdevice_unregister_net_notifiers(nb, net); | |
c67625a1 PE |
1829 | |
1830 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1831 | goto unlock; |
1da177e4 | 1832 | } |
d1b19dff | 1833 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1834 | |
1835 | /** | |
722c9a0c | 1836 | * unregister_netdevice_notifier - unregister a network notifier block |
1837 | * @nb: notifier | |
1da177e4 | 1838 | * |
722c9a0c | 1839 | * Unregister a notifier previously registered by |
1840 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1841 | * kernel structures and may then be reused. A negative errno code | |
1842 | * is returned on a failure. | |
7d3d43da | 1843 | * |
722c9a0c | 1844 | * After unregistering unregister and down device events are synthesized |
1845 | * for all devices on the device list to the removed notifier to remove | |
1846 | * the need for special case cleanup code. | |
1da177e4 LT |
1847 | */ |
1848 | ||
1849 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1850 | { | |
7d3d43da | 1851 | struct net *net; |
9f514950 HX |
1852 | int err; |
1853 | ||
328fbe74 KT |
1854 | /* Close race with setup_net() and cleanup_net() */ |
1855 | down_write(&pernet_ops_rwsem); | |
9f514950 | 1856 | rtnl_lock(); |
f07d5b94 | 1857 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1858 | if (err) |
1859 | goto unlock; | |
1860 | ||
48b3a137 JP |
1861 | for_each_net(net) |
1862 | call_netdevice_unregister_net_notifiers(nb, net); | |
1863 | ||
7d3d43da | 1864 | unlock: |
9f514950 | 1865 | rtnl_unlock(); |
328fbe74 | 1866 | up_write(&pernet_ops_rwsem); |
9f514950 | 1867 | return err; |
1da177e4 | 1868 | } |
d1b19dff | 1869 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1870 | |
1f637703 JP |
1871 | static int __register_netdevice_notifier_net(struct net *net, |
1872 | struct notifier_block *nb, | |
1873 | bool ignore_call_fail) | |
1874 | { | |
1875 | int err; | |
1876 | ||
1877 | err = raw_notifier_chain_register(&net->netdev_chain, nb); | |
1878 | if (err) | |
1879 | return err; | |
1880 | if (dev_boot_phase) | |
1881 | return 0; | |
1882 | ||
1883 | err = call_netdevice_register_net_notifiers(nb, net); | |
1884 | if (err && !ignore_call_fail) | |
1885 | goto chain_unregister; | |
1886 | ||
1887 | return 0; | |
1888 | ||
1889 | chain_unregister: | |
1890 | raw_notifier_chain_unregister(&net->netdev_chain, nb); | |
1891 | return err; | |
1892 | } | |
1893 | ||
1894 | static int __unregister_netdevice_notifier_net(struct net *net, | |
1895 | struct notifier_block *nb) | |
1896 | { | |
1897 | int err; | |
1898 | ||
1899 | err = raw_notifier_chain_unregister(&net->netdev_chain, nb); | |
1900 | if (err) | |
1901 | return err; | |
1902 | ||
1903 | call_netdevice_unregister_net_notifiers(nb, net); | |
1904 | return 0; | |
1905 | } | |
1906 | ||
a30c7b42 JP |
1907 | /** |
1908 | * register_netdevice_notifier_net - register a per-netns network notifier block | |
1909 | * @net: network namespace | |
1910 | * @nb: notifier | |
1911 | * | |
1912 | * Register a notifier to be called when network device events occur. | |
1913 | * The notifier passed is linked into the kernel structures and must | |
1914 | * not be reused until it has been unregistered. A negative errno code | |
1915 | * is returned on a failure. | |
1916 | * | |
1917 | * When registered all registration and up events are replayed | |
1918 | * to the new notifier to allow device to have a race free | |
1919 | * view of the network device list. | |
1920 | */ | |
1921 | ||
1922 | int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) | |
1923 | { | |
1924 | int err; | |
1925 | ||
1926 | rtnl_lock(); | |
1f637703 | 1927 | err = __register_netdevice_notifier_net(net, nb, false); |
a30c7b42 JP |
1928 | rtnl_unlock(); |
1929 | return err; | |
a30c7b42 JP |
1930 | } |
1931 | EXPORT_SYMBOL(register_netdevice_notifier_net); | |
1932 | ||
1933 | /** | |
1934 | * unregister_netdevice_notifier_net - unregister a per-netns | |
1935 | * network notifier block | |
1936 | * @net: network namespace | |
1937 | * @nb: notifier | |
1938 | * | |
1939 | * Unregister a notifier previously registered by | |
1940 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1941 | * kernel structures and may then be reused. A negative errno code | |
1942 | * is returned on a failure. | |
1943 | * | |
1944 | * After unregistering unregister and down device events are synthesized | |
1945 | * for all devices on the device list to the removed notifier to remove | |
1946 | * the need for special case cleanup code. | |
1947 | */ | |
1948 | ||
1949 | int unregister_netdevice_notifier_net(struct net *net, | |
1950 | struct notifier_block *nb) | |
1951 | { | |
1952 | int err; | |
1953 | ||
1954 | rtnl_lock(); | |
1f637703 | 1955 | err = __unregister_netdevice_notifier_net(net, nb); |
a30c7b42 JP |
1956 | rtnl_unlock(); |
1957 | return err; | |
1958 | } | |
1959 | EXPORT_SYMBOL(unregister_netdevice_notifier_net); | |
a30c7b42 | 1960 | |
93642e14 JP |
1961 | int register_netdevice_notifier_dev_net(struct net_device *dev, |
1962 | struct notifier_block *nb, | |
1963 | struct netdev_net_notifier *nn) | |
1964 | { | |
1965 | int err; | |
a30c7b42 | 1966 | |
93642e14 JP |
1967 | rtnl_lock(); |
1968 | err = __register_netdevice_notifier_net(dev_net(dev), nb, false); | |
1969 | if (!err) { | |
1970 | nn->nb = nb; | |
1971 | list_add(&nn->list, &dev->net_notifier_list); | |
1972 | } | |
a30c7b42 JP |
1973 | rtnl_unlock(); |
1974 | return err; | |
1975 | } | |
93642e14 JP |
1976 | EXPORT_SYMBOL(register_netdevice_notifier_dev_net); |
1977 | ||
1978 | int unregister_netdevice_notifier_dev_net(struct net_device *dev, | |
1979 | struct notifier_block *nb, | |
1980 | struct netdev_net_notifier *nn) | |
1981 | { | |
1982 | int err; | |
1983 | ||
1984 | rtnl_lock(); | |
1985 | list_del(&nn->list); | |
1986 | err = __unregister_netdevice_notifier_net(dev_net(dev), nb); | |
1987 | rtnl_unlock(); | |
1988 | return err; | |
1989 | } | |
1990 | EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); | |
1991 | ||
1992 | static void move_netdevice_notifiers_dev_net(struct net_device *dev, | |
1993 | struct net *net) | |
1994 | { | |
1995 | struct netdev_net_notifier *nn; | |
1996 | ||
1997 | list_for_each_entry(nn, &dev->net_notifier_list, list) { | |
1998 | __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); | |
1999 | __register_netdevice_notifier_net(net, nn->nb, true); | |
2000 | } | |
2001 | } | |
a30c7b42 | 2002 | |
351638e7 JP |
2003 | /** |
2004 | * call_netdevice_notifiers_info - call all network notifier blocks | |
2005 | * @val: value passed unmodified to notifier function | |
351638e7 JP |
2006 | * @info: notifier information data |
2007 | * | |
2008 | * Call all network notifier blocks. Parameters and return value | |
2009 | * are as for raw_notifier_call_chain(). | |
2010 | */ | |
2011 | ||
1d143d9f | 2012 | static int call_netdevice_notifiers_info(unsigned long val, |
1d143d9f | 2013 | struct netdev_notifier_info *info) |
351638e7 | 2014 | { |
a30c7b42 JP |
2015 | struct net *net = dev_net(info->dev); |
2016 | int ret; | |
2017 | ||
351638e7 | 2018 | ASSERT_RTNL(); |
a30c7b42 JP |
2019 | |
2020 | /* Run per-netns notifier block chain first, then run the global one. | |
2021 | * Hopefully, one day, the global one is going to be removed after | |
2022 | * all notifier block registrators get converted to be per-netns. | |
2023 | */ | |
2024 | ret = raw_notifier_call_chain(&net->netdev_chain, val, info); | |
2025 | if (ret & NOTIFY_STOP_MASK) | |
2026 | return ret; | |
351638e7 JP |
2027 | return raw_notifier_call_chain(&netdev_chain, val, info); |
2028 | } | |
351638e7 | 2029 | |
26372605 PM |
2030 | static int call_netdevice_notifiers_extack(unsigned long val, |
2031 | struct net_device *dev, | |
2032 | struct netlink_ext_ack *extack) | |
2033 | { | |
2034 | struct netdev_notifier_info info = { | |
2035 | .dev = dev, | |
2036 | .extack = extack, | |
2037 | }; | |
2038 | ||
2039 | return call_netdevice_notifiers_info(val, &info); | |
2040 | } | |
2041 | ||
1da177e4 LT |
2042 | /** |
2043 | * call_netdevice_notifiers - call all network notifier blocks | |
2044 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 2045 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
2046 | * |
2047 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 2048 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
2049 | */ |
2050 | ||
ad7379d4 | 2051 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 2052 | { |
26372605 | 2053 | return call_netdevice_notifiers_extack(val, dev, NULL); |
1da177e4 | 2054 | } |
edf947f1 | 2055 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 2056 | |
af7d6cce SD |
2057 | /** |
2058 | * call_netdevice_notifiers_mtu - call all network notifier blocks | |
2059 | * @val: value passed unmodified to notifier function | |
2060 | * @dev: net_device pointer passed unmodified to notifier function | |
2061 | * @arg: additional u32 argument passed to the notifier function | |
2062 | * | |
2063 | * Call all network notifier blocks. Parameters and return value | |
2064 | * are as for raw_notifier_call_chain(). | |
2065 | */ | |
2066 | static int call_netdevice_notifiers_mtu(unsigned long val, | |
2067 | struct net_device *dev, u32 arg) | |
2068 | { | |
2069 | struct netdev_notifier_info_ext info = { | |
2070 | .info.dev = dev, | |
2071 | .ext.mtu = arg, | |
2072 | }; | |
2073 | ||
2074 | BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); | |
2075 | ||
2076 | return call_netdevice_notifiers_info(val, &info.info); | |
2077 | } | |
2078 | ||
1cf51900 | 2079 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 2080 | static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); |
4577139b DB |
2081 | |
2082 | void net_inc_ingress_queue(void) | |
2083 | { | |
aabf6772 | 2084 | static_branch_inc(&ingress_needed_key); |
4577139b DB |
2085 | } |
2086 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
2087 | ||
2088 | void net_dec_ingress_queue(void) | |
2089 | { | |
aabf6772 | 2090 | static_branch_dec(&ingress_needed_key); |
4577139b DB |
2091 | } |
2092 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
2093 | #endif | |
2094 | ||
1f211a1b | 2095 | #ifdef CONFIG_NET_EGRESS |
aabf6772 | 2096 | static DEFINE_STATIC_KEY_FALSE(egress_needed_key); |
1f211a1b DB |
2097 | |
2098 | void net_inc_egress_queue(void) | |
2099 | { | |
aabf6772 | 2100 | static_branch_inc(&egress_needed_key); |
1f211a1b DB |
2101 | } |
2102 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
2103 | ||
2104 | void net_dec_egress_queue(void) | |
2105 | { | |
aabf6772 | 2106 | static_branch_dec(&egress_needed_key); |
1f211a1b DB |
2107 | } |
2108 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
2109 | #endif | |
2110 | ||
39e83922 | 2111 | static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); |
e9666d10 | 2112 | #ifdef CONFIG_JUMP_LABEL |
b90e5794 | 2113 | static atomic_t netstamp_needed_deferred; |
13baa00a | 2114 | static atomic_t netstamp_wanted; |
5fa8bbda | 2115 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 2116 | { |
b90e5794 | 2117 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 2118 | int wanted; |
b90e5794 | 2119 | |
13baa00a ED |
2120 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
2121 | if (wanted > 0) | |
39e83922 | 2122 | static_branch_enable(&netstamp_needed_key); |
13baa00a | 2123 | else |
39e83922 | 2124 | static_branch_disable(&netstamp_needed_key); |
5fa8bbda ED |
2125 | } |
2126 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 2127 | #endif |
5fa8bbda ED |
2128 | |
2129 | void net_enable_timestamp(void) | |
2130 | { | |
e9666d10 | 2131 | #ifdef CONFIG_JUMP_LABEL |
13baa00a ED |
2132 | int wanted; |
2133 | ||
2134 | while (1) { | |
2135 | wanted = atomic_read(&netstamp_wanted); | |
2136 | if (wanted <= 0) | |
2137 | break; | |
2138 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) | |
2139 | return; | |
2140 | } | |
2141 | atomic_inc(&netstamp_needed_deferred); | |
2142 | schedule_work(&netstamp_work); | |
2143 | #else | |
39e83922 | 2144 | static_branch_inc(&netstamp_needed_key); |
13baa00a | 2145 | #endif |
1da177e4 | 2146 | } |
d1b19dff | 2147 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
2148 | |
2149 | void net_disable_timestamp(void) | |
2150 | { | |
e9666d10 | 2151 | #ifdef CONFIG_JUMP_LABEL |
13baa00a ED |
2152 | int wanted; |
2153 | ||
2154 | while (1) { | |
2155 | wanted = atomic_read(&netstamp_wanted); | |
2156 | if (wanted <= 1) | |
2157 | break; | |
2158 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) | |
2159 | return; | |
2160 | } | |
2161 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
2162 | schedule_work(&netstamp_work); |
2163 | #else | |
39e83922 | 2164 | static_branch_dec(&netstamp_needed_key); |
5fa8bbda | 2165 | #endif |
1da177e4 | 2166 | } |
d1b19dff | 2167 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 2168 | |
3b098e2d | 2169 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 2170 | { |
2456e855 | 2171 | skb->tstamp = 0; |
39e83922 | 2172 | if (static_branch_unlikely(&netstamp_needed_key)) |
a61bbcf2 | 2173 | __net_timestamp(skb); |
1da177e4 LT |
2174 | } |
2175 | ||
39e83922 DB |
2176 | #define net_timestamp_check(COND, SKB) \ |
2177 | if (static_branch_unlikely(&netstamp_needed_key)) { \ | |
2178 | if ((COND) && !(SKB)->tstamp) \ | |
2179 | __net_timestamp(SKB); \ | |
2180 | } \ | |
3b098e2d | 2181 | |
f4b05d27 | 2182 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 DL |
2183 | { |
2184 | unsigned int len; | |
2185 | ||
2186 | if (!(dev->flags & IFF_UP)) | |
2187 | return false; | |
2188 | ||
2189 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
2190 | if (skb->len <= len) | |
2191 | return true; | |
2192 | ||
2193 | /* if TSO is enabled, we don't care about the length as the packet | |
2194 | * could be forwarded without being segmented before | |
2195 | */ | |
2196 | if (skb_is_gso(skb)) | |
2197 | return true; | |
2198 | ||
2199 | return false; | |
2200 | } | |
1ee481fb | 2201 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 2202 | |
a0265d28 HX |
2203 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
2204 | { | |
4e3264d2 | 2205 | int ret = ____dev_forward_skb(dev, skb); |
a0265d28 | 2206 | |
4e3264d2 MKL |
2207 | if (likely(!ret)) { |
2208 | skb->protocol = eth_type_trans(skb, dev); | |
2209 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
2210 | } | |
a0265d28 | 2211 | |
4e3264d2 | 2212 | return ret; |
a0265d28 HX |
2213 | } |
2214 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
2215 | ||
44540960 AB |
2216 | /** |
2217 | * dev_forward_skb - loopback an skb to another netif | |
2218 | * | |
2219 | * @dev: destination network device | |
2220 | * @skb: buffer to forward | |
2221 | * | |
2222 | * return values: | |
2223 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 2224 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
2225 | * |
2226 | * dev_forward_skb can be used for injecting an skb from the | |
2227 | * start_xmit function of one device into the receive queue | |
2228 | * of another device. | |
2229 | * | |
2230 | * The receiving device may be in another namespace, so | |
2231 | * we have to clear all information in the skb that could | |
2232 | * impact namespace isolation. | |
2233 | */ | |
2234 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
2235 | { | |
a0265d28 | 2236 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
2237 | } |
2238 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
2239 | ||
71d9dec2 CG |
2240 | static inline int deliver_skb(struct sk_buff *skb, |
2241 | struct packet_type *pt_prev, | |
2242 | struct net_device *orig_dev) | |
2243 | { | |
1f8b977a | 2244 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
1080e512 | 2245 | return -ENOMEM; |
63354797 | 2246 | refcount_inc(&skb->users); |
71d9dec2 CG |
2247 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
2248 | } | |
2249 | ||
7866a621 SN |
2250 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
2251 | struct packet_type **pt, | |
fbcb2170 JP |
2252 | struct net_device *orig_dev, |
2253 | __be16 type, | |
7866a621 SN |
2254 | struct list_head *ptype_list) |
2255 | { | |
2256 | struct packet_type *ptype, *pt_prev = *pt; | |
2257 | ||
2258 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
2259 | if (ptype->type != type) | |
2260 | continue; | |
2261 | if (pt_prev) | |
fbcb2170 | 2262 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
2263 | pt_prev = ptype; |
2264 | } | |
2265 | *pt = pt_prev; | |
2266 | } | |
2267 | ||
c0de08d0 EL |
2268 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
2269 | { | |
a3d744e9 | 2270 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
2271 | return false; |
2272 | ||
2273 | if (ptype->id_match) | |
2274 | return ptype->id_match(ptype, skb->sk); | |
2275 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
2276 | return true; | |
2277 | ||
2278 | return false; | |
2279 | } | |
2280 | ||
9f9a742d MR |
2281 | /** |
2282 | * dev_nit_active - return true if any network interface taps are in use | |
2283 | * | |
2284 | * @dev: network device to check for the presence of taps | |
2285 | */ | |
2286 | bool dev_nit_active(struct net_device *dev) | |
2287 | { | |
2288 | return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); | |
2289 | } | |
2290 | EXPORT_SYMBOL_GPL(dev_nit_active); | |
2291 | ||
1da177e4 LT |
2292 | /* |
2293 | * Support routine. Sends outgoing frames to any network | |
2294 | * taps currently in use. | |
2295 | */ | |
2296 | ||
74b20582 | 2297 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
2298 | { |
2299 | struct packet_type *ptype; | |
71d9dec2 CG |
2300 | struct sk_buff *skb2 = NULL; |
2301 | struct packet_type *pt_prev = NULL; | |
7866a621 | 2302 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 2303 | |
1da177e4 | 2304 | rcu_read_lock(); |
7866a621 SN |
2305 | again: |
2306 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
fa788d98 VW |
2307 | if (ptype->ignore_outgoing) |
2308 | continue; | |
2309 | ||
1da177e4 LT |
2310 | /* Never send packets back to the socket |
2311 | * they originated from - MvS (miquels@drinkel.ow.org) | |
2312 | */ | |
7866a621 SN |
2313 | if (skb_loop_sk(ptype, skb)) |
2314 | continue; | |
71d9dec2 | 2315 | |
7866a621 SN |
2316 | if (pt_prev) { |
2317 | deliver_skb(skb2, pt_prev, skb->dev); | |
2318 | pt_prev = ptype; | |
2319 | continue; | |
2320 | } | |
1da177e4 | 2321 | |
7866a621 SN |
2322 | /* need to clone skb, done only once */ |
2323 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
2324 | if (!skb2) | |
2325 | goto out_unlock; | |
70978182 | 2326 | |
7866a621 | 2327 | net_timestamp_set(skb2); |
1da177e4 | 2328 | |
7866a621 SN |
2329 | /* skb->nh should be correctly |
2330 | * set by sender, so that the second statement is | |
2331 | * just protection against buggy protocols. | |
2332 | */ | |
2333 | skb_reset_mac_header(skb2); | |
2334 | ||
2335 | if (skb_network_header(skb2) < skb2->data || | |
2336 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
2337 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
2338 | ntohs(skb2->protocol), | |
2339 | dev->name); | |
2340 | skb_reset_network_header(skb2); | |
1da177e4 | 2341 | } |
7866a621 SN |
2342 | |
2343 | skb2->transport_header = skb2->network_header; | |
2344 | skb2->pkt_type = PACKET_OUTGOING; | |
2345 | pt_prev = ptype; | |
2346 | } | |
2347 | ||
2348 | if (ptype_list == &ptype_all) { | |
2349 | ptype_list = &dev->ptype_all; | |
2350 | goto again; | |
1da177e4 | 2351 | } |
7866a621 | 2352 | out_unlock: |
581fe0ea WB |
2353 | if (pt_prev) { |
2354 | if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) | |
2355 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
2356 | else | |
2357 | kfree_skb(skb2); | |
2358 | } | |
1da177e4 LT |
2359 | rcu_read_unlock(); |
2360 | } | |
74b20582 | 2361 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 2362 | |
2c53040f BH |
2363 | /** |
2364 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
2365 | * @dev: Network device |
2366 | * @txq: number of queues available | |
2367 | * | |
2368 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
2369 | * valid. To resolve this verify the tc mapping remains valid and if | |
2370 | * not NULL the mapping. With no priorities mapping to this | |
2371 | * offset/count pair it will no longer be used. In the worst case TC0 | |
2372 | * is invalid nothing can be done so disable priority mappings. If is | |
2373 | * expected that drivers will fix this mapping if they can before | |
2374 | * calling netif_set_real_num_tx_queues. | |
2375 | */ | |
bb134d22 | 2376 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
2377 | { |
2378 | int i; | |
2379 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2380 | ||
2381 | /* If TC0 is invalidated disable TC mapping */ | |
2382 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 2383 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
2384 | dev->num_tc = 0; |
2385 | return; | |
2386 | } | |
2387 | ||
2388 | /* Invalidated prio to tc mappings set to TC0 */ | |
2389 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
2390 | int q = netdev_get_prio_tc_map(dev, i); | |
2391 | ||
2392 | tc = &dev->tc_to_txq[q]; | |
2393 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
2394 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
2395 | i, q); | |
4f57c087 JF |
2396 | netdev_set_prio_tc_map(dev, i, 0); |
2397 | } | |
2398 | } | |
2399 | } | |
2400 | ||
8d059b0f AD |
2401 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
2402 | { | |
2403 | if (dev->num_tc) { | |
2404 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2405 | int i; | |
2406 | ||
ffcfe25b | 2407 | /* walk through the TCs and see if it falls into any of them */ |
8d059b0f AD |
2408 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { |
2409 | if ((txq - tc->offset) < tc->count) | |
2410 | return i; | |
2411 | } | |
2412 | ||
ffcfe25b | 2413 | /* didn't find it, just return -1 to indicate no match */ |
8d059b0f AD |
2414 | return -1; |
2415 | } | |
2416 | ||
2417 | return 0; | |
2418 | } | |
8a5f2166 | 2419 | EXPORT_SYMBOL(netdev_txq_to_tc); |
8d059b0f | 2420 | |
537c00de | 2421 | #ifdef CONFIG_XPS |
04157469 AN |
2422 | struct static_key xps_needed __read_mostly; |
2423 | EXPORT_SYMBOL(xps_needed); | |
2424 | struct static_key xps_rxqs_needed __read_mostly; | |
2425 | EXPORT_SYMBOL(xps_rxqs_needed); | |
537c00de AD |
2426 | static DEFINE_MUTEX(xps_map_mutex); |
2427 | #define xmap_dereference(P) \ | |
2428 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
2429 | ||
6234f874 AD |
2430 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
2431 | int tci, u16 index) | |
537c00de | 2432 | { |
10cdc3f3 AD |
2433 | struct xps_map *map = NULL; |
2434 | int pos; | |
537c00de | 2435 | |
10cdc3f3 | 2436 | if (dev_maps) |
80d19669 | 2437 | map = xmap_dereference(dev_maps->attr_map[tci]); |
6234f874 AD |
2438 | if (!map) |
2439 | return false; | |
537c00de | 2440 | |
6234f874 AD |
2441 | for (pos = map->len; pos--;) { |
2442 | if (map->queues[pos] != index) | |
2443 | continue; | |
2444 | ||
2445 | if (map->len > 1) { | |
2446 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2447 | break; |
537c00de | 2448 | } |
6234f874 | 2449 | |
80d19669 | 2450 | RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); |
6234f874 AD |
2451 | kfree_rcu(map, rcu); |
2452 | return false; | |
537c00de AD |
2453 | } |
2454 | ||
6234f874 | 2455 | return true; |
10cdc3f3 AD |
2456 | } |
2457 | ||
6234f874 AD |
2458 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2459 | struct xps_dev_maps *dev_maps, | |
2460 | int cpu, u16 offset, u16 count) | |
2461 | { | |
184c449f AD |
2462 | int num_tc = dev->num_tc ? : 1; |
2463 | bool active = false; | |
2464 | int tci; | |
6234f874 | 2465 | |
184c449f AD |
2466 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2467 | int i, j; | |
2468 | ||
2469 | for (i = count, j = offset; i--; j++) { | |
6358d49a | 2470 | if (!remove_xps_queue(dev_maps, tci, j)) |
184c449f AD |
2471 | break; |
2472 | } | |
2473 | ||
2474 | active |= i < 0; | |
6234f874 AD |
2475 | } |
2476 | ||
184c449f | 2477 | return active; |
6234f874 AD |
2478 | } |
2479 | ||
867d0ad4 SD |
2480 | static void reset_xps_maps(struct net_device *dev, |
2481 | struct xps_dev_maps *dev_maps, | |
2482 | bool is_rxqs_map) | |
2483 | { | |
2484 | if (is_rxqs_map) { | |
2485 | static_key_slow_dec_cpuslocked(&xps_rxqs_needed); | |
2486 | RCU_INIT_POINTER(dev->xps_rxqs_map, NULL); | |
2487 | } else { | |
2488 | RCU_INIT_POINTER(dev->xps_cpus_map, NULL); | |
2489 | } | |
2490 | static_key_slow_dec_cpuslocked(&xps_needed); | |
2491 | kfree_rcu(dev_maps, rcu); | |
2492 | } | |
2493 | ||
80d19669 AN |
2494 | static void clean_xps_maps(struct net_device *dev, const unsigned long *mask, |
2495 | struct xps_dev_maps *dev_maps, unsigned int nr_ids, | |
2496 | u16 offset, u16 count, bool is_rxqs_map) | |
2497 | { | |
2498 | bool active = false; | |
2499 | int i, j; | |
2500 | ||
2501 | for (j = -1; j = netif_attrmask_next(j, mask, nr_ids), | |
2502 | j < nr_ids;) | |
2503 | active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, | |
2504 | count); | |
867d0ad4 SD |
2505 | if (!active) |
2506 | reset_xps_maps(dev, dev_maps, is_rxqs_map); | |
80d19669 | 2507 | |
f28c020f SD |
2508 | if (!is_rxqs_map) { |
2509 | for (i = offset + (count - 1); count--; i--) { | |
2510 | netdev_queue_numa_node_write( | |
2511 | netdev_get_tx_queue(dev, i), | |
2512 | NUMA_NO_NODE); | |
80d19669 | 2513 | } |
80d19669 AN |
2514 | } |
2515 | } | |
2516 | ||
6234f874 AD |
2517 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, |
2518 | u16 count) | |
10cdc3f3 | 2519 | { |
80d19669 | 2520 | const unsigned long *possible_mask = NULL; |
10cdc3f3 | 2521 | struct xps_dev_maps *dev_maps; |
80d19669 | 2522 | unsigned int nr_ids; |
10cdc3f3 | 2523 | |
04157469 AN |
2524 | if (!static_key_false(&xps_needed)) |
2525 | return; | |
10cdc3f3 | 2526 | |
4d99f660 | 2527 | cpus_read_lock(); |
04157469 | 2528 | mutex_lock(&xps_map_mutex); |
10cdc3f3 | 2529 | |
04157469 AN |
2530 | if (static_key_false(&xps_rxqs_needed)) { |
2531 | dev_maps = xmap_dereference(dev->xps_rxqs_map); | |
2532 | if (dev_maps) { | |
2533 | nr_ids = dev->num_rx_queues; | |
2534 | clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, | |
2535 | offset, count, true); | |
2536 | } | |
537c00de AD |
2537 | } |
2538 | ||
80d19669 AN |
2539 | dev_maps = xmap_dereference(dev->xps_cpus_map); |
2540 | if (!dev_maps) | |
2541 | goto out_no_maps; | |
2542 | ||
2543 | if (num_possible_cpus() > 1) | |
2544 | possible_mask = cpumask_bits(cpu_possible_mask); | |
2545 | nr_ids = nr_cpu_ids; | |
2546 | clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count, | |
2547 | false); | |
024e9679 | 2548 | |
537c00de AD |
2549 | out_no_maps: |
2550 | mutex_unlock(&xps_map_mutex); | |
4d99f660 | 2551 | cpus_read_unlock(); |
537c00de AD |
2552 | } |
2553 | ||
6234f874 AD |
2554 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2555 | { | |
2556 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2557 | } | |
2558 | ||
80d19669 AN |
2559 | static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, |
2560 | u16 index, bool is_rxqs_map) | |
01c5f864 AD |
2561 | { |
2562 | struct xps_map *new_map; | |
2563 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2564 | int i, pos; | |
2565 | ||
2566 | for (pos = 0; map && pos < map->len; pos++) { | |
2567 | if (map->queues[pos] != index) | |
2568 | continue; | |
2569 | return map; | |
2570 | } | |
2571 | ||
80d19669 | 2572 | /* Need to add tx-queue to this CPU's/rx-queue's existing map */ |
01c5f864 AD |
2573 | if (map) { |
2574 | if (pos < map->alloc_len) | |
2575 | return map; | |
2576 | ||
2577 | alloc_len = map->alloc_len * 2; | |
2578 | } | |
2579 | ||
80d19669 AN |
2580 | /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's |
2581 | * map | |
2582 | */ | |
2583 | if (is_rxqs_map) | |
2584 | new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); | |
2585 | else | |
2586 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2587 | cpu_to_node(attr_index)); | |
01c5f864 AD |
2588 | if (!new_map) |
2589 | return NULL; | |
2590 | ||
2591 | for (i = 0; i < pos; i++) | |
2592 | new_map->queues[i] = map->queues[i]; | |
2593 | new_map->alloc_len = alloc_len; | |
2594 | new_map->len = pos; | |
2595 | ||
2596 | return new_map; | |
2597 | } | |
2598 | ||
4d99f660 | 2599 | /* Must be called under cpus_read_lock */ |
80d19669 AN |
2600 | int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, |
2601 | u16 index, bool is_rxqs_map) | |
537c00de | 2602 | { |
80d19669 | 2603 | const unsigned long *online_mask = NULL, *possible_mask = NULL; |
01c5f864 | 2604 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
80d19669 | 2605 | int i, j, tci, numa_node_id = -2; |
184c449f | 2606 | int maps_sz, num_tc = 1, tc = 0; |
537c00de | 2607 | struct xps_map *map, *new_map; |
01c5f864 | 2608 | bool active = false; |
80d19669 | 2609 | unsigned int nr_ids; |
537c00de | 2610 | |
184c449f | 2611 | if (dev->num_tc) { |
ffcfe25b | 2612 | /* Do not allow XPS on subordinate device directly */ |
184c449f | 2613 | num_tc = dev->num_tc; |
ffcfe25b AD |
2614 | if (num_tc < 0) |
2615 | return -EINVAL; | |
2616 | ||
2617 | /* If queue belongs to subordinate dev use its map */ | |
2618 | dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; | |
2619 | ||
184c449f AD |
2620 | tc = netdev_txq_to_tc(dev, index); |
2621 | if (tc < 0) | |
2622 | return -EINVAL; | |
2623 | } | |
2624 | ||
537c00de | 2625 | mutex_lock(&xps_map_mutex); |
80d19669 AN |
2626 | if (is_rxqs_map) { |
2627 | maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); | |
2628 | dev_maps = xmap_dereference(dev->xps_rxqs_map); | |
2629 | nr_ids = dev->num_rx_queues; | |
2630 | } else { | |
2631 | maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); | |
2632 | if (num_possible_cpus() > 1) { | |
2633 | online_mask = cpumask_bits(cpu_online_mask); | |
2634 | possible_mask = cpumask_bits(cpu_possible_mask); | |
2635 | } | |
2636 | dev_maps = xmap_dereference(dev->xps_cpus_map); | |
2637 | nr_ids = nr_cpu_ids; | |
2638 | } | |
537c00de | 2639 | |
80d19669 AN |
2640 | if (maps_sz < L1_CACHE_BYTES) |
2641 | maps_sz = L1_CACHE_BYTES; | |
537c00de | 2642 | |
01c5f864 | 2643 | /* allocate memory for queue storage */ |
80d19669 AN |
2644 | for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), |
2645 | j < nr_ids;) { | |
01c5f864 AD |
2646 | if (!new_dev_maps) |
2647 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2648 | if (!new_dev_maps) { |
2649 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2650 | return -ENOMEM; |
2bb60cb9 | 2651 | } |
01c5f864 | 2652 | |
80d19669 AN |
2653 | tci = j * num_tc + tc; |
2654 | map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) : | |
01c5f864 AD |
2655 | NULL; |
2656 | ||
80d19669 | 2657 | map = expand_xps_map(map, j, index, is_rxqs_map); |
01c5f864 AD |
2658 | if (!map) |
2659 | goto error; | |
2660 | ||
80d19669 | 2661 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); |
01c5f864 AD |
2662 | } |
2663 | ||
2664 | if (!new_dev_maps) | |
2665 | goto out_no_new_maps; | |
2666 | ||
867d0ad4 SD |
2667 | if (!dev_maps) { |
2668 | /* Increment static keys at most once per type */ | |
2669 | static_key_slow_inc_cpuslocked(&xps_needed); | |
2670 | if (is_rxqs_map) | |
2671 | static_key_slow_inc_cpuslocked(&xps_rxqs_needed); | |
2672 | } | |
04157469 | 2673 | |
80d19669 AN |
2674 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2675 | j < nr_ids;) { | |
184c449f | 2676 | /* copy maps belonging to foreign traffic classes */ |
80d19669 | 2677 | for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) { |
184c449f | 2678 | /* fill in the new device map from the old device map */ |
80d19669 AN |
2679 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2680 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
184c449f AD |
2681 | } |
2682 | ||
2683 | /* We need to explicitly update tci as prevous loop | |
2684 | * could break out early if dev_maps is NULL. | |
2685 | */ | |
80d19669 | 2686 | tci = j * num_tc + tc; |
184c449f | 2687 | |
80d19669 AN |
2688 | if (netif_attr_test_mask(j, mask, nr_ids) && |
2689 | netif_attr_test_online(j, online_mask, nr_ids)) { | |
2690 | /* add tx-queue to CPU/rx-queue maps */ | |
01c5f864 AD |
2691 | int pos = 0; |
2692 | ||
80d19669 | 2693 | map = xmap_dereference(new_dev_maps->attr_map[tci]); |
01c5f864 AD |
2694 | while ((pos < map->len) && (map->queues[pos] != index)) |
2695 | pos++; | |
2696 | ||
2697 | if (pos == map->len) | |
2698 | map->queues[map->len++] = index; | |
537c00de | 2699 | #ifdef CONFIG_NUMA |
80d19669 AN |
2700 | if (!is_rxqs_map) { |
2701 | if (numa_node_id == -2) | |
2702 | numa_node_id = cpu_to_node(j); | |
2703 | else if (numa_node_id != cpu_to_node(j)) | |
2704 | numa_node_id = -1; | |
2705 | } | |
537c00de | 2706 | #endif |
01c5f864 AD |
2707 | } else if (dev_maps) { |
2708 | /* fill in the new device map from the old device map */ | |
80d19669 AN |
2709 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2710 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
537c00de | 2711 | } |
01c5f864 | 2712 | |
184c449f AD |
2713 | /* copy maps belonging to foreign traffic classes */ |
2714 | for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { | |
2715 | /* fill in the new device map from the old device map */ | |
80d19669 AN |
2716 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2717 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
184c449f | 2718 | } |
537c00de AD |
2719 | } |
2720 | ||
80d19669 AN |
2721 | if (is_rxqs_map) |
2722 | rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps); | |
2723 | else | |
2724 | rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps); | |
01c5f864 | 2725 | |
537c00de | 2726 | /* Cleanup old maps */ |
184c449f AD |
2727 | if (!dev_maps) |
2728 | goto out_no_old_maps; | |
2729 | ||
80d19669 AN |
2730 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2731 | j < nr_ids;) { | |
2732 | for (i = num_tc, tci = j * num_tc; i--; tci++) { | |
2733 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
2734 | map = xmap_dereference(dev_maps->attr_map[tci]); | |
01c5f864 AD |
2735 | if (map && map != new_map) |
2736 | kfree_rcu(map, rcu); | |
2737 | } | |
537c00de AD |
2738 | } |
2739 | ||
184c449f AD |
2740 | kfree_rcu(dev_maps, rcu); |
2741 | ||
2742 | out_no_old_maps: | |
01c5f864 AD |
2743 | dev_maps = new_dev_maps; |
2744 | active = true; | |
537c00de | 2745 | |
01c5f864 | 2746 | out_no_new_maps: |
80d19669 AN |
2747 | if (!is_rxqs_map) { |
2748 | /* update Tx queue numa node */ | |
2749 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), | |
2750 | (numa_node_id >= 0) ? | |
2751 | numa_node_id : NUMA_NO_NODE); | |
2752 | } | |
537c00de | 2753 | |
01c5f864 AD |
2754 | if (!dev_maps) |
2755 | goto out_no_maps; | |
2756 | ||
80d19669 AN |
2757 | /* removes tx-queue from unused CPUs/rx-queues */ |
2758 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), | |
2759 | j < nr_ids;) { | |
2760 | for (i = tc, tci = j * num_tc; i--; tci++) | |
184c449f | 2761 | active |= remove_xps_queue(dev_maps, tci, index); |
80d19669 AN |
2762 | if (!netif_attr_test_mask(j, mask, nr_ids) || |
2763 | !netif_attr_test_online(j, online_mask, nr_ids)) | |
184c449f AD |
2764 | active |= remove_xps_queue(dev_maps, tci, index); |
2765 | for (i = num_tc - tc, tci++; --i; tci++) | |
2766 | active |= remove_xps_queue(dev_maps, tci, index); | |
01c5f864 AD |
2767 | } |
2768 | ||
2769 | /* free map if not active */ | |
867d0ad4 SD |
2770 | if (!active) |
2771 | reset_xps_maps(dev, dev_maps, is_rxqs_map); | |
01c5f864 AD |
2772 | |
2773 | out_no_maps: | |
537c00de AD |
2774 | mutex_unlock(&xps_map_mutex); |
2775 | ||
2776 | return 0; | |
2777 | error: | |
01c5f864 | 2778 | /* remove any maps that we added */ |
80d19669 AN |
2779 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2780 | j < nr_ids;) { | |
2781 | for (i = num_tc, tci = j * num_tc; i--; tci++) { | |
2782 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
184c449f | 2783 | map = dev_maps ? |
80d19669 | 2784 | xmap_dereference(dev_maps->attr_map[tci]) : |
184c449f AD |
2785 | NULL; |
2786 | if (new_map && new_map != map) | |
2787 | kfree(new_map); | |
2788 | } | |
01c5f864 AD |
2789 | } |
2790 | ||
537c00de AD |
2791 | mutex_unlock(&xps_map_mutex); |
2792 | ||
537c00de AD |
2793 | kfree(new_dev_maps); |
2794 | return -ENOMEM; | |
2795 | } | |
4d99f660 | 2796 | EXPORT_SYMBOL_GPL(__netif_set_xps_queue); |
80d19669 AN |
2797 | |
2798 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, | |
2799 | u16 index) | |
2800 | { | |
4d99f660 AV |
2801 | int ret; |
2802 | ||
2803 | cpus_read_lock(); | |
2804 | ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false); | |
2805 | cpus_read_unlock(); | |
2806 | ||
2807 | return ret; | |
80d19669 | 2808 | } |
537c00de AD |
2809 | EXPORT_SYMBOL(netif_set_xps_queue); |
2810 | ||
2811 | #endif | |
ffcfe25b AD |
2812 | static void netdev_unbind_all_sb_channels(struct net_device *dev) |
2813 | { | |
2814 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2815 | ||
2816 | /* Unbind any subordinate channels */ | |
2817 | while (txq-- != &dev->_tx[0]) { | |
2818 | if (txq->sb_dev) | |
2819 | netdev_unbind_sb_channel(dev, txq->sb_dev); | |
2820 | } | |
2821 | } | |
2822 | ||
9cf1f6a8 AD |
2823 | void netdev_reset_tc(struct net_device *dev) |
2824 | { | |
6234f874 AD |
2825 | #ifdef CONFIG_XPS |
2826 | netif_reset_xps_queues_gt(dev, 0); | |
2827 | #endif | |
ffcfe25b AD |
2828 | netdev_unbind_all_sb_channels(dev); |
2829 | ||
2830 | /* Reset TC configuration of device */ | |
9cf1f6a8 AD |
2831 | dev->num_tc = 0; |
2832 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2833 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2834 | } | |
2835 | EXPORT_SYMBOL(netdev_reset_tc); | |
2836 | ||
2837 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2838 | { | |
2839 | if (tc >= dev->num_tc) | |
2840 | return -EINVAL; | |
2841 | ||
6234f874 AD |
2842 | #ifdef CONFIG_XPS |
2843 | netif_reset_xps_queues(dev, offset, count); | |
2844 | #endif | |
9cf1f6a8 AD |
2845 | dev->tc_to_txq[tc].count = count; |
2846 | dev->tc_to_txq[tc].offset = offset; | |
2847 | return 0; | |
2848 | } | |
2849 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2850 | ||
2851 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2852 | { | |
2853 | if (num_tc > TC_MAX_QUEUE) | |
2854 | return -EINVAL; | |
2855 | ||
6234f874 AD |
2856 | #ifdef CONFIG_XPS |
2857 | netif_reset_xps_queues_gt(dev, 0); | |
2858 | #endif | |
ffcfe25b AD |
2859 | netdev_unbind_all_sb_channels(dev); |
2860 | ||
9cf1f6a8 AD |
2861 | dev->num_tc = num_tc; |
2862 | return 0; | |
2863 | } | |
2864 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2865 | ||
ffcfe25b AD |
2866 | void netdev_unbind_sb_channel(struct net_device *dev, |
2867 | struct net_device *sb_dev) | |
2868 | { | |
2869 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2870 | ||
2871 | #ifdef CONFIG_XPS | |
2872 | netif_reset_xps_queues_gt(sb_dev, 0); | |
2873 | #endif | |
2874 | memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); | |
2875 | memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); | |
2876 | ||
2877 | while (txq-- != &dev->_tx[0]) { | |
2878 | if (txq->sb_dev == sb_dev) | |
2879 | txq->sb_dev = NULL; | |
2880 | } | |
2881 | } | |
2882 | EXPORT_SYMBOL(netdev_unbind_sb_channel); | |
2883 | ||
2884 | int netdev_bind_sb_channel_queue(struct net_device *dev, | |
2885 | struct net_device *sb_dev, | |
2886 | u8 tc, u16 count, u16 offset) | |
2887 | { | |
2888 | /* Make certain the sb_dev and dev are already configured */ | |
2889 | if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) | |
2890 | return -EINVAL; | |
2891 | ||
2892 | /* We cannot hand out queues we don't have */ | |
2893 | if ((offset + count) > dev->real_num_tx_queues) | |
2894 | return -EINVAL; | |
2895 | ||
2896 | /* Record the mapping */ | |
2897 | sb_dev->tc_to_txq[tc].count = count; | |
2898 | sb_dev->tc_to_txq[tc].offset = offset; | |
2899 | ||
2900 | /* Provide a way for Tx queue to find the tc_to_txq map or | |
2901 | * XPS map for itself. | |
2902 | */ | |
2903 | while (count--) | |
2904 | netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; | |
2905 | ||
2906 | return 0; | |
2907 | } | |
2908 | EXPORT_SYMBOL(netdev_bind_sb_channel_queue); | |
2909 | ||
2910 | int netdev_set_sb_channel(struct net_device *dev, u16 channel) | |
2911 | { | |
2912 | /* Do not use a multiqueue device to represent a subordinate channel */ | |
2913 | if (netif_is_multiqueue(dev)) | |
2914 | return -ENODEV; | |
2915 | ||
2916 | /* We allow channels 1 - 32767 to be used for subordinate channels. | |
2917 | * Channel 0 is meant to be "native" mode and used only to represent | |
2918 | * the main root device. We allow writing 0 to reset the device back | |
2919 | * to normal mode after being used as a subordinate channel. | |
2920 | */ | |
2921 | if (channel > S16_MAX) | |
2922 | return -EINVAL; | |
2923 | ||
2924 | dev->num_tc = -channel; | |
2925 | ||
2926 | return 0; | |
2927 | } | |
2928 | EXPORT_SYMBOL(netdev_set_sb_channel); | |
2929 | ||
f0796d5c JF |
2930 | /* |
2931 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
3a053b1a | 2932 | * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. |
f0796d5c | 2933 | */ |
e6484930 | 2934 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2935 | { |
ac5b7019 | 2936 | bool disabling; |
1d24eb48 TH |
2937 | int rc; |
2938 | ||
ac5b7019 JK |
2939 | disabling = txq < dev->real_num_tx_queues; |
2940 | ||
e6484930 TH |
2941 | if (txq < 1 || txq > dev->num_tx_queues) |
2942 | return -EINVAL; | |
f0796d5c | 2943 | |
5c56580b BH |
2944 | if (dev->reg_state == NETREG_REGISTERED || |
2945 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2946 | ASSERT_RTNL(); |
2947 | ||
1d24eb48 TH |
2948 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2949 | txq); | |
bf264145 TH |
2950 | if (rc) |
2951 | return rc; | |
2952 | ||
4f57c087 JF |
2953 | if (dev->num_tc) |
2954 | netif_setup_tc(dev, txq); | |
2955 | ||
ac5b7019 JK |
2956 | dev->real_num_tx_queues = txq; |
2957 | ||
2958 | if (disabling) { | |
2959 | synchronize_net(); | |
e6484930 | 2960 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2961 | #ifdef CONFIG_XPS |
2962 | netif_reset_xps_queues_gt(dev, txq); | |
2963 | #endif | |
2964 | } | |
ac5b7019 JK |
2965 | } else { |
2966 | dev->real_num_tx_queues = txq; | |
f0796d5c | 2967 | } |
e6484930 | 2968 | |
e6484930 | 2969 | return 0; |
f0796d5c JF |
2970 | } |
2971 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2972 | |
a953be53 | 2973 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2974 | /** |
2975 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2976 | * @dev: Network device | |
2977 | * @rxq: Actual number of RX queues | |
2978 | * | |
2979 | * This must be called either with the rtnl_lock held or before | |
2980 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2981 | * negative error code. If called before registration, it always |
2982 | * succeeds. | |
62fe0b40 BH |
2983 | */ |
2984 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2985 | { | |
2986 | int rc; | |
2987 | ||
bd25fa7b TH |
2988 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2989 | return -EINVAL; | |
2990 | ||
62fe0b40 BH |
2991 | if (dev->reg_state == NETREG_REGISTERED) { |
2992 | ASSERT_RTNL(); | |
2993 | ||
62fe0b40 BH |
2994 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2995 | rxq); | |
2996 | if (rc) | |
2997 | return rc; | |
62fe0b40 BH |
2998 | } |
2999 | ||
3000 | dev->real_num_rx_queues = rxq; | |
3001 | return 0; | |
3002 | } | |
3003 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
3004 | #endif | |
3005 | ||
2c53040f BH |
3006 | /** |
3007 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
3008 | * |
3009 | * This routine should set an upper limit on the number of RSS queues | |
3010 | * used by default by multiqueue devices. | |
3011 | */ | |
a55b138b | 3012 | int netif_get_num_default_rss_queues(void) |
16917b87 | 3013 | { |
40e4e713 HS |
3014 | return is_kdump_kernel() ? |
3015 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
3016 | } |
3017 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
3018 | ||
3bcb846c | 3019 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 3020 | { |
def82a1d JP |
3021 | struct softnet_data *sd; |
3022 | unsigned long flags; | |
56079431 | 3023 | |
def82a1d | 3024 | local_irq_save(flags); |
903ceff7 | 3025 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
3026 | q->next_sched = NULL; |
3027 | *sd->output_queue_tailp = q; | |
3028 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
3029 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
3030 | local_irq_restore(flags); | |
3031 | } | |
3032 | ||
3033 | void __netif_schedule(struct Qdisc *q) | |
3034 | { | |
3035 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
3036 | __netif_reschedule(q); | |
56079431 DV |
3037 | } |
3038 | EXPORT_SYMBOL(__netif_schedule); | |
3039 | ||
e6247027 ED |
3040 | struct dev_kfree_skb_cb { |
3041 | enum skb_free_reason reason; | |
3042 | }; | |
3043 | ||
3044 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 3045 | { |
e6247027 ED |
3046 | return (struct dev_kfree_skb_cb *)skb->cb; |
3047 | } | |
3048 | ||
46e5da40 JF |
3049 | void netif_schedule_queue(struct netdev_queue *txq) |
3050 | { | |
3051 | rcu_read_lock(); | |
5be5515a | 3052 | if (!netif_xmit_stopped(txq)) { |
46e5da40 JF |
3053 | struct Qdisc *q = rcu_dereference(txq->qdisc); |
3054 | ||
3055 | __netif_schedule(q); | |
3056 | } | |
3057 | rcu_read_unlock(); | |
3058 | } | |
3059 | EXPORT_SYMBOL(netif_schedule_queue); | |
3060 | ||
46e5da40 JF |
3061 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
3062 | { | |
3063 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
3064 | struct Qdisc *q; | |
3065 | ||
3066 | rcu_read_lock(); | |
3067 | q = rcu_dereference(dev_queue->qdisc); | |
3068 | __netif_schedule(q); | |
3069 | rcu_read_unlock(); | |
3070 | } | |
3071 | } | |
3072 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
3073 | ||
e6247027 | 3074 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 3075 | { |
e6247027 | 3076 | unsigned long flags; |
56079431 | 3077 | |
9899886d MJ |
3078 | if (unlikely(!skb)) |
3079 | return; | |
3080 | ||
63354797 | 3081 | if (likely(refcount_read(&skb->users) == 1)) { |
e6247027 | 3082 | smp_rmb(); |
63354797 RE |
3083 | refcount_set(&skb->users, 0); |
3084 | } else if (likely(!refcount_dec_and_test(&skb->users))) { | |
e6247027 | 3085 | return; |
bea3348e | 3086 | } |
e6247027 ED |
3087 | get_kfree_skb_cb(skb)->reason = reason; |
3088 | local_irq_save(flags); | |
3089 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
3090 | __this_cpu_write(softnet_data.completion_queue, skb); | |
3091 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
3092 | local_irq_restore(flags); | |
56079431 | 3093 | } |
e6247027 | 3094 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 3095 | |
e6247027 | 3096 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
3097 | { |
3098 | if (in_irq() || irqs_disabled()) | |
e6247027 | 3099 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
3100 | else |
3101 | dev_kfree_skb(skb); | |
3102 | } | |
e6247027 | 3103 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
3104 | |
3105 | ||
bea3348e SH |
3106 | /** |
3107 | * netif_device_detach - mark device as removed | |
3108 | * @dev: network device | |
3109 | * | |
3110 | * Mark device as removed from system and therefore no longer available. | |
3111 | */ | |
56079431 DV |
3112 | void netif_device_detach(struct net_device *dev) |
3113 | { | |
3114 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
3115 | netif_running(dev)) { | |
d543103a | 3116 | netif_tx_stop_all_queues(dev); |
56079431 DV |
3117 | } |
3118 | } | |
3119 | EXPORT_SYMBOL(netif_device_detach); | |
3120 | ||
bea3348e SH |
3121 | /** |
3122 | * netif_device_attach - mark device as attached | |
3123 | * @dev: network device | |
3124 | * | |
3125 | * Mark device as attached from system and restart if needed. | |
3126 | */ | |
56079431 DV |
3127 | void netif_device_attach(struct net_device *dev) |
3128 | { | |
3129 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
3130 | netif_running(dev)) { | |
d543103a | 3131 | netif_tx_wake_all_queues(dev); |
4ec93edb | 3132 | __netdev_watchdog_up(dev); |
56079431 DV |
3133 | } |
3134 | } | |
3135 | EXPORT_SYMBOL(netif_device_attach); | |
3136 | ||
5605c762 JP |
3137 | /* |
3138 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
3139 | * to be used as a distribution range. | |
3140 | */ | |
eadec877 AD |
3141 | static u16 skb_tx_hash(const struct net_device *dev, |
3142 | const struct net_device *sb_dev, | |
3143 | struct sk_buff *skb) | |
5605c762 JP |
3144 | { |
3145 | u32 hash; | |
3146 | u16 qoffset = 0; | |
1b837d48 | 3147 | u16 qcount = dev->real_num_tx_queues; |
5605c762 | 3148 | |
eadec877 AD |
3149 | if (dev->num_tc) { |
3150 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
3151 | ||
3152 | qoffset = sb_dev->tc_to_txq[tc].offset; | |
3153 | qcount = sb_dev->tc_to_txq[tc].count; | |
3154 | } | |
3155 | ||
5605c762 JP |
3156 | if (skb_rx_queue_recorded(skb)) { |
3157 | hash = skb_get_rx_queue(skb); | |
6e11d157 AN |
3158 | if (hash >= qoffset) |
3159 | hash -= qoffset; | |
1b837d48 AD |
3160 | while (unlikely(hash >= qcount)) |
3161 | hash -= qcount; | |
eadec877 | 3162 | return hash + qoffset; |
5605c762 JP |
3163 | } |
3164 | ||
3165 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
3166 | } | |
5605c762 | 3167 | |
36c92474 BH |
3168 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
3169 | { | |
84d15ae5 | 3170 | static const netdev_features_t null_features; |
36c92474 | 3171 | struct net_device *dev = skb->dev; |
88ad4175 | 3172 | const char *name = ""; |
36c92474 | 3173 | |
c846ad9b BG |
3174 | if (!net_ratelimit()) |
3175 | return; | |
3176 | ||
88ad4175 BM |
3177 | if (dev) { |
3178 | if (dev->dev.parent) | |
3179 | name = dev_driver_string(dev->dev.parent); | |
3180 | else | |
3181 | name = netdev_name(dev); | |
3182 | } | |
6413139d WB |
3183 | skb_dump(KERN_WARNING, skb, false); |
3184 | WARN(1, "%s: caps=(%pNF, %pNF)\n", | |
88ad4175 | 3185 | name, dev ? &dev->features : &null_features, |
6413139d | 3186 | skb->sk ? &skb->sk->sk_route_caps : &null_features); |
36c92474 BH |
3187 | } |
3188 | ||
1da177e4 LT |
3189 | /* |
3190 | * Invalidate hardware checksum when packet is to be mangled, and | |
3191 | * complete checksum manually on outgoing path. | |
3192 | */ | |
84fa7933 | 3193 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 3194 | { |
d3bc23e7 | 3195 | __wsum csum; |
663ead3b | 3196 | int ret = 0, offset; |
1da177e4 | 3197 | |
84fa7933 | 3198 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
3199 | goto out_set_summed; |
3200 | ||
3201 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
3202 | skb_warn_bad_offload(skb); |
3203 | return -EINVAL; | |
1da177e4 LT |
3204 | } |
3205 | ||
cef401de ED |
3206 | /* Before computing a checksum, we should make sure no frag could |
3207 | * be modified by an external entity : checksum could be wrong. | |
3208 | */ | |
3209 | if (skb_has_shared_frag(skb)) { | |
3210 | ret = __skb_linearize(skb); | |
3211 | if (ret) | |
3212 | goto out; | |
3213 | } | |
3214 | ||
55508d60 | 3215 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
3216 | BUG_ON(offset >= skb_headlen(skb)); |
3217 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
3218 | ||
3219 | offset += skb->csum_offset; | |
3220 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
3221 | ||
8211fbfa HK |
3222 | ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); |
3223 | if (ret) | |
3224 | goto out; | |
1da177e4 | 3225 | |
4f2e4ad5 | 3226 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 3227 | out_set_summed: |
1da177e4 | 3228 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 3229 | out: |
1da177e4 LT |
3230 | return ret; |
3231 | } | |
d1b19dff | 3232 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 3233 | |
b72b5bf6 DC |
3234 | int skb_crc32c_csum_help(struct sk_buff *skb) |
3235 | { | |
3236 | __le32 crc32c_csum; | |
3237 | int ret = 0, offset, start; | |
3238 | ||
3239 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
3240 | goto out; | |
3241 | ||
3242 | if (unlikely(skb_is_gso(skb))) | |
3243 | goto out; | |
3244 | ||
3245 | /* Before computing a checksum, we should make sure no frag could | |
3246 | * be modified by an external entity : checksum could be wrong. | |
3247 | */ | |
3248 | if (unlikely(skb_has_shared_frag(skb))) { | |
3249 | ret = __skb_linearize(skb); | |
3250 | if (ret) | |
3251 | goto out; | |
3252 | } | |
3253 | start = skb_checksum_start_offset(skb); | |
3254 | offset = start + offsetof(struct sctphdr, checksum); | |
3255 | if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { | |
3256 | ret = -EINVAL; | |
3257 | goto out; | |
3258 | } | |
8211fbfa HK |
3259 | |
3260 | ret = skb_ensure_writable(skb, offset + sizeof(__le32)); | |
3261 | if (ret) | |
3262 | goto out; | |
3263 | ||
b72b5bf6 DC |
3264 | crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, |
3265 | skb->len - start, ~(__u32)0, | |
3266 | crc32c_csum_stub)); | |
3267 | *(__le32 *)(skb->data + offset) = crc32c_csum; | |
3268 | skb->ip_summed = CHECKSUM_NONE; | |
dba00306 | 3269 | skb->csum_not_inet = 0; |
b72b5bf6 DC |
3270 | out: |
3271 | return ret; | |
3272 | } | |
3273 | ||
53d6471c | 3274 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 3275 | { |
252e3346 | 3276 | __be16 type = skb->protocol; |
f6a78bfc | 3277 | |
19acc327 PS |
3278 | /* Tunnel gso handlers can set protocol to ethernet. */ |
3279 | if (type == htons(ETH_P_TEB)) { | |
3280 | struct ethhdr *eth; | |
3281 | ||
3282 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
3283 | return 0; | |
3284 | ||
1dfe82eb | 3285 | eth = (struct ethhdr *)skb->data; |
19acc327 PS |
3286 | type = eth->h_proto; |
3287 | } | |
3288 | ||
d4bcef3f | 3289 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
3290 | } |
3291 | ||
3292 | /** | |
3293 | * skb_mac_gso_segment - mac layer segmentation handler. | |
3294 | * @skb: buffer to segment | |
3295 | * @features: features for the output path (see dev->features) | |
3296 | */ | |
3297 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
3298 | netdev_features_t features) | |
3299 | { | |
3300 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
3301 | struct packet_offload *ptype; | |
53d6471c VY |
3302 | int vlan_depth = skb->mac_len; |
3303 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
3304 | |
3305 | if (unlikely(!type)) | |
3306 | return ERR_PTR(-EINVAL); | |
3307 | ||
53d6471c | 3308 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
3309 | |
3310 | rcu_read_lock(); | |
22061d80 | 3311 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 3312 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 3313 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
3314 | break; |
3315 | } | |
3316 | } | |
3317 | rcu_read_unlock(); | |
3318 | ||
98e399f8 | 3319 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 3320 | |
f6a78bfc HX |
3321 | return segs; |
3322 | } | |
05e8ef4a PS |
3323 | EXPORT_SYMBOL(skb_mac_gso_segment); |
3324 | ||
3325 | ||
3326 | /* openvswitch calls this on rx path, so we need a different check. | |
3327 | */ | |
3328 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
3329 | { | |
3330 | if (tx_path) | |
0c19f846 WB |
3331 | return skb->ip_summed != CHECKSUM_PARTIAL && |
3332 | skb->ip_summed != CHECKSUM_UNNECESSARY; | |
6e7bc478 ED |
3333 | |
3334 | return skb->ip_summed == CHECKSUM_NONE; | |
05e8ef4a PS |
3335 | } |
3336 | ||
3337 | /** | |
3338 | * __skb_gso_segment - Perform segmentation on skb. | |
3339 | * @skb: buffer to segment | |
3340 | * @features: features for the output path (see dev->features) | |
3341 | * @tx_path: whether it is called in TX path | |
3342 | * | |
3343 | * This function segments the given skb and returns a list of segments. | |
3344 | * | |
3345 | * It may return NULL if the skb requires no segmentation. This is | |
3346 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 | 3347 | * |
a08e7fd9 | 3348 | * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. |
05e8ef4a PS |
3349 | */ |
3350 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
3351 | netdev_features_t features, bool tx_path) | |
3352 | { | |
b2504a5d ED |
3353 | struct sk_buff *segs; |
3354 | ||
05e8ef4a PS |
3355 | if (unlikely(skb_needs_check(skb, tx_path))) { |
3356 | int err; | |
3357 | ||
b2504a5d | 3358 | /* We're going to init ->check field in TCP or UDP header */ |
a40e0a66 | 3359 | err = skb_cow_head(skb, 0); |
3360 | if (err < 0) | |
05e8ef4a PS |
3361 | return ERR_PTR(err); |
3362 | } | |
3363 | ||
802ab55a AD |
3364 | /* Only report GSO partial support if it will enable us to |
3365 | * support segmentation on this frame without needing additional | |
3366 | * work. | |
3367 | */ | |
3368 | if (features & NETIF_F_GSO_PARTIAL) { | |
3369 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
3370 | struct net_device *dev = skb->dev; | |
3371 | ||
3372 | partial_features |= dev->features & dev->gso_partial_features; | |
3373 | if (!skb_gso_ok(skb, features | partial_features)) | |
3374 | features &= ~NETIF_F_GSO_PARTIAL; | |
3375 | } | |
3376 | ||
a08e7fd9 | 3377 | BUILD_BUG_ON(SKB_GSO_CB_OFFSET + |
9207f9d4 KK |
3378 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); |
3379 | ||
68c33163 | 3380 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
3381 | SKB_GSO_CB(skb)->encap_level = 0; |
3382 | ||
05e8ef4a PS |
3383 | skb_reset_mac_header(skb); |
3384 | skb_reset_mac_len(skb); | |
3385 | ||
b2504a5d ED |
3386 | segs = skb_mac_gso_segment(skb, features); |
3387 | ||
3a1296a3 | 3388 | if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) |
b2504a5d ED |
3389 | skb_warn_bad_offload(skb); |
3390 | ||
3391 | return segs; | |
05e8ef4a | 3392 | } |
12b0004d | 3393 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 3394 | |
fb286bb2 HX |
3395 | /* Take action when hardware reception checksum errors are detected. */ |
3396 | #ifdef CONFIG_BUG | |
7fe50ac8 | 3397 | void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) |
fb286bb2 HX |
3398 | { |
3399 | if (net_ratelimit()) { | |
7b6cd1ce | 3400 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
6413139d | 3401 | skb_dump(KERN_ERR, skb, true); |
fb286bb2 HX |
3402 | dump_stack(); |
3403 | } | |
3404 | } | |
3405 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
3406 | #endif | |
3407 | ||
ab74cfeb | 3408 | /* XXX: check that highmem exists at all on the given machine. */ |
c1e756bf | 3409 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 3410 | { |
3d3a8533 | 3411 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 3412 | int i; |
f4563a75 | 3413 | |
5acbbd42 | 3414 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
3415 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
3416 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 3417 | |
ea2ab693 | 3418 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 3419 | return 1; |
ea2ab693 | 3420 | } |
5acbbd42 | 3421 | } |
3d3a8533 | 3422 | #endif |
1da177e4 LT |
3423 | return 0; |
3424 | } | |
1da177e4 | 3425 | |
3b392ddb SH |
3426 | /* If MPLS offload request, verify we are testing hardware MPLS features |
3427 | * instead of standard features for the netdev. | |
3428 | */ | |
d0edc7bf | 3429 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
3430 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
3431 | netdev_features_t features, | |
3432 | __be16 type) | |
3433 | { | |
25cd9ba0 | 3434 | if (eth_p_mpls(type)) |
3b392ddb SH |
3435 | features &= skb->dev->mpls_features; |
3436 | ||
3437 | return features; | |
3438 | } | |
3439 | #else | |
3440 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
3441 | netdev_features_t features, | |
3442 | __be16 type) | |
3443 | { | |
3444 | return features; | |
3445 | } | |
3446 | #endif | |
3447 | ||
c8f44aff | 3448 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 3449 | netdev_features_t features) |
f01a5236 | 3450 | { |
53d6471c | 3451 | int tmp; |
3b392ddb SH |
3452 | __be16 type; |
3453 | ||
3454 | type = skb_network_protocol(skb, &tmp); | |
3455 | features = net_mpls_features(skb, features, type); | |
53d6471c | 3456 | |
c0d680e5 | 3457 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 3458 | !can_checksum_protocol(features, type)) { |
996e8021 | 3459 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 3460 | } |
7be2c82c ED |
3461 | if (illegal_highdma(skb->dev, skb)) |
3462 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
3463 | |
3464 | return features; | |
3465 | } | |
3466 | ||
e38f3025 TM |
3467 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
3468 | struct net_device *dev, | |
3469 | netdev_features_t features) | |
3470 | { | |
3471 | return features; | |
3472 | } | |
3473 | EXPORT_SYMBOL(passthru_features_check); | |
3474 | ||
7ce23672 | 3475 | static netdev_features_t dflt_features_check(struct sk_buff *skb, |
8cb65d00 TM |
3476 | struct net_device *dev, |
3477 | netdev_features_t features) | |
3478 | { | |
3479 | return vlan_features_check(skb, features); | |
3480 | } | |
3481 | ||
cbc53e08 AD |
3482 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
3483 | struct net_device *dev, | |
3484 | netdev_features_t features) | |
3485 | { | |
3486 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
3487 | ||
3488 | if (gso_segs > dev->gso_max_segs) | |
3489 | return features & ~NETIF_F_GSO_MASK; | |
3490 | ||
802ab55a AD |
3491 | /* Support for GSO partial features requires software |
3492 | * intervention before we can actually process the packets | |
3493 | * so we need to strip support for any partial features now | |
3494 | * and we can pull them back in after we have partially | |
3495 | * segmented the frame. | |
3496 | */ | |
3497 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
3498 | features &= ~dev->gso_partial_features; | |
3499 | ||
3500 | /* Make sure to clear the IPv4 ID mangling feature if the | |
3501 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
3502 | */ |
3503 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
3504 | struct iphdr *iph = skb->encapsulation ? | |
3505 | inner_ip_hdr(skb) : ip_hdr(skb); | |
3506 | ||
3507 | if (!(iph->frag_off & htons(IP_DF))) | |
3508 | features &= ~NETIF_F_TSO_MANGLEID; | |
3509 | } | |
3510 | ||
3511 | return features; | |
3512 | } | |
3513 | ||
c1e756bf | 3514 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 3515 | { |
5f35227e | 3516 | struct net_device *dev = skb->dev; |
fcbeb976 | 3517 | netdev_features_t features = dev->features; |
58e998c6 | 3518 | |
cbc53e08 AD |
3519 | if (skb_is_gso(skb)) |
3520 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 3521 | |
5f35227e JG |
3522 | /* If encapsulation offload request, verify we are testing |
3523 | * hardware encapsulation features instead of standard | |
3524 | * features for the netdev | |
3525 | */ | |
3526 | if (skb->encapsulation) | |
3527 | features &= dev->hw_enc_features; | |
3528 | ||
f5a7fb88 TM |
3529 | if (skb_vlan_tagged(skb)) |
3530 | features = netdev_intersect_features(features, | |
3531 | dev->vlan_features | | |
3532 | NETIF_F_HW_VLAN_CTAG_TX | | |
3533 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 3534 | |
5f35227e JG |
3535 | if (dev->netdev_ops->ndo_features_check) |
3536 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
3537 | features); | |
8cb65d00 TM |
3538 | else |
3539 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 3540 | |
c1e756bf | 3541 | return harmonize_features(skb, features); |
58e998c6 | 3542 | } |
c1e756bf | 3543 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 3544 | |
2ea25513 | 3545 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 3546 | struct netdev_queue *txq, bool more) |
f6a78bfc | 3547 | { |
2ea25513 DM |
3548 | unsigned int len; |
3549 | int rc; | |
00829823 | 3550 | |
9f9a742d | 3551 | if (dev_nit_active(dev)) |
2ea25513 | 3552 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 3553 | |
2ea25513 DM |
3554 | len = skb->len; |
3555 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 3556 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 3557 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 3558 | |
2ea25513 DM |
3559 | return rc; |
3560 | } | |
7b9c6090 | 3561 | |
8dcda22a DM |
3562 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
3563 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
3564 | { |
3565 | struct sk_buff *skb = first; | |
3566 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 3567 | |
7f2e870f DM |
3568 | while (skb) { |
3569 | struct sk_buff *next = skb->next; | |
fc70fb64 | 3570 | |
a8305bff | 3571 | skb_mark_not_on_list(skb); |
95f6b3dd | 3572 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
3573 | if (unlikely(!dev_xmit_complete(rc))) { |
3574 | skb->next = next; | |
3575 | goto out; | |
3576 | } | |
6afff0ca | 3577 | |
7f2e870f | 3578 | skb = next; |
fe60faa5 | 3579 | if (netif_tx_queue_stopped(txq) && skb) { |
7f2e870f DM |
3580 | rc = NETDEV_TX_BUSY; |
3581 | break; | |
9ccb8975 | 3582 | } |
7f2e870f | 3583 | } |
9ccb8975 | 3584 | |
7f2e870f DM |
3585 | out: |
3586 | *ret = rc; | |
3587 | return skb; | |
3588 | } | |
b40863c6 | 3589 | |
1ff0dc94 ED |
3590 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
3591 | netdev_features_t features) | |
f6a78bfc | 3592 | { |
df8a39de | 3593 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
3594 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
3595 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
3596 | return skb; |
3597 | } | |
f6a78bfc | 3598 | |
43c26a1a DC |
3599 | int skb_csum_hwoffload_help(struct sk_buff *skb, |
3600 | const netdev_features_t features) | |
3601 | { | |
3602 | if (unlikely(skb->csum_not_inet)) | |
3603 | return !!(features & NETIF_F_SCTP_CRC) ? 0 : | |
3604 | skb_crc32c_csum_help(skb); | |
3605 | ||
3606 | return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb); | |
3607 | } | |
3608 | EXPORT_SYMBOL(skb_csum_hwoffload_help); | |
3609 | ||
f53c7239 | 3610 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) |
eae3f88e DM |
3611 | { |
3612 | netdev_features_t features; | |
f6a78bfc | 3613 | |
eae3f88e DM |
3614 | features = netif_skb_features(skb); |
3615 | skb = validate_xmit_vlan(skb, features); | |
3616 | if (unlikely(!skb)) | |
3617 | goto out_null; | |
7b9c6090 | 3618 | |
ebf4e808 IL |
3619 | skb = sk_validate_xmit_skb(skb, dev); |
3620 | if (unlikely(!skb)) | |
3621 | goto out_null; | |
3622 | ||
8b86a61d | 3623 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
3624 | struct sk_buff *segs; |
3625 | ||
3626 | segs = skb_gso_segment(skb, features); | |
cecda693 | 3627 | if (IS_ERR(segs)) { |
af6dabc9 | 3628 | goto out_kfree_skb; |
cecda693 JW |
3629 | } else if (segs) { |
3630 | consume_skb(skb); | |
3631 | skb = segs; | |
f6a78bfc | 3632 | } |
eae3f88e DM |
3633 | } else { |
3634 | if (skb_needs_linearize(skb, features) && | |
3635 | __skb_linearize(skb)) | |
3636 | goto out_kfree_skb; | |
4ec93edb | 3637 | |
eae3f88e DM |
3638 | /* If packet is not checksummed and device does not |
3639 | * support checksumming for this protocol, complete | |
3640 | * checksumming here. | |
3641 | */ | |
3642 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
3643 | if (skb->encapsulation) | |
3644 | skb_set_inner_transport_header(skb, | |
3645 | skb_checksum_start_offset(skb)); | |
3646 | else | |
3647 | skb_set_transport_header(skb, | |
3648 | skb_checksum_start_offset(skb)); | |
43c26a1a | 3649 | if (skb_csum_hwoffload_help(skb, features)) |
eae3f88e | 3650 | goto out_kfree_skb; |
7b9c6090 | 3651 | } |
0c772159 | 3652 | } |
7b9c6090 | 3653 | |
f53c7239 | 3654 | skb = validate_xmit_xfrm(skb, features, again); |
3dca3f38 | 3655 | |
eae3f88e | 3656 | return skb; |
fc70fb64 | 3657 | |
f6a78bfc HX |
3658 | out_kfree_skb: |
3659 | kfree_skb(skb); | |
eae3f88e | 3660 | out_null: |
d21fd63e | 3661 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3662 | return NULL; |
3663 | } | |
6afff0ca | 3664 | |
f53c7239 | 3665 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) |
55a93b3e ED |
3666 | { |
3667 | struct sk_buff *next, *head = NULL, *tail; | |
3668 | ||
bec3cfdc | 3669 | for (; skb != NULL; skb = next) { |
55a93b3e | 3670 | next = skb->next; |
a8305bff | 3671 | skb_mark_not_on_list(skb); |
bec3cfdc ED |
3672 | |
3673 | /* in case skb wont be segmented, point to itself */ | |
3674 | skb->prev = skb; | |
3675 | ||
f53c7239 | 3676 | skb = validate_xmit_skb(skb, dev, again); |
bec3cfdc ED |
3677 | if (!skb) |
3678 | continue; | |
55a93b3e | 3679 | |
bec3cfdc ED |
3680 | if (!head) |
3681 | head = skb; | |
3682 | else | |
3683 | tail->next = skb; | |
3684 | /* If skb was segmented, skb->prev points to | |
3685 | * the last segment. If not, it still contains skb. | |
3686 | */ | |
3687 | tail = skb->prev; | |
55a93b3e ED |
3688 | } |
3689 | return head; | |
f6a78bfc | 3690 | } |
104ba78c | 3691 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3692 | |
1def9238 ED |
3693 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3694 | { | |
3695 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3696 | ||
3697 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3698 | ||
3699 | /* To get more precise estimation of bytes sent on wire, | |
3700 | * we add to pkt_len the headers size of all segments | |
3701 | */ | |
a0dce875 | 3702 | if (shinfo->gso_size && skb_transport_header_was_set(skb)) { |
757b8b1d | 3703 | unsigned int hdr_len; |
15e5a030 | 3704 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3705 | |
757b8b1d ED |
3706 | /* mac layer + network layer */ |
3707 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3708 | ||
3709 | /* + transport layer */ | |
7c68d1a6 ED |
3710 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { |
3711 | const struct tcphdr *th; | |
3712 | struct tcphdr _tcphdr; | |
3713 | ||
3714 | th = skb_header_pointer(skb, skb_transport_offset(skb), | |
3715 | sizeof(_tcphdr), &_tcphdr); | |
3716 | if (likely(th)) | |
3717 | hdr_len += __tcp_hdrlen(th); | |
3718 | } else { | |
3719 | struct udphdr _udphdr; | |
3720 | ||
3721 | if (skb_header_pointer(skb, skb_transport_offset(skb), | |
3722 | sizeof(_udphdr), &_udphdr)) | |
3723 | hdr_len += sizeof(struct udphdr); | |
3724 | } | |
15e5a030 JW |
3725 | |
3726 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3727 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3728 | shinfo->gso_size); | |
3729 | ||
3730 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3731 | } |
3732 | } | |
3733 | ||
bbd8a0d3 KK |
3734 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3735 | struct net_device *dev, | |
3736 | struct netdev_queue *txq) | |
3737 | { | |
3738 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3739 | struct sk_buff *to_free = NULL; |
a2da570d | 3740 | bool contended; |
bbd8a0d3 KK |
3741 | int rc; |
3742 | ||
a2da570d | 3743 | qdisc_calculate_pkt_len(skb, q); |
6b3ba914 JF |
3744 | |
3745 | if (q->flags & TCQ_F_NOLOCK) { | |
379349e9 PA |
3746 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
3747 | qdisc_run(q); | |
6b3ba914 JF |
3748 | |
3749 | if (unlikely(to_free)) | |
3750 | kfree_skb_list(to_free); | |
3751 | return rc; | |
3752 | } | |
3753 | ||
79640a4c ED |
3754 | /* |
3755 | * Heuristic to force contended enqueues to serialize on a | |
3756 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3757 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3758 | * often and dequeue packets faster. |
79640a4c | 3759 | */ |
a2da570d | 3760 | contended = qdisc_is_running(q); |
79640a4c ED |
3761 | if (unlikely(contended)) |
3762 | spin_lock(&q->busylock); | |
3763 | ||
bbd8a0d3 KK |
3764 | spin_lock(root_lock); |
3765 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3766 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3767 | rc = NET_XMIT_DROP; |
3768 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3769 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3770 | /* |
3771 | * This is a work-conserving queue; there are no old skbs | |
3772 | * waiting to be sent out; and the qdisc is not running - | |
3773 | * xmit the skb directly. | |
3774 | */ | |
bfe0d029 | 3775 | |
bfe0d029 ED |
3776 | qdisc_bstats_update(q, skb); |
3777 | ||
55a93b3e | 3778 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3779 | if (unlikely(contended)) { |
3780 | spin_unlock(&q->busylock); | |
3781 | contended = false; | |
3782 | } | |
bbd8a0d3 | 3783 | __qdisc_run(q); |
6c148184 | 3784 | } |
bbd8a0d3 | 3785 | |
6c148184 | 3786 | qdisc_run_end(q); |
bbd8a0d3 KK |
3787 | rc = NET_XMIT_SUCCESS; |
3788 | } else { | |
520ac30f | 3789 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3790 | if (qdisc_run_begin(q)) { |
3791 | if (unlikely(contended)) { | |
3792 | spin_unlock(&q->busylock); | |
3793 | contended = false; | |
3794 | } | |
3795 | __qdisc_run(q); | |
6c148184 | 3796 | qdisc_run_end(q); |
79640a4c | 3797 | } |
bbd8a0d3 KK |
3798 | } |
3799 | spin_unlock(root_lock); | |
520ac30f ED |
3800 | if (unlikely(to_free)) |
3801 | kfree_skb_list(to_free); | |
79640a4c ED |
3802 | if (unlikely(contended)) |
3803 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3804 | return rc; |
3805 | } | |
3806 | ||
86f8515f | 3807 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3808 | static void skb_update_prio(struct sk_buff *skb) |
3809 | { | |
4dcb31d4 ED |
3810 | const struct netprio_map *map; |
3811 | const struct sock *sk; | |
3812 | unsigned int prioidx; | |
5bc1421e | 3813 | |
4dcb31d4 ED |
3814 | if (skb->priority) |
3815 | return; | |
3816 | map = rcu_dereference_bh(skb->dev->priomap); | |
3817 | if (!map) | |
3818 | return; | |
3819 | sk = skb_to_full_sk(skb); | |
3820 | if (!sk) | |
3821 | return; | |
91c68ce2 | 3822 | |
4dcb31d4 ED |
3823 | prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); |
3824 | ||
3825 | if (prioidx < map->priomap_len) | |
3826 | skb->priority = map->priomap[prioidx]; | |
5bc1421e NH |
3827 | } |
3828 | #else | |
3829 | #define skb_update_prio(skb) | |
3830 | #endif | |
3831 | ||
95603e22 MM |
3832 | /** |
3833 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3834 | * @net: network namespace this loopback is happening in |
3835 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3836 | * @skb: buffer to transmit |
3837 | */ | |
0c4b51f0 | 3838 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3839 | { |
3840 | skb_reset_mac_header(skb); | |
3841 | __skb_pull(skb, skb_network_offset(skb)); | |
3842 | skb->pkt_type = PACKET_LOOPBACK; | |
3843 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3844 | WARN_ON(!skb_dst(skb)); | |
3845 | skb_dst_force(skb); | |
3846 | netif_rx_ni(skb); | |
3847 | return 0; | |
3848 | } | |
3849 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3850 | ||
1f211a1b DB |
3851 | #ifdef CONFIG_NET_EGRESS |
3852 | static struct sk_buff * | |
3853 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3854 | { | |
46209401 | 3855 | struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); |
1f211a1b DB |
3856 | struct tcf_result cl_res; |
3857 | ||
46209401 | 3858 | if (!miniq) |
1f211a1b DB |
3859 | return skb; |
3860 | ||
8dc07fdb | 3861 | /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ |
46209401 | 3862 | mini_qdisc_bstats_cpu_update(miniq, skb); |
1f211a1b | 3863 | |
46209401 | 3864 | switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { |
1f211a1b DB |
3865 | case TC_ACT_OK: |
3866 | case TC_ACT_RECLASSIFY: | |
3867 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3868 | break; | |
3869 | case TC_ACT_SHOT: | |
46209401 | 3870 | mini_qdisc_qstats_cpu_drop(miniq); |
1f211a1b | 3871 | *ret = NET_XMIT_DROP; |
7e2c3aea DB |
3872 | kfree_skb(skb); |
3873 | return NULL; | |
1f211a1b DB |
3874 | case TC_ACT_STOLEN: |
3875 | case TC_ACT_QUEUED: | |
e25ea21f | 3876 | case TC_ACT_TRAP: |
1f211a1b | 3877 | *ret = NET_XMIT_SUCCESS; |
7e2c3aea | 3878 | consume_skb(skb); |
1f211a1b DB |
3879 | return NULL; |
3880 | case TC_ACT_REDIRECT: | |
3881 | /* No need to push/pop skb's mac_header here on egress! */ | |
3882 | skb_do_redirect(skb); | |
3883 | *ret = NET_XMIT_SUCCESS; | |
3884 | return NULL; | |
3885 | default: | |
3886 | break; | |
3887 | } | |
357b6cc5 | 3888 | |
1f211a1b DB |
3889 | return skb; |
3890 | } | |
3891 | #endif /* CONFIG_NET_EGRESS */ | |
3892 | ||
fc9bab24 AN |
3893 | #ifdef CONFIG_XPS |
3894 | static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, | |
3895 | struct xps_dev_maps *dev_maps, unsigned int tci) | |
3896 | { | |
3897 | struct xps_map *map; | |
3898 | int queue_index = -1; | |
3899 | ||
3900 | if (dev->num_tc) { | |
3901 | tci *= dev->num_tc; | |
3902 | tci += netdev_get_prio_tc_map(dev, skb->priority); | |
3903 | } | |
3904 | ||
3905 | map = rcu_dereference(dev_maps->attr_map[tci]); | |
3906 | if (map) { | |
3907 | if (map->len == 1) | |
3908 | queue_index = map->queues[0]; | |
3909 | else | |
3910 | queue_index = map->queues[reciprocal_scale( | |
3911 | skb_get_hash(skb), map->len)]; | |
3912 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3913 | queue_index = -1; | |
3914 | } | |
3915 | return queue_index; | |
3916 | } | |
3917 | #endif | |
3918 | ||
eadec877 AD |
3919 | static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, |
3920 | struct sk_buff *skb) | |
638b2a69 JP |
3921 | { |
3922 | #ifdef CONFIG_XPS | |
3923 | struct xps_dev_maps *dev_maps; | |
fc9bab24 | 3924 | struct sock *sk = skb->sk; |
638b2a69 JP |
3925 | int queue_index = -1; |
3926 | ||
04157469 AN |
3927 | if (!static_key_false(&xps_needed)) |
3928 | return -1; | |
3929 | ||
638b2a69 | 3930 | rcu_read_lock(); |
fc9bab24 AN |
3931 | if (!static_key_false(&xps_rxqs_needed)) |
3932 | goto get_cpus_map; | |
3933 | ||
eadec877 | 3934 | dev_maps = rcu_dereference(sb_dev->xps_rxqs_map); |
638b2a69 | 3935 | if (dev_maps) { |
fc9bab24 | 3936 | int tci = sk_rx_queue_get(sk); |
184c449f | 3937 | |
fc9bab24 AN |
3938 | if (tci >= 0 && tci < dev->num_rx_queues) |
3939 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
3940 | tci); | |
3941 | } | |
184c449f | 3942 | |
fc9bab24 AN |
3943 | get_cpus_map: |
3944 | if (queue_index < 0) { | |
eadec877 | 3945 | dev_maps = rcu_dereference(sb_dev->xps_cpus_map); |
fc9bab24 AN |
3946 | if (dev_maps) { |
3947 | unsigned int tci = skb->sender_cpu - 1; | |
3948 | ||
3949 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
3950 | tci); | |
638b2a69 JP |
3951 | } |
3952 | } | |
3953 | rcu_read_unlock(); | |
3954 | ||
3955 | return queue_index; | |
3956 | #else | |
3957 | return -1; | |
3958 | #endif | |
3959 | } | |
3960 | ||
a4ea8a3d | 3961 | u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, |
a350ecce | 3962 | struct net_device *sb_dev) |
a4ea8a3d AD |
3963 | { |
3964 | return 0; | |
3965 | } | |
3966 | EXPORT_SYMBOL(dev_pick_tx_zero); | |
3967 | ||
3968 | u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, | |
a350ecce | 3969 | struct net_device *sb_dev) |
a4ea8a3d AD |
3970 | { |
3971 | return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; | |
3972 | } | |
3973 | EXPORT_SYMBOL(dev_pick_tx_cpu_id); | |
3974 | ||
b71b5837 PA |
3975 | u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, |
3976 | struct net_device *sb_dev) | |
638b2a69 JP |
3977 | { |
3978 | struct sock *sk = skb->sk; | |
3979 | int queue_index = sk_tx_queue_get(sk); | |
3980 | ||
eadec877 AD |
3981 | sb_dev = sb_dev ? : dev; |
3982 | ||
638b2a69 JP |
3983 | if (queue_index < 0 || skb->ooo_okay || |
3984 | queue_index >= dev->real_num_tx_queues) { | |
eadec877 | 3985 | int new_index = get_xps_queue(dev, sb_dev, skb); |
f4563a75 | 3986 | |
638b2a69 | 3987 | if (new_index < 0) |
eadec877 | 3988 | new_index = skb_tx_hash(dev, sb_dev, skb); |
638b2a69 JP |
3989 | |
3990 | if (queue_index != new_index && sk && | |
004a5d01 | 3991 | sk_fullsock(sk) && |
638b2a69 JP |
3992 | rcu_access_pointer(sk->sk_dst_cache)) |
3993 | sk_tx_queue_set(sk, new_index); | |
3994 | ||
3995 | queue_index = new_index; | |
3996 | } | |
3997 | ||
3998 | return queue_index; | |
3999 | } | |
b71b5837 | 4000 | EXPORT_SYMBOL(netdev_pick_tx); |
638b2a69 | 4001 | |
4bd97d51 PA |
4002 | struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, |
4003 | struct sk_buff *skb, | |
4004 | struct net_device *sb_dev) | |
638b2a69 JP |
4005 | { |
4006 | int queue_index = 0; | |
4007 | ||
4008 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
4009 | u32 sender_cpu = skb->sender_cpu - 1; |
4010 | ||
4011 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
4012 | skb->sender_cpu = raw_smp_processor_id() + 1; |
4013 | #endif | |
4014 | ||
4015 | if (dev->real_num_tx_queues != 1) { | |
4016 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 4017 | |
638b2a69 | 4018 | if (ops->ndo_select_queue) |
a350ecce | 4019 | queue_index = ops->ndo_select_queue(dev, skb, sb_dev); |
638b2a69 | 4020 | else |
4bd97d51 | 4021 | queue_index = netdev_pick_tx(dev, skb, sb_dev); |
638b2a69 | 4022 | |
d584527c | 4023 | queue_index = netdev_cap_txqueue(dev, queue_index); |
638b2a69 JP |
4024 | } |
4025 | ||
4026 | skb_set_queue_mapping(skb, queue_index); | |
4027 | return netdev_get_tx_queue(dev, queue_index); | |
4028 | } | |
4029 | ||
d29f749e | 4030 | /** |
9d08dd3d | 4031 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 4032 | * @skb: buffer to transmit |
eadec877 | 4033 | * @sb_dev: suboordinate device used for L2 forwarding offload |
d29f749e DJ |
4034 | * |
4035 | * Queue a buffer for transmission to a network device. The caller must | |
4036 | * have set the device and priority and built the buffer before calling | |
4037 | * this function. The function can be called from an interrupt. | |
4038 | * | |
4039 | * A negative errno code is returned on a failure. A success does not | |
4040 | * guarantee the frame will be transmitted as it may be dropped due | |
4041 | * to congestion or traffic shaping. | |
4042 | * | |
4043 | * ----------------------------------------------------------------------------------- | |
4044 | * I notice this method can also return errors from the queue disciplines, | |
4045 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
4046 | * be positive. | |
4047 | * | |
4048 | * Regardless of the return value, the skb is consumed, so it is currently | |
4049 | * difficult to retry a send to this method. (You can bump the ref count | |
4050 | * before sending to hold a reference for retry if you are careful.) | |
4051 | * | |
4052 | * When calling this method, interrupts MUST be enabled. This is because | |
4053 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
4054 | * --BLG | |
4055 | */ | |
eadec877 | 4056 | static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) |
1da177e4 LT |
4057 | { |
4058 | struct net_device *dev = skb->dev; | |
dc2b4847 | 4059 | struct netdev_queue *txq; |
1da177e4 LT |
4060 | struct Qdisc *q; |
4061 | int rc = -ENOMEM; | |
f53c7239 | 4062 | bool again = false; |
1da177e4 | 4063 | |
6d1ccff6 ED |
4064 | skb_reset_mac_header(skb); |
4065 | ||
e7fd2885 WB |
4066 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
4067 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
4068 | ||
4ec93edb YH |
4069 | /* Disable soft irqs for various locks below. Also |
4070 | * stops preemption for RCU. | |
1da177e4 | 4071 | */ |
4ec93edb | 4072 | rcu_read_lock_bh(); |
1da177e4 | 4073 | |
5bc1421e NH |
4074 | skb_update_prio(skb); |
4075 | ||
1f211a1b DB |
4076 | qdisc_pkt_len_init(skb); |
4077 | #ifdef CONFIG_NET_CLS_ACT | |
8dc07fdb | 4078 | skb->tc_at_ingress = 0; |
357b6cc5 | 4079 | # ifdef CONFIG_NET_EGRESS |
aabf6772 | 4080 | if (static_branch_unlikely(&egress_needed_key)) { |
1f211a1b DB |
4081 | skb = sch_handle_egress(skb, &rc, dev); |
4082 | if (!skb) | |
4083 | goto out; | |
4084 | } | |
357b6cc5 | 4085 | # endif |
1f211a1b | 4086 | #endif |
02875878 ED |
4087 | /* If device/qdisc don't need skb->dst, release it right now while |
4088 | * its hot in this cpu cache. | |
4089 | */ | |
4090 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
4091 | skb_dst_drop(skb); | |
4092 | else | |
4093 | skb_dst_force(skb); | |
4094 | ||
4bd97d51 | 4095 | txq = netdev_core_pick_tx(dev, skb, sb_dev); |
a898def2 | 4096 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 4097 | |
cf66ba58 | 4098 | trace_net_dev_queue(skb); |
1da177e4 | 4099 | if (q->enqueue) { |
bbd8a0d3 | 4100 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 4101 | goto out; |
1da177e4 LT |
4102 | } |
4103 | ||
4104 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 4105 | * loopback, all the sorts of tunnels... |
1da177e4 | 4106 | |
eb13da1a | 4107 | * Really, it is unlikely that netif_tx_lock protection is necessary |
4108 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
4109 | * counters.) | |
4110 | * However, it is possible, that they rely on protection | |
4111 | * made by us here. | |
1da177e4 | 4112 | |
eb13da1a | 4113 | * Check this and shot the lock. It is not prone from deadlocks. |
4114 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
4115 | */ |
4116 | if (dev->flags & IFF_UP) { | |
4117 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
4118 | ||
c773e847 | 4119 | if (txq->xmit_lock_owner != cpu) { |
97cdcf37 | 4120 | if (dev_xmit_recursion()) |
745e20f1 ED |
4121 | goto recursion_alert; |
4122 | ||
f53c7239 | 4123 | skb = validate_xmit_skb(skb, dev, &again); |
1f59533f | 4124 | if (!skb) |
d21fd63e | 4125 | goto out; |
1f59533f | 4126 | |
c773e847 | 4127 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 4128 | |
73466498 | 4129 | if (!netif_xmit_stopped(txq)) { |
97cdcf37 | 4130 | dev_xmit_recursion_inc(); |
ce93718f | 4131 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
97cdcf37 | 4132 | dev_xmit_recursion_dec(); |
572a9d7b | 4133 | if (dev_xmit_complete(rc)) { |
c773e847 | 4134 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
4135 | goto out; |
4136 | } | |
4137 | } | |
c773e847 | 4138 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
4139 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
4140 | dev->name); | |
1da177e4 LT |
4141 | } else { |
4142 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
4143 | * unfortunately |
4144 | */ | |
4145 | recursion_alert: | |
e87cc472 JP |
4146 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
4147 | dev->name); | |
1da177e4 LT |
4148 | } |
4149 | } | |
4150 | ||
4151 | rc = -ENETDOWN; | |
d4828d85 | 4152 | rcu_read_unlock_bh(); |
1da177e4 | 4153 | |
015f0688 | 4154 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 4155 | kfree_skb_list(skb); |
1da177e4 LT |
4156 | return rc; |
4157 | out: | |
d4828d85 | 4158 | rcu_read_unlock_bh(); |
1da177e4 LT |
4159 | return rc; |
4160 | } | |
f663dd9a | 4161 | |
2b4aa3ce | 4162 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
4163 | { |
4164 | return __dev_queue_xmit(skb, NULL); | |
4165 | } | |
2b4aa3ce | 4166 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 4167 | |
eadec877 | 4168 | int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) |
f663dd9a | 4169 | { |
eadec877 | 4170 | return __dev_queue_xmit(skb, sb_dev); |
f663dd9a JW |
4171 | } |
4172 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
4173 | ||
865b03f2 MK |
4174 | int dev_direct_xmit(struct sk_buff *skb, u16 queue_id) |
4175 | { | |
4176 | struct net_device *dev = skb->dev; | |
4177 | struct sk_buff *orig_skb = skb; | |
4178 | struct netdev_queue *txq; | |
4179 | int ret = NETDEV_TX_BUSY; | |
4180 | bool again = false; | |
4181 | ||
4182 | if (unlikely(!netif_running(dev) || | |
4183 | !netif_carrier_ok(dev))) | |
4184 | goto drop; | |
4185 | ||
4186 | skb = validate_xmit_skb_list(skb, dev, &again); | |
4187 | if (skb != orig_skb) | |
4188 | goto drop; | |
4189 | ||
4190 | skb_set_queue_mapping(skb, queue_id); | |
4191 | txq = skb_get_tx_queue(dev, skb); | |
4192 | ||
4193 | local_bh_disable(); | |
4194 | ||
4195 | HARD_TX_LOCK(dev, txq, smp_processor_id()); | |
4196 | if (!netif_xmit_frozen_or_drv_stopped(txq)) | |
4197 | ret = netdev_start_xmit(skb, dev, txq, false); | |
4198 | HARD_TX_UNLOCK(dev, txq); | |
4199 | ||
4200 | local_bh_enable(); | |
4201 | ||
4202 | if (!dev_xmit_complete(ret)) | |
4203 | kfree_skb(skb); | |
4204 | ||
4205 | return ret; | |
4206 | drop: | |
4207 | atomic_long_inc(&dev->tx_dropped); | |
4208 | kfree_skb_list(skb); | |
4209 | return NET_XMIT_DROP; | |
4210 | } | |
4211 | EXPORT_SYMBOL(dev_direct_xmit); | |
1da177e4 | 4212 | |
eb13da1a | 4213 | /************************************************************************* |
4214 | * Receiver routines | |
4215 | *************************************************************************/ | |
1da177e4 | 4216 | |
6b2bedc3 | 4217 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
4218 | EXPORT_SYMBOL(netdev_max_backlog); |
4219 | ||
3b098e2d | 4220 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 | 4221 | int netdev_budget __read_mostly = 300; |
a4837980 KK |
4222 | /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ |
4223 | unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; | |
3d48b53f MT |
4224 | int weight_p __read_mostly = 64; /* old backlog weight */ |
4225 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
4226 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
4227 | int dev_rx_weight __read_mostly = 64; | |
4228 | int dev_tx_weight __read_mostly = 64; | |
323ebb61 EC |
4229 | /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ |
4230 | int gro_normal_batch __read_mostly = 8; | |
1da177e4 | 4231 | |
eecfd7c4 ED |
4232 | /* Called with irq disabled */ |
4233 | static inline void ____napi_schedule(struct softnet_data *sd, | |
4234 | struct napi_struct *napi) | |
4235 | { | |
4236 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
4237 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4238 | } | |
4239 | ||
bfb564e7 KK |
4240 | #ifdef CONFIG_RPS |
4241 | ||
4242 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 4243 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 4244 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
4245 | u32 rps_cpu_mask __read_mostly; |
4246 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 4247 | |
dc05360f | 4248 | struct static_key_false rps_needed __read_mostly; |
3df97ba8 | 4249 | EXPORT_SYMBOL(rps_needed); |
dc05360f | 4250 | struct static_key_false rfs_needed __read_mostly; |
13bfff25 | 4251 | EXPORT_SYMBOL(rfs_needed); |
adc9300e | 4252 | |
c445477d BH |
4253 | static struct rps_dev_flow * |
4254 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4255 | struct rps_dev_flow *rflow, u16 next_cpu) | |
4256 | { | |
a31196b0 | 4257 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
4258 | #ifdef CONFIG_RFS_ACCEL |
4259 | struct netdev_rx_queue *rxqueue; | |
4260 | struct rps_dev_flow_table *flow_table; | |
4261 | struct rps_dev_flow *old_rflow; | |
4262 | u32 flow_id; | |
4263 | u16 rxq_index; | |
4264 | int rc; | |
4265 | ||
4266 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
4267 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
4268 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
4269 | goto out; |
4270 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
4271 | if (rxq_index == skb_get_rx_queue(skb)) | |
4272 | goto out; | |
4273 | ||
4274 | rxqueue = dev->_rx + rxq_index; | |
4275 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4276 | if (!flow_table) | |
4277 | goto out; | |
61b905da | 4278 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
4279 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
4280 | rxq_index, flow_id); | |
4281 | if (rc < 0) | |
4282 | goto out; | |
4283 | old_rflow = rflow; | |
4284 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
4285 | rflow->filter = rc; |
4286 | if (old_rflow->filter == rflow->filter) | |
4287 | old_rflow->filter = RPS_NO_FILTER; | |
4288 | out: | |
4289 | #endif | |
4290 | rflow->last_qtail = | |
09994d1b | 4291 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
4292 | } |
4293 | ||
09994d1b | 4294 | rflow->cpu = next_cpu; |
c445477d BH |
4295 | return rflow; |
4296 | } | |
4297 | ||
bfb564e7 KK |
4298 | /* |
4299 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
4300 | * CPU from the RPS map of the receiving queue for a given skb. | |
4301 | * rcu_read_lock must be held on entry. | |
4302 | */ | |
4303 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4304 | struct rps_dev_flow **rflowp) | |
4305 | { | |
567e4b79 ED |
4306 | const struct rps_sock_flow_table *sock_flow_table; |
4307 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 4308 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 4309 | struct rps_map *map; |
bfb564e7 | 4310 | int cpu = -1; |
567e4b79 | 4311 | u32 tcpu; |
61b905da | 4312 | u32 hash; |
bfb564e7 KK |
4313 | |
4314 | if (skb_rx_queue_recorded(skb)) { | |
4315 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 4316 | |
62fe0b40 BH |
4317 | if (unlikely(index >= dev->real_num_rx_queues)) { |
4318 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4319 | "%s received packet on queue %u, but number " | |
4320 | "of RX queues is %u\n", | |
4321 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
4322 | goto done; |
4323 | } | |
567e4b79 ED |
4324 | rxqueue += index; |
4325 | } | |
bfb564e7 | 4326 | |
567e4b79 ED |
4327 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
4328 | ||
4329 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 4330 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 4331 | if (!flow_table && !map) |
bfb564e7 KK |
4332 | goto done; |
4333 | ||
2d47b459 | 4334 | skb_reset_network_header(skb); |
61b905da TH |
4335 | hash = skb_get_hash(skb); |
4336 | if (!hash) | |
bfb564e7 KK |
4337 | goto done; |
4338 | ||
fec5e652 TH |
4339 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
4340 | if (flow_table && sock_flow_table) { | |
fec5e652 | 4341 | struct rps_dev_flow *rflow; |
567e4b79 ED |
4342 | u32 next_cpu; |
4343 | u32 ident; | |
4344 | ||
4345 | /* First check into global flow table if there is a match */ | |
4346 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
4347 | if ((ident ^ hash) & ~rps_cpu_mask) | |
4348 | goto try_rps; | |
fec5e652 | 4349 | |
567e4b79 ED |
4350 | next_cpu = ident & rps_cpu_mask; |
4351 | ||
4352 | /* OK, now we know there is a match, | |
4353 | * we can look at the local (per receive queue) flow table | |
4354 | */ | |
61b905da | 4355 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
4356 | tcpu = rflow->cpu; |
4357 | ||
fec5e652 TH |
4358 | /* |
4359 | * If the desired CPU (where last recvmsg was done) is | |
4360 | * different from current CPU (one in the rx-queue flow | |
4361 | * table entry), switch if one of the following holds: | |
a31196b0 | 4362 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
4363 | * - Current CPU is offline. |
4364 | * - The current CPU's queue tail has advanced beyond the | |
4365 | * last packet that was enqueued using this table entry. | |
4366 | * This guarantees that all previous packets for the flow | |
4367 | * have been dequeued, thus preserving in order delivery. | |
4368 | */ | |
4369 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 4370 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 4371 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
4372 | rflow->last_qtail)) >= 0)) { |
4373 | tcpu = next_cpu; | |
c445477d | 4374 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 4375 | } |
c445477d | 4376 | |
a31196b0 | 4377 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
4378 | *rflowp = rflow; |
4379 | cpu = tcpu; | |
4380 | goto done; | |
4381 | } | |
4382 | } | |
4383 | ||
567e4b79 ED |
4384 | try_rps: |
4385 | ||
0a9627f2 | 4386 | if (map) { |
8fc54f68 | 4387 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
4388 | if (cpu_online(tcpu)) { |
4389 | cpu = tcpu; | |
4390 | goto done; | |
4391 | } | |
4392 | } | |
4393 | ||
4394 | done: | |
0a9627f2 TH |
4395 | return cpu; |
4396 | } | |
4397 | ||
c445477d BH |
4398 | #ifdef CONFIG_RFS_ACCEL |
4399 | ||
4400 | /** | |
4401 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
4402 | * @dev: Device on which the filter was set | |
4403 | * @rxq_index: RX queue index | |
4404 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
4405 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
4406 | * | |
4407 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
4408 | * this function for each installed filter and remove the filters for | |
4409 | * which it returns %true. | |
4410 | */ | |
4411 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
4412 | u32 flow_id, u16 filter_id) | |
4413 | { | |
4414 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
4415 | struct rps_dev_flow_table *flow_table; | |
4416 | struct rps_dev_flow *rflow; | |
4417 | bool expire = true; | |
a31196b0 | 4418 | unsigned int cpu; |
c445477d BH |
4419 | |
4420 | rcu_read_lock(); | |
4421 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4422 | if (flow_table && flow_id <= flow_table->mask) { | |
4423 | rflow = &flow_table->flows[flow_id]; | |
6aa7de05 | 4424 | cpu = READ_ONCE(rflow->cpu); |
a31196b0 | 4425 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
4426 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
4427 | rflow->last_qtail) < | |
4428 | (int)(10 * flow_table->mask))) | |
4429 | expire = false; | |
4430 | } | |
4431 | rcu_read_unlock(); | |
4432 | return expire; | |
4433 | } | |
4434 | EXPORT_SYMBOL(rps_may_expire_flow); | |
4435 | ||
4436 | #endif /* CONFIG_RFS_ACCEL */ | |
4437 | ||
0a9627f2 | 4438 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 4439 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 4440 | { |
e36fa2f7 ED |
4441 | struct softnet_data *sd = data; |
4442 | ||
eecfd7c4 | 4443 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 4444 | sd->received_rps++; |
0a9627f2 | 4445 | } |
e36fa2f7 | 4446 | |
fec5e652 | 4447 | #endif /* CONFIG_RPS */ |
0a9627f2 | 4448 | |
e36fa2f7 ED |
4449 | /* |
4450 | * Check if this softnet_data structure is another cpu one | |
4451 | * If yes, queue it to our IPI list and return 1 | |
4452 | * If no, return 0 | |
4453 | */ | |
4454 | static int rps_ipi_queued(struct softnet_data *sd) | |
4455 | { | |
4456 | #ifdef CONFIG_RPS | |
903ceff7 | 4457 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
4458 | |
4459 | if (sd != mysd) { | |
4460 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
4461 | mysd->rps_ipi_list = sd; | |
4462 | ||
4463 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4464 | return 1; | |
4465 | } | |
4466 | #endif /* CONFIG_RPS */ | |
4467 | return 0; | |
4468 | } | |
4469 | ||
99bbc707 WB |
4470 | #ifdef CONFIG_NET_FLOW_LIMIT |
4471 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
4472 | #endif | |
4473 | ||
4474 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
4475 | { | |
4476 | #ifdef CONFIG_NET_FLOW_LIMIT | |
4477 | struct sd_flow_limit *fl; | |
4478 | struct softnet_data *sd; | |
4479 | unsigned int old_flow, new_flow; | |
4480 | ||
4481 | if (qlen < (netdev_max_backlog >> 1)) | |
4482 | return false; | |
4483 | ||
903ceff7 | 4484 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
4485 | |
4486 | rcu_read_lock(); | |
4487 | fl = rcu_dereference(sd->flow_limit); | |
4488 | if (fl) { | |
3958afa1 | 4489 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
4490 | old_flow = fl->history[fl->history_head]; |
4491 | fl->history[fl->history_head] = new_flow; | |
4492 | ||
4493 | fl->history_head++; | |
4494 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
4495 | ||
4496 | if (likely(fl->buckets[old_flow])) | |
4497 | fl->buckets[old_flow]--; | |
4498 | ||
4499 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
4500 | fl->count++; | |
4501 | rcu_read_unlock(); | |
4502 | return true; | |
4503 | } | |
4504 | } | |
4505 | rcu_read_unlock(); | |
4506 | #endif | |
4507 | return false; | |
4508 | } | |
4509 | ||
0a9627f2 TH |
4510 | /* |
4511 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
4512 | * queue (may be a remote CPU queue). | |
4513 | */ | |
fec5e652 TH |
4514 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
4515 | unsigned int *qtail) | |
0a9627f2 | 4516 | { |
e36fa2f7 | 4517 | struct softnet_data *sd; |
0a9627f2 | 4518 | unsigned long flags; |
99bbc707 | 4519 | unsigned int qlen; |
0a9627f2 | 4520 | |
e36fa2f7 | 4521 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
4522 | |
4523 | local_irq_save(flags); | |
0a9627f2 | 4524 | |
e36fa2f7 | 4525 | rps_lock(sd); |
e9e4dd32 JA |
4526 | if (!netif_running(skb->dev)) |
4527 | goto drop; | |
99bbc707 WB |
4528 | qlen = skb_queue_len(&sd->input_pkt_queue); |
4529 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 4530 | if (qlen) { |
0a9627f2 | 4531 | enqueue: |
e36fa2f7 | 4532 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 4533 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 4534 | rps_unlock(sd); |
152102c7 | 4535 | local_irq_restore(flags); |
0a9627f2 TH |
4536 | return NET_RX_SUCCESS; |
4537 | } | |
4538 | ||
ebda37c2 ED |
4539 | /* Schedule NAPI for backlog device |
4540 | * We can use non atomic operation since we own the queue lock | |
4541 | */ | |
4542 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 4543 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 4544 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
4545 | } |
4546 | goto enqueue; | |
4547 | } | |
4548 | ||
e9e4dd32 | 4549 | drop: |
dee42870 | 4550 | sd->dropped++; |
e36fa2f7 | 4551 | rps_unlock(sd); |
0a9627f2 | 4552 | |
0a9627f2 TH |
4553 | local_irq_restore(flags); |
4554 | ||
caf586e5 | 4555 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
4556 | kfree_skb(skb); |
4557 | return NET_RX_DROP; | |
4558 | } | |
1da177e4 | 4559 | |
e817f856 JDB |
4560 | static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) |
4561 | { | |
4562 | struct net_device *dev = skb->dev; | |
4563 | struct netdev_rx_queue *rxqueue; | |
4564 | ||
4565 | rxqueue = dev->_rx; | |
4566 | ||
4567 | if (skb_rx_queue_recorded(skb)) { | |
4568 | u16 index = skb_get_rx_queue(skb); | |
4569 | ||
4570 | if (unlikely(index >= dev->real_num_rx_queues)) { | |
4571 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4572 | "%s received packet on queue %u, but number " | |
4573 | "of RX queues is %u\n", | |
4574 | dev->name, index, dev->real_num_rx_queues); | |
4575 | ||
4576 | return rxqueue; /* Return first rxqueue */ | |
4577 | } | |
4578 | rxqueue += index; | |
4579 | } | |
4580 | return rxqueue; | |
4581 | } | |
4582 | ||
d4455169 | 4583 | static u32 netif_receive_generic_xdp(struct sk_buff *skb, |
02671e23 | 4584 | struct xdp_buff *xdp, |
d4455169 JF |
4585 | struct bpf_prog *xdp_prog) |
4586 | { | |
e817f856 | 4587 | struct netdev_rx_queue *rxqueue; |
198d83bb | 4588 | void *orig_data, *orig_data_end; |
de8f3a83 | 4589 | u32 metalen, act = XDP_DROP; |
29724956 JDB |
4590 | __be16 orig_eth_type; |
4591 | struct ethhdr *eth; | |
4592 | bool orig_bcast; | |
d4455169 JF |
4593 | int hlen, off; |
4594 | u32 mac_len; | |
4595 | ||
4596 | /* Reinjected packets coming from act_mirred or similar should | |
4597 | * not get XDP generic processing. | |
4598 | */ | |
2c64605b | 4599 | if (skb_is_redirected(skb)) |
d4455169 JF |
4600 | return XDP_PASS; |
4601 | ||
de8f3a83 DB |
4602 | /* XDP packets must be linear and must have sufficient headroom |
4603 | * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also | |
4604 | * native XDP provides, thus we need to do it here as well. | |
4605 | */ | |
ad1e03b2 | 4606 | if (skb_cloned(skb) || skb_is_nonlinear(skb) || |
de8f3a83 DB |
4607 | skb_headroom(skb) < XDP_PACKET_HEADROOM) { |
4608 | int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); | |
4609 | int troom = skb->tail + skb->data_len - skb->end; | |
4610 | ||
4611 | /* In case we have to go down the path and also linearize, | |
4612 | * then lets do the pskb_expand_head() work just once here. | |
4613 | */ | |
4614 | if (pskb_expand_head(skb, | |
4615 | hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, | |
4616 | troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) | |
4617 | goto do_drop; | |
2d17d8d7 | 4618 | if (skb_linearize(skb)) |
de8f3a83 DB |
4619 | goto do_drop; |
4620 | } | |
d4455169 JF |
4621 | |
4622 | /* The XDP program wants to see the packet starting at the MAC | |
4623 | * header. | |
4624 | */ | |
4625 | mac_len = skb->data - skb_mac_header(skb); | |
4626 | hlen = skb_headlen(skb) + mac_len; | |
02671e23 BT |
4627 | xdp->data = skb->data - mac_len; |
4628 | xdp->data_meta = xdp->data; | |
4629 | xdp->data_end = xdp->data + hlen; | |
4630 | xdp->data_hard_start = skb->data - skb_headroom(skb); | |
a075767b JDB |
4631 | |
4632 | /* SKB "head" area always have tailroom for skb_shared_info */ | |
4633 | xdp->frame_sz = (void *)skb_end_pointer(skb) - xdp->data_hard_start; | |
4634 | xdp->frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
4635 | ||
02671e23 BT |
4636 | orig_data_end = xdp->data_end; |
4637 | orig_data = xdp->data; | |
29724956 JDB |
4638 | eth = (struct ethhdr *)xdp->data; |
4639 | orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); | |
4640 | orig_eth_type = eth->h_proto; | |
d4455169 | 4641 | |
e817f856 | 4642 | rxqueue = netif_get_rxqueue(skb); |
02671e23 | 4643 | xdp->rxq = &rxqueue->xdp_rxq; |
e817f856 | 4644 | |
02671e23 | 4645 | act = bpf_prog_run_xdp(xdp_prog, xdp); |
d4455169 | 4646 | |
065af355 | 4647 | /* check if bpf_xdp_adjust_head was used */ |
02671e23 | 4648 | off = xdp->data - orig_data; |
065af355 JDB |
4649 | if (off) { |
4650 | if (off > 0) | |
4651 | __skb_pull(skb, off); | |
4652 | else if (off < 0) | |
4653 | __skb_push(skb, -off); | |
4654 | ||
4655 | skb->mac_header += off; | |
4656 | skb_reset_network_header(skb); | |
4657 | } | |
d4455169 | 4658 | |
a075767b JDB |
4659 | /* check if bpf_xdp_adjust_tail was used */ |
4660 | off = xdp->data_end - orig_data_end; | |
f7613120 | 4661 | if (off != 0) { |
02671e23 | 4662 | skb_set_tail_pointer(skb, xdp->data_end - xdp->data); |
a075767b | 4663 | skb->len += off; /* positive on grow, negative on shrink */ |
f7613120 | 4664 | } |
198d83bb | 4665 | |
29724956 JDB |
4666 | /* check if XDP changed eth hdr such SKB needs update */ |
4667 | eth = (struct ethhdr *)xdp->data; | |
4668 | if ((orig_eth_type != eth->h_proto) || | |
4669 | (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { | |
4670 | __skb_push(skb, ETH_HLEN); | |
4671 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4672 | } | |
4673 | ||
d4455169 | 4674 | switch (act) { |
6103aa96 | 4675 | case XDP_REDIRECT: |
d4455169 JF |
4676 | case XDP_TX: |
4677 | __skb_push(skb, mac_len); | |
de8f3a83 | 4678 | break; |
d4455169 | 4679 | case XDP_PASS: |
02671e23 | 4680 | metalen = xdp->data - xdp->data_meta; |
de8f3a83 DB |
4681 | if (metalen) |
4682 | skb_metadata_set(skb, metalen); | |
d4455169 | 4683 | break; |
d4455169 JF |
4684 | default: |
4685 | bpf_warn_invalid_xdp_action(act); | |
4686 | /* fall through */ | |
4687 | case XDP_ABORTED: | |
4688 | trace_xdp_exception(skb->dev, xdp_prog, act); | |
4689 | /* fall through */ | |
4690 | case XDP_DROP: | |
4691 | do_drop: | |
4692 | kfree_skb(skb); | |
4693 | break; | |
4694 | } | |
4695 | ||
4696 | return act; | |
4697 | } | |
4698 | ||
4699 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
4700 | * network taps in order to match in-driver-XDP behavior. | |
4701 | */ | |
7c497478 | 4702 | void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) |
d4455169 JF |
4703 | { |
4704 | struct net_device *dev = skb->dev; | |
4705 | struct netdev_queue *txq; | |
4706 | bool free_skb = true; | |
4707 | int cpu, rc; | |
4708 | ||
4bd97d51 | 4709 | txq = netdev_core_pick_tx(dev, skb, NULL); |
d4455169 JF |
4710 | cpu = smp_processor_id(); |
4711 | HARD_TX_LOCK(dev, txq, cpu); | |
4712 | if (!netif_xmit_stopped(txq)) { | |
4713 | rc = netdev_start_xmit(skb, dev, txq, 0); | |
4714 | if (dev_xmit_complete(rc)) | |
4715 | free_skb = false; | |
4716 | } | |
4717 | HARD_TX_UNLOCK(dev, txq); | |
4718 | if (free_skb) { | |
4719 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
4720 | kfree_skb(skb); | |
4721 | } | |
4722 | } | |
4723 | ||
02786475 | 4724 | static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); |
d4455169 | 4725 | |
7c497478 | 4726 | int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) |
d4455169 | 4727 | { |
d4455169 | 4728 | if (xdp_prog) { |
02671e23 BT |
4729 | struct xdp_buff xdp; |
4730 | u32 act; | |
6103aa96 | 4731 | int err; |
d4455169 | 4732 | |
02671e23 | 4733 | act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); |
d4455169 | 4734 | if (act != XDP_PASS) { |
6103aa96 JF |
4735 | switch (act) { |
4736 | case XDP_REDIRECT: | |
2facaad6 | 4737 | err = xdp_do_generic_redirect(skb->dev, skb, |
02671e23 | 4738 | &xdp, xdp_prog); |
6103aa96 JF |
4739 | if (err) |
4740 | goto out_redir; | |
02671e23 | 4741 | break; |
6103aa96 | 4742 | case XDP_TX: |
d4455169 | 4743 | generic_xdp_tx(skb, xdp_prog); |
6103aa96 JF |
4744 | break; |
4745 | } | |
d4455169 JF |
4746 | return XDP_DROP; |
4747 | } | |
4748 | } | |
4749 | return XDP_PASS; | |
6103aa96 | 4750 | out_redir: |
6103aa96 JF |
4751 | kfree_skb(skb); |
4752 | return XDP_DROP; | |
d4455169 | 4753 | } |
7c497478 | 4754 | EXPORT_SYMBOL_GPL(do_xdp_generic); |
d4455169 | 4755 | |
ae78dbfa | 4756 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 4757 | { |
b0e28f1e | 4758 | int ret; |
1da177e4 | 4759 | |
588f0330 | 4760 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 4761 | |
cf66ba58 | 4762 | trace_netif_rx(skb); |
d4455169 | 4763 | |
df334545 | 4764 | #ifdef CONFIG_RPS |
dc05360f | 4765 | if (static_branch_unlikely(&rps_needed)) { |
fec5e652 | 4766 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
4767 | int cpu; |
4768 | ||
cece1945 | 4769 | preempt_disable(); |
b0e28f1e | 4770 | rcu_read_lock(); |
fec5e652 TH |
4771 | |
4772 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
4773 | if (cpu < 0) |
4774 | cpu = smp_processor_id(); | |
fec5e652 TH |
4775 | |
4776 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4777 | ||
b0e28f1e | 4778 | rcu_read_unlock(); |
cece1945 | 4779 | preempt_enable(); |
adc9300e ED |
4780 | } else |
4781 | #endif | |
fec5e652 TH |
4782 | { |
4783 | unsigned int qtail; | |
f4563a75 | 4784 | |
fec5e652 TH |
4785 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); |
4786 | put_cpu(); | |
4787 | } | |
b0e28f1e | 4788 | return ret; |
1da177e4 | 4789 | } |
ae78dbfa BH |
4790 | |
4791 | /** | |
4792 | * netif_rx - post buffer to the network code | |
4793 | * @skb: buffer to post | |
4794 | * | |
4795 | * This function receives a packet from a device driver and queues it for | |
4796 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
4797 | * may be dropped during processing for congestion control or by the | |
4798 | * protocol layers. | |
4799 | * | |
4800 | * return values: | |
4801 | * NET_RX_SUCCESS (no congestion) | |
4802 | * NET_RX_DROP (packet was dropped) | |
4803 | * | |
4804 | */ | |
4805 | ||
4806 | int netif_rx(struct sk_buff *skb) | |
4807 | { | |
b0e3f1bd GB |
4808 | int ret; |
4809 | ||
ae78dbfa BH |
4810 | trace_netif_rx_entry(skb); |
4811 | ||
b0e3f1bd GB |
4812 | ret = netif_rx_internal(skb); |
4813 | trace_netif_rx_exit(ret); | |
4814 | ||
4815 | return ret; | |
ae78dbfa | 4816 | } |
d1b19dff | 4817 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
4818 | |
4819 | int netif_rx_ni(struct sk_buff *skb) | |
4820 | { | |
4821 | int err; | |
4822 | ||
ae78dbfa BH |
4823 | trace_netif_rx_ni_entry(skb); |
4824 | ||
1da177e4 | 4825 | preempt_disable(); |
ae78dbfa | 4826 | err = netif_rx_internal(skb); |
1da177e4 LT |
4827 | if (local_softirq_pending()) |
4828 | do_softirq(); | |
4829 | preempt_enable(); | |
b0e3f1bd | 4830 | trace_netif_rx_ni_exit(err); |
1da177e4 LT |
4831 | |
4832 | return err; | |
4833 | } | |
1da177e4 LT |
4834 | EXPORT_SYMBOL(netif_rx_ni); |
4835 | ||
0766f788 | 4836 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 4837 | { |
903ceff7 | 4838 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
4839 | |
4840 | if (sd->completion_queue) { | |
4841 | struct sk_buff *clist; | |
4842 | ||
4843 | local_irq_disable(); | |
4844 | clist = sd->completion_queue; | |
4845 | sd->completion_queue = NULL; | |
4846 | local_irq_enable(); | |
4847 | ||
4848 | while (clist) { | |
4849 | struct sk_buff *skb = clist; | |
f4563a75 | 4850 | |
1da177e4 LT |
4851 | clist = clist->next; |
4852 | ||
63354797 | 4853 | WARN_ON(refcount_read(&skb->users)); |
e6247027 ED |
4854 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
4855 | trace_consume_skb(skb); | |
4856 | else | |
4857 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
4858 | |
4859 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
4860 | __kfree_skb(skb); | |
4861 | else | |
4862 | __kfree_skb_defer(skb); | |
1da177e4 | 4863 | } |
15fad714 JDB |
4864 | |
4865 | __kfree_skb_flush(); | |
1da177e4 LT |
4866 | } |
4867 | ||
4868 | if (sd->output_queue) { | |
37437bb2 | 4869 | struct Qdisc *head; |
1da177e4 LT |
4870 | |
4871 | local_irq_disable(); | |
4872 | head = sd->output_queue; | |
4873 | sd->output_queue = NULL; | |
a9cbd588 | 4874 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
4875 | local_irq_enable(); |
4876 | ||
4877 | while (head) { | |
37437bb2 | 4878 | struct Qdisc *q = head; |
6b3ba914 | 4879 | spinlock_t *root_lock = NULL; |
37437bb2 | 4880 | |
1da177e4 LT |
4881 | head = head->next_sched; |
4882 | ||
6b3ba914 JF |
4883 | if (!(q->flags & TCQ_F_NOLOCK)) { |
4884 | root_lock = qdisc_lock(q); | |
4885 | spin_lock(root_lock); | |
4886 | } | |
3bcb846c ED |
4887 | /* We need to make sure head->next_sched is read |
4888 | * before clearing __QDISC_STATE_SCHED | |
4889 | */ | |
4890 | smp_mb__before_atomic(); | |
4891 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
4892 | qdisc_run(q); | |
6b3ba914 JF |
4893 | if (root_lock) |
4894 | spin_unlock(root_lock); | |
1da177e4 LT |
4895 | } |
4896 | } | |
f53c7239 SK |
4897 | |
4898 | xfrm_dev_backlog(sd); | |
1da177e4 LT |
4899 | } |
4900 | ||
181402a5 | 4901 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
4902 | /* This hook is defined here for ATM LANE */ |
4903 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
4904 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 4905 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 4906 | #endif |
1da177e4 | 4907 | |
1f211a1b DB |
4908 | static inline struct sk_buff * |
4909 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
4910 | struct net_device *orig_dev) | |
f697c3e8 | 4911 | { |
e7582bab | 4912 | #ifdef CONFIG_NET_CLS_ACT |
46209401 | 4913 | struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); |
d2788d34 | 4914 | struct tcf_result cl_res; |
24824a09 | 4915 | |
c9e99fd0 DB |
4916 | /* If there's at least one ingress present somewhere (so |
4917 | * we get here via enabled static key), remaining devices | |
4918 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 4919 | * out here. |
c9e99fd0 | 4920 | */ |
46209401 | 4921 | if (!miniq) |
4577139b | 4922 | return skb; |
46209401 | 4923 | |
f697c3e8 HX |
4924 | if (*pt_prev) { |
4925 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4926 | *pt_prev = NULL; | |
1da177e4 LT |
4927 | } |
4928 | ||
3365495c | 4929 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
8dc07fdb | 4930 | skb->tc_at_ingress = 1; |
46209401 | 4931 | mini_qdisc_bstats_cpu_update(miniq, skb); |
c9e99fd0 | 4932 | |
7d17c544 PB |
4933 | switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list, |
4934 | &cl_res, false)) { | |
d2788d34 DB |
4935 | case TC_ACT_OK: |
4936 | case TC_ACT_RECLASSIFY: | |
4937 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
4938 | break; | |
4939 | case TC_ACT_SHOT: | |
46209401 | 4940 | mini_qdisc_qstats_cpu_drop(miniq); |
8a3a4c6e ED |
4941 | kfree_skb(skb); |
4942 | return NULL; | |
d2788d34 DB |
4943 | case TC_ACT_STOLEN: |
4944 | case TC_ACT_QUEUED: | |
e25ea21f | 4945 | case TC_ACT_TRAP: |
8a3a4c6e | 4946 | consume_skb(skb); |
d2788d34 | 4947 | return NULL; |
27b29f63 AS |
4948 | case TC_ACT_REDIRECT: |
4949 | /* skb_mac_header check was done by cls/act_bpf, so | |
4950 | * we can safely push the L2 header back before | |
4951 | * redirecting to another netdev | |
4952 | */ | |
4953 | __skb_push(skb, skb->mac_len); | |
4954 | skb_do_redirect(skb); | |
4955 | return NULL; | |
720f22fe | 4956 | case TC_ACT_CONSUMED: |
cd11b164 | 4957 | return NULL; |
d2788d34 DB |
4958 | default: |
4959 | break; | |
f697c3e8 | 4960 | } |
e7582bab | 4961 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
4962 | return skb; |
4963 | } | |
1da177e4 | 4964 | |
24b27fc4 MB |
4965 | /** |
4966 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
4967 | * @dev: device to check | |
4968 | * | |
4969 | * Check if a receive handler is already registered for a given device. | |
4970 | * Return true if there one. | |
4971 | * | |
4972 | * The caller must hold the rtnl_mutex. | |
4973 | */ | |
4974 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
4975 | { | |
4976 | ASSERT_RTNL(); | |
4977 | return dev && rtnl_dereference(dev->rx_handler); | |
4978 | } | |
4979 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
4980 | ||
ab95bfe0 JP |
4981 | /** |
4982 | * netdev_rx_handler_register - register receive handler | |
4983 | * @dev: device to register a handler for | |
4984 | * @rx_handler: receive handler to register | |
93e2c32b | 4985 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 4986 | * |
e227867f | 4987 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
4988 | * called from __netif_receive_skb. A negative errno code is returned |
4989 | * on a failure. | |
4990 | * | |
4991 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
4992 | * |
4993 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
4994 | */ |
4995 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
4996 | rx_handler_func_t *rx_handler, |
4997 | void *rx_handler_data) | |
ab95bfe0 | 4998 | { |
1b7cd004 | 4999 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
5000 | return -EBUSY; |
5001 | ||
f5426250 PA |
5002 | if (dev->priv_flags & IFF_NO_RX_HANDLER) |
5003 | return -EINVAL; | |
5004 | ||
00cfec37 | 5005 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 5006 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
5007 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
5008 | ||
5009 | return 0; | |
5010 | } | |
5011 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
5012 | ||
5013 | /** | |
5014 | * netdev_rx_handler_unregister - unregister receive handler | |
5015 | * @dev: device to unregister a handler from | |
5016 | * | |
166ec369 | 5017 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
5018 | * |
5019 | * The caller must hold the rtnl_mutex. | |
5020 | */ | |
5021 | void netdev_rx_handler_unregister(struct net_device *dev) | |
5022 | { | |
5023 | ||
5024 | ASSERT_RTNL(); | |
a9b3cd7f | 5025 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
5026 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
5027 | * section has a guarantee to see a non NULL rx_handler_data | |
5028 | * as well. | |
5029 | */ | |
5030 | synchronize_net(); | |
a9b3cd7f | 5031 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
5032 | } |
5033 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
5034 | ||
b4b9e355 MG |
5035 | /* |
5036 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
5037 | * the special handling of PFMEMALLOC skbs. | |
5038 | */ | |
5039 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
5040 | { | |
5041 | switch (skb->protocol) { | |
2b8837ae JP |
5042 | case htons(ETH_P_ARP): |
5043 | case htons(ETH_P_IP): | |
5044 | case htons(ETH_P_IPV6): | |
5045 | case htons(ETH_P_8021Q): | |
5046 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
5047 | return true; |
5048 | default: | |
5049 | return false; | |
5050 | } | |
5051 | } | |
5052 | ||
e687ad60 PN |
5053 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
5054 | int *ret, struct net_device *orig_dev) | |
5055 | { | |
5056 | if (nf_hook_ingress_active(skb)) { | |
2c1e2703 AC |
5057 | int ingress_retval; |
5058 | ||
e687ad60 PN |
5059 | if (*pt_prev) { |
5060 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
5061 | *pt_prev = NULL; | |
5062 | } | |
5063 | ||
2c1e2703 AC |
5064 | rcu_read_lock(); |
5065 | ingress_retval = nf_hook_ingress(skb); | |
5066 | rcu_read_unlock(); | |
5067 | return ingress_retval; | |
e687ad60 PN |
5068 | } |
5069 | return 0; | |
5070 | } | |
e687ad60 | 5071 | |
c0bbbdc3 | 5072 | static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, |
88eb1944 | 5073 | struct packet_type **ppt_prev) |
1da177e4 LT |
5074 | { |
5075 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 5076 | rx_handler_func_t *rx_handler; |
c0bbbdc3 | 5077 | struct sk_buff *skb = *pskb; |
f2ccd8fa | 5078 | struct net_device *orig_dev; |
8a4eb573 | 5079 | bool deliver_exact = false; |
1da177e4 | 5080 | int ret = NET_RX_DROP; |
252e3346 | 5081 | __be16 type; |
1da177e4 | 5082 | |
588f0330 | 5083 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 5084 | |
cf66ba58 | 5085 | trace_netif_receive_skb(skb); |
9b22ea56 | 5086 | |
cc9bd5ce | 5087 | orig_dev = skb->dev; |
8f903c70 | 5088 | |
c1d2bbe1 | 5089 | skb_reset_network_header(skb); |
fda55eca ED |
5090 | if (!skb_transport_header_was_set(skb)) |
5091 | skb_reset_transport_header(skb); | |
0b5c9db1 | 5092 | skb_reset_mac_len(skb); |
1da177e4 LT |
5093 | |
5094 | pt_prev = NULL; | |
5095 | ||
63d8ea7f | 5096 | another_round: |
b6858177 | 5097 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
5098 | |
5099 | __this_cpu_inc(softnet_data.processed); | |
5100 | ||
458bf2f2 SH |
5101 | if (static_branch_unlikely(&generic_xdp_needed_key)) { |
5102 | int ret2; | |
5103 | ||
5104 | preempt_disable(); | |
5105 | ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
5106 | preempt_enable(); | |
5107 | ||
c0bbbdc3 BS |
5108 | if (ret2 != XDP_PASS) { |
5109 | ret = NET_RX_DROP; | |
5110 | goto out; | |
5111 | } | |
458bf2f2 SH |
5112 | skb_reset_mac_len(skb); |
5113 | } | |
5114 | ||
8ad227ff PM |
5115 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
5116 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 5117 | skb = skb_vlan_untag(skb); |
bcc6d479 | 5118 | if (unlikely(!skb)) |
2c17d27c | 5119 | goto out; |
bcc6d479 JP |
5120 | } |
5121 | ||
e7246e12 WB |
5122 | if (skb_skip_tc_classify(skb)) |
5123 | goto skip_classify; | |
1da177e4 | 5124 | |
9754e293 | 5125 | if (pfmemalloc) |
b4b9e355 MG |
5126 | goto skip_taps; |
5127 | ||
1da177e4 | 5128 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
5129 | if (pt_prev) |
5130 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5131 | pt_prev = ptype; | |
5132 | } | |
5133 | ||
5134 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
5135 | if (pt_prev) | |
5136 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5137 | pt_prev = ptype; | |
1da177e4 LT |
5138 | } |
5139 | ||
b4b9e355 | 5140 | skip_taps: |
1cf51900 | 5141 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 5142 | if (static_branch_unlikely(&ingress_needed_key)) { |
1f211a1b | 5143 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); |
4577139b | 5144 | if (!skb) |
2c17d27c | 5145 | goto out; |
e687ad60 PN |
5146 | |
5147 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 5148 | goto out; |
4577139b | 5149 | } |
1cf51900 | 5150 | #endif |
2c64605b | 5151 | skb_reset_redirect(skb); |
e7246e12 | 5152 | skip_classify: |
9754e293 | 5153 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
5154 | goto drop; |
5155 | ||
df8a39de | 5156 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
5157 | if (pt_prev) { |
5158 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5159 | pt_prev = NULL; | |
5160 | } | |
48cc32d3 | 5161 | if (vlan_do_receive(&skb)) |
2425717b JF |
5162 | goto another_round; |
5163 | else if (unlikely(!skb)) | |
2c17d27c | 5164 | goto out; |
2425717b JF |
5165 | } |
5166 | ||
48cc32d3 | 5167 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
5168 | if (rx_handler) { |
5169 | if (pt_prev) { | |
5170 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5171 | pt_prev = NULL; | |
5172 | } | |
8a4eb573 JP |
5173 | switch (rx_handler(&skb)) { |
5174 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 5175 | ret = NET_RX_SUCCESS; |
2c17d27c | 5176 | goto out; |
8a4eb573 | 5177 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 5178 | goto another_round; |
8a4eb573 JP |
5179 | case RX_HANDLER_EXACT: |
5180 | deliver_exact = true; | |
5181 | case RX_HANDLER_PASS: | |
5182 | break; | |
5183 | default: | |
5184 | BUG(); | |
5185 | } | |
ab95bfe0 | 5186 | } |
1da177e4 | 5187 | |
df8a39de | 5188 | if (unlikely(skb_vlan_tag_present(skb))) { |
36b2f61a GV |
5189 | check_vlan_id: |
5190 | if (skb_vlan_tag_get_id(skb)) { | |
5191 | /* Vlan id is non 0 and vlan_do_receive() above couldn't | |
5192 | * find vlan device. | |
5193 | */ | |
d4b812de | 5194 | skb->pkt_type = PACKET_OTHERHOST; |
36b2f61a GV |
5195 | } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
5196 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
5197 | /* Outer header is 802.1P with vlan 0, inner header is | |
5198 | * 802.1Q or 802.1AD and vlan_do_receive() above could | |
5199 | * not find vlan dev for vlan id 0. | |
5200 | */ | |
5201 | __vlan_hwaccel_clear_tag(skb); | |
5202 | skb = skb_vlan_untag(skb); | |
5203 | if (unlikely(!skb)) | |
5204 | goto out; | |
5205 | if (vlan_do_receive(&skb)) | |
5206 | /* After stripping off 802.1P header with vlan 0 | |
5207 | * vlan dev is found for inner header. | |
5208 | */ | |
5209 | goto another_round; | |
5210 | else if (unlikely(!skb)) | |
5211 | goto out; | |
5212 | else | |
5213 | /* We have stripped outer 802.1P vlan 0 header. | |
5214 | * But could not find vlan dev. | |
5215 | * check again for vlan id to set OTHERHOST. | |
5216 | */ | |
5217 | goto check_vlan_id; | |
5218 | } | |
d4b812de ED |
5219 | /* Note: we might in the future use prio bits |
5220 | * and set skb->priority like in vlan_do_receive() | |
5221 | * For the time being, just ignore Priority Code Point | |
5222 | */ | |
b1817524 | 5223 | __vlan_hwaccel_clear_tag(skb); |
d4b812de | 5224 | } |
48cc32d3 | 5225 | |
7866a621 SN |
5226 | type = skb->protocol; |
5227 | ||
63d8ea7f | 5228 | /* deliver only exact match when indicated */ |
7866a621 SN |
5229 | if (likely(!deliver_exact)) { |
5230 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
5231 | &ptype_base[ntohs(type) & | |
5232 | PTYPE_HASH_MASK]); | |
5233 | } | |
1f3c8804 | 5234 | |
7866a621 SN |
5235 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
5236 | &orig_dev->ptype_specific); | |
5237 | ||
5238 | if (unlikely(skb->dev != orig_dev)) { | |
5239 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
5240 | &skb->dev->ptype_specific); | |
1da177e4 LT |
5241 | } |
5242 | ||
5243 | if (pt_prev) { | |
1f8b977a | 5244 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
0e698bf6 | 5245 | goto drop; |
88eb1944 | 5246 | *ppt_prev = pt_prev; |
1da177e4 | 5247 | } else { |
b4b9e355 | 5248 | drop: |
6e7333d3 JW |
5249 | if (!deliver_exact) |
5250 | atomic_long_inc(&skb->dev->rx_dropped); | |
5251 | else | |
5252 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
5253 | kfree_skb(skb); |
5254 | /* Jamal, now you will not able to escape explaining | |
5255 | * me how you were going to use this. :-) | |
5256 | */ | |
5257 | ret = NET_RX_DROP; | |
5258 | } | |
5259 | ||
2c17d27c | 5260 | out: |
c0bbbdc3 BS |
5261 | /* The invariant here is that if *ppt_prev is not NULL |
5262 | * then skb should also be non-NULL. | |
5263 | * | |
5264 | * Apparently *ppt_prev assignment above holds this invariant due to | |
5265 | * skb dereferencing near it. | |
5266 | */ | |
5267 | *pskb = skb; | |
9754e293 DM |
5268 | return ret; |
5269 | } | |
5270 | ||
88eb1944 EC |
5271 | static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) |
5272 | { | |
5273 | struct net_device *orig_dev = skb->dev; | |
5274 | struct packet_type *pt_prev = NULL; | |
5275 | int ret; | |
5276 | ||
c0bbbdc3 | 5277 | ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); |
88eb1944 | 5278 | if (pt_prev) |
f5737cba PA |
5279 | ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, |
5280 | skb->dev, pt_prev, orig_dev); | |
88eb1944 EC |
5281 | return ret; |
5282 | } | |
5283 | ||
1c601d82 JDB |
5284 | /** |
5285 | * netif_receive_skb_core - special purpose version of netif_receive_skb | |
5286 | * @skb: buffer to process | |
5287 | * | |
5288 | * More direct receive version of netif_receive_skb(). It should | |
5289 | * only be used by callers that have a need to skip RPS and Generic XDP. | |
2de9780f | 5290 | * Caller must also take care of handling if ``(page_is_)pfmemalloc``. |
1c601d82 JDB |
5291 | * |
5292 | * This function may only be called from softirq context and interrupts | |
5293 | * should be enabled. | |
5294 | * | |
5295 | * Return values (usually ignored): | |
5296 | * NET_RX_SUCCESS: no congestion | |
5297 | * NET_RX_DROP: packet was dropped | |
5298 | */ | |
5299 | int netif_receive_skb_core(struct sk_buff *skb) | |
5300 | { | |
5301 | int ret; | |
5302 | ||
5303 | rcu_read_lock(); | |
88eb1944 | 5304 | ret = __netif_receive_skb_one_core(skb, false); |
1c601d82 JDB |
5305 | rcu_read_unlock(); |
5306 | ||
5307 | return ret; | |
5308 | } | |
5309 | EXPORT_SYMBOL(netif_receive_skb_core); | |
5310 | ||
88eb1944 EC |
5311 | static inline void __netif_receive_skb_list_ptype(struct list_head *head, |
5312 | struct packet_type *pt_prev, | |
5313 | struct net_device *orig_dev) | |
4ce0017a EC |
5314 | { |
5315 | struct sk_buff *skb, *next; | |
5316 | ||
88eb1944 EC |
5317 | if (!pt_prev) |
5318 | return; | |
5319 | if (list_empty(head)) | |
5320 | return; | |
17266ee9 | 5321 | if (pt_prev->list_func != NULL) |
fdf71426 PA |
5322 | INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, |
5323 | ip_list_rcv, head, pt_prev, orig_dev); | |
17266ee9 | 5324 | else |
9a5a90d1 AL |
5325 | list_for_each_entry_safe(skb, next, head, list) { |
5326 | skb_list_del_init(skb); | |
fdf71426 | 5327 | pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
9a5a90d1 | 5328 | } |
88eb1944 EC |
5329 | } |
5330 | ||
5331 | static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) | |
5332 | { | |
5333 | /* Fast-path assumptions: | |
5334 | * - There is no RX handler. | |
5335 | * - Only one packet_type matches. | |
5336 | * If either of these fails, we will end up doing some per-packet | |
5337 | * processing in-line, then handling the 'last ptype' for the whole | |
5338 | * sublist. This can't cause out-of-order delivery to any single ptype, | |
5339 | * because the 'last ptype' must be constant across the sublist, and all | |
5340 | * other ptypes are handled per-packet. | |
5341 | */ | |
5342 | /* Current (common) ptype of sublist */ | |
5343 | struct packet_type *pt_curr = NULL; | |
5344 | /* Current (common) orig_dev of sublist */ | |
5345 | struct net_device *od_curr = NULL; | |
5346 | struct list_head sublist; | |
5347 | struct sk_buff *skb, *next; | |
5348 | ||
9af86f93 | 5349 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5350 | list_for_each_entry_safe(skb, next, head, list) { |
5351 | struct net_device *orig_dev = skb->dev; | |
5352 | struct packet_type *pt_prev = NULL; | |
5353 | ||
22f6bbb7 | 5354 | skb_list_del_init(skb); |
c0bbbdc3 | 5355 | __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); |
9af86f93 EC |
5356 | if (!pt_prev) |
5357 | continue; | |
88eb1944 EC |
5358 | if (pt_curr != pt_prev || od_curr != orig_dev) { |
5359 | /* dispatch old sublist */ | |
88eb1944 EC |
5360 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
5361 | /* start new sublist */ | |
9af86f93 | 5362 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5363 | pt_curr = pt_prev; |
5364 | od_curr = orig_dev; | |
5365 | } | |
9af86f93 | 5366 | list_add_tail(&skb->list, &sublist); |
88eb1944 EC |
5367 | } |
5368 | ||
5369 | /* dispatch final sublist */ | |
9af86f93 | 5370 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
4ce0017a EC |
5371 | } |
5372 | ||
9754e293 DM |
5373 | static int __netif_receive_skb(struct sk_buff *skb) |
5374 | { | |
5375 | int ret; | |
5376 | ||
5377 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 5378 | unsigned int noreclaim_flag; |
9754e293 DM |
5379 | |
5380 | /* | |
5381 | * PFMEMALLOC skbs are special, they should | |
5382 | * - be delivered to SOCK_MEMALLOC sockets only | |
5383 | * - stay away from userspace | |
5384 | * - have bounded memory usage | |
5385 | * | |
5386 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
5387 | * context down to all allocation sites. | |
5388 | */ | |
f1083048 | 5389 | noreclaim_flag = memalloc_noreclaim_save(); |
88eb1944 | 5390 | ret = __netif_receive_skb_one_core(skb, true); |
f1083048 | 5391 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 | 5392 | } else |
88eb1944 | 5393 | ret = __netif_receive_skb_one_core(skb, false); |
9754e293 | 5394 | |
1da177e4 LT |
5395 | return ret; |
5396 | } | |
0a9627f2 | 5397 | |
4ce0017a EC |
5398 | static void __netif_receive_skb_list(struct list_head *head) |
5399 | { | |
5400 | unsigned long noreclaim_flag = 0; | |
5401 | struct sk_buff *skb, *next; | |
5402 | bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ | |
5403 | ||
5404 | list_for_each_entry_safe(skb, next, head, list) { | |
5405 | if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { | |
5406 | struct list_head sublist; | |
5407 | ||
5408 | /* Handle the previous sublist */ | |
5409 | list_cut_before(&sublist, head, &skb->list); | |
b9f463d6 EC |
5410 | if (!list_empty(&sublist)) |
5411 | __netif_receive_skb_list_core(&sublist, pfmemalloc); | |
4ce0017a EC |
5412 | pfmemalloc = !pfmemalloc; |
5413 | /* See comments in __netif_receive_skb */ | |
5414 | if (pfmemalloc) | |
5415 | noreclaim_flag = memalloc_noreclaim_save(); | |
5416 | else | |
5417 | memalloc_noreclaim_restore(noreclaim_flag); | |
5418 | } | |
5419 | } | |
5420 | /* Handle the remaining sublist */ | |
b9f463d6 EC |
5421 | if (!list_empty(head)) |
5422 | __netif_receive_skb_list_core(head, pfmemalloc); | |
4ce0017a EC |
5423 | /* Restore pflags */ |
5424 | if (pfmemalloc) | |
5425 | memalloc_noreclaim_restore(noreclaim_flag); | |
5426 | } | |
5427 | ||
f4e63525 | 5428 | static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) |
b5cdae32 | 5429 | { |
58038695 | 5430 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); |
b5cdae32 DM |
5431 | struct bpf_prog *new = xdp->prog; |
5432 | int ret = 0; | |
5433 | ||
fbee97fe DA |
5434 | if (new) { |
5435 | u32 i; | |
5436 | ||
5437 | /* generic XDP does not work with DEVMAPs that can | |
5438 | * have a bpf_prog installed on an entry | |
5439 | */ | |
5440 | for (i = 0; i < new->aux->used_map_cnt; i++) { | |
5441 | if (dev_map_can_have_prog(new->aux->used_maps[i])) | |
5442 | return -EINVAL; | |
5443 | } | |
5444 | } | |
5445 | ||
b5cdae32 | 5446 | switch (xdp->command) { |
58038695 | 5447 | case XDP_SETUP_PROG: |
b5cdae32 DM |
5448 | rcu_assign_pointer(dev->xdp_prog, new); |
5449 | if (old) | |
5450 | bpf_prog_put(old); | |
5451 | ||
5452 | if (old && !new) { | |
02786475 | 5453 | static_branch_dec(&generic_xdp_needed_key); |
b5cdae32 | 5454 | } else if (new && !old) { |
02786475 | 5455 | static_branch_inc(&generic_xdp_needed_key); |
b5cdae32 | 5456 | dev_disable_lro(dev); |
56f5aa77 | 5457 | dev_disable_gro_hw(dev); |
b5cdae32 DM |
5458 | } |
5459 | break; | |
b5cdae32 DM |
5460 | |
5461 | case XDP_QUERY_PROG: | |
58038695 | 5462 | xdp->prog_id = old ? old->aux->id : 0; |
b5cdae32 DM |
5463 | break; |
5464 | ||
5465 | default: | |
5466 | ret = -EINVAL; | |
5467 | break; | |
5468 | } | |
5469 | ||
5470 | return ret; | |
5471 | } | |
5472 | ||
ae78dbfa | 5473 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 5474 | { |
2c17d27c JA |
5475 | int ret; |
5476 | ||
588f0330 | 5477 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 5478 | |
c1f19b51 RC |
5479 | if (skb_defer_rx_timestamp(skb)) |
5480 | return NET_RX_SUCCESS; | |
5481 | ||
bbbe211c | 5482 | rcu_read_lock(); |
df334545 | 5483 | #ifdef CONFIG_RPS |
dc05360f | 5484 | if (static_branch_unlikely(&rps_needed)) { |
3b098e2d | 5485 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 5486 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 5487 | |
3b098e2d ED |
5488 | if (cpu >= 0) { |
5489 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
5490 | rcu_read_unlock(); | |
adc9300e | 5491 | return ret; |
3b098e2d | 5492 | } |
fec5e652 | 5493 | } |
1e94d72f | 5494 | #endif |
2c17d27c JA |
5495 | ret = __netif_receive_skb(skb); |
5496 | rcu_read_unlock(); | |
5497 | return ret; | |
0a9627f2 | 5498 | } |
ae78dbfa | 5499 | |
7da517a3 EC |
5500 | static void netif_receive_skb_list_internal(struct list_head *head) |
5501 | { | |
7da517a3 | 5502 | struct sk_buff *skb, *next; |
8c057efa | 5503 | struct list_head sublist; |
7da517a3 | 5504 | |
8c057efa | 5505 | INIT_LIST_HEAD(&sublist); |
7da517a3 EC |
5506 | list_for_each_entry_safe(skb, next, head, list) { |
5507 | net_timestamp_check(netdev_tstamp_prequeue, skb); | |
22f6bbb7 | 5508 | skb_list_del_init(skb); |
8c057efa EC |
5509 | if (!skb_defer_rx_timestamp(skb)) |
5510 | list_add_tail(&skb->list, &sublist); | |
7da517a3 | 5511 | } |
8c057efa | 5512 | list_splice_init(&sublist, head); |
7da517a3 | 5513 | |
7da517a3 EC |
5514 | rcu_read_lock(); |
5515 | #ifdef CONFIG_RPS | |
dc05360f | 5516 | if (static_branch_unlikely(&rps_needed)) { |
7da517a3 EC |
5517 | list_for_each_entry_safe(skb, next, head, list) { |
5518 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
5519 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
5520 | ||
5521 | if (cpu >= 0) { | |
8c057efa | 5522 | /* Will be handled, remove from list */ |
22f6bbb7 | 5523 | skb_list_del_init(skb); |
8c057efa | 5524 | enqueue_to_backlog(skb, cpu, &rflow->last_qtail); |
7da517a3 EC |
5525 | } |
5526 | } | |
5527 | } | |
5528 | #endif | |
5529 | __netif_receive_skb_list(head); | |
5530 | rcu_read_unlock(); | |
5531 | } | |
5532 | ||
ae78dbfa BH |
5533 | /** |
5534 | * netif_receive_skb - process receive buffer from network | |
5535 | * @skb: buffer to process | |
5536 | * | |
5537 | * netif_receive_skb() is the main receive data processing function. | |
5538 | * It always succeeds. The buffer may be dropped during processing | |
5539 | * for congestion control or by the protocol layers. | |
5540 | * | |
5541 | * This function may only be called from softirq context and interrupts | |
5542 | * should be enabled. | |
5543 | * | |
5544 | * Return values (usually ignored): | |
5545 | * NET_RX_SUCCESS: no congestion | |
5546 | * NET_RX_DROP: packet was dropped | |
5547 | */ | |
04eb4489 | 5548 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa | 5549 | { |
b0e3f1bd GB |
5550 | int ret; |
5551 | ||
ae78dbfa BH |
5552 | trace_netif_receive_skb_entry(skb); |
5553 | ||
b0e3f1bd GB |
5554 | ret = netif_receive_skb_internal(skb); |
5555 | trace_netif_receive_skb_exit(ret); | |
5556 | ||
5557 | return ret; | |
ae78dbfa | 5558 | } |
04eb4489 | 5559 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 5560 | |
f6ad8c1b EC |
5561 | /** |
5562 | * netif_receive_skb_list - process many receive buffers from network | |
5563 | * @head: list of skbs to process. | |
5564 | * | |
7da517a3 EC |
5565 | * Since return value of netif_receive_skb() is normally ignored, and |
5566 | * wouldn't be meaningful for a list, this function returns void. | |
f6ad8c1b EC |
5567 | * |
5568 | * This function may only be called from softirq context and interrupts | |
5569 | * should be enabled. | |
5570 | */ | |
5571 | void netif_receive_skb_list(struct list_head *head) | |
5572 | { | |
7da517a3 | 5573 | struct sk_buff *skb; |
f6ad8c1b | 5574 | |
b9f463d6 EC |
5575 | if (list_empty(head)) |
5576 | return; | |
b0e3f1bd GB |
5577 | if (trace_netif_receive_skb_list_entry_enabled()) { |
5578 | list_for_each_entry(skb, head, list) | |
5579 | trace_netif_receive_skb_list_entry(skb); | |
5580 | } | |
7da517a3 | 5581 | netif_receive_skb_list_internal(head); |
b0e3f1bd | 5582 | trace_netif_receive_skb_list_exit(0); |
f6ad8c1b EC |
5583 | } |
5584 | EXPORT_SYMBOL(netif_receive_skb_list); | |
5585 | ||
41852497 | 5586 | DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
5587 | |
5588 | /* Network device is going away, flush any packets still pending */ | |
5589 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 5590 | { |
6e583ce5 | 5591 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
5592 | struct softnet_data *sd; |
5593 | ||
5594 | local_bh_disable(); | |
5595 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 5596 | |
145dd5f9 | 5597 | local_irq_disable(); |
e36fa2f7 | 5598 | rps_lock(sd); |
6e7676c1 | 5599 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 5600 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 5601 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 5602 | kfree_skb(skb); |
76cc8b13 | 5603 | input_queue_head_incr(sd); |
6e583ce5 | 5604 | } |
6e7676c1 | 5605 | } |
e36fa2f7 | 5606 | rps_unlock(sd); |
145dd5f9 | 5607 | local_irq_enable(); |
6e7676c1 CG |
5608 | |
5609 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 5610 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
5611 | __skb_unlink(skb, &sd->process_queue); |
5612 | kfree_skb(skb); | |
76cc8b13 | 5613 | input_queue_head_incr(sd); |
6e7676c1 CG |
5614 | } |
5615 | } | |
145dd5f9 PA |
5616 | local_bh_enable(); |
5617 | } | |
5618 | ||
41852497 | 5619 | static void flush_all_backlogs(void) |
145dd5f9 PA |
5620 | { |
5621 | unsigned int cpu; | |
5622 | ||
5623 | get_online_cpus(); | |
5624 | ||
41852497 ED |
5625 | for_each_online_cpu(cpu) |
5626 | queue_work_on(cpu, system_highpri_wq, | |
5627 | per_cpu_ptr(&flush_works, cpu)); | |
145dd5f9 PA |
5628 | |
5629 | for_each_online_cpu(cpu) | |
41852497 | 5630 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 PA |
5631 | |
5632 | put_online_cpus(); | |
6e583ce5 SH |
5633 | } |
5634 | ||
c8079432 MM |
5635 | /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ |
5636 | static void gro_normal_list(struct napi_struct *napi) | |
5637 | { | |
5638 | if (!napi->rx_count) | |
5639 | return; | |
5640 | netif_receive_skb_list_internal(&napi->rx_list); | |
5641 | INIT_LIST_HEAD(&napi->rx_list); | |
5642 | napi->rx_count = 0; | |
5643 | } | |
5644 | ||
5645 | /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, | |
5646 | * pass the whole batch up to the stack. | |
5647 | */ | |
5648 | static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb) | |
5649 | { | |
5650 | list_add_tail(&skb->list, &napi->rx_list); | |
5651 | if (++napi->rx_count >= gro_normal_batch) | |
5652 | gro_normal_list(napi); | |
5653 | } | |
5654 | ||
aaa5d90b PA |
5655 | INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int)); |
5656 | INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int)); | |
c8079432 | 5657 | static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 | 5658 | { |
22061d80 | 5659 | struct packet_offload *ptype; |
d565b0a1 | 5660 | __be16 type = skb->protocol; |
22061d80 | 5661 | struct list_head *head = &offload_base; |
d565b0a1 HX |
5662 | int err = -ENOENT; |
5663 | ||
c3c7c254 ED |
5664 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
5665 | ||
fc59f9a3 HX |
5666 | if (NAPI_GRO_CB(skb)->count == 1) { |
5667 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 5668 | goto out; |
fc59f9a3 | 5669 | } |
d565b0a1 HX |
5670 | |
5671 | rcu_read_lock(); | |
5672 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 5673 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
5674 | continue; |
5675 | ||
aaa5d90b PA |
5676 | err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, |
5677 | ipv6_gro_complete, inet_gro_complete, | |
5678 | skb, 0); | |
d565b0a1 HX |
5679 | break; |
5680 | } | |
5681 | rcu_read_unlock(); | |
5682 | ||
5683 | if (err) { | |
5684 | WARN_ON(&ptype->list == head); | |
5685 | kfree_skb(skb); | |
5686 | return NET_RX_SUCCESS; | |
5687 | } | |
5688 | ||
5689 | out: | |
c8079432 MM |
5690 | gro_normal_one(napi, skb); |
5691 | return NET_RX_SUCCESS; | |
d565b0a1 HX |
5692 | } |
5693 | ||
6312fe77 | 5694 | static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, |
07d78363 | 5695 | bool flush_old) |
d565b0a1 | 5696 | { |
6312fe77 | 5697 | struct list_head *head = &napi->gro_hash[index].list; |
d4546c25 | 5698 | struct sk_buff *skb, *p; |
2e71a6f8 | 5699 | |
07d78363 | 5700 | list_for_each_entry_safe_reverse(skb, p, head, list) { |
2e71a6f8 ED |
5701 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) |
5702 | return; | |
992cba7e | 5703 | skb_list_del_init(skb); |
c8079432 | 5704 | napi_gro_complete(napi, skb); |
6312fe77 | 5705 | napi->gro_hash[index].count--; |
d565b0a1 | 5706 | } |
d9f37d01 LR |
5707 | |
5708 | if (!napi->gro_hash[index].count) | |
5709 | __clear_bit(index, &napi->gro_bitmask); | |
d565b0a1 | 5710 | } |
07d78363 | 5711 | |
6312fe77 | 5712 | /* napi->gro_hash[].list contains packets ordered by age. |
07d78363 DM |
5713 | * youngest packets at the head of it. |
5714 | * Complete skbs in reverse order to reduce latencies. | |
5715 | */ | |
5716 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
5717 | { | |
42519ede ED |
5718 | unsigned long bitmask = napi->gro_bitmask; |
5719 | unsigned int i, base = ~0U; | |
07d78363 | 5720 | |
42519ede ED |
5721 | while ((i = ffs(bitmask)) != 0) { |
5722 | bitmask >>= i; | |
5723 | base += i; | |
5724 | __napi_gro_flush_chain(napi, base, flush_old); | |
d9f37d01 | 5725 | } |
07d78363 | 5726 | } |
86cac58b | 5727 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 5728 | |
07d78363 DM |
5729 | static struct list_head *gro_list_prepare(struct napi_struct *napi, |
5730 | struct sk_buff *skb) | |
89c5fa33 | 5731 | { |
89c5fa33 | 5732 | unsigned int maclen = skb->dev->hard_header_len; |
0b4cec8c | 5733 | u32 hash = skb_get_hash_raw(skb); |
07d78363 | 5734 | struct list_head *head; |
d4546c25 | 5735 | struct sk_buff *p; |
89c5fa33 | 5736 | |
6312fe77 | 5737 | head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list; |
07d78363 | 5738 | list_for_each_entry(p, head, list) { |
89c5fa33 ED |
5739 | unsigned long diffs; |
5740 | ||
0b4cec8c TH |
5741 | NAPI_GRO_CB(p)->flush = 0; |
5742 | ||
5743 | if (hash != skb_get_hash_raw(p)) { | |
5744 | NAPI_GRO_CB(p)->same_flow = 0; | |
5745 | continue; | |
5746 | } | |
5747 | ||
89c5fa33 | 5748 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
b1817524 MM |
5749 | diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); |
5750 | if (skb_vlan_tag_present(p)) | |
fc5141cb | 5751 | diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); |
ce87fc6c | 5752 | diffs |= skb_metadata_dst_cmp(p, skb); |
de8f3a83 | 5753 | diffs |= skb_metadata_differs(p, skb); |
89c5fa33 ED |
5754 | if (maclen == ETH_HLEN) |
5755 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 5756 | skb_mac_header(skb)); |
89c5fa33 ED |
5757 | else if (!diffs) |
5758 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 5759 | skb_mac_header(skb), |
89c5fa33 ED |
5760 | maclen); |
5761 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 | 5762 | } |
07d78363 DM |
5763 | |
5764 | return head; | |
89c5fa33 ED |
5765 | } |
5766 | ||
299603e8 JC |
5767 | static void skb_gro_reset_offset(struct sk_buff *skb) |
5768 | { | |
5769 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
5770 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
5771 | ||
5772 | NAPI_GRO_CB(skb)->data_offset = 0; | |
5773 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
5774 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
5775 | ||
8aef998d | 5776 | if (!skb_headlen(skb) && pinfo->nr_frags && |
299603e8 JC |
5777 | !PageHighMem(skb_frag_page(frag0))) { |
5778 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
7cfd5fd5 ED |
5779 | NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, |
5780 | skb_frag_size(frag0), | |
5781 | skb->end - skb->tail); | |
89c5fa33 ED |
5782 | } |
5783 | } | |
5784 | ||
a50e233c ED |
5785 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
5786 | { | |
5787 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
5788 | ||
5789 | BUG_ON(skb->end - skb->tail < grow); | |
5790 | ||
5791 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
5792 | ||
5793 | skb->data_len -= grow; | |
5794 | skb->tail += grow; | |
5795 | ||
b54c9d5b | 5796 | skb_frag_off_add(&pinfo->frags[0], grow); |
a50e233c ED |
5797 | skb_frag_size_sub(&pinfo->frags[0], grow); |
5798 | ||
5799 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
5800 | skb_frag_unref(skb, 0); | |
5801 | memmove(pinfo->frags, pinfo->frags + 1, | |
5802 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
5803 | } | |
5804 | } | |
5805 | ||
c8079432 | 5806 | static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) |
07d78363 | 5807 | { |
6312fe77 | 5808 | struct sk_buff *oldest; |
07d78363 | 5809 | |
6312fe77 | 5810 | oldest = list_last_entry(head, struct sk_buff, list); |
07d78363 | 5811 | |
6312fe77 | 5812 | /* We are called with head length >= MAX_GRO_SKBS, so this is |
07d78363 DM |
5813 | * impossible. |
5814 | */ | |
5815 | if (WARN_ON_ONCE(!oldest)) | |
5816 | return; | |
5817 | ||
d9f37d01 LR |
5818 | /* Do not adjust napi->gro_hash[].count, caller is adding a new |
5819 | * SKB to the chain. | |
07d78363 | 5820 | */ |
ece23711 | 5821 | skb_list_del_init(oldest); |
c8079432 | 5822 | napi_gro_complete(napi, oldest); |
07d78363 DM |
5823 | } |
5824 | ||
aaa5d90b PA |
5825 | INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *, |
5826 | struct sk_buff *)); | |
5827 | INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *, | |
5828 | struct sk_buff *)); | |
bb728820 | 5829 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 | 5830 | { |
6312fe77 | 5831 | u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); |
d4546c25 | 5832 | struct list_head *head = &offload_base; |
22061d80 | 5833 | struct packet_offload *ptype; |
d565b0a1 | 5834 | __be16 type = skb->protocol; |
07d78363 | 5835 | struct list_head *gro_head; |
d4546c25 | 5836 | struct sk_buff *pp = NULL; |
5b252f0c | 5837 | enum gro_result ret; |
d4546c25 | 5838 | int same_flow; |
a50e233c | 5839 | int grow; |
d565b0a1 | 5840 | |
b5cdae32 | 5841 | if (netif_elide_gro(skb->dev)) |
d565b0a1 HX |
5842 | goto normal; |
5843 | ||
07d78363 | 5844 | gro_head = gro_list_prepare(napi, skb); |
89c5fa33 | 5845 | |
d565b0a1 HX |
5846 | rcu_read_lock(); |
5847 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 5848 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
5849 | continue; |
5850 | ||
86911732 | 5851 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 5852 | skb_reset_mac_len(skb); |
d565b0a1 | 5853 | NAPI_GRO_CB(skb)->same_flow = 0; |
d61d072e | 5854 | NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); |
5d38a079 | 5855 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 5856 | NAPI_GRO_CB(skb)->encap_mark = 0; |
fcd91dd4 | 5857 | NAPI_GRO_CB(skb)->recursion_counter = 0; |
a0ca153f | 5858 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 5859 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 5860 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 5861 | |
662880f4 TH |
5862 | /* Setup for GRO checksum validation */ |
5863 | switch (skb->ip_summed) { | |
5864 | case CHECKSUM_COMPLETE: | |
5865 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
5866 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
5867 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
5868 | break; | |
5869 | case CHECKSUM_UNNECESSARY: | |
5870 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
5871 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
5872 | break; | |
5873 | default: | |
5874 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
5875 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
5876 | } | |
d565b0a1 | 5877 | |
aaa5d90b PA |
5878 | pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, |
5879 | ipv6_gro_receive, inet_gro_receive, | |
5880 | gro_head, skb); | |
d565b0a1 HX |
5881 | break; |
5882 | } | |
5883 | rcu_read_unlock(); | |
5884 | ||
5885 | if (&ptype->list == head) | |
5886 | goto normal; | |
5887 | ||
45586c70 | 5888 | if (PTR_ERR(pp) == -EINPROGRESS) { |
25393d3f SK |
5889 | ret = GRO_CONSUMED; |
5890 | goto ok; | |
5891 | } | |
5892 | ||
0da2afd5 | 5893 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 5894 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 5895 | |
d565b0a1 | 5896 | if (pp) { |
992cba7e | 5897 | skb_list_del_init(pp); |
c8079432 | 5898 | napi_gro_complete(napi, pp); |
6312fe77 | 5899 | napi->gro_hash[hash].count--; |
d565b0a1 HX |
5900 | } |
5901 | ||
0da2afd5 | 5902 | if (same_flow) |
d565b0a1 HX |
5903 | goto ok; |
5904 | ||
600adc18 | 5905 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 5906 | goto normal; |
d565b0a1 | 5907 | |
6312fe77 | 5908 | if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) { |
c8079432 | 5909 | gro_flush_oldest(napi, gro_head); |
600adc18 | 5910 | } else { |
6312fe77 | 5911 | napi->gro_hash[hash].count++; |
600adc18 | 5912 | } |
d565b0a1 | 5913 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 5914 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 5915 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 5916 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
07d78363 | 5917 | list_add(&skb->list, gro_head); |
5d0d9be8 | 5918 | ret = GRO_HELD; |
d565b0a1 | 5919 | |
ad0f9904 | 5920 | pull: |
a50e233c ED |
5921 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
5922 | if (grow > 0) | |
5923 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 5924 | ok: |
d9f37d01 LR |
5925 | if (napi->gro_hash[hash].count) { |
5926 | if (!test_bit(hash, &napi->gro_bitmask)) | |
5927 | __set_bit(hash, &napi->gro_bitmask); | |
5928 | } else if (test_bit(hash, &napi->gro_bitmask)) { | |
5929 | __clear_bit(hash, &napi->gro_bitmask); | |
5930 | } | |
5931 | ||
5d0d9be8 | 5932 | return ret; |
d565b0a1 HX |
5933 | |
5934 | normal: | |
ad0f9904 HX |
5935 | ret = GRO_NORMAL; |
5936 | goto pull; | |
5d38a079 | 5937 | } |
96e93eab | 5938 | |
bf5a755f JC |
5939 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
5940 | { | |
5941 | struct list_head *offload_head = &offload_base; | |
5942 | struct packet_offload *ptype; | |
5943 | ||
5944 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
5945 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
5946 | continue; | |
5947 | return ptype; | |
5948 | } | |
5949 | return NULL; | |
5950 | } | |
e27a2f83 | 5951 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
5952 | |
5953 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
5954 | { | |
5955 | struct list_head *offload_head = &offload_base; | |
5956 | struct packet_offload *ptype; | |
5957 | ||
5958 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
5959 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
5960 | continue; | |
5961 | return ptype; | |
5962 | } | |
5963 | return NULL; | |
5964 | } | |
e27a2f83 | 5965 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 5966 | |
e44699d2 MK |
5967 | static void napi_skb_free_stolen_head(struct sk_buff *skb) |
5968 | { | |
5969 | skb_dst_drop(skb); | |
174e2381 | 5970 | skb_ext_put(skb); |
e44699d2 MK |
5971 | kmem_cache_free(skbuff_head_cache, skb); |
5972 | } | |
5973 | ||
6570bc79 AL |
5974 | static gro_result_t napi_skb_finish(struct napi_struct *napi, |
5975 | struct sk_buff *skb, | |
5976 | gro_result_t ret) | |
5d38a079 | 5977 | { |
5d0d9be8 HX |
5978 | switch (ret) { |
5979 | case GRO_NORMAL: | |
6570bc79 | 5980 | gro_normal_one(napi, skb); |
c7c4b3b6 | 5981 | break; |
5d38a079 | 5982 | |
5d0d9be8 | 5983 | case GRO_DROP: |
5d38a079 HX |
5984 | kfree_skb(skb); |
5985 | break; | |
5b252f0c | 5986 | |
daa86548 | 5987 | case GRO_MERGED_FREE: |
e44699d2 MK |
5988 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
5989 | napi_skb_free_stolen_head(skb); | |
5990 | else | |
d7e8883c | 5991 | __kfree_skb(skb); |
daa86548 ED |
5992 | break; |
5993 | ||
5b252f0c BH |
5994 | case GRO_HELD: |
5995 | case GRO_MERGED: | |
25393d3f | 5996 | case GRO_CONSUMED: |
5b252f0c | 5997 | break; |
5d38a079 HX |
5998 | } |
5999 | ||
c7c4b3b6 | 6000 | return ret; |
5d0d9be8 | 6001 | } |
5d0d9be8 | 6002 | |
c7c4b3b6 | 6003 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 6004 | { |
b0e3f1bd GB |
6005 | gro_result_t ret; |
6006 | ||
93f93a44 | 6007 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 6008 | trace_napi_gro_receive_entry(skb); |
86911732 | 6009 | |
a50e233c ED |
6010 | skb_gro_reset_offset(skb); |
6011 | ||
6570bc79 | 6012 | ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); |
b0e3f1bd GB |
6013 | trace_napi_gro_receive_exit(ret); |
6014 | ||
6015 | return ret; | |
d565b0a1 HX |
6016 | } |
6017 | EXPORT_SYMBOL(napi_gro_receive); | |
6018 | ||
d0c2b0d2 | 6019 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 6020 | { |
93a35f59 ED |
6021 | if (unlikely(skb->pfmemalloc)) { |
6022 | consume_skb(skb); | |
6023 | return; | |
6024 | } | |
96e93eab | 6025 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
6026 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
6027 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
b1817524 | 6028 | __vlan_hwaccel_clear_tag(skb); |
66c46d74 | 6029 | skb->dev = napi->dev; |
6d152e23 | 6030 | skb->skb_iif = 0; |
33d9a2c7 ED |
6031 | |
6032 | /* eth_type_trans() assumes pkt_type is PACKET_HOST */ | |
6033 | skb->pkt_type = PACKET_HOST; | |
6034 | ||
c3caf119 JC |
6035 | skb->encapsulation = 0; |
6036 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 6037 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
174e2381 | 6038 | skb_ext_reset(skb); |
96e93eab HX |
6039 | |
6040 | napi->skb = skb; | |
6041 | } | |
96e93eab | 6042 | |
76620aaf | 6043 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 6044 | { |
5d38a079 | 6045 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
6046 | |
6047 | if (!skb) { | |
fd11a83d | 6048 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
6049 | if (skb) { |
6050 | napi->skb = skb; | |
6051 | skb_mark_napi_id(skb, napi); | |
6052 | } | |
80595d59 | 6053 | } |
96e93eab HX |
6054 | return skb; |
6055 | } | |
76620aaf | 6056 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 6057 | |
a50e233c ED |
6058 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
6059 | struct sk_buff *skb, | |
6060 | gro_result_t ret) | |
96e93eab | 6061 | { |
5d0d9be8 HX |
6062 | switch (ret) { |
6063 | case GRO_NORMAL: | |
a50e233c ED |
6064 | case GRO_HELD: |
6065 | __skb_push(skb, ETH_HLEN); | |
6066 | skb->protocol = eth_type_trans(skb, skb->dev); | |
323ebb61 EC |
6067 | if (ret == GRO_NORMAL) |
6068 | gro_normal_one(napi, skb); | |
86911732 | 6069 | break; |
5d38a079 | 6070 | |
5d0d9be8 | 6071 | case GRO_DROP: |
5d0d9be8 HX |
6072 | napi_reuse_skb(napi, skb); |
6073 | break; | |
5b252f0c | 6074 | |
e44699d2 MK |
6075 | case GRO_MERGED_FREE: |
6076 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) | |
6077 | napi_skb_free_stolen_head(skb); | |
6078 | else | |
6079 | napi_reuse_skb(napi, skb); | |
6080 | break; | |
6081 | ||
5b252f0c | 6082 | case GRO_MERGED: |
25393d3f | 6083 | case GRO_CONSUMED: |
5b252f0c | 6084 | break; |
5d0d9be8 | 6085 | } |
5d38a079 | 6086 | |
c7c4b3b6 | 6087 | return ret; |
5d38a079 | 6088 | } |
5d0d9be8 | 6089 | |
a50e233c ED |
6090 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
6091 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
6092 | * We copy ethernet header into skb->data to have a common layout. | |
6093 | */ | |
4adb9c4a | 6094 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
6095 | { |
6096 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
6097 | const struct ethhdr *eth; |
6098 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
6099 | |
6100 | napi->skb = NULL; | |
6101 | ||
a50e233c ED |
6102 | skb_reset_mac_header(skb); |
6103 | skb_gro_reset_offset(skb); | |
6104 | ||
a50e233c ED |
6105 | if (unlikely(skb_gro_header_hard(skb, hlen))) { |
6106 | eth = skb_gro_header_slow(skb, hlen, 0); | |
6107 | if (unlikely(!eth)) { | |
4da46ceb AC |
6108 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
6109 | __func__, napi->dev->name); | |
a50e233c ED |
6110 | napi_reuse_skb(napi, skb); |
6111 | return NULL; | |
6112 | } | |
6113 | } else { | |
a4270d67 | 6114 | eth = (const struct ethhdr *)skb->data; |
a50e233c ED |
6115 | gro_pull_from_frag0(skb, hlen); |
6116 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
6117 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 6118 | } |
a50e233c ED |
6119 | __skb_pull(skb, hlen); |
6120 | ||
6121 | /* | |
6122 | * This works because the only protocols we care about don't require | |
6123 | * special handling. | |
6124 | * We'll fix it up properly in napi_frags_finish() | |
6125 | */ | |
6126 | skb->protocol = eth->h_proto; | |
76620aaf | 6127 | |
76620aaf HX |
6128 | return skb; |
6129 | } | |
76620aaf | 6130 | |
c7c4b3b6 | 6131 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 6132 | { |
b0e3f1bd | 6133 | gro_result_t ret; |
76620aaf | 6134 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
6135 | |
6136 | if (!skb) | |
c7c4b3b6 | 6137 | return GRO_DROP; |
5d0d9be8 | 6138 | |
ae78dbfa BH |
6139 | trace_napi_gro_frags_entry(skb); |
6140 | ||
b0e3f1bd GB |
6141 | ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
6142 | trace_napi_gro_frags_exit(ret); | |
6143 | ||
6144 | return ret; | |
5d0d9be8 | 6145 | } |
5d38a079 HX |
6146 | EXPORT_SYMBOL(napi_gro_frags); |
6147 | ||
573e8fca TH |
6148 | /* Compute the checksum from gro_offset and return the folded value |
6149 | * after adding in any pseudo checksum. | |
6150 | */ | |
6151 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
6152 | { | |
6153 | __wsum wsum; | |
6154 | __sum16 sum; | |
6155 | ||
6156 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
6157 | ||
6158 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
6159 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
14641931 | 6160 | /* See comments in __skb_checksum_complete(). */ |
573e8fca TH |
6161 | if (likely(!sum)) { |
6162 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
6163 | !skb->csum_complete_sw) | |
7fe50ac8 | 6164 | netdev_rx_csum_fault(skb->dev, skb); |
573e8fca TH |
6165 | } |
6166 | ||
6167 | NAPI_GRO_CB(skb)->csum = wsum; | |
6168 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
6169 | ||
6170 | return sum; | |
6171 | } | |
6172 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
6173 | ||
773fc8f6 | 6174 | static void net_rps_send_ipi(struct softnet_data *remsd) |
6175 | { | |
6176 | #ifdef CONFIG_RPS | |
6177 | while (remsd) { | |
6178 | struct softnet_data *next = remsd->rps_ipi_next; | |
6179 | ||
6180 | if (cpu_online(remsd->cpu)) | |
6181 | smp_call_function_single_async(remsd->cpu, &remsd->csd); | |
6182 | remsd = next; | |
6183 | } | |
6184 | #endif | |
6185 | } | |
6186 | ||
e326bed2 | 6187 | /* |
855abcf0 | 6188 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
6189 | * Note: called with local irq disabled, but exits with local irq enabled. |
6190 | */ | |
6191 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
6192 | { | |
6193 | #ifdef CONFIG_RPS | |
6194 | struct softnet_data *remsd = sd->rps_ipi_list; | |
6195 | ||
6196 | if (remsd) { | |
6197 | sd->rps_ipi_list = NULL; | |
6198 | ||
6199 | local_irq_enable(); | |
6200 | ||
6201 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
773fc8f6 | 6202 | net_rps_send_ipi(remsd); |
e326bed2 ED |
6203 | } else |
6204 | #endif | |
6205 | local_irq_enable(); | |
6206 | } | |
6207 | ||
d75b1ade ED |
6208 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
6209 | { | |
6210 | #ifdef CONFIG_RPS | |
6211 | return sd->rps_ipi_list != NULL; | |
6212 | #else | |
6213 | return false; | |
6214 | #endif | |
6215 | } | |
6216 | ||
bea3348e | 6217 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 6218 | { |
eecfd7c4 | 6219 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
6220 | bool again = true; |
6221 | int work = 0; | |
1da177e4 | 6222 | |
e326bed2 ED |
6223 | /* Check if we have pending ipi, its better to send them now, |
6224 | * not waiting net_rx_action() end. | |
6225 | */ | |
d75b1ade | 6226 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
6227 | local_irq_disable(); |
6228 | net_rps_action_and_irq_enable(sd); | |
6229 | } | |
d75b1ade | 6230 | |
3d48b53f | 6231 | napi->weight = dev_rx_weight; |
145dd5f9 | 6232 | while (again) { |
1da177e4 | 6233 | struct sk_buff *skb; |
6e7676c1 CG |
6234 | |
6235 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 6236 | rcu_read_lock(); |
6e7676c1 | 6237 | __netif_receive_skb(skb); |
2c17d27c | 6238 | rcu_read_unlock(); |
76cc8b13 | 6239 | input_queue_head_incr(sd); |
145dd5f9 | 6240 | if (++work >= quota) |
76cc8b13 | 6241 | return work; |
145dd5f9 | 6242 | |
6e7676c1 | 6243 | } |
1da177e4 | 6244 | |
145dd5f9 | 6245 | local_irq_disable(); |
e36fa2f7 | 6246 | rps_lock(sd); |
11ef7a89 | 6247 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
6248 | /* |
6249 | * Inline a custom version of __napi_complete(). | |
6250 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
6251 | * and NAPI_STATE_SCHED is the only possible flag set |
6252 | * on backlog. | |
6253 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
6254 | * and we dont need an smp_mb() memory barrier. |
6255 | */ | |
eecfd7c4 | 6256 | napi->state = 0; |
145dd5f9 PA |
6257 | again = false; |
6258 | } else { | |
6259 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
6260 | &sd->process_queue); | |
bea3348e | 6261 | } |
e36fa2f7 | 6262 | rps_unlock(sd); |
145dd5f9 | 6263 | local_irq_enable(); |
6e7676c1 | 6264 | } |
1da177e4 | 6265 | |
bea3348e SH |
6266 | return work; |
6267 | } | |
1da177e4 | 6268 | |
bea3348e SH |
6269 | /** |
6270 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 6271 | * @n: entry to schedule |
bea3348e | 6272 | * |
bc9ad166 ED |
6273 | * The entry's receive function will be scheduled to run. |
6274 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 6275 | */ |
b5606c2d | 6276 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
6277 | { |
6278 | unsigned long flags; | |
1da177e4 | 6279 | |
bea3348e | 6280 | local_irq_save(flags); |
903ceff7 | 6281 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 6282 | local_irq_restore(flags); |
1da177e4 | 6283 | } |
bea3348e SH |
6284 | EXPORT_SYMBOL(__napi_schedule); |
6285 | ||
39e6c820 ED |
6286 | /** |
6287 | * napi_schedule_prep - check if napi can be scheduled | |
6288 | * @n: napi context | |
6289 | * | |
6290 | * Test if NAPI routine is already running, and if not mark | |
6291 | * it as running. This is used as a condition variable | |
6292 | * insure only one NAPI poll instance runs. We also make | |
6293 | * sure there is no pending NAPI disable. | |
6294 | */ | |
6295 | bool napi_schedule_prep(struct napi_struct *n) | |
6296 | { | |
6297 | unsigned long val, new; | |
6298 | ||
6299 | do { | |
6300 | val = READ_ONCE(n->state); | |
6301 | if (unlikely(val & NAPIF_STATE_DISABLE)) | |
6302 | return false; | |
6303 | new = val | NAPIF_STATE_SCHED; | |
6304 | ||
6305 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
6306 | * This was suggested by Alexander Duyck, as compiler | |
6307 | * emits better code than : | |
6308 | * if (val & NAPIF_STATE_SCHED) | |
6309 | * new |= NAPIF_STATE_MISSED; | |
6310 | */ | |
6311 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
6312 | NAPIF_STATE_MISSED; | |
6313 | } while (cmpxchg(&n->state, val, new) != val); | |
6314 | ||
6315 | return !(val & NAPIF_STATE_SCHED); | |
6316 | } | |
6317 | EXPORT_SYMBOL(napi_schedule_prep); | |
6318 | ||
bc9ad166 ED |
6319 | /** |
6320 | * __napi_schedule_irqoff - schedule for receive | |
6321 | * @n: entry to schedule | |
6322 | * | |
6323 | * Variant of __napi_schedule() assuming hard irqs are masked | |
6324 | */ | |
6325 | void __napi_schedule_irqoff(struct napi_struct *n) | |
6326 | { | |
6327 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
6328 | } | |
6329 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
6330 | ||
364b6055 | 6331 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 6332 | { |
6f8b12d6 ED |
6333 | unsigned long flags, val, new, timeout = 0; |
6334 | bool ret = true; | |
d565b0a1 HX |
6335 | |
6336 | /* | |
217f6974 ED |
6337 | * 1) Don't let napi dequeue from the cpu poll list |
6338 | * just in case its running on a different cpu. | |
6339 | * 2) If we are busy polling, do nothing here, we have | |
6340 | * the guarantee we will be called later. | |
d565b0a1 | 6341 | */ |
217f6974 ED |
6342 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
6343 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 6344 | return false; |
d565b0a1 | 6345 | |
6f8b12d6 ED |
6346 | if (work_done) { |
6347 | if (n->gro_bitmask) | |
7e417a66 ED |
6348 | timeout = READ_ONCE(n->dev->gro_flush_timeout); |
6349 | n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); | |
6f8b12d6 ED |
6350 | } |
6351 | if (n->defer_hard_irqs_count > 0) { | |
6352 | n->defer_hard_irqs_count--; | |
7e417a66 | 6353 | timeout = READ_ONCE(n->dev->gro_flush_timeout); |
6f8b12d6 ED |
6354 | if (timeout) |
6355 | ret = false; | |
6356 | } | |
6357 | if (n->gro_bitmask) { | |
605108ac PA |
6358 | /* When the NAPI instance uses a timeout and keeps postponing |
6359 | * it, we need to bound somehow the time packets are kept in | |
6360 | * the GRO layer | |
6361 | */ | |
6362 | napi_gro_flush(n, !!timeout); | |
3b47d303 | 6363 | } |
c8079432 MM |
6364 | |
6365 | gro_normal_list(n); | |
6366 | ||
02c1602e | 6367 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
6368 | /* If n->poll_list is not empty, we need to mask irqs */ |
6369 | local_irq_save(flags); | |
02c1602e | 6370 | list_del_init(&n->poll_list); |
d75b1ade ED |
6371 | local_irq_restore(flags); |
6372 | } | |
39e6c820 ED |
6373 | |
6374 | do { | |
6375 | val = READ_ONCE(n->state); | |
6376 | ||
6377 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); | |
6378 | ||
6379 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); | |
6380 | ||
6381 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
6382 | * because we will call napi->poll() one more time. | |
6383 | * This C code was suggested by Alexander Duyck to help gcc. | |
6384 | */ | |
6385 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
6386 | NAPIF_STATE_SCHED; | |
6387 | } while (cmpxchg(&n->state, val, new) != val); | |
6388 | ||
6389 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
6390 | __napi_schedule(n); | |
6391 | return false; | |
6392 | } | |
6393 | ||
6f8b12d6 ED |
6394 | if (timeout) |
6395 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
6396 | HRTIMER_MODE_REL_PINNED); | |
6397 | return ret; | |
d565b0a1 | 6398 | } |
3b47d303 | 6399 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 6400 | |
af12fa6e | 6401 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 6402 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
6403 | { |
6404 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
6405 | struct napi_struct *napi; | |
6406 | ||
6407 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
6408 | if (napi->napi_id == napi_id) | |
6409 | return napi; | |
6410 | ||
6411 | return NULL; | |
6412 | } | |
02d62e86 ED |
6413 | |
6414 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
217f6974 | 6415 | |
ce6aea93 | 6416 | #define BUSY_POLL_BUDGET 8 |
217f6974 ED |
6417 | |
6418 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) | |
6419 | { | |
6420 | int rc; | |
6421 | ||
39e6c820 ED |
6422 | /* Busy polling means there is a high chance device driver hard irq |
6423 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
6424 | * set in napi_schedule_prep(). | |
6425 | * Since we are about to call napi->poll() once more, we can safely | |
6426 | * clear NAPI_STATE_MISSED. | |
6427 | * | |
6428 | * Note: x86 could use a single "lock and ..." instruction | |
6429 | * to perform these two clear_bit() | |
6430 | */ | |
6431 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
6432 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
6433 | ||
6434 | local_bh_disable(); | |
6435 | ||
6436 | /* All we really want here is to re-enable device interrupts. | |
6437 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
6438 | */ | |
6439 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
323ebb61 EC |
6440 | /* We can't gro_normal_list() here, because napi->poll() might have |
6441 | * rearmed the napi (napi_complete_done()) in which case it could | |
6442 | * already be running on another CPU. | |
6443 | */ | |
1e22391e | 6444 | trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); |
217f6974 | 6445 | netpoll_poll_unlock(have_poll_lock); |
323ebb61 EC |
6446 | if (rc == BUSY_POLL_BUDGET) { |
6447 | /* As the whole budget was spent, we still own the napi so can | |
6448 | * safely handle the rx_list. | |
6449 | */ | |
6450 | gro_normal_list(napi); | |
217f6974 | 6451 | __napi_schedule(napi); |
323ebb61 | 6452 | } |
217f6974 | 6453 | local_bh_enable(); |
217f6974 ED |
6454 | } |
6455 | ||
7db6b048 SS |
6456 | void napi_busy_loop(unsigned int napi_id, |
6457 | bool (*loop_end)(void *, unsigned long), | |
6458 | void *loop_end_arg) | |
02d62e86 | 6459 | { |
7db6b048 | 6460 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 6461 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 6462 | void *have_poll_lock = NULL; |
02d62e86 | 6463 | struct napi_struct *napi; |
217f6974 ED |
6464 | |
6465 | restart: | |
217f6974 | 6466 | napi_poll = NULL; |
02d62e86 | 6467 | |
2a028ecb | 6468 | rcu_read_lock(); |
02d62e86 | 6469 | |
545cd5e5 | 6470 | napi = napi_by_id(napi_id); |
02d62e86 ED |
6471 | if (!napi) |
6472 | goto out; | |
6473 | ||
217f6974 ED |
6474 | preempt_disable(); |
6475 | for (;;) { | |
2b5cd0df AD |
6476 | int work = 0; |
6477 | ||
2a028ecb | 6478 | local_bh_disable(); |
217f6974 ED |
6479 | if (!napi_poll) { |
6480 | unsigned long val = READ_ONCE(napi->state); | |
6481 | ||
6482 | /* If multiple threads are competing for this napi, | |
6483 | * we avoid dirtying napi->state as much as we can. | |
6484 | */ | |
6485 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
6486 | NAPIF_STATE_IN_BUSY_POLL)) | |
6487 | goto count; | |
6488 | if (cmpxchg(&napi->state, val, | |
6489 | val | NAPIF_STATE_IN_BUSY_POLL | | |
6490 | NAPIF_STATE_SCHED) != val) | |
6491 | goto count; | |
6492 | have_poll_lock = netpoll_poll_lock(napi); | |
6493 | napi_poll = napi->poll; | |
6494 | } | |
2b5cd0df AD |
6495 | work = napi_poll(napi, BUSY_POLL_BUDGET); |
6496 | trace_napi_poll(napi, work, BUSY_POLL_BUDGET); | |
323ebb61 | 6497 | gro_normal_list(napi); |
217f6974 | 6498 | count: |
2b5cd0df | 6499 | if (work > 0) |
7db6b048 | 6500 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 6501 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
2a028ecb | 6502 | local_bh_enable(); |
02d62e86 | 6503 | |
7db6b048 | 6504 | if (!loop_end || loop_end(loop_end_arg, start_time)) |
217f6974 | 6505 | break; |
02d62e86 | 6506 | |
217f6974 ED |
6507 | if (unlikely(need_resched())) { |
6508 | if (napi_poll) | |
6509 | busy_poll_stop(napi, have_poll_lock); | |
6510 | preempt_enable(); | |
6511 | rcu_read_unlock(); | |
6512 | cond_resched(); | |
7db6b048 | 6513 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 6514 | return; |
217f6974 ED |
6515 | goto restart; |
6516 | } | |
6cdf89b1 | 6517 | cpu_relax(); |
217f6974 ED |
6518 | } |
6519 | if (napi_poll) | |
6520 | busy_poll_stop(napi, have_poll_lock); | |
6521 | preempt_enable(); | |
02d62e86 | 6522 | out: |
2a028ecb | 6523 | rcu_read_unlock(); |
02d62e86 | 6524 | } |
7db6b048 | 6525 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
6526 | |
6527 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 6528 | |
149d6ad8 | 6529 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 6530 | { |
d64b5e85 ED |
6531 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
6532 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 6533 | return; |
af12fa6e | 6534 | |
52bd2d62 | 6535 | spin_lock(&napi_hash_lock); |
af12fa6e | 6536 | |
545cd5e5 | 6537 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 6538 | do { |
545cd5e5 AD |
6539 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
6540 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
6541 | } while (napi_by_id(napi_gen_id)); |
6542 | napi->napi_id = napi_gen_id; | |
af12fa6e | 6543 | |
52bd2d62 ED |
6544 | hlist_add_head_rcu(&napi->napi_hash_node, |
6545 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 6546 | |
52bd2d62 | 6547 | spin_unlock(&napi_hash_lock); |
af12fa6e | 6548 | } |
af12fa6e ET |
6549 | |
6550 | /* Warning : caller is responsible to make sure rcu grace period | |
6551 | * is respected before freeing memory containing @napi | |
6552 | */ | |
34cbe27e | 6553 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 6554 | { |
34cbe27e ED |
6555 | bool rcu_sync_needed = false; |
6556 | ||
af12fa6e ET |
6557 | spin_lock(&napi_hash_lock); |
6558 | ||
34cbe27e ED |
6559 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
6560 | rcu_sync_needed = true; | |
af12fa6e | 6561 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 6562 | } |
af12fa6e | 6563 | spin_unlock(&napi_hash_lock); |
34cbe27e | 6564 | return rcu_sync_needed; |
af12fa6e ET |
6565 | } |
6566 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
6567 | ||
3b47d303 ED |
6568 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
6569 | { | |
6570 | struct napi_struct *napi; | |
6571 | ||
6572 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
6573 | |
6574 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
6575 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
6576 | */ | |
6f8b12d6 | 6577 | if (!napi_disable_pending(napi) && |
39e6c820 ED |
6578 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) |
6579 | __napi_schedule_irqoff(napi); | |
3b47d303 ED |
6580 | |
6581 | return HRTIMER_NORESTART; | |
6582 | } | |
6583 | ||
7c4ec749 | 6584 | static void init_gro_hash(struct napi_struct *napi) |
d565b0a1 | 6585 | { |
07d78363 DM |
6586 | int i; |
6587 | ||
6312fe77 LR |
6588 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6589 | INIT_LIST_HEAD(&napi->gro_hash[i].list); | |
6590 | napi->gro_hash[i].count = 0; | |
6591 | } | |
7c4ec749 DM |
6592 | napi->gro_bitmask = 0; |
6593 | } | |
6594 | ||
6595 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
6596 | int (*poll)(struct napi_struct *, int), int weight) | |
6597 | { | |
6598 | INIT_LIST_HEAD(&napi->poll_list); | |
6599 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); | |
6600 | napi->timer.function = napi_watchdog; | |
6601 | init_gro_hash(napi); | |
5d38a079 | 6602 | napi->skb = NULL; |
323ebb61 EC |
6603 | INIT_LIST_HEAD(&napi->rx_list); |
6604 | napi->rx_count = 0; | |
d565b0a1 | 6605 | napi->poll = poll; |
82dc3c63 | 6606 | if (weight > NAPI_POLL_WEIGHT) |
bf29e9e9 QC |
6607 | netdev_err_once(dev, "%s() called with weight %d\n", __func__, |
6608 | weight); | |
d565b0a1 HX |
6609 | napi->weight = weight; |
6610 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 6611 | napi->dev = dev; |
5d38a079 | 6612 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
6613 | napi->poll_owner = -1; |
6614 | #endif | |
6615 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 6616 | napi_hash_add(napi); |
d565b0a1 HX |
6617 | } |
6618 | EXPORT_SYMBOL(netif_napi_add); | |
6619 | ||
3b47d303 ED |
6620 | void napi_disable(struct napi_struct *n) |
6621 | { | |
6622 | might_sleep(); | |
6623 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
6624 | ||
6625 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
6626 | msleep(1); | |
2d8bff12 NH |
6627 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
6628 | msleep(1); | |
3b47d303 ED |
6629 | |
6630 | hrtimer_cancel(&n->timer); | |
6631 | ||
6632 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
6633 | } | |
6634 | EXPORT_SYMBOL(napi_disable); | |
6635 | ||
07d78363 | 6636 | static void flush_gro_hash(struct napi_struct *napi) |
d4546c25 | 6637 | { |
07d78363 | 6638 | int i; |
d4546c25 | 6639 | |
07d78363 DM |
6640 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6641 | struct sk_buff *skb, *n; | |
6642 | ||
6312fe77 | 6643 | list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) |
07d78363 | 6644 | kfree_skb(skb); |
6312fe77 | 6645 | napi->gro_hash[i].count = 0; |
07d78363 | 6646 | } |
d4546c25 DM |
6647 | } |
6648 | ||
93d05d4a | 6649 | /* Must be called in process context */ |
d565b0a1 HX |
6650 | void netif_napi_del(struct napi_struct *napi) |
6651 | { | |
93d05d4a ED |
6652 | might_sleep(); |
6653 | if (napi_hash_del(napi)) | |
6654 | synchronize_net(); | |
d7b06636 | 6655 | list_del_init(&napi->dev_list); |
76620aaf | 6656 | napi_free_frags(napi); |
d565b0a1 | 6657 | |
07d78363 | 6658 | flush_gro_hash(napi); |
d9f37d01 | 6659 | napi->gro_bitmask = 0; |
d565b0a1 HX |
6660 | } |
6661 | EXPORT_SYMBOL(netif_napi_del); | |
6662 | ||
726ce70e HX |
6663 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
6664 | { | |
6665 | void *have; | |
6666 | int work, weight; | |
6667 | ||
6668 | list_del_init(&n->poll_list); | |
6669 | ||
6670 | have = netpoll_poll_lock(n); | |
6671 | ||
6672 | weight = n->weight; | |
6673 | ||
6674 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
6675 | * with netpoll's poll_napi(). Only the entity which | |
6676 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
6677 | * actually make the ->poll() call. Therefore we avoid | |
6678 | * accidentally calling ->poll() when NAPI is not scheduled. | |
6679 | */ | |
6680 | work = 0; | |
6681 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
6682 | work = n->poll(n, weight); | |
1db19db7 | 6683 | trace_napi_poll(n, work, weight); |
726ce70e HX |
6684 | } |
6685 | ||
6686 | WARN_ON_ONCE(work > weight); | |
6687 | ||
6688 | if (likely(work < weight)) | |
6689 | goto out_unlock; | |
6690 | ||
6691 | /* Drivers must not modify the NAPI state if they | |
6692 | * consume the entire weight. In such cases this code | |
6693 | * still "owns" the NAPI instance and therefore can | |
6694 | * move the instance around on the list at-will. | |
6695 | */ | |
6696 | if (unlikely(napi_disable_pending(n))) { | |
6697 | napi_complete(n); | |
6698 | goto out_unlock; | |
6699 | } | |
6700 | ||
d9f37d01 | 6701 | if (n->gro_bitmask) { |
726ce70e HX |
6702 | /* flush too old packets |
6703 | * If HZ < 1000, flush all packets. | |
6704 | */ | |
6705 | napi_gro_flush(n, HZ >= 1000); | |
6706 | } | |
6707 | ||
c8079432 MM |
6708 | gro_normal_list(n); |
6709 | ||
001ce546 HX |
6710 | /* Some drivers may have called napi_schedule |
6711 | * prior to exhausting their budget. | |
6712 | */ | |
6713 | if (unlikely(!list_empty(&n->poll_list))) { | |
6714 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
6715 | n->dev ? n->dev->name : "backlog"); | |
6716 | goto out_unlock; | |
6717 | } | |
6718 | ||
726ce70e HX |
6719 | list_add_tail(&n->poll_list, repoll); |
6720 | ||
6721 | out_unlock: | |
6722 | netpoll_poll_unlock(have); | |
6723 | ||
6724 | return work; | |
6725 | } | |
6726 | ||
0766f788 | 6727 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 6728 | { |
903ceff7 | 6729 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e MW |
6730 | unsigned long time_limit = jiffies + |
6731 | usecs_to_jiffies(netdev_budget_usecs); | |
51b0bded | 6732 | int budget = netdev_budget; |
d75b1ade ED |
6733 | LIST_HEAD(list); |
6734 | LIST_HEAD(repoll); | |
53fb95d3 | 6735 | |
1da177e4 | 6736 | local_irq_disable(); |
d75b1ade ED |
6737 | list_splice_init(&sd->poll_list, &list); |
6738 | local_irq_enable(); | |
1da177e4 | 6739 | |
ceb8d5bf | 6740 | for (;;) { |
bea3348e | 6741 | struct napi_struct *n; |
1da177e4 | 6742 | |
ceb8d5bf HX |
6743 | if (list_empty(&list)) { |
6744 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
f52dffe0 | 6745 | goto out; |
ceb8d5bf HX |
6746 | break; |
6747 | } | |
6748 | ||
6bd373eb HX |
6749 | n = list_first_entry(&list, struct napi_struct, poll_list); |
6750 | budget -= napi_poll(n, &repoll); | |
6751 | ||
d75b1ade | 6752 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
6753 | * Allow this to run for 2 jiffies since which will allow |
6754 | * an average latency of 1.5/HZ. | |
bea3348e | 6755 | */ |
ceb8d5bf HX |
6756 | if (unlikely(budget <= 0 || |
6757 | time_after_eq(jiffies, time_limit))) { | |
6758 | sd->time_squeeze++; | |
6759 | break; | |
6760 | } | |
1da177e4 | 6761 | } |
d75b1ade | 6762 | |
d75b1ade ED |
6763 | local_irq_disable(); |
6764 | ||
6765 | list_splice_tail_init(&sd->poll_list, &list); | |
6766 | list_splice_tail(&repoll, &list); | |
6767 | list_splice(&list, &sd->poll_list); | |
6768 | if (!list_empty(&sd->poll_list)) | |
6769 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6770 | ||
e326bed2 | 6771 | net_rps_action_and_irq_enable(sd); |
f52dffe0 ED |
6772 | out: |
6773 | __kfree_skb_flush(); | |
1da177e4 LT |
6774 | } |
6775 | ||
aa9d8560 | 6776 | struct netdev_adjacent { |
9ff162a8 | 6777 | struct net_device *dev; |
5d261913 VF |
6778 | |
6779 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 6780 | bool master; |
5d261913 | 6781 | |
32b6d34f TY |
6782 | /* lookup ignore flag */ |
6783 | bool ignore; | |
6784 | ||
5d261913 VF |
6785 | /* counter for the number of times this device was added to us */ |
6786 | u16 ref_nr; | |
6787 | ||
402dae96 VF |
6788 | /* private field for the users */ |
6789 | void *private; | |
6790 | ||
9ff162a8 JP |
6791 | struct list_head list; |
6792 | struct rcu_head rcu; | |
9ff162a8 JP |
6793 | }; |
6794 | ||
6ea29da1 | 6795 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 6796 | struct list_head *adj_list) |
9ff162a8 | 6797 | { |
5d261913 | 6798 | struct netdev_adjacent *adj; |
5d261913 | 6799 | |
2f268f12 | 6800 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
6801 | if (adj->dev == adj_dev) |
6802 | return adj; | |
9ff162a8 JP |
6803 | } |
6804 | return NULL; | |
6805 | } | |
6806 | ||
32b6d34f | 6807 | static int ____netdev_has_upper_dev(struct net_device *upper_dev, void *data) |
f1170fd4 DA |
6808 | { |
6809 | struct net_device *dev = data; | |
6810 | ||
6811 | return upper_dev == dev; | |
6812 | } | |
6813 | ||
9ff162a8 JP |
6814 | /** |
6815 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
6816 | * @dev: device | |
6817 | * @upper_dev: upper device to check | |
6818 | * | |
6819 | * Find out if a device is linked to specified upper device and return true | |
6820 | * in case it is. Note that this checks only immediate upper device, | |
6821 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
6822 | */ | |
6823 | bool netdev_has_upper_dev(struct net_device *dev, | |
6824 | struct net_device *upper_dev) | |
6825 | { | |
6826 | ASSERT_RTNL(); | |
6827 | ||
32b6d34f | 6828 | return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, |
f1170fd4 | 6829 | upper_dev); |
9ff162a8 JP |
6830 | } |
6831 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
6832 | ||
1a3f060c DA |
6833 | /** |
6834 | * netdev_has_upper_dev_all - Check if device is linked to an upper device | |
6835 | * @dev: device | |
6836 | * @upper_dev: upper device to check | |
6837 | * | |
6838 | * Find out if a device is linked to specified upper device and return true | |
6839 | * in case it is. Note that this checks the entire upper device chain. | |
6840 | * The caller must hold rcu lock. | |
6841 | */ | |
6842 | ||
1a3f060c DA |
6843 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
6844 | struct net_device *upper_dev) | |
6845 | { | |
32b6d34f | 6846 | return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, |
1a3f060c DA |
6847 | upper_dev); |
6848 | } | |
6849 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
6850 | ||
9ff162a8 JP |
6851 | /** |
6852 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
6853 | * @dev: device | |
6854 | * | |
6855 | * Find out if a device is linked to an upper device and return true in case | |
6856 | * it is. The caller must hold the RTNL lock. | |
6857 | */ | |
25cc72a3 | 6858 | bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
6859 | { |
6860 | ASSERT_RTNL(); | |
6861 | ||
f1170fd4 | 6862 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 6863 | } |
25cc72a3 | 6864 | EXPORT_SYMBOL(netdev_has_any_upper_dev); |
9ff162a8 JP |
6865 | |
6866 | /** | |
6867 | * netdev_master_upper_dev_get - Get master upper device | |
6868 | * @dev: device | |
6869 | * | |
6870 | * Find a master upper device and return pointer to it or NULL in case | |
6871 | * it's not there. The caller must hold the RTNL lock. | |
6872 | */ | |
6873 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
6874 | { | |
aa9d8560 | 6875 | struct netdev_adjacent *upper; |
9ff162a8 JP |
6876 | |
6877 | ASSERT_RTNL(); | |
6878 | ||
2f268f12 | 6879 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
6880 | return NULL; |
6881 | ||
2f268f12 | 6882 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 6883 | struct netdev_adjacent, list); |
9ff162a8 JP |
6884 | if (likely(upper->master)) |
6885 | return upper->dev; | |
6886 | return NULL; | |
6887 | } | |
6888 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
6889 | ||
32b6d34f TY |
6890 | static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) |
6891 | { | |
6892 | struct netdev_adjacent *upper; | |
6893 | ||
6894 | ASSERT_RTNL(); | |
6895 | ||
6896 | if (list_empty(&dev->adj_list.upper)) | |
6897 | return NULL; | |
6898 | ||
6899 | upper = list_first_entry(&dev->adj_list.upper, | |
6900 | struct netdev_adjacent, list); | |
6901 | if (likely(upper->master) && !upper->ignore) | |
6902 | return upper->dev; | |
6903 | return NULL; | |
6904 | } | |
6905 | ||
0f524a80 DA |
6906 | /** |
6907 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
6908 | * @dev: device | |
6909 | * | |
6910 | * Find out if a device is linked to a lower device and return true in case | |
6911 | * it is. The caller must hold the RTNL lock. | |
6912 | */ | |
6913 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
6914 | { | |
6915 | ASSERT_RTNL(); | |
6916 | ||
6917 | return !list_empty(&dev->adj_list.lower); | |
6918 | } | |
6919 | ||
b6ccba4c VF |
6920 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
6921 | { | |
6922 | struct netdev_adjacent *adj; | |
6923 | ||
6924 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
6925 | ||
6926 | return adj->private; | |
6927 | } | |
6928 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
6929 | ||
44a40855 VY |
6930 | /** |
6931 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
6932 | * @dev: device | |
6933 | * @iter: list_head ** of the current position | |
6934 | * | |
6935 | * Gets the next device from the dev's upper list, starting from iter | |
6936 | * position. The caller must hold RCU read lock. | |
6937 | */ | |
6938 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
6939 | struct list_head **iter) | |
6940 | { | |
6941 | struct netdev_adjacent *upper; | |
6942 | ||
6943 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
6944 | ||
6945 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6946 | ||
6947 | if (&upper->list == &dev->adj_list.upper) | |
6948 | return NULL; | |
6949 | ||
6950 | *iter = &upper->list; | |
6951 | ||
6952 | return upper->dev; | |
6953 | } | |
6954 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
6955 | ||
32b6d34f TY |
6956 | static struct net_device *__netdev_next_upper_dev(struct net_device *dev, |
6957 | struct list_head **iter, | |
6958 | bool *ignore) | |
5343da4c TY |
6959 | { |
6960 | struct netdev_adjacent *upper; | |
6961 | ||
6962 | upper = list_entry((*iter)->next, struct netdev_adjacent, list); | |
6963 | ||
6964 | if (&upper->list == &dev->adj_list.upper) | |
6965 | return NULL; | |
6966 | ||
6967 | *iter = &upper->list; | |
32b6d34f | 6968 | *ignore = upper->ignore; |
5343da4c TY |
6969 | |
6970 | return upper->dev; | |
6971 | } | |
6972 | ||
1a3f060c DA |
6973 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
6974 | struct list_head **iter) | |
6975 | { | |
6976 | struct netdev_adjacent *upper; | |
6977 | ||
6978 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
6979 | ||
6980 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6981 | ||
6982 | if (&upper->list == &dev->adj_list.upper) | |
6983 | return NULL; | |
6984 | ||
6985 | *iter = &upper->list; | |
6986 | ||
6987 | return upper->dev; | |
6988 | } | |
6989 | ||
32b6d34f TY |
6990 | static int __netdev_walk_all_upper_dev(struct net_device *dev, |
6991 | int (*fn)(struct net_device *dev, | |
6992 | void *data), | |
6993 | void *data) | |
5343da4c TY |
6994 | { |
6995 | struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
6996 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
6997 | int ret, cur = 0; | |
32b6d34f | 6998 | bool ignore; |
5343da4c TY |
6999 | |
7000 | now = dev; | |
7001 | iter = &dev->adj_list.upper; | |
7002 | ||
7003 | while (1) { | |
7004 | if (now != dev) { | |
7005 | ret = fn(now, data); | |
7006 | if (ret) | |
7007 | return ret; | |
7008 | } | |
7009 | ||
7010 | next = NULL; | |
7011 | while (1) { | |
32b6d34f | 7012 | udev = __netdev_next_upper_dev(now, &iter, &ignore); |
5343da4c TY |
7013 | if (!udev) |
7014 | break; | |
32b6d34f TY |
7015 | if (ignore) |
7016 | continue; | |
5343da4c TY |
7017 | |
7018 | next = udev; | |
7019 | niter = &udev->adj_list.upper; | |
7020 | dev_stack[cur] = now; | |
7021 | iter_stack[cur++] = iter; | |
7022 | break; | |
7023 | } | |
7024 | ||
7025 | if (!next) { | |
7026 | if (!cur) | |
7027 | return 0; | |
7028 | next = dev_stack[--cur]; | |
7029 | niter = iter_stack[cur]; | |
7030 | } | |
7031 | ||
7032 | now = next; | |
7033 | iter = niter; | |
7034 | } | |
7035 | ||
7036 | return 0; | |
7037 | } | |
7038 | ||
1a3f060c DA |
7039 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, |
7040 | int (*fn)(struct net_device *dev, | |
7041 | void *data), | |
7042 | void *data) | |
7043 | { | |
5343da4c TY |
7044 | struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; |
7045 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7046 | int ret, cur = 0; | |
1a3f060c | 7047 | |
5343da4c TY |
7048 | now = dev; |
7049 | iter = &dev->adj_list.upper; | |
1a3f060c | 7050 | |
5343da4c TY |
7051 | while (1) { |
7052 | if (now != dev) { | |
7053 | ret = fn(now, data); | |
7054 | if (ret) | |
7055 | return ret; | |
7056 | } | |
7057 | ||
7058 | next = NULL; | |
7059 | while (1) { | |
7060 | udev = netdev_next_upper_dev_rcu(now, &iter); | |
7061 | if (!udev) | |
7062 | break; | |
7063 | ||
7064 | next = udev; | |
7065 | niter = &udev->adj_list.upper; | |
7066 | dev_stack[cur] = now; | |
7067 | iter_stack[cur++] = iter; | |
7068 | break; | |
7069 | } | |
7070 | ||
7071 | if (!next) { | |
7072 | if (!cur) | |
7073 | return 0; | |
7074 | next = dev_stack[--cur]; | |
7075 | niter = iter_stack[cur]; | |
7076 | } | |
7077 | ||
7078 | now = next; | |
7079 | iter = niter; | |
1a3f060c DA |
7080 | } |
7081 | ||
7082 | return 0; | |
7083 | } | |
7084 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
7085 | ||
32b6d34f TY |
7086 | static bool __netdev_has_upper_dev(struct net_device *dev, |
7087 | struct net_device *upper_dev) | |
7088 | { | |
7089 | ASSERT_RTNL(); | |
7090 | ||
7091 | return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, | |
7092 | upper_dev); | |
7093 | } | |
7094 | ||
31088a11 VF |
7095 | /** |
7096 | * netdev_lower_get_next_private - Get the next ->private from the | |
7097 | * lower neighbour list | |
7098 | * @dev: device | |
7099 | * @iter: list_head ** of the current position | |
7100 | * | |
7101 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
7102 | * list, starting from iter position. The caller must hold either hold the | |
7103 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 7104 | * list will remain unchanged. |
31088a11 VF |
7105 | */ |
7106 | void *netdev_lower_get_next_private(struct net_device *dev, | |
7107 | struct list_head **iter) | |
7108 | { | |
7109 | struct netdev_adjacent *lower; | |
7110 | ||
7111 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
7112 | ||
7113 | if (&lower->list == &dev->adj_list.lower) | |
7114 | return NULL; | |
7115 | ||
6859e7df | 7116 | *iter = lower->list.next; |
31088a11 VF |
7117 | |
7118 | return lower->private; | |
7119 | } | |
7120 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
7121 | ||
7122 | /** | |
7123 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
7124 | * lower neighbour list, RCU | |
7125 | * variant | |
7126 | * @dev: device | |
7127 | * @iter: list_head ** of the current position | |
7128 | * | |
7129 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
7130 | * list, starting from iter position. The caller must hold RCU read lock. | |
7131 | */ | |
7132 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
7133 | struct list_head **iter) | |
7134 | { | |
7135 | struct netdev_adjacent *lower; | |
7136 | ||
7137 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
7138 | ||
7139 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7140 | ||
7141 | if (&lower->list == &dev->adj_list.lower) | |
7142 | return NULL; | |
7143 | ||
6859e7df | 7144 | *iter = &lower->list; |
31088a11 VF |
7145 | |
7146 | return lower->private; | |
7147 | } | |
7148 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
7149 | ||
4085ebe8 VY |
7150 | /** |
7151 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
7152 | * list | |
7153 | * @dev: device | |
7154 | * @iter: list_head ** of the current position | |
7155 | * | |
7156 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
7157 | * list, starting from iter position. The caller must hold RTNL lock or | |
7158 | * its own locking that guarantees that the neighbour lower | |
b469139e | 7159 | * list will remain unchanged. |
4085ebe8 VY |
7160 | */ |
7161 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
7162 | { | |
7163 | struct netdev_adjacent *lower; | |
7164 | ||
cfdd28be | 7165 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
7166 | |
7167 | if (&lower->list == &dev->adj_list.lower) | |
7168 | return NULL; | |
7169 | ||
cfdd28be | 7170 | *iter = lower->list.next; |
4085ebe8 VY |
7171 | |
7172 | return lower->dev; | |
7173 | } | |
7174 | EXPORT_SYMBOL(netdev_lower_get_next); | |
7175 | ||
1a3f060c DA |
7176 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
7177 | struct list_head **iter) | |
7178 | { | |
7179 | struct netdev_adjacent *lower; | |
7180 | ||
46b5ab1a | 7181 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
7182 | |
7183 | if (&lower->list == &dev->adj_list.lower) | |
7184 | return NULL; | |
7185 | ||
46b5ab1a | 7186 | *iter = &lower->list; |
1a3f060c DA |
7187 | |
7188 | return lower->dev; | |
7189 | } | |
7190 | ||
32b6d34f TY |
7191 | static struct net_device *__netdev_next_lower_dev(struct net_device *dev, |
7192 | struct list_head **iter, | |
7193 | bool *ignore) | |
7194 | { | |
7195 | struct netdev_adjacent *lower; | |
7196 | ||
7197 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
7198 | ||
7199 | if (&lower->list == &dev->adj_list.lower) | |
7200 | return NULL; | |
7201 | ||
7202 | *iter = &lower->list; | |
7203 | *ignore = lower->ignore; | |
7204 | ||
7205 | return lower->dev; | |
7206 | } | |
7207 | ||
1a3f060c DA |
7208 | int netdev_walk_all_lower_dev(struct net_device *dev, |
7209 | int (*fn)(struct net_device *dev, | |
7210 | void *data), | |
7211 | void *data) | |
7212 | { | |
5343da4c TY |
7213 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; |
7214 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7215 | int ret, cur = 0; | |
1a3f060c | 7216 | |
5343da4c TY |
7217 | now = dev; |
7218 | iter = &dev->adj_list.lower; | |
1a3f060c | 7219 | |
5343da4c TY |
7220 | while (1) { |
7221 | if (now != dev) { | |
7222 | ret = fn(now, data); | |
7223 | if (ret) | |
7224 | return ret; | |
7225 | } | |
7226 | ||
7227 | next = NULL; | |
7228 | while (1) { | |
7229 | ldev = netdev_next_lower_dev(now, &iter); | |
7230 | if (!ldev) | |
7231 | break; | |
7232 | ||
7233 | next = ldev; | |
7234 | niter = &ldev->adj_list.lower; | |
7235 | dev_stack[cur] = now; | |
7236 | iter_stack[cur++] = iter; | |
7237 | break; | |
7238 | } | |
7239 | ||
7240 | if (!next) { | |
7241 | if (!cur) | |
7242 | return 0; | |
7243 | next = dev_stack[--cur]; | |
7244 | niter = iter_stack[cur]; | |
7245 | } | |
7246 | ||
7247 | now = next; | |
7248 | iter = niter; | |
1a3f060c DA |
7249 | } |
7250 | ||
7251 | return 0; | |
7252 | } | |
7253 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
7254 | ||
32b6d34f TY |
7255 | static int __netdev_walk_all_lower_dev(struct net_device *dev, |
7256 | int (*fn)(struct net_device *dev, | |
7257 | void *data), | |
7258 | void *data) | |
7259 | { | |
7260 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7261 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7262 | int ret, cur = 0; | |
7263 | bool ignore; | |
7264 | ||
7265 | now = dev; | |
7266 | iter = &dev->adj_list.lower; | |
7267 | ||
7268 | while (1) { | |
7269 | if (now != dev) { | |
7270 | ret = fn(now, data); | |
7271 | if (ret) | |
7272 | return ret; | |
7273 | } | |
7274 | ||
7275 | next = NULL; | |
7276 | while (1) { | |
7277 | ldev = __netdev_next_lower_dev(now, &iter, &ignore); | |
7278 | if (!ldev) | |
7279 | break; | |
7280 | if (ignore) | |
7281 | continue; | |
7282 | ||
7283 | next = ldev; | |
7284 | niter = &ldev->adj_list.lower; | |
7285 | dev_stack[cur] = now; | |
7286 | iter_stack[cur++] = iter; | |
7287 | break; | |
7288 | } | |
7289 | ||
7290 | if (!next) { | |
7291 | if (!cur) | |
7292 | return 0; | |
7293 | next = dev_stack[--cur]; | |
7294 | niter = iter_stack[cur]; | |
7295 | } | |
7296 | ||
7297 | now = next; | |
7298 | iter = niter; | |
7299 | } | |
7300 | ||
7301 | return 0; | |
7302 | } | |
7303 | ||
7151affe TY |
7304 | struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
7305 | struct list_head **iter) | |
1a3f060c DA |
7306 | { |
7307 | struct netdev_adjacent *lower; | |
7308 | ||
7309 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7310 | if (&lower->list == &dev->adj_list.lower) | |
7311 | return NULL; | |
7312 | ||
7313 | *iter = &lower->list; | |
7314 | ||
7315 | return lower->dev; | |
7316 | } | |
7151affe | 7317 | EXPORT_SYMBOL(netdev_next_lower_dev_rcu); |
1a3f060c | 7318 | |
5343da4c TY |
7319 | static u8 __netdev_upper_depth(struct net_device *dev) |
7320 | { | |
7321 | struct net_device *udev; | |
7322 | struct list_head *iter; | |
7323 | u8 max_depth = 0; | |
32b6d34f | 7324 | bool ignore; |
5343da4c TY |
7325 | |
7326 | for (iter = &dev->adj_list.upper, | |
32b6d34f | 7327 | udev = __netdev_next_upper_dev(dev, &iter, &ignore); |
5343da4c | 7328 | udev; |
32b6d34f TY |
7329 | udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { |
7330 | if (ignore) | |
7331 | continue; | |
5343da4c TY |
7332 | if (max_depth < udev->upper_level) |
7333 | max_depth = udev->upper_level; | |
7334 | } | |
7335 | ||
7336 | return max_depth; | |
7337 | } | |
7338 | ||
7339 | static u8 __netdev_lower_depth(struct net_device *dev) | |
1a3f060c DA |
7340 | { |
7341 | struct net_device *ldev; | |
7342 | struct list_head *iter; | |
5343da4c | 7343 | u8 max_depth = 0; |
32b6d34f | 7344 | bool ignore; |
1a3f060c DA |
7345 | |
7346 | for (iter = &dev->adj_list.lower, | |
32b6d34f | 7347 | ldev = __netdev_next_lower_dev(dev, &iter, &ignore); |
1a3f060c | 7348 | ldev; |
32b6d34f TY |
7349 | ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { |
7350 | if (ignore) | |
7351 | continue; | |
5343da4c TY |
7352 | if (max_depth < ldev->lower_level) |
7353 | max_depth = ldev->lower_level; | |
7354 | } | |
1a3f060c | 7355 | |
5343da4c TY |
7356 | return max_depth; |
7357 | } | |
7358 | ||
7359 | static int __netdev_update_upper_level(struct net_device *dev, void *data) | |
7360 | { | |
7361 | dev->upper_level = __netdev_upper_depth(dev) + 1; | |
7362 | return 0; | |
7363 | } | |
7364 | ||
7365 | static int __netdev_update_lower_level(struct net_device *dev, void *data) | |
7366 | { | |
7367 | dev->lower_level = __netdev_lower_depth(dev) + 1; | |
7368 | return 0; | |
7369 | } | |
7370 | ||
7371 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
7372 | int (*fn)(struct net_device *dev, | |
7373 | void *data), | |
7374 | void *data) | |
7375 | { | |
7376 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7377 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7378 | int ret, cur = 0; | |
7379 | ||
7380 | now = dev; | |
7381 | iter = &dev->adj_list.lower; | |
7382 | ||
7383 | while (1) { | |
7384 | if (now != dev) { | |
7385 | ret = fn(now, data); | |
7386 | if (ret) | |
7387 | return ret; | |
7388 | } | |
7389 | ||
7390 | next = NULL; | |
7391 | while (1) { | |
7392 | ldev = netdev_next_lower_dev_rcu(now, &iter); | |
7393 | if (!ldev) | |
7394 | break; | |
7395 | ||
7396 | next = ldev; | |
7397 | niter = &ldev->adj_list.lower; | |
7398 | dev_stack[cur] = now; | |
7399 | iter_stack[cur++] = iter; | |
7400 | break; | |
7401 | } | |
7402 | ||
7403 | if (!next) { | |
7404 | if (!cur) | |
7405 | return 0; | |
7406 | next = dev_stack[--cur]; | |
7407 | niter = iter_stack[cur]; | |
7408 | } | |
7409 | ||
7410 | now = next; | |
7411 | iter = niter; | |
1a3f060c DA |
7412 | } |
7413 | ||
7414 | return 0; | |
7415 | } | |
7416 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
7417 | ||
e001bfad | 7418 | /** |
7419 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
7420 | * lower neighbour list, RCU | |
7421 | * variant | |
7422 | * @dev: device | |
7423 | * | |
7424 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
7425 | * list. The caller must hold RCU read lock. | |
7426 | */ | |
7427 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
7428 | { | |
7429 | struct netdev_adjacent *lower; | |
7430 | ||
7431 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
7432 | struct netdev_adjacent, list); | |
7433 | if (lower) | |
7434 | return lower->private; | |
7435 | return NULL; | |
7436 | } | |
7437 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
7438 | ||
9ff162a8 JP |
7439 | /** |
7440 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
7441 | * @dev: device | |
7442 | * | |
7443 | * Find a master upper device and return pointer to it or NULL in case | |
7444 | * it's not there. The caller must hold the RCU read lock. | |
7445 | */ | |
7446 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
7447 | { | |
aa9d8560 | 7448 | struct netdev_adjacent *upper; |
9ff162a8 | 7449 | |
2f268f12 | 7450 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 7451 | struct netdev_adjacent, list); |
9ff162a8 JP |
7452 | if (upper && likely(upper->master)) |
7453 | return upper->dev; | |
7454 | return NULL; | |
7455 | } | |
7456 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
7457 | ||
0a59f3a9 | 7458 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
7459 | struct net_device *adj_dev, |
7460 | struct list_head *dev_list) | |
7461 | { | |
7462 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 7463 | |
3ee32707 VF |
7464 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
7465 | "upper_%s" : "lower_%s", adj_dev->name); | |
7466 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
7467 | linkname); | |
7468 | } | |
0a59f3a9 | 7469 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
7470 | char *name, |
7471 | struct list_head *dev_list) | |
7472 | { | |
7473 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 7474 | |
3ee32707 VF |
7475 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
7476 | "upper_%s" : "lower_%s", name); | |
7477 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
7478 | } | |
7479 | ||
7ce64c79 AF |
7480 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
7481 | struct net_device *adj_dev, | |
7482 | struct list_head *dev_list) | |
7483 | { | |
7484 | return (dev_list == &dev->adj_list.upper || | |
7485 | dev_list == &dev->adj_list.lower) && | |
7486 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
7487 | } | |
3ee32707 | 7488 | |
5d261913 VF |
7489 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
7490 | struct net_device *adj_dev, | |
7863c054 | 7491 | struct list_head *dev_list, |
402dae96 | 7492 | void *private, bool master) |
5d261913 VF |
7493 | { |
7494 | struct netdev_adjacent *adj; | |
842d67a7 | 7495 | int ret; |
5d261913 | 7496 | |
6ea29da1 | 7497 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
7498 | |
7499 | if (adj) { | |
790510d9 | 7500 | adj->ref_nr += 1; |
67b62f98 DA |
7501 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
7502 | dev->name, adj_dev->name, adj->ref_nr); | |
7503 | ||
5d261913 VF |
7504 | return 0; |
7505 | } | |
7506 | ||
7507 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
7508 | if (!adj) | |
7509 | return -ENOMEM; | |
7510 | ||
7511 | adj->dev = adj_dev; | |
7512 | adj->master = master; | |
790510d9 | 7513 | adj->ref_nr = 1; |
402dae96 | 7514 | adj->private = private; |
32b6d34f | 7515 | adj->ignore = false; |
5d261913 | 7516 | dev_hold(adj_dev); |
2f268f12 | 7517 | |
67b62f98 DA |
7518 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
7519 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 7520 | |
7ce64c79 | 7521 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 7522 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
7523 | if (ret) |
7524 | goto free_adj; | |
7525 | } | |
7526 | ||
7863c054 | 7527 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
7528 | if (master) { |
7529 | ret = sysfs_create_link(&(dev->dev.kobj), | |
7530 | &(adj_dev->dev.kobj), "master"); | |
7531 | if (ret) | |
5831d66e | 7532 | goto remove_symlinks; |
842d67a7 | 7533 | |
7863c054 | 7534 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 7535 | } else { |
7863c054 | 7536 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 7537 | } |
5d261913 VF |
7538 | |
7539 | return 0; | |
842d67a7 | 7540 | |
5831d66e | 7541 | remove_symlinks: |
7ce64c79 | 7542 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 7543 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
7544 | free_adj: |
7545 | kfree(adj); | |
974daef7 | 7546 | dev_put(adj_dev); |
842d67a7 VF |
7547 | |
7548 | return ret; | |
5d261913 VF |
7549 | } |
7550 | ||
1d143d9f | 7551 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
7552 | struct net_device *adj_dev, | |
93409033 | 7553 | u16 ref_nr, |
1d143d9f | 7554 | struct list_head *dev_list) |
5d261913 VF |
7555 | { |
7556 | struct netdev_adjacent *adj; | |
7557 | ||
67b62f98 DA |
7558 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
7559 | dev->name, adj_dev->name, ref_nr); | |
7560 | ||
6ea29da1 | 7561 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 7562 | |
2f268f12 | 7563 | if (!adj) { |
67b62f98 | 7564 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 7565 | dev->name, adj_dev->name); |
67b62f98 DA |
7566 | WARN_ON(1); |
7567 | return; | |
2f268f12 | 7568 | } |
5d261913 | 7569 | |
93409033 | 7570 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
7571 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
7572 | dev->name, adj_dev->name, ref_nr, | |
7573 | adj->ref_nr - ref_nr); | |
93409033 | 7574 | adj->ref_nr -= ref_nr; |
5d261913 VF |
7575 | return; |
7576 | } | |
7577 | ||
842d67a7 VF |
7578 | if (adj->master) |
7579 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
7580 | ||
7ce64c79 | 7581 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 7582 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 7583 | |
5d261913 | 7584 | list_del_rcu(&adj->list); |
67b62f98 | 7585 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 7586 | adj_dev->name, dev->name, adj_dev->name); |
5d261913 VF |
7587 | dev_put(adj_dev); |
7588 | kfree_rcu(adj, rcu); | |
7589 | } | |
7590 | ||
1d143d9f | 7591 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
7592 | struct net_device *upper_dev, | |
7593 | struct list_head *up_list, | |
7594 | struct list_head *down_list, | |
7595 | void *private, bool master) | |
5d261913 VF |
7596 | { |
7597 | int ret; | |
7598 | ||
790510d9 | 7599 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 7600 | private, master); |
5d261913 VF |
7601 | if (ret) |
7602 | return ret; | |
7603 | ||
790510d9 | 7604 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 7605 | private, false); |
5d261913 | 7606 | if (ret) { |
790510d9 | 7607 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
7608 | return ret; |
7609 | } | |
7610 | ||
7611 | return 0; | |
7612 | } | |
7613 | ||
1d143d9f | 7614 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
7615 | struct net_device *upper_dev, | |
93409033 | 7616 | u16 ref_nr, |
1d143d9f | 7617 | struct list_head *up_list, |
7618 | struct list_head *down_list) | |
5d261913 | 7619 | { |
93409033 AC |
7620 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
7621 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
7622 | } |
7623 | ||
1d143d9f | 7624 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
7625 | struct net_device *upper_dev, | |
7626 | void *private, bool master) | |
2f268f12 | 7627 | { |
f1170fd4 DA |
7628 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
7629 | &dev->adj_list.upper, | |
7630 | &upper_dev->adj_list.lower, | |
7631 | private, master); | |
5d261913 VF |
7632 | } |
7633 | ||
1d143d9f | 7634 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
7635 | struct net_device *upper_dev) | |
2f268f12 | 7636 | { |
93409033 | 7637 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
7638 | &dev->adj_list.upper, |
7639 | &upper_dev->adj_list.lower); | |
7640 | } | |
5d261913 | 7641 | |
9ff162a8 | 7642 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 7643 | struct net_device *upper_dev, bool master, |
42ab19ee DA |
7644 | void *upper_priv, void *upper_info, |
7645 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7646 | { |
51d0c047 DA |
7647 | struct netdev_notifier_changeupper_info changeupper_info = { |
7648 | .info = { | |
7649 | .dev = dev, | |
42ab19ee | 7650 | .extack = extack, |
51d0c047 DA |
7651 | }, |
7652 | .upper_dev = upper_dev, | |
7653 | .master = master, | |
7654 | .linking = true, | |
7655 | .upper_info = upper_info, | |
7656 | }; | |
50d629e7 | 7657 | struct net_device *master_dev; |
5d261913 | 7658 | int ret = 0; |
9ff162a8 JP |
7659 | |
7660 | ASSERT_RTNL(); | |
7661 | ||
7662 | if (dev == upper_dev) | |
7663 | return -EBUSY; | |
7664 | ||
7665 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
32b6d34f | 7666 | if (__netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
7667 | return -EBUSY; |
7668 | ||
5343da4c TY |
7669 | if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) |
7670 | return -EMLINK; | |
7671 | ||
50d629e7 | 7672 | if (!master) { |
32b6d34f | 7673 | if (__netdev_has_upper_dev(dev, upper_dev)) |
50d629e7 MM |
7674 | return -EEXIST; |
7675 | } else { | |
32b6d34f | 7676 | master_dev = __netdev_master_upper_dev_get(dev); |
50d629e7 MM |
7677 | if (master_dev) |
7678 | return master_dev == upper_dev ? -EEXIST : -EBUSY; | |
7679 | } | |
9ff162a8 | 7680 | |
51d0c047 | 7681 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
7682 | &changeupper_info.info); |
7683 | ret = notifier_to_errno(ret); | |
7684 | if (ret) | |
7685 | return ret; | |
7686 | ||
6dffb044 | 7687 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 7688 | master); |
5d261913 VF |
7689 | if (ret) |
7690 | return ret; | |
9ff162a8 | 7691 | |
51d0c047 | 7692 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
b03804e7 IS |
7693 | &changeupper_info.info); |
7694 | ret = notifier_to_errno(ret); | |
7695 | if (ret) | |
f1170fd4 | 7696 | goto rollback; |
b03804e7 | 7697 | |
5343da4c | 7698 | __netdev_update_upper_level(dev, NULL); |
32b6d34f | 7699 | __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); |
5343da4c TY |
7700 | |
7701 | __netdev_update_lower_level(upper_dev, NULL); | |
32b6d34f TY |
7702 | __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, |
7703 | NULL); | |
5343da4c | 7704 | |
9ff162a8 | 7705 | return 0; |
5d261913 | 7706 | |
f1170fd4 | 7707 | rollback: |
2f268f12 | 7708 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
7709 | |
7710 | return ret; | |
9ff162a8 JP |
7711 | } |
7712 | ||
7713 | /** | |
7714 | * netdev_upper_dev_link - Add a link to the upper device | |
7715 | * @dev: device | |
7716 | * @upper_dev: new upper device | |
7a006d59 | 7717 | * @extack: netlink extended ack |
9ff162a8 JP |
7718 | * |
7719 | * Adds a link to device which is upper to this one. The caller must hold | |
7720 | * the RTNL lock. On a failure a negative errno code is returned. | |
7721 | * On success the reference counts are adjusted and the function | |
7722 | * returns zero. | |
7723 | */ | |
7724 | int netdev_upper_dev_link(struct net_device *dev, | |
42ab19ee DA |
7725 | struct net_device *upper_dev, |
7726 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7727 | { |
42ab19ee DA |
7728 | return __netdev_upper_dev_link(dev, upper_dev, false, |
7729 | NULL, NULL, extack); | |
9ff162a8 JP |
7730 | } |
7731 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
7732 | ||
7733 | /** | |
7734 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
7735 | * @dev: device | |
7736 | * @upper_dev: new upper device | |
6dffb044 | 7737 | * @upper_priv: upper device private |
29bf24af | 7738 | * @upper_info: upper info to be passed down via notifier |
7a006d59 | 7739 | * @extack: netlink extended ack |
9ff162a8 JP |
7740 | * |
7741 | * Adds a link to device which is upper to this one. In this case, only | |
7742 | * one master upper device can be linked, although other non-master devices | |
7743 | * might be linked as well. The caller must hold the RTNL lock. | |
7744 | * On a failure a negative errno code is returned. On success the reference | |
7745 | * counts are adjusted and the function returns zero. | |
7746 | */ | |
7747 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 7748 | struct net_device *upper_dev, |
42ab19ee DA |
7749 | void *upper_priv, void *upper_info, |
7750 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7751 | { |
29bf24af | 7752 | return __netdev_upper_dev_link(dev, upper_dev, true, |
42ab19ee | 7753 | upper_priv, upper_info, extack); |
9ff162a8 JP |
7754 | } |
7755 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
7756 | ||
7757 | /** | |
7758 | * netdev_upper_dev_unlink - Removes a link to upper device | |
7759 | * @dev: device | |
7760 | * @upper_dev: new upper device | |
7761 | * | |
7762 | * Removes a link to device which is upper to this one. The caller must hold | |
7763 | * the RTNL lock. | |
7764 | */ | |
7765 | void netdev_upper_dev_unlink(struct net_device *dev, | |
7766 | struct net_device *upper_dev) | |
7767 | { | |
51d0c047 DA |
7768 | struct netdev_notifier_changeupper_info changeupper_info = { |
7769 | .info = { | |
7770 | .dev = dev, | |
7771 | }, | |
7772 | .upper_dev = upper_dev, | |
7773 | .linking = false, | |
7774 | }; | |
f4563a75 | 7775 | |
9ff162a8 JP |
7776 | ASSERT_RTNL(); |
7777 | ||
0e4ead9d | 7778 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; |
0e4ead9d | 7779 | |
51d0c047 | 7780 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
7781 | &changeupper_info.info); |
7782 | ||
2f268f12 | 7783 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 7784 | |
51d0c047 | 7785 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
0e4ead9d | 7786 | &changeupper_info.info); |
5343da4c TY |
7787 | |
7788 | __netdev_update_upper_level(dev, NULL); | |
32b6d34f | 7789 | __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); |
5343da4c TY |
7790 | |
7791 | __netdev_update_lower_level(upper_dev, NULL); | |
32b6d34f TY |
7792 | __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, |
7793 | NULL); | |
9ff162a8 JP |
7794 | } |
7795 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
7796 | ||
32b6d34f TY |
7797 | static void __netdev_adjacent_dev_set(struct net_device *upper_dev, |
7798 | struct net_device *lower_dev, | |
7799 | bool val) | |
7800 | { | |
7801 | struct netdev_adjacent *adj; | |
7802 | ||
7803 | adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); | |
7804 | if (adj) | |
7805 | adj->ignore = val; | |
7806 | ||
7807 | adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); | |
7808 | if (adj) | |
7809 | adj->ignore = val; | |
7810 | } | |
7811 | ||
7812 | static void netdev_adjacent_dev_disable(struct net_device *upper_dev, | |
7813 | struct net_device *lower_dev) | |
7814 | { | |
7815 | __netdev_adjacent_dev_set(upper_dev, lower_dev, true); | |
7816 | } | |
7817 | ||
7818 | static void netdev_adjacent_dev_enable(struct net_device *upper_dev, | |
7819 | struct net_device *lower_dev) | |
7820 | { | |
7821 | __netdev_adjacent_dev_set(upper_dev, lower_dev, false); | |
7822 | } | |
7823 | ||
7824 | int netdev_adjacent_change_prepare(struct net_device *old_dev, | |
7825 | struct net_device *new_dev, | |
7826 | struct net_device *dev, | |
7827 | struct netlink_ext_ack *extack) | |
7828 | { | |
7829 | int err; | |
7830 | ||
7831 | if (!new_dev) | |
7832 | return 0; | |
7833 | ||
7834 | if (old_dev && new_dev != old_dev) | |
7835 | netdev_adjacent_dev_disable(dev, old_dev); | |
7836 | ||
7837 | err = netdev_upper_dev_link(new_dev, dev, extack); | |
7838 | if (err) { | |
7839 | if (old_dev && new_dev != old_dev) | |
7840 | netdev_adjacent_dev_enable(dev, old_dev); | |
7841 | return err; | |
7842 | } | |
7843 | ||
7844 | return 0; | |
7845 | } | |
7846 | EXPORT_SYMBOL(netdev_adjacent_change_prepare); | |
7847 | ||
7848 | void netdev_adjacent_change_commit(struct net_device *old_dev, | |
7849 | struct net_device *new_dev, | |
7850 | struct net_device *dev) | |
7851 | { | |
7852 | if (!new_dev || !old_dev) | |
7853 | return; | |
7854 | ||
7855 | if (new_dev == old_dev) | |
7856 | return; | |
7857 | ||
7858 | netdev_adjacent_dev_enable(dev, old_dev); | |
7859 | netdev_upper_dev_unlink(old_dev, dev); | |
7860 | } | |
7861 | EXPORT_SYMBOL(netdev_adjacent_change_commit); | |
7862 | ||
7863 | void netdev_adjacent_change_abort(struct net_device *old_dev, | |
7864 | struct net_device *new_dev, | |
7865 | struct net_device *dev) | |
7866 | { | |
7867 | if (!new_dev) | |
7868 | return; | |
7869 | ||
7870 | if (old_dev && new_dev != old_dev) | |
7871 | netdev_adjacent_dev_enable(dev, old_dev); | |
7872 | ||
7873 | netdev_upper_dev_unlink(new_dev, dev); | |
7874 | } | |
7875 | EXPORT_SYMBOL(netdev_adjacent_change_abort); | |
7876 | ||
61bd3857 MS |
7877 | /** |
7878 | * netdev_bonding_info_change - Dispatch event about slave change | |
7879 | * @dev: device | |
4a26e453 | 7880 | * @bonding_info: info to dispatch |
61bd3857 MS |
7881 | * |
7882 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
7883 | * The caller must hold the RTNL lock. | |
7884 | */ | |
7885 | void netdev_bonding_info_change(struct net_device *dev, | |
7886 | struct netdev_bonding_info *bonding_info) | |
7887 | { | |
51d0c047 DA |
7888 | struct netdev_notifier_bonding_info info = { |
7889 | .info.dev = dev, | |
7890 | }; | |
61bd3857 MS |
7891 | |
7892 | memcpy(&info.bonding_info, bonding_info, | |
7893 | sizeof(struct netdev_bonding_info)); | |
51d0c047 | 7894 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, |
61bd3857 MS |
7895 | &info.info); |
7896 | } | |
7897 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
7898 | ||
cff9f12b MG |
7899 | /** |
7900 | * netdev_get_xmit_slave - Get the xmit slave of master device | |
7901 | * @skb: The packet | |
7902 | * @all_slaves: assume all the slaves are active | |
7903 | * | |
7904 | * The reference counters are not incremented so the caller must be | |
7905 | * careful with locks. The caller must hold RCU lock. | |
7906 | * %NULL is returned if no slave is found. | |
7907 | */ | |
7908 | ||
7909 | struct net_device *netdev_get_xmit_slave(struct net_device *dev, | |
7910 | struct sk_buff *skb, | |
7911 | bool all_slaves) | |
7912 | { | |
7913 | const struct net_device_ops *ops = dev->netdev_ops; | |
7914 | ||
7915 | if (!ops->ndo_get_xmit_slave) | |
7916 | return NULL; | |
7917 | return ops->ndo_get_xmit_slave(dev, skb, all_slaves); | |
7918 | } | |
7919 | EXPORT_SYMBOL(netdev_get_xmit_slave); | |
7920 | ||
2ce1ee17 | 7921 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
7922 | { |
7923 | struct netdev_adjacent *iter; | |
7924 | ||
7925 | struct net *net = dev_net(dev); | |
7926 | ||
7927 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 7928 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7929 | continue; |
7930 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7931 | &iter->dev->adj_list.lower); | |
7932 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
7933 | &dev->adj_list.upper); | |
7934 | } | |
7935 | ||
7936 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 7937 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7938 | continue; |
7939 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7940 | &iter->dev->adj_list.upper); | |
7941 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
7942 | &dev->adj_list.lower); | |
7943 | } | |
7944 | } | |
7945 | ||
2ce1ee17 | 7946 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
7947 | { |
7948 | struct netdev_adjacent *iter; | |
7949 | ||
7950 | struct net *net = dev_net(dev); | |
7951 | ||
7952 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 7953 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7954 | continue; |
7955 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
7956 | &iter->dev->adj_list.lower); | |
7957 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
7958 | &dev->adj_list.upper); | |
7959 | } | |
7960 | ||
7961 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 7962 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7963 | continue; |
7964 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
7965 | &iter->dev->adj_list.upper); | |
7966 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
7967 | &dev->adj_list.lower); | |
7968 | } | |
7969 | } | |
7970 | ||
5bb025fa | 7971 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 7972 | { |
5bb025fa | 7973 | struct netdev_adjacent *iter; |
402dae96 | 7974 | |
4c75431a AF |
7975 | struct net *net = dev_net(dev); |
7976 | ||
5bb025fa | 7977 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 7978 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 7979 | continue; |
5bb025fa VF |
7980 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
7981 | &iter->dev->adj_list.lower); | |
7982 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7983 | &iter->dev->adj_list.lower); | |
7984 | } | |
402dae96 | 7985 | |
5bb025fa | 7986 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 7987 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 7988 | continue; |
5bb025fa VF |
7989 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
7990 | &iter->dev->adj_list.upper); | |
7991 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7992 | &iter->dev->adj_list.upper); | |
7993 | } | |
402dae96 | 7994 | } |
402dae96 VF |
7995 | |
7996 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
7997 | struct net_device *lower_dev) | |
7998 | { | |
7999 | struct netdev_adjacent *lower; | |
8000 | ||
8001 | if (!lower_dev) | |
8002 | return NULL; | |
6ea29da1 | 8003 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
8004 | if (!lower) |
8005 | return NULL; | |
8006 | ||
8007 | return lower->private; | |
8008 | } | |
8009 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
8010 | ||
4085ebe8 | 8011 | |
04d48266 JP |
8012 | /** |
8013 | * netdev_lower_change - Dispatch event about lower device state change | |
8014 | * @lower_dev: device | |
8015 | * @lower_state_info: state to dispatch | |
8016 | * | |
8017 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
8018 | * The caller must hold the RTNL lock. | |
8019 | */ | |
8020 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
8021 | void *lower_state_info) | |
8022 | { | |
51d0c047 DA |
8023 | struct netdev_notifier_changelowerstate_info changelowerstate_info = { |
8024 | .info.dev = lower_dev, | |
8025 | }; | |
04d48266 JP |
8026 | |
8027 | ASSERT_RTNL(); | |
8028 | changelowerstate_info.lower_state_info = lower_state_info; | |
51d0c047 | 8029 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, |
04d48266 JP |
8030 | &changelowerstate_info.info); |
8031 | } | |
8032 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
8033 | ||
b6c40d68 PM |
8034 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
8035 | { | |
d314774c SH |
8036 | const struct net_device_ops *ops = dev->netdev_ops; |
8037 | ||
d2615bf4 | 8038 | if (ops->ndo_change_rx_flags) |
d314774c | 8039 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
8040 | } |
8041 | ||
991fb3f7 | 8042 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 8043 | { |
b536db93 | 8044 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
8045 | kuid_t uid; |
8046 | kgid_t gid; | |
1da177e4 | 8047 | |
24023451 PM |
8048 | ASSERT_RTNL(); |
8049 | ||
dad9b335 WC |
8050 | dev->flags |= IFF_PROMISC; |
8051 | dev->promiscuity += inc; | |
8052 | if (dev->promiscuity == 0) { | |
8053 | /* | |
8054 | * Avoid overflow. | |
8055 | * If inc causes overflow, untouch promisc and return error. | |
8056 | */ | |
8057 | if (inc < 0) | |
8058 | dev->flags &= ~IFF_PROMISC; | |
8059 | else { | |
8060 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
8061 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
8062 | dev->name); | |
dad9b335 WC |
8063 | return -EOVERFLOW; |
8064 | } | |
8065 | } | |
52609c0b | 8066 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
8067 | pr_info("device %s %s promiscuous mode\n", |
8068 | dev->name, | |
8069 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
8070 | if (audit_enabled) { |
8071 | current_uid_gid(&uid, &gid); | |
cdfb6b34 RGB |
8072 | audit_log(audit_context(), GFP_ATOMIC, |
8073 | AUDIT_ANOM_PROMISCUOUS, | |
8074 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
8075 | dev->name, (dev->flags & IFF_PROMISC), | |
8076 | (old_flags & IFF_PROMISC), | |
8077 | from_kuid(&init_user_ns, audit_get_loginuid(current)), | |
8078 | from_kuid(&init_user_ns, uid), | |
8079 | from_kgid(&init_user_ns, gid), | |
8080 | audit_get_sessionid(current)); | |
8192b0c4 | 8081 | } |
24023451 | 8082 | |
b6c40d68 | 8083 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 8084 | } |
991fb3f7 ND |
8085 | if (notify) |
8086 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 8087 | return 0; |
1da177e4 LT |
8088 | } |
8089 | ||
4417da66 PM |
8090 | /** |
8091 | * dev_set_promiscuity - update promiscuity count on a device | |
8092 | * @dev: device | |
8093 | * @inc: modifier | |
8094 | * | |
8095 | * Add or remove promiscuity from a device. While the count in the device | |
8096 | * remains above zero the interface remains promiscuous. Once it hits zero | |
8097 | * the device reverts back to normal filtering operation. A negative inc | |
8098 | * value is used to drop promiscuity on the device. | |
dad9b335 | 8099 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 8100 | */ |
dad9b335 | 8101 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 8102 | { |
b536db93 | 8103 | unsigned int old_flags = dev->flags; |
dad9b335 | 8104 | int err; |
4417da66 | 8105 | |
991fb3f7 | 8106 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 8107 | if (err < 0) |
dad9b335 | 8108 | return err; |
4417da66 PM |
8109 | if (dev->flags != old_flags) |
8110 | dev_set_rx_mode(dev); | |
dad9b335 | 8111 | return err; |
4417da66 | 8112 | } |
d1b19dff | 8113 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 8114 | |
991fb3f7 | 8115 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 8116 | { |
991fb3f7 | 8117 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 8118 | |
24023451 PM |
8119 | ASSERT_RTNL(); |
8120 | ||
1da177e4 | 8121 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
8122 | dev->allmulti += inc; |
8123 | if (dev->allmulti == 0) { | |
8124 | /* | |
8125 | * Avoid overflow. | |
8126 | * If inc causes overflow, untouch allmulti and return error. | |
8127 | */ | |
8128 | if (inc < 0) | |
8129 | dev->flags &= ~IFF_ALLMULTI; | |
8130 | else { | |
8131 | dev->allmulti -= inc; | |
7b6cd1ce JP |
8132 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
8133 | dev->name); | |
dad9b335 WC |
8134 | return -EOVERFLOW; |
8135 | } | |
8136 | } | |
24023451 | 8137 | if (dev->flags ^ old_flags) { |
b6c40d68 | 8138 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 8139 | dev_set_rx_mode(dev); |
991fb3f7 ND |
8140 | if (notify) |
8141 | __dev_notify_flags(dev, old_flags, | |
8142 | dev->gflags ^ old_gflags); | |
24023451 | 8143 | } |
dad9b335 | 8144 | return 0; |
4417da66 | 8145 | } |
991fb3f7 ND |
8146 | |
8147 | /** | |
8148 | * dev_set_allmulti - update allmulti count on a device | |
8149 | * @dev: device | |
8150 | * @inc: modifier | |
8151 | * | |
8152 | * Add or remove reception of all multicast frames to a device. While the | |
8153 | * count in the device remains above zero the interface remains listening | |
8154 | * to all interfaces. Once it hits zero the device reverts back to normal | |
8155 | * filtering operation. A negative @inc value is used to drop the counter | |
8156 | * when releasing a resource needing all multicasts. | |
8157 | * Return 0 if successful or a negative errno code on error. | |
8158 | */ | |
8159 | ||
8160 | int dev_set_allmulti(struct net_device *dev, int inc) | |
8161 | { | |
8162 | return __dev_set_allmulti(dev, inc, true); | |
8163 | } | |
d1b19dff | 8164 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
8165 | |
8166 | /* | |
8167 | * Upload unicast and multicast address lists to device and | |
8168 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 8169 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
8170 | * are present. |
8171 | */ | |
8172 | void __dev_set_rx_mode(struct net_device *dev) | |
8173 | { | |
d314774c SH |
8174 | const struct net_device_ops *ops = dev->netdev_ops; |
8175 | ||
4417da66 PM |
8176 | /* dev_open will call this function so the list will stay sane. */ |
8177 | if (!(dev->flags&IFF_UP)) | |
8178 | return; | |
8179 | ||
8180 | if (!netif_device_present(dev)) | |
40b77c94 | 8181 | return; |
4417da66 | 8182 | |
01789349 | 8183 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
8184 | /* Unicast addresses changes may only happen under the rtnl, |
8185 | * therefore calling __dev_set_promiscuity here is safe. | |
8186 | */ | |
32e7bfc4 | 8187 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 8188 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 8189 | dev->uc_promisc = true; |
32e7bfc4 | 8190 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 8191 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 8192 | dev->uc_promisc = false; |
4417da66 | 8193 | } |
4417da66 | 8194 | } |
01789349 JP |
8195 | |
8196 | if (ops->ndo_set_rx_mode) | |
8197 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
8198 | } |
8199 | ||
8200 | void dev_set_rx_mode(struct net_device *dev) | |
8201 | { | |
b9e40857 | 8202 | netif_addr_lock_bh(dev); |
4417da66 | 8203 | __dev_set_rx_mode(dev); |
b9e40857 | 8204 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
8205 | } |
8206 | ||
f0db275a SH |
8207 | /** |
8208 | * dev_get_flags - get flags reported to userspace | |
8209 | * @dev: device | |
8210 | * | |
8211 | * Get the combination of flag bits exported through APIs to userspace. | |
8212 | */ | |
95c96174 | 8213 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 8214 | { |
95c96174 | 8215 | unsigned int flags; |
1da177e4 LT |
8216 | |
8217 | flags = (dev->flags & ~(IFF_PROMISC | | |
8218 | IFF_ALLMULTI | | |
b00055aa SR |
8219 | IFF_RUNNING | |
8220 | IFF_LOWER_UP | | |
8221 | IFF_DORMANT)) | | |
1da177e4 LT |
8222 | (dev->gflags & (IFF_PROMISC | |
8223 | IFF_ALLMULTI)); | |
8224 | ||
b00055aa SR |
8225 | if (netif_running(dev)) { |
8226 | if (netif_oper_up(dev)) | |
8227 | flags |= IFF_RUNNING; | |
8228 | if (netif_carrier_ok(dev)) | |
8229 | flags |= IFF_LOWER_UP; | |
8230 | if (netif_dormant(dev)) | |
8231 | flags |= IFF_DORMANT; | |
8232 | } | |
1da177e4 LT |
8233 | |
8234 | return flags; | |
8235 | } | |
d1b19dff | 8236 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 8237 | |
6d040321 PM |
8238 | int __dev_change_flags(struct net_device *dev, unsigned int flags, |
8239 | struct netlink_ext_ack *extack) | |
1da177e4 | 8240 | { |
b536db93 | 8241 | unsigned int old_flags = dev->flags; |
bd380811 | 8242 | int ret; |
1da177e4 | 8243 | |
24023451 PM |
8244 | ASSERT_RTNL(); |
8245 | ||
1da177e4 LT |
8246 | /* |
8247 | * Set the flags on our device. | |
8248 | */ | |
8249 | ||
8250 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
8251 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
8252 | IFF_AUTOMEDIA)) | | |
8253 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
8254 | IFF_ALLMULTI)); | |
8255 | ||
8256 | /* | |
8257 | * Load in the correct multicast list now the flags have changed. | |
8258 | */ | |
8259 | ||
b6c40d68 PM |
8260 | if ((old_flags ^ flags) & IFF_MULTICAST) |
8261 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 8262 | |
4417da66 | 8263 | dev_set_rx_mode(dev); |
1da177e4 LT |
8264 | |
8265 | /* | |
8266 | * Have we downed the interface. We handle IFF_UP ourselves | |
8267 | * according to user attempts to set it, rather than blindly | |
8268 | * setting it. | |
8269 | */ | |
8270 | ||
8271 | ret = 0; | |
7051b88a | 8272 | if ((old_flags ^ flags) & IFF_UP) { |
8273 | if (old_flags & IFF_UP) | |
8274 | __dev_close(dev); | |
8275 | else | |
40c900aa | 8276 | ret = __dev_open(dev, extack); |
7051b88a | 8277 | } |
1da177e4 | 8278 | |
1da177e4 | 8279 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 8280 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 8281 | unsigned int old_flags = dev->flags; |
d1b19dff | 8282 | |
1da177e4 | 8283 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
8284 | |
8285 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
8286 | if (dev->flags != old_flags) | |
8287 | dev_set_rx_mode(dev); | |
1da177e4 LT |
8288 | } |
8289 | ||
8290 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 8291 | * is important. Some (broken) drivers set IFF_PROMISC, when |
8292 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
8293 | */ |
8294 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
8295 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
8296 | ||
1da177e4 | 8297 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 8298 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
8299 | } |
8300 | ||
bd380811 PM |
8301 | return ret; |
8302 | } | |
8303 | ||
a528c219 ND |
8304 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
8305 | unsigned int gchanges) | |
bd380811 PM |
8306 | { |
8307 | unsigned int changes = dev->flags ^ old_flags; | |
8308 | ||
a528c219 | 8309 | if (gchanges) |
7f294054 | 8310 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 8311 | |
bd380811 PM |
8312 | if (changes & IFF_UP) { |
8313 | if (dev->flags & IFF_UP) | |
8314 | call_netdevice_notifiers(NETDEV_UP, dev); | |
8315 | else | |
8316 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
8317 | } | |
8318 | ||
8319 | if (dev->flags & IFF_UP && | |
be9efd36 | 8320 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
51d0c047 DA |
8321 | struct netdev_notifier_change_info change_info = { |
8322 | .info = { | |
8323 | .dev = dev, | |
8324 | }, | |
8325 | .flags_changed = changes, | |
8326 | }; | |
be9efd36 | 8327 | |
51d0c047 | 8328 | call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); |
be9efd36 | 8329 | } |
bd380811 PM |
8330 | } |
8331 | ||
8332 | /** | |
8333 | * dev_change_flags - change device settings | |
8334 | * @dev: device | |
8335 | * @flags: device state flags | |
567c5e13 | 8336 | * @extack: netlink extended ack |
bd380811 PM |
8337 | * |
8338 | * Change settings on device based state flags. The flags are | |
8339 | * in the userspace exported format. | |
8340 | */ | |
567c5e13 PM |
8341 | int dev_change_flags(struct net_device *dev, unsigned int flags, |
8342 | struct netlink_ext_ack *extack) | |
bd380811 | 8343 | { |
b536db93 | 8344 | int ret; |
991fb3f7 | 8345 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 | 8346 | |
6d040321 | 8347 | ret = __dev_change_flags(dev, flags, extack); |
bd380811 PM |
8348 | if (ret < 0) |
8349 | return ret; | |
8350 | ||
991fb3f7 | 8351 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 8352 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
8353 | return ret; |
8354 | } | |
d1b19dff | 8355 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 8356 | |
f51048c3 | 8357 | int __dev_set_mtu(struct net_device *dev, int new_mtu) |
2315dc91 VF |
8358 | { |
8359 | const struct net_device_ops *ops = dev->netdev_ops; | |
8360 | ||
8361 | if (ops->ndo_change_mtu) | |
8362 | return ops->ndo_change_mtu(dev, new_mtu); | |
8363 | ||
501a90c9 ED |
8364 | /* Pairs with all the lockless reads of dev->mtu in the stack */ |
8365 | WRITE_ONCE(dev->mtu, new_mtu); | |
2315dc91 VF |
8366 | return 0; |
8367 | } | |
f51048c3 | 8368 | EXPORT_SYMBOL(__dev_set_mtu); |
2315dc91 | 8369 | |
d836f5c6 ED |
8370 | int dev_validate_mtu(struct net_device *dev, int new_mtu, |
8371 | struct netlink_ext_ack *extack) | |
8372 | { | |
8373 | /* MTU must be positive, and in range */ | |
8374 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
8375 | NL_SET_ERR_MSG(extack, "mtu less than device minimum"); | |
8376 | return -EINVAL; | |
8377 | } | |
8378 | ||
8379 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
8380 | NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); | |
8381 | return -EINVAL; | |
8382 | } | |
8383 | return 0; | |
8384 | } | |
8385 | ||
f0db275a | 8386 | /** |
7a4c53be | 8387 | * dev_set_mtu_ext - Change maximum transfer unit |
f0db275a SH |
8388 | * @dev: device |
8389 | * @new_mtu: new transfer unit | |
7a4c53be | 8390 | * @extack: netlink extended ack |
f0db275a SH |
8391 | * |
8392 | * Change the maximum transfer size of the network device. | |
8393 | */ | |
7a4c53be SH |
8394 | int dev_set_mtu_ext(struct net_device *dev, int new_mtu, |
8395 | struct netlink_ext_ack *extack) | |
1da177e4 | 8396 | { |
2315dc91 | 8397 | int err, orig_mtu; |
1da177e4 LT |
8398 | |
8399 | if (new_mtu == dev->mtu) | |
8400 | return 0; | |
8401 | ||
d836f5c6 ED |
8402 | err = dev_validate_mtu(dev, new_mtu, extack); |
8403 | if (err) | |
8404 | return err; | |
1da177e4 LT |
8405 | |
8406 | if (!netif_device_present(dev)) | |
8407 | return -ENODEV; | |
8408 | ||
1d486bfb VF |
8409 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
8410 | err = notifier_to_errno(err); | |
8411 | if (err) | |
8412 | return err; | |
d314774c | 8413 | |
2315dc91 VF |
8414 | orig_mtu = dev->mtu; |
8415 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 8416 | |
2315dc91 | 8417 | if (!err) { |
af7d6cce SD |
8418 | err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
8419 | orig_mtu); | |
2315dc91 VF |
8420 | err = notifier_to_errno(err); |
8421 | if (err) { | |
8422 | /* setting mtu back and notifying everyone again, | |
8423 | * so that they have a chance to revert changes. | |
8424 | */ | |
8425 | __dev_set_mtu(dev, orig_mtu); | |
af7d6cce SD |
8426 | call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
8427 | new_mtu); | |
2315dc91 VF |
8428 | } |
8429 | } | |
1da177e4 LT |
8430 | return err; |
8431 | } | |
7a4c53be SH |
8432 | |
8433 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
8434 | { | |
8435 | struct netlink_ext_ack extack; | |
8436 | int err; | |
8437 | ||
a6bcfc89 | 8438 | memset(&extack, 0, sizeof(extack)); |
7a4c53be | 8439 | err = dev_set_mtu_ext(dev, new_mtu, &extack); |
a6bcfc89 | 8440 | if (err && extack._msg) |
7a4c53be SH |
8441 | net_err_ratelimited("%s: %s\n", dev->name, extack._msg); |
8442 | return err; | |
8443 | } | |
d1b19dff | 8444 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 8445 | |
6a643ddb CW |
8446 | /** |
8447 | * dev_change_tx_queue_len - Change TX queue length of a netdevice | |
8448 | * @dev: device | |
8449 | * @new_len: new tx queue length | |
8450 | */ | |
8451 | int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) | |
8452 | { | |
8453 | unsigned int orig_len = dev->tx_queue_len; | |
8454 | int res; | |
8455 | ||
8456 | if (new_len != (unsigned int)new_len) | |
8457 | return -ERANGE; | |
8458 | ||
8459 | if (new_len != orig_len) { | |
8460 | dev->tx_queue_len = new_len; | |
8461 | res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); | |
8462 | res = notifier_to_errno(res); | |
7effaf06 TT |
8463 | if (res) |
8464 | goto err_rollback; | |
8465 | res = dev_qdisc_change_tx_queue_len(dev); | |
8466 | if (res) | |
8467 | goto err_rollback; | |
6a643ddb CW |
8468 | } |
8469 | ||
8470 | return 0; | |
7effaf06 TT |
8471 | |
8472 | err_rollback: | |
8473 | netdev_err(dev, "refused to change device tx_queue_len\n"); | |
8474 | dev->tx_queue_len = orig_len; | |
8475 | return res; | |
6a643ddb CW |
8476 | } |
8477 | ||
cbda10fa VD |
8478 | /** |
8479 | * dev_set_group - Change group this device belongs to | |
8480 | * @dev: device | |
8481 | * @new_group: group this device should belong to | |
8482 | */ | |
8483 | void dev_set_group(struct net_device *dev, int new_group) | |
8484 | { | |
8485 | dev->group = new_group; | |
8486 | } | |
8487 | EXPORT_SYMBOL(dev_set_group); | |
8488 | ||
d59cdf94 PM |
8489 | /** |
8490 | * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. | |
8491 | * @dev: device | |
8492 | * @addr: new address | |
8493 | * @extack: netlink extended ack | |
8494 | */ | |
8495 | int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, | |
8496 | struct netlink_ext_ack *extack) | |
8497 | { | |
8498 | struct netdev_notifier_pre_changeaddr_info info = { | |
8499 | .info.dev = dev, | |
8500 | .info.extack = extack, | |
8501 | .dev_addr = addr, | |
8502 | }; | |
8503 | int rc; | |
8504 | ||
8505 | rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); | |
8506 | return notifier_to_errno(rc); | |
8507 | } | |
8508 | EXPORT_SYMBOL(dev_pre_changeaddr_notify); | |
8509 | ||
f0db275a SH |
8510 | /** |
8511 | * dev_set_mac_address - Change Media Access Control Address | |
8512 | * @dev: device | |
8513 | * @sa: new address | |
3a37a963 | 8514 | * @extack: netlink extended ack |
f0db275a SH |
8515 | * |
8516 | * Change the hardware (MAC) address of the device | |
8517 | */ | |
3a37a963 PM |
8518 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, |
8519 | struct netlink_ext_ack *extack) | |
1da177e4 | 8520 | { |
d314774c | 8521 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
8522 | int err; |
8523 | ||
d314774c | 8524 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
8525 | return -EOPNOTSUPP; |
8526 | if (sa->sa_family != dev->type) | |
8527 | return -EINVAL; | |
8528 | if (!netif_device_present(dev)) | |
8529 | return -ENODEV; | |
d59cdf94 PM |
8530 | err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); |
8531 | if (err) | |
8532 | return err; | |
d314774c | 8533 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
8534 | if (err) |
8535 | return err; | |
fbdeca2d | 8536 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 8537 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 8538 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 8539 | return 0; |
1da177e4 | 8540 | } |
d1b19dff | 8541 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 8542 | |
4bf84c35 JP |
8543 | /** |
8544 | * dev_change_carrier - Change device carrier | |
8545 | * @dev: device | |
691b3b7e | 8546 | * @new_carrier: new value |
4bf84c35 JP |
8547 | * |
8548 | * Change device carrier | |
8549 | */ | |
8550 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
8551 | { | |
8552 | const struct net_device_ops *ops = dev->netdev_ops; | |
8553 | ||
8554 | if (!ops->ndo_change_carrier) | |
8555 | return -EOPNOTSUPP; | |
8556 | if (!netif_device_present(dev)) | |
8557 | return -ENODEV; | |
8558 | return ops->ndo_change_carrier(dev, new_carrier); | |
8559 | } | |
8560 | EXPORT_SYMBOL(dev_change_carrier); | |
8561 | ||
66b52b0d JP |
8562 | /** |
8563 | * dev_get_phys_port_id - Get device physical port ID | |
8564 | * @dev: device | |
8565 | * @ppid: port ID | |
8566 | * | |
8567 | * Get device physical port ID | |
8568 | */ | |
8569 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 8570 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
8571 | { |
8572 | const struct net_device_ops *ops = dev->netdev_ops; | |
8573 | ||
8574 | if (!ops->ndo_get_phys_port_id) | |
8575 | return -EOPNOTSUPP; | |
8576 | return ops->ndo_get_phys_port_id(dev, ppid); | |
8577 | } | |
8578 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
8579 | ||
db24a904 DA |
8580 | /** |
8581 | * dev_get_phys_port_name - Get device physical port name | |
8582 | * @dev: device | |
8583 | * @name: port name | |
ed49e650 | 8584 | * @len: limit of bytes to copy to name |
db24a904 DA |
8585 | * |
8586 | * Get device physical port name | |
8587 | */ | |
8588 | int dev_get_phys_port_name(struct net_device *dev, | |
8589 | char *name, size_t len) | |
8590 | { | |
8591 | const struct net_device_ops *ops = dev->netdev_ops; | |
af3836df | 8592 | int err; |
db24a904 | 8593 | |
af3836df JP |
8594 | if (ops->ndo_get_phys_port_name) { |
8595 | err = ops->ndo_get_phys_port_name(dev, name, len); | |
8596 | if (err != -EOPNOTSUPP) | |
8597 | return err; | |
8598 | } | |
8599 | return devlink_compat_phys_port_name_get(dev, name, len); | |
db24a904 DA |
8600 | } |
8601 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
8602 | ||
d6abc596 FF |
8603 | /** |
8604 | * dev_get_port_parent_id - Get the device's port parent identifier | |
8605 | * @dev: network device | |
8606 | * @ppid: pointer to a storage for the port's parent identifier | |
8607 | * @recurse: allow/disallow recursion to lower devices | |
8608 | * | |
8609 | * Get the devices's port parent identifier | |
8610 | */ | |
8611 | int dev_get_port_parent_id(struct net_device *dev, | |
8612 | struct netdev_phys_item_id *ppid, | |
8613 | bool recurse) | |
8614 | { | |
8615 | const struct net_device_ops *ops = dev->netdev_ops; | |
8616 | struct netdev_phys_item_id first = { }; | |
8617 | struct net_device *lower_dev; | |
8618 | struct list_head *iter; | |
7e1146e8 JP |
8619 | int err; |
8620 | ||
8621 | if (ops->ndo_get_port_parent_id) { | |
8622 | err = ops->ndo_get_port_parent_id(dev, ppid); | |
8623 | if (err != -EOPNOTSUPP) | |
8624 | return err; | |
8625 | } | |
d6abc596 | 8626 | |
7e1146e8 JP |
8627 | err = devlink_compat_switch_id_get(dev, ppid); |
8628 | if (!err || err != -EOPNOTSUPP) | |
8629 | return err; | |
d6abc596 FF |
8630 | |
8631 | if (!recurse) | |
7e1146e8 | 8632 | return -EOPNOTSUPP; |
d6abc596 FF |
8633 | |
8634 | netdev_for_each_lower_dev(dev, lower_dev, iter) { | |
8635 | err = dev_get_port_parent_id(lower_dev, ppid, recurse); | |
8636 | if (err) | |
8637 | break; | |
8638 | if (!first.id_len) | |
8639 | first = *ppid; | |
8640 | else if (memcmp(&first, ppid, sizeof(*ppid))) | |
8641 | return -ENODATA; | |
8642 | } | |
8643 | ||
8644 | return err; | |
8645 | } | |
8646 | EXPORT_SYMBOL(dev_get_port_parent_id); | |
8647 | ||
8648 | /** | |
8649 | * netdev_port_same_parent_id - Indicate if two network devices have | |
8650 | * the same port parent identifier | |
8651 | * @a: first network device | |
8652 | * @b: second network device | |
8653 | */ | |
8654 | bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) | |
8655 | { | |
8656 | struct netdev_phys_item_id a_id = { }; | |
8657 | struct netdev_phys_item_id b_id = { }; | |
8658 | ||
8659 | if (dev_get_port_parent_id(a, &a_id, true) || | |
8660 | dev_get_port_parent_id(b, &b_id, true)) | |
8661 | return false; | |
8662 | ||
8663 | return netdev_phys_item_id_same(&a_id, &b_id); | |
8664 | } | |
8665 | EXPORT_SYMBOL(netdev_port_same_parent_id); | |
8666 | ||
d746d707 AK |
8667 | /** |
8668 | * dev_change_proto_down - update protocol port state information | |
8669 | * @dev: device | |
8670 | * @proto_down: new value | |
8671 | * | |
8672 | * This info can be used by switch drivers to set the phys state of the | |
8673 | * port. | |
8674 | */ | |
8675 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
8676 | { | |
8677 | const struct net_device_ops *ops = dev->netdev_ops; | |
8678 | ||
8679 | if (!ops->ndo_change_proto_down) | |
8680 | return -EOPNOTSUPP; | |
8681 | if (!netif_device_present(dev)) | |
8682 | return -ENODEV; | |
8683 | return ops->ndo_change_proto_down(dev, proto_down); | |
8684 | } | |
8685 | EXPORT_SYMBOL(dev_change_proto_down); | |
8686 | ||
b5899679 AR |
8687 | /** |
8688 | * dev_change_proto_down_generic - generic implementation for | |
8689 | * ndo_change_proto_down that sets carrier according to | |
8690 | * proto_down. | |
8691 | * | |
8692 | * @dev: device | |
8693 | * @proto_down: new value | |
8694 | */ | |
8695 | int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) | |
8696 | { | |
8697 | if (proto_down) | |
8698 | netif_carrier_off(dev); | |
8699 | else | |
8700 | netif_carrier_on(dev); | |
8701 | dev->proto_down = proto_down; | |
8702 | return 0; | |
8703 | } | |
8704 | EXPORT_SYMBOL(dev_change_proto_down_generic); | |
8705 | ||
a25717d2 JK |
8706 | u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op, |
8707 | enum bpf_netdev_command cmd) | |
d67b9cd2 | 8708 | { |
a25717d2 | 8709 | struct netdev_bpf xdp; |
d67b9cd2 | 8710 | |
a25717d2 JK |
8711 | if (!bpf_op) |
8712 | return 0; | |
118b4aa2 | 8713 | |
a25717d2 JK |
8714 | memset(&xdp, 0, sizeof(xdp)); |
8715 | xdp.command = cmd; | |
118b4aa2 | 8716 | |
a25717d2 JK |
8717 | /* Query must always succeed. */ |
8718 | WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG); | |
58038695 | 8719 | |
6b867589 | 8720 | return xdp.prog_id; |
d67b9cd2 DB |
8721 | } |
8722 | ||
f4e63525 | 8723 | static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op, |
32d60277 | 8724 | struct netlink_ext_ack *extack, u32 flags, |
d67b9cd2 DB |
8725 | struct bpf_prog *prog) |
8726 | { | |
7e6897f9 BT |
8727 | bool non_hw = !(flags & XDP_FLAGS_HW_MODE); |
8728 | struct bpf_prog *prev_prog = NULL; | |
f4e63525 | 8729 | struct netdev_bpf xdp; |
7e6897f9 BT |
8730 | int err; |
8731 | ||
8732 | if (non_hw) { | |
8733 | prev_prog = bpf_prog_by_id(__dev_xdp_query(dev, bpf_op, | |
8734 | XDP_QUERY_PROG)); | |
8735 | if (IS_ERR(prev_prog)) | |
8736 | prev_prog = NULL; | |
8737 | } | |
d67b9cd2 DB |
8738 | |
8739 | memset(&xdp, 0, sizeof(xdp)); | |
ee5d032f JK |
8740 | if (flags & XDP_FLAGS_HW_MODE) |
8741 | xdp.command = XDP_SETUP_PROG_HW; | |
8742 | else | |
8743 | xdp.command = XDP_SETUP_PROG; | |
d67b9cd2 | 8744 | xdp.extack = extack; |
32d60277 | 8745 | xdp.flags = flags; |
d67b9cd2 DB |
8746 | xdp.prog = prog; |
8747 | ||
7e6897f9 BT |
8748 | err = bpf_op(dev, &xdp); |
8749 | if (!err && non_hw) | |
8750 | bpf_prog_change_xdp(prev_prog, prog); | |
8751 | ||
8752 | if (prev_prog) | |
8753 | bpf_prog_put(prev_prog); | |
8754 | ||
8755 | return err; | |
d67b9cd2 DB |
8756 | } |
8757 | ||
bd0b2e7f JK |
8758 | static void dev_xdp_uninstall(struct net_device *dev) |
8759 | { | |
8760 | struct netdev_bpf xdp; | |
8761 | bpf_op_t ndo_bpf; | |
8762 | ||
8763 | /* Remove generic XDP */ | |
8764 | WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL)); | |
8765 | ||
8766 | /* Remove from the driver */ | |
8767 | ndo_bpf = dev->netdev_ops->ndo_bpf; | |
8768 | if (!ndo_bpf) | |
8769 | return; | |
8770 | ||
a25717d2 JK |
8771 | memset(&xdp, 0, sizeof(xdp)); |
8772 | xdp.command = XDP_QUERY_PROG; | |
8773 | WARN_ON(ndo_bpf(dev, &xdp)); | |
8774 | if (xdp.prog_id) | |
8775 | WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, | |
8776 | NULL)); | |
bd0b2e7f | 8777 | |
a25717d2 JK |
8778 | /* Remove HW offload */ |
8779 | memset(&xdp, 0, sizeof(xdp)); | |
8780 | xdp.command = XDP_QUERY_PROG_HW; | |
8781 | if (!ndo_bpf(dev, &xdp) && xdp.prog_id) | |
8782 | WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, | |
8783 | NULL)); | |
bd0b2e7f JK |
8784 | } |
8785 | ||
a7862b45 BB |
8786 | /** |
8787 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
8788 | * @dev: device | |
b5d60989 | 8789 | * @extack: netlink extended ack |
a7862b45 | 8790 | * @fd: new program fd or negative value to clear |
92234c8f | 8791 | * @expected_fd: old program fd that userspace expects to replace or clear |
85de8576 | 8792 | * @flags: xdp-related flags |
a7862b45 BB |
8793 | * |
8794 | * Set or clear a bpf program for a device | |
8795 | */ | |
ddf9f970 | 8796 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, |
92234c8f | 8797 | int fd, int expected_fd, u32 flags) |
a7862b45 BB |
8798 | { |
8799 | const struct net_device_ops *ops = dev->netdev_ops; | |
a25717d2 | 8800 | enum bpf_netdev_command query; |
92234c8f | 8801 | u32 prog_id, expected_id = 0; |
f4e63525 | 8802 | bpf_op_t bpf_op, bpf_chk; |
dfa74909 | 8803 | struct bpf_prog *prog; |
9ee963d6 | 8804 | bool offload; |
a7862b45 BB |
8805 | int err; |
8806 | ||
85de8576 DB |
8807 | ASSERT_RTNL(); |
8808 | ||
9ee963d6 JK |
8809 | offload = flags & XDP_FLAGS_HW_MODE; |
8810 | query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG; | |
a25717d2 | 8811 | |
f4e63525 | 8812 | bpf_op = bpf_chk = ops->ndo_bpf; |
01dde20c MF |
8813 | if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) { |
8814 | NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode"); | |
0489df9a | 8815 | return -EOPNOTSUPP; |
01dde20c | 8816 | } |
f4e63525 JK |
8817 | if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE)) |
8818 | bpf_op = generic_xdp_install; | |
8819 | if (bpf_op == bpf_chk) | |
8820 | bpf_chk = generic_xdp_install; | |
b5cdae32 | 8821 | |
92234c8f THJ |
8822 | prog_id = __dev_xdp_query(dev, bpf_op, query); |
8823 | if (flags & XDP_FLAGS_REPLACE) { | |
8824 | if (expected_fd >= 0) { | |
8825 | prog = bpf_prog_get_type_dev(expected_fd, | |
8826 | BPF_PROG_TYPE_XDP, | |
8827 | bpf_op == ops->ndo_bpf); | |
8828 | if (IS_ERR(prog)) | |
8829 | return PTR_ERR(prog); | |
8830 | expected_id = prog->aux->id; | |
8831 | bpf_prog_put(prog); | |
8832 | } | |
c14a9f63 | 8833 | |
92234c8f THJ |
8834 | if (prog_id != expected_id) { |
8835 | NL_SET_ERR_MSG(extack, "Active program does not match expected"); | |
8836 | return -EEXIST; | |
8837 | } | |
8838 | } | |
8839 | if (fd >= 0) { | |
9ee963d6 | 8840 | if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) { |
01dde20c | 8841 | NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time"); |
d67b9cd2 | 8842 | return -EEXIST; |
01dde20c | 8843 | } |
c14a9f63 | 8844 | |
c14a9f63 | 8845 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && prog_id) { |
01dde20c | 8846 | NL_SET_ERR_MSG(extack, "XDP program already attached"); |
d67b9cd2 | 8847 | return -EBUSY; |
01dde20c | 8848 | } |
85de8576 | 8849 | |
288b3de5 JK |
8850 | prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, |
8851 | bpf_op == ops->ndo_bpf); | |
a7862b45 BB |
8852 | if (IS_ERR(prog)) |
8853 | return PTR_ERR(prog); | |
441a3303 | 8854 | |
9ee963d6 | 8855 | if (!offload && bpf_prog_is_dev_bound(prog->aux)) { |
441a3303 JK |
8856 | NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported"); |
8857 | bpf_prog_put(prog); | |
8858 | return -EINVAL; | |
8859 | } | |
c14a9f63 | 8860 | |
fbee97fe DA |
8861 | if (prog->expected_attach_type == BPF_XDP_DEVMAP) { |
8862 | NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); | |
8863 | bpf_prog_put(prog); | |
8864 | return -EINVAL; | |
8865 | } | |
8866 | ||
aefc3e72 JK |
8867 | /* prog->aux->id may be 0 for orphaned device-bound progs */ |
8868 | if (prog->aux->id && prog->aux->id == prog_id) { | |
c14a9f63 MM |
8869 | bpf_prog_put(prog); |
8870 | return 0; | |
8871 | } | |
8872 | } else { | |
92234c8f | 8873 | if (!prog_id) |
c14a9f63 | 8874 | return 0; |
dfa74909 | 8875 | prog = NULL; |
a7862b45 BB |
8876 | } |
8877 | ||
f4e63525 | 8878 | err = dev_xdp_install(dev, bpf_op, extack, flags, prog); |
a7862b45 BB |
8879 | if (err < 0 && prog) |
8880 | bpf_prog_put(prog); | |
8881 | ||
8882 | return err; | |
8883 | } | |
a7862b45 | 8884 | |
1da177e4 LT |
8885 | /** |
8886 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 8887 | * @net: the applicable net namespace |
1da177e4 LT |
8888 | * |
8889 | * Returns a suitable unique value for a new device interface | |
8890 | * number. The caller must hold the rtnl semaphore or the | |
8891 | * dev_base_lock to be sure it remains unique. | |
8892 | */ | |
881d966b | 8893 | static int dev_new_index(struct net *net) |
1da177e4 | 8894 | { |
aa79e66e | 8895 | int ifindex = net->ifindex; |
f4563a75 | 8896 | |
1da177e4 LT |
8897 | for (;;) { |
8898 | if (++ifindex <= 0) | |
8899 | ifindex = 1; | |
881d966b | 8900 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 8901 | return net->ifindex = ifindex; |
1da177e4 LT |
8902 | } |
8903 | } | |
8904 | ||
1da177e4 | 8905 | /* Delayed registration/unregisteration */ |
3b5b34fd | 8906 | static LIST_HEAD(net_todo_list); |
200b916f | 8907 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 8908 | |
6f05f629 | 8909 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 8910 | { |
1da177e4 | 8911 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 8912 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
8913 | } |
8914 | ||
9b5e383c | 8915 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 8916 | { |
e93737b0 | 8917 | struct net_device *dev, *tmp; |
5cde2829 | 8918 | LIST_HEAD(close_head); |
9b5e383c | 8919 | |
93ee31f1 DL |
8920 | BUG_ON(dev_boot_phase); |
8921 | ASSERT_RTNL(); | |
8922 | ||
e93737b0 | 8923 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 8924 | /* Some devices call without registering |
e93737b0 KK |
8925 | * for initialization unwind. Remove those |
8926 | * devices and proceed with the remaining. | |
9b5e383c ED |
8927 | */ |
8928 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
8929 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
8930 | dev->name, dev); | |
93ee31f1 | 8931 | |
9b5e383c | 8932 | WARN_ON(1); |
e93737b0 KK |
8933 | list_del(&dev->unreg_list); |
8934 | continue; | |
9b5e383c | 8935 | } |
449f4544 | 8936 | dev->dismantle = true; |
9b5e383c | 8937 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 8938 | } |
93ee31f1 | 8939 | |
44345724 | 8940 | /* If device is running, close it first. */ |
5cde2829 EB |
8941 | list_for_each_entry(dev, head, unreg_list) |
8942 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 8943 | dev_close_many(&close_head, true); |
93ee31f1 | 8944 | |
44345724 | 8945 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
8946 | /* And unlink it from device chain. */ |
8947 | unlist_netdevice(dev); | |
93ee31f1 | 8948 | |
9b5e383c ED |
8949 | dev->reg_state = NETREG_UNREGISTERING; |
8950 | } | |
41852497 | 8951 | flush_all_backlogs(); |
93ee31f1 DL |
8952 | |
8953 | synchronize_net(); | |
8954 | ||
9b5e383c | 8955 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
8956 | struct sk_buff *skb = NULL; |
8957 | ||
9b5e383c ED |
8958 | /* Shutdown queueing discipline. */ |
8959 | dev_shutdown(dev); | |
93ee31f1 | 8960 | |
bd0b2e7f | 8961 | dev_xdp_uninstall(dev); |
93ee31f1 | 8962 | |
9b5e383c | 8963 | /* Notify protocols, that we are about to destroy |
eb13da1a | 8964 | * this device. They should clean all the things. |
8965 | */ | |
9b5e383c | 8966 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
93ee31f1 | 8967 | |
395eea6c MB |
8968 | if (!dev->rtnl_link_ops || |
8969 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
3d3ea5af | 8970 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, |
38e01b30 | 8971 | GFP_KERNEL, NULL, 0); |
395eea6c | 8972 | |
9b5e383c ED |
8973 | /* |
8974 | * Flush the unicast and multicast chains | |
8975 | */ | |
a748ee24 | 8976 | dev_uc_flush(dev); |
22bedad3 | 8977 | dev_mc_flush(dev); |
93ee31f1 | 8978 | |
36fbf1e5 | 8979 | netdev_name_node_alt_flush(dev); |
ff927412 JP |
8980 | netdev_name_node_free(dev->name_node); |
8981 | ||
9b5e383c ED |
8982 | if (dev->netdev_ops->ndo_uninit) |
8983 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 8984 | |
395eea6c MB |
8985 | if (skb) |
8986 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 8987 | |
9ff162a8 JP |
8988 | /* Notifier chain MUST detach us all upper devices. */ |
8989 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
0f524a80 | 8990 | WARN_ON(netdev_has_any_lower_dev(dev)); |
93ee31f1 | 8991 | |
9b5e383c ED |
8992 | /* Remove entries from kobject tree */ |
8993 | netdev_unregister_kobject(dev); | |
024e9679 AD |
8994 | #ifdef CONFIG_XPS |
8995 | /* Remove XPS queueing entries */ | |
8996 | netif_reset_xps_queues_gt(dev, 0); | |
8997 | #endif | |
9b5e383c | 8998 | } |
93ee31f1 | 8999 | |
850a545b | 9000 | synchronize_net(); |
395264d5 | 9001 | |
a5ee1551 | 9002 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
9003 | dev_put(dev); |
9004 | } | |
9005 | ||
9006 | static void rollback_registered(struct net_device *dev) | |
9007 | { | |
9008 | LIST_HEAD(single); | |
9009 | ||
9010 | list_add(&dev->unreg_list, &single); | |
9011 | rollback_registered_many(&single); | |
ceaaec98 | 9012 | list_del(&single); |
93ee31f1 DL |
9013 | } |
9014 | ||
fd867d51 JW |
9015 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
9016 | struct net_device *upper, netdev_features_t features) | |
9017 | { | |
9018 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
9019 | netdev_features_t feature; | |
5ba3f7d6 | 9020 | int feature_bit; |
fd867d51 | 9021 | |
3b89ea9c | 9022 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 9023 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
9024 | if (!(upper->wanted_features & feature) |
9025 | && (features & feature)) { | |
9026 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
9027 | &feature, upper->name); | |
9028 | features &= ~feature; | |
9029 | } | |
9030 | } | |
9031 | ||
9032 | return features; | |
9033 | } | |
9034 | ||
9035 | static void netdev_sync_lower_features(struct net_device *upper, | |
9036 | struct net_device *lower, netdev_features_t features) | |
9037 | { | |
9038 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
9039 | netdev_features_t feature; | |
5ba3f7d6 | 9040 | int feature_bit; |
fd867d51 | 9041 | |
3b89ea9c | 9042 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 9043 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
9044 | if (!(features & feature) && (lower->features & feature)) { |
9045 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
9046 | &feature, lower->name); | |
9047 | lower->wanted_features &= ~feature; | |
dd912306 | 9048 | __netdev_update_features(lower); |
fd867d51 JW |
9049 | |
9050 | if (unlikely(lower->features & feature)) | |
9051 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
9052 | &feature, lower->name); | |
dd912306 CW |
9053 | else |
9054 | netdev_features_change(lower); | |
fd867d51 JW |
9055 | } |
9056 | } | |
9057 | } | |
9058 | ||
c8f44aff MM |
9059 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
9060 | netdev_features_t features) | |
b63365a2 | 9061 | { |
57422dc5 MM |
9062 | /* Fix illegal checksum combinations */ |
9063 | if ((features & NETIF_F_HW_CSUM) && | |
9064 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 9065 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
9066 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
9067 | } | |
9068 | ||
b63365a2 | 9069 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 9070 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 9071 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 9072 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
9073 | } |
9074 | ||
ec5f0615 PS |
9075 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
9076 | !(features & NETIF_F_IP_CSUM)) { | |
9077 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
9078 | features &= ~NETIF_F_TSO; | |
9079 | features &= ~NETIF_F_TSO_ECN; | |
9080 | } | |
9081 | ||
9082 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
9083 | !(features & NETIF_F_IPV6_CSUM)) { | |
9084 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
9085 | features &= ~NETIF_F_TSO6; | |
9086 | } | |
9087 | ||
b1dc497b AD |
9088 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
9089 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
9090 | features &= ~NETIF_F_TSO_MANGLEID; | |
9091 | ||
31d8b9e0 BH |
9092 | /* TSO ECN requires that TSO is present as well. */ |
9093 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
9094 | features &= ~NETIF_F_TSO_ECN; | |
9095 | ||
212b573f MM |
9096 | /* Software GSO depends on SG. */ |
9097 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 9098 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
9099 | features &= ~NETIF_F_GSO; |
9100 | } | |
9101 | ||
802ab55a AD |
9102 | /* GSO partial features require GSO partial be set */ |
9103 | if ((features & dev->gso_partial_features) && | |
9104 | !(features & NETIF_F_GSO_PARTIAL)) { | |
9105 | netdev_dbg(dev, | |
9106 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
9107 | features &= ~dev->gso_partial_features; | |
9108 | } | |
9109 | ||
fb1f5f79 MC |
9110 | if (!(features & NETIF_F_RXCSUM)) { |
9111 | /* NETIF_F_GRO_HW implies doing RXCSUM since every packet | |
9112 | * successfully merged by hardware must also have the | |
9113 | * checksum verified by hardware. If the user does not | |
9114 | * want to enable RXCSUM, logically, we should disable GRO_HW. | |
9115 | */ | |
9116 | if (features & NETIF_F_GRO_HW) { | |
9117 | netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); | |
9118 | features &= ~NETIF_F_GRO_HW; | |
9119 | } | |
9120 | } | |
9121 | ||
de8d5ab2 GP |
9122 | /* LRO/HW-GRO features cannot be combined with RX-FCS */ |
9123 | if (features & NETIF_F_RXFCS) { | |
9124 | if (features & NETIF_F_LRO) { | |
9125 | netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); | |
9126 | features &= ~NETIF_F_LRO; | |
9127 | } | |
9128 | ||
9129 | if (features & NETIF_F_GRO_HW) { | |
9130 | netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); | |
9131 | features &= ~NETIF_F_GRO_HW; | |
9132 | } | |
e6c6a929 GP |
9133 | } |
9134 | ||
b63365a2 HX |
9135 | return features; |
9136 | } | |
b63365a2 | 9137 | |
6cb6a27c | 9138 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 9139 | { |
fd867d51 | 9140 | struct net_device *upper, *lower; |
c8f44aff | 9141 | netdev_features_t features; |
fd867d51 | 9142 | struct list_head *iter; |
e7868a85 | 9143 | int err = -1; |
5455c699 | 9144 | |
87267485 MM |
9145 | ASSERT_RTNL(); |
9146 | ||
5455c699 MM |
9147 | features = netdev_get_wanted_features(dev); |
9148 | ||
9149 | if (dev->netdev_ops->ndo_fix_features) | |
9150 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
9151 | ||
9152 | /* driver might be less strict about feature dependencies */ | |
9153 | features = netdev_fix_features(dev, features); | |
9154 | ||
fd867d51 JW |
9155 | /* some features can't be enabled if they're off an an upper device */ |
9156 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
9157 | features = netdev_sync_upper_features(dev, upper, features); | |
9158 | ||
5455c699 | 9159 | if (dev->features == features) |
e7868a85 | 9160 | goto sync_lower; |
5455c699 | 9161 | |
c8f44aff MM |
9162 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
9163 | &dev->features, &features); | |
5455c699 MM |
9164 | |
9165 | if (dev->netdev_ops->ndo_set_features) | |
9166 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
9167 | else |
9168 | err = 0; | |
5455c699 | 9169 | |
6cb6a27c | 9170 | if (unlikely(err < 0)) { |
5455c699 | 9171 | netdev_err(dev, |
c8f44aff MM |
9172 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
9173 | err, &features, &dev->features); | |
17b85d29 NA |
9174 | /* return non-0 since some features might have changed and |
9175 | * it's better to fire a spurious notification than miss it | |
9176 | */ | |
9177 | return -1; | |
6cb6a27c MM |
9178 | } |
9179 | ||
e7868a85 | 9180 | sync_lower: |
fd867d51 JW |
9181 | /* some features must be disabled on lower devices when disabled |
9182 | * on an upper device (think: bonding master or bridge) | |
9183 | */ | |
9184 | netdev_for_each_lower_dev(dev, lower, iter) | |
9185 | netdev_sync_lower_features(dev, lower, features); | |
9186 | ||
ae847f40 SD |
9187 | if (!err) { |
9188 | netdev_features_t diff = features ^ dev->features; | |
9189 | ||
9190 | if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
9191 | /* udp_tunnel_{get,drop}_rx_info both need | |
9192 | * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the | |
9193 | * device, or they won't do anything. | |
9194 | * Thus we need to update dev->features | |
9195 | * *before* calling udp_tunnel_get_rx_info, | |
9196 | * but *after* calling udp_tunnel_drop_rx_info. | |
9197 | */ | |
9198 | if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
9199 | dev->features = features; | |
9200 | udp_tunnel_get_rx_info(dev); | |
9201 | } else { | |
9202 | udp_tunnel_drop_rx_info(dev); | |
9203 | } | |
9204 | } | |
9205 | ||
9daae9bd GP |
9206 | if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { |
9207 | if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { | |
9208 | dev->features = features; | |
9209 | err |= vlan_get_rx_ctag_filter_info(dev); | |
9210 | } else { | |
9211 | vlan_drop_rx_ctag_filter_info(dev); | |
9212 | } | |
9213 | } | |
9214 | ||
9215 | if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { | |
9216 | if (features & NETIF_F_HW_VLAN_STAG_FILTER) { | |
9217 | dev->features = features; | |
9218 | err |= vlan_get_rx_stag_filter_info(dev); | |
9219 | } else { | |
9220 | vlan_drop_rx_stag_filter_info(dev); | |
9221 | } | |
9222 | } | |
9223 | ||
6cb6a27c | 9224 | dev->features = features; |
ae847f40 | 9225 | } |
6cb6a27c | 9226 | |
e7868a85 | 9227 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
9228 | } |
9229 | ||
afe12cc8 MM |
9230 | /** |
9231 | * netdev_update_features - recalculate device features | |
9232 | * @dev: the device to check | |
9233 | * | |
9234 | * Recalculate dev->features set and send notifications if it | |
9235 | * has changed. Should be called after driver or hardware dependent | |
9236 | * conditions might have changed that influence the features. | |
9237 | */ | |
6cb6a27c MM |
9238 | void netdev_update_features(struct net_device *dev) |
9239 | { | |
9240 | if (__netdev_update_features(dev)) | |
9241 | netdev_features_change(dev); | |
5455c699 MM |
9242 | } |
9243 | EXPORT_SYMBOL(netdev_update_features); | |
9244 | ||
afe12cc8 MM |
9245 | /** |
9246 | * netdev_change_features - recalculate device features | |
9247 | * @dev: the device to check | |
9248 | * | |
9249 | * Recalculate dev->features set and send notifications even | |
9250 | * if they have not changed. Should be called instead of | |
9251 | * netdev_update_features() if also dev->vlan_features might | |
9252 | * have changed to allow the changes to be propagated to stacked | |
9253 | * VLAN devices. | |
9254 | */ | |
9255 | void netdev_change_features(struct net_device *dev) | |
9256 | { | |
9257 | __netdev_update_features(dev); | |
9258 | netdev_features_change(dev); | |
9259 | } | |
9260 | EXPORT_SYMBOL(netdev_change_features); | |
9261 | ||
fc4a7489 PM |
9262 | /** |
9263 | * netif_stacked_transfer_operstate - transfer operstate | |
9264 | * @rootdev: the root or lower level device to transfer state from | |
9265 | * @dev: the device to transfer operstate to | |
9266 | * | |
9267 | * Transfer operational state from root to device. This is normally | |
9268 | * called when a stacking relationship exists between the root | |
9269 | * device and the device(a leaf device). | |
9270 | */ | |
9271 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
9272 | struct net_device *dev) | |
9273 | { | |
9274 | if (rootdev->operstate == IF_OPER_DORMANT) | |
9275 | netif_dormant_on(dev); | |
9276 | else | |
9277 | netif_dormant_off(dev); | |
9278 | ||
eec517cd AL |
9279 | if (rootdev->operstate == IF_OPER_TESTING) |
9280 | netif_testing_on(dev); | |
9281 | else | |
9282 | netif_testing_off(dev); | |
9283 | ||
0575c86b ZS |
9284 | if (netif_carrier_ok(rootdev)) |
9285 | netif_carrier_on(dev); | |
9286 | else | |
9287 | netif_carrier_off(dev); | |
fc4a7489 PM |
9288 | } |
9289 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
9290 | ||
1b4bf461 ED |
9291 | static int netif_alloc_rx_queues(struct net_device *dev) |
9292 | { | |
1b4bf461 | 9293 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 9294 | struct netdev_rx_queue *rx; |
10595902 | 9295 | size_t sz = count * sizeof(*rx); |
e817f856 | 9296 | int err = 0; |
1b4bf461 | 9297 | |
bd25fa7b | 9298 | BUG_ON(count < 1); |
1b4bf461 | 9299 | |
dcda9b04 | 9300 | rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
9301 | if (!rx) |
9302 | return -ENOMEM; | |
9303 | ||
bd25fa7b TH |
9304 | dev->_rx = rx; |
9305 | ||
e817f856 | 9306 | for (i = 0; i < count; i++) { |
fe822240 | 9307 | rx[i].dev = dev; |
e817f856 JDB |
9308 | |
9309 | /* XDP RX-queue setup */ | |
9310 | err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i); | |
9311 | if (err < 0) | |
9312 | goto err_rxq_info; | |
9313 | } | |
1b4bf461 | 9314 | return 0; |
e817f856 JDB |
9315 | |
9316 | err_rxq_info: | |
9317 | /* Rollback successful reg's and free other resources */ | |
9318 | while (i--) | |
9319 | xdp_rxq_info_unreg(&rx[i].xdp_rxq); | |
141b52a9 | 9320 | kvfree(dev->_rx); |
e817f856 JDB |
9321 | dev->_rx = NULL; |
9322 | return err; | |
9323 | } | |
9324 | ||
9325 | static void netif_free_rx_queues(struct net_device *dev) | |
9326 | { | |
9327 | unsigned int i, count = dev->num_rx_queues; | |
e817f856 JDB |
9328 | |
9329 | /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ | |
9330 | if (!dev->_rx) | |
9331 | return; | |
9332 | ||
e817f856 | 9333 | for (i = 0; i < count; i++) |
82aaff2f JK |
9334 | xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); |
9335 | ||
9336 | kvfree(dev->_rx); | |
1b4bf461 ED |
9337 | } |
9338 | ||
aa942104 CG |
9339 | static void netdev_init_one_queue(struct net_device *dev, |
9340 | struct netdev_queue *queue, void *_unused) | |
9341 | { | |
9342 | /* Initialize queue lock */ | |
9343 | spin_lock_init(&queue->_xmit_lock); | |
1a33e10e | 9344 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); |
aa942104 | 9345 | queue->xmit_lock_owner = -1; |
b236da69 | 9346 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 9347 | queue->dev = dev; |
114cf580 TH |
9348 | #ifdef CONFIG_BQL |
9349 | dql_init(&queue->dql, HZ); | |
9350 | #endif | |
aa942104 CG |
9351 | } |
9352 | ||
60877a32 ED |
9353 | static void netif_free_tx_queues(struct net_device *dev) |
9354 | { | |
4cb28970 | 9355 | kvfree(dev->_tx); |
60877a32 ED |
9356 | } |
9357 | ||
e6484930 TH |
9358 | static int netif_alloc_netdev_queues(struct net_device *dev) |
9359 | { | |
9360 | unsigned int count = dev->num_tx_queues; | |
9361 | struct netdev_queue *tx; | |
60877a32 | 9362 | size_t sz = count * sizeof(*tx); |
e6484930 | 9363 | |
d339727c ED |
9364 | if (count < 1 || count > 0xffff) |
9365 | return -EINVAL; | |
62b5942a | 9366 | |
dcda9b04 | 9367 | tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
9368 | if (!tx) |
9369 | return -ENOMEM; | |
9370 | ||
e6484930 | 9371 | dev->_tx = tx; |
1d24eb48 | 9372 | |
e6484930 TH |
9373 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
9374 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
9375 | |
9376 | return 0; | |
e6484930 TH |
9377 | } |
9378 | ||
a2029240 DV |
9379 | void netif_tx_stop_all_queues(struct net_device *dev) |
9380 | { | |
9381 | unsigned int i; | |
9382 | ||
9383 | for (i = 0; i < dev->num_tx_queues; i++) { | |
9384 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 9385 | |
a2029240 DV |
9386 | netif_tx_stop_queue(txq); |
9387 | } | |
9388 | } | |
9389 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
9390 | ||
1da177e4 LT |
9391 | /** |
9392 | * register_netdevice - register a network device | |
9393 | * @dev: device to register | |
9394 | * | |
9395 | * Take a completed network device structure and add it to the kernel | |
9396 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
9397 | * chain. 0 is returned on success. A negative errno code is returned | |
9398 | * on a failure to set up the device, or if the name is a duplicate. | |
9399 | * | |
9400 | * Callers must hold the rtnl semaphore. You may want | |
9401 | * register_netdev() instead of this. | |
9402 | * | |
9403 | * BUGS: | |
9404 | * The locking appears insufficient to guarantee two parallel registers | |
9405 | * will not get the same name. | |
9406 | */ | |
9407 | ||
9408 | int register_netdevice(struct net_device *dev) | |
9409 | { | |
1da177e4 | 9410 | int ret; |
d314774c | 9411 | struct net *net = dev_net(dev); |
1da177e4 | 9412 | |
e283de3a FF |
9413 | BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < |
9414 | NETDEV_FEATURE_COUNT); | |
1da177e4 LT |
9415 | BUG_ON(dev_boot_phase); |
9416 | ASSERT_RTNL(); | |
9417 | ||
b17a7c17 SH |
9418 | might_sleep(); |
9419 | ||
1da177e4 LT |
9420 | /* When net_device's are persistent, this will be fatal. */ |
9421 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 9422 | BUG_ON(!net); |
1da177e4 | 9423 | |
9000edb7 JK |
9424 | ret = ethtool_check_ops(dev->ethtool_ops); |
9425 | if (ret) | |
9426 | return ret; | |
9427 | ||
f1f28aa3 | 9428 | spin_lock_init(&dev->addr_list_lock); |
845e0ebb | 9429 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 9430 | |
828de4f6 | 9431 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
9432 | if (ret < 0) |
9433 | goto out; | |
9434 | ||
9077f052 | 9435 | ret = -ENOMEM; |
ff927412 JP |
9436 | dev->name_node = netdev_name_node_head_alloc(dev); |
9437 | if (!dev->name_node) | |
9438 | goto out; | |
9439 | ||
1da177e4 | 9440 | /* Init, if this function is available */ |
d314774c SH |
9441 | if (dev->netdev_ops->ndo_init) { |
9442 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
9443 | if (ret) { |
9444 | if (ret > 0) | |
9445 | ret = -EIO; | |
42c17fa6 | 9446 | goto err_free_name; |
1da177e4 LT |
9447 | } |
9448 | } | |
4ec93edb | 9449 | |
f646968f PM |
9450 | if (((dev->hw_features | dev->features) & |
9451 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
9452 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
9453 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
9454 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
9455 | ret = -EINVAL; | |
9456 | goto err_uninit; | |
9457 | } | |
9458 | ||
9c7dafbf PE |
9459 | ret = -EBUSY; |
9460 | if (!dev->ifindex) | |
9461 | dev->ifindex = dev_new_index(net); | |
9462 | else if (__dev_get_by_index(net, dev->ifindex)) | |
9463 | goto err_uninit; | |
9464 | ||
5455c699 MM |
9465 | /* Transfer changeable features to wanted_features and enable |
9466 | * software offloads (GSO and GRO). | |
9467 | */ | |
1a3c998f | 9468 | dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); |
14d1232f | 9469 | dev->features |= NETIF_F_SOFT_FEATURES; |
d764a122 SD |
9470 | |
9471 | if (dev->netdev_ops->ndo_udp_tunnel_add) { | |
9472 | dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
9473 | dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
9474 | } | |
9475 | ||
14d1232f | 9476 | dev->wanted_features = dev->features & dev->hw_features; |
1da177e4 | 9477 | |
cbc53e08 | 9478 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 9479 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 9480 | |
7f348a60 AD |
9481 | /* If IPv4 TCP segmentation offload is supported we should also |
9482 | * allow the device to enable segmenting the frame with the option | |
9483 | * of ignoring a static IP ID value. This doesn't enable the | |
9484 | * feature itself but allows the user to enable it later. | |
9485 | */ | |
cbc53e08 AD |
9486 | if (dev->hw_features & NETIF_F_TSO) |
9487 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
9488 | if (dev->vlan_features & NETIF_F_TSO) |
9489 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
9490 | if (dev->mpls_features & NETIF_F_TSO) | |
9491 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
9492 | if (dev->hw_enc_features & NETIF_F_TSO) | |
9493 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 9494 | |
1180e7d6 | 9495 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 9496 | */ |
1180e7d6 | 9497 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 9498 | |
ee579677 PS |
9499 | /* Make NETIF_F_SG inheritable to tunnel devices. |
9500 | */ | |
802ab55a | 9501 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 9502 | |
0d89d203 SH |
9503 | /* Make NETIF_F_SG inheritable to MPLS. |
9504 | */ | |
9505 | dev->mpls_features |= NETIF_F_SG; | |
9506 | ||
7ffbe3fd JB |
9507 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
9508 | ret = notifier_to_errno(ret); | |
9509 | if (ret) | |
9510 | goto err_uninit; | |
9511 | ||
8b41d188 | 9512 | ret = netdev_register_kobject(dev); |
cb626bf5 JH |
9513 | if (ret) { |
9514 | dev->reg_state = NETREG_UNREGISTERED; | |
7ce1b0ed | 9515 | goto err_uninit; |
cb626bf5 | 9516 | } |
b17a7c17 SH |
9517 | dev->reg_state = NETREG_REGISTERED; |
9518 | ||
6cb6a27c | 9519 | __netdev_update_features(dev); |
8e9b59b2 | 9520 | |
1da177e4 LT |
9521 | /* |
9522 | * Default initial state at registry is that the | |
9523 | * device is present. | |
9524 | */ | |
9525 | ||
9526 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
9527 | ||
8f4cccbb BH |
9528 | linkwatch_init_dev(dev); |
9529 | ||
1da177e4 | 9530 | dev_init_scheduler(dev); |
1da177e4 | 9531 | dev_hold(dev); |
ce286d32 | 9532 | list_netdevice(dev); |
7bf23575 | 9533 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 9534 | |
948b337e JP |
9535 | /* If the device has permanent device address, driver should |
9536 | * set dev_addr and also addr_assign_type should be set to | |
9537 | * NET_ADDR_PERM (default value). | |
9538 | */ | |
9539 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
9540 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
9541 | ||
1da177e4 | 9542 | /* Notify protocols, that a new device appeared. */ |
056925ab | 9543 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 9544 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
9545 | if (ret) { |
9546 | rollback_registered(dev); | |
10cc514f SAK |
9547 | rcu_barrier(); |
9548 | ||
93ee31f1 DL |
9549 | dev->reg_state = NETREG_UNREGISTERED; |
9550 | } | |
d90a909e EB |
9551 | /* |
9552 | * Prevent userspace races by waiting until the network | |
9553 | * device is fully setup before sending notifications. | |
9554 | */ | |
a2835763 PM |
9555 | if (!dev->rtnl_link_ops || |
9556 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 9557 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
9558 | |
9559 | out: | |
9560 | return ret; | |
7ce1b0ed HX |
9561 | |
9562 | err_uninit: | |
d314774c SH |
9563 | if (dev->netdev_ops->ndo_uninit) |
9564 | dev->netdev_ops->ndo_uninit(dev); | |
cf124db5 DM |
9565 | if (dev->priv_destructor) |
9566 | dev->priv_destructor(dev); | |
42c17fa6 DC |
9567 | err_free_name: |
9568 | netdev_name_node_free(dev->name_node); | |
7ce1b0ed | 9569 | goto out; |
1da177e4 | 9570 | } |
d1b19dff | 9571 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 9572 | |
937f1ba5 BH |
9573 | /** |
9574 | * init_dummy_netdev - init a dummy network device for NAPI | |
9575 | * @dev: device to init | |
9576 | * | |
9577 | * This takes a network device structure and initialize the minimum | |
9578 | * amount of fields so it can be used to schedule NAPI polls without | |
9579 | * registering a full blown interface. This is to be used by drivers | |
9580 | * that need to tie several hardware interfaces to a single NAPI | |
9581 | * poll scheduler due to HW limitations. | |
9582 | */ | |
9583 | int init_dummy_netdev(struct net_device *dev) | |
9584 | { | |
9585 | /* Clear everything. Note we don't initialize spinlocks | |
9586 | * are they aren't supposed to be taken by any of the | |
9587 | * NAPI code and this dummy netdev is supposed to be | |
9588 | * only ever used for NAPI polls | |
9589 | */ | |
9590 | memset(dev, 0, sizeof(struct net_device)); | |
9591 | ||
9592 | /* make sure we BUG if trying to hit standard | |
9593 | * register/unregister code path | |
9594 | */ | |
9595 | dev->reg_state = NETREG_DUMMY; | |
9596 | ||
937f1ba5 BH |
9597 | /* NAPI wants this */ |
9598 | INIT_LIST_HEAD(&dev->napi_list); | |
9599 | ||
9600 | /* a dummy interface is started by default */ | |
9601 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
9602 | set_bit(__LINK_STATE_START, &dev->state); | |
9603 | ||
35edfdc7 JE |
9604 | /* napi_busy_loop stats accounting wants this */ |
9605 | dev_net_set(dev, &init_net); | |
9606 | ||
29b4433d ED |
9607 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
9608 | * because users of this 'device' dont need to change | |
9609 | * its refcount. | |
9610 | */ | |
9611 | ||
937f1ba5 BH |
9612 | return 0; |
9613 | } | |
9614 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
9615 | ||
9616 | ||
1da177e4 LT |
9617 | /** |
9618 | * register_netdev - register a network device | |
9619 | * @dev: device to register | |
9620 | * | |
9621 | * Take a completed network device structure and add it to the kernel | |
9622 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
9623 | * chain. 0 is returned on success. A negative errno code is returned | |
9624 | * on a failure to set up the device, or if the name is a duplicate. | |
9625 | * | |
38b4da38 | 9626 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
9627 | * and expands the device name if you passed a format string to |
9628 | * alloc_netdev. | |
9629 | */ | |
9630 | int register_netdev(struct net_device *dev) | |
9631 | { | |
9632 | int err; | |
9633 | ||
b0f3debc KT |
9634 | if (rtnl_lock_killable()) |
9635 | return -EINTR; | |
1da177e4 | 9636 | err = register_netdevice(dev); |
1da177e4 LT |
9637 | rtnl_unlock(); |
9638 | return err; | |
9639 | } | |
9640 | EXPORT_SYMBOL(register_netdev); | |
9641 | ||
29b4433d ED |
9642 | int netdev_refcnt_read(const struct net_device *dev) |
9643 | { | |
9644 | int i, refcnt = 0; | |
9645 | ||
9646 | for_each_possible_cpu(i) | |
9647 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
9648 | return refcnt; | |
9649 | } | |
9650 | EXPORT_SYMBOL(netdev_refcnt_read); | |
9651 | ||
2c53040f | 9652 | /** |
1da177e4 | 9653 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 9654 | * @dev: target net_device |
1da177e4 LT |
9655 | * |
9656 | * This is called when unregistering network devices. | |
9657 | * | |
9658 | * Any protocol or device that holds a reference should register | |
9659 | * for netdevice notification, and cleanup and put back the | |
9660 | * reference if they receive an UNREGISTER event. | |
9661 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 9662 | * call dev_put. |
1da177e4 LT |
9663 | */ |
9664 | static void netdev_wait_allrefs(struct net_device *dev) | |
9665 | { | |
9666 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 9667 | int refcnt; |
1da177e4 | 9668 | |
e014debe ED |
9669 | linkwatch_forget_dev(dev); |
9670 | ||
1da177e4 | 9671 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
9672 | refcnt = netdev_refcnt_read(dev); |
9673 | ||
9674 | while (refcnt != 0) { | |
1da177e4 | 9675 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 9676 | rtnl_lock(); |
1da177e4 LT |
9677 | |
9678 | /* Rebroadcast unregister notification */ | |
056925ab | 9679 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 9680 | |
748e2d93 | 9681 | __rtnl_unlock(); |
0115e8e3 | 9682 | rcu_barrier(); |
748e2d93 ED |
9683 | rtnl_lock(); |
9684 | ||
1da177e4 LT |
9685 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
9686 | &dev->state)) { | |
9687 | /* We must not have linkwatch events | |
9688 | * pending on unregister. If this | |
9689 | * happens, we simply run the queue | |
9690 | * unscheduled, resulting in a noop | |
9691 | * for this device. | |
9692 | */ | |
9693 | linkwatch_run_queue(); | |
9694 | } | |
9695 | ||
6756ae4b | 9696 | __rtnl_unlock(); |
1da177e4 LT |
9697 | |
9698 | rebroadcast_time = jiffies; | |
9699 | } | |
9700 | ||
9701 | msleep(250); | |
9702 | ||
29b4433d ED |
9703 | refcnt = netdev_refcnt_read(dev); |
9704 | ||
d7c04b05 | 9705 | if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
9706 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
9707 | dev->name, refcnt); | |
1da177e4 LT |
9708 | warning_time = jiffies; |
9709 | } | |
9710 | } | |
9711 | } | |
9712 | ||
9713 | /* The sequence is: | |
9714 | * | |
9715 | * rtnl_lock(); | |
9716 | * ... | |
9717 | * register_netdevice(x1); | |
9718 | * register_netdevice(x2); | |
9719 | * ... | |
9720 | * unregister_netdevice(y1); | |
9721 | * unregister_netdevice(y2); | |
9722 | * ... | |
9723 | * rtnl_unlock(); | |
9724 | * free_netdev(y1); | |
9725 | * free_netdev(y2); | |
9726 | * | |
58ec3b4d | 9727 | * We are invoked by rtnl_unlock(). |
1da177e4 | 9728 | * This allows us to deal with problems: |
b17a7c17 | 9729 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
9730 | * without deadlocking with linkwatch via keventd. |
9731 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
9732 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
9733 | * |
9734 | * We must not return until all unregister events added during | |
9735 | * the interval the lock was held have been completed. | |
1da177e4 | 9736 | */ |
1da177e4 LT |
9737 | void netdev_run_todo(void) |
9738 | { | |
626ab0e6 | 9739 | struct list_head list; |
1da177e4 | 9740 | |
1da177e4 | 9741 | /* Snapshot list, allow later requests */ |
626ab0e6 | 9742 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
9743 | |
9744 | __rtnl_unlock(); | |
626ab0e6 | 9745 | |
0115e8e3 ED |
9746 | |
9747 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
9748 | if (!list_empty(&list)) |
9749 | rcu_barrier(); | |
9750 | ||
1da177e4 LT |
9751 | while (!list_empty(&list)) { |
9752 | struct net_device *dev | |
e5e26d75 | 9753 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
9754 | list_del(&dev->todo_list); |
9755 | ||
b17a7c17 | 9756 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 9757 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
9758 | dev->name, dev->reg_state); |
9759 | dump_stack(); | |
9760 | continue; | |
9761 | } | |
1da177e4 | 9762 | |
b17a7c17 | 9763 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 9764 | |
b17a7c17 | 9765 | netdev_wait_allrefs(dev); |
1da177e4 | 9766 | |
b17a7c17 | 9767 | /* paranoia */ |
29b4433d | 9768 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
9769 | BUG_ON(!list_empty(&dev->ptype_all)); |
9770 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
9771 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
9772 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
330c7272 | 9773 | #if IS_ENABLED(CONFIG_DECNET) |
547b792c | 9774 | WARN_ON(dev->dn_ptr); |
330c7272 | 9775 | #endif |
cf124db5 DM |
9776 | if (dev->priv_destructor) |
9777 | dev->priv_destructor(dev); | |
9778 | if (dev->needs_free_netdev) | |
9779 | free_netdev(dev); | |
9093bbb2 | 9780 | |
50624c93 EB |
9781 | /* Report a network device has been unregistered */ |
9782 | rtnl_lock(); | |
9783 | dev_net(dev)->dev_unreg_count--; | |
9784 | __rtnl_unlock(); | |
9785 | wake_up(&netdev_unregistering_wq); | |
9786 | ||
9093bbb2 SH |
9787 | /* Free network device */ |
9788 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 9789 | } |
1da177e4 LT |
9790 | } |
9791 | ||
9256645a JW |
9792 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
9793 | * all the same fields in the same order as net_device_stats, with only | |
9794 | * the type differing, but rtnl_link_stats64 may have additional fields | |
9795 | * at the end for newer counters. | |
3cfde79c | 9796 | */ |
77a1abf5 ED |
9797 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
9798 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
9799 | { |
9800 | #if BITS_PER_LONG == 64 | |
9256645a | 9801 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
9af9959e | 9802 | memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); |
9256645a JW |
9803 | /* zero out counters that only exist in rtnl_link_stats64 */ |
9804 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
9805 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 9806 | #else |
9256645a | 9807 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
9808 | const unsigned long *src = (const unsigned long *)netdev_stats; |
9809 | u64 *dst = (u64 *)stats64; | |
9810 | ||
9256645a | 9811 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
9812 | for (i = 0; i < n; i++) |
9813 | dst[i] = src[i]; | |
9256645a JW |
9814 | /* zero out counters that only exist in rtnl_link_stats64 */ |
9815 | memset((char *)stats64 + n * sizeof(u64), 0, | |
9816 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
9817 | #endif |
9818 | } | |
77a1abf5 | 9819 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 9820 | |
eeda3fd6 SH |
9821 | /** |
9822 | * dev_get_stats - get network device statistics | |
9823 | * @dev: device to get statistics from | |
28172739 | 9824 | * @storage: place to store stats |
eeda3fd6 | 9825 | * |
d7753516 BH |
9826 | * Get network statistics from device. Return @storage. |
9827 | * The device driver may provide its own method by setting | |
9828 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
9829 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 9830 | */ |
d7753516 BH |
9831 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
9832 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 9833 | { |
eeda3fd6 SH |
9834 | const struct net_device_ops *ops = dev->netdev_ops; |
9835 | ||
28172739 ED |
9836 | if (ops->ndo_get_stats64) { |
9837 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
9838 | ops->ndo_get_stats64(dev, storage); |
9839 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 9840 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
9841 | } else { |
9842 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 9843 | } |
6f64ec74 ED |
9844 | storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); |
9845 | storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); | |
9846 | storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); | |
28172739 | 9847 | return storage; |
c45d286e | 9848 | } |
eeda3fd6 | 9849 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 9850 | |
24824a09 | 9851 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 9852 | { |
24824a09 | 9853 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 9854 | |
24824a09 ED |
9855 | #ifdef CONFIG_NET_CLS_ACT |
9856 | if (queue) | |
9857 | return queue; | |
9858 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
9859 | if (!queue) | |
9860 | return NULL; | |
9861 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 9862 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
9863 | queue->qdisc_sleeping = &noop_qdisc; |
9864 | rcu_assign_pointer(dev->ingress_queue, queue); | |
9865 | #endif | |
9866 | return queue; | |
bb949fbd DM |
9867 | } |
9868 | ||
2c60db03 ED |
9869 | static const struct ethtool_ops default_ethtool_ops; |
9870 | ||
d07d7507 SG |
9871 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
9872 | const struct ethtool_ops *ops) | |
9873 | { | |
9874 | if (dev->ethtool_ops == &default_ethtool_ops) | |
9875 | dev->ethtool_ops = ops; | |
9876 | } | |
9877 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
9878 | ||
74d332c1 ED |
9879 | void netdev_freemem(struct net_device *dev) |
9880 | { | |
9881 | char *addr = (char *)dev - dev->padded; | |
9882 | ||
4cb28970 | 9883 | kvfree(addr); |
74d332c1 ED |
9884 | } |
9885 | ||
1da177e4 | 9886 | /** |
722c9a0c | 9887 | * alloc_netdev_mqs - allocate network device |
9888 | * @sizeof_priv: size of private data to allocate space for | |
9889 | * @name: device name format string | |
9890 | * @name_assign_type: origin of device name | |
9891 | * @setup: callback to initialize device | |
9892 | * @txqs: the number of TX subqueues to allocate | |
9893 | * @rxqs: the number of RX subqueues to allocate | |
9894 | * | |
9895 | * Allocates a struct net_device with private data area for driver use | |
9896 | * and performs basic initialization. Also allocates subqueue structs | |
9897 | * for each queue on the device. | |
1da177e4 | 9898 | */ |
36909ea4 | 9899 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 9900 | unsigned char name_assign_type, |
36909ea4 TH |
9901 | void (*setup)(struct net_device *), |
9902 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 9903 | { |
1da177e4 | 9904 | struct net_device *dev; |
52a59bd5 | 9905 | unsigned int alloc_size; |
1ce8e7b5 | 9906 | struct net_device *p; |
1da177e4 | 9907 | |
b6fe17d6 SH |
9908 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
9909 | ||
36909ea4 | 9910 | if (txqs < 1) { |
7b6cd1ce | 9911 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
9912 | return NULL; |
9913 | } | |
9914 | ||
36909ea4 | 9915 | if (rxqs < 1) { |
7b6cd1ce | 9916 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
9917 | return NULL; |
9918 | } | |
36909ea4 | 9919 | |
fd2ea0a7 | 9920 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
9921 | if (sizeof_priv) { |
9922 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 9923 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
9924 | alloc_size += sizeof_priv; |
9925 | } | |
9926 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 9927 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 9928 | |
dcda9b04 | 9929 | p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
62b5942a | 9930 | if (!p) |
1da177e4 | 9931 | return NULL; |
1da177e4 | 9932 | |
1ce8e7b5 | 9933 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 9934 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 9935 | |
29b4433d ED |
9936 | dev->pcpu_refcnt = alloc_percpu(int); |
9937 | if (!dev->pcpu_refcnt) | |
74d332c1 | 9938 | goto free_dev; |
ab9c73cc | 9939 | |
ab9c73cc | 9940 | if (dev_addr_init(dev)) |
29b4433d | 9941 | goto free_pcpu; |
ab9c73cc | 9942 | |
22bedad3 | 9943 | dev_mc_init(dev); |
a748ee24 | 9944 | dev_uc_init(dev); |
ccffad25 | 9945 | |
c346dca1 | 9946 | dev_net_set(dev, &init_net); |
1da177e4 | 9947 | |
8d3bdbd5 | 9948 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 9949 | dev->gso_max_segs = GSO_MAX_SEGS; |
5343da4c TY |
9950 | dev->upper_level = 1; |
9951 | dev->lower_level = 1; | |
8d3bdbd5 | 9952 | |
8d3bdbd5 DM |
9953 | INIT_LIST_HEAD(&dev->napi_list); |
9954 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 9955 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 9956 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
9957 | INIT_LIST_HEAD(&dev->adj_list.upper); |
9958 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
9959 | INIT_LIST_HEAD(&dev->ptype_all); |
9960 | INIT_LIST_HEAD(&dev->ptype_specific); | |
93642e14 | 9961 | INIT_LIST_HEAD(&dev->net_notifier_list); |
59cc1f61 JK |
9962 | #ifdef CONFIG_NET_SCHED |
9963 | hash_init(dev->qdisc_hash); | |
9964 | #endif | |
02875878 | 9965 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
9966 | setup(dev); |
9967 | ||
a813104d | 9968 | if (!dev->tx_queue_len) { |
f84bb1ea | 9969 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 9970 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 9971 | } |
906470c1 | 9972 | |
36909ea4 TH |
9973 | dev->num_tx_queues = txqs; |
9974 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 9975 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 9976 | goto free_all; |
e8a0464c | 9977 | |
36909ea4 TH |
9978 | dev->num_rx_queues = rxqs; |
9979 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 9980 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 9981 | goto free_all; |
0a9627f2 | 9982 | |
1da177e4 | 9983 | strcpy(dev->name, name); |
c835a677 | 9984 | dev->name_assign_type = name_assign_type; |
cbda10fa | 9985 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
9986 | if (!dev->ethtool_ops) |
9987 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 | 9988 | |
357b6cc5 | 9989 | nf_hook_ingress_init(dev); |
e687ad60 | 9990 | |
1da177e4 | 9991 | return dev; |
ab9c73cc | 9992 | |
8d3bdbd5 DM |
9993 | free_all: |
9994 | free_netdev(dev); | |
9995 | return NULL; | |
9996 | ||
29b4433d ED |
9997 | free_pcpu: |
9998 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
9999 | free_dev: |
10000 | netdev_freemem(dev); | |
ab9c73cc | 10001 | return NULL; |
1da177e4 | 10002 | } |
36909ea4 | 10003 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
10004 | |
10005 | /** | |
722c9a0c | 10006 | * free_netdev - free network device |
10007 | * @dev: device | |
1da177e4 | 10008 | * |
722c9a0c | 10009 | * This function does the last stage of destroying an allocated device |
10010 | * interface. The reference to the device object is released. If this | |
10011 | * is the last reference then it will be freed.Must be called in process | |
10012 | * context. | |
1da177e4 LT |
10013 | */ |
10014 | void free_netdev(struct net_device *dev) | |
10015 | { | |
d565b0a1 HX |
10016 | struct napi_struct *p, *n; |
10017 | ||
93d05d4a | 10018 | might_sleep(); |
60877a32 | 10019 | netif_free_tx_queues(dev); |
e817f856 | 10020 | netif_free_rx_queues(dev); |
e8a0464c | 10021 | |
33d480ce | 10022 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 10023 | |
f001fde5 JP |
10024 | /* Flush device addresses */ |
10025 | dev_addr_flush(dev); | |
10026 | ||
d565b0a1 HX |
10027 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
10028 | netif_napi_del(p); | |
10029 | ||
29b4433d ED |
10030 | free_percpu(dev->pcpu_refcnt); |
10031 | dev->pcpu_refcnt = NULL; | |
75ccae62 THJ |
10032 | free_percpu(dev->xdp_bulkq); |
10033 | dev->xdp_bulkq = NULL; | |
29b4433d | 10034 | |
3041a069 | 10035 | /* Compatibility with error handling in drivers */ |
1da177e4 | 10036 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 10037 | netdev_freemem(dev); |
1da177e4 LT |
10038 | return; |
10039 | } | |
10040 | ||
10041 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
10042 | dev->reg_state = NETREG_RELEASED; | |
10043 | ||
43cb76d9 GKH |
10044 | /* will free via device release */ |
10045 | put_device(&dev->dev); | |
1da177e4 | 10046 | } |
d1b19dff | 10047 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 10048 | |
f0db275a SH |
10049 | /** |
10050 | * synchronize_net - Synchronize with packet receive processing | |
10051 | * | |
10052 | * Wait for packets currently being received to be done. | |
10053 | * Does not block later packets from starting. | |
10054 | */ | |
4ec93edb | 10055 | void synchronize_net(void) |
1da177e4 LT |
10056 | { |
10057 | might_sleep(); | |
be3fc413 ED |
10058 | if (rtnl_is_locked()) |
10059 | synchronize_rcu_expedited(); | |
10060 | else | |
10061 | synchronize_rcu(); | |
1da177e4 | 10062 | } |
d1b19dff | 10063 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
10064 | |
10065 | /** | |
44a0873d | 10066 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 10067 | * @dev: device |
44a0873d | 10068 | * @head: list |
6ebfbc06 | 10069 | * |
1da177e4 | 10070 | * This function shuts down a device interface and removes it |
d59b54b1 | 10071 | * from the kernel tables. |
44a0873d | 10072 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
10073 | * |
10074 | * Callers must hold the rtnl semaphore. You may want | |
10075 | * unregister_netdev() instead of this. | |
10076 | */ | |
10077 | ||
44a0873d | 10078 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 10079 | { |
a6620712 HX |
10080 | ASSERT_RTNL(); |
10081 | ||
44a0873d | 10082 | if (head) { |
9fdce099 | 10083 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
10084 | } else { |
10085 | rollback_registered(dev); | |
10086 | /* Finish processing unregister after unlock */ | |
10087 | net_set_todo(dev); | |
10088 | } | |
1da177e4 | 10089 | } |
44a0873d | 10090 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 10091 | |
9b5e383c ED |
10092 | /** |
10093 | * unregister_netdevice_many - unregister many devices | |
10094 | * @head: list of devices | |
87757a91 ED |
10095 | * |
10096 | * Note: As most callers use a stack allocated list_head, | |
10097 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
10098 | */ |
10099 | void unregister_netdevice_many(struct list_head *head) | |
10100 | { | |
10101 | struct net_device *dev; | |
10102 | ||
10103 | if (!list_empty(head)) { | |
10104 | rollback_registered_many(head); | |
10105 | list_for_each_entry(dev, head, unreg_list) | |
10106 | net_set_todo(dev); | |
87757a91 | 10107 | list_del(head); |
9b5e383c ED |
10108 | } |
10109 | } | |
63c8099d | 10110 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 10111 | |
1da177e4 LT |
10112 | /** |
10113 | * unregister_netdev - remove device from the kernel | |
10114 | * @dev: device | |
10115 | * | |
10116 | * This function shuts down a device interface and removes it | |
d59b54b1 | 10117 | * from the kernel tables. |
1da177e4 LT |
10118 | * |
10119 | * This is just a wrapper for unregister_netdevice that takes | |
10120 | * the rtnl semaphore. In general you want to use this and not | |
10121 | * unregister_netdevice. | |
10122 | */ | |
10123 | void unregister_netdev(struct net_device *dev) | |
10124 | { | |
10125 | rtnl_lock(); | |
10126 | unregister_netdevice(dev); | |
10127 | rtnl_unlock(); | |
10128 | } | |
1da177e4 LT |
10129 | EXPORT_SYMBOL(unregister_netdev); |
10130 | ||
ce286d32 EB |
10131 | /** |
10132 | * dev_change_net_namespace - move device to different nethost namespace | |
10133 | * @dev: device | |
10134 | * @net: network namespace | |
10135 | * @pat: If not NULL name pattern to try if the current device name | |
10136 | * is already taken in the destination network namespace. | |
10137 | * | |
10138 | * This function shuts down a device interface and moves it | |
10139 | * to a new network namespace. On success 0 is returned, on | |
10140 | * a failure a netagive errno code is returned. | |
10141 | * | |
10142 | * Callers must hold the rtnl semaphore. | |
10143 | */ | |
10144 | ||
10145 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
10146 | { | |
ef6a4c88 | 10147 | struct net *net_old = dev_net(dev); |
38e01b30 | 10148 | int err, new_nsid, new_ifindex; |
ce286d32 EB |
10149 | |
10150 | ASSERT_RTNL(); | |
10151 | ||
10152 | /* Don't allow namespace local devices to be moved. */ | |
10153 | err = -EINVAL; | |
10154 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
10155 | goto out; | |
10156 | ||
10157 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
10158 | if (dev->reg_state != NETREG_REGISTERED) |
10159 | goto out; | |
10160 | ||
10161 | /* Get out if there is nothing todo */ | |
10162 | err = 0; | |
ef6a4c88 | 10163 | if (net_eq(net_old, net)) |
ce286d32 EB |
10164 | goto out; |
10165 | ||
10166 | /* Pick the destination device name, and ensure | |
10167 | * we can use it in the destination network namespace. | |
10168 | */ | |
10169 | err = -EEXIST; | |
d9031024 | 10170 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
10171 | /* We get here if we can't use the current device name */ |
10172 | if (!pat) | |
10173 | goto out; | |
7892bd08 LR |
10174 | err = dev_get_valid_name(net, dev, pat); |
10175 | if (err < 0) | |
ce286d32 EB |
10176 | goto out; |
10177 | } | |
10178 | ||
10179 | /* | |
10180 | * And now a mini version of register_netdevice unregister_netdevice. | |
10181 | */ | |
10182 | ||
10183 | /* If device is running close it first. */ | |
9b772652 | 10184 | dev_close(dev); |
ce286d32 EB |
10185 | |
10186 | /* And unlink it from device chain */ | |
ce286d32 EB |
10187 | unlist_netdevice(dev); |
10188 | ||
10189 | synchronize_net(); | |
10190 | ||
10191 | /* Shutdown queueing discipline. */ | |
10192 | dev_shutdown(dev); | |
10193 | ||
10194 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 10195 | * this device. They should clean all the things. |
10196 | * | |
10197 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
10198 | * This is wanted because this way 8021q and macvlan know | |
10199 | * the device is just moving and can keep their slaves up. | |
10200 | */ | |
ce286d32 | 10201 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 | 10202 | rcu_barrier(); |
38e01b30 | 10203 | |
d4e4fdf9 | 10204 | new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); |
38e01b30 ND |
10205 | /* If there is an ifindex conflict assign a new one */ |
10206 | if (__dev_get_by_index(net, dev->ifindex)) | |
10207 | new_ifindex = dev_new_index(net); | |
10208 | else | |
10209 | new_ifindex = dev->ifindex; | |
10210 | ||
10211 | rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, | |
10212 | new_ifindex); | |
ce286d32 EB |
10213 | |
10214 | /* | |
10215 | * Flush the unicast and multicast chains | |
10216 | */ | |
a748ee24 | 10217 | dev_uc_flush(dev); |
22bedad3 | 10218 | dev_mc_flush(dev); |
ce286d32 | 10219 | |
4e66ae2e SH |
10220 | /* Send a netdev-removed uevent to the old namespace */ |
10221 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 10222 | netdev_adjacent_del_links(dev); |
4e66ae2e | 10223 | |
93642e14 JP |
10224 | /* Move per-net netdevice notifiers that are following the netdevice */ |
10225 | move_netdevice_notifiers_dev_net(dev, net); | |
10226 | ||
ce286d32 | 10227 | /* Actually switch the network namespace */ |
c346dca1 | 10228 | dev_net_set(dev, net); |
38e01b30 | 10229 | dev->ifindex = new_ifindex; |
ce286d32 | 10230 | |
4e66ae2e SH |
10231 | /* Send a netdev-add uevent to the new namespace */ |
10232 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 10233 | netdev_adjacent_add_links(dev); |
4e66ae2e | 10234 | |
8b41d188 | 10235 | /* Fixup kobjects */ |
a1b3f594 | 10236 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 10237 | WARN_ON(err); |
ce286d32 | 10238 | |
ef6a4c88 CB |
10239 | /* Adapt owner in case owning user namespace of target network |
10240 | * namespace is different from the original one. | |
10241 | */ | |
10242 | err = netdev_change_owner(dev, net_old, net); | |
10243 | WARN_ON(err); | |
10244 | ||
ce286d32 EB |
10245 | /* Add the device back in the hashes */ |
10246 | list_netdevice(dev); | |
10247 | ||
10248 | /* Notify protocols, that a new device appeared. */ | |
10249 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
10250 | ||
d90a909e EB |
10251 | /* |
10252 | * Prevent userspace races by waiting until the network | |
10253 | * device is fully setup before sending notifications. | |
10254 | */ | |
7f294054 | 10255 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 10256 | |
ce286d32 EB |
10257 | synchronize_net(); |
10258 | err = 0; | |
10259 | out: | |
10260 | return err; | |
10261 | } | |
463d0183 | 10262 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 10263 | |
f0bf90de | 10264 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
10265 | { |
10266 | struct sk_buff **list_skb; | |
1da177e4 | 10267 | struct sk_buff *skb; |
f0bf90de | 10268 | unsigned int cpu; |
97d8b6e3 | 10269 | struct softnet_data *sd, *oldsd, *remsd = NULL; |
1da177e4 | 10270 | |
1da177e4 LT |
10271 | local_irq_disable(); |
10272 | cpu = smp_processor_id(); | |
10273 | sd = &per_cpu(softnet_data, cpu); | |
10274 | oldsd = &per_cpu(softnet_data, oldcpu); | |
10275 | ||
10276 | /* Find end of our completion_queue. */ | |
10277 | list_skb = &sd->completion_queue; | |
10278 | while (*list_skb) | |
10279 | list_skb = &(*list_skb)->next; | |
10280 | /* Append completion queue from offline CPU. */ | |
10281 | *list_skb = oldsd->completion_queue; | |
10282 | oldsd->completion_queue = NULL; | |
10283 | ||
1da177e4 | 10284 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
10285 | if (oldsd->output_queue) { |
10286 | *sd->output_queue_tailp = oldsd->output_queue; | |
10287 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
10288 | oldsd->output_queue = NULL; | |
10289 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
10290 | } | |
ac64da0b ED |
10291 | /* Append NAPI poll list from offline CPU, with one exception : |
10292 | * process_backlog() must be called by cpu owning percpu backlog. | |
10293 | * We properly handle process_queue & input_pkt_queue later. | |
10294 | */ | |
10295 | while (!list_empty(&oldsd->poll_list)) { | |
10296 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
10297 | struct napi_struct, | |
10298 | poll_list); | |
10299 | ||
10300 | list_del_init(&napi->poll_list); | |
10301 | if (napi->poll == process_backlog) | |
10302 | napi->state = 0; | |
10303 | else | |
10304 | ____napi_schedule(sd, napi); | |
264524d5 | 10305 | } |
1da177e4 LT |
10306 | |
10307 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
10308 | local_irq_enable(); | |
10309 | ||
773fc8f6 | 10310 | #ifdef CONFIG_RPS |
10311 | remsd = oldsd->rps_ipi_list; | |
10312 | oldsd->rps_ipi_list = NULL; | |
10313 | #endif | |
10314 | /* send out pending IPI's on offline CPU */ | |
10315 | net_rps_send_ipi(remsd); | |
10316 | ||
1da177e4 | 10317 | /* Process offline CPU's input_pkt_queue */ |
76cc8b13 | 10318 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 10319 | netif_rx_ni(skb); |
76cc8b13 | 10320 | input_queue_head_incr(oldsd); |
fec5e652 | 10321 | } |
ac64da0b | 10322 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 10323 | netif_rx_ni(skb); |
76cc8b13 TH |
10324 | input_queue_head_incr(oldsd); |
10325 | } | |
1da177e4 | 10326 | |
f0bf90de | 10327 | return 0; |
1da177e4 | 10328 | } |
1da177e4 | 10329 | |
7f353bf2 | 10330 | /** |
b63365a2 HX |
10331 | * netdev_increment_features - increment feature set by one |
10332 | * @all: current feature set | |
10333 | * @one: new feature set | |
10334 | * @mask: mask feature set | |
7f353bf2 HX |
10335 | * |
10336 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
10337 | * @one to the master device with current feature set @all. Will not |
10338 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 10339 | */ |
c8f44aff MM |
10340 | netdev_features_t netdev_increment_features(netdev_features_t all, |
10341 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 10342 | { |
c8cd0989 | 10343 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 10344 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 10345 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 10346 | |
a188222b | 10347 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 10348 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 10349 | |
1742f183 | 10350 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
10351 | if (all & NETIF_F_HW_CSUM) |
10352 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
10353 | |
10354 | return all; | |
10355 | } | |
b63365a2 | 10356 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 10357 | |
430f03cd | 10358 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
10359 | { |
10360 | int i; | |
10361 | struct hlist_head *hash; | |
10362 | ||
6da2ec56 | 10363 | hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); |
30d97d35 PE |
10364 | if (hash != NULL) |
10365 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
10366 | INIT_HLIST_HEAD(&hash[i]); | |
10367 | ||
10368 | return hash; | |
10369 | } | |
10370 | ||
881d966b | 10371 | /* Initialize per network namespace state */ |
4665079c | 10372 | static int __net_init netdev_init(struct net *net) |
881d966b | 10373 | { |
d9f37d01 | 10374 | BUILD_BUG_ON(GRO_HASH_BUCKETS > |
c593642c | 10375 | 8 * sizeof_field(struct napi_struct, gro_bitmask)); |
d9f37d01 | 10376 | |
734b6541 RM |
10377 | if (net != &init_net) |
10378 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 10379 | |
30d97d35 PE |
10380 | net->dev_name_head = netdev_create_hash(); |
10381 | if (net->dev_name_head == NULL) | |
10382 | goto err_name; | |
881d966b | 10383 | |
30d97d35 PE |
10384 | net->dev_index_head = netdev_create_hash(); |
10385 | if (net->dev_index_head == NULL) | |
10386 | goto err_idx; | |
881d966b | 10387 | |
a30c7b42 JP |
10388 | RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); |
10389 | ||
881d966b | 10390 | return 0; |
30d97d35 PE |
10391 | |
10392 | err_idx: | |
10393 | kfree(net->dev_name_head); | |
10394 | err_name: | |
10395 | return -ENOMEM; | |
881d966b EB |
10396 | } |
10397 | ||
f0db275a SH |
10398 | /** |
10399 | * netdev_drivername - network driver for the device | |
10400 | * @dev: network device | |
f0db275a SH |
10401 | * |
10402 | * Determine network driver for device. | |
10403 | */ | |
3019de12 | 10404 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 10405 | { |
cf04a4c7 SH |
10406 | const struct device_driver *driver; |
10407 | const struct device *parent; | |
3019de12 | 10408 | const char *empty = ""; |
6579e57b AV |
10409 | |
10410 | parent = dev->dev.parent; | |
6579e57b | 10411 | if (!parent) |
3019de12 | 10412 | return empty; |
6579e57b AV |
10413 | |
10414 | driver = parent->driver; | |
10415 | if (driver && driver->name) | |
3019de12 DM |
10416 | return driver->name; |
10417 | return empty; | |
6579e57b AV |
10418 | } |
10419 | ||
6ea754eb JP |
10420 | static void __netdev_printk(const char *level, const struct net_device *dev, |
10421 | struct va_format *vaf) | |
256df2f3 | 10422 | { |
b004ff49 | 10423 | if (dev && dev->dev.parent) { |
6ea754eb JP |
10424 | dev_printk_emit(level[1] - '0', |
10425 | dev->dev.parent, | |
10426 | "%s %s %s%s: %pV", | |
10427 | dev_driver_string(dev->dev.parent), | |
10428 | dev_name(dev->dev.parent), | |
10429 | netdev_name(dev), netdev_reg_state(dev), | |
10430 | vaf); | |
b004ff49 | 10431 | } else if (dev) { |
6ea754eb JP |
10432 | printk("%s%s%s: %pV", |
10433 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 10434 | } else { |
6ea754eb | 10435 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 10436 | } |
256df2f3 JP |
10437 | } |
10438 | ||
6ea754eb JP |
10439 | void netdev_printk(const char *level, const struct net_device *dev, |
10440 | const char *format, ...) | |
256df2f3 JP |
10441 | { |
10442 | struct va_format vaf; | |
10443 | va_list args; | |
256df2f3 JP |
10444 | |
10445 | va_start(args, format); | |
10446 | ||
10447 | vaf.fmt = format; | |
10448 | vaf.va = &args; | |
10449 | ||
6ea754eb | 10450 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 10451 | |
256df2f3 | 10452 | va_end(args); |
256df2f3 JP |
10453 | } |
10454 | EXPORT_SYMBOL(netdev_printk); | |
10455 | ||
10456 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 10457 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 10458 | { \ |
256df2f3 JP |
10459 | struct va_format vaf; \ |
10460 | va_list args; \ | |
10461 | \ | |
10462 | va_start(args, fmt); \ | |
10463 | \ | |
10464 | vaf.fmt = fmt; \ | |
10465 | vaf.va = &args; \ | |
10466 | \ | |
6ea754eb | 10467 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 10468 | \ |
256df2f3 | 10469 | va_end(args); \ |
256df2f3 JP |
10470 | } \ |
10471 | EXPORT_SYMBOL(func); | |
10472 | ||
10473 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
10474 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
10475 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
10476 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
10477 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
10478 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
10479 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
10480 | ||
4665079c | 10481 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
10482 | { |
10483 | kfree(net->dev_name_head); | |
10484 | kfree(net->dev_index_head); | |
ee21b18b VA |
10485 | if (net != &init_net) |
10486 | WARN_ON_ONCE(!list_empty(&net->dev_base_head)); | |
881d966b EB |
10487 | } |
10488 | ||
022cbae6 | 10489 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
10490 | .init = netdev_init, |
10491 | .exit = netdev_exit, | |
10492 | }; | |
10493 | ||
4665079c | 10494 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 10495 | { |
e008b5fc | 10496 | struct net_device *dev, *aux; |
ce286d32 | 10497 | /* |
e008b5fc | 10498 | * Push all migratable network devices back to the |
ce286d32 EB |
10499 | * initial network namespace |
10500 | */ | |
10501 | rtnl_lock(); | |
e008b5fc | 10502 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 10503 | int err; |
aca51397 | 10504 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
10505 | |
10506 | /* Ignore unmoveable devices (i.e. loopback) */ | |
10507 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
10508 | continue; | |
10509 | ||
e008b5fc EB |
10510 | /* Leave virtual devices for the generic cleanup */ |
10511 | if (dev->rtnl_link_ops) | |
10512 | continue; | |
d0c082ce | 10513 | |
25985edc | 10514 | /* Push remaining network devices to init_net */ |
aca51397 | 10515 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
55b40dbf JP |
10516 | if (__dev_get_by_name(&init_net, fb_name)) |
10517 | snprintf(fb_name, IFNAMSIZ, "dev%%d"); | |
aca51397 | 10518 | err = dev_change_net_namespace(dev, &init_net, fb_name); |
ce286d32 | 10519 | if (err) { |
7b6cd1ce JP |
10520 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
10521 | __func__, dev->name, err); | |
aca51397 | 10522 | BUG(); |
ce286d32 EB |
10523 | } |
10524 | } | |
10525 | rtnl_unlock(); | |
10526 | } | |
10527 | ||
50624c93 EB |
10528 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
10529 | { | |
10530 | /* Return with the rtnl_lock held when there are no network | |
10531 | * devices unregistering in any network namespace in net_list. | |
10532 | */ | |
10533 | struct net *net; | |
10534 | bool unregistering; | |
ff960a73 | 10535 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 10536 | |
ff960a73 | 10537 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 10538 | for (;;) { |
50624c93 EB |
10539 | unregistering = false; |
10540 | rtnl_lock(); | |
10541 | list_for_each_entry(net, net_list, exit_list) { | |
10542 | if (net->dev_unreg_count > 0) { | |
10543 | unregistering = true; | |
10544 | break; | |
10545 | } | |
10546 | } | |
10547 | if (!unregistering) | |
10548 | break; | |
10549 | __rtnl_unlock(); | |
ff960a73 PZ |
10550 | |
10551 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 10552 | } |
ff960a73 | 10553 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
10554 | } |
10555 | ||
04dc7f6b EB |
10556 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
10557 | { | |
10558 | /* At exit all network devices most be removed from a network | |
b595076a | 10559 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
10560 | * Do this across as many network namespaces as possible to |
10561 | * improve batching efficiency. | |
10562 | */ | |
10563 | struct net_device *dev; | |
10564 | struct net *net; | |
10565 | LIST_HEAD(dev_kill_list); | |
10566 | ||
50624c93 EB |
10567 | /* To prevent network device cleanup code from dereferencing |
10568 | * loopback devices or network devices that have been freed | |
10569 | * wait here for all pending unregistrations to complete, | |
10570 | * before unregistring the loopback device and allowing the | |
10571 | * network namespace be freed. | |
10572 | * | |
10573 | * The netdev todo list containing all network devices | |
10574 | * unregistrations that happen in default_device_exit_batch | |
10575 | * will run in the rtnl_unlock() at the end of | |
10576 | * default_device_exit_batch. | |
10577 | */ | |
10578 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
10579 | list_for_each_entry(net, net_list, exit_list) { |
10580 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 10581 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
10582 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
10583 | else | |
10584 | unregister_netdevice_queue(dev, &dev_kill_list); | |
10585 | } | |
10586 | } | |
10587 | unregister_netdevice_many(&dev_kill_list); | |
10588 | rtnl_unlock(); | |
10589 | } | |
10590 | ||
022cbae6 | 10591 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 10592 | .exit = default_device_exit, |
04dc7f6b | 10593 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
10594 | }; |
10595 | ||
1da177e4 LT |
10596 | /* |
10597 | * Initialize the DEV module. At boot time this walks the device list and | |
10598 | * unhooks any devices that fail to initialise (normally hardware not | |
10599 | * present) and leaves us with a valid list of present and active devices. | |
10600 | * | |
10601 | */ | |
10602 | ||
10603 | /* | |
10604 | * This is called single threaded during boot, so no need | |
10605 | * to take the rtnl semaphore. | |
10606 | */ | |
10607 | static int __init net_dev_init(void) | |
10608 | { | |
10609 | int i, rc = -ENOMEM; | |
10610 | ||
10611 | BUG_ON(!dev_boot_phase); | |
10612 | ||
1da177e4 LT |
10613 | if (dev_proc_init()) |
10614 | goto out; | |
10615 | ||
8b41d188 | 10616 | if (netdev_kobject_init()) |
1da177e4 LT |
10617 | goto out; |
10618 | ||
10619 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 10620 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
10621 | INIT_LIST_HEAD(&ptype_base[i]); |
10622 | ||
62532da9 VY |
10623 | INIT_LIST_HEAD(&offload_base); |
10624 | ||
881d966b EB |
10625 | if (register_pernet_subsys(&netdev_net_ops)) |
10626 | goto out; | |
1da177e4 LT |
10627 | |
10628 | /* | |
10629 | * Initialise the packet receive queues. | |
10630 | */ | |
10631 | ||
6f912042 | 10632 | for_each_possible_cpu(i) { |
41852497 | 10633 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 10634 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 10635 | |
41852497 ED |
10636 | INIT_WORK(flush, flush_backlog); |
10637 | ||
e36fa2f7 | 10638 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 10639 | skb_queue_head_init(&sd->process_queue); |
f53c7239 SK |
10640 | #ifdef CONFIG_XFRM_OFFLOAD |
10641 | skb_queue_head_init(&sd->xfrm_backlog); | |
10642 | #endif | |
e36fa2f7 | 10643 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 10644 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 10645 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
10646 | sd->csd.func = rps_trigger_softirq; |
10647 | sd->csd.info = sd; | |
e36fa2f7 | 10648 | sd->cpu = i; |
1e94d72f | 10649 | #endif |
0a9627f2 | 10650 | |
7c4ec749 | 10651 | init_gro_hash(&sd->backlog); |
e36fa2f7 ED |
10652 | sd->backlog.poll = process_backlog; |
10653 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
10654 | } |
10655 | ||
1da177e4 LT |
10656 | dev_boot_phase = 0; |
10657 | ||
505d4f73 EB |
10658 | /* The loopback device is special if any other network devices |
10659 | * is present in a network namespace the loopback device must | |
10660 | * be present. Since we now dynamically allocate and free the | |
10661 | * loopback device ensure this invariant is maintained by | |
10662 | * keeping the loopback device as the first device on the | |
10663 | * list of network devices. Ensuring the loopback devices | |
10664 | * is the first device that appears and the last network device | |
10665 | * that disappears. | |
10666 | */ | |
10667 | if (register_pernet_device(&loopback_net_ops)) | |
10668 | goto out; | |
10669 | ||
10670 | if (register_pernet_device(&default_device_ops)) | |
10671 | goto out; | |
10672 | ||
962cf36c CM |
10673 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
10674 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 10675 | |
f0bf90de SAS |
10676 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
10677 | NULL, dev_cpu_dead); | |
10678 | WARN_ON(rc < 0); | |
1da177e4 LT |
10679 | rc = 0; |
10680 | out: | |
10681 | return rc; | |
10682 | } | |
10683 | ||
10684 | subsys_initcall(net_dev_init); |