]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * NET3 Protocol independent device support routines. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Derived from the non IP parts of dev.c 1.0.19 | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * | |
14 | * Additional Authors: | |
15 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
16 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
17 | * David Hinds <dahinds@users.sourceforge.net> | |
18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
19 | * Adam Sulmicki <adam@cfar.umd.edu> | |
20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
21 | * | |
22 | * Changes: | |
23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
24 | * to 2 if register_netdev gets called | |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
27 | * Alan Cox : device private ioctl copies fields back. | |
28 | * Alan Cox : Transmit queue code does relevant | |
29 | * stunts to keep the queue safe. | |
30 | * Alan Cox : Fixed double lock. | |
31 | * Alan Cox : Fixed promisc NULL pointer trap | |
32 | * ???????? : Support the full private ioctl range | |
33 | * Alan Cox : Moved ioctl permission check into | |
34 | * drivers | |
35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
36 | * Alan Cox : 100 backlog just doesn't cut it when | |
37 | * you start doing multicast video 8) | |
38 | * Alan Cox : Rewrote net_bh and list manager. | |
39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. | |
40 | * Alan Cox : Took out transmit every packet pass | |
41 | * Saved a few bytes in the ioctl handler | |
42 | * Alan Cox : Network driver sets packet type before | |
43 | * calling netif_rx. Saves a function | |
44 | * call a packet. | |
45 | * Alan Cox : Hashed net_bh() | |
46 | * Richard Kooijman: Timestamp fixes. | |
47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
48 | * Alan Cox : Device lock protection. | |
49 | * Alan Cox : Fixed nasty side effect of device close | |
50 | * changes. | |
51 | * Rudi Cilibrasi : Pass the right thing to | |
52 | * set_mac_address() | |
53 | * Dave Miller : 32bit quantity for the device lock to | |
54 | * make it work out on a Sparc. | |
55 | * Bjorn Ekwall : Added KERNELD hack. | |
56 | * Alan Cox : Cleaned up the backlog initialise. | |
57 | * Craig Metz : SIOCGIFCONF fix if space for under | |
58 | * 1 device. | |
59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
60 | * is no device open function. | |
61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
63 | * Cyrus Durgin : Cleaned for KMOD | |
64 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
65 | * A network device unload needs to purge | |
66 | * the backlog queue. | |
67 | * Paul Rusty Russell : SIOCSIFNAME | |
68 | * Pekka Riikonen : Netdev boot-time settings code | |
69 | * Andrew Morton : Make unregister_netdevice wait | |
70 | * indefinitely on dev->refcnt | |
71 | * J Hadi Salim : - Backlog queue sampling | |
72 | * - netif_rx() feedback | |
73 | */ | |
74 | ||
75 | #include <asm/uaccess.h> | |
1da177e4 | 76 | #include <linux/bitops.h> |
4fc268d2 | 77 | #include <linux/capability.h> |
1da177e4 LT |
78 | #include <linux/cpu.h> |
79 | #include <linux/types.h> | |
80 | #include <linux/kernel.h> | |
08e9897d | 81 | #include <linux/hash.h> |
5a0e3ad6 | 82 | #include <linux/slab.h> |
1da177e4 | 83 | #include <linux/sched.h> |
4a3e2f71 | 84 | #include <linux/mutex.h> |
1da177e4 LT |
85 | #include <linux/string.h> |
86 | #include <linux/mm.h> | |
87 | #include <linux/socket.h> | |
88 | #include <linux/sockios.h> | |
89 | #include <linux/errno.h> | |
90 | #include <linux/interrupt.h> | |
91 | #include <linux/if_ether.h> | |
92 | #include <linux/netdevice.h> | |
93 | #include <linux/etherdevice.h> | |
0187bdfb | 94 | #include <linux/ethtool.h> |
1da177e4 LT |
95 | #include <linux/notifier.h> |
96 | #include <linux/skbuff.h> | |
457c4cbc | 97 | #include <net/net_namespace.h> |
1da177e4 LT |
98 | #include <net/sock.h> |
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 LT |
103 | #include <net/pkt_sched.h> |
104 | #include <net/checksum.h> | |
44540960 | 105 | #include <net/xfrm.h> |
1da177e4 LT |
106 | #include <linux/highmem.h> |
107 | #include <linux/init.h> | |
1da177e4 | 108 | #include <linux/module.h> |
1da177e4 LT |
109 | #include <linux/netpoll.h> |
110 | #include <linux/rcupdate.h> | |
111 | #include <linux/delay.h> | |
1da177e4 | 112 | #include <net/iw_handler.h> |
1da177e4 | 113 | #include <asm/current.h> |
5bdb9886 | 114 | #include <linux/audit.h> |
db217334 | 115 | #include <linux/dmaengine.h> |
f6a78bfc | 116 | #include <linux/err.h> |
c7fa9d18 | 117 | #include <linux/ctype.h> |
723e98b7 | 118 | #include <linux/if_arp.h> |
6de329e2 | 119 | #include <linux/if_vlan.h> |
8f0f2223 | 120 | #include <linux/ip.h> |
ad55dcaf | 121 | #include <net/ip.h> |
25cd9ba0 | 122 | #include <net/mpls.h> |
8f0f2223 DM |
123 | #include <linux/ipv6.h> |
124 | #include <linux/in.h> | |
b6b2fed1 DM |
125 | #include <linux/jhash.h> |
126 | #include <linux/random.h> | |
9cbc1cb8 | 127 | #include <trace/events/napi.h> |
cf66ba58 | 128 | #include <trace/events/net.h> |
07dc22e7 | 129 | #include <trace/events/skb.h> |
5acbbd42 | 130 | #include <linux/pci.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> |
e687ad60 | 139 | #include <linux/netfilter_ingress.h> |
1da177e4 | 140 | |
342709ef PE |
141 | #include "net-sysfs.h" |
142 | ||
d565b0a1 HX |
143 | /* Instead of increasing this, you should create a hash table. */ |
144 | #define MAX_GRO_SKBS 8 | |
145 | ||
5d38a079 HX |
146 | /* This should be increased if a protocol with a bigger head is added. */ |
147 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
148 | ||
1da177e4 | 149 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 150 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
151 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
152 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 153 | static struct list_head offload_base __read_mostly; |
1da177e4 | 154 | |
ae78dbfa | 155 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
156 | static int call_netdevice_notifiers_info(unsigned long val, |
157 | struct net_device *dev, | |
158 | struct netdev_notifier_info *info); | |
ae78dbfa | 159 | |
1da177e4 | 160 | /* |
7562f876 | 161 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
162 | * semaphore. |
163 | * | |
c6d14c84 | 164 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
165 | * |
166 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 167 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
168 | * actual updates. This allows pure readers to access the list even |
169 | * while a writer is preparing to update it. | |
170 | * | |
171 | * To put it another way, dev_base_lock is held for writing only to | |
172 | * protect against pure readers; the rtnl semaphore provides the | |
173 | * protection against other writers. | |
174 | * | |
175 | * See, for example usages, register_netdevice() and | |
176 | * unregister_netdevice(), which must be called with the rtnl | |
177 | * semaphore held. | |
178 | */ | |
1da177e4 | 179 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
180 | EXPORT_SYMBOL(dev_base_lock); |
181 | ||
af12fa6e ET |
182 | /* protects napi_hash addition/deletion and napi_gen_id */ |
183 | static DEFINE_SPINLOCK(napi_hash_lock); | |
184 | ||
185 | static unsigned int napi_gen_id; | |
186 | static DEFINE_HASHTABLE(napi_hash, 8); | |
187 | ||
18afa4b0 | 188 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 189 | |
4e985ada TG |
190 | static inline void dev_base_seq_inc(struct net *net) |
191 | { | |
192 | while (++net->dev_base_seq == 0); | |
193 | } | |
194 | ||
881d966b | 195 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 196 | { |
95c96174 ED |
197 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
198 | ||
08e9897d | 199 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
200 | } |
201 | ||
881d966b | 202 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 203 | { |
7c28bd0b | 204 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
205 | } |
206 | ||
e36fa2f7 | 207 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
208 | { |
209 | #ifdef CONFIG_RPS | |
e36fa2f7 | 210 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
211 | #endif |
212 | } | |
213 | ||
e36fa2f7 | 214 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
215 | { |
216 | #ifdef CONFIG_RPS | |
e36fa2f7 | 217 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
218 | #endif |
219 | } | |
220 | ||
ce286d32 | 221 | /* Device list insertion */ |
53759be9 | 222 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 223 | { |
c346dca1 | 224 | struct net *net = dev_net(dev); |
ce286d32 EB |
225 | |
226 | ASSERT_RTNL(); | |
227 | ||
228 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 229 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 230 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
231 | hlist_add_head_rcu(&dev->index_hlist, |
232 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 233 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
234 | |
235 | dev_base_seq_inc(net); | |
ce286d32 EB |
236 | } |
237 | ||
fb699dfd ED |
238 | /* Device list removal |
239 | * caller must respect a RCU grace period before freeing/reusing dev | |
240 | */ | |
ce286d32 EB |
241 | static void unlist_netdevice(struct net_device *dev) |
242 | { | |
243 | ASSERT_RTNL(); | |
244 | ||
245 | /* Unlink dev from the device chain */ | |
246 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 247 | list_del_rcu(&dev->dev_list); |
72c9528b | 248 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 249 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 250 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
251 | |
252 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
253 | } |
254 | ||
1da177e4 LT |
255 | /* |
256 | * Our notifier list | |
257 | */ | |
258 | ||
f07d5b94 | 259 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
260 | |
261 | /* | |
262 | * Device drivers call our routines to queue packets here. We empty the | |
263 | * queue in the local softnet handler. | |
264 | */ | |
bea3348e | 265 | |
9958da05 | 266 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 267 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 268 | |
cf508b12 | 269 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 270 | /* |
c773e847 | 271 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
272 | * according to dev->type |
273 | */ | |
274 | static const unsigned short netdev_lock_type[] = | |
275 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
276 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
277 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
278 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
279 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
280 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
281 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
282 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
283 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
284 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
285 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
286 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
287 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
288 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
289 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 290 | |
36cbd3dc | 291 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
292 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
293 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
294 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
295 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
296 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
297 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
298 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
299 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
300 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
301 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
302 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
303 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
304 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
305 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
306 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
307 | |
308 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 309 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
310 | |
311 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
312 | { | |
313 | int i; | |
314 | ||
315 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
316 | if (netdev_lock_type[i] == dev_type) | |
317 | return i; | |
318 | /* the last key is used by default */ | |
319 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
320 | } | |
321 | ||
cf508b12 DM |
322 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
323 | unsigned short dev_type) | |
723e98b7 JP |
324 | { |
325 | int i; | |
326 | ||
327 | i = netdev_lock_pos(dev_type); | |
328 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
329 | netdev_lock_name[i]); | |
330 | } | |
cf508b12 DM |
331 | |
332 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
333 | { | |
334 | int i; | |
335 | ||
336 | i = netdev_lock_pos(dev->type); | |
337 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
338 | &netdev_addr_lock_key[i], | |
339 | netdev_lock_name[i]); | |
340 | } | |
723e98b7 | 341 | #else |
cf508b12 DM |
342 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
343 | unsigned short dev_type) | |
344 | { | |
345 | } | |
346 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
347 | { |
348 | } | |
349 | #endif | |
1da177e4 LT |
350 | |
351 | /******************************************************************************* | |
352 | ||
353 | Protocol management and registration routines | |
354 | ||
355 | *******************************************************************************/ | |
356 | ||
1da177e4 LT |
357 | /* |
358 | * Add a protocol ID to the list. Now that the input handler is | |
359 | * smarter we can dispense with all the messy stuff that used to be | |
360 | * here. | |
361 | * | |
362 | * BEWARE!!! Protocol handlers, mangling input packets, | |
363 | * MUST BE last in hash buckets and checking protocol handlers | |
364 | * MUST start from promiscuous ptype_all chain in net_bh. | |
365 | * It is true now, do not change it. | |
366 | * Explanation follows: if protocol handler, mangling packet, will | |
367 | * be the first on list, it is not able to sense, that packet | |
368 | * is cloned and should be copied-on-write, so that it will | |
369 | * change it and subsequent readers will get broken packet. | |
370 | * --ANK (980803) | |
371 | */ | |
372 | ||
c07b68e8 ED |
373 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
374 | { | |
375 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 376 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 377 | else |
7866a621 SN |
378 | return pt->dev ? &pt->dev->ptype_specific : |
379 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
380 | } |
381 | ||
1da177e4 LT |
382 | /** |
383 | * dev_add_pack - add packet handler | |
384 | * @pt: packet type declaration | |
385 | * | |
386 | * Add a protocol handler to the networking stack. The passed &packet_type | |
387 | * is linked into kernel lists and may not be freed until it has been | |
388 | * removed from the kernel lists. | |
389 | * | |
4ec93edb | 390 | * This call does not sleep therefore it can not |
1da177e4 LT |
391 | * guarantee all CPU's that are in middle of receiving packets |
392 | * will see the new packet type (until the next received packet). | |
393 | */ | |
394 | ||
395 | void dev_add_pack(struct packet_type *pt) | |
396 | { | |
c07b68e8 | 397 | struct list_head *head = ptype_head(pt); |
1da177e4 | 398 | |
c07b68e8 ED |
399 | spin_lock(&ptype_lock); |
400 | list_add_rcu(&pt->list, head); | |
401 | spin_unlock(&ptype_lock); | |
1da177e4 | 402 | } |
d1b19dff | 403 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 404 | |
1da177e4 LT |
405 | /** |
406 | * __dev_remove_pack - remove packet handler | |
407 | * @pt: packet type declaration | |
408 | * | |
409 | * Remove a protocol handler that was previously added to the kernel | |
410 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
411 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 412 | * returns. |
1da177e4 LT |
413 | * |
414 | * The packet type might still be in use by receivers | |
415 | * and must not be freed until after all the CPU's have gone | |
416 | * through a quiescent state. | |
417 | */ | |
418 | void __dev_remove_pack(struct packet_type *pt) | |
419 | { | |
c07b68e8 | 420 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
421 | struct packet_type *pt1; |
422 | ||
c07b68e8 | 423 | spin_lock(&ptype_lock); |
1da177e4 LT |
424 | |
425 | list_for_each_entry(pt1, head, list) { | |
426 | if (pt == pt1) { | |
427 | list_del_rcu(&pt->list); | |
428 | goto out; | |
429 | } | |
430 | } | |
431 | ||
7b6cd1ce | 432 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 433 | out: |
c07b68e8 | 434 | spin_unlock(&ptype_lock); |
1da177e4 | 435 | } |
d1b19dff ED |
436 | EXPORT_SYMBOL(__dev_remove_pack); |
437 | ||
1da177e4 LT |
438 | /** |
439 | * dev_remove_pack - remove packet handler | |
440 | * @pt: packet type declaration | |
441 | * | |
442 | * Remove a protocol handler that was previously added to the kernel | |
443 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
444 | * from the kernel lists and can be freed or reused once this function | |
445 | * returns. | |
446 | * | |
447 | * This call sleeps to guarantee that no CPU is looking at the packet | |
448 | * type after return. | |
449 | */ | |
450 | void dev_remove_pack(struct packet_type *pt) | |
451 | { | |
452 | __dev_remove_pack(pt); | |
4ec93edb | 453 | |
1da177e4 LT |
454 | synchronize_net(); |
455 | } | |
d1b19dff | 456 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 457 | |
62532da9 VY |
458 | |
459 | /** | |
460 | * dev_add_offload - register offload handlers | |
461 | * @po: protocol offload declaration | |
462 | * | |
463 | * Add protocol offload handlers to the networking stack. The passed | |
464 | * &proto_offload is linked into kernel lists and may not be freed until | |
465 | * it has been removed from the kernel lists. | |
466 | * | |
467 | * This call does not sleep therefore it can not | |
468 | * guarantee all CPU's that are in middle of receiving packets | |
469 | * will see the new offload handlers (until the next received packet). | |
470 | */ | |
471 | void dev_add_offload(struct packet_offload *po) | |
472 | { | |
bdef7de4 | 473 | struct packet_offload *elem; |
62532da9 VY |
474 | |
475 | spin_lock(&offload_lock); | |
bdef7de4 DM |
476 | list_for_each_entry(elem, &offload_base, list) { |
477 | if (po->priority < elem->priority) | |
478 | break; | |
479 | } | |
480 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
481 | spin_unlock(&offload_lock); |
482 | } | |
483 | EXPORT_SYMBOL(dev_add_offload); | |
484 | ||
485 | /** | |
486 | * __dev_remove_offload - remove offload handler | |
487 | * @po: packet offload declaration | |
488 | * | |
489 | * Remove a protocol offload handler that was previously added to the | |
490 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
491 | * is removed from the kernel lists and can be freed or reused once this | |
492 | * function returns. | |
493 | * | |
494 | * The packet type might still be in use by receivers | |
495 | * and must not be freed until after all the CPU's have gone | |
496 | * through a quiescent state. | |
497 | */ | |
1d143d9f | 498 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
499 | { |
500 | struct list_head *head = &offload_base; | |
501 | struct packet_offload *po1; | |
502 | ||
c53aa505 | 503 | spin_lock(&offload_lock); |
62532da9 VY |
504 | |
505 | list_for_each_entry(po1, head, list) { | |
506 | if (po == po1) { | |
507 | list_del_rcu(&po->list); | |
508 | goto out; | |
509 | } | |
510 | } | |
511 | ||
512 | pr_warn("dev_remove_offload: %p not found\n", po); | |
513 | out: | |
c53aa505 | 514 | spin_unlock(&offload_lock); |
62532da9 | 515 | } |
62532da9 VY |
516 | |
517 | /** | |
518 | * dev_remove_offload - remove packet offload handler | |
519 | * @po: packet offload declaration | |
520 | * | |
521 | * Remove a packet offload handler that was previously added to the kernel | |
522 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
523 | * removed from the kernel lists and can be freed or reused once this | |
524 | * function returns. | |
525 | * | |
526 | * This call sleeps to guarantee that no CPU is looking at the packet | |
527 | * type after return. | |
528 | */ | |
529 | void dev_remove_offload(struct packet_offload *po) | |
530 | { | |
531 | __dev_remove_offload(po); | |
532 | ||
533 | synchronize_net(); | |
534 | } | |
535 | EXPORT_SYMBOL(dev_remove_offload); | |
536 | ||
1da177e4 LT |
537 | /****************************************************************************** |
538 | ||
539 | Device Boot-time Settings Routines | |
540 | ||
541 | *******************************************************************************/ | |
542 | ||
543 | /* Boot time configuration table */ | |
544 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
545 | ||
546 | /** | |
547 | * netdev_boot_setup_add - add new setup entry | |
548 | * @name: name of the device | |
549 | * @map: configured settings for the device | |
550 | * | |
551 | * Adds new setup entry to the dev_boot_setup list. The function | |
552 | * returns 0 on error and 1 on success. This is a generic routine to | |
553 | * all netdevices. | |
554 | */ | |
555 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
556 | { | |
557 | struct netdev_boot_setup *s; | |
558 | int i; | |
559 | ||
560 | s = dev_boot_setup; | |
561 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
562 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
563 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 564 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
565 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
566 | break; | |
567 | } | |
568 | } | |
569 | ||
570 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
571 | } | |
572 | ||
573 | /** | |
574 | * netdev_boot_setup_check - check boot time settings | |
575 | * @dev: the netdevice | |
576 | * | |
577 | * Check boot time settings for the device. | |
578 | * The found settings are set for the device to be used | |
579 | * later in the device probing. | |
580 | * Returns 0 if no settings found, 1 if they are. | |
581 | */ | |
582 | int netdev_boot_setup_check(struct net_device *dev) | |
583 | { | |
584 | struct netdev_boot_setup *s = dev_boot_setup; | |
585 | int i; | |
586 | ||
587 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
588 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 589 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
590 | dev->irq = s[i].map.irq; |
591 | dev->base_addr = s[i].map.base_addr; | |
592 | dev->mem_start = s[i].map.mem_start; | |
593 | dev->mem_end = s[i].map.mem_end; | |
594 | return 1; | |
595 | } | |
596 | } | |
597 | return 0; | |
598 | } | |
d1b19dff | 599 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
600 | |
601 | ||
602 | /** | |
603 | * netdev_boot_base - get address from boot time settings | |
604 | * @prefix: prefix for network device | |
605 | * @unit: id for network device | |
606 | * | |
607 | * Check boot time settings for the base address of device. | |
608 | * The found settings are set for the device to be used | |
609 | * later in the device probing. | |
610 | * Returns 0 if no settings found. | |
611 | */ | |
612 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
613 | { | |
614 | const struct netdev_boot_setup *s = dev_boot_setup; | |
615 | char name[IFNAMSIZ]; | |
616 | int i; | |
617 | ||
618 | sprintf(name, "%s%d", prefix, unit); | |
619 | ||
620 | /* | |
621 | * If device already registered then return base of 1 | |
622 | * to indicate not to probe for this interface | |
623 | */ | |
881d966b | 624 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
625 | return 1; |
626 | ||
627 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
628 | if (!strcmp(name, s[i].name)) | |
629 | return s[i].map.base_addr; | |
630 | return 0; | |
631 | } | |
632 | ||
633 | /* | |
634 | * Saves at boot time configured settings for any netdevice. | |
635 | */ | |
636 | int __init netdev_boot_setup(char *str) | |
637 | { | |
638 | int ints[5]; | |
639 | struct ifmap map; | |
640 | ||
641 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
642 | if (!str || !*str) | |
643 | return 0; | |
644 | ||
645 | /* Save settings */ | |
646 | memset(&map, 0, sizeof(map)); | |
647 | if (ints[0] > 0) | |
648 | map.irq = ints[1]; | |
649 | if (ints[0] > 1) | |
650 | map.base_addr = ints[2]; | |
651 | if (ints[0] > 2) | |
652 | map.mem_start = ints[3]; | |
653 | if (ints[0] > 3) | |
654 | map.mem_end = ints[4]; | |
655 | ||
656 | /* Add new entry to the list */ | |
657 | return netdev_boot_setup_add(str, &map); | |
658 | } | |
659 | ||
660 | __setup("netdev=", netdev_boot_setup); | |
661 | ||
662 | /******************************************************************************* | |
663 | ||
664 | Device Interface Subroutines | |
665 | ||
666 | *******************************************************************************/ | |
667 | ||
a54acb3a ND |
668 | /** |
669 | * dev_get_iflink - get 'iflink' value of a interface | |
670 | * @dev: targeted interface | |
671 | * | |
672 | * Indicates the ifindex the interface is linked to. | |
673 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
674 | */ | |
675 | ||
676 | int dev_get_iflink(const struct net_device *dev) | |
677 | { | |
678 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
679 | return dev->netdev_ops->ndo_get_iflink(dev); | |
680 | ||
7a66bbc9 | 681 | return dev->ifindex; |
a54acb3a ND |
682 | } |
683 | EXPORT_SYMBOL(dev_get_iflink); | |
684 | ||
fc4099f1 PS |
685 | /** |
686 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
687 | * @dev: targeted interface | |
688 | * @skb: The packet. | |
689 | * | |
690 | * For better visibility of tunnel traffic OVS needs to retrieve | |
691 | * egress tunnel information for a packet. Following API allows | |
692 | * user to get this info. | |
693 | */ | |
694 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
695 | { | |
696 | struct ip_tunnel_info *info; | |
697 | ||
698 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
699 | return -EINVAL; | |
700 | ||
701 | info = skb_tunnel_info_unclone(skb); | |
702 | if (!info) | |
703 | return -ENOMEM; | |
704 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
705 | return -EINVAL; | |
706 | ||
707 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
708 | } | |
709 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
710 | ||
1da177e4 LT |
711 | /** |
712 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 713 | * @net: the applicable net namespace |
1da177e4 LT |
714 | * @name: name to find |
715 | * | |
716 | * Find an interface by name. Must be called under RTNL semaphore | |
717 | * or @dev_base_lock. If the name is found a pointer to the device | |
718 | * is returned. If the name is not found then %NULL is returned. The | |
719 | * reference counters are not incremented so the caller must be | |
720 | * careful with locks. | |
721 | */ | |
722 | ||
881d966b | 723 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 724 | { |
0bd8d536 ED |
725 | struct net_device *dev; |
726 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 727 | |
b67bfe0d | 728 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
729 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
730 | return dev; | |
0bd8d536 | 731 | |
1da177e4 LT |
732 | return NULL; |
733 | } | |
d1b19dff | 734 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 735 | |
72c9528b ED |
736 | /** |
737 | * dev_get_by_name_rcu - find a device by its name | |
738 | * @net: the applicable net namespace | |
739 | * @name: name to find | |
740 | * | |
741 | * Find an interface by name. | |
742 | * If the name is found a pointer to the device is returned. | |
743 | * If the name is not found then %NULL is returned. | |
744 | * The reference counters are not incremented so the caller must be | |
745 | * careful with locks. The caller must hold RCU lock. | |
746 | */ | |
747 | ||
748 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
749 | { | |
72c9528b ED |
750 | struct net_device *dev; |
751 | struct hlist_head *head = dev_name_hash(net, name); | |
752 | ||
b67bfe0d | 753 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
754 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
755 | return dev; | |
756 | ||
757 | return NULL; | |
758 | } | |
759 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
760 | ||
1da177e4 LT |
761 | /** |
762 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 763 | * @net: the applicable net namespace |
1da177e4 LT |
764 | * @name: name to find |
765 | * | |
766 | * Find an interface by name. This can be called from any | |
767 | * context and does its own locking. The returned handle has | |
768 | * the usage count incremented and the caller must use dev_put() to | |
769 | * release it when it is no longer needed. %NULL is returned if no | |
770 | * matching device is found. | |
771 | */ | |
772 | ||
881d966b | 773 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
774 | { |
775 | struct net_device *dev; | |
776 | ||
72c9528b ED |
777 | rcu_read_lock(); |
778 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
779 | if (dev) |
780 | dev_hold(dev); | |
72c9528b | 781 | rcu_read_unlock(); |
1da177e4 LT |
782 | return dev; |
783 | } | |
d1b19dff | 784 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
785 | |
786 | /** | |
787 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 788 | * @net: the applicable net namespace |
1da177e4 LT |
789 | * @ifindex: index of device |
790 | * | |
791 | * Search for an interface by index. Returns %NULL if the device | |
792 | * is not found or a pointer to the device. The device has not | |
793 | * had its reference counter increased so the caller must be careful | |
794 | * about locking. The caller must hold either the RTNL semaphore | |
795 | * or @dev_base_lock. | |
796 | */ | |
797 | ||
881d966b | 798 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 799 | { |
0bd8d536 ED |
800 | struct net_device *dev; |
801 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 802 | |
b67bfe0d | 803 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
804 | if (dev->ifindex == ifindex) |
805 | return dev; | |
0bd8d536 | 806 | |
1da177e4 LT |
807 | return NULL; |
808 | } | |
d1b19dff | 809 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 810 | |
fb699dfd ED |
811 | /** |
812 | * dev_get_by_index_rcu - find a device by its ifindex | |
813 | * @net: the applicable net namespace | |
814 | * @ifindex: index of device | |
815 | * | |
816 | * Search for an interface by index. Returns %NULL if the device | |
817 | * is not found or a pointer to the device. The device has not | |
818 | * had its reference counter increased so the caller must be careful | |
819 | * about locking. The caller must hold RCU lock. | |
820 | */ | |
821 | ||
822 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
823 | { | |
fb699dfd ED |
824 | struct net_device *dev; |
825 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
826 | ||
b67bfe0d | 827 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
828 | if (dev->ifindex == ifindex) |
829 | return dev; | |
830 | ||
831 | return NULL; | |
832 | } | |
833 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
834 | ||
1da177e4 LT |
835 | |
836 | /** | |
837 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 838 | * @net: the applicable net namespace |
1da177e4 LT |
839 | * @ifindex: index of device |
840 | * | |
841 | * Search for an interface by index. Returns NULL if the device | |
842 | * is not found or a pointer to the device. The device returned has | |
843 | * had a reference added and the pointer is safe until the user calls | |
844 | * dev_put to indicate they have finished with it. | |
845 | */ | |
846 | ||
881d966b | 847 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
848 | { |
849 | struct net_device *dev; | |
850 | ||
fb699dfd ED |
851 | rcu_read_lock(); |
852 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
853 | if (dev) |
854 | dev_hold(dev); | |
fb699dfd | 855 | rcu_read_unlock(); |
1da177e4 LT |
856 | return dev; |
857 | } | |
d1b19dff | 858 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 859 | |
5dbe7c17 NS |
860 | /** |
861 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
862 | * @net: network namespace | |
863 | * @name: a pointer to the buffer where the name will be stored. | |
864 | * @ifindex: the ifindex of the interface to get the name from. | |
865 | * | |
866 | * The use of raw_seqcount_begin() and cond_resched() before | |
867 | * retrying is required as we want to give the writers a chance | |
868 | * to complete when CONFIG_PREEMPT is not set. | |
869 | */ | |
870 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
871 | { | |
872 | struct net_device *dev; | |
873 | unsigned int seq; | |
874 | ||
875 | retry: | |
876 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
877 | rcu_read_lock(); | |
878 | dev = dev_get_by_index_rcu(net, ifindex); | |
879 | if (!dev) { | |
880 | rcu_read_unlock(); | |
881 | return -ENODEV; | |
882 | } | |
883 | ||
884 | strcpy(name, dev->name); | |
885 | rcu_read_unlock(); | |
886 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
887 | cond_resched(); | |
888 | goto retry; | |
889 | } | |
890 | ||
891 | return 0; | |
892 | } | |
893 | ||
1da177e4 | 894 | /** |
941666c2 | 895 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 896 | * @net: the applicable net namespace |
1da177e4 LT |
897 | * @type: media type of device |
898 | * @ha: hardware address | |
899 | * | |
900 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
901 | * is not found or a pointer to the device. |
902 | * The caller must hold RCU or RTNL. | |
941666c2 | 903 | * The returned device has not had its ref count increased |
1da177e4 LT |
904 | * and the caller must therefore be careful about locking |
905 | * | |
1da177e4 LT |
906 | */ |
907 | ||
941666c2 ED |
908 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
909 | const char *ha) | |
1da177e4 LT |
910 | { |
911 | struct net_device *dev; | |
912 | ||
941666c2 | 913 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
914 | if (dev->type == type && |
915 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
916 | return dev; |
917 | ||
918 | return NULL; | |
1da177e4 | 919 | } |
941666c2 | 920 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 921 | |
881d966b | 922 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
923 | { |
924 | struct net_device *dev; | |
925 | ||
4e9cac2b | 926 | ASSERT_RTNL(); |
881d966b | 927 | for_each_netdev(net, dev) |
4e9cac2b | 928 | if (dev->type == type) |
7562f876 PE |
929 | return dev; |
930 | ||
931 | return NULL; | |
4e9cac2b | 932 | } |
4e9cac2b PM |
933 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
934 | ||
881d966b | 935 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 936 | { |
99fe3c39 | 937 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 938 | |
99fe3c39 ED |
939 | rcu_read_lock(); |
940 | for_each_netdev_rcu(net, dev) | |
941 | if (dev->type == type) { | |
942 | dev_hold(dev); | |
943 | ret = dev; | |
944 | break; | |
945 | } | |
946 | rcu_read_unlock(); | |
947 | return ret; | |
1da177e4 | 948 | } |
1da177e4 LT |
949 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
950 | ||
951 | /** | |
6c555490 | 952 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 953 | * @net: the applicable net namespace |
1da177e4 LT |
954 | * @if_flags: IFF_* values |
955 | * @mask: bitmask of bits in if_flags to check | |
956 | * | |
957 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 958 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 959 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
960 | */ |
961 | ||
6c555490 WC |
962 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
963 | unsigned short mask) | |
1da177e4 | 964 | { |
7562f876 | 965 | struct net_device *dev, *ret; |
1da177e4 | 966 | |
6c555490 WC |
967 | ASSERT_RTNL(); |
968 | ||
7562f876 | 969 | ret = NULL; |
6c555490 | 970 | for_each_netdev(net, dev) { |
1da177e4 | 971 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 972 | ret = dev; |
1da177e4 LT |
973 | break; |
974 | } | |
975 | } | |
7562f876 | 976 | return ret; |
1da177e4 | 977 | } |
6c555490 | 978 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
979 | |
980 | /** | |
981 | * dev_valid_name - check if name is okay for network device | |
982 | * @name: name string | |
983 | * | |
984 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
985 | * to allow sysfs to work. We also disallow any kind of |
986 | * whitespace. | |
1da177e4 | 987 | */ |
95f050bf | 988 | bool dev_valid_name(const char *name) |
1da177e4 | 989 | { |
c7fa9d18 | 990 | if (*name == '\0') |
95f050bf | 991 | return false; |
b6fe17d6 | 992 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 993 | return false; |
c7fa9d18 | 994 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 995 | return false; |
c7fa9d18 DM |
996 | |
997 | while (*name) { | |
a4176a93 | 998 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 999 | return false; |
c7fa9d18 DM |
1000 | name++; |
1001 | } | |
95f050bf | 1002 | return true; |
1da177e4 | 1003 | } |
d1b19dff | 1004 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1005 | |
1006 | /** | |
b267b179 EB |
1007 | * __dev_alloc_name - allocate a name for a device |
1008 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1009 | * @name: name format string |
b267b179 | 1010 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1011 | * |
1012 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1013 | * id. It scans list of devices to build up a free map, then chooses |
1014 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1015 | * while allocating the name and adding the device in order to avoid | |
1016 | * duplicates. | |
1017 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1018 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1019 | */ |
1020 | ||
b267b179 | 1021 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1022 | { |
1023 | int i = 0; | |
1da177e4 LT |
1024 | const char *p; |
1025 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1026 | unsigned long *inuse; |
1da177e4 LT |
1027 | struct net_device *d; |
1028 | ||
1029 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
1030 | if (p) { | |
1031 | /* | |
1032 | * Verify the string as this thing may have come from | |
1033 | * the user. There must be either one "%d" and no other "%" | |
1034 | * characters. | |
1035 | */ | |
1036 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1037 | return -EINVAL; | |
1038 | ||
1039 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1040 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1041 | if (!inuse) |
1042 | return -ENOMEM; | |
1043 | ||
881d966b | 1044 | for_each_netdev(net, d) { |
1da177e4 LT |
1045 | if (!sscanf(d->name, name, &i)) |
1046 | continue; | |
1047 | if (i < 0 || i >= max_netdevices) | |
1048 | continue; | |
1049 | ||
1050 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1051 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1052 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1053 | set_bit(i, inuse); | |
1054 | } | |
1055 | ||
1056 | i = find_first_zero_bit(inuse, max_netdevices); | |
1057 | free_page((unsigned long) inuse); | |
1058 | } | |
1059 | ||
d9031024 OP |
1060 | if (buf != name) |
1061 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1062 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1063 | return i; |
1da177e4 LT |
1064 | |
1065 | /* It is possible to run out of possible slots | |
1066 | * when the name is long and there isn't enough space left | |
1067 | * for the digits, or if all bits are used. | |
1068 | */ | |
1069 | return -ENFILE; | |
1070 | } | |
1071 | ||
b267b179 EB |
1072 | /** |
1073 | * dev_alloc_name - allocate a name for a device | |
1074 | * @dev: device | |
1075 | * @name: name format string | |
1076 | * | |
1077 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1078 | * id. It scans list of devices to build up a free map, then chooses | |
1079 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1080 | * while allocating the name and adding the device in order to avoid | |
1081 | * duplicates. | |
1082 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1083 | * Returns the number of the unit assigned or a negative errno code. | |
1084 | */ | |
1085 | ||
1086 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1087 | { | |
1088 | char buf[IFNAMSIZ]; | |
1089 | struct net *net; | |
1090 | int ret; | |
1091 | ||
c346dca1 YH |
1092 | BUG_ON(!dev_net(dev)); |
1093 | net = dev_net(dev); | |
b267b179 EB |
1094 | ret = __dev_alloc_name(net, name, buf); |
1095 | if (ret >= 0) | |
1096 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1097 | return ret; | |
1098 | } | |
d1b19dff | 1099 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1100 | |
828de4f6 G |
1101 | static int dev_alloc_name_ns(struct net *net, |
1102 | struct net_device *dev, | |
1103 | const char *name) | |
d9031024 | 1104 | { |
828de4f6 G |
1105 | char buf[IFNAMSIZ]; |
1106 | int ret; | |
8ce6cebc | 1107 | |
828de4f6 G |
1108 | ret = __dev_alloc_name(net, name, buf); |
1109 | if (ret >= 0) | |
1110 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1111 | return ret; | |
1112 | } | |
1113 | ||
1114 | static int dev_get_valid_name(struct net *net, | |
1115 | struct net_device *dev, | |
1116 | const char *name) | |
1117 | { | |
1118 | BUG_ON(!net); | |
8ce6cebc | 1119 | |
d9031024 OP |
1120 | if (!dev_valid_name(name)) |
1121 | return -EINVAL; | |
1122 | ||
1c5cae81 | 1123 | if (strchr(name, '%')) |
828de4f6 | 1124 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1125 | else if (__dev_get_by_name(net, name)) |
1126 | return -EEXIST; | |
8ce6cebc DL |
1127 | else if (dev->name != name) |
1128 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1129 | |
1130 | return 0; | |
1131 | } | |
1da177e4 LT |
1132 | |
1133 | /** | |
1134 | * dev_change_name - change name of a device | |
1135 | * @dev: device | |
1136 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1137 | * | |
1138 | * Change name of a device, can pass format strings "eth%d". | |
1139 | * for wildcarding. | |
1140 | */ | |
cf04a4c7 | 1141 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1142 | { |
238fa362 | 1143 | unsigned char old_assign_type; |
fcc5a03a | 1144 | char oldname[IFNAMSIZ]; |
1da177e4 | 1145 | int err = 0; |
fcc5a03a | 1146 | int ret; |
881d966b | 1147 | struct net *net; |
1da177e4 LT |
1148 | |
1149 | ASSERT_RTNL(); | |
c346dca1 | 1150 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1151 | |
c346dca1 | 1152 | net = dev_net(dev); |
1da177e4 LT |
1153 | if (dev->flags & IFF_UP) |
1154 | return -EBUSY; | |
1155 | ||
30e6c9fa | 1156 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1157 | |
1158 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1159 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1160 | return 0; |
c91f6df2 | 1161 | } |
c8d90dca | 1162 | |
fcc5a03a HX |
1163 | memcpy(oldname, dev->name, IFNAMSIZ); |
1164 | ||
828de4f6 | 1165 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1166 | if (err < 0) { |
30e6c9fa | 1167 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1168 | return err; |
c91f6df2 | 1169 | } |
1da177e4 | 1170 | |
6fe82a39 VF |
1171 | if (oldname[0] && !strchr(oldname, '%')) |
1172 | netdev_info(dev, "renamed from %s\n", oldname); | |
1173 | ||
238fa362 TG |
1174 | old_assign_type = dev->name_assign_type; |
1175 | dev->name_assign_type = NET_NAME_RENAMED; | |
1176 | ||
fcc5a03a | 1177 | rollback: |
a1b3f594 EB |
1178 | ret = device_rename(&dev->dev, dev->name); |
1179 | if (ret) { | |
1180 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1181 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1182 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1183 | return ret; |
dcc99773 | 1184 | } |
7f988eab | 1185 | |
30e6c9fa | 1186 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1187 | |
5bb025fa VF |
1188 | netdev_adjacent_rename_links(dev, oldname); |
1189 | ||
7f988eab | 1190 | write_lock_bh(&dev_base_lock); |
372b2312 | 1191 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1192 | write_unlock_bh(&dev_base_lock); |
1193 | ||
1194 | synchronize_rcu(); | |
1195 | ||
1196 | write_lock_bh(&dev_base_lock); | |
1197 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1198 | write_unlock_bh(&dev_base_lock); |
1199 | ||
056925ab | 1200 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1201 | ret = notifier_to_errno(ret); |
1202 | ||
1203 | if (ret) { | |
91e9c07b ED |
1204 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1205 | if (err >= 0) { | |
fcc5a03a | 1206 | err = ret; |
30e6c9fa | 1207 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1208 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1209 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1210 | dev->name_assign_type = old_assign_type; |
1211 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1212 | goto rollback; |
91e9c07b | 1213 | } else { |
7b6cd1ce | 1214 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1215 | dev->name, ret); |
fcc5a03a HX |
1216 | } |
1217 | } | |
1da177e4 LT |
1218 | |
1219 | return err; | |
1220 | } | |
1221 | ||
0b815a1a SH |
1222 | /** |
1223 | * dev_set_alias - change ifalias of a device | |
1224 | * @dev: device | |
1225 | * @alias: name up to IFALIASZ | |
f0db275a | 1226 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1227 | * |
1228 | * Set ifalias for a device, | |
1229 | */ | |
1230 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1231 | { | |
7364e445 AK |
1232 | char *new_ifalias; |
1233 | ||
0b815a1a SH |
1234 | ASSERT_RTNL(); |
1235 | ||
1236 | if (len >= IFALIASZ) | |
1237 | return -EINVAL; | |
1238 | ||
96ca4a2c | 1239 | if (!len) { |
388dfc2d SK |
1240 | kfree(dev->ifalias); |
1241 | dev->ifalias = NULL; | |
96ca4a2c OH |
1242 | return 0; |
1243 | } | |
1244 | ||
7364e445 AK |
1245 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1246 | if (!new_ifalias) | |
0b815a1a | 1247 | return -ENOMEM; |
7364e445 | 1248 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1249 | |
1250 | strlcpy(dev->ifalias, alias, len+1); | |
1251 | return len; | |
1252 | } | |
1253 | ||
1254 | ||
d8a33ac4 | 1255 | /** |
3041a069 | 1256 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1257 | * @dev: device to cause notification |
1258 | * | |
1259 | * Called to indicate a device has changed features. | |
1260 | */ | |
1261 | void netdev_features_change(struct net_device *dev) | |
1262 | { | |
056925ab | 1263 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1264 | } |
1265 | EXPORT_SYMBOL(netdev_features_change); | |
1266 | ||
1da177e4 LT |
1267 | /** |
1268 | * netdev_state_change - device changes state | |
1269 | * @dev: device to cause notification | |
1270 | * | |
1271 | * Called to indicate a device has changed state. This function calls | |
1272 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1273 | * to the routing socket. | |
1274 | */ | |
1275 | void netdev_state_change(struct net_device *dev) | |
1276 | { | |
1277 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1278 | struct netdev_notifier_change_info change_info; |
1279 | ||
1280 | change_info.flags_changed = 0; | |
1281 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1282 | &change_info.info); | |
7f294054 | 1283 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1284 | } |
1285 | } | |
d1b19dff | 1286 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1287 | |
ee89bab1 AW |
1288 | /** |
1289 | * netdev_notify_peers - notify network peers about existence of @dev | |
1290 | * @dev: network device | |
1291 | * | |
1292 | * Generate traffic such that interested network peers are aware of | |
1293 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1294 | * a device wants to inform the rest of the network about some sort of | |
1295 | * reconfiguration such as a failover event or virtual machine | |
1296 | * migration. | |
1297 | */ | |
1298 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1299 | { |
ee89bab1 AW |
1300 | rtnl_lock(); |
1301 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1302 | rtnl_unlock(); | |
c1da4ac7 | 1303 | } |
ee89bab1 | 1304 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1305 | |
bd380811 | 1306 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1307 | { |
d314774c | 1308 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1309 | int ret; |
1da177e4 | 1310 | |
e46b66bc BH |
1311 | ASSERT_RTNL(); |
1312 | ||
1da177e4 LT |
1313 | if (!netif_device_present(dev)) |
1314 | return -ENODEV; | |
1315 | ||
ca99ca14 NH |
1316 | /* Block netpoll from trying to do any rx path servicing. |
1317 | * If we don't do this there is a chance ndo_poll_controller | |
1318 | * or ndo_poll may be running while we open the device | |
1319 | */ | |
66b5552f | 1320 | netpoll_poll_disable(dev); |
ca99ca14 | 1321 | |
3b8bcfd5 JB |
1322 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1323 | ret = notifier_to_errno(ret); | |
1324 | if (ret) | |
1325 | return ret; | |
1326 | ||
1da177e4 | 1327 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1328 | |
d314774c SH |
1329 | if (ops->ndo_validate_addr) |
1330 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1331 | |
d314774c SH |
1332 | if (!ret && ops->ndo_open) |
1333 | ret = ops->ndo_open(dev); | |
1da177e4 | 1334 | |
66b5552f | 1335 | netpoll_poll_enable(dev); |
ca99ca14 | 1336 | |
bada339b JG |
1337 | if (ret) |
1338 | clear_bit(__LINK_STATE_START, &dev->state); | |
1339 | else { | |
1da177e4 | 1340 | dev->flags |= IFF_UP; |
4417da66 | 1341 | dev_set_rx_mode(dev); |
1da177e4 | 1342 | dev_activate(dev); |
7bf23575 | 1343 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1344 | } |
bada339b | 1345 | |
1da177e4 LT |
1346 | return ret; |
1347 | } | |
1348 | ||
1349 | /** | |
bd380811 PM |
1350 | * dev_open - prepare an interface for use. |
1351 | * @dev: device to open | |
1da177e4 | 1352 | * |
bd380811 PM |
1353 | * Takes a device from down to up state. The device's private open |
1354 | * function is invoked and then the multicast lists are loaded. Finally | |
1355 | * the device is moved into the up state and a %NETDEV_UP message is | |
1356 | * sent to the netdev notifier chain. | |
1357 | * | |
1358 | * Calling this function on an active interface is a nop. On a failure | |
1359 | * a negative errno code is returned. | |
1da177e4 | 1360 | */ |
bd380811 PM |
1361 | int dev_open(struct net_device *dev) |
1362 | { | |
1363 | int ret; | |
1364 | ||
bd380811 PM |
1365 | if (dev->flags & IFF_UP) |
1366 | return 0; | |
1367 | ||
bd380811 PM |
1368 | ret = __dev_open(dev); |
1369 | if (ret < 0) | |
1370 | return ret; | |
1371 | ||
7f294054 | 1372 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1373 | call_netdevice_notifiers(NETDEV_UP, dev); |
1374 | ||
1375 | return ret; | |
1376 | } | |
1377 | EXPORT_SYMBOL(dev_open); | |
1378 | ||
44345724 | 1379 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1380 | { |
44345724 | 1381 | struct net_device *dev; |
e46b66bc | 1382 | |
bd380811 | 1383 | ASSERT_RTNL(); |
9d5010db DM |
1384 | might_sleep(); |
1385 | ||
5cde2829 | 1386 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1387 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1388 | netpoll_poll_disable(dev); |
3f4df206 | 1389 | |
44345724 | 1390 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1391 | |
44345724 | 1392 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1393 | |
44345724 OP |
1394 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1395 | * can be even on different cpu. So just clear netif_running(). | |
1396 | * | |
1397 | * dev->stop() will invoke napi_disable() on all of it's | |
1398 | * napi_struct instances on this device. | |
1399 | */ | |
4e857c58 | 1400 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1401 | } |
1da177e4 | 1402 | |
44345724 | 1403 | dev_deactivate_many(head); |
d8b2a4d2 | 1404 | |
5cde2829 | 1405 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1406 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1407 | |
44345724 OP |
1408 | /* |
1409 | * Call the device specific close. This cannot fail. | |
1410 | * Only if device is UP | |
1411 | * | |
1412 | * We allow it to be called even after a DETACH hot-plug | |
1413 | * event. | |
1414 | */ | |
1415 | if (ops->ndo_stop) | |
1416 | ops->ndo_stop(dev); | |
1417 | ||
44345724 | 1418 | dev->flags &= ~IFF_UP; |
66b5552f | 1419 | netpoll_poll_enable(dev); |
44345724 OP |
1420 | } |
1421 | ||
1422 | return 0; | |
1423 | } | |
1424 | ||
1425 | static int __dev_close(struct net_device *dev) | |
1426 | { | |
f87e6f47 | 1427 | int retval; |
44345724 OP |
1428 | LIST_HEAD(single); |
1429 | ||
5cde2829 | 1430 | list_add(&dev->close_list, &single); |
f87e6f47 LT |
1431 | retval = __dev_close_many(&single); |
1432 | list_del(&single); | |
ca99ca14 | 1433 | |
f87e6f47 | 1434 | return retval; |
44345724 OP |
1435 | } |
1436 | ||
99c4a26a | 1437 | int dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1438 | { |
1439 | struct net_device *dev, *tmp; | |
1da177e4 | 1440 | |
5cde2829 EB |
1441 | /* Remove the devices that don't need to be closed */ |
1442 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1443 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1444 | list_del_init(&dev->close_list); |
44345724 OP |
1445 | |
1446 | __dev_close_many(head); | |
1da177e4 | 1447 | |
5cde2829 | 1448 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1449 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1450 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1451 | if (unlink) |
1452 | list_del_init(&dev->close_list); | |
44345724 | 1453 | } |
bd380811 PM |
1454 | |
1455 | return 0; | |
1456 | } | |
99c4a26a | 1457 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1458 | |
1459 | /** | |
1460 | * dev_close - shutdown an interface. | |
1461 | * @dev: device to shutdown | |
1462 | * | |
1463 | * This function moves an active device into down state. A | |
1464 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1465 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1466 | * chain. | |
1467 | */ | |
1468 | int dev_close(struct net_device *dev) | |
1469 | { | |
e14a5993 ED |
1470 | if (dev->flags & IFF_UP) { |
1471 | LIST_HEAD(single); | |
1da177e4 | 1472 | |
5cde2829 | 1473 | list_add(&dev->close_list, &single); |
99c4a26a | 1474 | dev_close_many(&single, true); |
e14a5993 ED |
1475 | list_del(&single); |
1476 | } | |
da6e378b | 1477 | return 0; |
1da177e4 | 1478 | } |
d1b19dff | 1479 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1480 | |
1481 | ||
0187bdfb BH |
1482 | /** |
1483 | * dev_disable_lro - disable Large Receive Offload on a device | |
1484 | * @dev: device | |
1485 | * | |
1486 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1487 | * called under RTNL. This is needed if received packets may be | |
1488 | * forwarded to another interface. | |
1489 | */ | |
1490 | void dev_disable_lro(struct net_device *dev) | |
1491 | { | |
fbe168ba MK |
1492 | struct net_device *lower_dev; |
1493 | struct list_head *iter; | |
529d0489 | 1494 | |
bc5787c6 MM |
1495 | dev->wanted_features &= ~NETIF_F_LRO; |
1496 | netdev_update_features(dev); | |
27660515 | 1497 | |
22d5969f MM |
1498 | if (unlikely(dev->features & NETIF_F_LRO)) |
1499 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1500 | |
1501 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1502 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1503 | } |
1504 | EXPORT_SYMBOL(dev_disable_lro); | |
1505 | ||
351638e7 JP |
1506 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1507 | struct net_device *dev) | |
1508 | { | |
1509 | struct netdev_notifier_info info; | |
1510 | ||
1511 | netdev_notifier_info_init(&info, dev); | |
1512 | return nb->notifier_call(nb, val, &info); | |
1513 | } | |
0187bdfb | 1514 | |
881d966b EB |
1515 | static int dev_boot_phase = 1; |
1516 | ||
1da177e4 LT |
1517 | /** |
1518 | * register_netdevice_notifier - register a network notifier block | |
1519 | * @nb: notifier | |
1520 | * | |
1521 | * Register a notifier to be called when network device events occur. | |
1522 | * The notifier passed is linked into the kernel structures and must | |
1523 | * not be reused until it has been unregistered. A negative errno code | |
1524 | * is returned on a failure. | |
1525 | * | |
1526 | * When registered all registration and up events are replayed | |
4ec93edb | 1527 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1528 | * view of the network device list. |
1529 | */ | |
1530 | ||
1531 | int register_netdevice_notifier(struct notifier_block *nb) | |
1532 | { | |
1533 | struct net_device *dev; | |
fcc5a03a | 1534 | struct net_device *last; |
881d966b | 1535 | struct net *net; |
1da177e4 LT |
1536 | int err; |
1537 | ||
1538 | rtnl_lock(); | |
f07d5b94 | 1539 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1540 | if (err) |
1541 | goto unlock; | |
881d966b EB |
1542 | if (dev_boot_phase) |
1543 | goto unlock; | |
1544 | for_each_net(net) { | |
1545 | for_each_netdev(net, dev) { | |
351638e7 | 1546 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1547 | err = notifier_to_errno(err); |
1548 | if (err) | |
1549 | goto rollback; | |
1550 | ||
1551 | if (!(dev->flags & IFF_UP)) | |
1552 | continue; | |
1da177e4 | 1553 | |
351638e7 | 1554 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1555 | } |
1da177e4 | 1556 | } |
fcc5a03a HX |
1557 | |
1558 | unlock: | |
1da177e4 LT |
1559 | rtnl_unlock(); |
1560 | return err; | |
fcc5a03a HX |
1561 | |
1562 | rollback: | |
1563 | last = dev; | |
881d966b EB |
1564 | for_each_net(net) { |
1565 | for_each_netdev(net, dev) { | |
1566 | if (dev == last) | |
8f891489 | 1567 | goto outroll; |
fcc5a03a | 1568 | |
881d966b | 1569 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1570 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1571 | dev); | |
1572 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1573 | } |
351638e7 | 1574 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1575 | } |
fcc5a03a | 1576 | } |
c67625a1 | 1577 | |
8f891489 | 1578 | outroll: |
c67625a1 | 1579 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1580 | goto unlock; |
1da177e4 | 1581 | } |
d1b19dff | 1582 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1583 | |
1584 | /** | |
1585 | * unregister_netdevice_notifier - unregister a network notifier block | |
1586 | * @nb: notifier | |
1587 | * | |
1588 | * Unregister a notifier previously registered by | |
1589 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1590 | * kernel structures and may then be reused. A negative errno code | |
1591 | * is returned on a failure. | |
7d3d43da EB |
1592 | * |
1593 | * After unregistering unregister and down device events are synthesized | |
1594 | * for all devices on the device list to the removed notifier to remove | |
1595 | * the need for special case cleanup code. | |
1da177e4 LT |
1596 | */ |
1597 | ||
1598 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1599 | { | |
7d3d43da EB |
1600 | struct net_device *dev; |
1601 | struct net *net; | |
9f514950 HX |
1602 | int err; |
1603 | ||
1604 | rtnl_lock(); | |
f07d5b94 | 1605 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1606 | if (err) |
1607 | goto unlock; | |
1608 | ||
1609 | for_each_net(net) { | |
1610 | for_each_netdev(net, dev) { | |
1611 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1612 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1613 | dev); | |
1614 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1615 | } |
351638e7 | 1616 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1617 | } |
1618 | } | |
1619 | unlock: | |
9f514950 HX |
1620 | rtnl_unlock(); |
1621 | return err; | |
1da177e4 | 1622 | } |
d1b19dff | 1623 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1624 | |
351638e7 JP |
1625 | /** |
1626 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1627 | * @val: value passed unmodified to notifier function | |
1628 | * @dev: net_device pointer passed unmodified to notifier function | |
1629 | * @info: notifier information data | |
1630 | * | |
1631 | * Call all network notifier blocks. Parameters and return value | |
1632 | * are as for raw_notifier_call_chain(). | |
1633 | */ | |
1634 | ||
1d143d9f | 1635 | static int call_netdevice_notifiers_info(unsigned long val, |
1636 | struct net_device *dev, | |
1637 | struct netdev_notifier_info *info) | |
351638e7 JP |
1638 | { |
1639 | ASSERT_RTNL(); | |
1640 | netdev_notifier_info_init(info, dev); | |
1641 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1642 | } | |
351638e7 | 1643 | |
1da177e4 LT |
1644 | /** |
1645 | * call_netdevice_notifiers - call all network notifier blocks | |
1646 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1647 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1648 | * |
1649 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1650 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1651 | */ |
1652 | ||
ad7379d4 | 1653 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1654 | { |
351638e7 JP |
1655 | struct netdev_notifier_info info; |
1656 | ||
1657 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1658 | } |
edf947f1 | 1659 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1660 | |
1cf51900 | 1661 | #ifdef CONFIG_NET_INGRESS |
4577139b DB |
1662 | static struct static_key ingress_needed __read_mostly; |
1663 | ||
1664 | void net_inc_ingress_queue(void) | |
1665 | { | |
1666 | static_key_slow_inc(&ingress_needed); | |
1667 | } | |
1668 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
1669 | ||
1670 | void net_dec_ingress_queue(void) | |
1671 | { | |
1672 | static_key_slow_dec(&ingress_needed); | |
1673 | } | |
1674 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
1675 | #endif | |
1676 | ||
c5905afb | 1677 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1678 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1679 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1680 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1681 | * static_key_slow_dec() calls. |
b90e5794 ED |
1682 | */ |
1683 | static atomic_t netstamp_needed_deferred; | |
1684 | #endif | |
1da177e4 LT |
1685 | |
1686 | void net_enable_timestamp(void) | |
1687 | { | |
b90e5794 ED |
1688 | #ifdef HAVE_JUMP_LABEL |
1689 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1690 | ||
1691 | if (deferred) { | |
1692 | while (--deferred) | |
c5905afb | 1693 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1694 | return; |
1695 | } | |
1696 | #endif | |
c5905afb | 1697 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1698 | } |
d1b19dff | 1699 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1700 | |
1701 | void net_disable_timestamp(void) | |
1702 | { | |
b90e5794 ED |
1703 | #ifdef HAVE_JUMP_LABEL |
1704 | if (in_interrupt()) { | |
1705 | atomic_inc(&netstamp_needed_deferred); | |
1706 | return; | |
1707 | } | |
1708 | #endif | |
c5905afb | 1709 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1710 | } |
d1b19dff | 1711 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1712 | |
3b098e2d | 1713 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1714 | { |
588f0330 | 1715 | skb->tstamp.tv64 = 0; |
c5905afb | 1716 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1717 | __net_timestamp(skb); |
1da177e4 LT |
1718 | } |
1719 | ||
588f0330 | 1720 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1721 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1722 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1723 | __net_timestamp(SKB); \ | |
1724 | } \ | |
3b098e2d | 1725 | |
1ee481fb | 1726 | bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb) |
79b569f0 DL |
1727 | { |
1728 | unsigned int len; | |
1729 | ||
1730 | if (!(dev->flags & IFF_UP)) | |
1731 | return false; | |
1732 | ||
1733 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1734 | if (skb->len <= len) | |
1735 | return true; | |
1736 | ||
1737 | /* if TSO is enabled, we don't care about the length as the packet | |
1738 | * could be forwarded without being segmented before | |
1739 | */ | |
1740 | if (skb_is_gso(skb)) | |
1741 | return true; | |
1742 | ||
1743 | return false; | |
1744 | } | |
1ee481fb | 1745 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1746 | |
a0265d28 HX |
1747 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1748 | { | |
bbbf2df0 WB |
1749 | if (skb_orphan_frags(skb, GFP_ATOMIC) || |
1750 | unlikely(!is_skb_forwardable(dev, skb))) { | |
a0265d28 HX |
1751 | atomic_long_inc(&dev->rx_dropped); |
1752 | kfree_skb(skb); | |
1753 | return NET_RX_DROP; | |
1754 | } | |
1755 | ||
1756 | skb_scrub_packet(skb, true); | |
08b4b8ea | 1757 | skb->priority = 0; |
a0265d28 | 1758 | skb->protocol = eth_type_trans(skb, dev); |
2c26d34b | 1759 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); |
a0265d28 HX |
1760 | |
1761 | return 0; | |
1762 | } | |
1763 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1764 | ||
44540960 AB |
1765 | /** |
1766 | * dev_forward_skb - loopback an skb to another netif | |
1767 | * | |
1768 | * @dev: destination network device | |
1769 | * @skb: buffer to forward | |
1770 | * | |
1771 | * return values: | |
1772 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1773 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1774 | * |
1775 | * dev_forward_skb can be used for injecting an skb from the | |
1776 | * start_xmit function of one device into the receive queue | |
1777 | * of another device. | |
1778 | * | |
1779 | * The receiving device may be in another namespace, so | |
1780 | * we have to clear all information in the skb that could | |
1781 | * impact namespace isolation. | |
1782 | */ | |
1783 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1784 | { | |
a0265d28 | 1785 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1786 | } |
1787 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1788 | ||
71d9dec2 CG |
1789 | static inline int deliver_skb(struct sk_buff *skb, |
1790 | struct packet_type *pt_prev, | |
1791 | struct net_device *orig_dev) | |
1792 | { | |
1080e512 MT |
1793 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1794 | return -ENOMEM; | |
71d9dec2 CG |
1795 | atomic_inc(&skb->users); |
1796 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1797 | } | |
1798 | ||
7866a621 SN |
1799 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
1800 | struct packet_type **pt, | |
fbcb2170 JP |
1801 | struct net_device *orig_dev, |
1802 | __be16 type, | |
7866a621 SN |
1803 | struct list_head *ptype_list) |
1804 | { | |
1805 | struct packet_type *ptype, *pt_prev = *pt; | |
1806 | ||
1807 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1808 | if (ptype->type != type) | |
1809 | continue; | |
1810 | if (pt_prev) | |
fbcb2170 | 1811 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
1812 | pt_prev = ptype; |
1813 | } | |
1814 | *pt = pt_prev; | |
1815 | } | |
1816 | ||
c0de08d0 EL |
1817 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1818 | { | |
a3d744e9 | 1819 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1820 | return false; |
1821 | ||
1822 | if (ptype->id_match) | |
1823 | return ptype->id_match(ptype, skb->sk); | |
1824 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1825 | return true; | |
1826 | ||
1827 | return false; | |
1828 | } | |
1829 | ||
1da177e4 LT |
1830 | /* |
1831 | * Support routine. Sends outgoing frames to any network | |
1832 | * taps currently in use. | |
1833 | */ | |
1834 | ||
f6a78bfc | 1835 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1836 | { |
1837 | struct packet_type *ptype; | |
71d9dec2 CG |
1838 | struct sk_buff *skb2 = NULL; |
1839 | struct packet_type *pt_prev = NULL; | |
7866a621 | 1840 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 1841 | |
1da177e4 | 1842 | rcu_read_lock(); |
7866a621 SN |
1843 | again: |
1844 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1da177e4 LT |
1845 | /* Never send packets back to the socket |
1846 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1847 | */ | |
7866a621 SN |
1848 | if (skb_loop_sk(ptype, skb)) |
1849 | continue; | |
71d9dec2 | 1850 | |
7866a621 SN |
1851 | if (pt_prev) { |
1852 | deliver_skb(skb2, pt_prev, skb->dev); | |
1853 | pt_prev = ptype; | |
1854 | continue; | |
1855 | } | |
1da177e4 | 1856 | |
7866a621 SN |
1857 | /* need to clone skb, done only once */ |
1858 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1859 | if (!skb2) | |
1860 | goto out_unlock; | |
70978182 | 1861 | |
7866a621 | 1862 | net_timestamp_set(skb2); |
1da177e4 | 1863 | |
7866a621 SN |
1864 | /* skb->nh should be correctly |
1865 | * set by sender, so that the second statement is | |
1866 | * just protection against buggy protocols. | |
1867 | */ | |
1868 | skb_reset_mac_header(skb2); | |
1869 | ||
1870 | if (skb_network_header(skb2) < skb2->data || | |
1871 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
1872 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
1873 | ntohs(skb2->protocol), | |
1874 | dev->name); | |
1875 | skb_reset_network_header(skb2); | |
1da177e4 | 1876 | } |
7866a621 SN |
1877 | |
1878 | skb2->transport_header = skb2->network_header; | |
1879 | skb2->pkt_type = PACKET_OUTGOING; | |
1880 | pt_prev = ptype; | |
1881 | } | |
1882 | ||
1883 | if (ptype_list == &ptype_all) { | |
1884 | ptype_list = &dev->ptype_all; | |
1885 | goto again; | |
1da177e4 | 1886 | } |
7866a621 | 1887 | out_unlock: |
71d9dec2 CG |
1888 | if (pt_prev) |
1889 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1890 | rcu_read_unlock(); |
1891 | } | |
1892 | ||
2c53040f BH |
1893 | /** |
1894 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1895 | * @dev: Network device |
1896 | * @txq: number of queues available | |
1897 | * | |
1898 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1899 | * valid. To resolve this verify the tc mapping remains valid and if | |
1900 | * not NULL the mapping. With no priorities mapping to this | |
1901 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1902 | * is invalid nothing can be done so disable priority mappings. If is | |
1903 | * expected that drivers will fix this mapping if they can before | |
1904 | * calling netif_set_real_num_tx_queues. | |
1905 | */ | |
bb134d22 | 1906 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1907 | { |
1908 | int i; | |
1909 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1910 | ||
1911 | /* If TC0 is invalidated disable TC mapping */ | |
1912 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1913 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1914 | dev->num_tc = 0; |
1915 | return; | |
1916 | } | |
1917 | ||
1918 | /* Invalidated prio to tc mappings set to TC0 */ | |
1919 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1920 | int q = netdev_get_prio_tc_map(dev, i); | |
1921 | ||
1922 | tc = &dev->tc_to_txq[q]; | |
1923 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1924 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1925 | i, q); | |
4f57c087 JF |
1926 | netdev_set_prio_tc_map(dev, i, 0); |
1927 | } | |
1928 | } | |
1929 | } | |
1930 | ||
537c00de AD |
1931 | #ifdef CONFIG_XPS |
1932 | static DEFINE_MUTEX(xps_map_mutex); | |
1933 | #define xmap_dereference(P) \ | |
1934 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1935 | ||
10cdc3f3 AD |
1936 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1937 | int cpu, u16 index) | |
537c00de | 1938 | { |
10cdc3f3 AD |
1939 | struct xps_map *map = NULL; |
1940 | int pos; | |
537c00de | 1941 | |
10cdc3f3 AD |
1942 | if (dev_maps) |
1943 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1944 | |
10cdc3f3 AD |
1945 | for (pos = 0; map && pos < map->len; pos++) { |
1946 | if (map->queues[pos] == index) { | |
537c00de AD |
1947 | if (map->len > 1) { |
1948 | map->queues[pos] = map->queues[--map->len]; | |
1949 | } else { | |
10cdc3f3 | 1950 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1951 | kfree_rcu(map, rcu); |
1952 | map = NULL; | |
1953 | } | |
10cdc3f3 | 1954 | break; |
537c00de | 1955 | } |
537c00de AD |
1956 | } |
1957 | ||
10cdc3f3 AD |
1958 | return map; |
1959 | } | |
1960 | ||
024e9679 | 1961 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
10cdc3f3 AD |
1962 | { |
1963 | struct xps_dev_maps *dev_maps; | |
024e9679 | 1964 | int cpu, i; |
10cdc3f3 AD |
1965 | bool active = false; |
1966 | ||
1967 | mutex_lock(&xps_map_mutex); | |
1968 | dev_maps = xmap_dereference(dev->xps_maps); | |
1969 | ||
1970 | if (!dev_maps) | |
1971 | goto out_no_maps; | |
1972 | ||
1973 | for_each_possible_cpu(cpu) { | |
024e9679 AD |
1974 | for (i = index; i < dev->num_tx_queues; i++) { |
1975 | if (!remove_xps_queue(dev_maps, cpu, i)) | |
1976 | break; | |
1977 | } | |
1978 | if (i == dev->num_tx_queues) | |
10cdc3f3 AD |
1979 | active = true; |
1980 | } | |
1981 | ||
1982 | if (!active) { | |
537c00de AD |
1983 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
1984 | kfree_rcu(dev_maps, rcu); | |
1985 | } | |
1986 | ||
024e9679 AD |
1987 | for (i = index; i < dev->num_tx_queues; i++) |
1988 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), | |
1989 | NUMA_NO_NODE); | |
1990 | ||
537c00de AD |
1991 | out_no_maps: |
1992 | mutex_unlock(&xps_map_mutex); | |
1993 | } | |
1994 | ||
01c5f864 AD |
1995 | static struct xps_map *expand_xps_map(struct xps_map *map, |
1996 | int cpu, u16 index) | |
1997 | { | |
1998 | struct xps_map *new_map; | |
1999 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2000 | int i, pos; | |
2001 | ||
2002 | for (pos = 0; map && pos < map->len; pos++) { | |
2003 | if (map->queues[pos] != index) | |
2004 | continue; | |
2005 | return map; | |
2006 | } | |
2007 | ||
2008 | /* Need to add queue to this CPU's existing map */ | |
2009 | if (map) { | |
2010 | if (pos < map->alloc_len) | |
2011 | return map; | |
2012 | ||
2013 | alloc_len = map->alloc_len * 2; | |
2014 | } | |
2015 | ||
2016 | /* Need to allocate new map to store queue on this CPU's map */ | |
2017 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2018 | cpu_to_node(cpu)); | |
2019 | if (!new_map) | |
2020 | return NULL; | |
2021 | ||
2022 | for (i = 0; i < pos; i++) | |
2023 | new_map->queues[i] = map->queues[i]; | |
2024 | new_map->alloc_len = alloc_len; | |
2025 | new_map->len = pos; | |
2026 | ||
2027 | return new_map; | |
2028 | } | |
2029 | ||
3573540c MT |
2030 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
2031 | u16 index) | |
537c00de | 2032 | { |
01c5f864 | 2033 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 2034 | struct xps_map *map, *new_map; |
537c00de | 2035 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
2036 | int cpu, numa_node_id = -2; |
2037 | bool active = false; | |
537c00de AD |
2038 | |
2039 | mutex_lock(&xps_map_mutex); | |
2040 | ||
2041 | dev_maps = xmap_dereference(dev->xps_maps); | |
2042 | ||
01c5f864 AD |
2043 | /* allocate memory for queue storage */ |
2044 | for_each_online_cpu(cpu) { | |
2045 | if (!cpumask_test_cpu(cpu, mask)) | |
2046 | continue; | |
2047 | ||
2048 | if (!new_dev_maps) | |
2049 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2050 | if (!new_dev_maps) { |
2051 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2052 | return -ENOMEM; |
2bb60cb9 | 2053 | } |
01c5f864 AD |
2054 | |
2055 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2056 | NULL; | |
2057 | ||
2058 | map = expand_xps_map(map, cpu, index); | |
2059 | if (!map) | |
2060 | goto error; | |
2061 | ||
2062 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
2063 | } | |
2064 | ||
2065 | if (!new_dev_maps) | |
2066 | goto out_no_new_maps; | |
2067 | ||
537c00de | 2068 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
2069 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
2070 | /* add queue to CPU maps */ | |
2071 | int pos = 0; | |
2072 | ||
2073 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2074 | while ((pos < map->len) && (map->queues[pos] != index)) | |
2075 | pos++; | |
2076 | ||
2077 | if (pos == map->len) | |
2078 | map->queues[map->len++] = index; | |
537c00de | 2079 | #ifdef CONFIG_NUMA |
537c00de AD |
2080 | if (numa_node_id == -2) |
2081 | numa_node_id = cpu_to_node(cpu); | |
2082 | else if (numa_node_id != cpu_to_node(cpu)) | |
2083 | numa_node_id = -1; | |
537c00de | 2084 | #endif |
01c5f864 AD |
2085 | } else if (dev_maps) { |
2086 | /* fill in the new device map from the old device map */ | |
2087 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2088 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 2089 | } |
01c5f864 | 2090 | |
537c00de AD |
2091 | } |
2092 | ||
01c5f864 AD |
2093 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2094 | ||
537c00de | 2095 | /* Cleanup old maps */ |
01c5f864 AD |
2096 | if (dev_maps) { |
2097 | for_each_possible_cpu(cpu) { | |
2098 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2099 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2100 | if (map && map != new_map) | |
2101 | kfree_rcu(map, rcu); | |
2102 | } | |
537c00de | 2103 | |
01c5f864 | 2104 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
2105 | } |
2106 | ||
01c5f864 AD |
2107 | dev_maps = new_dev_maps; |
2108 | active = true; | |
537c00de | 2109 | |
01c5f864 AD |
2110 | out_no_new_maps: |
2111 | /* update Tx queue numa node */ | |
537c00de AD |
2112 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2113 | (numa_node_id >= 0) ? numa_node_id : | |
2114 | NUMA_NO_NODE); | |
2115 | ||
01c5f864 AD |
2116 | if (!dev_maps) |
2117 | goto out_no_maps; | |
2118 | ||
2119 | /* removes queue from unused CPUs */ | |
2120 | for_each_possible_cpu(cpu) { | |
2121 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
2122 | continue; | |
2123 | ||
2124 | if (remove_xps_queue(dev_maps, cpu, index)) | |
2125 | active = true; | |
2126 | } | |
2127 | ||
2128 | /* free map if not active */ | |
2129 | if (!active) { | |
2130 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2131 | kfree_rcu(dev_maps, rcu); | |
2132 | } | |
2133 | ||
2134 | out_no_maps: | |
537c00de AD |
2135 | mutex_unlock(&xps_map_mutex); |
2136 | ||
2137 | return 0; | |
2138 | error: | |
01c5f864 AD |
2139 | /* remove any maps that we added */ |
2140 | for_each_possible_cpu(cpu) { | |
2141 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2142 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2143 | NULL; | |
2144 | if (new_map && new_map != map) | |
2145 | kfree(new_map); | |
2146 | } | |
2147 | ||
537c00de AD |
2148 | mutex_unlock(&xps_map_mutex); |
2149 | ||
537c00de AD |
2150 | kfree(new_dev_maps); |
2151 | return -ENOMEM; | |
2152 | } | |
2153 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2154 | ||
2155 | #endif | |
f0796d5c JF |
2156 | /* |
2157 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2158 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2159 | */ | |
e6484930 | 2160 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2161 | { |
1d24eb48 TH |
2162 | int rc; |
2163 | ||
e6484930 TH |
2164 | if (txq < 1 || txq > dev->num_tx_queues) |
2165 | return -EINVAL; | |
f0796d5c | 2166 | |
5c56580b BH |
2167 | if (dev->reg_state == NETREG_REGISTERED || |
2168 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2169 | ASSERT_RTNL(); |
2170 | ||
1d24eb48 TH |
2171 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2172 | txq); | |
bf264145 TH |
2173 | if (rc) |
2174 | return rc; | |
2175 | ||
4f57c087 JF |
2176 | if (dev->num_tc) |
2177 | netif_setup_tc(dev, txq); | |
2178 | ||
024e9679 | 2179 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2180 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2181 | #ifdef CONFIG_XPS |
2182 | netif_reset_xps_queues_gt(dev, txq); | |
2183 | #endif | |
2184 | } | |
f0796d5c | 2185 | } |
e6484930 TH |
2186 | |
2187 | dev->real_num_tx_queues = txq; | |
2188 | return 0; | |
f0796d5c JF |
2189 | } |
2190 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2191 | |
a953be53 | 2192 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2193 | /** |
2194 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2195 | * @dev: Network device | |
2196 | * @rxq: Actual number of RX queues | |
2197 | * | |
2198 | * This must be called either with the rtnl_lock held or before | |
2199 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2200 | * negative error code. If called before registration, it always |
2201 | * succeeds. | |
62fe0b40 BH |
2202 | */ |
2203 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2204 | { | |
2205 | int rc; | |
2206 | ||
bd25fa7b TH |
2207 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2208 | return -EINVAL; | |
2209 | ||
62fe0b40 BH |
2210 | if (dev->reg_state == NETREG_REGISTERED) { |
2211 | ASSERT_RTNL(); | |
2212 | ||
62fe0b40 BH |
2213 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2214 | rxq); | |
2215 | if (rc) | |
2216 | return rc; | |
62fe0b40 BH |
2217 | } |
2218 | ||
2219 | dev->real_num_rx_queues = rxq; | |
2220 | return 0; | |
2221 | } | |
2222 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2223 | #endif | |
2224 | ||
2c53040f BH |
2225 | /** |
2226 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2227 | * |
2228 | * This routine should set an upper limit on the number of RSS queues | |
2229 | * used by default by multiqueue devices. | |
2230 | */ | |
a55b138b | 2231 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
2232 | { |
2233 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
2234 | } | |
2235 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2236 | ||
def82a1d | 2237 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 2238 | { |
def82a1d JP |
2239 | struct softnet_data *sd; |
2240 | unsigned long flags; | |
56079431 | 2241 | |
def82a1d | 2242 | local_irq_save(flags); |
903ceff7 | 2243 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2244 | q->next_sched = NULL; |
2245 | *sd->output_queue_tailp = q; | |
2246 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2247 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2248 | local_irq_restore(flags); | |
2249 | } | |
2250 | ||
2251 | void __netif_schedule(struct Qdisc *q) | |
2252 | { | |
2253 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2254 | __netif_reschedule(q); | |
56079431 DV |
2255 | } |
2256 | EXPORT_SYMBOL(__netif_schedule); | |
2257 | ||
e6247027 ED |
2258 | struct dev_kfree_skb_cb { |
2259 | enum skb_free_reason reason; | |
2260 | }; | |
2261 | ||
2262 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2263 | { |
e6247027 ED |
2264 | return (struct dev_kfree_skb_cb *)skb->cb; |
2265 | } | |
2266 | ||
46e5da40 JF |
2267 | void netif_schedule_queue(struct netdev_queue *txq) |
2268 | { | |
2269 | rcu_read_lock(); | |
2270 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2271 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2272 | ||
2273 | __netif_schedule(q); | |
2274 | } | |
2275 | rcu_read_unlock(); | |
2276 | } | |
2277 | EXPORT_SYMBOL(netif_schedule_queue); | |
2278 | ||
2279 | /** | |
2280 | * netif_wake_subqueue - allow sending packets on subqueue | |
2281 | * @dev: network device | |
2282 | * @queue_index: sub queue index | |
2283 | * | |
2284 | * Resume individual transmit queue of a device with multiple transmit queues. | |
2285 | */ | |
2286 | void netif_wake_subqueue(struct net_device *dev, u16 queue_index) | |
2287 | { | |
2288 | struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index); | |
2289 | ||
2290 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) { | |
2291 | struct Qdisc *q; | |
2292 | ||
2293 | rcu_read_lock(); | |
2294 | q = rcu_dereference(txq->qdisc); | |
2295 | __netif_schedule(q); | |
2296 | rcu_read_unlock(); | |
2297 | } | |
2298 | } | |
2299 | EXPORT_SYMBOL(netif_wake_subqueue); | |
2300 | ||
2301 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) | |
2302 | { | |
2303 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2304 | struct Qdisc *q; | |
2305 | ||
2306 | rcu_read_lock(); | |
2307 | q = rcu_dereference(dev_queue->qdisc); | |
2308 | __netif_schedule(q); | |
2309 | rcu_read_unlock(); | |
2310 | } | |
2311 | } | |
2312 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2313 | ||
e6247027 | 2314 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2315 | { |
e6247027 | 2316 | unsigned long flags; |
56079431 | 2317 | |
e6247027 ED |
2318 | if (likely(atomic_read(&skb->users) == 1)) { |
2319 | smp_rmb(); | |
2320 | atomic_set(&skb->users, 0); | |
2321 | } else if (likely(!atomic_dec_and_test(&skb->users))) { | |
2322 | return; | |
bea3348e | 2323 | } |
e6247027 ED |
2324 | get_kfree_skb_cb(skb)->reason = reason; |
2325 | local_irq_save(flags); | |
2326 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2327 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2328 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2329 | local_irq_restore(flags); | |
56079431 | 2330 | } |
e6247027 | 2331 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2332 | |
e6247027 | 2333 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2334 | { |
2335 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2336 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2337 | else |
2338 | dev_kfree_skb(skb); | |
2339 | } | |
e6247027 | 2340 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2341 | |
2342 | ||
bea3348e SH |
2343 | /** |
2344 | * netif_device_detach - mark device as removed | |
2345 | * @dev: network device | |
2346 | * | |
2347 | * Mark device as removed from system and therefore no longer available. | |
2348 | */ | |
56079431 DV |
2349 | void netif_device_detach(struct net_device *dev) |
2350 | { | |
2351 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2352 | netif_running(dev)) { | |
d543103a | 2353 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2354 | } |
2355 | } | |
2356 | EXPORT_SYMBOL(netif_device_detach); | |
2357 | ||
bea3348e SH |
2358 | /** |
2359 | * netif_device_attach - mark device as attached | |
2360 | * @dev: network device | |
2361 | * | |
2362 | * Mark device as attached from system and restart if needed. | |
2363 | */ | |
56079431 DV |
2364 | void netif_device_attach(struct net_device *dev) |
2365 | { | |
2366 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2367 | netif_running(dev)) { | |
d543103a | 2368 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2369 | __netdev_watchdog_up(dev); |
56079431 DV |
2370 | } |
2371 | } | |
2372 | EXPORT_SYMBOL(netif_device_attach); | |
2373 | ||
5605c762 JP |
2374 | /* |
2375 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2376 | * to be used as a distribution range. | |
2377 | */ | |
2378 | u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, | |
2379 | unsigned int num_tx_queues) | |
2380 | { | |
2381 | u32 hash; | |
2382 | u16 qoffset = 0; | |
2383 | u16 qcount = num_tx_queues; | |
2384 | ||
2385 | if (skb_rx_queue_recorded(skb)) { | |
2386 | hash = skb_get_rx_queue(skb); | |
2387 | while (unlikely(hash >= num_tx_queues)) | |
2388 | hash -= num_tx_queues; | |
2389 | return hash; | |
2390 | } | |
2391 | ||
2392 | if (dev->num_tc) { | |
2393 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
2394 | qoffset = dev->tc_to_txq[tc].offset; | |
2395 | qcount = dev->tc_to_txq[tc].count; | |
2396 | } | |
2397 | ||
2398 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
2399 | } | |
2400 | EXPORT_SYMBOL(__skb_tx_hash); | |
2401 | ||
36c92474 BH |
2402 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2403 | { | |
65e9d2fa | 2404 | static const netdev_features_t null_features = 0; |
36c92474 | 2405 | struct net_device *dev = skb->dev; |
88ad4175 | 2406 | const char *name = ""; |
36c92474 | 2407 | |
c846ad9b BG |
2408 | if (!net_ratelimit()) |
2409 | return; | |
2410 | ||
88ad4175 BM |
2411 | if (dev) { |
2412 | if (dev->dev.parent) | |
2413 | name = dev_driver_string(dev->dev.parent); | |
2414 | else | |
2415 | name = netdev_name(dev); | |
2416 | } | |
36c92474 BH |
2417 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " |
2418 | "gso_type=%d ip_summed=%d\n", | |
88ad4175 | 2419 | name, dev ? &dev->features : &null_features, |
65e9d2fa | 2420 | skb->sk ? &skb->sk->sk_route_caps : &null_features, |
36c92474 BH |
2421 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2422 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2423 | } | |
2424 | ||
1da177e4 LT |
2425 | /* |
2426 | * Invalidate hardware checksum when packet is to be mangled, and | |
2427 | * complete checksum manually on outgoing path. | |
2428 | */ | |
84fa7933 | 2429 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2430 | { |
d3bc23e7 | 2431 | __wsum csum; |
663ead3b | 2432 | int ret = 0, offset; |
1da177e4 | 2433 | |
84fa7933 | 2434 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2435 | goto out_set_summed; |
2436 | ||
2437 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2438 | skb_warn_bad_offload(skb); |
2439 | return -EINVAL; | |
1da177e4 LT |
2440 | } |
2441 | ||
cef401de ED |
2442 | /* Before computing a checksum, we should make sure no frag could |
2443 | * be modified by an external entity : checksum could be wrong. | |
2444 | */ | |
2445 | if (skb_has_shared_frag(skb)) { | |
2446 | ret = __skb_linearize(skb); | |
2447 | if (ret) | |
2448 | goto out; | |
2449 | } | |
2450 | ||
55508d60 | 2451 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2452 | BUG_ON(offset >= skb_headlen(skb)); |
2453 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2454 | ||
2455 | offset += skb->csum_offset; | |
2456 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2457 | ||
2458 | if (skb_cloned(skb) && | |
2459 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2460 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2461 | if (ret) | |
2462 | goto out; | |
2463 | } | |
2464 | ||
a030847e | 2465 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2466 | out_set_summed: |
1da177e4 | 2467 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2468 | out: |
1da177e4 LT |
2469 | return ret; |
2470 | } | |
d1b19dff | 2471 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2472 | |
53d6471c | 2473 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2474 | { |
252e3346 | 2475 | __be16 type = skb->protocol; |
f6a78bfc | 2476 | |
19acc327 PS |
2477 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2478 | if (type == htons(ETH_P_TEB)) { | |
2479 | struct ethhdr *eth; | |
2480 | ||
2481 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2482 | return 0; | |
2483 | ||
2484 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2485 | type = eth->h_proto; | |
2486 | } | |
2487 | ||
d4bcef3f | 2488 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
2489 | } |
2490 | ||
2491 | /** | |
2492 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2493 | * @skb: buffer to segment | |
2494 | * @features: features for the output path (see dev->features) | |
2495 | */ | |
2496 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2497 | netdev_features_t features) | |
2498 | { | |
2499 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2500 | struct packet_offload *ptype; | |
53d6471c VY |
2501 | int vlan_depth = skb->mac_len; |
2502 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2503 | |
2504 | if (unlikely(!type)) | |
2505 | return ERR_PTR(-EINVAL); | |
2506 | ||
53d6471c | 2507 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2508 | |
2509 | rcu_read_lock(); | |
22061d80 | 2510 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2511 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2512 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2513 | break; |
2514 | } | |
2515 | } | |
2516 | rcu_read_unlock(); | |
2517 | ||
98e399f8 | 2518 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2519 | |
f6a78bfc HX |
2520 | return segs; |
2521 | } | |
05e8ef4a PS |
2522 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2523 | ||
2524 | ||
2525 | /* openvswitch calls this on rx path, so we need a different check. | |
2526 | */ | |
2527 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2528 | { | |
2529 | if (tx_path) | |
2530 | return skb->ip_summed != CHECKSUM_PARTIAL; | |
2531 | else | |
2532 | return skb->ip_summed == CHECKSUM_NONE; | |
2533 | } | |
2534 | ||
2535 | /** | |
2536 | * __skb_gso_segment - Perform segmentation on skb. | |
2537 | * @skb: buffer to segment | |
2538 | * @features: features for the output path (see dev->features) | |
2539 | * @tx_path: whether it is called in TX path | |
2540 | * | |
2541 | * This function segments the given skb and returns a list of segments. | |
2542 | * | |
2543 | * It may return NULL if the skb requires no segmentation. This is | |
2544 | * only possible when GSO is used for verifying header integrity. | |
2545 | */ | |
2546 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2547 | netdev_features_t features, bool tx_path) | |
2548 | { | |
2549 | if (unlikely(skb_needs_check(skb, tx_path))) { | |
2550 | int err; | |
2551 | ||
2552 | skb_warn_bad_offload(skb); | |
2553 | ||
a40e0a66 | 2554 | err = skb_cow_head(skb, 0); |
2555 | if (err < 0) | |
05e8ef4a PS |
2556 | return ERR_PTR(err); |
2557 | } | |
2558 | ||
68c33163 | 2559 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2560 | SKB_GSO_CB(skb)->encap_level = 0; |
2561 | ||
05e8ef4a PS |
2562 | skb_reset_mac_header(skb); |
2563 | skb_reset_mac_len(skb); | |
2564 | ||
2565 | return skb_mac_gso_segment(skb, features); | |
2566 | } | |
12b0004d | 2567 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2568 | |
fb286bb2 HX |
2569 | /* Take action when hardware reception checksum errors are detected. */ |
2570 | #ifdef CONFIG_BUG | |
2571 | void netdev_rx_csum_fault(struct net_device *dev) | |
2572 | { | |
2573 | if (net_ratelimit()) { | |
7b6cd1ce | 2574 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2575 | dump_stack(); |
2576 | } | |
2577 | } | |
2578 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2579 | #endif | |
2580 | ||
1da177e4 LT |
2581 | /* Actually, we should eliminate this check as soon as we know, that: |
2582 | * 1. IOMMU is present and allows to map all the memory. | |
2583 | * 2. No high memory really exists on this machine. | |
2584 | */ | |
2585 | ||
c1e756bf | 2586 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2587 | { |
3d3a8533 | 2588 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2589 | int i; |
5acbbd42 | 2590 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2591 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2592 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2593 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2594 | return 1; |
ea2ab693 | 2595 | } |
5acbbd42 | 2596 | } |
1da177e4 | 2597 | |
5acbbd42 FT |
2598 | if (PCI_DMA_BUS_IS_PHYS) { |
2599 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2600 | |
9092c658 ED |
2601 | if (!pdev) |
2602 | return 0; | |
5acbbd42 | 2603 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2604 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2605 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2606 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2607 | return 1; | |
2608 | } | |
2609 | } | |
3d3a8533 | 2610 | #endif |
1da177e4 LT |
2611 | return 0; |
2612 | } | |
1da177e4 | 2613 | |
3b392ddb SH |
2614 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2615 | * instead of standard features for the netdev. | |
2616 | */ | |
d0edc7bf | 2617 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
2618 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
2619 | netdev_features_t features, | |
2620 | __be16 type) | |
2621 | { | |
25cd9ba0 | 2622 | if (eth_p_mpls(type)) |
3b392ddb SH |
2623 | features &= skb->dev->mpls_features; |
2624 | ||
2625 | return features; | |
2626 | } | |
2627 | #else | |
2628 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2629 | netdev_features_t features, | |
2630 | __be16 type) | |
2631 | { | |
2632 | return features; | |
2633 | } | |
2634 | #endif | |
2635 | ||
c8f44aff | 2636 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2637 | netdev_features_t features) |
f01a5236 | 2638 | { |
53d6471c | 2639 | int tmp; |
3b392ddb SH |
2640 | __be16 type; |
2641 | ||
2642 | type = skb_network_protocol(skb, &tmp); | |
2643 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2644 | |
c0d680e5 | 2645 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2646 | !can_checksum_protocol(features, type)) { |
f01a5236 | 2647 | features &= ~NETIF_F_ALL_CSUM; |
c1e756bf | 2648 | } else if (illegal_highdma(skb->dev, skb)) { |
f01a5236 JG |
2649 | features &= ~NETIF_F_SG; |
2650 | } | |
2651 | ||
2652 | return features; | |
2653 | } | |
2654 | ||
e38f3025 TM |
2655 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
2656 | struct net_device *dev, | |
2657 | netdev_features_t features) | |
2658 | { | |
2659 | return features; | |
2660 | } | |
2661 | EXPORT_SYMBOL(passthru_features_check); | |
2662 | ||
8cb65d00 TM |
2663 | static netdev_features_t dflt_features_check(const struct sk_buff *skb, |
2664 | struct net_device *dev, | |
2665 | netdev_features_t features) | |
2666 | { | |
2667 | return vlan_features_check(skb, features); | |
2668 | } | |
2669 | ||
c1e756bf | 2670 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 2671 | { |
5f35227e | 2672 | struct net_device *dev = skb->dev; |
fcbeb976 ED |
2673 | netdev_features_t features = dev->features; |
2674 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
58e998c6 | 2675 | |
fcbeb976 | 2676 | if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs) |
30b678d8 BH |
2677 | features &= ~NETIF_F_GSO_MASK; |
2678 | ||
5f35227e JG |
2679 | /* If encapsulation offload request, verify we are testing |
2680 | * hardware encapsulation features instead of standard | |
2681 | * features for the netdev | |
2682 | */ | |
2683 | if (skb->encapsulation) | |
2684 | features &= dev->hw_enc_features; | |
2685 | ||
f5a7fb88 TM |
2686 | if (skb_vlan_tagged(skb)) |
2687 | features = netdev_intersect_features(features, | |
2688 | dev->vlan_features | | |
2689 | NETIF_F_HW_VLAN_CTAG_TX | | |
2690 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2691 | |
5f35227e JG |
2692 | if (dev->netdev_ops->ndo_features_check) |
2693 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
2694 | features); | |
8cb65d00 TM |
2695 | else |
2696 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 2697 | |
c1e756bf | 2698 | return harmonize_features(skb, features); |
58e998c6 | 2699 | } |
c1e756bf | 2700 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2701 | |
2ea25513 | 2702 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2703 | struct netdev_queue *txq, bool more) |
f6a78bfc | 2704 | { |
2ea25513 DM |
2705 | unsigned int len; |
2706 | int rc; | |
00829823 | 2707 | |
7866a621 | 2708 | if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) |
2ea25513 | 2709 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 2710 | |
2ea25513 DM |
2711 | len = skb->len; |
2712 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 2713 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 2714 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 2715 | |
2ea25513 DM |
2716 | return rc; |
2717 | } | |
7b9c6090 | 2718 | |
8dcda22a DM |
2719 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
2720 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
2721 | { |
2722 | struct sk_buff *skb = first; | |
2723 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 2724 | |
7f2e870f DM |
2725 | while (skb) { |
2726 | struct sk_buff *next = skb->next; | |
fc70fb64 | 2727 | |
7f2e870f | 2728 | skb->next = NULL; |
95f6b3dd | 2729 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
2730 | if (unlikely(!dev_xmit_complete(rc))) { |
2731 | skb->next = next; | |
2732 | goto out; | |
2733 | } | |
6afff0ca | 2734 | |
7f2e870f DM |
2735 | skb = next; |
2736 | if (netif_xmit_stopped(txq) && skb) { | |
2737 | rc = NETDEV_TX_BUSY; | |
2738 | break; | |
9ccb8975 | 2739 | } |
7f2e870f | 2740 | } |
9ccb8975 | 2741 | |
7f2e870f DM |
2742 | out: |
2743 | *ret = rc; | |
2744 | return skb; | |
2745 | } | |
b40863c6 | 2746 | |
1ff0dc94 ED |
2747 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
2748 | netdev_features_t features) | |
f6a78bfc | 2749 | { |
df8a39de | 2750 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
2751 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
2752 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
2753 | return skb; |
2754 | } | |
f6a78bfc | 2755 | |
55a93b3e | 2756 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
2757 | { |
2758 | netdev_features_t features; | |
f6a78bfc | 2759 | |
eae3f88e DM |
2760 | if (skb->next) |
2761 | return skb; | |
068a2de5 | 2762 | |
eae3f88e DM |
2763 | features = netif_skb_features(skb); |
2764 | skb = validate_xmit_vlan(skb, features); | |
2765 | if (unlikely(!skb)) | |
2766 | goto out_null; | |
7b9c6090 | 2767 | |
8b86a61d | 2768 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
2769 | struct sk_buff *segs; |
2770 | ||
2771 | segs = skb_gso_segment(skb, features); | |
cecda693 | 2772 | if (IS_ERR(segs)) { |
af6dabc9 | 2773 | goto out_kfree_skb; |
cecda693 JW |
2774 | } else if (segs) { |
2775 | consume_skb(skb); | |
2776 | skb = segs; | |
f6a78bfc | 2777 | } |
eae3f88e DM |
2778 | } else { |
2779 | if (skb_needs_linearize(skb, features) && | |
2780 | __skb_linearize(skb)) | |
2781 | goto out_kfree_skb; | |
4ec93edb | 2782 | |
eae3f88e DM |
2783 | /* If packet is not checksummed and device does not |
2784 | * support checksumming for this protocol, complete | |
2785 | * checksumming here. | |
2786 | */ | |
2787 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
2788 | if (skb->encapsulation) | |
2789 | skb_set_inner_transport_header(skb, | |
2790 | skb_checksum_start_offset(skb)); | |
2791 | else | |
2792 | skb_set_transport_header(skb, | |
2793 | skb_checksum_start_offset(skb)); | |
2794 | if (!(features & NETIF_F_ALL_CSUM) && | |
2795 | skb_checksum_help(skb)) | |
2796 | goto out_kfree_skb; | |
7b9c6090 | 2797 | } |
0c772159 | 2798 | } |
7b9c6090 | 2799 | |
eae3f88e | 2800 | return skb; |
fc70fb64 | 2801 | |
f6a78bfc HX |
2802 | out_kfree_skb: |
2803 | kfree_skb(skb); | |
eae3f88e DM |
2804 | out_null: |
2805 | return NULL; | |
2806 | } | |
6afff0ca | 2807 | |
55a93b3e ED |
2808 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
2809 | { | |
2810 | struct sk_buff *next, *head = NULL, *tail; | |
2811 | ||
bec3cfdc | 2812 | for (; skb != NULL; skb = next) { |
55a93b3e ED |
2813 | next = skb->next; |
2814 | skb->next = NULL; | |
bec3cfdc ED |
2815 | |
2816 | /* in case skb wont be segmented, point to itself */ | |
2817 | skb->prev = skb; | |
2818 | ||
55a93b3e | 2819 | skb = validate_xmit_skb(skb, dev); |
bec3cfdc ED |
2820 | if (!skb) |
2821 | continue; | |
55a93b3e | 2822 | |
bec3cfdc ED |
2823 | if (!head) |
2824 | head = skb; | |
2825 | else | |
2826 | tail->next = skb; | |
2827 | /* If skb was segmented, skb->prev points to | |
2828 | * the last segment. If not, it still contains skb. | |
2829 | */ | |
2830 | tail = skb->prev; | |
55a93b3e ED |
2831 | } |
2832 | return head; | |
f6a78bfc HX |
2833 | } |
2834 | ||
1def9238 ED |
2835 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
2836 | { | |
2837 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2838 | ||
2839 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2840 | ||
2841 | /* To get more precise estimation of bytes sent on wire, | |
2842 | * we add to pkt_len the headers size of all segments | |
2843 | */ | |
2844 | if (shinfo->gso_size) { | |
757b8b1d | 2845 | unsigned int hdr_len; |
15e5a030 | 2846 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 2847 | |
757b8b1d ED |
2848 | /* mac layer + network layer */ |
2849 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
2850 | ||
2851 | /* + transport layer */ | |
1def9238 ED |
2852 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
2853 | hdr_len += tcp_hdrlen(skb); | |
2854 | else | |
2855 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
2856 | |
2857 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
2858 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
2859 | shinfo->gso_size); | |
2860 | ||
2861 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
2862 | } |
2863 | } | |
2864 | ||
bbd8a0d3 KK |
2865 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2866 | struct net_device *dev, | |
2867 | struct netdev_queue *txq) | |
2868 | { | |
2869 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2870 | bool contended; |
bbd8a0d3 KK |
2871 | int rc; |
2872 | ||
1def9238 | 2873 | qdisc_pkt_len_init(skb); |
a2da570d | 2874 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
2875 | /* |
2876 | * Heuristic to force contended enqueues to serialize on a | |
2877 | * separate lock before trying to get qdisc main lock. | |
9bf2b8c2 YX |
2878 | * This permits __QDISC___STATE_RUNNING owner to get the lock more |
2879 | * often and dequeue packets faster. | |
79640a4c | 2880 | */ |
a2da570d | 2881 | contended = qdisc_is_running(q); |
79640a4c ED |
2882 | if (unlikely(contended)) |
2883 | spin_lock(&q->busylock); | |
2884 | ||
bbd8a0d3 KK |
2885 | spin_lock(root_lock); |
2886 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2887 | kfree_skb(skb); | |
2888 | rc = NET_XMIT_DROP; | |
2889 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2890 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2891 | /* |
2892 | * This is a work-conserving queue; there are no old skbs | |
2893 | * waiting to be sent out; and the qdisc is not running - | |
2894 | * xmit the skb directly. | |
2895 | */ | |
bfe0d029 | 2896 | |
bfe0d029 ED |
2897 | qdisc_bstats_update(q, skb); |
2898 | ||
55a93b3e | 2899 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
2900 | if (unlikely(contended)) { |
2901 | spin_unlock(&q->busylock); | |
2902 | contended = false; | |
2903 | } | |
bbd8a0d3 | 2904 | __qdisc_run(q); |
79640a4c | 2905 | } else |
bc135b23 | 2906 | qdisc_run_end(q); |
bbd8a0d3 KK |
2907 | |
2908 | rc = NET_XMIT_SUCCESS; | |
2909 | } else { | |
a2da570d | 2910 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2911 | if (qdisc_run_begin(q)) { |
2912 | if (unlikely(contended)) { | |
2913 | spin_unlock(&q->busylock); | |
2914 | contended = false; | |
2915 | } | |
2916 | __qdisc_run(q); | |
2917 | } | |
bbd8a0d3 KK |
2918 | } |
2919 | spin_unlock(root_lock); | |
79640a4c ED |
2920 | if (unlikely(contended)) |
2921 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2922 | return rc; |
2923 | } | |
2924 | ||
86f8515f | 2925 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
2926 | static void skb_update_prio(struct sk_buff *skb) |
2927 | { | |
6977a79d | 2928 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 2929 | |
91c68ce2 ED |
2930 | if (!skb->priority && skb->sk && map) { |
2931 | unsigned int prioidx = skb->sk->sk_cgrp_prioidx; | |
2932 | ||
2933 | if (prioidx < map->priomap_len) | |
2934 | skb->priority = map->priomap[prioidx]; | |
2935 | } | |
5bc1421e NH |
2936 | } |
2937 | #else | |
2938 | #define skb_update_prio(skb) | |
2939 | #endif | |
2940 | ||
f60e5990 | 2941 | DEFINE_PER_CPU(int, xmit_recursion); |
2942 | EXPORT_SYMBOL(xmit_recursion); | |
2943 | ||
11a766ce | 2944 | #define RECURSION_LIMIT 10 |
745e20f1 | 2945 | |
95603e22 MM |
2946 | /** |
2947 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
2948 | * @net: network namespace this loopback is happening in |
2949 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
2950 | * @skb: buffer to transmit |
2951 | */ | |
0c4b51f0 | 2952 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
2953 | { |
2954 | skb_reset_mac_header(skb); | |
2955 | __skb_pull(skb, skb_network_offset(skb)); | |
2956 | skb->pkt_type = PACKET_LOOPBACK; | |
2957 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2958 | WARN_ON(!skb_dst(skb)); | |
2959 | skb_dst_force(skb); | |
2960 | netif_rx_ni(skb); | |
2961 | return 0; | |
2962 | } | |
2963 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2964 | ||
638b2a69 JP |
2965 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
2966 | { | |
2967 | #ifdef CONFIG_XPS | |
2968 | struct xps_dev_maps *dev_maps; | |
2969 | struct xps_map *map; | |
2970 | int queue_index = -1; | |
2971 | ||
2972 | rcu_read_lock(); | |
2973 | dev_maps = rcu_dereference(dev->xps_maps); | |
2974 | if (dev_maps) { | |
2975 | map = rcu_dereference( | |
2976 | dev_maps->cpu_map[skb->sender_cpu - 1]); | |
2977 | if (map) { | |
2978 | if (map->len == 1) | |
2979 | queue_index = map->queues[0]; | |
2980 | else | |
2981 | queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), | |
2982 | map->len)]; | |
2983 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
2984 | queue_index = -1; | |
2985 | } | |
2986 | } | |
2987 | rcu_read_unlock(); | |
2988 | ||
2989 | return queue_index; | |
2990 | #else | |
2991 | return -1; | |
2992 | #endif | |
2993 | } | |
2994 | ||
2995 | static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) | |
2996 | { | |
2997 | struct sock *sk = skb->sk; | |
2998 | int queue_index = sk_tx_queue_get(sk); | |
2999 | ||
3000 | if (queue_index < 0 || skb->ooo_okay || | |
3001 | queue_index >= dev->real_num_tx_queues) { | |
3002 | int new_index = get_xps_queue(dev, skb); | |
3003 | if (new_index < 0) | |
3004 | new_index = skb_tx_hash(dev, skb); | |
3005 | ||
3006 | if (queue_index != new_index && sk && | |
004a5d01 | 3007 | sk_fullsock(sk) && |
638b2a69 JP |
3008 | rcu_access_pointer(sk->sk_dst_cache)) |
3009 | sk_tx_queue_set(sk, new_index); | |
3010 | ||
3011 | queue_index = new_index; | |
3012 | } | |
3013 | ||
3014 | return queue_index; | |
3015 | } | |
3016 | ||
3017 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, | |
3018 | struct sk_buff *skb, | |
3019 | void *accel_priv) | |
3020 | { | |
3021 | int queue_index = 0; | |
3022 | ||
3023 | #ifdef CONFIG_XPS | |
3024 | if (skb->sender_cpu == 0) | |
3025 | skb->sender_cpu = raw_smp_processor_id() + 1; | |
3026 | #endif | |
3027 | ||
3028 | if (dev->real_num_tx_queues != 1) { | |
3029 | const struct net_device_ops *ops = dev->netdev_ops; | |
3030 | if (ops->ndo_select_queue) | |
3031 | queue_index = ops->ndo_select_queue(dev, skb, accel_priv, | |
3032 | __netdev_pick_tx); | |
3033 | else | |
3034 | queue_index = __netdev_pick_tx(dev, skb); | |
3035 | ||
3036 | if (!accel_priv) | |
3037 | queue_index = netdev_cap_txqueue(dev, queue_index); | |
3038 | } | |
3039 | ||
3040 | skb_set_queue_mapping(skb, queue_index); | |
3041 | return netdev_get_tx_queue(dev, queue_index); | |
3042 | } | |
3043 | ||
d29f749e | 3044 | /** |
9d08dd3d | 3045 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3046 | * @skb: buffer to transmit |
9d08dd3d | 3047 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
3048 | * |
3049 | * Queue a buffer for transmission to a network device. The caller must | |
3050 | * have set the device and priority and built the buffer before calling | |
3051 | * this function. The function can be called from an interrupt. | |
3052 | * | |
3053 | * A negative errno code is returned on a failure. A success does not | |
3054 | * guarantee the frame will be transmitted as it may be dropped due | |
3055 | * to congestion or traffic shaping. | |
3056 | * | |
3057 | * ----------------------------------------------------------------------------------- | |
3058 | * I notice this method can also return errors from the queue disciplines, | |
3059 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3060 | * be positive. | |
3061 | * | |
3062 | * Regardless of the return value, the skb is consumed, so it is currently | |
3063 | * difficult to retry a send to this method. (You can bump the ref count | |
3064 | * before sending to hold a reference for retry if you are careful.) | |
3065 | * | |
3066 | * When calling this method, interrupts MUST be enabled. This is because | |
3067 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3068 | * --BLG | |
3069 | */ | |
0a59f3a9 | 3070 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
3071 | { |
3072 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3073 | struct netdev_queue *txq; |
1da177e4 LT |
3074 | struct Qdisc *q; |
3075 | int rc = -ENOMEM; | |
3076 | ||
6d1ccff6 ED |
3077 | skb_reset_mac_header(skb); |
3078 | ||
e7fd2885 WB |
3079 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3080 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3081 | ||
4ec93edb YH |
3082 | /* Disable soft irqs for various locks below. Also |
3083 | * stops preemption for RCU. | |
1da177e4 | 3084 | */ |
4ec93edb | 3085 | rcu_read_lock_bh(); |
1da177e4 | 3086 | |
5bc1421e NH |
3087 | skb_update_prio(skb); |
3088 | ||
02875878 ED |
3089 | /* If device/qdisc don't need skb->dst, release it right now while |
3090 | * its hot in this cpu cache. | |
3091 | */ | |
3092 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
3093 | skb_dst_drop(skb); | |
3094 | else | |
3095 | skb_dst_force(skb); | |
3096 | ||
0c4f691f SF |
3097 | #ifdef CONFIG_NET_SWITCHDEV |
3098 | /* Don't forward if offload device already forwarded */ | |
3099 | if (skb->offload_fwd_mark && | |
3100 | skb->offload_fwd_mark == dev->offload_fwd_mark) { | |
3101 | consume_skb(skb); | |
3102 | rc = NET_XMIT_SUCCESS; | |
3103 | goto out; | |
3104 | } | |
3105 | #endif | |
3106 | ||
f663dd9a | 3107 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 3108 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 3109 | |
1da177e4 | 3110 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 3111 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 3112 | #endif |
cf66ba58 | 3113 | trace_net_dev_queue(skb); |
1da177e4 | 3114 | if (q->enqueue) { |
bbd8a0d3 | 3115 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 3116 | goto out; |
1da177e4 LT |
3117 | } |
3118 | ||
3119 | /* The device has no queue. Common case for software devices: | |
3120 | loopback, all the sorts of tunnels... | |
3121 | ||
932ff279 HX |
3122 | Really, it is unlikely that netif_tx_lock protection is necessary |
3123 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
3124 | counters.) |
3125 | However, it is possible, that they rely on protection | |
3126 | made by us here. | |
3127 | ||
3128 | Check this and shot the lock. It is not prone from deadlocks. | |
3129 | Either shot noqueue qdisc, it is even simpler 8) | |
3130 | */ | |
3131 | if (dev->flags & IFF_UP) { | |
3132 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
3133 | ||
c773e847 | 3134 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 3135 | |
745e20f1 ED |
3136 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
3137 | goto recursion_alert; | |
3138 | ||
1f59533f JDB |
3139 | skb = validate_xmit_skb(skb, dev); |
3140 | if (!skb) | |
3141 | goto drop; | |
3142 | ||
c773e847 | 3143 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 3144 | |
73466498 | 3145 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 3146 | __this_cpu_inc(xmit_recursion); |
ce93718f | 3147 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 3148 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 3149 | if (dev_xmit_complete(rc)) { |
c773e847 | 3150 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
3151 | goto out; |
3152 | } | |
3153 | } | |
c773e847 | 3154 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
3155 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
3156 | dev->name); | |
1da177e4 LT |
3157 | } else { |
3158 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
3159 | * unfortunately |
3160 | */ | |
3161 | recursion_alert: | |
e87cc472 JP |
3162 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
3163 | dev->name); | |
1da177e4 LT |
3164 | } |
3165 | } | |
3166 | ||
3167 | rc = -ENETDOWN; | |
1f59533f | 3168 | drop: |
d4828d85 | 3169 | rcu_read_unlock_bh(); |
1da177e4 | 3170 | |
015f0688 | 3171 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 3172 | kfree_skb_list(skb); |
1da177e4 LT |
3173 | return rc; |
3174 | out: | |
d4828d85 | 3175 | rcu_read_unlock_bh(); |
1da177e4 LT |
3176 | return rc; |
3177 | } | |
f663dd9a | 3178 | |
2b4aa3ce | 3179 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
3180 | { |
3181 | return __dev_queue_xmit(skb, NULL); | |
3182 | } | |
2b4aa3ce | 3183 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3184 | |
f663dd9a JW |
3185 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3186 | { | |
3187 | return __dev_queue_xmit(skb, accel_priv); | |
3188 | } | |
3189 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3190 | ||
1da177e4 LT |
3191 | |
3192 | /*======================================================================= | |
3193 | Receiver routines | |
3194 | =======================================================================*/ | |
3195 | ||
6b2bedc3 | 3196 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3197 | EXPORT_SYMBOL(netdev_max_backlog); |
3198 | ||
3b098e2d | 3199 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
3200 | int netdev_budget __read_mostly = 300; |
3201 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 3202 | |
eecfd7c4 ED |
3203 | /* Called with irq disabled */ |
3204 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3205 | struct napi_struct *napi) | |
3206 | { | |
3207 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3208 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3209 | } | |
3210 | ||
bfb564e7 KK |
3211 | #ifdef CONFIG_RPS |
3212 | ||
3213 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3214 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 3215 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
3216 | u32 rps_cpu_mask __read_mostly; |
3217 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 3218 | |
c5905afb | 3219 | struct static_key rps_needed __read_mostly; |
adc9300e | 3220 | |
c445477d BH |
3221 | static struct rps_dev_flow * |
3222 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3223 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3224 | { | |
a31196b0 | 3225 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
3226 | #ifdef CONFIG_RFS_ACCEL |
3227 | struct netdev_rx_queue *rxqueue; | |
3228 | struct rps_dev_flow_table *flow_table; | |
3229 | struct rps_dev_flow *old_rflow; | |
3230 | u32 flow_id; | |
3231 | u16 rxq_index; | |
3232 | int rc; | |
3233 | ||
3234 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3235 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3236 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3237 | goto out; |
3238 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3239 | if (rxq_index == skb_get_rx_queue(skb)) | |
3240 | goto out; | |
3241 | ||
3242 | rxqueue = dev->_rx + rxq_index; | |
3243 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3244 | if (!flow_table) | |
3245 | goto out; | |
61b905da | 3246 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3247 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3248 | rxq_index, flow_id); | |
3249 | if (rc < 0) | |
3250 | goto out; | |
3251 | old_rflow = rflow; | |
3252 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3253 | rflow->filter = rc; |
3254 | if (old_rflow->filter == rflow->filter) | |
3255 | old_rflow->filter = RPS_NO_FILTER; | |
3256 | out: | |
3257 | #endif | |
3258 | rflow->last_qtail = | |
09994d1b | 3259 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3260 | } |
3261 | ||
09994d1b | 3262 | rflow->cpu = next_cpu; |
c445477d BH |
3263 | return rflow; |
3264 | } | |
3265 | ||
bfb564e7 KK |
3266 | /* |
3267 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3268 | * CPU from the RPS map of the receiving queue for a given skb. | |
3269 | * rcu_read_lock must be held on entry. | |
3270 | */ | |
3271 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3272 | struct rps_dev_flow **rflowp) | |
3273 | { | |
567e4b79 ED |
3274 | const struct rps_sock_flow_table *sock_flow_table; |
3275 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 3276 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 3277 | struct rps_map *map; |
bfb564e7 | 3278 | int cpu = -1; |
567e4b79 | 3279 | u32 tcpu; |
61b905da | 3280 | u32 hash; |
bfb564e7 KK |
3281 | |
3282 | if (skb_rx_queue_recorded(skb)) { | |
3283 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 3284 | |
62fe0b40 BH |
3285 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3286 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3287 | "%s received packet on queue %u, but number " | |
3288 | "of RX queues is %u\n", | |
3289 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3290 | goto done; |
3291 | } | |
567e4b79 ED |
3292 | rxqueue += index; |
3293 | } | |
bfb564e7 | 3294 | |
567e4b79 ED |
3295 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
3296 | ||
3297 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 3298 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 3299 | if (!flow_table && !map) |
bfb564e7 KK |
3300 | goto done; |
3301 | ||
2d47b459 | 3302 | skb_reset_network_header(skb); |
61b905da TH |
3303 | hash = skb_get_hash(skb); |
3304 | if (!hash) | |
bfb564e7 KK |
3305 | goto done; |
3306 | ||
fec5e652 TH |
3307 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
3308 | if (flow_table && sock_flow_table) { | |
fec5e652 | 3309 | struct rps_dev_flow *rflow; |
567e4b79 ED |
3310 | u32 next_cpu; |
3311 | u32 ident; | |
3312 | ||
3313 | /* First check into global flow table if there is a match */ | |
3314 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
3315 | if ((ident ^ hash) & ~rps_cpu_mask) | |
3316 | goto try_rps; | |
fec5e652 | 3317 | |
567e4b79 ED |
3318 | next_cpu = ident & rps_cpu_mask; |
3319 | ||
3320 | /* OK, now we know there is a match, | |
3321 | * we can look at the local (per receive queue) flow table | |
3322 | */ | |
61b905da | 3323 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3324 | tcpu = rflow->cpu; |
3325 | ||
fec5e652 TH |
3326 | /* |
3327 | * If the desired CPU (where last recvmsg was done) is | |
3328 | * different from current CPU (one in the rx-queue flow | |
3329 | * table entry), switch if one of the following holds: | |
a31196b0 | 3330 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
3331 | * - Current CPU is offline. |
3332 | * - The current CPU's queue tail has advanced beyond the | |
3333 | * last packet that was enqueued using this table entry. | |
3334 | * This guarantees that all previous packets for the flow | |
3335 | * have been dequeued, thus preserving in order delivery. | |
3336 | */ | |
3337 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 3338 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 3339 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
3340 | rflow->last_qtail)) >= 0)) { |
3341 | tcpu = next_cpu; | |
c445477d | 3342 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3343 | } |
c445477d | 3344 | |
a31196b0 | 3345 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
3346 | *rflowp = rflow; |
3347 | cpu = tcpu; | |
3348 | goto done; | |
3349 | } | |
3350 | } | |
3351 | ||
567e4b79 ED |
3352 | try_rps: |
3353 | ||
0a9627f2 | 3354 | if (map) { |
8fc54f68 | 3355 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3356 | if (cpu_online(tcpu)) { |
3357 | cpu = tcpu; | |
3358 | goto done; | |
3359 | } | |
3360 | } | |
3361 | ||
3362 | done: | |
0a9627f2 TH |
3363 | return cpu; |
3364 | } | |
3365 | ||
c445477d BH |
3366 | #ifdef CONFIG_RFS_ACCEL |
3367 | ||
3368 | /** | |
3369 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3370 | * @dev: Device on which the filter was set | |
3371 | * @rxq_index: RX queue index | |
3372 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3373 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3374 | * | |
3375 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3376 | * this function for each installed filter and remove the filters for | |
3377 | * which it returns %true. | |
3378 | */ | |
3379 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3380 | u32 flow_id, u16 filter_id) | |
3381 | { | |
3382 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3383 | struct rps_dev_flow_table *flow_table; | |
3384 | struct rps_dev_flow *rflow; | |
3385 | bool expire = true; | |
a31196b0 | 3386 | unsigned int cpu; |
c445477d BH |
3387 | |
3388 | rcu_read_lock(); | |
3389 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3390 | if (flow_table && flow_id <= flow_table->mask) { | |
3391 | rflow = &flow_table->flows[flow_id]; | |
3392 | cpu = ACCESS_ONCE(rflow->cpu); | |
a31196b0 | 3393 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
3394 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
3395 | rflow->last_qtail) < | |
3396 | (int)(10 * flow_table->mask))) | |
3397 | expire = false; | |
3398 | } | |
3399 | rcu_read_unlock(); | |
3400 | return expire; | |
3401 | } | |
3402 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3403 | ||
3404 | #endif /* CONFIG_RFS_ACCEL */ | |
3405 | ||
0a9627f2 | 3406 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3407 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3408 | { |
e36fa2f7 ED |
3409 | struct softnet_data *sd = data; |
3410 | ||
eecfd7c4 | 3411 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3412 | sd->received_rps++; |
0a9627f2 | 3413 | } |
e36fa2f7 | 3414 | |
fec5e652 | 3415 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3416 | |
e36fa2f7 ED |
3417 | /* |
3418 | * Check if this softnet_data structure is another cpu one | |
3419 | * If yes, queue it to our IPI list and return 1 | |
3420 | * If no, return 0 | |
3421 | */ | |
3422 | static int rps_ipi_queued(struct softnet_data *sd) | |
3423 | { | |
3424 | #ifdef CONFIG_RPS | |
903ceff7 | 3425 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
3426 | |
3427 | if (sd != mysd) { | |
3428 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3429 | mysd->rps_ipi_list = sd; | |
3430 | ||
3431 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3432 | return 1; | |
3433 | } | |
3434 | #endif /* CONFIG_RPS */ | |
3435 | return 0; | |
3436 | } | |
3437 | ||
99bbc707 WB |
3438 | #ifdef CONFIG_NET_FLOW_LIMIT |
3439 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3440 | #endif | |
3441 | ||
3442 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3443 | { | |
3444 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3445 | struct sd_flow_limit *fl; | |
3446 | struct softnet_data *sd; | |
3447 | unsigned int old_flow, new_flow; | |
3448 | ||
3449 | if (qlen < (netdev_max_backlog >> 1)) | |
3450 | return false; | |
3451 | ||
903ceff7 | 3452 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
3453 | |
3454 | rcu_read_lock(); | |
3455 | fl = rcu_dereference(sd->flow_limit); | |
3456 | if (fl) { | |
3958afa1 | 3457 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3458 | old_flow = fl->history[fl->history_head]; |
3459 | fl->history[fl->history_head] = new_flow; | |
3460 | ||
3461 | fl->history_head++; | |
3462 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3463 | ||
3464 | if (likely(fl->buckets[old_flow])) | |
3465 | fl->buckets[old_flow]--; | |
3466 | ||
3467 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3468 | fl->count++; | |
3469 | rcu_read_unlock(); | |
3470 | return true; | |
3471 | } | |
3472 | } | |
3473 | rcu_read_unlock(); | |
3474 | #endif | |
3475 | return false; | |
3476 | } | |
3477 | ||
0a9627f2 TH |
3478 | /* |
3479 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3480 | * queue (may be a remote CPU queue). | |
3481 | */ | |
fec5e652 TH |
3482 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3483 | unsigned int *qtail) | |
0a9627f2 | 3484 | { |
e36fa2f7 | 3485 | struct softnet_data *sd; |
0a9627f2 | 3486 | unsigned long flags; |
99bbc707 | 3487 | unsigned int qlen; |
0a9627f2 | 3488 | |
e36fa2f7 | 3489 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3490 | |
3491 | local_irq_save(flags); | |
0a9627f2 | 3492 | |
e36fa2f7 | 3493 | rps_lock(sd); |
e9e4dd32 JA |
3494 | if (!netif_running(skb->dev)) |
3495 | goto drop; | |
99bbc707 WB |
3496 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3497 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 3498 | if (qlen) { |
0a9627f2 | 3499 | enqueue: |
e36fa2f7 | 3500 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3501 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3502 | rps_unlock(sd); |
152102c7 | 3503 | local_irq_restore(flags); |
0a9627f2 TH |
3504 | return NET_RX_SUCCESS; |
3505 | } | |
3506 | ||
ebda37c2 ED |
3507 | /* Schedule NAPI for backlog device |
3508 | * We can use non atomic operation since we own the queue lock | |
3509 | */ | |
3510 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3511 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3512 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3513 | } |
3514 | goto enqueue; | |
3515 | } | |
3516 | ||
e9e4dd32 | 3517 | drop: |
dee42870 | 3518 | sd->dropped++; |
e36fa2f7 | 3519 | rps_unlock(sd); |
0a9627f2 | 3520 | |
0a9627f2 TH |
3521 | local_irq_restore(flags); |
3522 | ||
caf586e5 | 3523 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3524 | kfree_skb(skb); |
3525 | return NET_RX_DROP; | |
3526 | } | |
1da177e4 | 3527 | |
ae78dbfa | 3528 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3529 | { |
b0e28f1e | 3530 | int ret; |
1da177e4 | 3531 | |
588f0330 | 3532 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3533 | |
cf66ba58 | 3534 | trace_netif_rx(skb); |
df334545 | 3535 | #ifdef CONFIG_RPS |
c5905afb | 3536 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3537 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3538 | int cpu; |
3539 | ||
cece1945 | 3540 | preempt_disable(); |
b0e28f1e | 3541 | rcu_read_lock(); |
fec5e652 TH |
3542 | |
3543 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3544 | if (cpu < 0) |
3545 | cpu = smp_processor_id(); | |
fec5e652 TH |
3546 | |
3547 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3548 | ||
b0e28f1e | 3549 | rcu_read_unlock(); |
cece1945 | 3550 | preempt_enable(); |
adc9300e ED |
3551 | } else |
3552 | #endif | |
fec5e652 TH |
3553 | { |
3554 | unsigned int qtail; | |
3555 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3556 | put_cpu(); | |
3557 | } | |
b0e28f1e | 3558 | return ret; |
1da177e4 | 3559 | } |
ae78dbfa BH |
3560 | |
3561 | /** | |
3562 | * netif_rx - post buffer to the network code | |
3563 | * @skb: buffer to post | |
3564 | * | |
3565 | * This function receives a packet from a device driver and queues it for | |
3566 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3567 | * may be dropped during processing for congestion control or by the | |
3568 | * protocol layers. | |
3569 | * | |
3570 | * return values: | |
3571 | * NET_RX_SUCCESS (no congestion) | |
3572 | * NET_RX_DROP (packet was dropped) | |
3573 | * | |
3574 | */ | |
3575 | ||
3576 | int netif_rx(struct sk_buff *skb) | |
3577 | { | |
3578 | trace_netif_rx_entry(skb); | |
3579 | ||
3580 | return netif_rx_internal(skb); | |
3581 | } | |
d1b19dff | 3582 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3583 | |
3584 | int netif_rx_ni(struct sk_buff *skb) | |
3585 | { | |
3586 | int err; | |
3587 | ||
ae78dbfa BH |
3588 | trace_netif_rx_ni_entry(skb); |
3589 | ||
1da177e4 | 3590 | preempt_disable(); |
ae78dbfa | 3591 | err = netif_rx_internal(skb); |
1da177e4 LT |
3592 | if (local_softirq_pending()) |
3593 | do_softirq(); | |
3594 | preempt_enable(); | |
3595 | ||
3596 | return err; | |
3597 | } | |
1da177e4 LT |
3598 | EXPORT_SYMBOL(netif_rx_ni); |
3599 | ||
1da177e4 LT |
3600 | static void net_tx_action(struct softirq_action *h) |
3601 | { | |
903ceff7 | 3602 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
3603 | |
3604 | if (sd->completion_queue) { | |
3605 | struct sk_buff *clist; | |
3606 | ||
3607 | local_irq_disable(); | |
3608 | clist = sd->completion_queue; | |
3609 | sd->completion_queue = NULL; | |
3610 | local_irq_enable(); | |
3611 | ||
3612 | while (clist) { | |
3613 | struct sk_buff *skb = clist; | |
3614 | clist = clist->next; | |
3615 | ||
547b792c | 3616 | WARN_ON(atomic_read(&skb->users)); |
e6247027 ED |
3617 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
3618 | trace_consume_skb(skb); | |
3619 | else | |
3620 | trace_kfree_skb(skb, net_tx_action); | |
1da177e4 LT |
3621 | __kfree_skb(skb); |
3622 | } | |
3623 | } | |
3624 | ||
3625 | if (sd->output_queue) { | |
37437bb2 | 3626 | struct Qdisc *head; |
1da177e4 LT |
3627 | |
3628 | local_irq_disable(); | |
3629 | head = sd->output_queue; | |
3630 | sd->output_queue = NULL; | |
a9cbd588 | 3631 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3632 | local_irq_enable(); |
3633 | ||
3634 | while (head) { | |
37437bb2 DM |
3635 | struct Qdisc *q = head; |
3636 | spinlock_t *root_lock; | |
3637 | ||
1da177e4 LT |
3638 | head = head->next_sched; |
3639 | ||
5fb66229 | 3640 | root_lock = qdisc_lock(q); |
37437bb2 | 3641 | if (spin_trylock(root_lock)) { |
4e857c58 | 3642 | smp_mb__before_atomic(); |
def82a1d JP |
3643 | clear_bit(__QDISC_STATE_SCHED, |
3644 | &q->state); | |
37437bb2 DM |
3645 | qdisc_run(q); |
3646 | spin_unlock(root_lock); | |
1da177e4 | 3647 | } else { |
195648bb | 3648 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3649 | &q->state)) { |
195648bb | 3650 | __netif_reschedule(q); |
e8a83e10 | 3651 | } else { |
4e857c58 | 3652 | smp_mb__before_atomic(); |
e8a83e10 JP |
3653 | clear_bit(__QDISC_STATE_SCHED, |
3654 | &q->state); | |
3655 | } | |
1da177e4 LT |
3656 | } |
3657 | } | |
3658 | } | |
3659 | } | |
3660 | ||
ab95bfe0 JP |
3661 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3662 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3663 | /* This hook is defined here for ATM LANE */ |
3664 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3665 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3666 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3667 | #endif |
1da177e4 | 3668 | |
f697c3e8 HX |
3669 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3670 | struct packet_type **pt_prev, | |
3671 | int *ret, struct net_device *orig_dev) | |
3672 | { | |
e7582bab | 3673 | #ifdef CONFIG_NET_CLS_ACT |
d2788d34 DB |
3674 | struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list); |
3675 | struct tcf_result cl_res; | |
24824a09 | 3676 | |
c9e99fd0 DB |
3677 | /* If there's at least one ingress present somewhere (so |
3678 | * we get here via enabled static key), remaining devices | |
3679 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 3680 | * out here. |
c9e99fd0 | 3681 | */ |
d2788d34 | 3682 | if (!cl) |
4577139b | 3683 | return skb; |
f697c3e8 HX |
3684 | if (*pt_prev) { |
3685 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3686 | *pt_prev = NULL; | |
1da177e4 LT |
3687 | } |
3688 | ||
3365495c | 3689 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
c9e99fd0 | 3690 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); |
24ea591d | 3691 | qdisc_bstats_cpu_update(cl->q, skb); |
c9e99fd0 | 3692 | |
3b3ae880 | 3693 | switch (tc_classify(skb, cl, &cl_res, false)) { |
d2788d34 DB |
3694 | case TC_ACT_OK: |
3695 | case TC_ACT_RECLASSIFY: | |
3696 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3697 | break; | |
3698 | case TC_ACT_SHOT: | |
24ea591d | 3699 | qdisc_qstats_cpu_drop(cl->q); |
d2788d34 DB |
3700 | case TC_ACT_STOLEN: |
3701 | case TC_ACT_QUEUED: | |
3702 | kfree_skb(skb); | |
3703 | return NULL; | |
27b29f63 AS |
3704 | case TC_ACT_REDIRECT: |
3705 | /* skb_mac_header check was done by cls/act_bpf, so | |
3706 | * we can safely push the L2 header back before | |
3707 | * redirecting to another netdev | |
3708 | */ | |
3709 | __skb_push(skb, skb->mac_len); | |
3710 | skb_do_redirect(skb); | |
3711 | return NULL; | |
d2788d34 DB |
3712 | default: |
3713 | break; | |
f697c3e8 | 3714 | } |
e7582bab | 3715 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
3716 | return skb; |
3717 | } | |
1da177e4 | 3718 | |
ab95bfe0 JP |
3719 | /** |
3720 | * netdev_rx_handler_register - register receive handler | |
3721 | * @dev: device to register a handler for | |
3722 | * @rx_handler: receive handler to register | |
93e2c32b | 3723 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 3724 | * |
e227867f | 3725 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
3726 | * called from __netif_receive_skb. A negative errno code is returned |
3727 | * on a failure. | |
3728 | * | |
3729 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3730 | * |
3731 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3732 | */ |
3733 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3734 | rx_handler_func_t *rx_handler, |
3735 | void *rx_handler_data) | |
ab95bfe0 JP |
3736 | { |
3737 | ASSERT_RTNL(); | |
3738 | ||
3739 | if (dev->rx_handler) | |
3740 | return -EBUSY; | |
3741 | ||
00cfec37 | 3742 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 3743 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3744 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3745 | ||
3746 | return 0; | |
3747 | } | |
3748 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3749 | ||
3750 | /** | |
3751 | * netdev_rx_handler_unregister - unregister receive handler | |
3752 | * @dev: device to unregister a handler from | |
3753 | * | |
166ec369 | 3754 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
3755 | * |
3756 | * The caller must hold the rtnl_mutex. | |
3757 | */ | |
3758 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3759 | { | |
3760 | ||
3761 | ASSERT_RTNL(); | |
a9b3cd7f | 3762 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
3763 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
3764 | * section has a guarantee to see a non NULL rx_handler_data | |
3765 | * as well. | |
3766 | */ | |
3767 | synchronize_net(); | |
a9b3cd7f | 3768 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
3769 | } |
3770 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3771 | ||
b4b9e355 MG |
3772 | /* |
3773 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3774 | * the special handling of PFMEMALLOC skbs. | |
3775 | */ | |
3776 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3777 | { | |
3778 | switch (skb->protocol) { | |
2b8837ae JP |
3779 | case htons(ETH_P_ARP): |
3780 | case htons(ETH_P_IP): | |
3781 | case htons(ETH_P_IPV6): | |
3782 | case htons(ETH_P_8021Q): | |
3783 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
3784 | return true; |
3785 | default: | |
3786 | return false; | |
3787 | } | |
3788 | } | |
3789 | ||
e687ad60 PN |
3790 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
3791 | int *ret, struct net_device *orig_dev) | |
3792 | { | |
e7582bab | 3793 | #ifdef CONFIG_NETFILTER_INGRESS |
e687ad60 PN |
3794 | if (nf_hook_ingress_active(skb)) { |
3795 | if (*pt_prev) { | |
3796 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3797 | *pt_prev = NULL; | |
3798 | } | |
3799 | ||
3800 | return nf_hook_ingress(skb); | |
3801 | } | |
e7582bab | 3802 | #endif /* CONFIG_NETFILTER_INGRESS */ |
e687ad60 PN |
3803 | return 0; |
3804 | } | |
e687ad60 | 3805 | |
9754e293 | 3806 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
3807 | { |
3808 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3809 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3810 | struct net_device *orig_dev; |
8a4eb573 | 3811 | bool deliver_exact = false; |
1da177e4 | 3812 | int ret = NET_RX_DROP; |
252e3346 | 3813 | __be16 type; |
1da177e4 | 3814 | |
588f0330 | 3815 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3816 | |
cf66ba58 | 3817 | trace_netif_receive_skb(skb); |
9b22ea56 | 3818 | |
cc9bd5ce | 3819 | orig_dev = skb->dev; |
8f903c70 | 3820 | |
c1d2bbe1 | 3821 | skb_reset_network_header(skb); |
fda55eca ED |
3822 | if (!skb_transport_header_was_set(skb)) |
3823 | skb_reset_transport_header(skb); | |
0b5c9db1 | 3824 | skb_reset_mac_len(skb); |
1da177e4 LT |
3825 | |
3826 | pt_prev = NULL; | |
3827 | ||
63d8ea7f | 3828 | another_round: |
b6858177 | 3829 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
3830 | |
3831 | __this_cpu_inc(softnet_data.processed); | |
3832 | ||
8ad227ff PM |
3833 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
3834 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 3835 | skb = skb_vlan_untag(skb); |
bcc6d479 | 3836 | if (unlikely(!skb)) |
2c17d27c | 3837 | goto out; |
bcc6d479 JP |
3838 | } |
3839 | ||
1da177e4 LT |
3840 | #ifdef CONFIG_NET_CLS_ACT |
3841 | if (skb->tc_verd & TC_NCLS) { | |
3842 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3843 | goto ncls; | |
3844 | } | |
3845 | #endif | |
3846 | ||
9754e293 | 3847 | if (pfmemalloc) |
b4b9e355 MG |
3848 | goto skip_taps; |
3849 | ||
1da177e4 | 3850 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
3851 | if (pt_prev) |
3852 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3853 | pt_prev = ptype; | |
3854 | } | |
3855 | ||
3856 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
3857 | if (pt_prev) | |
3858 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3859 | pt_prev = ptype; | |
1da177e4 LT |
3860 | } |
3861 | ||
b4b9e355 | 3862 | skip_taps: |
1cf51900 | 3863 | #ifdef CONFIG_NET_INGRESS |
4577139b DB |
3864 | if (static_key_false(&ingress_needed)) { |
3865 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); | |
3866 | if (!skb) | |
2c17d27c | 3867 | goto out; |
e687ad60 PN |
3868 | |
3869 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 3870 | goto out; |
4577139b | 3871 | } |
1cf51900 PN |
3872 | #endif |
3873 | #ifdef CONFIG_NET_CLS_ACT | |
4577139b | 3874 | skb->tc_verd = 0; |
1da177e4 LT |
3875 | ncls: |
3876 | #endif | |
9754e293 | 3877 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
3878 | goto drop; |
3879 | ||
df8a39de | 3880 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
3881 | if (pt_prev) { |
3882 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3883 | pt_prev = NULL; | |
3884 | } | |
48cc32d3 | 3885 | if (vlan_do_receive(&skb)) |
2425717b JF |
3886 | goto another_round; |
3887 | else if (unlikely(!skb)) | |
2c17d27c | 3888 | goto out; |
2425717b JF |
3889 | } |
3890 | ||
48cc32d3 | 3891 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
3892 | if (rx_handler) { |
3893 | if (pt_prev) { | |
3894 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3895 | pt_prev = NULL; | |
3896 | } | |
8a4eb573 JP |
3897 | switch (rx_handler(&skb)) { |
3898 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 3899 | ret = NET_RX_SUCCESS; |
2c17d27c | 3900 | goto out; |
8a4eb573 | 3901 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3902 | goto another_round; |
8a4eb573 JP |
3903 | case RX_HANDLER_EXACT: |
3904 | deliver_exact = true; | |
3905 | case RX_HANDLER_PASS: | |
3906 | break; | |
3907 | default: | |
3908 | BUG(); | |
3909 | } | |
ab95bfe0 | 3910 | } |
1da177e4 | 3911 | |
df8a39de JP |
3912 | if (unlikely(skb_vlan_tag_present(skb))) { |
3913 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
3914 | skb->pkt_type = PACKET_OTHERHOST; |
3915 | /* Note: we might in the future use prio bits | |
3916 | * and set skb->priority like in vlan_do_receive() | |
3917 | * For the time being, just ignore Priority Code Point | |
3918 | */ | |
3919 | skb->vlan_tci = 0; | |
3920 | } | |
48cc32d3 | 3921 | |
7866a621 SN |
3922 | type = skb->protocol; |
3923 | ||
63d8ea7f | 3924 | /* deliver only exact match when indicated */ |
7866a621 SN |
3925 | if (likely(!deliver_exact)) { |
3926 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
3927 | &ptype_base[ntohs(type) & | |
3928 | PTYPE_HASH_MASK]); | |
3929 | } | |
1f3c8804 | 3930 | |
7866a621 SN |
3931 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
3932 | &orig_dev->ptype_specific); | |
3933 | ||
3934 | if (unlikely(skb->dev != orig_dev)) { | |
3935 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
3936 | &skb->dev->ptype_specific); | |
1da177e4 LT |
3937 | } |
3938 | ||
3939 | if (pt_prev) { | |
1080e512 | 3940 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 3941 | goto drop; |
1080e512 MT |
3942 | else |
3943 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 3944 | } else { |
b4b9e355 | 3945 | drop: |
caf586e5 | 3946 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3947 | kfree_skb(skb); |
3948 | /* Jamal, now you will not able to escape explaining | |
3949 | * me how you were going to use this. :-) | |
3950 | */ | |
3951 | ret = NET_RX_DROP; | |
3952 | } | |
3953 | ||
2c17d27c | 3954 | out: |
9754e293 DM |
3955 | return ret; |
3956 | } | |
3957 | ||
3958 | static int __netif_receive_skb(struct sk_buff *skb) | |
3959 | { | |
3960 | int ret; | |
3961 | ||
3962 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
3963 | unsigned long pflags = current->flags; | |
3964 | ||
3965 | /* | |
3966 | * PFMEMALLOC skbs are special, they should | |
3967 | * - be delivered to SOCK_MEMALLOC sockets only | |
3968 | * - stay away from userspace | |
3969 | * - have bounded memory usage | |
3970 | * | |
3971 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
3972 | * context down to all allocation sites. | |
3973 | */ | |
3974 | current->flags |= PF_MEMALLOC; | |
3975 | ret = __netif_receive_skb_core(skb, true); | |
3976 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
3977 | } else | |
3978 | ret = __netif_receive_skb_core(skb, false); | |
3979 | ||
1da177e4 LT |
3980 | return ret; |
3981 | } | |
0a9627f2 | 3982 | |
ae78dbfa | 3983 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 3984 | { |
2c17d27c JA |
3985 | int ret; |
3986 | ||
588f0330 | 3987 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3988 | |
c1f19b51 RC |
3989 | if (skb_defer_rx_timestamp(skb)) |
3990 | return NET_RX_SUCCESS; | |
3991 | ||
2c17d27c JA |
3992 | rcu_read_lock(); |
3993 | ||
df334545 | 3994 | #ifdef CONFIG_RPS |
c5905afb | 3995 | if (static_key_false(&rps_needed)) { |
3b098e2d | 3996 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 3997 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 3998 | |
3b098e2d ED |
3999 | if (cpu >= 0) { |
4000 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4001 | rcu_read_unlock(); | |
adc9300e | 4002 | return ret; |
3b098e2d | 4003 | } |
fec5e652 | 4004 | } |
1e94d72f | 4005 | #endif |
2c17d27c JA |
4006 | ret = __netif_receive_skb(skb); |
4007 | rcu_read_unlock(); | |
4008 | return ret; | |
0a9627f2 | 4009 | } |
ae78dbfa BH |
4010 | |
4011 | /** | |
4012 | * netif_receive_skb - process receive buffer from network | |
4013 | * @skb: buffer to process | |
4014 | * | |
4015 | * netif_receive_skb() is the main receive data processing function. | |
4016 | * It always succeeds. The buffer may be dropped during processing | |
4017 | * for congestion control or by the protocol layers. | |
4018 | * | |
4019 | * This function may only be called from softirq context and interrupts | |
4020 | * should be enabled. | |
4021 | * | |
4022 | * Return values (usually ignored): | |
4023 | * NET_RX_SUCCESS: no congestion | |
4024 | * NET_RX_DROP: packet was dropped | |
4025 | */ | |
04eb4489 | 4026 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa BH |
4027 | { |
4028 | trace_netif_receive_skb_entry(skb); | |
4029 | ||
4030 | return netif_receive_skb_internal(skb); | |
4031 | } | |
04eb4489 | 4032 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 4033 | |
88751275 ED |
4034 | /* Network device is going away, flush any packets still pending |
4035 | * Called with irqs disabled. | |
4036 | */ | |
152102c7 | 4037 | static void flush_backlog(void *arg) |
6e583ce5 | 4038 | { |
152102c7 | 4039 | struct net_device *dev = arg; |
903ceff7 | 4040 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
6e583ce5 SH |
4041 | struct sk_buff *skb, *tmp; |
4042 | ||
e36fa2f7 | 4043 | rps_lock(sd); |
6e7676c1 | 4044 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 4045 | if (skb->dev == dev) { |
e36fa2f7 | 4046 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 4047 | kfree_skb(skb); |
76cc8b13 | 4048 | input_queue_head_incr(sd); |
6e583ce5 | 4049 | } |
6e7676c1 | 4050 | } |
e36fa2f7 | 4051 | rps_unlock(sd); |
6e7676c1 CG |
4052 | |
4053 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
4054 | if (skb->dev == dev) { | |
4055 | __skb_unlink(skb, &sd->process_queue); | |
4056 | kfree_skb(skb); | |
76cc8b13 | 4057 | input_queue_head_incr(sd); |
6e7676c1 CG |
4058 | } |
4059 | } | |
6e583ce5 SH |
4060 | } |
4061 | ||
d565b0a1 HX |
4062 | static int napi_gro_complete(struct sk_buff *skb) |
4063 | { | |
22061d80 | 4064 | struct packet_offload *ptype; |
d565b0a1 | 4065 | __be16 type = skb->protocol; |
22061d80 | 4066 | struct list_head *head = &offload_base; |
d565b0a1 HX |
4067 | int err = -ENOENT; |
4068 | ||
c3c7c254 ED |
4069 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
4070 | ||
fc59f9a3 HX |
4071 | if (NAPI_GRO_CB(skb)->count == 1) { |
4072 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 4073 | goto out; |
fc59f9a3 | 4074 | } |
d565b0a1 HX |
4075 | |
4076 | rcu_read_lock(); | |
4077 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4078 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
4079 | continue; |
4080 | ||
299603e8 | 4081 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
4082 | break; |
4083 | } | |
4084 | rcu_read_unlock(); | |
4085 | ||
4086 | if (err) { | |
4087 | WARN_ON(&ptype->list == head); | |
4088 | kfree_skb(skb); | |
4089 | return NET_RX_SUCCESS; | |
4090 | } | |
4091 | ||
4092 | out: | |
ae78dbfa | 4093 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
4094 | } |
4095 | ||
2e71a6f8 ED |
4096 | /* napi->gro_list contains packets ordered by age. |
4097 | * youngest packets at the head of it. | |
4098 | * Complete skbs in reverse order to reduce latencies. | |
4099 | */ | |
4100 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 4101 | { |
2e71a6f8 | 4102 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 4103 | |
2e71a6f8 ED |
4104 | /* scan list and build reverse chain */ |
4105 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
4106 | skb->prev = prev; | |
4107 | prev = skb; | |
4108 | } | |
4109 | ||
4110 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 4111 | skb->next = NULL; |
2e71a6f8 ED |
4112 | |
4113 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
4114 | return; | |
4115 | ||
4116 | prev = skb->prev; | |
d565b0a1 | 4117 | napi_gro_complete(skb); |
2e71a6f8 | 4118 | napi->gro_count--; |
d565b0a1 HX |
4119 | } |
4120 | ||
4121 | napi->gro_list = NULL; | |
4122 | } | |
86cac58b | 4123 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 4124 | |
89c5fa33 ED |
4125 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
4126 | { | |
4127 | struct sk_buff *p; | |
4128 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 4129 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
4130 | |
4131 | for (p = napi->gro_list; p; p = p->next) { | |
4132 | unsigned long diffs; | |
4133 | ||
0b4cec8c TH |
4134 | NAPI_GRO_CB(p)->flush = 0; |
4135 | ||
4136 | if (hash != skb_get_hash_raw(p)) { | |
4137 | NAPI_GRO_CB(p)->same_flow = 0; | |
4138 | continue; | |
4139 | } | |
4140 | ||
89c5fa33 ED |
4141 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
4142 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
4143 | if (maclen == ETH_HLEN) | |
4144 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 4145 | skb_mac_header(skb)); |
89c5fa33 ED |
4146 | else if (!diffs) |
4147 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 4148 | skb_mac_header(skb), |
89c5fa33 ED |
4149 | maclen); |
4150 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
4151 | } |
4152 | } | |
4153 | ||
299603e8 JC |
4154 | static void skb_gro_reset_offset(struct sk_buff *skb) |
4155 | { | |
4156 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4157 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
4158 | ||
4159 | NAPI_GRO_CB(skb)->data_offset = 0; | |
4160 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
4161 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
4162 | ||
4163 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
4164 | pinfo->nr_frags && | |
4165 | !PageHighMem(skb_frag_page(frag0))) { | |
4166 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
4167 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
89c5fa33 ED |
4168 | } |
4169 | } | |
4170 | ||
a50e233c ED |
4171 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
4172 | { | |
4173 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4174 | ||
4175 | BUG_ON(skb->end - skb->tail < grow); | |
4176 | ||
4177 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
4178 | ||
4179 | skb->data_len -= grow; | |
4180 | skb->tail += grow; | |
4181 | ||
4182 | pinfo->frags[0].page_offset += grow; | |
4183 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
4184 | ||
4185 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
4186 | skb_frag_unref(skb, 0); | |
4187 | memmove(pinfo->frags, pinfo->frags + 1, | |
4188 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
4189 | } | |
4190 | } | |
4191 | ||
bb728820 | 4192 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
4193 | { |
4194 | struct sk_buff **pp = NULL; | |
22061d80 | 4195 | struct packet_offload *ptype; |
d565b0a1 | 4196 | __be16 type = skb->protocol; |
22061d80 | 4197 | struct list_head *head = &offload_base; |
0da2afd5 | 4198 | int same_flow; |
5b252f0c | 4199 | enum gro_result ret; |
a50e233c | 4200 | int grow; |
d565b0a1 | 4201 | |
9c62a68d | 4202 | if (!(skb->dev->features & NETIF_F_GRO)) |
d565b0a1 HX |
4203 | goto normal; |
4204 | ||
5a212329 | 4205 | if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad) |
f17f5c91 HX |
4206 | goto normal; |
4207 | ||
89c5fa33 ED |
4208 | gro_list_prepare(napi, skb); |
4209 | ||
d565b0a1 HX |
4210 | rcu_read_lock(); |
4211 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4212 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4213 | continue; |
4214 | ||
86911732 | 4215 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4216 | skb_reset_mac_len(skb); |
d565b0a1 HX |
4217 | NAPI_GRO_CB(skb)->same_flow = 0; |
4218 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 4219 | NAPI_GRO_CB(skb)->free = 0; |
b582ef09 | 4220 | NAPI_GRO_CB(skb)->udp_mark = 0; |
15e2396d | 4221 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 4222 | |
662880f4 TH |
4223 | /* Setup for GRO checksum validation */ |
4224 | switch (skb->ip_summed) { | |
4225 | case CHECKSUM_COMPLETE: | |
4226 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4227 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4228 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4229 | break; | |
4230 | case CHECKSUM_UNNECESSARY: | |
4231 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4232 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4233 | break; | |
4234 | default: | |
4235 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4236 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4237 | } | |
d565b0a1 | 4238 | |
f191a1d1 | 4239 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4240 | break; |
4241 | } | |
4242 | rcu_read_unlock(); | |
4243 | ||
4244 | if (&ptype->list == head) | |
4245 | goto normal; | |
4246 | ||
0da2afd5 | 4247 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4248 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4249 | |
d565b0a1 HX |
4250 | if (pp) { |
4251 | struct sk_buff *nskb = *pp; | |
4252 | ||
4253 | *pp = nskb->next; | |
4254 | nskb->next = NULL; | |
4255 | napi_gro_complete(nskb); | |
4ae5544f | 4256 | napi->gro_count--; |
d565b0a1 HX |
4257 | } |
4258 | ||
0da2afd5 | 4259 | if (same_flow) |
d565b0a1 HX |
4260 | goto ok; |
4261 | ||
600adc18 | 4262 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4263 | goto normal; |
d565b0a1 | 4264 | |
600adc18 ED |
4265 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4266 | struct sk_buff *nskb = napi->gro_list; | |
4267 | ||
4268 | /* locate the end of the list to select the 'oldest' flow */ | |
4269 | while (nskb->next) { | |
4270 | pp = &nskb->next; | |
4271 | nskb = *pp; | |
4272 | } | |
4273 | *pp = NULL; | |
4274 | nskb->next = NULL; | |
4275 | napi_gro_complete(nskb); | |
4276 | } else { | |
4277 | napi->gro_count++; | |
4278 | } | |
d565b0a1 | 4279 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4280 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4281 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4282 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4283 | skb->next = napi->gro_list; |
4284 | napi->gro_list = skb; | |
5d0d9be8 | 4285 | ret = GRO_HELD; |
d565b0a1 | 4286 | |
ad0f9904 | 4287 | pull: |
a50e233c ED |
4288 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4289 | if (grow > 0) | |
4290 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4291 | ok: |
5d0d9be8 | 4292 | return ret; |
d565b0a1 HX |
4293 | |
4294 | normal: | |
ad0f9904 HX |
4295 | ret = GRO_NORMAL; |
4296 | goto pull; | |
5d38a079 | 4297 | } |
96e93eab | 4298 | |
bf5a755f JC |
4299 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4300 | { | |
4301 | struct list_head *offload_head = &offload_base; | |
4302 | struct packet_offload *ptype; | |
4303 | ||
4304 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4305 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4306 | continue; | |
4307 | return ptype; | |
4308 | } | |
4309 | return NULL; | |
4310 | } | |
e27a2f83 | 4311 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4312 | |
4313 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4314 | { | |
4315 | struct list_head *offload_head = &offload_base; | |
4316 | struct packet_offload *ptype; | |
4317 | ||
4318 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4319 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4320 | continue; | |
4321 | return ptype; | |
4322 | } | |
4323 | return NULL; | |
4324 | } | |
e27a2f83 | 4325 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4326 | |
bb728820 | 4327 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4328 | { |
5d0d9be8 HX |
4329 | switch (ret) { |
4330 | case GRO_NORMAL: | |
ae78dbfa | 4331 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4332 | ret = GRO_DROP; |
4333 | break; | |
5d38a079 | 4334 | |
5d0d9be8 | 4335 | case GRO_DROP: |
5d38a079 HX |
4336 | kfree_skb(skb); |
4337 | break; | |
5b252f0c | 4338 | |
daa86548 | 4339 | case GRO_MERGED_FREE: |
d7e8883c ED |
4340 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
4341 | kmem_cache_free(skbuff_head_cache, skb); | |
4342 | else | |
4343 | __kfree_skb(skb); | |
daa86548 ED |
4344 | break; |
4345 | ||
5b252f0c BH |
4346 | case GRO_HELD: |
4347 | case GRO_MERGED: | |
4348 | break; | |
5d38a079 HX |
4349 | } |
4350 | ||
c7c4b3b6 | 4351 | return ret; |
5d0d9be8 | 4352 | } |
5d0d9be8 | 4353 | |
c7c4b3b6 | 4354 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4355 | { |
ae78dbfa | 4356 | trace_napi_gro_receive_entry(skb); |
86911732 | 4357 | |
a50e233c ED |
4358 | skb_gro_reset_offset(skb); |
4359 | ||
89c5fa33 | 4360 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4361 | } |
4362 | EXPORT_SYMBOL(napi_gro_receive); | |
4363 | ||
d0c2b0d2 | 4364 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4365 | { |
93a35f59 ED |
4366 | if (unlikely(skb->pfmemalloc)) { |
4367 | consume_skb(skb); | |
4368 | return; | |
4369 | } | |
96e93eab | 4370 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4371 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4372 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4373 | skb->vlan_tci = 0; |
66c46d74 | 4374 | skb->dev = napi->dev; |
6d152e23 | 4375 | skb->skb_iif = 0; |
c3caf119 JC |
4376 | skb->encapsulation = 0; |
4377 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4378 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
96e93eab HX |
4379 | |
4380 | napi->skb = skb; | |
4381 | } | |
96e93eab | 4382 | |
76620aaf | 4383 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4384 | { |
5d38a079 | 4385 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4386 | |
4387 | if (!skb) { | |
fd11a83d | 4388 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
84b9cd63 | 4389 | napi->skb = skb; |
80595d59 | 4390 | } |
96e93eab HX |
4391 | return skb; |
4392 | } | |
76620aaf | 4393 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4394 | |
a50e233c ED |
4395 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4396 | struct sk_buff *skb, | |
4397 | gro_result_t ret) | |
96e93eab | 4398 | { |
5d0d9be8 HX |
4399 | switch (ret) { |
4400 | case GRO_NORMAL: | |
a50e233c ED |
4401 | case GRO_HELD: |
4402 | __skb_push(skb, ETH_HLEN); | |
4403 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4404 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4405 | ret = GRO_DROP; |
86911732 | 4406 | break; |
5d38a079 | 4407 | |
5d0d9be8 | 4408 | case GRO_DROP: |
5d0d9be8 HX |
4409 | case GRO_MERGED_FREE: |
4410 | napi_reuse_skb(napi, skb); | |
4411 | break; | |
5b252f0c BH |
4412 | |
4413 | case GRO_MERGED: | |
4414 | break; | |
5d0d9be8 | 4415 | } |
5d38a079 | 4416 | |
c7c4b3b6 | 4417 | return ret; |
5d38a079 | 4418 | } |
5d0d9be8 | 4419 | |
a50e233c ED |
4420 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
4421 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
4422 | * We copy ethernet header into skb->data to have a common layout. | |
4423 | */ | |
4adb9c4a | 4424 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4425 | { |
4426 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
4427 | const struct ethhdr *eth; |
4428 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
4429 | |
4430 | napi->skb = NULL; | |
4431 | ||
a50e233c ED |
4432 | skb_reset_mac_header(skb); |
4433 | skb_gro_reset_offset(skb); | |
4434 | ||
4435 | eth = skb_gro_header_fast(skb, 0); | |
4436 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
4437 | eth = skb_gro_header_slow(skb, hlen, 0); | |
4438 | if (unlikely(!eth)) { | |
4439 | napi_reuse_skb(napi, skb); | |
4440 | return NULL; | |
4441 | } | |
4442 | } else { | |
4443 | gro_pull_from_frag0(skb, hlen); | |
4444 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
4445 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 4446 | } |
a50e233c ED |
4447 | __skb_pull(skb, hlen); |
4448 | ||
4449 | /* | |
4450 | * This works because the only protocols we care about don't require | |
4451 | * special handling. | |
4452 | * We'll fix it up properly in napi_frags_finish() | |
4453 | */ | |
4454 | skb->protocol = eth->h_proto; | |
76620aaf | 4455 | |
76620aaf HX |
4456 | return skb; |
4457 | } | |
76620aaf | 4458 | |
c7c4b3b6 | 4459 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4460 | { |
76620aaf | 4461 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4462 | |
4463 | if (!skb) | |
c7c4b3b6 | 4464 | return GRO_DROP; |
5d0d9be8 | 4465 | |
ae78dbfa BH |
4466 | trace_napi_gro_frags_entry(skb); |
4467 | ||
89c5fa33 | 4468 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4469 | } |
5d38a079 HX |
4470 | EXPORT_SYMBOL(napi_gro_frags); |
4471 | ||
573e8fca TH |
4472 | /* Compute the checksum from gro_offset and return the folded value |
4473 | * after adding in any pseudo checksum. | |
4474 | */ | |
4475 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
4476 | { | |
4477 | __wsum wsum; | |
4478 | __sum16 sum; | |
4479 | ||
4480 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
4481 | ||
4482 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
4483 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
4484 | if (likely(!sum)) { | |
4485 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
4486 | !skb->csum_complete_sw) | |
4487 | netdev_rx_csum_fault(skb->dev); | |
4488 | } | |
4489 | ||
4490 | NAPI_GRO_CB(skb)->csum = wsum; | |
4491 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4492 | ||
4493 | return sum; | |
4494 | } | |
4495 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
4496 | ||
e326bed2 | 4497 | /* |
855abcf0 | 4498 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
4499 | * Note: called with local irq disabled, but exits with local irq enabled. |
4500 | */ | |
4501 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4502 | { | |
4503 | #ifdef CONFIG_RPS | |
4504 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4505 | ||
4506 | if (remsd) { | |
4507 | sd->rps_ipi_list = NULL; | |
4508 | ||
4509 | local_irq_enable(); | |
4510 | ||
4511 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
4512 | while (remsd) { | |
4513 | struct softnet_data *next = remsd->rps_ipi_next; | |
4514 | ||
4515 | if (cpu_online(remsd->cpu)) | |
c46fff2a | 4516 | smp_call_function_single_async(remsd->cpu, |
fce8ad15 | 4517 | &remsd->csd); |
e326bed2 ED |
4518 | remsd = next; |
4519 | } | |
4520 | } else | |
4521 | #endif | |
4522 | local_irq_enable(); | |
4523 | } | |
4524 | ||
d75b1ade ED |
4525 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
4526 | { | |
4527 | #ifdef CONFIG_RPS | |
4528 | return sd->rps_ipi_list != NULL; | |
4529 | #else | |
4530 | return false; | |
4531 | #endif | |
4532 | } | |
4533 | ||
bea3348e | 4534 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
4535 | { |
4536 | int work = 0; | |
eecfd7c4 | 4537 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 4538 | |
e326bed2 ED |
4539 | /* Check if we have pending ipi, its better to send them now, |
4540 | * not waiting net_rx_action() end. | |
4541 | */ | |
d75b1ade | 4542 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
4543 | local_irq_disable(); |
4544 | net_rps_action_and_irq_enable(sd); | |
4545 | } | |
d75b1ade | 4546 | |
bea3348e | 4547 | napi->weight = weight_p; |
6e7676c1 | 4548 | local_irq_disable(); |
11ef7a89 | 4549 | while (1) { |
1da177e4 | 4550 | struct sk_buff *skb; |
6e7676c1 CG |
4551 | |
4552 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 4553 | rcu_read_lock(); |
6e7676c1 CG |
4554 | local_irq_enable(); |
4555 | __netif_receive_skb(skb); | |
2c17d27c | 4556 | rcu_read_unlock(); |
6e7676c1 | 4557 | local_irq_disable(); |
76cc8b13 TH |
4558 | input_queue_head_incr(sd); |
4559 | if (++work >= quota) { | |
4560 | local_irq_enable(); | |
4561 | return work; | |
4562 | } | |
6e7676c1 | 4563 | } |
1da177e4 | 4564 | |
e36fa2f7 | 4565 | rps_lock(sd); |
11ef7a89 | 4566 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
4567 | /* |
4568 | * Inline a custom version of __napi_complete(). | |
4569 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
4570 | * and NAPI_STATE_SCHED is the only possible flag set |
4571 | * on backlog. | |
4572 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
4573 | * and we dont need an smp_mb() memory barrier. |
4574 | */ | |
eecfd7c4 | 4575 | napi->state = 0; |
11ef7a89 | 4576 | rps_unlock(sd); |
eecfd7c4 | 4577 | |
11ef7a89 | 4578 | break; |
bea3348e | 4579 | } |
11ef7a89 TH |
4580 | |
4581 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
4582 | &sd->process_queue); | |
e36fa2f7 | 4583 | rps_unlock(sd); |
6e7676c1 CG |
4584 | } |
4585 | local_irq_enable(); | |
1da177e4 | 4586 | |
bea3348e SH |
4587 | return work; |
4588 | } | |
1da177e4 | 4589 | |
bea3348e SH |
4590 | /** |
4591 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4592 | * @n: entry to schedule |
bea3348e | 4593 | * |
bc9ad166 ED |
4594 | * The entry's receive function will be scheduled to run. |
4595 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 4596 | */ |
b5606c2d | 4597 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4598 | { |
4599 | unsigned long flags; | |
1da177e4 | 4600 | |
bea3348e | 4601 | local_irq_save(flags); |
903ceff7 | 4602 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 4603 | local_irq_restore(flags); |
1da177e4 | 4604 | } |
bea3348e SH |
4605 | EXPORT_SYMBOL(__napi_schedule); |
4606 | ||
bc9ad166 ED |
4607 | /** |
4608 | * __napi_schedule_irqoff - schedule for receive | |
4609 | * @n: entry to schedule | |
4610 | * | |
4611 | * Variant of __napi_schedule() assuming hard irqs are masked | |
4612 | */ | |
4613 | void __napi_schedule_irqoff(struct napi_struct *n) | |
4614 | { | |
4615 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
4616 | } | |
4617 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
4618 | ||
d565b0a1 HX |
4619 | void __napi_complete(struct napi_struct *n) |
4620 | { | |
4621 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
d565b0a1 | 4622 | |
d75b1ade | 4623 | list_del_init(&n->poll_list); |
4e857c58 | 4624 | smp_mb__before_atomic(); |
d565b0a1 HX |
4625 | clear_bit(NAPI_STATE_SCHED, &n->state); |
4626 | } | |
4627 | EXPORT_SYMBOL(__napi_complete); | |
4628 | ||
3b47d303 | 4629 | void napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 HX |
4630 | { |
4631 | unsigned long flags; | |
4632 | ||
4633 | /* | |
4634 | * don't let napi dequeue from the cpu poll list | |
4635 | * just in case its running on a different cpu | |
4636 | */ | |
4637 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4638 | return; | |
4639 | ||
3b47d303 ED |
4640 | if (n->gro_list) { |
4641 | unsigned long timeout = 0; | |
d75b1ade | 4642 | |
3b47d303 ED |
4643 | if (work_done) |
4644 | timeout = n->dev->gro_flush_timeout; | |
4645 | ||
4646 | if (timeout) | |
4647 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
4648 | HRTIMER_MODE_REL_PINNED); | |
4649 | else | |
4650 | napi_gro_flush(n, false); | |
4651 | } | |
d75b1ade ED |
4652 | if (likely(list_empty(&n->poll_list))) { |
4653 | WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state)); | |
4654 | } else { | |
4655 | /* If n->poll_list is not empty, we need to mask irqs */ | |
4656 | local_irq_save(flags); | |
4657 | __napi_complete(n); | |
4658 | local_irq_restore(flags); | |
4659 | } | |
d565b0a1 | 4660 | } |
3b47d303 | 4661 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 4662 | |
af12fa6e ET |
4663 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
4664 | struct napi_struct *napi_by_id(unsigned int napi_id) | |
4665 | { | |
4666 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
4667 | struct napi_struct *napi; | |
4668 | ||
4669 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
4670 | if (napi->napi_id == napi_id) | |
4671 | return napi; | |
4672 | ||
4673 | return NULL; | |
4674 | } | |
4675 | EXPORT_SYMBOL_GPL(napi_by_id); | |
4676 | ||
4677 | void napi_hash_add(struct napi_struct *napi) | |
4678 | { | |
4679 | if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) { | |
4680 | ||
4681 | spin_lock(&napi_hash_lock); | |
4682 | ||
4683 | /* 0 is not a valid id, we also skip an id that is taken | |
4684 | * we expect both events to be extremely rare | |
4685 | */ | |
4686 | napi->napi_id = 0; | |
4687 | while (!napi->napi_id) { | |
4688 | napi->napi_id = ++napi_gen_id; | |
4689 | if (napi_by_id(napi->napi_id)) | |
4690 | napi->napi_id = 0; | |
4691 | } | |
4692 | ||
4693 | hlist_add_head_rcu(&napi->napi_hash_node, | |
4694 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
4695 | ||
4696 | spin_unlock(&napi_hash_lock); | |
4697 | } | |
4698 | } | |
4699 | EXPORT_SYMBOL_GPL(napi_hash_add); | |
4700 | ||
4701 | /* Warning : caller is responsible to make sure rcu grace period | |
4702 | * is respected before freeing memory containing @napi | |
4703 | */ | |
4704 | void napi_hash_del(struct napi_struct *napi) | |
4705 | { | |
4706 | spin_lock(&napi_hash_lock); | |
4707 | ||
4708 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) | |
4709 | hlist_del_rcu(&napi->napi_hash_node); | |
4710 | ||
4711 | spin_unlock(&napi_hash_lock); | |
4712 | } | |
4713 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
4714 | ||
3b47d303 ED |
4715 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
4716 | { | |
4717 | struct napi_struct *napi; | |
4718 | ||
4719 | napi = container_of(timer, struct napi_struct, timer); | |
4720 | if (napi->gro_list) | |
4721 | napi_schedule(napi); | |
4722 | ||
4723 | return HRTIMER_NORESTART; | |
4724 | } | |
4725 | ||
d565b0a1 HX |
4726 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
4727 | int (*poll)(struct napi_struct *, int), int weight) | |
4728 | { | |
4729 | INIT_LIST_HEAD(&napi->poll_list); | |
3b47d303 ED |
4730 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
4731 | napi->timer.function = napi_watchdog; | |
4ae5544f | 4732 | napi->gro_count = 0; |
d565b0a1 | 4733 | napi->gro_list = NULL; |
5d38a079 | 4734 | napi->skb = NULL; |
d565b0a1 | 4735 | napi->poll = poll; |
82dc3c63 ED |
4736 | if (weight > NAPI_POLL_WEIGHT) |
4737 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
4738 | weight, dev->name); | |
d565b0a1 HX |
4739 | napi->weight = weight; |
4740 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 4741 | napi->dev = dev; |
5d38a079 | 4742 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
4743 | spin_lock_init(&napi->poll_lock); |
4744 | napi->poll_owner = -1; | |
4745 | #endif | |
4746 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
4747 | } | |
4748 | EXPORT_SYMBOL(netif_napi_add); | |
4749 | ||
3b47d303 ED |
4750 | void napi_disable(struct napi_struct *n) |
4751 | { | |
4752 | might_sleep(); | |
4753 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
4754 | ||
4755 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
4756 | msleep(1); | |
2d8bff12 NH |
4757 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
4758 | msleep(1); | |
3b47d303 ED |
4759 | |
4760 | hrtimer_cancel(&n->timer); | |
4761 | ||
4762 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
4763 | } | |
4764 | EXPORT_SYMBOL(napi_disable); | |
4765 | ||
d565b0a1 HX |
4766 | void netif_napi_del(struct napi_struct *napi) |
4767 | { | |
d7b06636 | 4768 | list_del_init(&napi->dev_list); |
76620aaf | 4769 | napi_free_frags(napi); |
d565b0a1 | 4770 | |
289dccbe | 4771 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 4772 | napi->gro_list = NULL; |
4ae5544f | 4773 | napi->gro_count = 0; |
d565b0a1 HX |
4774 | } |
4775 | EXPORT_SYMBOL(netif_napi_del); | |
4776 | ||
726ce70e HX |
4777 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
4778 | { | |
4779 | void *have; | |
4780 | int work, weight; | |
4781 | ||
4782 | list_del_init(&n->poll_list); | |
4783 | ||
4784 | have = netpoll_poll_lock(n); | |
4785 | ||
4786 | weight = n->weight; | |
4787 | ||
4788 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
4789 | * with netpoll's poll_napi(). Only the entity which | |
4790 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
4791 | * actually make the ->poll() call. Therefore we avoid | |
4792 | * accidentally calling ->poll() when NAPI is not scheduled. | |
4793 | */ | |
4794 | work = 0; | |
4795 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
4796 | work = n->poll(n, weight); | |
4797 | trace_napi_poll(n); | |
4798 | } | |
4799 | ||
4800 | WARN_ON_ONCE(work > weight); | |
4801 | ||
4802 | if (likely(work < weight)) | |
4803 | goto out_unlock; | |
4804 | ||
4805 | /* Drivers must not modify the NAPI state if they | |
4806 | * consume the entire weight. In such cases this code | |
4807 | * still "owns" the NAPI instance and therefore can | |
4808 | * move the instance around on the list at-will. | |
4809 | */ | |
4810 | if (unlikely(napi_disable_pending(n))) { | |
4811 | napi_complete(n); | |
4812 | goto out_unlock; | |
4813 | } | |
4814 | ||
4815 | if (n->gro_list) { | |
4816 | /* flush too old packets | |
4817 | * If HZ < 1000, flush all packets. | |
4818 | */ | |
4819 | napi_gro_flush(n, HZ >= 1000); | |
4820 | } | |
4821 | ||
001ce546 HX |
4822 | /* Some drivers may have called napi_schedule |
4823 | * prior to exhausting their budget. | |
4824 | */ | |
4825 | if (unlikely(!list_empty(&n->poll_list))) { | |
4826 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
4827 | n->dev ? n->dev->name : "backlog"); | |
4828 | goto out_unlock; | |
4829 | } | |
4830 | ||
726ce70e HX |
4831 | list_add_tail(&n->poll_list, repoll); |
4832 | ||
4833 | out_unlock: | |
4834 | netpoll_poll_unlock(have); | |
4835 | ||
4836 | return work; | |
4837 | } | |
4838 | ||
1da177e4 LT |
4839 | static void net_rx_action(struct softirq_action *h) |
4840 | { | |
903ceff7 | 4841 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
24f8b238 | 4842 | unsigned long time_limit = jiffies + 2; |
51b0bded | 4843 | int budget = netdev_budget; |
d75b1ade ED |
4844 | LIST_HEAD(list); |
4845 | LIST_HEAD(repoll); | |
53fb95d3 | 4846 | |
1da177e4 | 4847 | local_irq_disable(); |
d75b1ade ED |
4848 | list_splice_init(&sd->poll_list, &list); |
4849 | local_irq_enable(); | |
1da177e4 | 4850 | |
ceb8d5bf | 4851 | for (;;) { |
bea3348e | 4852 | struct napi_struct *n; |
1da177e4 | 4853 | |
ceb8d5bf HX |
4854 | if (list_empty(&list)) { |
4855 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
4856 | return; | |
4857 | break; | |
4858 | } | |
4859 | ||
6bd373eb HX |
4860 | n = list_first_entry(&list, struct napi_struct, poll_list); |
4861 | budget -= napi_poll(n, &repoll); | |
4862 | ||
d75b1ade | 4863 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
4864 | * Allow this to run for 2 jiffies since which will allow |
4865 | * an average latency of 1.5/HZ. | |
bea3348e | 4866 | */ |
ceb8d5bf HX |
4867 | if (unlikely(budget <= 0 || |
4868 | time_after_eq(jiffies, time_limit))) { | |
4869 | sd->time_squeeze++; | |
4870 | break; | |
4871 | } | |
1da177e4 | 4872 | } |
d75b1ade | 4873 | |
d75b1ade ED |
4874 | local_irq_disable(); |
4875 | ||
4876 | list_splice_tail_init(&sd->poll_list, &list); | |
4877 | list_splice_tail(&repoll, &list); | |
4878 | list_splice(&list, &sd->poll_list); | |
4879 | if (!list_empty(&sd->poll_list)) | |
4880 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4881 | ||
e326bed2 | 4882 | net_rps_action_and_irq_enable(sd); |
1da177e4 LT |
4883 | } |
4884 | ||
aa9d8560 | 4885 | struct netdev_adjacent { |
9ff162a8 | 4886 | struct net_device *dev; |
5d261913 VF |
4887 | |
4888 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 4889 | bool master; |
5d261913 | 4890 | |
5d261913 VF |
4891 | /* counter for the number of times this device was added to us */ |
4892 | u16 ref_nr; | |
4893 | ||
402dae96 VF |
4894 | /* private field for the users */ |
4895 | void *private; | |
4896 | ||
9ff162a8 JP |
4897 | struct list_head list; |
4898 | struct rcu_head rcu; | |
9ff162a8 JP |
4899 | }; |
4900 | ||
6ea29da1 | 4901 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 4902 | struct list_head *adj_list) |
9ff162a8 | 4903 | { |
5d261913 | 4904 | struct netdev_adjacent *adj; |
5d261913 | 4905 | |
2f268f12 | 4906 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
4907 | if (adj->dev == adj_dev) |
4908 | return adj; | |
9ff162a8 JP |
4909 | } |
4910 | return NULL; | |
4911 | } | |
4912 | ||
4913 | /** | |
4914 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
4915 | * @dev: device | |
4916 | * @upper_dev: upper device to check | |
4917 | * | |
4918 | * Find out if a device is linked to specified upper device and return true | |
4919 | * in case it is. Note that this checks only immediate upper device, | |
4920 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
4921 | */ | |
4922 | bool netdev_has_upper_dev(struct net_device *dev, | |
4923 | struct net_device *upper_dev) | |
4924 | { | |
4925 | ASSERT_RTNL(); | |
4926 | ||
6ea29da1 | 4927 | return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper); |
9ff162a8 JP |
4928 | } |
4929 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
4930 | ||
4931 | /** | |
4932 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
4933 | * @dev: device | |
4934 | * | |
4935 | * Find out if a device is linked to an upper device and return true in case | |
4936 | * it is. The caller must hold the RTNL lock. | |
4937 | */ | |
1d143d9f | 4938 | static bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
4939 | { |
4940 | ASSERT_RTNL(); | |
4941 | ||
2f268f12 | 4942 | return !list_empty(&dev->all_adj_list.upper); |
9ff162a8 | 4943 | } |
9ff162a8 JP |
4944 | |
4945 | /** | |
4946 | * netdev_master_upper_dev_get - Get master upper device | |
4947 | * @dev: device | |
4948 | * | |
4949 | * Find a master upper device and return pointer to it or NULL in case | |
4950 | * it's not there. The caller must hold the RTNL lock. | |
4951 | */ | |
4952 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
4953 | { | |
aa9d8560 | 4954 | struct netdev_adjacent *upper; |
9ff162a8 JP |
4955 | |
4956 | ASSERT_RTNL(); | |
4957 | ||
2f268f12 | 4958 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
4959 | return NULL; |
4960 | ||
2f268f12 | 4961 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 4962 | struct netdev_adjacent, list); |
9ff162a8 JP |
4963 | if (likely(upper->master)) |
4964 | return upper->dev; | |
4965 | return NULL; | |
4966 | } | |
4967 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
4968 | ||
b6ccba4c VF |
4969 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
4970 | { | |
4971 | struct netdev_adjacent *adj; | |
4972 | ||
4973 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
4974 | ||
4975 | return adj->private; | |
4976 | } | |
4977 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
4978 | ||
44a40855 VY |
4979 | /** |
4980 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
4981 | * @dev: device | |
4982 | * @iter: list_head ** of the current position | |
4983 | * | |
4984 | * Gets the next device from the dev's upper list, starting from iter | |
4985 | * position. The caller must hold RCU read lock. | |
4986 | */ | |
4987 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
4988 | struct list_head **iter) | |
4989 | { | |
4990 | struct netdev_adjacent *upper; | |
4991 | ||
4992 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
4993 | ||
4994 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4995 | ||
4996 | if (&upper->list == &dev->adj_list.upper) | |
4997 | return NULL; | |
4998 | ||
4999 | *iter = &upper->list; | |
5000 | ||
5001 | return upper->dev; | |
5002 | } | |
5003 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
5004 | ||
31088a11 VF |
5005 | /** |
5006 | * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list | |
48311f46 VF |
5007 | * @dev: device |
5008 | * @iter: list_head ** of the current position | |
5009 | * | |
5010 | * Gets the next device from the dev's upper list, starting from iter | |
5011 | * position. The caller must hold RCU read lock. | |
5012 | */ | |
2f268f12 VF |
5013 | struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, |
5014 | struct list_head **iter) | |
48311f46 VF |
5015 | { |
5016 | struct netdev_adjacent *upper; | |
5017 | ||
85328240 | 5018 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); |
48311f46 VF |
5019 | |
5020 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5021 | ||
2f268f12 | 5022 | if (&upper->list == &dev->all_adj_list.upper) |
48311f46 VF |
5023 | return NULL; |
5024 | ||
5025 | *iter = &upper->list; | |
5026 | ||
5027 | return upper->dev; | |
5028 | } | |
2f268f12 | 5029 | EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); |
48311f46 | 5030 | |
31088a11 VF |
5031 | /** |
5032 | * netdev_lower_get_next_private - Get the next ->private from the | |
5033 | * lower neighbour list | |
5034 | * @dev: device | |
5035 | * @iter: list_head ** of the current position | |
5036 | * | |
5037 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5038 | * list, starting from iter position. The caller must hold either hold the | |
5039 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 5040 | * list will remain unchanged. |
31088a11 VF |
5041 | */ |
5042 | void *netdev_lower_get_next_private(struct net_device *dev, | |
5043 | struct list_head **iter) | |
5044 | { | |
5045 | struct netdev_adjacent *lower; | |
5046 | ||
5047 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
5048 | ||
5049 | if (&lower->list == &dev->adj_list.lower) | |
5050 | return NULL; | |
5051 | ||
6859e7df | 5052 | *iter = lower->list.next; |
31088a11 VF |
5053 | |
5054 | return lower->private; | |
5055 | } | |
5056 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
5057 | ||
5058 | /** | |
5059 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
5060 | * lower neighbour list, RCU | |
5061 | * variant | |
5062 | * @dev: device | |
5063 | * @iter: list_head ** of the current position | |
5064 | * | |
5065 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5066 | * list, starting from iter position. The caller must hold RCU read lock. | |
5067 | */ | |
5068 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
5069 | struct list_head **iter) | |
5070 | { | |
5071 | struct netdev_adjacent *lower; | |
5072 | ||
5073 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5074 | ||
5075 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5076 | ||
5077 | if (&lower->list == &dev->adj_list.lower) | |
5078 | return NULL; | |
5079 | ||
6859e7df | 5080 | *iter = &lower->list; |
31088a11 VF |
5081 | |
5082 | return lower->private; | |
5083 | } | |
5084 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
5085 | ||
4085ebe8 VY |
5086 | /** |
5087 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
5088 | * list | |
5089 | * @dev: device | |
5090 | * @iter: list_head ** of the current position | |
5091 | * | |
5092 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
5093 | * list, starting from iter position. The caller must hold RTNL lock or | |
5094 | * its own locking that guarantees that the neighbour lower | |
b469139e | 5095 | * list will remain unchanged. |
4085ebe8 VY |
5096 | */ |
5097 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
5098 | { | |
5099 | struct netdev_adjacent *lower; | |
5100 | ||
5101 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
5102 | ||
5103 | if (&lower->list == &dev->adj_list.lower) | |
5104 | return NULL; | |
5105 | ||
5106 | *iter = &lower->list; | |
5107 | ||
5108 | return lower->dev; | |
5109 | } | |
5110 | EXPORT_SYMBOL(netdev_lower_get_next); | |
5111 | ||
e001bfad | 5112 | /** |
5113 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
5114 | * lower neighbour list, RCU | |
5115 | * variant | |
5116 | * @dev: device | |
5117 | * | |
5118 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
5119 | * list. The caller must hold RCU read lock. | |
5120 | */ | |
5121 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
5122 | { | |
5123 | struct netdev_adjacent *lower; | |
5124 | ||
5125 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
5126 | struct netdev_adjacent, list); | |
5127 | if (lower) | |
5128 | return lower->private; | |
5129 | return NULL; | |
5130 | } | |
5131 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
5132 | ||
9ff162a8 JP |
5133 | /** |
5134 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
5135 | * @dev: device | |
5136 | * | |
5137 | * Find a master upper device and return pointer to it or NULL in case | |
5138 | * it's not there. The caller must hold the RCU read lock. | |
5139 | */ | |
5140 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
5141 | { | |
aa9d8560 | 5142 | struct netdev_adjacent *upper; |
9ff162a8 | 5143 | |
2f268f12 | 5144 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 5145 | struct netdev_adjacent, list); |
9ff162a8 JP |
5146 | if (upper && likely(upper->master)) |
5147 | return upper->dev; | |
5148 | return NULL; | |
5149 | } | |
5150 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
5151 | ||
0a59f3a9 | 5152 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
5153 | struct net_device *adj_dev, |
5154 | struct list_head *dev_list) | |
5155 | { | |
5156 | char linkname[IFNAMSIZ+7]; | |
5157 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
5158 | "upper_%s" : "lower_%s", adj_dev->name); | |
5159 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
5160 | linkname); | |
5161 | } | |
0a59f3a9 | 5162 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
5163 | char *name, |
5164 | struct list_head *dev_list) | |
5165 | { | |
5166 | char linkname[IFNAMSIZ+7]; | |
5167 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
5168 | "upper_%s" : "lower_%s", name); | |
5169 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
5170 | } | |
5171 | ||
7ce64c79 AF |
5172 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
5173 | struct net_device *adj_dev, | |
5174 | struct list_head *dev_list) | |
5175 | { | |
5176 | return (dev_list == &dev->adj_list.upper || | |
5177 | dev_list == &dev->adj_list.lower) && | |
5178 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
5179 | } | |
3ee32707 | 5180 | |
5d261913 VF |
5181 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
5182 | struct net_device *adj_dev, | |
7863c054 | 5183 | struct list_head *dev_list, |
402dae96 | 5184 | void *private, bool master) |
5d261913 VF |
5185 | { |
5186 | struct netdev_adjacent *adj; | |
842d67a7 | 5187 | int ret; |
5d261913 | 5188 | |
6ea29da1 | 5189 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
5190 | |
5191 | if (adj) { | |
5d261913 VF |
5192 | adj->ref_nr++; |
5193 | return 0; | |
5194 | } | |
5195 | ||
5196 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
5197 | if (!adj) | |
5198 | return -ENOMEM; | |
5199 | ||
5200 | adj->dev = adj_dev; | |
5201 | adj->master = master; | |
5d261913 | 5202 | adj->ref_nr = 1; |
402dae96 | 5203 | adj->private = private; |
5d261913 | 5204 | dev_hold(adj_dev); |
2f268f12 VF |
5205 | |
5206 | pr_debug("dev_hold for %s, because of link added from %s to %s\n", | |
5207 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 | 5208 | |
7ce64c79 | 5209 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 5210 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
5211 | if (ret) |
5212 | goto free_adj; | |
5213 | } | |
5214 | ||
7863c054 | 5215 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
5216 | if (master) { |
5217 | ret = sysfs_create_link(&(dev->dev.kobj), | |
5218 | &(adj_dev->dev.kobj), "master"); | |
5219 | if (ret) | |
5831d66e | 5220 | goto remove_symlinks; |
842d67a7 | 5221 | |
7863c054 | 5222 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 5223 | } else { |
7863c054 | 5224 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 5225 | } |
5d261913 VF |
5226 | |
5227 | return 0; | |
842d67a7 | 5228 | |
5831d66e | 5229 | remove_symlinks: |
7ce64c79 | 5230 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5231 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
5232 | free_adj: |
5233 | kfree(adj); | |
974daef7 | 5234 | dev_put(adj_dev); |
842d67a7 VF |
5235 | |
5236 | return ret; | |
5d261913 VF |
5237 | } |
5238 | ||
1d143d9f | 5239 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
5240 | struct net_device *adj_dev, | |
5241 | struct list_head *dev_list) | |
5d261913 VF |
5242 | { |
5243 | struct netdev_adjacent *adj; | |
5244 | ||
6ea29da1 | 5245 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 5246 | |
2f268f12 VF |
5247 | if (!adj) { |
5248 | pr_err("tried to remove device %s from %s\n", | |
5249 | dev->name, adj_dev->name); | |
5d261913 | 5250 | BUG(); |
2f268f12 | 5251 | } |
5d261913 VF |
5252 | |
5253 | if (adj->ref_nr > 1) { | |
2f268f12 VF |
5254 | pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name, |
5255 | adj->ref_nr-1); | |
5d261913 VF |
5256 | adj->ref_nr--; |
5257 | return; | |
5258 | } | |
5259 | ||
842d67a7 VF |
5260 | if (adj->master) |
5261 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
5262 | ||
7ce64c79 | 5263 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5264 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 5265 | |
5d261913 | 5266 | list_del_rcu(&adj->list); |
2f268f12 VF |
5267 | pr_debug("dev_put for %s, because link removed from %s to %s\n", |
5268 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 VF |
5269 | dev_put(adj_dev); |
5270 | kfree_rcu(adj, rcu); | |
5271 | } | |
5272 | ||
1d143d9f | 5273 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
5274 | struct net_device *upper_dev, | |
5275 | struct list_head *up_list, | |
5276 | struct list_head *down_list, | |
5277 | void *private, bool master) | |
5d261913 VF |
5278 | { |
5279 | int ret; | |
5280 | ||
402dae96 VF |
5281 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private, |
5282 | master); | |
5d261913 VF |
5283 | if (ret) |
5284 | return ret; | |
5285 | ||
402dae96 VF |
5286 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private, |
5287 | false); | |
5d261913 | 5288 | if (ret) { |
2f268f12 | 5289 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5d261913 VF |
5290 | return ret; |
5291 | } | |
5292 | ||
5293 | return 0; | |
5294 | } | |
5295 | ||
1d143d9f | 5296 | static int __netdev_adjacent_dev_link(struct net_device *dev, |
5297 | struct net_device *upper_dev) | |
5d261913 | 5298 | { |
2f268f12 VF |
5299 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
5300 | &dev->all_adj_list.upper, | |
5301 | &upper_dev->all_adj_list.lower, | |
402dae96 | 5302 | NULL, false); |
5d261913 VF |
5303 | } |
5304 | ||
1d143d9f | 5305 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
5306 | struct net_device *upper_dev, | |
5307 | struct list_head *up_list, | |
5308 | struct list_head *down_list) | |
5d261913 | 5309 | { |
2f268f12 VF |
5310 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5311 | __netdev_adjacent_dev_remove(upper_dev, dev, down_list); | |
5d261913 VF |
5312 | } |
5313 | ||
1d143d9f | 5314 | static void __netdev_adjacent_dev_unlink(struct net_device *dev, |
5315 | struct net_device *upper_dev) | |
5d261913 | 5316 | { |
2f268f12 VF |
5317 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, |
5318 | &dev->all_adj_list.upper, | |
5319 | &upper_dev->all_adj_list.lower); | |
5320 | } | |
5321 | ||
1d143d9f | 5322 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
5323 | struct net_device *upper_dev, | |
5324 | void *private, bool master) | |
2f268f12 VF |
5325 | { |
5326 | int ret = __netdev_adjacent_dev_link(dev, upper_dev); | |
5327 | ||
5328 | if (ret) | |
5329 | return ret; | |
5330 | ||
5331 | ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, | |
5332 | &dev->adj_list.upper, | |
5333 | &upper_dev->adj_list.lower, | |
402dae96 | 5334 | private, master); |
2f268f12 VF |
5335 | if (ret) { |
5336 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5337 | return ret; | |
5338 | } | |
5339 | ||
5340 | return 0; | |
5d261913 VF |
5341 | } |
5342 | ||
1d143d9f | 5343 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
5344 | struct net_device *upper_dev) | |
2f268f12 VF |
5345 | { |
5346 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5347 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, | |
5348 | &dev->adj_list.upper, | |
5349 | &upper_dev->adj_list.lower); | |
5350 | } | |
5d261913 | 5351 | |
9ff162a8 | 5352 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 VF |
5353 | struct net_device *upper_dev, bool master, |
5354 | void *private) | |
9ff162a8 | 5355 | { |
0e4ead9d | 5356 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 VF |
5357 | struct netdev_adjacent *i, *j, *to_i, *to_j; |
5358 | int ret = 0; | |
9ff162a8 JP |
5359 | |
5360 | ASSERT_RTNL(); | |
5361 | ||
5362 | if (dev == upper_dev) | |
5363 | return -EBUSY; | |
5364 | ||
5365 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
6ea29da1 | 5366 | if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper)) |
9ff162a8 JP |
5367 | return -EBUSY; |
5368 | ||
6ea29da1 | 5369 | if (__netdev_find_adj(upper_dev, &dev->adj_list.upper)) |
9ff162a8 JP |
5370 | return -EEXIST; |
5371 | ||
5372 | if (master && netdev_master_upper_dev_get(dev)) | |
5373 | return -EBUSY; | |
5374 | ||
0e4ead9d JP |
5375 | changeupper_info.upper_dev = upper_dev; |
5376 | changeupper_info.master = master; | |
5377 | changeupper_info.linking = true; | |
5378 | ||
573c7ba0 JP |
5379 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
5380 | &changeupper_info.info); | |
5381 | ret = notifier_to_errno(ret); | |
5382 | if (ret) | |
5383 | return ret; | |
5384 | ||
402dae96 VF |
5385 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private, |
5386 | master); | |
5d261913 VF |
5387 | if (ret) |
5388 | return ret; | |
9ff162a8 | 5389 | |
5d261913 | 5390 | /* Now that we linked these devs, make all the upper_dev's |
2f268f12 | 5391 | * all_adj_list.upper visible to every dev's all_adj_list.lower an |
5d261913 VF |
5392 | * versa, and don't forget the devices itself. All of these |
5393 | * links are non-neighbours. | |
5394 | */ | |
2f268f12 VF |
5395 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5396 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5397 | pr_debug("Interlinking %s with %s, non-neighbour\n", | |
5398 | i->dev->name, j->dev->name); | |
5d261913 VF |
5399 | ret = __netdev_adjacent_dev_link(i->dev, j->dev); |
5400 | if (ret) | |
5401 | goto rollback_mesh; | |
5402 | } | |
5403 | } | |
5404 | ||
5405 | /* add dev to every upper_dev's upper device */ | |
2f268f12 VF |
5406 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5407 | pr_debug("linking %s's upper device %s with %s\n", | |
5408 | upper_dev->name, i->dev->name, dev->name); | |
5d261913 VF |
5409 | ret = __netdev_adjacent_dev_link(dev, i->dev); |
5410 | if (ret) | |
5411 | goto rollback_upper_mesh; | |
5412 | } | |
5413 | ||
5414 | /* add upper_dev to every dev's lower device */ | |
2f268f12 VF |
5415 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5416 | pr_debug("linking %s's lower device %s with %s\n", dev->name, | |
5417 | i->dev->name, upper_dev->name); | |
5d261913 VF |
5418 | ret = __netdev_adjacent_dev_link(i->dev, upper_dev); |
5419 | if (ret) | |
5420 | goto rollback_lower_mesh; | |
5421 | } | |
9ff162a8 | 5422 | |
0e4ead9d JP |
5423 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
5424 | &changeupper_info.info); | |
9ff162a8 | 5425 | return 0; |
5d261913 VF |
5426 | |
5427 | rollback_lower_mesh: | |
5428 | to_i = i; | |
2f268f12 | 5429 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5d261913 VF |
5430 | if (i == to_i) |
5431 | break; | |
5432 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); | |
5433 | } | |
5434 | ||
5435 | i = NULL; | |
5436 | ||
5437 | rollback_upper_mesh: | |
5438 | to_i = i; | |
2f268f12 | 5439 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5d261913 VF |
5440 | if (i == to_i) |
5441 | break; | |
5442 | __netdev_adjacent_dev_unlink(dev, i->dev); | |
5443 | } | |
5444 | ||
5445 | i = j = NULL; | |
5446 | ||
5447 | rollback_mesh: | |
5448 | to_i = i; | |
5449 | to_j = j; | |
2f268f12 VF |
5450 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5451 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5d261913 VF |
5452 | if (i == to_i && j == to_j) |
5453 | break; | |
5454 | __netdev_adjacent_dev_unlink(i->dev, j->dev); | |
5455 | } | |
5456 | if (i == to_i) | |
5457 | break; | |
5458 | } | |
5459 | ||
2f268f12 | 5460 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5461 | |
5462 | return ret; | |
9ff162a8 JP |
5463 | } |
5464 | ||
5465 | /** | |
5466 | * netdev_upper_dev_link - Add a link to the upper device | |
5467 | * @dev: device | |
5468 | * @upper_dev: new upper device | |
5469 | * | |
5470 | * Adds a link to device which is upper to this one. The caller must hold | |
5471 | * the RTNL lock. On a failure a negative errno code is returned. | |
5472 | * On success the reference counts are adjusted and the function | |
5473 | * returns zero. | |
5474 | */ | |
5475 | int netdev_upper_dev_link(struct net_device *dev, | |
5476 | struct net_device *upper_dev) | |
5477 | { | |
402dae96 | 5478 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL); |
9ff162a8 JP |
5479 | } |
5480 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
5481 | ||
5482 | /** | |
5483 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
5484 | * @dev: device | |
5485 | * @upper_dev: new upper device | |
5486 | * | |
5487 | * Adds a link to device which is upper to this one. In this case, only | |
5488 | * one master upper device can be linked, although other non-master devices | |
5489 | * might be linked as well. The caller must hold the RTNL lock. | |
5490 | * On a failure a negative errno code is returned. On success the reference | |
5491 | * counts are adjusted and the function returns zero. | |
5492 | */ | |
5493 | int netdev_master_upper_dev_link(struct net_device *dev, | |
5494 | struct net_device *upper_dev) | |
5495 | { | |
402dae96 | 5496 | return __netdev_upper_dev_link(dev, upper_dev, true, NULL); |
9ff162a8 JP |
5497 | } |
5498 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
5499 | ||
402dae96 VF |
5500 | int netdev_master_upper_dev_link_private(struct net_device *dev, |
5501 | struct net_device *upper_dev, | |
5502 | void *private) | |
5503 | { | |
5504 | return __netdev_upper_dev_link(dev, upper_dev, true, private); | |
5505 | } | |
5506 | EXPORT_SYMBOL(netdev_master_upper_dev_link_private); | |
5507 | ||
9ff162a8 JP |
5508 | /** |
5509 | * netdev_upper_dev_unlink - Removes a link to upper device | |
5510 | * @dev: device | |
5511 | * @upper_dev: new upper device | |
5512 | * | |
5513 | * Removes a link to device which is upper to this one. The caller must hold | |
5514 | * the RTNL lock. | |
5515 | */ | |
5516 | void netdev_upper_dev_unlink(struct net_device *dev, | |
5517 | struct net_device *upper_dev) | |
5518 | { | |
0e4ead9d | 5519 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 | 5520 | struct netdev_adjacent *i, *j; |
9ff162a8 JP |
5521 | ASSERT_RTNL(); |
5522 | ||
0e4ead9d JP |
5523 | changeupper_info.upper_dev = upper_dev; |
5524 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; | |
5525 | changeupper_info.linking = false; | |
5526 | ||
573c7ba0 JP |
5527 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
5528 | &changeupper_info.info); | |
5529 | ||
2f268f12 | 5530 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5531 | |
5532 | /* Here is the tricky part. We must remove all dev's lower | |
5533 | * devices from all upper_dev's upper devices and vice | |
5534 | * versa, to maintain the graph relationship. | |
5535 | */ | |
2f268f12 VF |
5536 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5537 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) | |
5d261913 VF |
5538 | __netdev_adjacent_dev_unlink(i->dev, j->dev); |
5539 | ||
5540 | /* remove also the devices itself from lower/upper device | |
5541 | * list | |
5542 | */ | |
2f268f12 | 5543 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5d261913 VF |
5544 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); |
5545 | ||
2f268f12 | 5546 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) |
5d261913 VF |
5547 | __netdev_adjacent_dev_unlink(dev, i->dev); |
5548 | ||
0e4ead9d JP |
5549 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
5550 | &changeupper_info.info); | |
9ff162a8 JP |
5551 | } |
5552 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
5553 | ||
61bd3857 MS |
5554 | /** |
5555 | * netdev_bonding_info_change - Dispatch event about slave change | |
5556 | * @dev: device | |
4a26e453 | 5557 | * @bonding_info: info to dispatch |
61bd3857 MS |
5558 | * |
5559 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
5560 | * The caller must hold the RTNL lock. | |
5561 | */ | |
5562 | void netdev_bonding_info_change(struct net_device *dev, | |
5563 | struct netdev_bonding_info *bonding_info) | |
5564 | { | |
5565 | struct netdev_notifier_bonding_info info; | |
5566 | ||
5567 | memcpy(&info.bonding_info, bonding_info, | |
5568 | sizeof(struct netdev_bonding_info)); | |
5569 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, | |
5570 | &info.info); | |
5571 | } | |
5572 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
5573 | ||
2ce1ee17 | 5574 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
5575 | { |
5576 | struct netdev_adjacent *iter; | |
5577 | ||
5578 | struct net *net = dev_net(dev); | |
5579 | ||
5580 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
5581 | if (!net_eq(net,dev_net(iter->dev))) | |
5582 | continue; | |
5583 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5584 | &iter->dev->adj_list.lower); | |
5585 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5586 | &dev->adj_list.upper); | |
5587 | } | |
5588 | ||
5589 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
5590 | if (!net_eq(net,dev_net(iter->dev))) | |
5591 | continue; | |
5592 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5593 | &iter->dev->adj_list.upper); | |
5594 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5595 | &dev->adj_list.lower); | |
5596 | } | |
5597 | } | |
5598 | ||
2ce1ee17 | 5599 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
5600 | { |
5601 | struct netdev_adjacent *iter; | |
5602 | ||
5603 | struct net *net = dev_net(dev); | |
5604 | ||
5605 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
5606 | if (!net_eq(net,dev_net(iter->dev))) | |
5607 | continue; | |
5608 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5609 | &iter->dev->adj_list.lower); | |
5610 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5611 | &dev->adj_list.upper); | |
5612 | } | |
5613 | ||
5614 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
5615 | if (!net_eq(net,dev_net(iter->dev))) | |
5616 | continue; | |
5617 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5618 | &iter->dev->adj_list.upper); | |
5619 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5620 | &dev->adj_list.lower); | |
5621 | } | |
5622 | } | |
5623 | ||
5bb025fa | 5624 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 5625 | { |
5bb025fa | 5626 | struct netdev_adjacent *iter; |
402dae96 | 5627 | |
4c75431a AF |
5628 | struct net *net = dev_net(dev); |
5629 | ||
5bb025fa | 5630 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
4c75431a AF |
5631 | if (!net_eq(net,dev_net(iter->dev))) |
5632 | continue; | |
5bb025fa VF |
5633 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
5634 | &iter->dev->adj_list.lower); | |
5635 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5636 | &iter->dev->adj_list.lower); | |
5637 | } | |
402dae96 | 5638 | |
5bb025fa | 5639 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
4c75431a AF |
5640 | if (!net_eq(net,dev_net(iter->dev))) |
5641 | continue; | |
5bb025fa VF |
5642 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
5643 | &iter->dev->adj_list.upper); | |
5644 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5645 | &iter->dev->adj_list.upper); | |
5646 | } | |
402dae96 | 5647 | } |
402dae96 VF |
5648 | |
5649 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
5650 | struct net_device *lower_dev) | |
5651 | { | |
5652 | struct netdev_adjacent *lower; | |
5653 | ||
5654 | if (!lower_dev) | |
5655 | return NULL; | |
6ea29da1 | 5656 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
5657 | if (!lower) |
5658 | return NULL; | |
5659 | ||
5660 | return lower->private; | |
5661 | } | |
5662 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
5663 | ||
4085ebe8 VY |
5664 | |
5665 | int dev_get_nest_level(struct net_device *dev, | |
5666 | bool (*type_check)(struct net_device *dev)) | |
5667 | { | |
5668 | struct net_device *lower = NULL; | |
5669 | struct list_head *iter; | |
5670 | int max_nest = -1; | |
5671 | int nest; | |
5672 | ||
5673 | ASSERT_RTNL(); | |
5674 | ||
5675 | netdev_for_each_lower_dev(dev, lower, iter) { | |
5676 | nest = dev_get_nest_level(lower, type_check); | |
5677 | if (max_nest < nest) | |
5678 | max_nest = nest; | |
5679 | } | |
5680 | ||
5681 | if (type_check(dev)) | |
5682 | max_nest++; | |
5683 | ||
5684 | return max_nest; | |
5685 | } | |
5686 | EXPORT_SYMBOL(dev_get_nest_level); | |
5687 | ||
b6c40d68 PM |
5688 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
5689 | { | |
d314774c SH |
5690 | const struct net_device_ops *ops = dev->netdev_ops; |
5691 | ||
d2615bf4 | 5692 | if (ops->ndo_change_rx_flags) |
d314774c | 5693 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
5694 | } |
5695 | ||
991fb3f7 | 5696 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5697 | { |
b536db93 | 5698 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
5699 | kuid_t uid; |
5700 | kgid_t gid; | |
1da177e4 | 5701 | |
24023451 PM |
5702 | ASSERT_RTNL(); |
5703 | ||
dad9b335 WC |
5704 | dev->flags |= IFF_PROMISC; |
5705 | dev->promiscuity += inc; | |
5706 | if (dev->promiscuity == 0) { | |
5707 | /* | |
5708 | * Avoid overflow. | |
5709 | * If inc causes overflow, untouch promisc and return error. | |
5710 | */ | |
5711 | if (inc < 0) | |
5712 | dev->flags &= ~IFF_PROMISC; | |
5713 | else { | |
5714 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
5715 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
5716 | dev->name); | |
dad9b335 WC |
5717 | return -EOVERFLOW; |
5718 | } | |
5719 | } | |
52609c0b | 5720 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
5721 | pr_info("device %s %s promiscuous mode\n", |
5722 | dev->name, | |
5723 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
5724 | if (audit_enabled) { |
5725 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
5726 | audit_log(current->audit_context, GFP_ATOMIC, |
5727 | AUDIT_ANOM_PROMISCUOUS, | |
5728 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
5729 | dev->name, (dev->flags & IFF_PROMISC), | |
5730 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 5731 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
5732 | from_kuid(&init_user_ns, uid), |
5733 | from_kgid(&init_user_ns, gid), | |
7759db82 | 5734 | audit_get_sessionid(current)); |
8192b0c4 | 5735 | } |
24023451 | 5736 | |
b6c40d68 | 5737 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 5738 | } |
991fb3f7 ND |
5739 | if (notify) |
5740 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 5741 | return 0; |
1da177e4 LT |
5742 | } |
5743 | ||
4417da66 PM |
5744 | /** |
5745 | * dev_set_promiscuity - update promiscuity count on a device | |
5746 | * @dev: device | |
5747 | * @inc: modifier | |
5748 | * | |
5749 | * Add or remove promiscuity from a device. While the count in the device | |
5750 | * remains above zero the interface remains promiscuous. Once it hits zero | |
5751 | * the device reverts back to normal filtering operation. A negative inc | |
5752 | * value is used to drop promiscuity on the device. | |
dad9b335 | 5753 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 5754 | */ |
dad9b335 | 5755 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 5756 | { |
b536db93 | 5757 | unsigned int old_flags = dev->flags; |
dad9b335 | 5758 | int err; |
4417da66 | 5759 | |
991fb3f7 | 5760 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 5761 | if (err < 0) |
dad9b335 | 5762 | return err; |
4417da66 PM |
5763 | if (dev->flags != old_flags) |
5764 | dev_set_rx_mode(dev); | |
dad9b335 | 5765 | return err; |
4417da66 | 5766 | } |
d1b19dff | 5767 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 5768 | |
991fb3f7 | 5769 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5770 | { |
991fb3f7 | 5771 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 5772 | |
24023451 PM |
5773 | ASSERT_RTNL(); |
5774 | ||
1da177e4 | 5775 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
5776 | dev->allmulti += inc; |
5777 | if (dev->allmulti == 0) { | |
5778 | /* | |
5779 | * Avoid overflow. | |
5780 | * If inc causes overflow, untouch allmulti and return error. | |
5781 | */ | |
5782 | if (inc < 0) | |
5783 | dev->flags &= ~IFF_ALLMULTI; | |
5784 | else { | |
5785 | dev->allmulti -= inc; | |
7b6cd1ce JP |
5786 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
5787 | dev->name); | |
dad9b335 WC |
5788 | return -EOVERFLOW; |
5789 | } | |
5790 | } | |
24023451 | 5791 | if (dev->flags ^ old_flags) { |
b6c40d68 | 5792 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 5793 | dev_set_rx_mode(dev); |
991fb3f7 ND |
5794 | if (notify) |
5795 | __dev_notify_flags(dev, old_flags, | |
5796 | dev->gflags ^ old_gflags); | |
24023451 | 5797 | } |
dad9b335 | 5798 | return 0; |
4417da66 | 5799 | } |
991fb3f7 ND |
5800 | |
5801 | /** | |
5802 | * dev_set_allmulti - update allmulti count on a device | |
5803 | * @dev: device | |
5804 | * @inc: modifier | |
5805 | * | |
5806 | * Add or remove reception of all multicast frames to a device. While the | |
5807 | * count in the device remains above zero the interface remains listening | |
5808 | * to all interfaces. Once it hits zero the device reverts back to normal | |
5809 | * filtering operation. A negative @inc value is used to drop the counter | |
5810 | * when releasing a resource needing all multicasts. | |
5811 | * Return 0 if successful or a negative errno code on error. | |
5812 | */ | |
5813 | ||
5814 | int dev_set_allmulti(struct net_device *dev, int inc) | |
5815 | { | |
5816 | return __dev_set_allmulti(dev, inc, true); | |
5817 | } | |
d1b19dff | 5818 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
5819 | |
5820 | /* | |
5821 | * Upload unicast and multicast address lists to device and | |
5822 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 5823 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
5824 | * are present. |
5825 | */ | |
5826 | void __dev_set_rx_mode(struct net_device *dev) | |
5827 | { | |
d314774c SH |
5828 | const struct net_device_ops *ops = dev->netdev_ops; |
5829 | ||
4417da66 PM |
5830 | /* dev_open will call this function so the list will stay sane. */ |
5831 | if (!(dev->flags&IFF_UP)) | |
5832 | return; | |
5833 | ||
5834 | if (!netif_device_present(dev)) | |
40b77c94 | 5835 | return; |
4417da66 | 5836 | |
01789349 | 5837 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
5838 | /* Unicast addresses changes may only happen under the rtnl, |
5839 | * therefore calling __dev_set_promiscuity here is safe. | |
5840 | */ | |
32e7bfc4 | 5841 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 5842 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 5843 | dev->uc_promisc = true; |
32e7bfc4 | 5844 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 5845 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 5846 | dev->uc_promisc = false; |
4417da66 | 5847 | } |
4417da66 | 5848 | } |
01789349 JP |
5849 | |
5850 | if (ops->ndo_set_rx_mode) | |
5851 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
5852 | } |
5853 | ||
5854 | void dev_set_rx_mode(struct net_device *dev) | |
5855 | { | |
b9e40857 | 5856 | netif_addr_lock_bh(dev); |
4417da66 | 5857 | __dev_set_rx_mode(dev); |
b9e40857 | 5858 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
5859 | } |
5860 | ||
f0db275a SH |
5861 | /** |
5862 | * dev_get_flags - get flags reported to userspace | |
5863 | * @dev: device | |
5864 | * | |
5865 | * Get the combination of flag bits exported through APIs to userspace. | |
5866 | */ | |
95c96174 | 5867 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 5868 | { |
95c96174 | 5869 | unsigned int flags; |
1da177e4 LT |
5870 | |
5871 | flags = (dev->flags & ~(IFF_PROMISC | | |
5872 | IFF_ALLMULTI | | |
b00055aa SR |
5873 | IFF_RUNNING | |
5874 | IFF_LOWER_UP | | |
5875 | IFF_DORMANT)) | | |
1da177e4 LT |
5876 | (dev->gflags & (IFF_PROMISC | |
5877 | IFF_ALLMULTI)); | |
5878 | ||
b00055aa SR |
5879 | if (netif_running(dev)) { |
5880 | if (netif_oper_up(dev)) | |
5881 | flags |= IFF_RUNNING; | |
5882 | if (netif_carrier_ok(dev)) | |
5883 | flags |= IFF_LOWER_UP; | |
5884 | if (netif_dormant(dev)) | |
5885 | flags |= IFF_DORMANT; | |
5886 | } | |
1da177e4 LT |
5887 | |
5888 | return flags; | |
5889 | } | |
d1b19dff | 5890 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 5891 | |
bd380811 | 5892 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 5893 | { |
b536db93 | 5894 | unsigned int old_flags = dev->flags; |
bd380811 | 5895 | int ret; |
1da177e4 | 5896 | |
24023451 PM |
5897 | ASSERT_RTNL(); |
5898 | ||
1da177e4 LT |
5899 | /* |
5900 | * Set the flags on our device. | |
5901 | */ | |
5902 | ||
5903 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
5904 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
5905 | IFF_AUTOMEDIA)) | | |
5906 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
5907 | IFF_ALLMULTI)); | |
5908 | ||
5909 | /* | |
5910 | * Load in the correct multicast list now the flags have changed. | |
5911 | */ | |
5912 | ||
b6c40d68 PM |
5913 | if ((old_flags ^ flags) & IFF_MULTICAST) |
5914 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 5915 | |
4417da66 | 5916 | dev_set_rx_mode(dev); |
1da177e4 LT |
5917 | |
5918 | /* | |
5919 | * Have we downed the interface. We handle IFF_UP ourselves | |
5920 | * according to user attempts to set it, rather than blindly | |
5921 | * setting it. | |
5922 | */ | |
5923 | ||
5924 | ret = 0; | |
d215d10f | 5925 | if ((old_flags ^ flags) & IFF_UP) |
bd380811 | 5926 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 | 5927 | |
1da177e4 | 5928 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 5929 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 5930 | unsigned int old_flags = dev->flags; |
d1b19dff | 5931 | |
1da177e4 | 5932 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
5933 | |
5934 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
5935 | if (dev->flags != old_flags) | |
5936 | dev_set_rx_mode(dev); | |
1da177e4 LT |
5937 | } |
5938 | ||
5939 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
5940 | is important. Some (broken) drivers set IFF_PROMISC, when | |
5941 | IFF_ALLMULTI is requested not asking us and not reporting. | |
5942 | */ | |
5943 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
5944 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
5945 | ||
1da177e4 | 5946 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 5947 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
5948 | } |
5949 | ||
bd380811 PM |
5950 | return ret; |
5951 | } | |
5952 | ||
a528c219 ND |
5953 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
5954 | unsigned int gchanges) | |
bd380811 PM |
5955 | { |
5956 | unsigned int changes = dev->flags ^ old_flags; | |
5957 | ||
a528c219 | 5958 | if (gchanges) |
7f294054 | 5959 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 5960 | |
bd380811 PM |
5961 | if (changes & IFF_UP) { |
5962 | if (dev->flags & IFF_UP) | |
5963 | call_netdevice_notifiers(NETDEV_UP, dev); | |
5964 | else | |
5965 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
5966 | } | |
5967 | ||
5968 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
5969 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
5970 | struct netdev_notifier_change_info change_info; | |
5971 | ||
5972 | change_info.flags_changed = changes; | |
5973 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
5974 | &change_info.info); | |
5975 | } | |
bd380811 PM |
5976 | } |
5977 | ||
5978 | /** | |
5979 | * dev_change_flags - change device settings | |
5980 | * @dev: device | |
5981 | * @flags: device state flags | |
5982 | * | |
5983 | * Change settings on device based state flags. The flags are | |
5984 | * in the userspace exported format. | |
5985 | */ | |
b536db93 | 5986 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 5987 | { |
b536db93 | 5988 | int ret; |
991fb3f7 | 5989 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
5990 | |
5991 | ret = __dev_change_flags(dev, flags); | |
5992 | if (ret < 0) | |
5993 | return ret; | |
5994 | ||
991fb3f7 | 5995 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 5996 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
5997 | return ret; |
5998 | } | |
d1b19dff | 5999 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 6000 | |
2315dc91 VF |
6001 | static int __dev_set_mtu(struct net_device *dev, int new_mtu) |
6002 | { | |
6003 | const struct net_device_ops *ops = dev->netdev_ops; | |
6004 | ||
6005 | if (ops->ndo_change_mtu) | |
6006 | return ops->ndo_change_mtu(dev, new_mtu); | |
6007 | ||
6008 | dev->mtu = new_mtu; | |
6009 | return 0; | |
6010 | } | |
6011 | ||
f0db275a SH |
6012 | /** |
6013 | * dev_set_mtu - Change maximum transfer unit | |
6014 | * @dev: device | |
6015 | * @new_mtu: new transfer unit | |
6016 | * | |
6017 | * Change the maximum transfer size of the network device. | |
6018 | */ | |
1da177e4 LT |
6019 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
6020 | { | |
2315dc91 | 6021 | int err, orig_mtu; |
1da177e4 LT |
6022 | |
6023 | if (new_mtu == dev->mtu) | |
6024 | return 0; | |
6025 | ||
6026 | /* MTU must be positive. */ | |
6027 | if (new_mtu < 0) | |
6028 | return -EINVAL; | |
6029 | ||
6030 | if (!netif_device_present(dev)) | |
6031 | return -ENODEV; | |
6032 | ||
1d486bfb VF |
6033 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
6034 | err = notifier_to_errno(err); | |
6035 | if (err) | |
6036 | return err; | |
d314774c | 6037 | |
2315dc91 VF |
6038 | orig_mtu = dev->mtu; |
6039 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 6040 | |
2315dc91 VF |
6041 | if (!err) { |
6042 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6043 | err = notifier_to_errno(err); | |
6044 | if (err) { | |
6045 | /* setting mtu back and notifying everyone again, | |
6046 | * so that they have a chance to revert changes. | |
6047 | */ | |
6048 | __dev_set_mtu(dev, orig_mtu); | |
6049 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6050 | } | |
6051 | } | |
1da177e4 LT |
6052 | return err; |
6053 | } | |
d1b19dff | 6054 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 6055 | |
cbda10fa VD |
6056 | /** |
6057 | * dev_set_group - Change group this device belongs to | |
6058 | * @dev: device | |
6059 | * @new_group: group this device should belong to | |
6060 | */ | |
6061 | void dev_set_group(struct net_device *dev, int new_group) | |
6062 | { | |
6063 | dev->group = new_group; | |
6064 | } | |
6065 | EXPORT_SYMBOL(dev_set_group); | |
6066 | ||
f0db275a SH |
6067 | /** |
6068 | * dev_set_mac_address - Change Media Access Control Address | |
6069 | * @dev: device | |
6070 | * @sa: new address | |
6071 | * | |
6072 | * Change the hardware (MAC) address of the device | |
6073 | */ | |
1da177e4 LT |
6074 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
6075 | { | |
d314774c | 6076 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
6077 | int err; |
6078 | ||
d314774c | 6079 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
6080 | return -EOPNOTSUPP; |
6081 | if (sa->sa_family != dev->type) | |
6082 | return -EINVAL; | |
6083 | if (!netif_device_present(dev)) | |
6084 | return -ENODEV; | |
d314774c | 6085 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
6086 | if (err) |
6087 | return err; | |
fbdeca2d | 6088 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 6089 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 6090 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 6091 | return 0; |
1da177e4 | 6092 | } |
d1b19dff | 6093 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 6094 | |
4bf84c35 JP |
6095 | /** |
6096 | * dev_change_carrier - Change device carrier | |
6097 | * @dev: device | |
691b3b7e | 6098 | * @new_carrier: new value |
4bf84c35 JP |
6099 | * |
6100 | * Change device carrier | |
6101 | */ | |
6102 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
6103 | { | |
6104 | const struct net_device_ops *ops = dev->netdev_ops; | |
6105 | ||
6106 | if (!ops->ndo_change_carrier) | |
6107 | return -EOPNOTSUPP; | |
6108 | if (!netif_device_present(dev)) | |
6109 | return -ENODEV; | |
6110 | return ops->ndo_change_carrier(dev, new_carrier); | |
6111 | } | |
6112 | EXPORT_SYMBOL(dev_change_carrier); | |
6113 | ||
66b52b0d JP |
6114 | /** |
6115 | * dev_get_phys_port_id - Get device physical port ID | |
6116 | * @dev: device | |
6117 | * @ppid: port ID | |
6118 | * | |
6119 | * Get device physical port ID | |
6120 | */ | |
6121 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 6122 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
6123 | { |
6124 | const struct net_device_ops *ops = dev->netdev_ops; | |
6125 | ||
6126 | if (!ops->ndo_get_phys_port_id) | |
6127 | return -EOPNOTSUPP; | |
6128 | return ops->ndo_get_phys_port_id(dev, ppid); | |
6129 | } | |
6130 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
6131 | ||
db24a904 DA |
6132 | /** |
6133 | * dev_get_phys_port_name - Get device physical port name | |
6134 | * @dev: device | |
6135 | * @name: port name | |
6136 | * | |
6137 | * Get device physical port name | |
6138 | */ | |
6139 | int dev_get_phys_port_name(struct net_device *dev, | |
6140 | char *name, size_t len) | |
6141 | { | |
6142 | const struct net_device_ops *ops = dev->netdev_ops; | |
6143 | ||
6144 | if (!ops->ndo_get_phys_port_name) | |
6145 | return -EOPNOTSUPP; | |
6146 | return ops->ndo_get_phys_port_name(dev, name, len); | |
6147 | } | |
6148 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
6149 | ||
d746d707 AK |
6150 | /** |
6151 | * dev_change_proto_down - update protocol port state information | |
6152 | * @dev: device | |
6153 | * @proto_down: new value | |
6154 | * | |
6155 | * This info can be used by switch drivers to set the phys state of the | |
6156 | * port. | |
6157 | */ | |
6158 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
6159 | { | |
6160 | const struct net_device_ops *ops = dev->netdev_ops; | |
6161 | ||
6162 | if (!ops->ndo_change_proto_down) | |
6163 | return -EOPNOTSUPP; | |
6164 | if (!netif_device_present(dev)) | |
6165 | return -ENODEV; | |
6166 | return ops->ndo_change_proto_down(dev, proto_down); | |
6167 | } | |
6168 | EXPORT_SYMBOL(dev_change_proto_down); | |
6169 | ||
1da177e4 LT |
6170 | /** |
6171 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 6172 | * @net: the applicable net namespace |
1da177e4 LT |
6173 | * |
6174 | * Returns a suitable unique value for a new device interface | |
6175 | * number. The caller must hold the rtnl semaphore or the | |
6176 | * dev_base_lock to be sure it remains unique. | |
6177 | */ | |
881d966b | 6178 | static int dev_new_index(struct net *net) |
1da177e4 | 6179 | { |
aa79e66e | 6180 | int ifindex = net->ifindex; |
1da177e4 LT |
6181 | for (;;) { |
6182 | if (++ifindex <= 0) | |
6183 | ifindex = 1; | |
881d966b | 6184 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 6185 | return net->ifindex = ifindex; |
1da177e4 LT |
6186 | } |
6187 | } | |
6188 | ||
1da177e4 | 6189 | /* Delayed registration/unregisteration */ |
3b5b34fd | 6190 | static LIST_HEAD(net_todo_list); |
200b916f | 6191 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 6192 | |
6f05f629 | 6193 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 6194 | { |
1da177e4 | 6195 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 6196 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
6197 | } |
6198 | ||
9b5e383c | 6199 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 6200 | { |
e93737b0 | 6201 | struct net_device *dev, *tmp; |
5cde2829 | 6202 | LIST_HEAD(close_head); |
9b5e383c | 6203 | |
93ee31f1 DL |
6204 | BUG_ON(dev_boot_phase); |
6205 | ASSERT_RTNL(); | |
6206 | ||
e93737b0 | 6207 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 6208 | /* Some devices call without registering |
e93737b0 KK |
6209 | * for initialization unwind. Remove those |
6210 | * devices and proceed with the remaining. | |
9b5e383c ED |
6211 | */ |
6212 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
6213 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
6214 | dev->name, dev); | |
93ee31f1 | 6215 | |
9b5e383c | 6216 | WARN_ON(1); |
e93737b0 KK |
6217 | list_del(&dev->unreg_list); |
6218 | continue; | |
9b5e383c | 6219 | } |
449f4544 | 6220 | dev->dismantle = true; |
9b5e383c | 6221 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 6222 | } |
93ee31f1 | 6223 | |
44345724 | 6224 | /* If device is running, close it first. */ |
5cde2829 EB |
6225 | list_for_each_entry(dev, head, unreg_list) |
6226 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 6227 | dev_close_many(&close_head, true); |
93ee31f1 | 6228 | |
44345724 | 6229 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
6230 | /* And unlink it from device chain. */ |
6231 | unlist_netdevice(dev); | |
93ee31f1 | 6232 | |
9b5e383c | 6233 | dev->reg_state = NETREG_UNREGISTERING; |
e9e4dd32 | 6234 | on_each_cpu(flush_backlog, dev, 1); |
9b5e383c | 6235 | } |
93ee31f1 DL |
6236 | |
6237 | synchronize_net(); | |
6238 | ||
9b5e383c | 6239 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
6240 | struct sk_buff *skb = NULL; |
6241 | ||
9b5e383c ED |
6242 | /* Shutdown queueing discipline. */ |
6243 | dev_shutdown(dev); | |
93ee31f1 DL |
6244 | |
6245 | ||
9b5e383c ED |
6246 | /* Notify protocols, that we are about to destroy |
6247 | this device. They should clean all the things. | |
6248 | */ | |
6249 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 6250 | |
395eea6c MB |
6251 | if (!dev->rtnl_link_ops || |
6252 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
6253 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, | |
6254 | GFP_KERNEL); | |
6255 | ||
9b5e383c ED |
6256 | /* |
6257 | * Flush the unicast and multicast chains | |
6258 | */ | |
a748ee24 | 6259 | dev_uc_flush(dev); |
22bedad3 | 6260 | dev_mc_flush(dev); |
93ee31f1 | 6261 | |
9b5e383c ED |
6262 | if (dev->netdev_ops->ndo_uninit) |
6263 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 6264 | |
395eea6c MB |
6265 | if (skb) |
6266 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 6267 | |
9ff162a8 JP |
6268 | /* Notifier chain MUST detach us all upper devices. */ |
6269 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 6270 | |
9b5e383c ED |
6271 | /* Remove entries from kobject tree */ |
6272 | netdev_unregister_kobject(dev); | |
024e9679 AD |
6273 | #ifdef CONFIG_XPS |
6274 | /* Remove XPS queueing entries */ | |
6275 | netif_reset_xps_queues_gt(dev, 0); | |
6276 | #endif | |
9b5e383c | 6277 | } |
93ee31f1 | 6278 | |
850a545b | 6279 | synchronize_net(); |
395264d5 | 6280 | |
a5ee1551 | 6281 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
6282 | dev_put(dev); |
6283 | } | |
6284 | ||
6285 | static void rollback_registered(struct net_device *dev) | |
6286 | { | |
6287 | LIST_HEAD(single); | |
6288 | ||
6289 | list_add(&dev->unreg_list, &single); | |
6290 | rollback_registered_many(&single); | |
ceaaec98 | 6291 | list_del(&single); |
93ee31f1 DL |
6292 | } |
6293 | ||
fd867d51 JW |
6294 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
6295 | struct net_device *upper, netdev_features_t features) | |
6296 | { | |
6297 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
6298 | netdev_features_t feature; | |
5ba3f7d6 | 6299 | int feature_bit; |
fd867d51 | 6300 | |
5ba3f7d6 JW |
6301 | for_each_netdev_feature(&upper_disables, feature_bit) { |
6302 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
6303 | if (!(upper->wanted_features & feature) |
6304 | && (features & feature)) { | |
6305 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
6306 | &feature, upper->name); | |
6307 | features &= ~feature; | |
6308 | } | |
6309 | } | |
6310 | ||
6311 | return features; | |
6312 | } | |
6313 | ||
6314 | static void netdev_sync_lower_features(struct net_device *upper, | |
6315 | struct net_device *lower, netdev_features_t features) | |
6316 | { | |
6317 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
6318 | netdev_features_t feature; | |
5ba3f7d6 | 6319 | int feature_bit; |
fd867d51 | 6320 | |
5ba3f7d6 JW |
6321 | for_each_netdev_feature(&upper_disables, feature_bit) { |
6322 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
6323 | if (!(features & feature) && (lower->features & feature)) { |
6324 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
6325 | &feature, lower->name); | |
6326 | lower->wanted_features &= ~feature; | |
6327 | netdev_update_features(lower); | |
6328 | ||
6329 | if (unlikely(lower->features & feature)) | |
6330 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
6331 | &feature, lower->name); | |
6332 | } | |
6333 | } | |
6334 | } | |
6335 | ||
c8f44aff MM |
6336 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
6337 | netdev_features_t features) | |
b63365a2 | 6338 | { |
57422dc5 MM |
6339 | /* Fix illegal checksum combinations */ |
6340 | if ((features & NETIF_F_HW_CSUM) && | |
6341 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 6342 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
6343 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
6344 | } | |
6345 | ||
b63365a2 | 6346 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 6347 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 6348 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 6349 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
6350 | } |
6351 | ||
ec5f0615 PS |
6352 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
6353 | !(features & NETIF_F_IP_CSUM)) { | |
6354 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
6355 | features &= ~NETIF_F_TSO; | |
6356 | features &= ~NETIF_F_TSO_ECN; | |
6357 | } | |
6358 | ||
6359 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
6360 | !(features & NETIF_F_IPV6_CSUM)) { | |
6361 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
6362 | features &= ~NETIF_F_TSO6; | |
6363 | } | |
6364 | ||
31d8b9e0 BH |
6365 | /* TSO ECN requires that TSO is present as well. */ |
6366 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
6367 | features &= ~NETIF_F_TSO_ECN; | |
6368 | ||
212b573f MM |
6369 | /* Software GSO depends on SG. */ |
6370 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 6371 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
6372 | features &= ~NETIF_F_GSO; |
6373 | } | |
6374 | ||
acd1130e | 6375 | /* UFO needs SG and checksumming */ |
b63365a2 | 6376 | if (features & NETIF_F_UFO) { |
79032644 MM |
6377 | /* maybe split UFO into V4 and V6? */ |
6378 | if (!((features & NETIF_F_GEN_CSUM) || | |
6379 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
6380 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 6381 | netdev_dbg(dev, |
acd1130e | 6382 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
6383 | features &= ~NETIF_F_UFO; |
6384 | } | |
6385 | ||
6386 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 6387 | netdev_dbg(dev, |
acd1130e | 6388 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
6389 | features &= ~NETIF_F_UFO; |
6390 | } | |
6391 | } | |
6392 | ||
d0290214 JP |
6393 | #ifdef CONFIG_NET_RX_BUSY_POLL |
6394 | if (dev->netdev_ops->ndo_busy_poll) | |
6395 | features |= NETIF_F_BUSY_POLL; | |
6396 | else | |
6397 | #endif | |
6398 | features &= ~NETIF_F_BUSY_POLL; | |
6399 | ||
b63365a2 HX |
6400 | return features; |
6401 | } | |
b63365a2 | 6402 | |
6cb6a27c | 6403 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 6404 | { |
fd867d51 | 6405 | struct net_device *upper, *lower; |
c8f44aff | 6406 | netdev_features_t features; |
fd867d51 | 6407 | struct list_head *iter; |
e7868a85 | 6408 | int err = -1; |
5455c699 | 6409 | |
87267485 MM |
6410 | ASSERT_RTNL(); |
6411 | ||
5455c699 MM |
6412 | features = netdev_get_wanted_features(dev); |
6413 | ||
6414 | if (dev->netdev_ops->ndo_fix_features) | |
6415 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
6416 | ||
6417 | /* driver might be less strict about feature dependencies */ | |
6418 | features = netdev_fix_features(dev, features); | |
6419 | ||
fd867d51 JW |
6420 | /* some features can't be enabled if they're off an an upper device */ |
6421 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
6422 | features = netdev_sync_upper_features(dev, upper, features); | |
6423 | ||
5455c699 | 6424 | if (dev->features == features) |
e7868a85 | 6425 | goto sync_lower; |
5455c699 | 6426 | |
c8f44aff MM |
6427 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
6428 | &dev->features, &features); | |
5455c699 MM |
6429 | |
6430 | if (dev->netdev_ops->ndo_set_features) | |
6431 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
6432 | else |
6433 | err = 0; | |
5455c699 | 6434 | |
6cb6a27c | 6435 | if (unlikely(err < 0)) { |
5455c699 | 6436 | netdev_err(dev, |
c8f44aff MM |
6437 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
6438 | err, &features, &dev->features); | |
00ee5927 | 6439 | return 0; |
6cb6a27c MM |
6440 | } |
6441 | ||
e7868a85 | 6442 | sync_lower: |
fd867d51 JW |
6443 | /* some features must be disabled on lower devices when disabled |
6444 | * on an upper device (think: bonding master or bridge) | |
6445 | */ | |
6446 | netdev_for_each_lower_dev(dev, lower, iter) | |
6447 | netdev_sync_lower_features(dev, lower, features); | |
6448 | ||
6cb6a27c MM |
6449 | if (!err) |
6450 | dev->features = features; | |
6451 | ||
e7868a85 | 6452 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
6453 | } |
6454 | ||
afe12cc8 MM |
6455 | /** |
6456 | * netdev_update_features - recalculate device features | |
6457 | * @dev: the device to check | |
6458 | * | |
6459 | * Recalculate dev->features set and send notifications if it | |
6460 | * has changed. Should be called after driver or hardware dependent | |
6461 | * conditions might have changed that influence the features. | |
6462 | */ | |
6cb6a27c MM |
6463 | void netdev_update_features(struct net_device *dev) |
6464 | { | |
6465 | if (__netdev_update_features(dev)) | |
6466 | netdev_features_change(dev); | |
5455c699 MM |
6467 | } |
6468 | EXPORT_SYMBOL(netdev_update_features); | |
6469 | ||
afe12cc8 MM |
6470 | /** |
6471 | * netdev_change_features - recalculate device features | |
6472 | * @dev: the device to check | |
6473 | * | |
6474 | * Recalculate dev->features set and send notifications even | |
6475 | * if they have not changed. Should be called instead of | |
6476 | * netdev_update_features() if also dev->vlan_features might | |
6477 | * have changed to allow the changes to be propagated to stacked | |
6478 | * VLAN devices. | |
6479 | */ | |
6480 | void netdev_change_features(struct net_device *dev) | |
6481 | { | |
6482 | __netdev_update_features(dev); | |
6483 | netdev_features_change(dev); | |
6484 | } | |
6485 | EXPORT_SYMBOL(netdev_change_features); | |
6486 | ||
fc4a7489 PM |
6487 | /** |
6488 | * netif_stacked_transfer_operstate - transfer operstate | |
6489 | * @rootdev: the root or lower level device to transfer state from | |
6490 | * @dev: the device to transfer operstate to | |
6491 | * | |
6492 | * Transfer operational state from root to device. This is normally | |
6493 | * called when a stacking relationship exists between the root | |
6494 | * device and the device(a leaf device). | |
6495 | */ | |
6496 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
6497 | struct net_device *dev) | |
6498 | { | |
6499 | if (rootdev->operstate == IF_OPER_DORMANT) | |
6500 | netif_dormant_on(dev); | |
6501 | else | |
6502 | netif_dormant_off(dev); | |
6503 | ||
6504 | if (netif_carrier_ok(rootdev)) { | |
6505 | if (!netif_carrier_ok(dev)) | |
6506 | netif_carrier_on(dev); | |
6507 | } else { | |
6508 | if (netif_carrier_ok(dev)) | |
6509 | netif_carrier_off(dev); | |
6510 | } | |
6511 | } | |
6512 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
6513 | ||
a953be53 | 6514 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
6515 | static int netif_alloc_rx_queues(struct net_device *dev) |
6516 | { | |
1b4bf461 | 6517 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 6518 | struct netdev_rx_queue *rx; |
10595902 | 6519 | size_t sz = count * sizeof(*rx); |
1b4bf461 | 6520 | |
bd25fa7b | 6521 | BUG_ON(count < 1); |
1b4bf461 | 6522 | |
10595902 PG |
6523 | rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6524 | if (!rx) { | |
6525 | rx = vzalloc(sz); | |
6526 | if (!rx) | |
6527 | return -ENOMEM; | |
6528 | } | |
bd25fa7b TH |
6529 | dev->_rx = rx; |
6530 | ||
bd25fa7b | 6531 | for (i = 0; i < count; i++) |
fe822240 | 6532 | rx[i].dev = dev; |
1b4bf461 ED |
6533 | return 0; |
6534 | } | |
bf264145 | 6535 | #endif |
1b4bf461 | 6536 | |
aa942104 CG |
6537 | static void netdev_init_one_queue(struct net_device *dev, |
6538 | struct netdev_queue *queue, void *_unused) | |
6539 | { | |
6540 | /* Initialize queue lock */ | |
6541 | spin_lock_init(&queue->_xmit_lock); | |
6542 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
6543 | queue->xmit_lock_owner = -1; | |
b236da69 | 6544 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 6545 | queue->dev = dev; |
114cf580 TH |
6546 | #ifdef CONFIG_BQL |
6547 | dql_init(&queue->dql, HZ); | |
6548 | #endif | |
aa942104 CG |
6549 | } |
6550 | ||
60877a32 ED |
6551 | static void netif_free_tx_queues(struct net_device *dev) |
6552 | { | |
4cb28970 | 6553 | kvfree(dev->_tx); |
60877a32 ED |
6554 | } |
6555 | ||
e6484930 TH |
6556 | static int netif_alloc_netdev_queues(struct net_device *dev) |
6557 | { | |
6558 | unsigned int count = dev->num_tx_queues; | |
6559 | struct netdev_queue *tx; | |
60877a32 | 6560 | size_t sz = count * sizeof(*tx); |
e6484930 | 6561 | |
d339727c ED |
6562 | if (count < 1 || count > 0xffff) |
6563 | return -EINVAL; | |
62b5942a | 6564 | |
60877a32 ED |
6565 | tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6566 | if (!tx) { | |
6567 | tx = vzalloc(sz); | |
6568 | if (!tx) | |
6569 | return -ENOMEM; | |
6570 | } | |
e6484930 | 6571 | dev->_tx = tx; |
1d24eb48 | 6572 | |
e6484930 TH |
6573 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
6574 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
6575 | |
6576 | return 0; | |
e6484930 TH |
6577 | } |
6578 | ||
a2029240 DV |
6579 | void netif_tx_stop_all_queues(struct net_device *dev) |
6580 | { | |
6581 | unsigned int i; | |
6582 | ||
6583 | for (i = 0; i < dev->num_tx_queues; i++) { | |
6584 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
6585 | netif_tx_stop_queue(txq); | |
6586 | } | |
6587 | } | |
6588 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
6589 | ||
1da177e4 LT |
6590 | /** |
6591 | * register_netdevice - register a network device | |
6592 | * @dev: device to register | |
6593 | * | |
6594 | * Take a completed network device structure and add it to the kernel | |
6595 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6596 | * chain. 0 is returned on success. A negative errno code is returned | |
6597 | * on a failure to set up the device, or if the name is a duplicate. | |
6598 | * | |
6599 | * Callers must hold the rtnl semaphore. You may want | |
6600 | * register_netdev() instead of this. | |
6601 | * | |
6602 | * BUGS: | |
6603 | * The locking appears insufficient to guarantee two parallel registers | |
6604 | * will not get the same name. | |
6605 | */ | |
6606 | ||
6607 | int register_netdevice(struct net_device *dev) | |
6608 | { | |
1da177e4 | 6609 | int ret; |
d314774c | 6610 | struct net *net = dev_net(dev); |
1da177e4 LT |
6611 | |
6612 | BUG_ON(dev_boot_phase); | |
6613 | ASSERT_RTNL(); | |
6614 | ||
b17a7c17 SH |
6615 | might_sleep(); |
6616 | ||
1da177e4 LT |
6617 | /* When net_device's are persistent, this will be fatal. */ |
6618 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 6619 | BUG_ON(!net); |
1da177e4 | 6620 | |
f1f28aa3 | 6621 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 6622 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 6623 | |
828de4f6 | 6624 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
6625 | if (ret < 0) |
6626 | goto out; | |
6627 | ||
1da177e4 | 6628 | /* Init, if this function is available */ |
d314774c SH |
6629 | if (dev->netdev_ops->ndo_init) { |
6630 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
6631 | if (ret) { |
6632 | if (ret > 0) | |
6633 | ret = -EIO; | |
90833aa4 | 6634 | goto out; |
1da177e4 LT |
6635 | } |
6636 | } | |
4ec93edb | 6637 | |
f646968f PM |
6638 | if (((dev->hw_features | dev->features) & |
6639 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
6640 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
6641 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
6642 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
6643 | ret = -EINVAL; | |
6644 | goto err_uninit; | |
6645 | } | |
6646 | ||
9c7dafbf PE |
6647 | ret = -EBUSY; |
6648 | if (!dev->ifindex) | |
6649 | dev->ifindex = dev_new_index(net); | |
6650 | else if (__dev_get_by_index(net, dev->ifindex)) | |
6651 | goto err_uninit; | |
6652 | ||
5455c699 MM |
6653 | /* Transfer changeable features to wanted_features and enable |
6654 | * software offloads (GSO and GRO). | |
6655 | */ | |
6656 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
6657 | dev->features |= NETIF_F_SOFT_FEATURES; |
6658 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 6659 | |
34324dc2 MM |
6660 | if (!(dev->flags & IFF_LOOPBACK)) { |
6661 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
c6e1a0d1 TH |
6662 | } |
6663 | ||
1180e7d6 | 6664 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 6665 | */ |
1180e7d6 | 6666 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 6667 | |
ee579677 PS |
6668 | /* Make NETIF_F_SG inheritable to tunnel devices. |
6669 | */ | |
6670 | dev->hw_enc_features |= NETIF_F_SG; | |
6671 | ||
0d89d203 SH |
6672 | /* Make NETIF_F_SG inheritable to MPLS. |
6673 | */ | |
6674 | dev->mpls_features |= NETIF_F_SG; | |
6675 | ||
7ffbe3fd JB |
6676 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
6677 | ret = notifier_to_errno(ret); | |
6678 | if (ret) | |
6679 | goto err_uninit; | |
6680 | ||
8b41d188 | 6681 | ret = netdev_register_kobject(dev); |
b17a7c17 | 6682 | if (ret) |
7ce1b0ed | 6683 | goto err_uninit; |
b17a7c17 SH |
6684 | dev->reg_state = NETREG_REGISTERED; |
6685 | ||
6cb6a27c | 6686 | __netdev_update_features(dev); |
8e9b59b2 | 6687 | |
1da177e4 LT |
6688 | /* |
6689 | * Default initial state at registry is that the | |
6690 | * device is present. | |
6691 | */ | |
6692 | ||
6693 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6694 | ||
8f4cccbb BH |
6695 | linkwatch_init_dev(dev); |
6696 | ||
1da177e4 | 6697 | dev_init_scheduler(dev); |
1da177e4 | 6698 | dev_hold(dev); |
ce286d32 | 6699 | list_netdevice(dev); |
7bf23575 | 6700 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 6701 | |
948b337e JP |
6702 | /* If the device has permanent device address, driver should |
6703 | * set dev_addr and also addr_assign_type should be set to | |
6704 | * NET_ADDR_PERM (default value). | |
6705 | */ | |
6706 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
6707 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
6708 | ||
1da177e4 | 6709 | /* Notify protocols, that a new device appeared. */ |
056925ab | 6710 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 6711 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
6712 | if (ret) { |
6713 | rollback_registered(dev); | |
6714 | dev->reg_state = NETREG_UNREGISTERED; | |
6715 | } | |
d90a909e EB |
6716 | /* |
6717 | * Prevent userspace races by waiting until the network | |
6718 | * device is fully setup before sending notifications. | |
6719 | */ | |
a2835763 PM |
6720 | if (!dev->rtnl_link_ops || |
6721 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 6722 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
6723 | |
6724 | out: | |
6725 | return ret; | |
7ce1b0ed HX |
6726 | |
6727 | err_uninit: | |
d314774c SH |
6728 | if (dev->netdev_ops->ndo_uninit) |
6729 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 6730 | goto out; |
1da177e4 | 6731 | } |
d1b19dff | 6732 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 6733 | |
937f1ba5 BH |
6734 | /** |
6735 | * init_dummy_netdev - init a dummy network device for NAPI | |
6736 | * @dev: device to init | |
6737 | * | |
6738 | * This takes a network device structure and initialize the minimum | |
6739 | * amount of fields so it can be used to schedule NAPI polls without | |
6740 | * registering a full blown interface. This is to be used by drivers | |
6741 | * that need to tie several hardware interfaces to a single NAPI | |
6742 | * poll scheduler due to HW limitations. | |
6743 | */ | |
6744 | int init_dummy_netdev(struct net_device *dev) | |
6745 | { | |
6746 | /* Clear everything. Note we don't initialize spinlocks | |
6747 | * are they aren't supposed to be taken by any of the | |
6748 | * NAPI code and this dummy netdev is supposed to be | |
6749 | * only ever used for NAPI polls | |
6750 | */ | |
6751 | memset(dev, 0, sizeof(struct net_device)); | |
6752 | ||
6753 | /* make sure we BUG if trying to hit standard | |
6754 | * register/unregister code path | |
6755 | */ | |
6756 | dev->reg_state = NETREG_DUMMY; | |
6757 | ||
937f1ba5 BH |
6758 | /* NAPI wants this */ |
6759 | INIT_LIST_HEAD(&dev->napi_list); | |
6760 | ||
6761 | /* a dummy interface is started by default */ | |
6762 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6763 | set_bit(__LINK_STATE_START, &dev->state); | |
6764 | ||
29b4433d ED |
6765 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
6766 | * because users of this 'device' dont need to change | |
6767 | * its refcount. | |
6768 | */ | |
6769 | ||
937f1ba5 BH |
6770 | return 0; |
6771 | } | |
6772 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
6773 | ||
6774 | ||
1da177e4 LT |
6775 | /** |
6776 | * register_netdev - register a network device | |
6777 | * @dev: device to register | |
6778 | * | |
6779 | * Take a completed network device structure and add it to the kernel | |
6780 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6781 | * chain. 0 is returned on success. A negative errno code is returned | |
6782 | * on a failure to set up the device, or if the name is a duplicate. | |
6783 | * | |
38b4da38 | 6784 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
6785 | * and expands the device name if you passed a format string to |
6786 | * alloc_netdev. | |
6787 | */ | |
6788 | int register_netdev(struct net_device *dev) | |
6789 | { | |
6790 | int err; | |
6791 | ||
6792 | rtnl_lock(); | |
1da177e4 | 6793 | err = register_netdevice(dev); |
1da177e4 LT |
6794 | rtnl_unlock(); |
6795 | return err; | |
6796 | } | |
6797 | EXPORT_SYMBOL(register_netdev); | |
6798 | ||
29b4433d ED |
6799 | int netdev_refcnt_read(const struct net_device *dev) |
6800 | { | |
6801 | int i, refcnt = 0; | |
6802 | ||
6803 | for_each_possible_cpu(i) | |
6804 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
6805 | return refcnt; | |
6806 | } | |
6807 | EXPORT_SYMBOL(netdev_refcnt_read); | |
6808 | ||
2c53040f | 6809 | /** |
1da177e4 | 6810 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 6811 | * @dev: target net_device |
1da177e4 LT |
6812 | * |
6813 | * This is called when unregistering network devices. | |
6814 | * | |
6815 | * Any protocol or device that holds a reference should register | |
6816 | * for netdevice notification, and cleanup and put back the | |
6817 | * reference if they receive an UNREGISTER event. | |
6818 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 6819 | * call dev_put. |
1da177e4 LT |
6820 | */ |
6821 | static void netdev_wait_allrefs(struct net_device *dev) | |
6822 | { | |
6823 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 6824 | int refcnt; |
1da177e4 | 6825 | |
e014debe ED |
6826 | linkwatch_forget_dev(dev); |
6827 | ||
1da177e4 | 6828 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
6829 | refcnt = netdev_refcnt_read(dev); |
6830 | ||
6831 | while (refcnt != 0) { | |
1da177e4 | 6832 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 6833 | rtnl_lock(); |
1da177e4 LT |
6834 | |
6835 | /* Rebroadcast unregister notification */ | |
056925ab | 6836 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 6837 | |
748e2d93 | 6838 | __rtnl_unlock(); |
0115e8e3 | 6839 | rcu_barrier(); |
748e2d93 ED |
6840 | rtnl_lock(); |
6841 | ||
0115e8e3 | 6842 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
6843 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
6844 | &dev->state)) { | |
6845 | /* We must not have linkwatch events | |
6846 | * pending on unregister. If this | |
6847 | * happens, we simply run the queue | |
6848 | * unscheduled, resulting in a noop | |
6849 | * for this device. | |
6850 | */ | |
6851 | linkwatch_run_queue(); | |
6852 | } | |
6853 | ||
6756ae4b | 6854 | __rtnl_unlock(); |
1da177e4 LT |
6855 | |
6856 | rebroadcast_time = jiffies; | |
6857 | } | |
6858 | ||
6859 | msleep(250); | |
6860 | ||
29b4433d ED |
6861 | refcnt = netdev_refcnt_read(dev); |
6862 | ||
1da177e4 | 6863 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
6864 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
6865 | dev->name, refcnt); | |
1da177e4 LT |
6866 | warning_time = jiffies; |
6867 | } | |
6868 | } | |
6869 | } | |
6870 | ||
6871 | /* The sequence is: | |
6872 | * | |
6873 | * rtnl_lock(); | |
6874 | * ... | |
6875 | * register_netdevice(x1); | |
6876 | * register_netdevice(x2); | |
6877 | * ... | |
6878 | * unregister_netdevice(y1); | |
6879 | * unregister_netdevice(y2); | |
6880 | * ... | |
6881 | * rtnl_unlock(); | |
6882 | * free_netdev(y1); | |
6883 | * free_netdev(y2); | |
6884 | * | |
58ec3b4d | 6885 | * We are invoked by rtnl_unlock(). |
1da177e4 | 6886 | * This allows us to deal with problems: |
b17a7c17 | 6887 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
6888 | * without deadlocking with linkwatch via keventd. |
6889 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
6890 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
6891 | * |
6892 | * We must not return until all unregister events added during | |
6893 | * the interval the lock was held have been completed. | |
1da177e4 | 6894 | */ |
1da177e4 LT |
6895 | void netdev_run_todo(void) |
6896 | { | |
626ab0e6 | 6897 | struct list_head list; |
1da177e4 | 6898 | |
1da177e4 | 6899 | /* Snapshot list, allow later requests */ |
626ab0e6 | 6900 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
6901 | |
6902 | __rtnl_unlock(); | |
626ab0e6 | 6903 | |
0115e8e3 ED |
6904 | |
6905 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
6906 | if (!list_empty(&list)) |
6907 | rcu_barrier(); | |
6908 | ||
1da177e4 LT |
6909 | while (!list_empty(&list)) { |
6910 | struct net_device *dev | |
e5e26d75 | 6911 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
6912 | list_del(&dev->todo_list); |
6913 | ||
748e2d93 | 6914 | rtnl_lock(); |
0115e8e3 | 6915 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 6916 | __rtnl_unlock(); |
0115e8e3 | 6917 | |
b17a7c17 | 6918 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 6919 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
6920 | dev->name, dev->reg_state); |
6921 | dump_stack(); | |
6922 | continue; | |
6923 | } | |
1da177e4 | 6924 | |
b17a7c17 | 6925 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 6926 | |
b17a7c17 | 6927 | netdev_wait_allrefs(dev); |
1da177e4 | 6928 | |
b17a7c17 | 6929 | /* paranoia */ |
29b4433d | 6930 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
6931 | BUG_ON(!list_empty(&dev->ptype_all)); |
6932 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
6933 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
6934 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 6935 | WARN_ON(dev->dn_ptr); |
1da177e4 | 6936 | |
b17a7c17 SH |
6937 | if (dev->destructor) |
6938 | dev->destructor(dev); | |
9093bbb2 | 6939 | |
50624c93 EB |
6940 | /* Report a network device has been unregistered */ |
6941 | rtnl_lock(); | |
6942 | dev_net(dev)->dev_unreg_count--; | |
6943 | __rtnl_unlock(); | |
6944 | wake_up(&netdev_unregistering_wq); | |
6945 | ||
9093bbb2 SH |
6946 | /* Free network device */ |
6947 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 6948 | } |
1da177e4 LT |
6949 | } |
6950 | ||
3cfde79c BH |
6951 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
6952 | * fields in the same order, with only the type differing. | |
6953 | */ | |
77a1abf5 ED |
6954 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
6955 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
6956 | { |
6957 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
6958 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
6959 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
6960 | #else |
6961 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
6962 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
6963 | u64 *dst = (u64 *)stats64; | |
6964 | ||
6965 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
6966 | sizeof(*stats64) / sizeof(u64)); | |
6967 | for (i = 0; i < n; i++) | |
6968 | dst[i] = src[i]; | |
6969 | #endif | |
6970 | } | |
77a1abf5 | 6971 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 6972 | |
eeda3fd6 SH |
6973 | /** |
6974 | * dev_get_stats - get network device statistics | |
6975 | * @dev: device to get statistics from | |
28172739 | 6976 | * @storage: place to store stats |
eeda3fd6 | 6977 | * |
d7753516 BH |
6978 | * Get network statistics from device. Return @storage. |
6979 | * The device driver may provide its own method by setting | |
6980 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
6981 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 6982 | */ |
d7753516 BH |
6983 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
6984 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 6985 | { |
eeda3fd6 SH |
6986 | const struct net_device_ops *ops = dev->netdev_ops; |
6987 | ||
28172739 ED |
6988 | if (ops->ndo_get_stats64) { |
6989 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
6990 | ops->ndo_get_stats64(dev, storage); |
6991 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 6992 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
6993 | } else { |
6994 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 6995 | } |
caf586e5 | 6996 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
015f0688 | 6997 | storage->tx_dropped += atomic_long_read(&dev->tx_dropped); |
28172739 | 6998 | return storage; |
c45d286e | 6999 | } |
eeda3fd6 | 7000 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 7001 | |
24824a09 | 7002 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 7003 | { |
24824a09 | 7004 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 7005 | |
24824a09 ED |
7006 | #ifdef CONFIG_NET_CLS_ACT |
7007 | if (queue) | |
7008 | return queue; | |
7009 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
7010 | if (!queue) | |
7011 | return NULL; | |
7012 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 7013 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
7014 | queue->qdisc_sleeping = &noop_qdisc; |
7015 | rcu_assign_pointer(dev->ingress_queue, queue); | |
7016 | #endif | |
7017 | return queue; | |
bb949fbd DM |
7018 | } |
7019 | ||
2c60db03 ED |
7020 | static const struct ethtool_ops default_ethtool_ops; |
7021 | ||
d07d7507 SG |
7022 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
7023 | const struct ethtool_ops *ops) | |
7024 | { | |
7025 | if (dev->ethtool_ops == &default_ethtool_ops) | |
7026 | dev->ethtool_ops = ops; | |
7027 | } | |
7028 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
7029 | ||
74d332c1 ED |
7030 | void netdev_freemem(struct net_device *dev) |
7031 | { | |
7032 | char *addr = (char *)dev - dev->padded; | |
7033 | ||
4cb28970 | 7034 | kvfree(addr); |
74d332c1 ED |
7035 | } |
7036 | ||
1da177e4 | 7037 | /** |
36909ea4 | 7038 | * alloc_netdev_mqs - allocate network device |
c835a677 TG |
7039 | * @sizeof_priv: size of private data to allocate space for |
7040 | * @name: device name format string | |
7041 | * @name_assign_type: origin of device name | |
7042 | * @setup: callback to initialize device | |
7043 | * @txqs: the number of TX subqueues to allocate | |
7044 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
7045 | * |
7046 | * Allocates a struct net_device with private data area for driver use | |
90e51adf | 7047 | * and performs basic initialization. Also allocates subqueue structs |
36909ea4 | 7048 | * for each queue on the device. |
1da177e4 | 7049 | */ |
36909ea4 | 7050 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 7051 | unsigned char name_assign_type, |
36909ea4 TH |
7052 | void (*setup)(struct net_device *), |
7053 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 7054 | { |
1da177e4 | 7055 | struct net_device *dev; |
7943986c | 7056 | size_t alloc_size; |
1ce8e7b5 | 7057 | struct net_device *p; |
1da177e4 | 7058 | |
b6fe17d6 SH |
7059 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
7060 | ||
36909ea4 | 7061 | if (txqs < 1) { |
7b6cd1ce | 7062 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
7063 | return NULL; |
7064 | } | |
7065 | ||
a953be53 | 7066 | #ifdef CONFIG_SYSFS |
36909ea4 | 7067 | if (rxqs < 1) { |
7b6cd1ce | 7068 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
7069 | return NULL; |
7070 | } | |
7071 | #endif | |
7072 | ||
fd2ea0a7 | 7073 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
7074 | if (sizeof_priv) { |
7075 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 7076 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
7077 | alloc_size += sizeof_priv; |
7078 | } | |
7079 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 7080 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 7081 | |
74d332c1 ED |
7082 | p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
7083 | if (!p) | |
7084 | p = vzalloc(alloc_size); | |
62b5942a | 7085 | if (!p) |
1da177e4 | 7086 | return NULL; |
1da177e4 | 7087 | |
1ce8e7b5 | 7088 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 7089 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 7090 | |
29b4433d ED |
7091 | dev->pcpu_refcnt = alloc_percpu(int); |
7092 | if (!dev->pcpu_refcnt) | |
74d332c1 | 7093 | goto free_dev; |
ab9c73cc | 7094 | |
ab9c73cc | 7095 | if (dev_addr_init(dev)) |
29b4433d | 7096 | goto free_pcpu; |
ab9c73cc | 7097 | |
22bedad3 | 7098 | dev_mc_init(dev); |
a748ee24 | 7099 | dev_uc_init(dev); |
ccffad25 | 7100 | |
c346dca1 | 7101 | dev_net_set(dev, &init_net); |
1da177e4 | 7102 | |
8d3bdbd5 | 7103 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 7104 | dev->gso_max_segs = GSO_MAX_SEGS; |
fcbeb976 | 7105 | dev->gso_min_segs = 0; |
8d3bdbd5 | 7106 | |
8d3bdbd5 DM |
7107 | INIT_LIST_HEAD(&dev->napi_list); |
7108 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 7109 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 7110 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
7111 | INIT_LIST_HEAD(&dev->adj_list.upper); |
7112 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7113 | INIT_LIST_HEAD(&dev->all_adj_list.upper); | |
7114 | INIT_LIST_HEAD(&dev->all_adj_list.lower); | |
7866a621 SN |
7115 | INIT_LIST_HEAD(&dev->ptype_all); |
7116 | INIT_LIST_HEAD(&dev->ptype_specific); | |
02875878 | 7117 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
7118 | setup(dev); |
7119 | ||
906470c1 | 7120 | if (!dev->tx_queue_len) |
f84bb1ea | 7121 | dev->priv_flags |= IFF_NO_QUEUE; |
906470c1 | 7122 | |
36909ea4 TH |
7123 | dev->num_tx_queues = txqs; |
7124 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 7125 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 7126 | goto free_all; |
e8a0464c | 7127 | |
a953be53 | 7128 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
7129 | dev->num_rx_queues = rxqs; |
7130 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 7131 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 7132 | goto free_all; |
df334545 | 7133 | #endif |
0a9627f2 | 7134 | |
1da177e4 | 7135 | strcpy(dev->name, name); |
c835a677 | 7136 | dev->name_assign_type = name_assign_type; |
cbda10fa | 7137 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
7138 | if (!dev->ethtool_ops) |
7139 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 PN |
7140 | |
7141 | nf_hook_ingress_init(dev); | |
7142 | ||
1da177e4 | 7143 | return dev; |
ab9c73cc | 7144 | |
8d3bdbd5 DM |
7145 | free_all: |
7146 | free_netdev(dev); | |
7147 | return NULL; | |
7148 | ||
29b4433d ED |
7149 | free_pcpu: |
7150 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
7151 | free_dev: |
7152 | netdev_freemem(dev); | |
ab9c73cc | 7153 | return NULL; |
1da177e4 | 7154 | } |
36909ea4 | 7155 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
7156 | |
7157 | /** | |
7158 | * free_netdev - free network device | |
7159 | * @dev: device | |
7160 | * | |
4ec93edb YH |
7161 | * This function does the last stage of destroying an allocated device |
7162 | * interface. The reference to the device object is released. | |
1da177e4 LT |
7163 | * If this is the last reference then it will be freed. |
7164 | */ | |
7165 | void free_netdev(struct net_device *dev) | |
7166 | { | |
d565b0a1 HX |
7167 | struct napi_struct *p, *n; |
7168 | ||
60877a32 | 7169 | netif_free_tx_queues(dev); |
a953be53 | 7170 | #ifdef CONFIG_SYSFS |
10595902 | 7171 | kvfree(dev->_rx); |
fe822240 | 7172 | #endif |
e8a0464c | 7173 | |
33d480ce | 7174 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 7175 | |
f001fde5 JP |
7176 | /* Flush device addresses */ |
7177 | dev_addr_flush(dev); | |
7178 | ||
d565b0a1 HX |
7179 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
7180 | netif_napi_del(p); | |
7181 | ||
29b4433d ED |
7182 | free_percpu(dev->pcpu_refcnt); |
7183 | dev->pcpu_refcnt = NULL; | |
7184 | ||
3041a069 | 7185 | /* Compatibility with error handling in drivers */ |
1da177e4 | 7186 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 7187 | netdev_freemem(dev); |
1da177e4 LT |
7188 | return; |
7189 | } | |
7190 | ||
7191 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
7192 | dev->reg_state = NETREG_RELEASED; | |
7193 | ||
43cb76d9 GKH |
7194 | /* will free via device release */ |
7195 | put_device(&dev->dev); | |
1da177e4 | 7196 | } |
d1b19dff | 7197 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 7198 | |
f0db275a SH |
7199 | /** |
7200 | * synchronize_net - Synchronize with packet receive processing | |
7201 | * | |
7202 | * Wait for packets currently being received to be done. | |
7203 | * Does not block later packets from starting. | |
7204 | */ | |
4ec93edb | 7205 | void synchronize_net(void) |
1da177e4 LT |
7206 | { |
7207 | might_sleep(); | |
be3fc413 ED |
7208 | if (rtnl_is_locked()) |
7209 | synchronize_rcu_expedited(); | |
7210 | else | |
7211 | synchronize_rcu(); | |
1da177e4 | 7212 | } |
d1b19dff | 7213 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
7214 | |
7215 | /** | |
44a0873d | 7216 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 7217 | * @dev: device |
44a0873d | 7218 | * @head: list |
6ebfbc06 | 7219 | * |
1da177e4 | 7220 | * This function shuts down a device interface and removes it |
d59b54b1 | 7221 | * from the kernel tables. |
44a0873d | 7222 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
7223 | * |
7224 | * Callers must hold the rtnl semaphore. You may want | |
7225 | * unregister_netdev() instead of this. | |
7226 | */ | |
7227 | ||
44a0873d | 7228 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 7229 | { |
a6620712 HX |
7230 | ASSERT_RTNL(); |
7231 | ||
44a0873d | 7232 | if (head) { |
9fdce099 | 7233 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
7234 | } else { |
7235 | rollback_registered(dev); | |
7236 | /* Finish processing unregister after unlock */ | |
7237 | net_set_todo(dev); | |
7238 | } | |
1da177e4 | 7239 | } |
44a0873d | 7240 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 7241 | |
9b5e383c ED |
7242 | /** |
7243 | * unregister_netdevice_many - unregister many devices | |
7244 | * @head: list of devices | |
87757a91 ED |
7245 | * |
7246 | * Note: As most callers use a stack allocated list_head, | |
7247 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
7248 | */ |
7249 | void unregister_netdevice_many(struct list_head *head) | |
7250 | { | |
7251 | struct net_device *dev; | |
7252 | ||
7253 | if (!list_empty(head)) { | |
7254 | rollback_registered_many(head); | |
7255 | list_for_each_entry(dev, head, unreg_list) | |
7256 | net_set_todo(dev); | |
87757a91 | 7257 | list_del(head); |
9b5e383c ED |
7258 | } |
7259 | } | |
63c8099d | 7260 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 7261 | |
1da177e4 LT |
7262 | /** |
7263 | * unregister_netdev - remove device from the kernel | |
7264 | * @dev: device | |
7265 | * | |
7266 | * This function shuts down a device interface and removes it | |
d59b54b1 | 7267 | * from the kernel tables. |
1da177e4 LT |
7268 | * |
7269 | * This is just a wrapper for unregister_netdevice that takes | |
7270 | * the rtnl semaphore. In general you want to use this and not | |
7271 | * unregister_netdevice. | |
7272 | */ | |
7273 | void unregister_netdev(struct net_device *dev) | |
7274 | { | |
7275 | rtnl_lock(); | |
7276 | unregister_netdevice(dev); | |
7277 | rtnl_unlock(); | |
7278 | } | |
1da177e4 LT |
7279 | EXPORT_SYMBOL(unregister_netdev); |
7280 | ||
ce286d32 EB |
7281 | /** |
7282 | * dev_change_net_namespace - move device to different nethost namespace | |
7283 | * @dev: device | |
7284 | * @net: network namespace | |
7285 | * @pat: If not NULL name pattern to try if the current device name | |
7286 | * is already taken in the destination network namespace. | |
7287 | * | |
7288 | * This function shuts down a device interface and moves it | |
7289 | * to a new network namespace. On success 0 is returned, on | |
7290 | * a failure a netagive errno code is returned. | |
7291 | * | |
7292 | * Callers must hold the rtnl semaphore. | |
7293 | */ | |
7294 | ||
7295 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
7296 | { | |
ce286d32 EB |
7297 | int err; |
7298 | ||
7299 | ASSERT_RTNL(); | |
7300 | ||
7301 | /* Don't allow namespace local devices to be moved. */ | |
7302 | err = -EINVAL; | |
7303 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7304 | goto out; | |
7305 | ||
7306 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
7307 | if (dev->reg_state != NETREG_REGISTERED) |
7308 | goto out; | |
7309 | ||
7310 | /* Get out if there is nothing todo */ | |
7311 | err = 0; | |
878628fb | 7312 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
7313 | goto out; |
7314 | ||
7315 | /* Pick the destination device name, and ensure | |
7316 | * we can use it in the destination network namespace. | |
7317 | */ | |
7318 | err = -EEXIST; | |
d9031024 | 7319 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
7320 | /* We get here if we can't use the current device name */ |
7321 | if (!pat) | |
7322 | goto out; | |
828de4f6 | 7323 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
7324 | goto out; |
7325 | } | |
7326 | ||
7327 | /* | |
7328 | * And now a mini version of register_netdevice unregister_netdevice. | |
7329 | */ | |
7330 | ||
7331 | /* If device is running close it first. */ | |
9b772652 | 7332 | dev_close(dev); |
ce286d32 EB |
7333 | |
7334 | /* And unlink it from device chain */ | |
7335 | err = -ENODEV; | |
7336 | unlist_netdevice(dev); | |
7337 | ||
7338 | synchronize_net(); | |
7339 | ||
7340 | /* Shutdown queueing discipline. */ | |
7341 | dev_shutdown(dev); | |
7342 | ||
7343 | /* Notify protocols, that we are about to destroy | |
7344 | this device. They should clean all the things. | |
3b27e105 DL |
7345 | |
7346 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
7347 | This is wanted because this way 8021q and macvlan know | |
7348 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
7349 | */ |
7350 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
7351 | rcu_barrier(); |
7352 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 7353 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
7354 | |
7355 | /* | |
7356 | * Flush the unicast and multicast chains | |
7357 | */ | |
a748ee24 | 7358 | dev_uc_flush(dev); |
22bedad3 | 7359 | dev_mc_flush(dev); |
ce286d32 | 7360 | |
4e66ae2e SH |
7361 | /* Send a netdev-removed uevent to the old namespace */ |
7362 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 7363 | netdev_adjacent_del_links(dev); |
4e66ae2e | 7364 | |
ce286d32 | 7365 | /* Actually switch the network namespace */ |
c346dca1 | 7366 | dev_net_set(dev, net); |
ce286d32 | 7367 | |
ce286d32 | 7368 | /* If there is an ifindex conflict assign a new one */ |
7a66bbc9 | 7369 | if (__dev_get_by_index(net, dev->ifindex)) |
ce286d32 | 7370 | dev->ifindex = dev_new_index(net); |
ce286d32 | 7371 | |
4e66ae2e SH |
7372 | /* Send a netdev-add uevent to the new namespace */ |
7373 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 7374 | netdev_adjacent_add_links(dev); |
4e66ae2e | 7375 | |
8b41d188 | 7376 | /* Fixup kobjects */ |
a1b3f594 | 7377 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 7378 | WARN_ON(err); |
ce286d32 EB |
7379 | |
7380 | /* Add the device back in the hashes */ | |
7381 | list_netdevice(dev); | |
7382 | ||
7383 | /* Notify protocols, that a new device appeared. */ | |
7384 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
7385 | ||
d90a909e EB |
7386 | /* |
7387 | * Prevent userspace races by waiting until the network | |
7388 | * device is fully setup before sending notifications. | |
7389 | */ | |
7f294054 | 7390 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 7391 | |
ce286d32 EB |
7392 | synchronize_net(); |
7393 | err = 0; | |
7394 | out: | |
7395 | return err; | |
7396 | } | |
463d0183 | 7397 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 7398 | |
1da177e4 LT |
7399 | static int dev_cpu_callback(struct notifier_block *nfb, |
7400 | unsigned long action, | |
7401 | void *ocpu) | |
7402 | { | |
7403 | struct sk_buff **list_skb; | |
1da177e4 LT |
7404 | struct sk_buff *skb; |
7405 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
7406 | struct softnet_data *sd, *oldsd; | |
7407 | ||
8bb78442 | 7408 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
7409 | return NOTIFY_OK; |
7410 | ||
7411 | local_irq_disable(); | |
7412 | cpu = smp_processor_id(); | |
7413 | sd = &per_cpu(softnet_data, cpu); | |
7414 | oldsd = &per_cpu(softnet_data, oldcpu); | |
7415 | ||
7416 | /* Find end of our completion_queue. */ | |
7417 | list_skb = &sd->completion_queue; | |
7418 | while (*list_skb) | |
7419 | list_skb = &(*list_skb)->next; | |
7420 | /* Append completion queue from offline CPU. */ | |
7421 | *list_skb = oldsd->completion_queue; | |
7422 | oldsd->completion_queue = NULL; | |
7423 | ||
1da177e4 | 7424 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
7425 | if (oldsd->output_queue) { |
7426 | *sd->output_queue_tailp = oldsd->output_queue; | |
7427 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
7428 | oldsd->output_queue = NULL; | |
7429 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
7430 | } | |
ac64da0b ED |
7431 | /* Append NAPI poll list from offline CPU, with one exception : |
7432 | * process_backlog() must be called by cpu owning percpu backlog. | |
7433 | * We properly handle process_queue & input_pkt_queue later. | |
7434 | */ | |
7435 | while (!list_empty(&oldsd->poll_list)) { | |
7436 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
7437 | struct napi_struct, | |
7438 | poll_list); | |
7439 | ||
7440 | list_del_init(&napi->poll_list); | |
7441 | if (napi->poll == process_backlog) | |
7442 | napi->state = 0; | |
7443 | else | |
7444 | ____napi_schedule(sd, napi); | |
264524d5 | 7445 | } |
1da177e4 LT |
7446 | |
7447 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
7448 | local_irq_enable(); | |
7449 | ||
7450 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 7451 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 7452 | netif_rx_ni(skb); |
76cc8b13 | 7453 | input_queue_head_incr(oldsd); |
fec5e652 | 7454 | } |
ac64da0b | 7455 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 7456 | netif_rx_ni(skb); |
76cc8b13 TH |
7457 | input_queue_head_incr(oldsd); |
7458 | } | |
1da177e4 LT |
7459 | |
7460 | return NOTIFY_OK; | |
7461 | } | |
1da177e4 LT |
7462 | |
7463 | ||
7f353bf2 | 7464 | /** |
b63365a2 HX |
7465 | * netdev_increment_features - increment feature set by one |
7466 | * @all: current feature set | |
7467 | * @one: new feature set | |
7468 | * @mask: mask feature set | |
7f353bf2 HX |
7469 | * |
7470 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
7471 | * @one to the master device with current feature set @all. Will not |
7472 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 7473 | */ |
c8f44aff MM |
7474 | netdev_features_t netdev_increment_features(netdev_features_t all, |
7475 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 7476 | { |
1742f183 MM |
7477 | if (mask & NETIF_F_GEN_CSUM) |
7478 | mask |= NETIF_F_ALL_CSUM; | |
7479 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 7480 | |
1742f183 MM |
7481 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
7482 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 7483 | |
1742f183 MM |
7484 | /* If one device supports hw checksumming, set for all. */ |
7485 | if (all & NETIF_F_GEN_CSUM) | |
7486 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
7487 | |
7488 | return all; | |
7489 | } | |
b63365a2 | 7490 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 7491 | |
430f03cd | 7492 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
7493 | { |
7494 | int i; | |
7495 | struct hlist_head *hash; | |
7496 | ||
7497 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
7498 | if (hash != NULL) | |
7499 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
7500 | INIT_HLIST_HEAD(&hash[i]); | |
7501 | ||
7502 | return hash; | |
7503 | } | |
7504 | ||
881d966b | 7505 | /* Initialize per network namespace state */ |
4665079c | 7506 | static int __net_init netdev_init(struct net *net) |
881d966b | 7507 | { |
734b6541 RM |
7508 | if (net != &init_net) |
7509 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 7510 | |
30d97d35 PE |
7511 | net->dev_name_head = netdev_create_hash(); |
7512 | if (net->dev_name_head == NULL) | |
7513 | goto err_name; | |
881d966b | 7514 | |
30d97d35 PE |
7515 | net->dev_index_head = netdev_create_hash(); |
7516 | if (net->dev_index_head == NULL) | |
7517 | goto err_idx; | |
881d966b EB |
7518 | |
7519 | return 0; | |
30d97d35 PE |
7520 | |
7521 | err_idx: | |
7522 | kfree(net->dev_name_head); | |
7523 | err_name: | |
7524 | return -ENOMEM; | |
881d966b EB |
7525 | } |
7526 | ||
f0db275a SH |
7527 | /** |
7528 | * netdev_drivername - network driver for the device | |
7529 | * @dev: network device | |
f0db275a SH |
7530 | * |
7531 | * Determine network driver for device. | |
7532 | */ | |
3019de12 | 7533 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 7534 | { |
cf04a4c7 SH |
7535 | const struct device_driver *driver; |
7536 | const struct device *parent; | |
3019de12 | 7537 | const char *empty = ""; |
6579e57b AV |
7538 | |
7539 | parent = dev->dev.parent; | |
6579e57b | 7540 | if (!parent) |
3019de12 | 7541 | return empty; |
6579e57b AV |
7542 | |
7543 | driver = parent->driver; | |
7544 | if (driver && driver->name) | |
3019de12 DM |
7545 | return driver->name; |
7546 | return empty; | |
6579e57b AV |
7547 | } |
7548 | ||
6ea754eb JP |
7549 | static void __netdev_printk(const char *level, const struct net_device *dev, |
7550 | struct va_format *vaf) | |
256df2f3 | 7551 | { |
b004ff49 | 7552 | if (dev && dev->dev.parent) { |
6ea754eb JP |
7553 | dev_printk_emit(level[1] - '0', |
7554 | dev->dev.parent, | |
7555 | "%s %s %s%s: %pV", | |
7556 | dev_driver_string(dev->dev.parent), | |
7557 | dev_name(dev->dev.parent), | |
7558 | netdev_name(dev), netdev_reg_state(dev), | |
7559 | vaf); | |
b004ff49 | 7560 | } else if (dev) { |
6ea754eb JP |
7561 | printk("%s%s%s: %pV", |
7562 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 7563 | } else { |
6ea754eb | 7564 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 7565 | } |
256df2f3 JP |
7566 | } |
7567 | ||
6ea754eb JP |
7568 | void netdev_printk(const char *level, const struct net_device *dev, |
7569 | const char *format, ...) | |
256df2f3 JP |
7570 | { |
7571 | struct va_format vaf; | |
7572 | va_list args; | |
256df2f3 JP |
7573 | |
7574 | va_start(args, format); | |
7575 | ||
7576 | vaf.fmt = format; | |
7577 | vaf.va = &args; | |
7578 | ||
6ea754eb | 7579 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 7580 | |
256df2f3 | 7581 | va_end(args); |
256df2f3 JP |
7582 | } |
7583 | EXPORT_SYMBOL(netdev_printk); | |
7584 | ||
7585 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 7586 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 7587 | { \ |
256df2f3 JP |
7588 | struct va_format vaf; \ |
7589 | va_list args; \ | |
7590 | \ | |
7591 | va_start(args, fmt); \ | |
7592 | \ | |
7593 | vaf.fmt = fmt; \ | |
7594 | vaf.va = &args; \ | |
7595 | \ | |
6ea754eb | 7596 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 7597 | \ |
256df2f3 | 7598 | va_end(args); \ |
256df2f3 JP |
7599 | } \ |
7600 | EXPORT_SYMBOL(func); | |
7601 | ||
7602 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
7603 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
7604 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
7605 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
7606 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
7607 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
7608 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
7609 | ||
4665079c | 7610 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
7611 | { |
7612 | kfree(net->dev_name_head); | |
7613 | kfree(net->dev_index_head); | |
7614 | } | |
7615 | ||
022cbae6 | 7616 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
7617 | .init = netdev_init, |
7618 | .exit = netdev_exit, | |
7619 | }; | |
7620 | ||
4665079c | 7621 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 7622 | { |
e008b5fc | 7623 | struct net_device *dev, *aux; |
ce286d32 | 7624 | /* |
e008b5fc | 7625 | * Push all migratable network devices back to the |
ce286d32 EB |
7626 | * initial network namespace |
7627 | */ | |
7628 | rtnl_lock(); | |
e008b5fc | 7629 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 7630 | int err; |
aca51397 | 7631 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
7632 | |
7633 | /* Ignore unmoveable devices (i.e. loopback) */ | |
7634 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7635 | continue; | |
7636 | ||
e008b5fc EB |
7637 | /* Leave virtual devices for the generic cleanup */ |
7638 | if (dev->rtnl_link_ops) | |
7639 | continue; | |
d0c082ce | 7640 | |
25985edc | 7641 | /* Push remaining network devices to init_net */ |
aca51397 PE |
7642 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
7643 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 7644 | if (err) { |
7b6cd1ce JP |
7645 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
7646 | __func__, dev->name, err); | |
aca51397 | 7647 | BUG(); |
ce286d32 EB |
7648 | } |
7649 | } | |
7650 | rtnl_unlock(); | |
7651 | } | |
7652 | ||
50624c93 EB |
7653 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
7654 | { | |
7655 | /* Return with the rtnl_lock held when there are no network | |
7656 | * devices unregistering in any network namespace in net_list. | |
7657 | */ | |
7658 | struct net *net; | |
7659 | bool unregistering; | |
ff960a73 | 7660 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 7661 | |
ff960a73 | 7662 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 7663 | for (;;) { |
50624c93 EB |
7664 | unregistering = false; |
7665 | rtnl_lock(); | |
7666 | list_for_each_entry(net, net_list, exit_list) { | |
7667 | if (net->dev_unreg_count > 0) { | |
7668 | unregistering = true; | |
7669 | break; | |
7670 | } | |
7671 | } | |
7672 | if (!unregistering) | |
7673 | break; | |
7674 | __rtnl_unlock(); | |
ff960a73 PZ |
7675 | |
7676 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 7677 | } |
ff960a73 | 7678 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
7679 | } |
7680 | ||
04dc7f6b EB |
7681 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
7682 | { | |
7683 | /* At exit all network devices most be removed from a network | |
b595076a | 7684 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
7685 | * Do this across as many network namespaces as possible to |
7686 | * improve batching efficiency. | |
7687 | */ | |
7688 | struct net_device *dev; | |
7689 | struct net *net; | |
7690 | LIST_HEAD(dev_kill_list); | |
7691 | ||
50624c93 EB |
7692 | /* To prevent network device cleanup code from dereferencing |
7693 | * loopback devices or network devices that have been freed | |
7694 | * wait here for all pending unregistrations to complete, | |
7695 | * before unregistring the loopback device and allowing the | |
7696 | * network namespace be freed. | |
7697 | * | |
7698 | * The netdev todo list containing all network devices | |
7699 | * unregistrations that happen in default_device_exit_batch | |
7700 | * will run in the rtnl_unlock() at the end of | |
7701 | * default_device_exit_batch. | |
7702 | */ | |
7703 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
7704 | list_for_each_entry(net, net_list, exit_list) { |
7705 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 7706 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
7707 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
7708 | else | |
7709 | unregister_netdevice_queue(dev, &dev_kill_list); | |
7710 | } | |
7711 | } | |
7712 | unregister_netdevice_many(&dev_kill_list); | |
7713 | rtnl_unlock(); | |
7714 | } | |
7715 | ||
022cbae6 | 7716 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 7717 | .exit = default_device_exit, |
04dc7f6b | 7718 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
7719 | }; |
7720 | ||
1da177e4 LT |
7721 | /* |
7722 | * Initialize the DEV module. At boot time this walks the device list and | |
7723 | * unhooks any devices that fail to initialise (normally hardware not | |
7724 | * present) and leaves us with a valid list of present and active devices. | |
7725 | * | |
7726 | */ | |
7727 | ||
7728 | /* | |
7729 | * This is called single threaded during boot, so no need | |
7730 | * to take the rtnl semaphore. | |
7731 | */ | |
7732 | static int __init net_dev_init(void) | |
7733 | { | |
7734 | int i, rc = -ENOMEM; | |
7735 | ||
7736 | BUG_ON(!dev_boot_phase); | |
7737 | ||
1da177e4 LT |
7738 | if (dev_proc_init()) |
7739 | goto out; | |
7740 | ||
8b41d188 | 7741 | if (netdev_kobject_init()) |
1da177e4 LT |
7742 | goto out; |
7743 | ||
7744 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 7745 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
7746 | INIT_LIST_HEAD(&ptype_base[i]); |
7747 | ||
62532da9 VY |
7748 | INIT_LIST_HEAD(&offload_base); |
7749 | ||
881d966b EB |
7750 | if (register_pernet_subsys(&netdev_net_ops)) |
7751 | goto out; | |
1da177e4 LT |
7752 | |
7753 | /* | |
7754 | * Initialise the packet receive queues. | |
7755 | */ | |
7756 | ||
6f912042 | 7757 | for_each_possible_cpu(i) { |
e36fa2f7 | 7758 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 7759 | |
e36fa2f7 | 7760 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 7761 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 7762 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 7763 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 7764 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
7765 | sd->csd.func = rps_trigger_softirq; |
7766 | sd->csd.info = sd; | |
e36fa2f7 | 7767 | sd->cpu = i; |
1e94d72f | 7768 | #endif |
0a9627f2 | 7769 | |
e36fa2f7 ED |
7770 | sd->backlog.poll = process_backlog; |
7771 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
7772 | } |
7773 | ||
1da177e4 LT |
7774 | dev_boot_phase = 0; |
7775 | ||
505d4f73 EB |
7776 | /* The loopback device is special if any other network devices |
7777 | * is present in a network namespace the loopback device must | |
7778 | * be present. Since we now dynamically allocate and free the | |
7779 | * loopback device ensure this invariant is maintained by | |
7780 | * keeping the loopback device as the first device on the | |
7781 | * list of network devices. Ensuring the loopback devices | |
7782 | * is the first device that appears and the last network device | |
7783 | * that disappears. | |
7784 | */ | |
7785 | if (register_pernet_device(&loopback_net_ops)) | |
7786 | goto out; | |
7787 | ||
7788 | if (register_pernet_device(&default_device_ops)) | |
7789 | goto out; | |
7790 | ||
962cf36c CM |
7791 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
7792 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
7793 | |
7794 | hotcpu_notifier(dev_cpu_callback, 0); | |
f38a9eb1 | 7795 | dst_subsys_init(); |
1da177e4 LT |
7796 | rc = 0; |
7797 | out: | |
7798 | return rc; | |
7799 | } | |
7800 | ||
7801 | subsys_initcall(net_dev_init); |