]>
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 LT |
101 | #include <net/dst.h> |
102 | #include <net/pkt_sched.h> | |
103 | #include <net/checksum.h> | |
44540960 | 104 | #include <net/xfrm.h> |
1da177e4 LT |
105 | #include <linux/highmem.h> |
106 | #include <linux/init.h> | |
1da177e4 | 107 | #include <linux/module.h> |
1da177e4 LT |
108 | #include <linux/netpoll.h> |
109 | #include <linux/rcupdate.h> | |
110 | #include <linux/delay.h> | |
1da177e4 | 111 | #include <net/iw_handler.h> |
1da177e4 | 112 | #include <asm/current.h> |
5bdb9886 | 113 | #include <linux/audit.h> |
db217334 | 114 | #include <linux/dmaengine.h> |
f6a78bfc | 115 | #include <linux/err.h> |
c7fa9d18 | 116 | #include <linux/ctype.h> |
723e98b7 | 117 | #include <linux/if_arp.h> |
6de329e2 | 118 | #include <linux/if_vlan.h> |
8f0f2223 | 119 | #include <linux/ip.h> |
ad55dcaf | 120 | #include <net/ip.h> |
8f0f2223 DM |
121 | #include <linux/ipv6.h> |
122 | #include <linux/in.h> | |
b6b2fed1 DM |
123 | #include <linux/jhash.h> |
124 | #include <linux/random.h> | |
9cbc1cb8 | 125 | #include <trace/events/napi.h> |
cf66ba58 | 126 | #include <trace/events/net.h> |
07dc22e7 | 127 | #include <trace/events/skb.h> |
5acbbd42 | 128 | #include <linux/pci.h> |
caeda9b9 | 129 | #include <linux/inetdevice.h> |
c445477d | 130 | #include <linux/cpu_rmap.h> |
c5905afb | 131 | #include <linux/static_key.h> |
af12fa6e | 132 | #include <linux/hashtable.h> |
60877a32 | 133 | #include <linux/vmalloc.h> |
529d0489 | 134 | #include <linux/if_macvlan.h> |
e7fd2885 | 135 | #include <linux/errqueue.h> |
1da177e4 | 136 | |
342709ef PE |
137 | #include "net-sysfs.h" |
138 | ||
d565b0a1 HX |
139 | /* Instead of increasing this, you should create a hash table. */ |
140 | #define MAX_GRO_SKBS 8 | |
141 | ||
5d38a079 HX |
142 | /* This should be increased if a protocol with a bigger head is added. */ |
143 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
144 | ||
1da177e4 | 145 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 146 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
147 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
148 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 149 | static struct list_head offload_base __read_mostly; |
1da177e4 | 150 | |
ae78dbfa | 151 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
152 | static int call_netdevice_notifiers_info(unsigned long val, |
153 | struct net_device *dev, | |
154 | struct netdev_notifier_info *info); | |
ae78dbfa | 155 | |
1da177e4 | 156 | /* |
7562f876 | 157 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
158 | * semaphore. |
159 | * | |
c6d14c84 | 160 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
161 | * |
162 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 163 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
164 | * actual updates. This allows pure readers to access the list even |
165 | * while a writer is preparing to update it. | |
166 | * | |
167 | * To put it another way, dev_base_lock is held for writing only to | |
168 | * protect against pure readers; the rtnl semaphore provides the | |
169 | * protection against other writers. | |
170 | * | |
171 | * See, for example usages, register_netdevice() and | |
172 | * unregister_netdevice(), which must be called with the rtnl | |
173 | * semaphore held. | |
174 | */ | |
1da177e4 | 175 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
176 | EXPORT_SYMBOL(dev_base_lock); |
177 | ||
af12fa6e ET |
178 | /* protects napi_hash addition/deletion and napi_gen_id */ |
179 | static DEFINE_SPINLOCK(napi_hash_lock); | |
180 | ||
181 | static unsigned int napi_gen_id; | |
182 | static DEFINE_HASHTABLE(napi_hash, 8); | |
183 | ||
18afa4b0 | 184 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 185 | |
4e985ada TG |
186 | static inline void dev_base_seq_inc(struct net *net) |
187 | { | |
188 | while (++net->dev_base_seq == 0); | |
189 | } | |
190 | ||
881d966b | 191 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 192 | { |
95c96174 ED |
193 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
194 | ||
08e9897d | 195 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
196 | } |
197 | ||
881d966b | 198 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 199 | { |
7c28bd0b | 200 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
201 | } |
202 | ||
e36fa2f7 | 203 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
204 | { |
205 | #ifdef CONFIG_RPS | |
e36fa2f7 | 206 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
207 | #endif |
208 | } | |
209 | ||
e36fa2f7 | 210 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
211 | { |
212 | #ifdef CONFIG_RPS | |
e36fa2f7 | 213 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
214 | #endif |
215 | } | |
216 | ||
ce286d32 | 217 | /* Device list insertion */ |
53759be9 | 218 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 219 | { |
c346dca1 | 220 | struct net *net = dev_net(dev); |
ce286d32 EB |
221 | |
222 | ASSERT_RTNL(); | |
223 | ||
224 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 225 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 226 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
227 | hlist_add_head_rcu(&dev->index_hlist, |
228 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 229 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
230 | |
231 | dev_base_seq_inc(net); | |
ce286d32 EB |
232 | } |
233 | ||
fb699dfd ED |
234 | /* Device list removal |
235 | * caller must respect a RCU grace period before freeing/reusing dev | |
236 | */ | |
ce286d32 EB |
237 | static void unlist_netdevice(struct net_device *dev) |
238 | { | |
239 | ASSERT_RTNL(); | |
240 | ||
241 | /* Unlink dev from the device chain */ | |
242 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 243 | list_del_rcu(&dev->dev_list); |
72c9528b | 244 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 245 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 246 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
247 | |
248 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
249 | } |
250 | ||
1da177e4 LT |
251 | /* |
252 | * Our notifier list | |
253 | */ | |
254 | ||
f07d5b94 | 255 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
256 | |
257 | /* | |
258 | * Device drivers call our routines to queue packets here. We empty the | |
259 | * queue in the local softnet handler. | |
260 | */ | |
bea3348e | 261 | |
9958da05 | 262 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 263 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 264 | |
cf508b12 | 265 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 266 | /* |
c773e847 | 267 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
268 | * according to dev->type |
269 | */ | |
270 | static const unsigned short netdev_lock_type[] = | |
271 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
272 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
273 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
274 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
275 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
276 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
277 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
278 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
279 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
280 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
281 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
282 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
283 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
284 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
285 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 286 | |
36cbd3dc | 287 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
288 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
289 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
290 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
291 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
292 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
293 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
294 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
295 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
296 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
297 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
298 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
299 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
300 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
301 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
302 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
303 | |
304 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 305 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
306 | |
307 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
308 | { | |
309 | int i; | |
310 | ||
311 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
312 | if (netdev_lock_type[i] == dev_type) | |
313 | return i; | |
314 | /* the last key is used by default */ | |
315 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
316 | } | |
317 | ||
cf508b12 DM |
318 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
319 | unsigned short dev_type) | |
723e98b7 JP |
320 | { |
321 | int i; | |
322 | ||
323 | i = netdev_lock_pos(dev_type); | |
324 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
325 | netdev_lock_name[i]); | |
326 | } | |
cf508b12 DM |
327 | |
328 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
329 | { | |
330 | int i; | |
331 | ||
332 | i = netdev_lock_pos(dev->type); | |
333 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
334 | &netdev_addr_lock_key[i], | |
335 | netdev_lock_name[i]); | |
336 | } | |
723e98b7 | 337 | #else |
cf508b12 DM |
338 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
339 | unsigned short dev_type) | |
340 | { | |
341 | } | |
342 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
343 | { |
344 | } | |
345 | #endif | |
1da177e4 LT |
346 | |
347 | /******************************************************************************* | |
348 | ||
349 | Protocol management and registration routines | |
350 | ||
351 | *******************************************************************************/ | |
352 | ||
1da177e4 LT |
353 | /* |
354 | * Add a protocol ID to the list. Now that the input handler is | |
355 | * smarter we can dispense with all the messy stuff that used to be | |
356 | * here. | |
357 | * | |
358 | * BEWARE!!! Protocol handlers, mangling input packets, | |
359 | * MUST BE last in hash buckets and checking protocol handlers | |
360 | * MUST start from promiscuous ptype_all chain in net_bh. | |
361 | * It is true now, do not change it. | |
362 | * Explanation follows: if protocol handler, mangling packet, will | |
363 | * be the first on list, it is not able to sense, that packet | |
364 | * is cloned and should be copied-on-write, so that it will | |
365 | * change it and subsequent readers will get broken packet. | |
366 | * --ANK (980803) | |
367 | */ | |
368 | ||
c07b68e8 ED |
369 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
370 | { | |
371 | if (pt->type == htons(ETH_P_ALL)) | |
372 | return &ptype_all; | |
373 | else | |
374 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
375 | } | |
376 | ||
1da177e4 LT |
377 | /** |
378 | * dev_add_pack - add packet handler | |
379 | * @pt: packet type declaration | |
380 | * | |
381 | * Add a protocol handler to the networking stack. The passed &packet_type | |
382 | * is linked into kernel lists and may not be freed until it has been | |
383 | * removed from the kernel lists. | |
384 | * | |
4ec93edb | 385 | * This call does not sleep therefore it can not |
1da177e4 LT |
386 | * guarantee all CPU's that are in middle of receiving packets |
387 | * will see the new packet type (until the next received packet). | |
388 | */ | |
389 | ||
390 | void dev_add_pack(struct packet_type *pt) | |
391 | { | |
c07b68e8 | 392 | struct list_head *head = ptype_head(pt); |
1da177e4 | 393 | |
c07b68e8 ED |
394 | spin_lock(&ptype_lock); |
395 | list_add_rcu(&pt->list, head); | |
396 | spin_unlock(&ptype_lock); | |
1da177e4 | 397 | } |
d1b19dff | 398 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 399 | |
1da177e4 LT |
400 | /** |
401 | * __dev_remove_pack - remove packet handler | |
402 | * @pt: packet type declaration | |
403 | * | |
404 | * Remove a protocol handler that was previously added to the kernel | |
405 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
406 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 407 | * returns. |
1da177e4 LT |
408 | * |
409 | * The packet type might still be in use by receivers | |
410 | * and must not be freed until after all the CPU's have gone | |
411 | * through a quiescent state. | |
412 | */ | |
413 | void __dev_remove_pack(struct packet_type *pt) | |
414 | { | |
c07b68e8 | 415 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
416 | struct packet_type *pt1; |
417 | ||
c07b68e8 | 418 | spin_lock(&ptype_lock); |
1da177e4 LT |
419 | |
420 | list_for_each_entry(pt1, head, list) { | |
421 | if (pt == pt1) { | |
422 | list_del_rcu(&pt->list); | |
423 | goto out; | |
424 | } | |
425 | } | |
426 | ||
7b6cd1ce | 427 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 428 | out: |
c07b68e8 | 429 | spin_unlock(&ptype_lock); |
1da177e4 | 430 | } |
d1b19dff ED |
431 | EXPORT_SYMBOL(__dev_remove_pack); |
432 | ||
1da177e4 LT |
433 | /** |
434 | * dev_remove_pack - remove packet handler | |
435 | * @pt: packet type declaration | |
436 | * | |
437 | * Remove a protocol handler that was previously added to the kernel | |
438 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
439 | * from the kernel lists and can be freed or reused once this function | |
440 | * returns. | |
441 | * | |
442 | * This call sleeps to guarantee that no CPU is looking at the packet | |
443 | * type after return. | |
444 | */ | |
445 | void dev_remove_pack(struct packet_type *pt) | |
446 | { | |
447 | __dev_remove_pack(pt); | |
4ec93edb | 448 | |
1da177e4 LT |
449 | synchronize_net(); |
450 | } | |
d1b19dff | 451 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 452 | |
62532da9 VY |
453 | |
454 | /** | |
455 | * dev_add_offload - register offload handlers | |
456 | * @po: protocol offload declaration | |
457 | * | |
458 | * Add protocol offload handlers to the networking stack. The passed | |
459 | * &proto_offload is linked into kernel lists and may not be freed until | |
460 | * it has been removed from the kernel lists. | |
461 | * | |
462 | * This call does not sleep therefore it can not | |
463 | * guarantee all CPU's that are in middle of receiving packets | |
464 | * will see the new offload handlers (until the next received packet). | |
465 | */ | |
466 | void dev_add_offload(struct packet_offload *po) | |
467 | { | |
468 | struct list_head *head = &offload_base; | |
469 | ||
470 | spin_lock(&offload_lock); | |
471 | list_add_rcu(&po->list, head); | |
472 | spin_unlock(&offload_lock); | |
473 | } | |
474 | EXPORT_SYMBOL(dev_add_offload); | |
475 | ||
476 | /** | |
477 | * __dev_remove_offload - remove offload handler | |
478 | * @po: packet offload declaration | |
479 | * | |
480 | * Remove a protocol offload handler that was previously added to the | |
481 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
482 | * is removed from the kernel lists and can be freed or reused once this | |
483 | * function returns. | |
484 | * | |
485 | * The packet type might still be in use by receivers | |
486 | * and must not be freed until after all the CPU's have gone | |
487 | * through a quiescent state. | |
488 | */ | |
1d143d9f | 489 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
490 | { |
491 | struct list_head *head = &offload_base; | |
492 | struct packet_offload *po1; | |
493 | ||
c53aa505 | 494 | spin_lock(&offload_lock); |
62532da9 VY |
495 | |
496 | list_for_each_entry(po1, head, list) { | |
497 | if (po == po1) { | |
498 | list_del_rcu(&po->list); | |
499 | goto out; | |
500 | } | |
501 | } | |
502 | ||
503 | pr_warn("dev_remove_offload: %p not found\n", po); | |
504 | out: | |
c53aa505 | 505 | spin_unlock(&offload_lock); |
62532da9 | 506 | } |
62532da9 VY |
507 | |
508 | /** | |
509 | * dev_remove_offload - remove packet offload handler | |
510 | * @po: packet offload declaration | |
511 | * | |
512 | * Remove a packet offload handler that was previously added to the kernel | |
513 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
514 | * removed from the kernel lists and can be freed or reused once this | |
515 | * function returns. | |
516 | * | |
517 | * This call sleeps to guarantee that no CPU is looking at the packet | |
518 | * type after return. | |
519 | */ | |
520 | void dev_remove_offload(struct packet_offload *po) | |
521 | { | |
522 | __dev_remove_offload(po); | |
523 | ||
524 | synchronize_net(); | |
525 | } | |
526 | EXPORT_SYMBOL(dev_remove_offload); | |
527 | ||
1da177e4 LT |
528 | /****************************************************************************** |
529 | ||
530 | Device Boot-time Settings Routines | |
531 | ||
532 | *******************************************************************************/ | |
533 | ||
534 | /* Boot time configuration table */ | |
535 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
536 | ||
537 | /** | |
538 | * netdev_boot_setup_add - add new setup entry | |
539 | * @name: name of the device | |
540 | * @map: configured settings for the device | |
541 | * | |
542 | * Adds new setup entry to the dev_boot_setup list. The function | |
543 | * returns 0 on error and 1 on success. This is a generic routine to | |
544 | * all netdevices. | |
545 | */ | |
546 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
547 | { | |
548 | struct netdev_boot_setup *s; | |
549 | int i; | |
550 | ||
551 | s = dev_boot_setup; | |
552 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
553 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
554 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 555 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
556 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
557 | break; | |
558 | } | |
559 | } | |
560 | ||
561 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
562 | } | |
563 | ||
564 | /** | |
565 | * netdev_boot_setup_check - check boot time settings | |
566 | * @dev: the netdevice | |
567 | * | |
568 | * Check boot time settings for the device. | |
569 | * The found settings are set for the device to be used | |
570 | * later in the device probing. | |
571 | * Returns 0 if no settings found, 1 if they are. | |
572 | */ | |
573 | int netdev_boot_setup_check(struct net_device *dev) | |
574 | { | |
575 | struct netdev_boot_setup *s = dev_boot_setup; | |
576 | int i; | |
577 | ||
578 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
579 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 580 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
581 | dev->irq = s[i].map.irq; |
582 | dev->base_addr = s[i].map.base_addr; | |
583 | dev->mem_start = s[i].map.mem_start; | |
584 | dev->mem_end = s[i].map.mem_end; | |
585 | return 1; | |
586 | } | |
587 | } | |
588 | return 0; | |
589 | } | |
d1b19dff | 590 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
591 | |
592 | ||
593 | /** | |
594 | * netdev_boot_base - get address from boot time settings | |
595 | * @prefix: prefix for network device | |
596 | * @unit: id for network device | |
597 | * | |
598 | * Check boot time settings for the base address of device. | |
599 | * The found settings are set for the device to be used | |
600 | * later in the device probing. | |
601 | * Returns 0 if no settings found. | |
602 | */ | |
603 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
604 | { | |
605 | const struct netdev_boot_setup *s = dev_boot_setup; | |
606 | char name[IFNAMSIZ]; | |
607 | int i; | |
608 | ||
609 | sprintf(name, "%s%d", prefix, unit); | |
610 | ||
611 | /* | |
612 | * If device already registered then return base of 1 | |
613 | * to indicate not to probe for this interface | |
614 | */ | |
881d966b | 615 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
616 | return 1; |
617 | ||
618 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
619 | if (!strcmp(name, s[i].name)) | |
620 | return s[i].map.base_addr; | |
621 | return 0; | |
622 | } | |
623 | ||
624 | /* | |
625 | * Saves at boot time configured settings for any netdevice. | |
626 | */ | |
627 | int __init netdev_boot_setup(char *str) | |
628 | { | |
629 | int ints[5]; | |
630 | struct ifmap map; | |
631 | ||
632 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
633 | if (!str || !*str) | |
634 | return 0; | |
635 | ||
636 | /* Save settings */ | |
637 | memset(&map, 0, sizeof(map)); | |
638 | if (ints[0] > 0) | |
639 | map.irq = ints[1]; | |
640 | if (ints[0] > 1) | |
641 | map.base_addr = ints[2]; | |
642 | if (ints[0] > 2) | |
643 | map.mem_start = ints[3]; | |
644 | if (ints[0] > 3) | |
645 | map.mem_end = ints[4]; | |
646 | ||
647 | /* Add new entry to the list */ | |
648 | return netdev_boot_setup_add(str, &map); | |
649 | } | |
650 | ||
651 | __setup("netdev=", netdev_boot_setup); | |
652 | ||
653 | /******************************************************************************* | |
654 | ||
655 | Device Interface Subroutines | |
656 | ||
657 | *******************************************************************************/ | |
658 | ||
659 | /** | |
660 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 661 | * @net: the applicable net namespace |
1da177e4 LT |
662 | * @name: name to find |
663 | * | |
664 | * Find an interface by name. Must be called under RTNL semaphore | |
665 | * or @dev_base_lock. If the name is found a pointer to the device | |
666 | * is returned. If the name is not found then %NULL is returned. The | |
667 | * reference counters are not incremented so the caller must be | |
668 | * careful with locks. | |
669 | */ | |
670 | ||
881d966b | 671 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 672 | { |
0bd8d536 ED |
673 | struct net_device *dev; |
674 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 675 | |
b67bfe0d | 676 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
677 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
678 | return dev; | |
0bd8d536 | 679 | |
1da177e4 LT |
680 | return NULL; |
681 | } | |
d1b19dff | 682 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 683 | |
72c9528b ED |
684 | /** |
685 | * dev_get_by_name_rcu - find a device by its name | |
686 | * @net: the applicable net namespace | |
687 | * @name: name to find | |
688 | * | |
689 | * Find an interface by name. | |
690 | * If the name is found a pointer to the device is returned. | |
691 | * If the name is not found then %NULL is returned. | |
692 | * The reference counters are not incremented so the caller must be | |
693 | * careful with locks. The caller must hold RCU lock. | |
694 | */ | |
695 | ||
696 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
697 | { | |
72c9528b ED |
698 | struct net_device *dev; |
699 | struct hlist_head *head = dev_name_hash(net, name); | |
700 | ||
b67bfe0d | 701 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
702 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
703 | return dev; | |
704 | ||
705 | return NULL; | |
706 | } | |
707 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
708 | ||
1da177e4 LT |
709 | /** |
710 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 711 | * @net: the applicable net namespace |
1da177e4 LT |
712 | * @name: name to find |
713 | * | |
714 | * Find an interface by name. This can be called from any | |
715 | * context and does its own locking. The returned handle has | |
716 | * the usage count incremented and the caller must use dev_put() to | |
717 | * release it when it is no longer needed. %NULL is returned if no | |
718 | * matching device is found. | |
719 | */ | |
720 | ||
881d966b | 721 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
722 | { |
723 | struct net_device *dev; | |
724 | ||
72c9528b ED |
725 | rcu_read_lock(); |
726 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
727 | if (dev) |
728 | dev_hold(dev); | |
72c9528b | 729 | rcu_read_unlock(); |
1da177e4 LT |
730 | return dev; |
731 | } | |
d1b19dff | 732 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
733 | |
734 | /** | |
735 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 736 | * @net: the applicable net namespace |
1da177e4 LT |
737 | * @ifindex: index of device |
738 | * | |
739 | * Search for an interface by index. Returns %NULL if the device | |
740 | * is not found or a pointer to the device. The device has not | |
741 | * had its reference counter increased so the caller must be careful | |
742 | * about locking. The caller must hold either the RTNL semaphore | |
743 | * or @dev_base_lock. | |
744 | */ | |
745 | ||
881d966b | 746 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 747 | { |
0bd8d536 ED |
748 | struct net_device *dev; |
749 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 750 | |
b67bfe0d | 751 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
752 | if (dev->ifindex == ifindex) |
753 | return dev; | |
0bd8d536 | 754 | |
1da177e4 LT |
755 | return NULL; |
756 | } | |
d1b19dff | 757 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 758 | |
fb699dfd ED |
759 | /** |
760 | * dev_get_by_index_rcu - find a device by its ifindex | |
761 | * @net: the applicable net namespace | |
762 | * @ifindex: index of device | |
763 | * | |
764 | * Search for an interface by index. Returns %NULL if the device | |
765 | * is not found or a pointer to the device. The device has not | |
766 | * had its reference counter increased so the caller must be careful | |
767 | * about locking. The caller must hold RCU lock. | |
768 | */ | |
769 | ||
770 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
771 | { | |
fb699dfd ED |
772 | struct net_device *dev; |
773 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
774 | ||
b67bfe0d | 775 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
776 | if (dev->ifindex == ifindex) |
777 | return dev; | |
778 | ||
779 | return NULL; | |
780 | } | |
781 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
782 | ||
1da177e4 LT |
783 | |
784 | /** | |
785 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 786 | * @net: the applicable net namespace |
1da177e4 LT |
787 | * @ifindex: index of device |
788 | * | |
789 | * Search for an interface by index. Returns NULL if the device | |
790 | * is not found or a pointer to the device. The device returned has | |
791 | * had a reference added and the pointer is safe until the user calls | |
792 | * dev_put to indicate they have finished with it. | |
793 | */ | |
794 | ||
881d966b | 795 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
796 | { |
797 | struct net_device *dev; | |
798 | ||
fb699dfd ED |
799 | rcu_read_lock(); |
800 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
801 | if (dev) |
802 | dev_hold(dev); | |
fb699dfd | 803 | rcu_read_unlock(); |
1da177e4 LT |
804 | return dev; |
805 | } | |
d1b19dff | 806 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 807 | |
5dbe7c17 NS |
808 | /** |
809 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
810 | * @net: network namespace | |
811 | * @name: a pointer to the buffer where the name will be stored. | |
812 | * @ifindex: the ifindex of the interface to get the name from. | |
813 | * | |
814 | * The use of raw_seqcount_begin() and cond_resched() before | |
815 | * retrying is required as we want to give the writers a chance | |
816 | * to complete when CONFIG_PREEMPT is not set. | |
817 | */ | |
818 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
819 | { | |
820 | struct net_device *dev; | |
821 | unsigned int seq; | |
822 | ||
823 | retry: | |
824 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
825 | rcu_read_lock(); | |
826 | dev = dev_get_by_index_rcu(net, ifindex); | |
827 | if (!dev) { | |
828 | rcu_read_unlock(); | |
829 | return -ENODEV; | |
830 | } | |
831 | ||
832 | strcpy(name, dev->name); | |
833 | rcu_read_unlock(); | |
834 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
835 | cond_resched(); | |
836 | goto retry; | |
837 | } | |
838 | ||
839 | return 0; | |
840 | } | |
841 | ||
1da177e4 | 842 | /** |
941666c2 | 843 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 844 | * @net: the applicable net namespace |
1da177e4 LT |
845 | * @type: media type of device |
846 | * @ha: hardware address | |
847 | * | |
848 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
849 | * is not found or a pointer to the device. |
850 | * The caller must hold RCU or RTNL. | |
941666c2 | 851 | * The returned device has not had its ref count increased |
1da177e4 LT |
852 | * and the caller must therefore be careful about locking |
853 | * | |
1da177e4 LT |
854 | */ |
855 | ||
941666c2 ED |
856 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
857 | const char *ha) | |
1da177e4 LT |
858 | { |
859 | struct net_device *dev; | |
860 | ||
941666c2 | 861 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
862 | if (dev->type == type && |
863 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
864 | return dev; |
865 | ||
866 | return NULL; | |
1da177e4 | 867 | } |
941666c2 | 868 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 869 | |
881d966b | 870 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
871 | { |
872 | struct net_device *dev; | |
873 | ||
4e9cac2b | 874 | ASSERT_RTNL(); |
881d966b | 875 | for_each_netdev(net, dev) |
4e9cac2b | 876 | if (dev->type == type) |
7562f876 PE |
877 | return dev; |
878 | ||
879 | return NULL; | |
4e9cac2b | 880 | } |
4e9cac2b PM |
881 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
882 | ||
881d966b | 883 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 884 | { |
99fe3c39 | 885 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 886 | |
99fe3c39 ED |
887 | rcu_read_lock(); |
888 | for_each_netdev_rcu(net, dev) | |
889 | if (dev->type == type) { | |
890 | dev_hold(dev); | |
891 | ret = dev; | |
892 | break; | |
893 | } | |
894 | rcu_read_unlock(); | |
895 | return ret; | |
1da177e4 | 896 | } |
1da177e4 LT |
897 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
898 | ||
899 | /** | |
6c555490 | 900 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 901 | * @net: the applicable net namespace |
1da177e4 LT |
902 | * @if_flags: IFF_* values |
903 | * @mask: bitmask of bits in if_flags to check | |
904 | * | |
905 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 906 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 907 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
908 | */ |
909 | ||
6c555490 WC |
910 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
911 | unsigned short mask) | |
1da177e4 | 912 | { |
7562f876 | 913 | struct net_device *dev, *ret; |
1da177e4 | 914 | |
6c555490 WC |
915 | ASSERT_RTNL(); |
916 | ||
7562f876 | 917 | ret = NULL; |
6c555490 | 918 | for_each_netdev(net, dev) { |
1da177e4 | 919 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 920 | ret = dev; |
1da177e4 LT |
921 | break; |
922 | } | |
923 | } | |
7562f876 | 924 | return ret; |
1da177e4 | 925 | } |
6c555490 | 926 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
927 | |
928 | /** | |
929 | * dev_valid_name - check if name is okay for network device | |
930 | * @name: name string | |
931 | * | |
932 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
933 | * to allow sysfs to work. We also disallow any kind of |
934 | * whitespace. | |
1da177e4 | 935 | */ |
95f050bf | 936 | bool dev_valid_name(const char *name) |
1da177e4 | 937 | { |
c7fa9d18 | 938 | if (*name == '\0') |
95f050bf | 939 | return false; |
b6fe17d6 | 940 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 941 | return false; |
c7fa9d18 | 942 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 943 | return false; |
c7fa9d18 DM |
944 | |
945 | while (*name) { | |
946 | if (*name == '/' || isspace(*name)) | |
95f050bf | 947 | return false; |
c7fa9d18 DM |
948 | name++; |
949 | } | |
95f050bf | 950 | return true; |
1da177e4 | 951 | } |
d1b19dff | 952 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
953 | |
954 | /** | |
b267b179 EB |
955 | * __dev_alloc_name - allocate a name for a device |
956 | * @net: network namespace to allocate the device name in | |
1da177e4 | 957 | * @name: name format string |
b267b179 | 958 | * @buf: scratch buffer and result name string |
1da177e4 LT |
959 | * |
960 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
961 | * id. It scans list of devices to build up a free map, then chooses |
962 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
963 | * while allocating the name and adding the device in order to avoid | |
964 | * duplicates. | |
965 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
966 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
967 | */ |
968 | ||
b267b179 | 969 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
970 | { |
971 | int i = 0; | |
1da177e4 LT |
972 | const char *p; |
973 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 974 | unsigned long *inuse; |
1da177e4 LT |
975 | struct net_device *d; |
976 | ||
977 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
978 | if (p) { | |
979 | /* | |
980 | * Verify the string as this thing may have come from | |
981 | * the user. There must be either one "%d" and no other "%" | |
982 | * characters. | |
983 | */ | |
984 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
985 | return -EINVAL; | |
986 | ||
987 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 988 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
989 | if (!inuse) |
990 | return -ENOMEM; | |
991 | ||
881d966b | 992 | for_each_netdev(net, d) { |
1da177e4 LT |
993 | if (!sscanf(d->name, name, &i)) |
994 | continue; | |
995 | if (i < 0 || i >= max_netdevices) | |
996 | continue; | |
997 | ||
998 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 999 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1000 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1001 | set_bit(i, inuse); | |
1002 | } | |
1003 | ||
1004 | i = find_first_zero_bit(inuse, max_netdevices); | |
1005 | free_page((unsigned long) inuse); | |
1006 | } | |
1007 | ||
d9031024 OP |
1008 | if (buf != name) |
1009 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1010 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1011 | return i; |
1da177e4 LT |
1012 | |
1013 | /* It is possible to run out of possible slots | |
1014 | * when the name is long and there isn't enough space left | |
1015 | * for the digits, or if all bits are used. | |
1016 | */ | |
1017 | return -ENFILE; | |
1018 | } | |
1019 | ||
b267b179 EB |
1020 | /** |
1021 | * dev_alloc_name - allocate a name for a device | |
1022 | * @dev: device | |
1023 | * @name: name format string | |
1024 | * | |
1025 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1026 | * id. It scans list of devices to build up a free map, then chooses | |
1027 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1028 | * while allocating the name and adding the device in order to avoid | |
1029 | * duplicates. | |
1030 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1031 | * Returns the number of the unit assigned or a negative errno code. | |
1032 | */ | |
1033 | ||
1034 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1035 | { | |
1036 | char buf[IFNAMSIZ]; | |
1037 | struct net *net; | |
1038 | int ret; | |
1039 | ||
c346dca1 YH |
1040 | BUG_ON(!dev_net(dev)); |
1041 | net = dev_net(dev); | |
b267b179 EB |
1042 | ret = __dev_alloc_name(net, name, buf); |
1043 | if (ret >= 0) | |
1044 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1045 | return ret; | |
1046 | } | |
d1b19dff | 1047 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1048 | |
828de4f6 G |
1049 | static int dev_alloc_name_ns(struct net *net, |
1050 | struct net_device *dev, | |
1051 | const char *name) | |
d9031024 | 1052 | { |
828de4f6 G |
1053 | char buf[IFNAMSIZ]; |
1054 | int ret; | |
8ce6cebc | 1055 | |
828de4f6 G |
1056 | ret = __dev_alloc_name(net, name, buf); |
1057 | if (ret >= 0) | |
1058 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1059 | return ret; | |
1060 | } | |
1061 | ||
1062 | static int dev_get_valid_name(struct net *net, | |
1063 | struct net_device *dev, | |
1064 | const char *name) | |
1065 | { | |
1066 | BUG_ON(!net); | |
8ce6cebc | 1067 | |
d9031024 OP |
1068 | if (!dev_valid_name(name)) |
1069 | return -EINVAL; | |
1070 | ||
1c5cae81 | 1071 | if (strchr(name, '%')) |
828de4f6 | 1072 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1073 | else if (__dev_get_by_name(net, name)) |
1074 | return -EEXIST; | |
8ce6cebc DL |
1075 | else if (dev->name != name) |
1076 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1077 | |
1078 | return 0; | |
1079 | } | |
1da177e4 LT |
1080 | |
1081 | /** | |
1082 | * dev_change_name - change name of a device | |
1083 | * @dev: device | |
1084 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1085 | * | |
1086 | * Change name of a device, can pass format strings "eth%d". | |
1087 | * for wildcarding. | |
1088 | */ | |
cf04a4c7 | 1089 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1090 | { |
238fa362 | 1091 | unsigned char old_assign_type; |
fcc5a03a | 1092 | char oldname[IFNAMSIZ]; |
1da177e4 | 1093 | int err = 0; |
fcc5a03a | 1094 | int ret; |
881d966b | 1095 | struct net *net; |
1da177e4 LT |
1096 | |
1097 | ASSERT_RTNL(); | |
c346dca1 | 1098 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1099 | |
c346dca1 | 1100 | net = dev_net(dev); |
1da177e4 LT |
1101 | if (dev->flags & IFF_UP) |
1102 | return -EBUSY; | |
1103 | ||
30e6c9fa | 1104 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1105 | |
1106 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1107 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1108 | return 0; |
c91f6df2 | 1109 | } |
c8d90dca | 1110 | |
fcc5a03a HX |
1111 | memcpy(oldname, dev->name, IFNAMSIZ); |
1112 | ||
828de4f6 | 1113 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1114 | if (err < 0) { |
30e6c9fa | 1115 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1116 | return err; |
c91f6df2 | 1117 | } |
1da177e4 | 1118 | |
6fe82a39 VF |
1119 | if (oldname[0] && !strchr(oldname, '%')) |
1120 | netdev_info(dev, "renamed from %s\n", oldname); | |
1121 | ||
238fa362 TG |
1122 | old_assign_type = dev->name_assign_type; |
1123 | dev->name_assign_type = NET_NAME_RENAMED; | |
1124 | ||
fcc5a03a | 1125 | rollback: |
a1b3f594 EB |
1126 | ret = device_rename(&dev->dev, dev->name); |
1127 | if (ret) { | |
1128 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1129 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1130 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1131 | return ret; |
dcc99773 | 1132 | } |
7f988eab | 1133 | |
30e6c9fa | 1134 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1135 | |
5bb025fa VF |
1136 | netdev_adjacent_rename_links(dev, oldname); |
1137 | ||
7f988eab | 1138 | write_lock_bh(&dev_base_lock); |
372b2312 | 1139 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1140 | write_unlock_bh(&dev_base_lock); |
1141 | ||
1142 | synchronize_rcu(); | |
1143 | ||
1144 | write_lock_bh(&dev_base_lock); | |
1145 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1146 | write_unlock_bh(&dev_base_lock); |
1147 | ||
056925ab | 1148 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1149 | ret = notifier_to_errno(ret); |
1150 | ||
1151 | if (ret) { | |
91e9c07b ED |
1152 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1153 | if (err >= 0) { | |
fcc5a03a | 1154 | err = ret; |
30e6c9fa | 1155 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1156 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1157 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1158 | dev->name_assign_type = old_assign_type; |
1159 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1160 | goto rollback; |
91e9c07b | 1161 | } else { |
7b6cd1ce | 1162 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1163 | dev->name, ret); |
fcc5a03a HX |
1164 | } |
1165 | } | |
1da177e4 LT |
1166 | |
1167 | return err; | |
1168 | } | |
1169 | ||
0b815a1a SH |
1170 | /** |
1171 | * dev_set_alias - change ifalias of a device | |
1172 | * @dev: device | |
1173 | * @alias: name up to IFALIASZ | |
f0db275a | 1174 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1175 | * |
1176 | * Set ifalias for a device, | |
1177 | */ | |
1178 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1179 | { | |
7364e445 AK |
1180 | char *new_ifalias; |
1181 | ||
0b815a1a SH |
1182 | ASSERT_RTNL(); |
1183 | ||
1184 | if (len >= IFALIASZ) | |
1185 | return -EINVAL; | |
1186 | ||
96ca4a2c | 1187 | if (!len) { |
388dfc2d SK |
1188 | kfree(dev->ifalias); |
1189 | dev->ifalias = NULL; | |
96ca4a2c OH |
1190 | return 0; |
1191 | } | |
1192 | ||
7364e445 AK |
1193 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1194 | if (!new_ifalias) | |
0b815a1a | 1195 | return -ENOMEM; |
7364e445 | 1196 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1197 | |
1198 | strlcpy(dev->ifalias, alias, len+1); | |
1199 | return len; | |
1200 | } | |
1201 | ||
1202 | ||
d8a33ac4 | 1203 | /** |
3041a069 | 1204 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1205 | * @dev: device to cause notification |
1206 | * | |
1207 | * Called to indicate a device has changed features. | |
1208 | */ | |
1209 | void netdev_features_change(struct net_device *dev) | |
1210 | { | |
056925ab | 1211 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1212 | } |
1213 | EXPORT_SYMBOL(netdev_features_change); | |
1214 | ||
1da177e4 LT |
1215 | /** |
1216 | * netdev_state_change - device changes state | |
1217 | * @dev: device to cause notification | |
1218 | * | |
1219 | * Called to indicate a device has changed state. This function calls | |
1220 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1221 | * to the routing socket. | |
1222 | */ | |
1223 | void netdev_state_change(struct net_device *dev) | |
1224 | { | |
1225 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1226 | struct netdev_notifier_change_info change_info; |
1227 | ||
1228 | change_info.flags_changed = 0; | |
1229 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1230 | &change_info.info); | |
7f294054 | 1231 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1232 | } |
1233 | } | |
d1b19dff | 1234 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1235 | |
ee89bab1 AW |
1236 | /** |
1237 | * netdev_notify_peers - notify network peers about existence of @dev | |
1238 | * @dev: network device | |
1239 | * | |
1240 | * Generate traffic such that interested network peers are aware of | |
1241 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1242 | * a device wants to inform the rest of the network about some sort of | |
1243 | * reconfiguration such as a failover event or virtual machine | |
1244 | * migration. | |
1245 | */ | |
1246 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1247 | { |
ee89bab1 AW |
1248 | rtnl_lock(); |
1249 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1250 | rtnl_unlock(); | |
c1da4ac7 | 1251 | } |
ee89bab1 | 1252 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1253 | |
bd380811 | 1254 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1255 | { |
d314774c | 1256 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1257 | int ret; |
1da177e4 | 1258 | |
e46b66bc BH |
1259 | ASSERT_RTNL(); |
1260 | ||
1da177e4 LT |
1261 | if (!netif_device_present(dev)) |
1262 | return -ENODEV; | |
1263 | ||
ca99ca14 NH |
1264 | /* Block netpoll from trying to do any rx path servicing. |
1265 | * If we don't do this there is a chance ndo_poll_controller | |
1266 | * or ndo_poll may be running while we open the device | |
1267 | */ | |
66b5552f | 1268 | netpoll_poll_disable(dev); |
ca99ca14 | 1269 | |
3b8bcfd5 JB |
1270 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1271 | ret = notifier_to_errno(ret); | |
1272 | if (ret) | |
1273 | return ret; | |
1274 | ||
1da177e4 | 1275 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1276 | |
d314774c SH |
1277 | if (ops->ndo_validate_addr) |
1278 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1279 | |
d314774c SH |
1280 | if (!ret && ops->ndo_open) |
1281 | ret = ops->ndo_open(dev); | |
1da177e4 | 1282 | |
66b5552f | 1283 | netpoll_poll_enable(dev); |
ca99ca14 | 1284 | |
bada339b JG |
1285 | if (ret) |
1286 | clear_bit(__LINK_STATE_START, &dev->state); | |
1287 | else { | |
1da177e4 | 1288 | dev->flags |= IFF_UP; |
b4bd07c2 | 1289 | net_dmaengine_get(); |
4417da66 | 1290 | dev_set_rx_mode(dev); |
1da177e4 | 1291 | dev_activate(dev); |
7bf23575 | 1292 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1293 | } |
bada339b | 1294 | |
1da177e4 LT |
1295 | return ret; |
1296 | } | |
1297 | ||
1298 | /** | |
bd380811 PM |
1299 | * dev_open - prepare an interface for use. |
1300 | * @dev: device to open | |
1da177e4 | 1301 | * |
bd380811 PM |
1302 | * Takes a device from down to up state. The device's private open |
1303 | * function is invoked and then the multicast lists are loaded. Finally | |
1304 | * the device is moved into the up state and a %NETDEV_UP message is | |
1305 | * sent to the netdev notifier chain. | |
1306 | * | |
1307 | * Calling this function on an active interface is a nop. On a failure | |
1308 | * a negative errno code is returned. | |
1da177e4 | 1309 | */ |
bd380811 PM |
1310 | int dev_open(struct net_device *dev) |
1311 | { | |
1312 | int ret; | |
1313 | ||
bd380811 PM |
1314 | if (dev->flags & IFF_UP) |
1315 | return 0; | |
1316 | ||
bd380811 PM |
1317 | ret = __dev_open(dev); |
1318 | if (ret < 0) | |
1319 | return ret; | |
1320 | ||
7f294054 | 1321 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1322 | call_netdevice_notifiers(NETDEV_UP, dev); |
1323 | ||
1324 | return ret; | |
1325 | } | |
1326 | EXPORT_SYMBOL(dev_open); | |
1327 | ||
44345724 | 1328 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1329 | { |
44345724 | 1330 | struct net_device *dev; |
e46b66bc | 1331 | |
bd380811 | 1332 | ASSERT_RTNL(); |
9d5010db DM |
1333 | might_sleep(); |
1334 | ||
5cde2829 | 1335 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1336 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1337 | netpoll_poll_disable(dev); |
3f4df206 | 1338 | |
44345724 | 1339 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1340 | |
44345724 | 1341 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1342 | |
44345724 OP |
1343 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1344 | * can be even on different cpu. So just clear netif_running(). | |
1345 | * | |
1346 | * dev->stop() will invoke napi_disable() on all of it's | |
1347 | * napi_struct instances on this device. | |
1348 | */ | |
4e857c58 | 1349 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1350 | } |
1da177e4 | 1351 | |
44345724 | 1352 | dev_deactivate_many(head); |
d8b2a4d2 | 1353 | |
5cde2829 | 1354 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1355 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1356 | |
44345724 OP |
1357 | /* |
1358 | * Call the device specific close. This cannot fail. | |
1359 | * Only if device is UP | |
1360 | * | |
1361 | * We allow it to be called even after a DETACH hot-plug | |
1362 | * event. | |
1363 | */ | |
1364 | if (ops->ndo_stop) | |
1365 | ops->ndo_stop(dev); | |
1366 | ||
44345724 | 1367 | dev->flags &= ~IFF_UP; |
44345724 | 1368 | net_dmaengine_put(); |
66b5552f | 1369 | netpoll_poll_enable(dev); |
44345724 OP |
1370 | } |
1371 | ||
1372 | return 0; | |
1373 | } | |
1374 | ||
1375 | static int __dev_close(struct net_device *dev) | |
1376 | { | |
f87e6f47 | 1377 | int retval; |
44345724 OP |
1378 | LIST_HEAD(single); |
1379 | ||
5cde2829 | 1380 | list_add(&dev->close_list, &single); |
f87e6f47 LT |
1381 | retval = __dev_close_many(&single); |
1382 | list_del(&single); | |
ca99ca14 | 1383 | |
f87e6f47 | 1384 | return retval; |
44345724 OP |
1385 | } |
1386 | ||
3fbd8758 | 1387 | static int dev_close_many(struct list_head *head) |
44345724 OP |
1388 | { |
1389 | struct net_device *dev, *tmp; | |
1da177e4 | 1390 | |
5cde2829 EB |
1391 | /* Remove the devices that don't need to be closed */ |
1392 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1393 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1394 | list_del_init(&dev->close_list); |
44345724 OP |
1395 | |
1396 | __dev_close_many(head); | |
1da177e4 | 1397 | |
5cde2829 | 1398 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1399 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1400 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
5cde2829 | 1401 | list_del_init(&dev->close_list); |
44345724 | 1402 | } |
bd380811 PM |
1403 | |
1404 | return 0; | |
1405 | } | |
1406 | ||
1407 | /** | |
1408 | * dev_close - shutdown an interface. | |
1409 | * @dev: device to shutdown | |
1410 | * | |
1411 | * This function moves an active device into down state. A | |
1412 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1413 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1414 | * chain. | |
1415 | */ | |
1416 | int dev_close(struct net_device *dev) | |
1417 | { | |
e14a5993 ED |
1418 | if (dev->flags & IFF_UP) { |
1419 | LIST_HEAD(single); | |
1da177e4 | 1420 | |
5cde2829 | 1421 | list_add(&dev->close_list, &single); |
e14a5993 ED |
1422 | dev_close_many(&single); |
1423 | list_del(&single); | |
1424 | } | |
da6e378b | 1425 | return 0; |
1da177e4 | 1426 | } |
d1b19dff | 1427 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1428 | |
1429 | ||
0187bdfb BH |
1430 | /** |
1431 | * dev_disable_lro - disable Large Receive Offload on a device | |
1432 | * @dev: device | |
1433 | * | |
1434 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1435 | * called under RTNL. This is needed if received packets may be | |
1436 | * forwarded to another interface. | |
1437 | */ | |
1438 | void dev_disable_lro(struct net_device *dev) | |
1439 | { | |
f11970e3 NH |
1440 | /* |
1441 | * If we're trying to disable lro on a vlan device | |
1442 | * use the underlying physical device instead | |
1443 | */ | |
1444 | if (is_vlan_dev(dev)) | |
1445 | dev = vlan_dev_real_dev(dev); | |
1446 | ||
529d0489 MK |
1447 | /* the same for macvlan devices */ |
1448 | if (netif_is_macvlan(dev)) | |
1449 | dev = macvlan_dev_real_dev(dev); | |
1450 | ||
bc5787c6 MM |
1451 | dev->wanted_features &= ~NETIF_F_LRO; |
1452 | netdev_update_features(dev); | |
27660515 | 1453 | |
22d5969f MM |
1454 | if (unlikely(dev->features & NETIF_F_LRO)) |
1455 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
0187bdfb BH |
1456 | } |
1457 | EXPORT_SYMBOL(dev_disable_lro); | |
1458 | ||
351638e7 JP |
1459 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1460 | struct net_device *dev) | |
1461 | { | |
1462 | struct netdev_notifier_info info; | |
1463 | ||
1464 | netdev_notifier_info_init(&info, dev); | |
1465 | return nb->notifier_call(nb, val, &info); | |
1466 | } | |
0187bdfb | 1467 | |
881d966b EB |
1468 | static int dev_boot_phase = 1; |
1469 | ||
1da177e4 LT |
1470 | /** |
1471 | * register_netdevice_notifier - register a network notifier block | |
1472 | * @nb: notifier | |
1473 | * | |
1474 | * Register a notifier to be called when network device events occur. | |
1475 | * The notifier passed is linked into the kernel structures and must | |
1476 | * not be reused until it has been unregistered. A negative errno code | |
1477 | * is returned on a failure. | |
1478 | * | |
1479 | * When registered all registration and up events are replayed | |
4ec93edb | 1480 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1481 | * view of the network device list. |
1482 | */ | |
1483 | ||
1484 | int register_netdevice_notifier(struct notifier_block *nb) | |
1485 | { | |
1486 | struct net_device *dev; | |
fcc5a03a | 1487 | struct net_device *last; |
881d966b | 1488 | struct net *net; |
1da177e4 LT |
1489 | int err; |
1490 | ||
1491 | rtnl_lock(); | |
f07d5b94 | 1492 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1493 | if (err) |
1494 | goto unlock; | |
881d966b EB |
1495 | if (dev_boot_phase) |
1496 | goto unlock; | |
1497 | for_each_net(net) { | |
1498 | for_each_netdev(net, dev) { | |
351638e7 | 1499 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1500 | err = notifier_to_errno(err); |
1501 | if (err) | |
1502 | goto rollback; | |
1503 | ||
1504 | if (!(dev->flags & IFF_UP)) | |
1505 | continue; | |
1da177e4 | 1506 | |
351638e7 | 1507 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1508 | } |
1da177e4 | 1509 | } |
fcc5a03a HX |
1510 | |
1511 | unlock: | |
1da177e4 LT |
1512 | rtnl_unlock(); |
1513 | return err; | |
fcc5a03a HX |
1514 | |
1515 | rollback: | |
1516 | last = dev; | |
881d966b EB |
1517 | for_each_net(net) { |
1518 | for_each_netdev(net, dev) { | |
1519 | if (dev == last) | |
8f891489 | 1520 | goto outroll; |
fcc5a03a | 1521 | |
881d966b | 1522 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1523 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1524 | dev); | |
1525 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1526 | } |
351638e7 | 1527 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1528 | } |
fcc5a03a | 1529 | } |
c67625a1 | 1530 | |
8f891489 | 1531 | outroll: |
c67625a1 | 1532 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1533 | goto unlock; |
1da177e4 | 1534 | } |
d1b19dff | 1535 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1536 | |
1537 | /** | |
1538 | * unregister_netdevice_notifier - unregister a network notifier block | |
1539 | * @nb: notifier | |
1540 | * | |
1541 | * Unregister a notifier previously registered by | |
1542 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1543 | * kernel structures and may then be reused. A negative errno code | |
1544 | * is returned on a failure. | |
7d3d43da EB |
1545 | * |
1546 | * After unregistering unregister and down device events are synthesized | |
1547 | * for all devices on the device list to the removed notifier to remove | |
1548 | * the need for special case cleanup code. | |
1da177e4 LT |
1549 | */ |
1550 | ||
1551 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1552 | { | |
7d3d43da EB |
1553 | struct net_device *dev; |
1554 | struct net *net; | |
9f514950 HX |
1555 | int err; |
1556 | ||
1557 | rtnl_lock(); | |
f07d5b94 | 1558 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1559 | if (err) |
1560 | goto unlock; | |
1561 | ||
1562 | for_each_net(net) { | |
1563 | for_each_netdev(net, dev) { | |
1564 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1565 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1566 | dev); | |
1567 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1568 | } |
351638e7 | 1569 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1570 | } |
1571 | } | |
1572 | unlock: | |
9f514950 HX |
1573 | rtnl_unlock(); |
1574 | return err; | |
1da177e4 | 1575 | } |
d1b19dff | 1576 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1577 | |
351638e7 JP |
1578 | /** |
1579 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1580 | * @val: value passed unmodified to notifier function | |
1581 | * @dev: net_device pointer passed unmodified to notifier function | |
1582 | * @info: notifier information data | |
1583 | * | |
1584 | * Call all network notifier blocks. Parameters and return value | |
1585 | * are as for raw_notifier_call_chain(). | |
1586 | */ | |
1587 | ||
1d143d9f | 1588 | static int call_netdevice_notifiers_info(unsigned long val, |
1589 | struct net_device *dev, | |
1590 | struct netdev_notifier_info *info) | |
351638e7 JP |
1591 | { |
1592 | ASSERT_RTNL(); | |
1593 | netdev_notifier_info_init(info, dev); | |
1594 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1595 | } | |
351638e7 | 1596 | |
1da177e4 LT |
1597 | /** |
1598 | * call_netdevice_notifiers - call all network notifier blocks | |
1599 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1600 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1601 | * |
1602 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1603 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1604 | */ |
1605 | ||
ad7379d4 | 1606 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1607 | { |
351638e7 JP |
1608 | struct netdev_notifier_info info; |
1609 | ||
1610 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1611 | } |
edf947f1 | 1612 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1613 | |
c5905afb | 1614 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1615 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1616 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1617 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1618 | * static_key_slow_dec() calls. |
b90e5794 ED |
1619 | */ |
1620 | static atomic_t netstamp_needed_deferred; | |
1621 | #endif | |
1da177e4 LT |
1622 | |
1623 | void net_enable_timestamp(void) | |
1624 | { | |
b90e5794 ED |
1625 | #ifdef HAVE_JUMP_LABEL |
1626 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1627 | ||
1628 | if (deferred) { | |
1629 | while (--deferred) | |
c5905afb | 1630 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1631 | return; |
1632 | } | |
1633 | #endif | |
c5905afb | 1634 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1635 | } |
d1b19dff | 1636 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1637 | |
1638 | void net_disable_timestamp(void) | |
1639 | { | |
b90e5794 ED |
1640 | #ifdef HAVE_JUMP_LABEL |
1641 | if (in_interrupt()) { | |
1642 | atomic_inc(&netstamp_needed_deferred); | |
1643 | return; | |
1644 | } | |
1645 | #endif | |
c5905afb | 1646 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1647 | } |
d1b19dff | 1648 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1649 | |
3b098e2d | 1650 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1651 | { |
588f0330 | 1652 | skb->tstamp.tv64 = 0; |
c5905afb | 1653 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1654 | __net_timestamp(skb); |
1da177e4 LT |
1655 | } |
1656 | ||
588f0330 | 1657 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1658 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1659 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1660 | __net_timestamp(SKB); \ | |
1661 | } \ | |
3b098e2d | 1662 | |
1ee481fb | 1663 | bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb) |
79b569f0 DL |
1664 | { |
1665 | unsigned int len; | |
1666 | ||
1667 | if (!(dev->flags & IFF_UP)) | |
1668 | return false; | |
1669 | ||
1670 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1671 | if (skb->len <= len) | |
1672 | return true; | |
1673 | ||
1674 | /* if TSO is enabled, we don't care about the length as the packet | |
1675 | * could be forwarded without being segmented before | |
1676 | */ | |
1677 | if (skb_is_gso(skb)) | |
1678 | return true; | |
1679 | ||
1680 | return false; | |
1681 | } | |
1ee481fb | 1682 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1683 | |
a0265d28 HX |
1684 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1685 | { | |
1686 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { | |
1687 | if (skb_copy_ubufs(skb, GFP_ATOMIC)) { | |
1688 | atomic_long_inc(&dev->rx_dropped); | |
1689 | kfree_skb(skb); | |
1690 | return NET_RX_DROP; | |
1691 | } | |
1692 | } | |
1693 | ||
1694 | if (unlikely(!is_skb_forwardable(dev, skb))) { | |
1695 | atomic_long_inc(&dev->rx_dropped); | |
1696 | kfree_skb(skb); | |
1697 | return NET_RX_DROP; | |
1698 | } | |
1699 | ||
1700 | skb_scrub_packet(skb, true); | |
1701 | skb->protocol = eth_type_trans(skb, dev); | |
1702 | ||
1703 | return 0; | |
1704 | } | |
1705 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1706 | ||
44540960 AB |
1707 | /** |
1708 | * dev_forward_skb - loopback an skb to another netif | |
1709 | * | |
1710 | * @dev: destination network device | |
1711 | * @skb: buffer to forward | |
1712 | * | |
1713 | * return values: | |
1714 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1715 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1716 | * |
1717 | * dev_forward_skb can be used for injecting an skb from the | |
1718 | * start_xmit function of one device into the receive queue | |
1719 | * of another device. | |
1720 | * | |
1721 | * The receiving device may be in another namespace, so | |
1722 | * we have to clear all information in the skb that could | |
1723 | * impact namespace isolation. | |
1724 | */ | |
1725 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1726 | { | |
a0265d28 | 1727 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1728 | } |
1729 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1730 | ||
71d9dec2 CG |
1731 | static inline int deliver_skb(struct sk_buff *skb, |
1732 | struct packet_type *pt_prev, | |
1733 | struct net_device *orig_dev) | |
1734 | { | |
1080e512 MT |
1735 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1736 | return -ENOMEM; | |
71d9dec2 CG |
1737 | atomic_inc(&skb->users); |
1738 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1739 | } | |
1740 | ||
c0de08d0 EL |
1741 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1742 | { | |
a3d744e9 | 1743 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1744 | return false; |
1745 | ||
1746 | if (ptype->id_match) | |
1747 | return ptype->id_match(ptype, skb->sk); | |
1748 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1749 | return true; | |
1750 | ||
1751 | return false; | |
1752 | } | |
1753 | ||
1da177e4 LT |
1754 | /* |
1755 | * Support routine. Sends outgoing frames to any network | |
1756 | * taps currently in use. | |
1757 | */ | |
1758 | ||
f6a78bfc | 1759 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1760 | { |
1761 | struct packet_type *ptype; | |
71d9dec2 CG |
1762 | struct sk_buff *skb2 = NULL; |
1763 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1764 | |
1da177e4 LT |
1765 | rcu_read_lock(); |
1766 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1767 | /* Never send packets back to the socket | |
1768 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1769 | */ | |
1770 | if ((ptype->dev == dev || !ptype->dev) && | |
c0de08d0 | 1771 | (!skb_loop_sk(ptype, skb))) { |
71d9dec2 CG |
1772 | if (pt_prev) { |
1773 | deliver_skb(skb2, pt_prev, skb->dev); | |
1774 | pt_prev = ptype; | |
1775 | continue; | |
1776 | } | |
1777 | ||
1778 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1779 | if (!skb2) |
1780 | break; | |
1781 | ||
70978182 ED |
1782 | net_timestamp_set(skb2); |
1783 | ||
1da177e4 LT |
1784 | /* skb->nh should be correctly |
1785 | set by sender, so that the second statement is | |
1786 | just protection against buggy protocols. | |
1787 | */ | |
459a98ed | 1788 | skb_reset_mac_header(skb2); |
1da177e4 | 1789 | |
d56f90a7 | 1790 | if (skb_network_header(skb2) < skb2->data || |
ced14f68 | 1791 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { |
e87cc472 JP |
1792 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", |
1793 | ntohs(skb2->protocol), | |
1794 | dev->name); | |
c1d2bbe1 | 1795 | skb_reset_network_header(skb2); |
1da177e4 LT |
1796 | } |
1797 | ||
b0e380b1 | 1798 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1799 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1800 | pt_prev = ptype; |
1da177e4 LT |
1801 | } |
1802 | } | |
71d9dec2 CG |
1803 | if (pt_prev) |
1804 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1805 | rcu_read_unlock(); |
1806 | } | |
1807 | ||
2c53040f BH |
1808 | /** |
1809 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1810 | * @dev: Network device |
1811 | * @txq: number of queues available | |
1812 | * | |
1813 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1814 | * valid. To resolve this verify the tc mapping remains valid and if | |
1815 | * not NULL the mapping. With no priorities mapping to this | |
1816 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1817 | * is invalid nothing can be done so disable priority mappings. If is | |
1818 | * expected that drivers will fix this mapping if they can before | |
1819 | * calling netif_set_real_num_tx_queues. | |
1820 | */ | |
bb134d22 | 1821 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1822 | { |
1823 | int i; | |
1824 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1825 | ||
1826 | /* If TC0 is invalidated disable TC mapping */ | |
1827 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1828 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1829 | dev->num_tc = 0; |
1830 | return; | |
1831 | } | |
1832 | ||
1833 | /* Invalidated prio to tc mappings set to TC0 */ | |
1834 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1835 | int q = netdev_get_prio_tc_map(dev, i); | |
1836 | ||
1837 | tc = &dev->tc_to_txq[q]; | |
1838 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1839 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1840 | i, q); | |
4f57c087 JF |
1841 | netdev_set_prio_tc_map(dev, i, 0); |
1842 | } | |
1843 | } | |
1844 | } | |
1845 | ||
537c00de AD |
1846 | #ifdef CONFIG_XPS |
1847 | static DEFINE_MUTEX(xps_map_mutex); | |
1848 | #define xmap_dereference(P) \ | |
1849 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1850 | ||
10cdc3f3 AD |
1851 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1852 | int cpu, u16 index) | |
537c00de | 1853 | { |
10cdc3f3 AD |
1854 | struct xps_map *map = NULL; |
1855 | int pos; | |
537c00de | 1856 | |
10cdc3f3 AD |
1857 | if (dev_maps) |
1858 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1859 | |
10cdc3f3 AD |
1860 | for (pos = 0; map && pos < map->len; pos++) { |
1861 | if (map->queues[pos] == index) { | |
537c00de AD |
1862 | if (map->len > 1) { |
1863 | map->queues[pos] = map->queues[--map->len]; | |
1864 | } else { | |
10cdc3f3 | 1865 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1866 | kfree_rcu(map, rcu); |
1867 | map = NULL; | |
1868 | } | |
10cdc3f3 | 1869 | break; |
537c00de | 1870 | } |
537c00de AD |
1871 | } |
1872 | ||
10cdc3f3 AD |
1873 | return map; |
1874 | } | |
1875 | ||
024e9679 | 1876 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
10cdc3f3 AD |
1877 | { |
1878 | struct xps_dev_maps *dev_maps; | |
024e9679 | 1879 | int cpu, i; |
10cdc3f3 AD |
1880 | bool active = false; |
1881 | ||
1882 | mutex_lock(&xps_map_mutex); | |
1883 | dev_maps = xmap_dereference(dev->xps_maps); | |
1884 | ||
1885 | if (!dev_maps) | |
1886 | goto out_no_maps; | |
1887 | ||
1888 | for_each_possible_cpu(cpu) { | |
024e9679 AD |
1889 | for (i = index; i < dev->num_tx_queues; i++) { |
1890 | if (!remove_xps_queue(dev_maps, cpu, i)) | |
1891 | break; | |
1892 | } | |
1893 | if (i == dev->num_tx_queues) | |
10cdc3f3 AD |
1894 | active = true; |
1895 | } | |
1896 | ||
1897 | if (!active) { | |
537c00de AD |
1898 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
1899 | kfree_rcu(dev_maps, rcu); | |
1900 | } | |
1901 | ||
024e9679 AD |
1902 | for (i = index; i < dev->num_tx_queues; i++) |
1903 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), | |
1904 | NUMA_NO_NODE); | |
1905 | ||
537c00de AD |
1906 | out_no_maps: |
1907 | mutex_unlock(&xps_map_mutex); | |
1908 | } | |
1909 | ||
01c5f864 AD |
1910 | static struct xps_map *expand_xps_map(struct xps_map *map, |
1911 | int cpu, u16 index) | |
1912 | { | |
1913 | struct xps_map *new_map; | |
1914 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
1915 | int i, pos; | |
1916 | ||
1917 | for (pos = 0; map && pos < map->len; pos++) { | |
1918 | if (map->queues[pos] != index) | |
1919 | continue; | |
1920 | return map; | |
1921 | } | |
1922 | ||
1923 | /* Need to add queue to this CPU's existing map */ | |
1924 | if (map) { | |
1925 | if (pos < map->alloc_len) | |
1926 | return map; | |
1927 | ||
1928 | alloc_len = map->alloc_len * 2; | |
1929 | } | |
1930 | ||
1931 | /* Need to allocate new map to store queue on this CPU's map */ | |
1932 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
1933 | cpu_to_node(cpu)); | |
1934 | if (!new_map) | |
1935 | return NULL; | |
1936 | ||
1937 | for (i = 0; i < pos; i++) | |
1938 | new_map->queues[i] = map->queues[i]; | |
1939 | new_map->alloc_len = alloc_len; | |
1940 | new_map->len = pos; | |
1941 | ||
1942 | return new_map; | |
1943 | } | |
1944 | ||
3573540c MT |
1945 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
1946 | u16 index) | |
537c00de | 1947 | { |
01c5f864 | 1948 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 1949 | struct xps_map *map, *new_map; |
537c00de | 1950 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
1951 | int cpu, numa_node_id = -2; |
1952 | bool active = false; | |
537c00de AD |
1953 | |
1954 | mutex_lock(&xps_map_mutex); | |
1955 | ||
1956 | dev_maps = xmap_dereference(dev->xps_maps); | |
1957 | ||
01c5f864 AD |
1958 | /* allocate memory for queue storage */ |
1959 | for_each_online_cpu(cpu) { | |
1960 | if (!cpumask_test_cpu(cpu, mask)) | |
1961 | continue; | |
1962 | ||
1963 | if (!new_dev_maps) | |
1964 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
1965 | if (!new_dev_maps) { |
1966 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 1967 | return -ENOMEM; |
2bb60cb9 | 1968 | } |
01c5f864 AD |
1969 | |
1970 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
1971 | NULL; | |
1972 | ||
1973 | map = expand_xps_map(map, cpu, index); | |
1974 | if (!map) | |
1975 | goto error; | |
1976 | ||
1977 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
1978 | } | |
1979 | ||
1980 | if (!new_dev_maps) | |
1981 | goto out_no_new_maps; | |
1982 | ||
537c00de | 1983 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
1984 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
1985 | /* add queue to CPU maps */ | |
1986 | int pos = 0; | |
1987 | ||
1988 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1989 | while ((pos < map->len) && (map->queues[pos] != index)) | |
1990 | pos++; | |
1991 | ||
1992 | if (pos == map->len) | |
1993 | map->queues[map->len++] = index; | |
537c00de | 1994 | #ifdef CONFIG_NUMA |
537c00de AD |
1995 | if (numa_node_id == -2) |
1996 | numa_node_id = cpu_to_node(cpu); | |
1997 | else if (numa_node_id != cpu_to_node(cpu)) | |
1998 | numa_node_id = -1; | |
537c00de | 1999 | #endif |
01c5f864 AD |
2000 | } else if (dev_maps) { |
2001 | /* fill in the new device map from the old device map */ | |
2002 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2003 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 2004 | } |
01c5f864 | 2005 | |
537c00de AD |
2006 | } |
2007 | ||
01c5f864 AD |
2008 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2009 | ||
537c00de | 2010 | /* Cleanup old maps */ |
01c5f864 AD |
2011 | if (dev_maps) { |
2012 | for_each_possible_cpu(cpu) { | |
2013 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2014 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2015 | if (map && map != new_map) | |
2016 | kfree_rcu(map, rcu); | |
2017 | } | |
537c00de | 2018 | |
01c5f864 | 2019 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
2020 | } |
2021 | ||
01c5f864 AD |
2022 | dev_maps = new_dev_maps; |
2023 | active = true; | |
537c00de | 2024 | |
01c5f864 AD |
2025 | out_no_new_maps: |
2026 | /* update Tx queue numa node */ | |
537c00de AD |
2027 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2028 | (numa_node_id >= 0) ? numa_node_id : | |
2029 | NUMA_NO_NODE); | |
2030 | ||
01c5f864 AD |
2031 | if (!dev_maps) |
2032 | goto out_no_maps; | |
2033 | ||
2034 | /* removes queue from unused CPUs */ | |
2035 | for_each_possible_cpu(cpu) { | |
2036 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
2037 | continue; | |
2038 | ||
2039 | if (remove_xps_queue(dev_maps, cpu, index)) | |
2040 | active = true; | |
2041 | } | |
2042 | ||
2043 | /* free map if not active */ | |
2044 | if (!active) { | |
2045 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2046 | kfree_rcu(dev_maps, rcu); | |
2047 | } | |
2048 | ||
2049 | out_no_maps: | |
537c00de AD |
2050 | mutex_unlock(&xps_map_mutex); |
2051 | ||
2052 | return 0; | |
2053 | error: | |
01c5f864 AD |
2054 | /* remove any maps that we added */ |
2055 | for_each_possible_cpu(cpu) { | |
2056 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2057 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2058 | NULL; | |
2059 | if (new_map && new_map != map) | |
2060 | kfree(new_map); | |
2061 | } | |
2062 | ||
537c00de AD |
2063 | mutex_unlock(&xps_map_mutex); |
2064 | ||
537c00de AD |
2065 | kfree(new_dev_maps); |
2066 | return -ENOMEM; | |
2067 | } | |
2068 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2069 | ||
2070 | #endif | |
f0796d5c JF |
2071 | /* |
2072 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2073 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2074 | */ | |
e6484930 | 2075 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2076 | { |
1d24eb48 TH |
2077 | int rc; |
2078 | ||
e6484930 TH |
2079 | if (txq < 1 || txq > dev->num_tx_queues) |
2080 | return -EINVAL; | |
f0796d5c | 2081 | |
5c56580b BH |
2082 | if (dev->reg_state == NETREG_REGISTERED || |
2083 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2084 | ASSERT_RTNL(); |
2085 | ||
1d24eb48 TH |
2086 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2087 | txq); | |
bf264145 TH |
2088 | if (rc) |
2089 | return rc; | |
2090 | ||
4f57c087 JF |
2091 | if (dev->num_tc) |
2092 | netif_setup_tc(dev, txq); | |
2093 | ||
024e9679 | 2094 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2095 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2096 | #ifdef CONFIG_XPS |
2097 | netif_reset_xps_queues_gt(dev, txq); | |
2098 | #endif | |
2099 | } | |
f0796d5c | 2100 | } |
e6484930 TH |
2101 | |
2102 | dev->real_num_tx_queues = txq; | |
2103 | return 0; | |
f0796d5c JF |
2104 | } |
2105 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2106 | |
a953be53 | 2107 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2108 | /** |
2109 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2110 | * @dev: Network device | |
2111 | * @rxq: Actual number of RX queues | |
2112 | * | |
2113 | * This must be called either with the rtnl_lock held or before | |
2114 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2115 | * negative error code. If called before registration, it always |
2116 | * succeeds. | |
62fe0b40 BH |
2117 | */ |
2118 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2119 | { | |
2120 | int rc; | |
2121 | ||
bd25fa7b TH |
2122 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2123 | return -EINVAL; | |
2124 | ||
62fe0b40 BH |
2125 | if (dev->reg_state == NETREG_REGISTERED) { |
2126 | ASSERT_RTNL(); | |
2127 | ||
62fe0b40 BH |
2128 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2129 | rxq); | |
2130 | if (rc) | |
2131 | return rc; | |
62fe0b40 BH |
2132 | } |
2133 | ||
2134 | dev->real_num_rx_queues = rxq; | |
2135 | return 0; | |
2136 | } | |
2137 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2138 | #endif | |
2139 | ||
2c53040f BH |
2140 | /** |
2141 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2142 | * |
2143 | * This routine should set an upper limit on the number of RSS queues | |
2144 | * used by default by multiqueue devices. | |
2145 | */ | |
a55b138b | 2146 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
2147 | { |
2148 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
2149 | } | |
2150 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2151 | ||
def82a1d | 2152 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 2153 | { |
def82a1d JP |
2154 | struct softnet_data *sd; |
2155 | unsigned long flags; | |
56079431 | 2156 | |
def82a1d JP |
2157 | local_irq_save(flags); |
2158 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
2159 | q->next_sched = NULL; |
2160 | *sd->output_queue_tailp = q; | |
2161 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2162 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2163 | local_irq_restore(flags); | |
2164 | } | |
2165 | ||
2166 | void __netif_schedule(struct Qdisc *q) | |
2167 | { | |
2168 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2169 | __netif_reschedule(q); | |
56079431 DV |
2170 | } |
2171 | EXPORT_SYMBOL(__netif_schedule); | |
2172 | ||
e6247027 ED |
2173 | struct dev_kfree_skb_cb { |
2174 | enum skb_free_reason reason; | |
2175 | }; | |
2176 | ||
2177 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2178 | { |
e6247027 ED |
2179 | return (struct dev_kfree_skb_cb *)skb->cb; |
2180 | } | |
2181 | ||
46e5da40 JF |
2182 | void netif_schedule_queue(struct netdev_queue *txq) |
2183 | { | |
2184 | rcu_read_lock(); | |
2185 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2186 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2187 | ||
2188 | __netif_schedule(q); | |
2189 | } | |
2190 | rcu_read_unlock(); | |
2191 | } | |
2192 | EXPORT_SYMBOL(netif_schedule_queue); | |
2193 | ||
2194 | /** | |
2195 | * netif_wake_subqueue - allow sending packets on subqueue | |
2196 | * @dev: network device | |
2197 | * @queue_index: sub queue index | |
2198 | * | |
2199 | * Resume individual transmit queue of a device with multiple transmit queues. | |
2200 | */ | |
2201 | void netif_wake_subqueue(struct net_device *dev, u16 queue_index) | |
2202 | { | |
2203 | struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index); | |
2204 | ||
2205 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) { | |
2206 | struct Qdisc *q; | |
2207 | ||
2208 | rcu_read_lock(); | |
2209 | q = rcu_dereference(txq->qdisc); | |
2210 | __netif_schedule(q); | |
2211 | rcu_read_unlock(); | |
2212 | } | |
2213 | } | |
2214 | EXPORT_SYMBOL(netif_wake_subqueue); | |
2215 | ||
2216 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) | |
2217 | { | |
2218 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2219 | struct Qdisc *q; | |
2220 | ||
2221 | rcu_read_lock(); | |
2222 | q = rcu_dereference(dev_queue->qdisc); | |
2223 | __netif_schedule(q); | |
2224 | rcu_read_unlock(); | |
2225 | } | |
2226 | } | |
2227 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2228 | ||
e6247027 | 2229 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2230 | { |
e6247027 | 2231 | unsigned long flags; |
56079431 | 2232 | |
e6247027 ED |
2233 | if (likely(atomic_read(&skb->users) == 1)) { |
2234 | smp_rmb(); | |
2235 | atomic_set(&skb->users, 0); | |
2236 | } else if (likely(!atomic_dec_and_test(&skb->users))) { | |
2237 | return; | |
bea3348e | 2238 | } |
e6247027 ED |
2239 | get_kfree_skb_cb(skb)->reason = reason; |
2240 | local_irq_save(flags); | |
2241 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2242 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2243 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2244 | local_irq_restore(flags); | |
56079431 | 2245 | } |
e6247027 | 2246 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2247 | |
e6247027 | 2248 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2249 | { |
2250 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2251 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2252 | else |
2253 | dev_kfree_skb(skb); | |
2254 | } | |
e6247027 | 2255 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2256 | |
2257 | ||
bea3348e SH |
2258 | /** |
2259 | * netif_device_detach - mark device as removed | |
2260 | * @dev: network device | |
2261 | * | |
2262 | * Mark device as removed from system and therefore no longer available. | |
2263 | */ | |
56079431 DV |
2264 | void netif_device_detach(struct net_device *dev) |
2265 | { | |
2266 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2267 | netif_running(dev)) { | |
d543103a | 2268 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2269 | } |
2270 | } | |
2271 | EXPORT_SYMBOL(netif_device_detach); | |
2272 | ||
bea3348e SH |
2273 | /** |
2274 | * netif_device_attach - mark device as attached | |
2275 | * @dev: network device | |
2276 | * | |
2277 | * Mark device as attached from system and restart if needed. | |
2278 | */ | |
56079431 DV |
2279 | void netif_device_attach(struct net_device *dev) |
2280 | { | |
2281 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2282 | netif_running(dev)) { | |
d543103a | 2283 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2284 | __netdev_watchdog_up(dev); |
56079431 DV |
2285 | } |
2286 | } | |
2287 | EXPORT_SYMBOL(netif_device_attach); | |
2288 | ||
36c92474 BH |
2289 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2290 | { | |
65e9d2fa | 2291 | static const netdev_features_t null_features = 0; |
36c92474 BH |
2292 | struct net_device *dev = skb->dev; |
2293 | const char *driver = ""; | |
2294 | ||
c846ad9b BG |
2295 | if (!net_ratelimit()) |
2296 | return; | |
2297 | ||
36c92474 BH |
2298 | if (dev && dev->dev.parent) |
2299 | driver = dev_driver_string(dev->dev.parent); | |
2300 | ||
2301 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " | |
2302 | "gso_type=%d ip_summed=%d\n", | |
65e9d2fa MM |
2303 | driver, dev ? &dev->features : &null_features, |
2304 | skb->sk ? &skb->sk->sk_route_caps : &null_features, | |
36c92474 BH |
2305 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2306 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2307 | } | |
2308 | ||
1da177e4 LT |
2309 | /* |
2310 | * Invalidate hardware checksum when packet is to be mangled, and | |
2311 | * complete checksum manually on outgoing path. | |
2312 | */ | |
84fa7933 | 2313 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2314 | { |
d3bc23e7 | 2315 | __wsum csum; |
663ead3b | 2316 | int ret = 0, offset; |
1da177e4 | 2317 | |
84fa7933 | 2318 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2319 | goto out_set_summed; |
2320 | ||
2321 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2322 | skb_warn_bad_offload(skb); |
2323 | return -EINVAL; | |
1da177e4 LT |
2324 | } |
2325 | ||
cef401de ED |
2326 | /* Before computing a checksum, we should make sure no frag could |
2327 | * be modified by an external entity : checksum could be wrong. | |
2328 | */ | |
2329 | if (skb_has_shared_frag(skb)) { | |
2330 | ret = __skb_linearize(skb); | |
2331 | if (ret) | |
2332 | goto out; | |
2333 | } | |
2334 | ||
55508d60 | 2335 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2336 | BUG_ON(offset >= skb_headlen(skb)); |
2337 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2338 | ||
2339 | offset += skb->csum_offset; | |
2340 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2341 | ||
2342 | if (skb_cloned(skb) && | |
2343 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2344 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2345 | if (ret) | |
2346 | goto out; | |
2347 | } | |
2348 | ||
a030847e | 2349 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2350 | out_set_summed: |
1da177e4 | 2351 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2352 | out: |
1da177e4 LT |
2353 | return ret; |
2354 | } | |
d1b19dff | 2355 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2356 | |
53d6471c | 2357 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2358 | { |
4b9b1cdf | 2359 | unsigned int vlan_depth = skb->mac_len; |
252e3346 | 2360 | __be16 type = skb->protocol; |
f6a78bfc | 2361 | |
19acc327 PS |
2362 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2363 | if (type == htons(ETH_P_TEB)) { | |
2364 | struct ethhdr *eth; | |
2365 | ||
2366 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2367 | return 0; | |
2368 | ||
2369 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2370 | type = eth->h_proto; | |
2371 | } | |
2372 | ||
4b9b1cdf NA |
2373 | /* if skb->protocol is 802.1Q/AD then the header should already be |
2374 | * present at mac_len - VLAN_HLEN (if mac_len > 0), or at | |
2375 | * ETH_HLEN otherwise | |
2376 | */ | |
2377 | if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) { | |
2378 | if (vlan_depth) { | |
80019d31 | 2379 | if (WARN_ON(vlan_depth < VLAN_HLEN)) |
4b9b1cdf NA |
2380 | return 0; |
2381 | vlan_depth -= VLAN_HLEN; | |
2382 | } else { | |
2383 | vlan_depth = ETH_HLEN; | |
2384 | } | |
2385 | do { | |
2386 | struct vlan_hdr *vh; | |
2387 | ||
2388 | if (unlikely(!pskb_may_pull(skb, | |
2389 | vlan_depth + VLAN_HLEN))) | |
2390 | return 0; | |
2391 | ||
2392 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); | |
2393 | type = vh->h_vlan_encapsulated_proto; | |
2394 | vlan_depth += VLAN_HLEN; | |
2395 | } while (type == htons(ETH_P_8021Q) || | |
2396 | type == htons(ETH_P_8021AD)); | |
7b9c6090 JG |
2397 | } |
2398 | ||
53d6471c VY |
2399 | *depth = vlan_depth; |
2400 | ||
ec5f0615 PS |
2401 | return type; |
2402 | } | |
2403 | ||
2404 | /** | |
2405 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2406 | * @skb: buffer to segment | |
2407 | * @features: features for the output path (see dev->features) | |
2408 | */ | |
2409 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2410 | netdev_features_t features) | |
2411 | { | |
2412 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2413 | struct packet_offload *ptype; | |
53d6471c VY |
2414 | int vlan_depth = skb->mac_len; |
2415 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2416 | |
2417 | if (unlikely(!type)) | |
2418 | return ERR_PTR(-EINVAL); | |
2419 | ||
53d6471c | 2420 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2421 | |
2422 | rcu_read_lock(); | |
22061d80 | 2423 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2424 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2425 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2426 | break; |
2427 | } | |
2428 | } | |
2429 | rcu_read_unlock(); | |
2430 | ||
98e399f8 | 2431 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2432 | |
f6a78bfc HX |
2433 | return segs; |
2434 | } | |
05e8ef4a PS |
2435 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2436 | ||
2437 | ||
2438 | /* openvswitch calls this on rx path, so we need a different check. | |
2439 | */ | |
2440 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2441 | { | |
2442 | if (tx_path) | |
2443 | return skb->ip_summed != CHECKSUM_PARTIAL; | |
2444 | else | |
2445 | return skb->ip_summed == CHECKSUM_NONE; | |
2446 | } | |
2447 | ||
2448 | /** | |
2449 | * __skb_gso_segment - Perform segmentation on skb. | |
2450 | * @skb: buffer to segment | |
2451 | * @features: features for the output path (see dev->features) | |
2452 | * @tx_path: whether it is called in TX path | |
2453 | * | |
2454 | * This function segments the given skb and returns a list of segments. | |
2455 | * | |
2456 | * It may return NULL if the skb requires no segmentation. This is | |
2457 | * only possible when GSO is used for verifying header integrity. | |
2458 | */ | |
2459 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2460 | netdev_features_t features, bool tx_path) | |
2461 | { | |
2462 | if (unlikely(skb_needs_check(skb, tx_path))) { | |
2463 | int err; | |
2464 | ||
2465 | skb_warn_bad_offload(skb); | |
2466 | ||
a40e0a66 | 2467 | err = skb_cow_head(skb, 0); |
2468 | if (err < 0) | |
05e8ef4a PS |
2469 | return ERR_PTR(err); |
2470 | } | |
2471 | ||
68c33163 | 2472 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2473 | SKB_GSO_CB(skb)->encap_level = 0; |
2474 | ||
05e8ef4a PS |
2475 | skb_reset_mac_header(skb); |
2476 | skb_reset_mac_len(skb); | |
2477 | ||
2478 | return skb_mac_gso_segment(skb, features); | |
2479 | } | |
12b0004d | 2480 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2481 | |
fb286bb2 HX |
2482 | /* Take action when hardware reception checksum errors are detected. */ |
2483 | #ifdef CONFIG_BUG | |
2484 | void netdev_rx_csum_fault(struct net_device *dev) | |
2485 | { | |
2486 | if (net_ratelimit()) { | |
7b6cd1ce | 2487 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2488 | dump_stack(); |
2489 | } | |
2490 | } | |
2491 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2492 | #endif | |
2493 | ||
1da177e4 LT |
2494 | /* Actually, we should eliminate this check as soon as we know, that: |
2495 | * 1. IOMMU is present and allows to map all the memory. | |
2496 | * 2. No high memory really exists on this machine. | |
2497 | */ | |
2498 | ||
c1e756bf | 2499 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2500 | { |
3d3a8533 | 2501 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2502 | int i; |
5acbbd42 | 2503 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2504 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2505 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2506 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2507 | return 1; |
ea2ab693 | 2508 | } |
5acbbd42 | 2509 | } |
1da177e4 | 2510 | |
5acbbd42 FT |
2511 | if (PCI_DMA_BUS_IS_PHYS) { |
2512 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2513 | |
9092c658 ED |
2514 | if (!pdev) |
2515 | return 0; | |
5acbbd42 | 2516 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2517 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2518 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2519 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2520 | return 1; | |
2521 | } | |
2522 | } | |
3d3a8533 | 2523 | #endif |
1da177e4 LT |
2524 | return 0; |
2525 | } | |
1da177e4 | 2526 | |
3b392ddb SH |
2527 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2528 | * instead of standard features for the netdev. | |
2529 | */ | |
2530 | #ifdef CONFIG_NET_MPLS_GSO | |
2531 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2532 | netdev_features_t features, | |
2533 | __be16 type) | |
2534 | { | |
2535 | if (type == htons(ETH_P_MPLS_UC) || type == htons(ETH_P_MPLS_MC)) | |
2536 | features &= skb->dev->mpls_features; | |
2537 | ||
2538 | return features; | |
2539 | } | |
2540 | #else | |
2541 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2542 | netdev_features_t features, | |
2543 | __be16 type) | |
2544 | { | |
2545 | return features; | |
2546 | } | |
2547 | #endif | |
2548 | ||
c8f44aff | 2549 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2550 | netdev_features_t features) |
f01a5236 | 2551 | { |
53d6471c | 2552 | int tmp; |
3b392ddb SH |
2553 | __be16 type; |
2554 | ||
2555 | type = skb_network_protocol(skb, &tmp); | |
2556 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2557 | |
c0d680e5 | 2558 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2559 | !can_checksum_protocol(features, type)) { |
f01a5236 | 2560 | features &= ~NETIF_F_ALL_CSUM; |
c1e756bf | 2561 | } else if (illegal_highdma(skb->dev, skb)) { |
f01a5236 JG |
2562 | features &= ~NETIF_F_SG; |
2563 | } | |
2564 | ||
2565 | return features; | |
2566 | } | |
2567 | ||
c1e756bf | 2568 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 JG |
2569 | { |
2570 | __be16 protocol = skb->protocol; | |
c1e756bf | 2571 | netdev_features_t features = skb->dev->features; |
58e998c6 | 2572 | |
c1e756bf | 2573 | if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs) |
30b678d8 BH |
2574 | features &= ~NETIF_F_GSO_MASK; |
2575 | ||
8ad227ff | 2576 | if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) { |
58e998c6 JG |
2577 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; |
2578 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 | 2579 | } else if (!vlan_tx_tag_present(skb)) { |
c1e756bf | 2580 | return harmonize_features(skb, features); |
f01a5236 | 2581 | } |
58e998c6 | 2582 | |
db115037 MK |
2583 | features = netdev_intersect_features(features, |
2584 | skb->dev->vlan_features | | |
2585 | NETIF_F_HW_VLAN_CTAG_TX | | |
2586 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2587 | |
cdbaa0bb | 2588 | if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) |
db115037 MK |
2589 | features = netdev_intersect_features(features, |
2590 | NETIF_F_SG | | |
2591 | NETIF_F_HIGHDMA | | |
2592 | NETIF_F_FRAGLIST | | |
2593 | NETIF_F_GEN_CSUM | | |
2594 | NETIF_F_HW_VLAN_CTAG_TX | | |
2595 | NETIF_F_HW_VLAN_STAG_TX); | |
cdbaa0bb | 2596 | |
c1e756bf | 2597 | return harmonize_features(skb, features); |
58e998c6 | 2598 | } |
c1e756bf | 2599 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2600 | |
2ea25513 | 2601 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2602 | struct netdev_queue *txq, bool more) |
2ea25513 DM |
2603 | { |
2604 | unsigned int len; | |
2605 | int rc; | |
2606 | ||
2607 | if (!list_empty(&ptype_all)) | |
2608 | dev_queue_xmit_nit(skb, dev); | |
2609 | ||
2610 | len = skb->len; | |
2611 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 2612 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 DM |
2613 | trace_net_dev_xmit(skb, rc, dev, len); |
2614 | ||
2615 | return rc; | |
2616 | } | |
2617 | ||
8dcda22a DM |
2618 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
2619 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
2620 | { |
2621 | struct sk_buff *skb = first; | |
2622 | int rc = NETDEV_TX_OK; | |
2623 | ||
2624 | while (skb) { | |
2625 | struct sk_buff *next = skb->next; | |
2626 | ||
2627 | skb->next = NULL; | |
95f6b3dd | 2628 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
2629 | if (unlikely(!dev_xmit_complete(rc))) { |
2630 | skb->next = next; | |
2631 | goto out; | |
2632 | } | |
2633 | ||
2634 | skb = next; | |
2635 | if (netif_xmit_stopped(txq) && skb) { | |
2636 | rc = NETDEV_TX_BUSY; | |
2637 | break; | |
2638 | } | |
2639 | } | |
2640 | ||
2641 | out: | |
2642 | *ret = rc; | |
2643 | return skb; | |
2644 | } | |
2645 | ||
eae3f88e | 2646 | struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, netdev_features_t features) |
f6a78bfc | 2647 | { |
eae3f88e DM |
2648 | if (vlan_tx_tag_present(skb) && |
2649 | !vlan_hw_offload_capable(features, skb->vlan_proto)) { | |
2650 | skb = __vlan_put_tag(skb, skb->vlan_proto, | |
2651 | vlan_tx_tag_get(skb)); | |
2652 | if (skb) | |
2653 | skb->vlan_tci = 0; | |
2654 | } | |
2655 | return skb; | |
2656 | } | |
00829823 | 2657 | |
55a93b3e | 2658 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
2659 | { |
2660 | netdev_features_t features; | |
fc741216 | 2661 | |
eae3f88e DM |
2662 | if (skb->next) |
2663 | return skb; | |
adf30907 | 2664 | |
eae3f88e DM |
2665 | /* If device doesn't need skb->dst, release it right now while |
2666 | * its hot in this cpu cache | |
2667 | */ | |
2668 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2669 | skb_dst_drop(skb); | |
fc741216 | 2670 | |
eae3f88e DM |
2671 | features = netif_skb_features(skb); |
2672 | skb = validate_xmit_vlan(skb, features); | |
2673 | if (unlikely(!skb)) | |
2674 | goto out_null; | |
7b9c6090 | 2675 | |
eae3f88e DM |
2676 | /* If encapsulation offload request, verify we are testing |
2677 | * hardware encapsulation features instead of standard | |
2678 | * features for the netdev | |
2679 | */ | |
2680 | if (skb->encapsulation) | |
2681 | features &= dev->hw_enc_features; | |
2682 | ||
2683 | if (netif_needs_gso(skb, features)) { | |
ce93718f DM |
2684 | struct sk_buff *segs; |
2685 | ||
2686 | segs = skb_gso_segment(skb, features); | |
cecda693 | 2687 | if (IS_ERR(segs)) { |
ce93718f | 2688 | segs = NULL; |
cecda693 JW |
2689 | } else if (segs) { |
2690 | consume_skb(skb); | |
2691 | skb = segs; | |
2692 | } | |
eae3f88e DM |
2693 | } else { |
2694 | if (skb_needs_linearize(skb, features) && | |
2695 | __skb_linearize(skb)) | |
2696 | goto out_kfree_skb; | |
2697 | ||
2698 | /* If packet is not checksummed and device does not | |
2699 | * support checksumming for this protocol, complete | |
2700 | * checksumming here. | |
2701 | */ | |
2702 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
2703 | if (skb->encapsulation) | |
2704 | skb_set_inner_transport_header(skb, | |
2705 | skb_checksum_start_offset(skb)); | |
2706 | else | |
2707 | skb_set_transport_header(skb, | |
2708 | skb_checksum_start_offset(skb)); | |
2709 | if (!(features & NETIF_F_ALL_CSUM) && | |
2710 | skb_checksum_help(skb)) | |
2711 | goto out_kfree_skb; | |
7b9c6090 | 2712 | } |
eae3f88e | 2713 | } |
7b9c6090 | 2714 | |
eae3f88e | 2715 | return skb; |
fc70fb64 | 2716 | |
eae3f88e DM |
2717 | out_kfree_skb: |
2718 | kfree_skb(skb); | |
2719 | out_null: | |
2720 | return NULL; | |
2721 | } | |
6afff0ca | 2722 | |
55a93b3e ED |
2723 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
2724 | { | |
2725 | struct sk_buff *next, *head = NULL, *tail; | |
2726 | ||
2727 | while (skb) { | |
2728 | next = skb->next; | |
2729 | skb->next = NULL; | |
2730 | skb = validate_xmit_skb(skb, dev); | |
2731 | if (skb) { | |
2732 | struct sk_buff *end = skb; | |
2733 | ||
2734 | while (end->next) | |
2735 | end = end->next; | |
2736 | if (!head) | |
2737 | head = skb; | |
2738 | else | |
2739 | tail->next = skb; | |
2740 | tail = end; | |
2741 | } | |
2742 | skb = next; | |
2743 | } | |
2744 | return head; | |
2745 | } | |
2746 | ||
1def9238 ED |
2747 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
2748 | { | |
2749 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2750 | ||
2751 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2752 | ||
2753 | /* To get more precise estimation of bytes sent on wire, | |
2754 | * we add to pkt_len the headers size of all segments | |
2755 | */ | |
2756 | if (shinfo->gso_size) { | |
757b8b1d | 2757 | unsigned int hdr_len; |
15e5a030 | 2758 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 2759 | |
757b8b1d ED |
2760 | /* mac layer + network layer */ |
2761 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
2762 | ||
2763 | /* + transport layer */ | |
1def9238 ED |
2764 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
2765 | hdr_len += tcp_hdrlen(skb); | |
2766 | else | |
2767 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
2768 | |
2769 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
2770 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
2771 | shinfo->gso_size); | |
2772 | ||
2773 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
2774 | } |
2775 | } | |
2776 | ||
bbd8a0d3 KK |
2777 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2778 | struct net_device *dev, | |
2779 | struct netdev_queue *txq) | |
2780 | { | |
2781 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2782 | bool contended; |
bbd8a0d3 KK |
2783 | int rc; |
2784 | ||
1def9238 | 2785 | qdisc_pkt_len_init(skb); |
a2da570d | 2786 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
2787 | /* |
2788 | * Heuristic to force contended enqueues to serialize on a | |
2789 | * separate lock before trying to get qdisc main lock. | |
9bf2b8c2 YX |
2790 | * This permits __QDISC___STATE_RUNNING owner to get the lock more |
2791 | * often and dequeue packets faster. | |
79640a4c | 2792 | */ |
a2da570d | 2793 | contended = qdisc_is_running(q); |
79640a4c ED |
2794 | if (unlikely(contended)) |
2795 | spin_lock(&q->busylock); | |
2796 | ||
bbd8a0d3 KK |
2797 | spin_lock(root_lock); |
2798 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2799 | kfree_skb(skb); | |
2800 | rc = NET_XMIT_DROP; | |
2801 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2802 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2803 | /* |
2804 | * This is a work-conserving queue; there are no old skbs | |
2805 | * waiting to be sent out; and the qdisc is not running - | |
2806 | * xmit the skb directly. | |
2807 | */ | |
7fee226a ED |
2808 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2809 | skb_dst_force(skb); | |
bfe0d029 | 2810 | |
bfe0d029 ED |
2811 | qdisc_bstats_update(q, skb); |
2812 | ||
55a93b3e | 2813 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
2814 | if (unlikely(contended)) { |
2815 | spin_unlock(&q->busylock); | |
2816 | contended = false; | |
2817 | } | |
bbd8a0d3 | 2818 | __qdisc_run(q); |
79640a4c | 2819 | } else |
bc135b23 | 2820 | qdisc_run_end(q); |
bbd8a0d3 KK |
2821 | |
2822 | rc = NET_XMIT_SUCCESS; | |
2823 | } else { | |
7fee226a | 2824 | skb_dst_force(skb); |
a2da570d | 2825 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2826 | if (qdisc_run_begin(q)) { |
2827 | if (unlikely(contended)) { | |
2828 | spin_unlock(&q->busylock); | |
2829 | contended = false; | |
2830 | } | |
2831 | __qdisc_run(q); | |
2832 | } | |
bbd8a0d3 KK |
2833 | } |
2834 | spin_unlock(root_lock); | |
79640a4c ED |
2835 | if (unlikely(contended)) |
2836 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2837 | return rc; |
2838 | } | |
2839 | ||
86f8515f | 2840 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
2841 | static void skb_update_prio(struct sk_buff *skb) |
2842 | { | |
6977a79d | 2843 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 2844 | |
91c68ce2 ED |
2845 | if (!skb->priority && skb->sk && map) { |
2846 | unsigned int prioidx = skb->sk->sk_cgrp_prioidx; | |
2847 | ||
2848 | if (prioidx < map->priomap_len) | |
2849 | skb->priority = map->priomap[prioidx]; | |
2850 | } | |
5bc1421e NH |
2851 | } |
2852 | #else | |
2853 | #define skb_update_prio(skb) | |
2854 | #endif | |
2855 | ||
745e20f1 | 2856 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2857 | #define RECURSION_LIMIT 10 |
745e20f1 | 2858 | |
95603e22 MM |
2859 | /** |
2860 | * dev_loopback_xmit - loop back @skb | |
2861 | * @skb: buffer to transmit | |
2862 | */ | |
2863 | int dev_loopback_xmit(struct sk_buff *skb) | |
2864 | { | |
2865 | skb_reset_mac_header(skb); | |
2866 | __skb_pull(skb, skb_network_offset(skb)); | |
2867 | skb->pkt_type = PACKET_LOOPBACK; | |
2868 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2869 | WARN_ON(!skb_dst(skb)); | |
2870 | skb_dst_force(skb); | |
2871 | netif_rx_ni(skb); | |
2872 | return 0; | |
2873 | } | |
2874 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2875 | ||
d29f749e | 2876 | /** |
9d08dd3d | 2877 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 2878 | * @skb: buffer to transmit |
9d08dd3d | 2879 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
2880 | * |
2881 | * Queue a buffer for transmission to a network device. The caller must | |
2882 | * have set the device and priority and built the buffer before calling | |
2883 | * this function. The function can be called from an interrupt. | |
2884 | * | |
2885 | * A negative errno code is returned on a failure. A success does not | |
2886 | * guarantee the frame will be transmitted as it may be dropped due | |
2887 | * to congestion or traffic shaping. | |
2888 | * | |
2889 | * ----------------------------------------------------------------------------------- | |
2890 | * I notice this method can also return errors from the queue disciplines, | |
2891 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2892 | * be positive. | |
2893 | * | |
2894 | * Regardless of the return value, the skb is consumed, so it is currently | |
2895 | * difficult to retry a send to this method. (You can bump the ref count | |
2896 | * before sending to hold a reference for retry if you are careful.) | |
2897 | * | |
2898 | * When calling this method, interrupts MUST be enabled. This is because | |
2899 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2900 | * --BLG | |
2901 | */ | |
0a59f3a9 | 2902 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
2903 | { |
2904 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2905 | struct netdev_queue *txq; |
1da177e4 LT |
2906 | struct Qdisc *q; |
2907 | int rc = -ENOMEM; | |
2908 | ||
6d1ccff6 ED |
2909 | skb_reset_mac_header(skb); |
2910 | ||
e7fd2885 WB |
2911 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
2912 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
2913 | ||
4ec93edb YH |
2914 | /* Disable soft irqs for various locks below. Also |
2915 | * stops preemption for RCU. | |
1da177e4 | 2916 | */ |
4ec93edb | 2917 | rcu_read_lock_bh(); |
1da177e4 | 2918 | |
5bc1421e NH |
2919 | skb_update_prio(skb); |
2920 | ||
f663dd9a | 2921 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 2922 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2923 | |
1da177e4 | 2924 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2925 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2926 | #endif |
cf66ba58 | 2927 | trace_net_dev_queue(skb); |
1da177e4 | 2928 | if (q->enqueue) { |
bbd8a0d3 | 2929 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2930 | goto out; |
1da177e4 LT |
2931 | } |
2932 | ||
2933 | /* The device has no queue. Common case for software devices: | |
2934 | loopback, all the sorts of tunnels... | |
2935 | ||
932ff279 HX |
2936 | Really, it is unlikely that netif_tx_lock protection is necessary |
2937 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2938 | counters.) |
2939 | However, it is possible, that they rely on protection | |
2940 | made by us here. | |
2941 | ||
2942 | Check this and shot the lock. It is not prone from deadlocks. | |
2943 | Either shot noqueue qdisc, it is even simpler 8) | |
2944 | */ | |
2945 | if (dev->flags & IFF_UP) { | |
2946 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2947 | ||
c773e847 | 2948 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2949 | |
745e20f1 ED |
2950 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2951 | goto recursion_alert; | |
2952 | ||
1f59533f JDB |
2953 | skb = validate_xmit_skb(skb, dev); |
2954 | if (!skb) | |
2955 | goto drop; | |
2956 | ||
c773e847 | 2957 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2958 | |
73466498 | 2959 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 2960 | __this_cpu_inc(xmit_recursion); |
ce93718f | 2961 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 2962 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2963 | if (dev_xmit_complete(rc)) { |
c773e847 | 2964 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2965 | goto out; |
2966 | } | |
2967 | } | |
c773e847 | 2968 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
2969 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
2970 | dev->name); | |
1da177e4 LT |
2971 | } else { |
2972 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2973 | * unfortunately |
2974 | */ | |
2975 | recursion_alert: | |
e87cc472 JP |
2976 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
2977 | dev->name); | |
1da177e4 LT |
2978 | } |
2979 | } | |
2980 | ||
2981 | rc = -ENETDOWN; | |
1f59533f | 2982 | drop: |
d4828d85 | 2983 | rcu_read_unlock_bh(); |
1da177e4 | 2984 | |
015f0688 | 2985 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 2986 | kfree_skb_list(skb); |
1da177e4 LT |
2987 | return rc; |
2988 | out: | |
d4828d85 | 2989 | rcu_read_unlock_bh(); |
1da177e4 LT |
2990 | return rc; |
2991 | } | |
f663dd9a JW |
2992 | |
2993 | int dev_queue_xmit(struct sk_buff *skb) | |
2994 | { | |
2995 | return __dev_queue_xmit(skb, NULL); | |
2996 | } | |
d1b19dff | 2997 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 2998 | |
f663dd9a JW |
2999 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3000 | { | |
3001 | return __dev_queue_xmit(skb, accel_priv); | |
3002 | } | |
3003 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3004 | ||
1da177e4 LT |
3005 | |
3006 | /*======================================================================= | |
3007 | Receiver routines | |
3008 | =======================================================================*/ | |
3009 | ||
6b2bedc3 | 3010 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3011 | EXPORT_SYMBOL(netdev_max_backlog); |
3012 | ||
3b098e2d | 3013 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
3014 | int netdev_budget __read_mostly = 300; |
3015 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 3016 | |
eecfd7c4 ED |
3017 | /* Called with irq disabled */ |
3018 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3019 | struct napi_struct *napi) | |
3020 | { | |
3021 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3022 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3023 | } | |
3024 | ||
bfb564e7 KK |
3025 | #ifdef CONFIG_RPS |
3026 | ||
3027 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3028 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
3029 | EXPORT_SYMBOL(rps_sock_flow_table); |
3030 | ||
c5905afb | 3031 | struct static_key rps_needed __read_mostly; |
adc9300e | 3032 | |
c445477d BH |
3033 | static struct rps_dev_flow * |
3034 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3035 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3036 | { | |
09994d1b | 3037 | if (next_cpu != RPS_NO_CPU) { |
c445477d BH |
3038 | #ifdef CONFIG_RFS_ACCEL |
3039 | struct netdev_rx_queue *rxqueue; | |
3040 | struct rps_dev_flow_table *flow_table; | |
3041 | struct rps_dev_flow *old_rflow; | |
3042 | u32 flow_id; | |
3043 | u16 rxq_index; | |
3044 | int rc; | |
3045 | ||
3046 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3047 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3048 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3049 | goto out; |
3050 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3051 | if (rxq_index == skb_get_rx_queue(skb)) | |
3052 | goto out; | |
3053 | ||
3054 | rxqueue = dev->_rx + rxq_index; | |
3055 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3056 | if (!flow_table) | |
3057 | goto out; | |
61b905da | 3058 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3059 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3060 | rxq_index, flow_id); | |
3061 | if (rc < 0) | |
3062 | goto out; | |
3063 | old_rflow = rflow; | |
3064 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3065 | rflow->filter = rc; |
3066 | if (old_rflow->filter == rflow->filter) | |
3067 | old_rflow->filter = RPS_NO_FILTER; | |
3068 | out: | |
3069 | #endif | |
3070 | rflow->last_qtail = | |
09994d1b | 3071 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3072 | } |
3073 | ||
09994d1b | 3074 | rflow->cpu = next_cpu; |
c445477d BH |
3075 | return rflow; |
3076 | } | |
3077 | ||
bfb564e7 KK |
3078 | /* |
3079 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3080 | * CPU from the RPS map of the receiving queue for a given skb. | |
3081 | * rcu_read_lock must be held on entry. | |
3082 | */ | |
3083 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3084 | struct rps_dev_flow **rflowp) | |
3085 | { | |
3086 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 3087 | struct rps_map *map; |
bfb564e7 KK |
3088 | struct rps_dev_flow_table *flow_table; |
3089 | struct rps_sock_flow_table *sock_flow_table; | |
3090 | int cpu = -1; | |
3091 | u16 tcpu; | |
61b905da | 3092 | u32 hash; |
bfb564e7 KK |
3093 | |
3094 | if (skb_rx_queue_recorded(skb)) { | |
3095 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
3096 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3097 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3098 | "%s received packet on queue %u, but number " | |
3099 | "of RX queues is %u\n", | |
3100 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3101 | goto done; |
3102 | } | |
3103 | rxqueue = dev->_rx + index; | |
3104 | } else | |
3105 | rxqueue = dev->_rx; | |
3106 | ||
6e3f7faf ED |
3107 | map = rcu_dereference(rxqueue->rps_map); |
3108 | if (map) { | |
85875236 | 3109 | if (map->len == 1 && |
33d480ce | 3110 | !rcu_access_pointer(rxqueue->rps_flow_table)) { |
6febfca9 CG |
3111 | tcpu = map->cpus[0]; |
3112 | if (cpu_online(tcpu)) | |
3113 | cpu = tcpu; | |
3114 | goto done; | |
3115 | } | |
33d480ce | 3116 | } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) { |
bfb564e7 | 3117 | goto done; |
6febfca9 | 3118 | } |
bfb564e7 | 3119 | |
2d47b459 | 3120 | skb_reset_network_header(skb); |
61b905da TH |
3121 | hash = skb_get_hash(skb); |
3122 | if (!hash) | |
bfb564e7 KK |
3123 | goto done; |
3124 | ||
fec5e652 TH |
3125 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
3126 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
3127 | if (flow_table && sock_flow_table) { | |
3128 | u16 next_cpu; | |
3129 | struct rps_dev_flow *rflow; | |
3130 | ||
61b905da | 3131 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3132 | tcpu = rflow->cpu; |
3133 | ||
61b905da | 3134 | next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask]; |
fec5e652 TH |
3135 | |
3136 | /* | |
3137 | * If the desired CPU (where last recvmsg was done) is | |
3138 | * different from current CPU (one in the rx-queue flow | |
3139 | * table entry), switch if one of the following holds: | |
3140 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
3141 | * - Current CPU is offline. | |
3142 | * - The current CPU's queue tail has advanced beyond the | |
3143 | * last packet that was enqueued using this table entry. | |
3144 | * This guarantees that all previous packets for the flow | |
3145 | * have been dequeued, thus preserving in order delivery. | |
3146 | */ | |
3147 | if (unlikely(tcpu != next_cpu) && | |
3148 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
3149 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
baefa31d TH |
3150 | rflow->last_qtail)) >= 0)) { |
3151 | tcpu = next_cpu; | |
c445477d | 3152 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3153 | } |
c445477d | 3154 | |
fec5e652 TH |
3155 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
3156 | *rflowp = rflow; | |
3157 | cpu = tcpu; | |
3158 | goto done; | |
3159 | } | |
3160 | } | |
3161 | ||
0a9627f2 | 3162 | if (map) { |
8fc54f68 | 3163 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3164 | if (cpu_online(tcpu)) { |
3165 | cpu = tcpu; | |
3166 | goto done; | |
3167 | } | |
3168 | } | |
3169 | ||
3170 | done: | |
0a9627f2 TH |
3171 | return cpu; |
3172 | } | |
3173 | ||
c445477d BH |
3174 | #ifdef CONFIG_RFS_ACCEL |
3175 | ||
3176 | /** | |
3177 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3178 | * @dev: Device on which the filter was set | |
3179 | * @rxq_index: RX queue index | |
3180 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3181 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3182 | * | |
3183 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3184 | * this function for each installed filter and remove the filters for | |
3185 | * which it returns %true. | |
3186 | */ | |
3187 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3188 | u32 flow_id, u16 filter_id) | |
3189 | { | |
3190 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3191 | struct rps_dev_flow_table *flow_table; | |
3192 | struct rps_dev_flow *rflow; | |
3193 | bool expire = true; | |
3194 | int cpu; | |
3195 | ||
3196 | rcu_read_lock(); | |
3197 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3198 | if (flow_table && flow_id <= flow_table->mask) { | |
3199 | rflow = &flow_table->flows[flow_id]; | |
3200 | cpu = ACCESS_ONCE(rflow->cpu); | |
3201 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
3202 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
3203 | rflow->last_qtail) < | |
3204 | (int)(10 * flow_table->mask))) | |
3205 | expire = false; | |
3206 | } | |
3207 | rcu_read_unlock(); | |
3208 | return expire; | |
3209 | } | |
3210 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3211 | ||
3212 | #endif /* CONFIG_RFS_ACCEL */ | |
3213 | ||
0a9627f2 | 3214 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3215 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3216 | { |
e36fa2f7 ED |
3217 | struct softnet_data *sd = data; |
3218 | ||
eecfd7c4 | 3219 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3220 | sd->received_rps++; |
0a9627f2 | 3221 | } |
e36fa2f7 | 3222 | |
fec5e652 | 3223 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3224 | |
e36fa2f7 ED |
3225 | /* |
3226 | * Check if this softnet_data structure is another cpu one | |
3227 | * If yes, queue it to our IPI list and return 1 | |
3228 | * If no, return 0 | |
3229 | */ | |
3230 | static int rps_ipi_queued(struct softnet_data *sd) | |
3231 | { | |
3232 | #ifdef CONFIG_RPS | |
3233 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
3234 | ||
3235 | if (sd != mysd) { | |
3236 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3237 | mysd->rps_ipi_list = sd; | |
3238 | ||
3239 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3240 | return 1; | |
3241 | } | |
3242 | #endif /* CONFIG_RPS */ | |
3243 | return 0; | |
3244 | } | |
3245 | ||
99bbc707 WB |
3246 | #ifdef CONFIG_NET_FLOW_LIMIT |
3247 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3248 | #endif | |
3249 | ||
3250 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3251 | { | |
3252 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3253 | struct sd_flow_limit *fl; | |
3254 | struct softnet_data *sd; | |
3255 | unsigned int old_flow, new_flow; | |
3256 | ||
3257 | if (qlen < (netdev_max_backlog >> 1)) | |
3258 | return false; | |
3259 | ||
3260 | sd = &__get_cpu_var(softnet_data); | |
3261 | ||
3262 | rcu_read_lock(); | |
3263 | fl = rcu_dereference(sd->flow_limit); | |
3264 | if (fl) { | |
3958afa1 | 3265 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3266 | old_flow = fl->history[fl->history_head]; |
3267 | fl->history[fl->history_head] = new_flow; | |
3268 | ||
3269 | fl->history_head++; | |
3270 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3271 | ||
3272 | if (likely(fl->buckets[old_flow])) | |
3273 | fl->buckets[old_flow]--; | |
3274 | ||
3275 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3276 | fl->count++; | |
3277 | rcu_read_unlock(); | |
3278 | return true; | |
3279 | } | |
3280 | } | |
3281 | rcu_read_unlock(); | |
3282 | #endif | |
3283 | return false; | |
3284 | } | |
3285 | ||
0a9627f2 TH |
3286 | /* |
3287 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3288 | * queue (may be a remote CPU queue). | |
3289 | */ | |
fec5e652 TH |
3290 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3291 | unsigned int *qtail) | |
0a9627f2 | 3292 | { |
e36fa2f7 | 3293 | struct softnet_data *sd; |
0a9627f2 | 3294 | unsigned long flags; |
99bbc707 | 3295 | unsigned int qlen; |
0a9627f2 | 3296 | |
e36fa2f7 | 3297 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3298 | |
3299 | local_irq_save(flags); | |
0a9627f2 | 3300 | |
e36fa2f7 | 3301 | rps_lock(sd); |
99bbc707 WB |
3302 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3303 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
6e7676c1 | 3304 | if (skb_queue_len(&sd->input_pkt_queue)) { |
0a9627f2 | 3305 | enqueue: |
e36fa2f7 | 3306 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3307 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3308 | rps_unlock(sd); |
152102c7 | 3309 | local_irq_restore(flags); |
0a9627f2 TH |
3310 | return NET_RX_SUCCESS; |
3311 | } | |
3312 | ||
ebda37c2 ED |
3313 | /* Schedule NAPI for backlog device |
3314 | * We can use non atomic operation since we own the queue lock | |
3315 | */ | |
3316 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3317 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3318 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3319 | } |
3320 | goto enqueue; | |
3321 | } | |
3322 | ||
dee42870 | 3323 | sd->dropped++; |
e36fa2f7 | 3324 | rps_unlock(sd); |
0a9627f2 | 3325 | |
0a9627f2 TH |
3326 | local_irq_restore(flags); |
3327 | ||
caf586e5 | 3328 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3329 | kfree_skb(skb); |
3330 | return NET_RX_DROP; | |
3331 | } | |
1da177e4 | 3332 | |
ae78dbfa | 3333 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3334 | { |
b0e28f1e | 3335 | int ret; |
1da177e4 | 3336 | |
588f0330 | 3337 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3338 | |
cf66ba58 | 3339 | trace_netif_rx(skb); |
df334545 | 3340 | #ifdef CONFIG_RPS |
c5905afb | 3341 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3342 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3343 | int cpu; |
3344 | ||
cece1945 | 3345 | preempt_disable(); |
b0e28f1e | 3346 | rcu_read_lock(); |
fec5e652 TH |
3347 | |
3348 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3349 | if (cpu < 0) |
3350 | cpu = smp_processor_id(); | |
fec5e652 TH |
3351 | |
3352 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3353 | ||
b0e28f1e | 3354 | rcu_read_unlock(); |
cece1945 | 3355 | preempt_enable(); |
adc9300e ED |
3356 | } else |
3357 | #endif | |
fec5e652 TH |
3358 | { |
3359 | unsigned int qtail; | |
3360 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3361 | put_cpu(); | |
3362 | } | |
b0e28f1e | 3363 | return ret; |
1da177e4 | 3364 | } |
ae78dbfa BH |
3365 | |
3366 | /** | |
3367 | * netif_rx - post buffer to the network code | |
3368 | * @skb: buffer to post | |
3369 | * | |
3370 | * This function receives a packet from a device driver and queues it for | |
3371 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3372 | * may be dropped during processing for congestion control or by the | |
3373 | * protocol layers. | |
3374 | * | |
3375 | * return values: | |
3376 | * NET_RX_SUCCESS (no congestion) | |
3377 | * NET_RX_DROP (packet was dropped) | |
3378 | * | |
3379 | */ | |
3380 | ||
3381 | int netif_rx(struct sk_buff *skb) | |
3382 | { | |
3383 | trace_netif_rx_entry(skb); | |
3384 | ||
3385 | return netif_rx_internal(skb); | |
3386 | } | |
d1b19dff | 3387 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3388 | |
3389 | int netif_rx_ni(struct sk_buff *skb) | |
3390 | { | |
3391 | int err; | |
3392 | ||
ae78dbfa BH |
3393 | trace_netif_rx_ni_entry(skb); |
3394 | ||
1da177e4 | 3395 | preempt_disable(); |
ae78dbfa | 3396 | err = netif_rx_internal(skb); |
1da177e4 LT |
3397 | if (local_softirq_pending()) |
3398 | do_softirq(); | |
3399 | preempt_enable(); | |
3400 | ||
3401 | return err; | |
3402 | } | |
1da177e4 LT |
3403 | EXPORT_SYMBOL(netif_rx_ni); |
3404 | ||
1da177e4 LT |
3405 | static void net_tx_action(struct softirq_action *h) |
3406 | { | |
3407 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
3408 | ||
3409 | if (sd->completion_queue) { | |
3410 | struct sk_buff *clist; | |
3411 | ||
3412 | local_irq_disable(); | |
3413 | clist = sd->completion_queue; | |
3414 | sd->completion_queue = NULL; | |
3415 | local_irq_enable(); | |
3416 | ||
3417 | while (clist) { | |
3418 | struct sk_buff *skb = clist; | |
3419 | clist = clist->next; | |
3420 | ||
547b792c | 3421 | WARN_ON(atomic_read(&skb->users)); |
e6247027 ED |
3422 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
3423 | trace_consume_skb(skb); | |
3424 | else | |
3425 | trace_kfree_skb(skb, net_tx_action); | |
1da177e4 LT |
3426 | __kfree_skb(skb); |
3427 | } | |
3428 | } | |
3429 | ||
3430 | if (sd->output_queue) { | |
37437bb2 | 3431 | struct Qdisc *head; |
1da177e4 LT |
3432 | |
3433 | local_irq_disable(); | |
3434 | head = sd->output_queue; | |
3435 | sd->output_queue = NULL; | |
a9cbd588 | 3436 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3437 | local_irq_enable(); |
3438 | ||
3439 | while (head) { | |
37437bb2 DM |
3440 | struct Qdisc *q = head; |
3441 | spinlock_t *root_lock; | |
3442 | ||
1da177e4 LT |
3443 | head = head->next_sched; |
3444 | ||
5fb66229 | 3445 | root_lock = qdisc_lock(q); |
37437bb2 | 3446 | if (spin_trylock(root_lock)) { |
4e857c58 | 3447 | smp_mb__before_atomic(); |
def82a1d JP |
3448 | clear_bit(__QDISC_STATE_SCHED, |
3449 | &q->state); | |
37437bb2 DM |
3450 | qdisc_run(q); |
3451 | spin_unlock(root_lock); | |
1da177e4 | 3452 | } else { |
195648bb | 3453 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3454 | &q->state)) { |
195648bb | 3455 | __netif_reschedule(q); |
e8a83e10 | 3456 | } else { |
4e857c58 | 3457 | smp_mb__before_atomic(); |
e8a83e10 JP |
3458 | clear_bit(__QDISC_STATE_SCHED, |
3459 | &q->state); | |
3460 | } | |
1da177e4 LT |
3461 | } |
3462 | } | |
3463 | } | |
3464 | } | |
3465 | ||
ab95bfe0 JP |
3466 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3467 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3468 | /* This hook is defined here for ATM LANE */ |
3469 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3470 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3471 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3472 | #endif |
1da177e4 | 3473 | |
1da177e4 LT |
3474 | #ifdef CONFIG_NET_CLS_ACT |
3475 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3476 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3477 | * a compare and 2 stores extra right now if we dont have it on | |
3478 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3479 | * NOTE: This doesn't stop any functionality; if you dont have |
3480 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3481 | * |
3482 | */ | |
24824a09 | 3483 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3484 | { |
1da177e4 | 3485 | struct net_device *dev = skb->dev; |
f697c3e8 | 3486 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3487 | int result = TC_ACT_OK; |
3488 | struct Qdisc *q; | |
4ec93edb | 3489 | |
de384830 | 3490 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
e87cc472 JP |
3491 | net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n", |
3492 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3493 | return TC_ACT_SHOT; |
3494 | } | |
1da177e4 | 3495 | |
f697c3e8 HX |
3496 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3497 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3498 | |
46e5da40 | 3499 | q = rcu_dereference(rxq->qdisc); |
8d50b53d | 3500 | if (q != &noop_qdisc) { |
83874000 | 3501 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3502 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3503 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3504 | spin_unlock(qdisc_lock(q)); |
3505 | } | |
f697c3e8 HX |
3506 | |
3507 | return result; | |
3508 | } | |
86e65da9 | 3509 | |
f697c3e8 HX |
3510 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3511 | struct packet_type **pt_prev, | |
3512 | int *ret, struct net_device *orig_dev) | |
3513 | { | |
24824a09 ED |
3514 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3515 | ||
46e5da40 | 3516 | if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc) |
f697c3e8 | 3517 | goto out; |
1da177e4 | 3518 | |
f697c3e8 HX |
3519 | if (*pt_prev) { |
3520 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3521 | *pt_prev = NULL; | |
1da177e4 LT |
3522 | } |
3523 | ||
24824a09 | 3524 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3525 | case TC_ACT_SHOT: |
3526 | case TC_ACT_STOLEN: | |
3527 | kfree_skb(skb); | |
3528 | return NULL; | |
3529 | } | |
3530 | ||
3531 | out: | |
3532 | skb->tc_verd = 0; | |
3533 | return skb; | |
1da177e4 LT |
3534 | } |
3535 | #endif | |
3536 | ||
ab95bfe0 JP |
3537 | /** |
3538 | * netdev_rx_handler_register - register receive handler | |
3539 | * @dev: device to register a handler for | |
3540 | * @rx_handler: receive handler to register | |
93e2c32b | 3541 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 3542 | * |
e227867f | 3543 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
3544 | * called from __netif_receive_skb. A negative errno code is returned |
3545 | * on a failure. | |
3546 | * | |
3547 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3548 | * |
3549 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3550 | */ |
3551 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3552 | rx_handler_func_t *rx_handler, |
3553 | void *rx_handler_data) | |
ab95bfe0 JP |
3554 | { |
3555 | ASSERT_RTNL(); | |
3556 | ||
3557 | if (dev->rx_handler) | |
3558 | return -EBUSY; | |
3559 | ||
00cfec37 | 3560 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 3561 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3562 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3563 | ||
3564 | return 0; | |
3565 | } | |
3566 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3567 | ||
3568 | /** | |
3569 | * netdev_rx_handler_unregister - unregister receive handler | |
3570 | * @dev: device to unregister a handler from | |
3571 | * | |
166ec369 | 3572 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
3573 | * |
3574 | * The caller must hold the rtnl_mutex. | |
3575 | */ | |
3576 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3577 | { | |
3578 | ||
3579 | ASSERT_RTNL(); | |
a9b3cd7f | 3580 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
3581 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
3582 | * section has a guarantee to see a non NULL rx_handler_data | |
3583 | * as well. | |
3584 | */ | |
3585 | synchronize_net(); | |
a9b3cd7f | 3586 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
3587 | } |
3588 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3589 | ||
b4b9e355 MG |
3590 | /* |
3591 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3592 | * the special handling of PFMEMALLOC skbs. | |
3593 | */ | |
3594 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3595 | { | |
3596 | switch (skb->protocol) { | |
2b8837ae JP |
3597 | case htons(ETH_P_ARP): |
3598 | case htons(ETH_P_IP): | |
3599 | case htons(ETH_P_IPV6): | |
3600 | case htons(ETH_P_8021Q): | |
3601 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
3602 | return true; |
3603 | default: | |
3604 | return false; | |
3605 | } | |
3606 | } | |
3607 | ||
9754e293 | 3608 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
3609 | { |
3610 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3611 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3612 | struct net_device *orig_dev; |
63d8ea7f | 3613 | struct net_device *null_or_dev; |
8a4eb573 | 3614 | bool deliver_exact = false; |
1da177e4 | 3615 | int ret = NET_RX_DROP; |
252e3346 | 3616 | __be16 type; |
1da177e4 | 3617 | |
588f0330 | 3618 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3619 | |
cf66ba58 | 3620 | trace_netif_receive_skb(skb); |
9b22ea56 | 3621 | |
cc9bd5ce | 3622 | orig_dev = skb->dev; |
8f903c70 | 3623 | |
c1d2bbe1 | 3624 | skb_reset_network_header(skb); |
fda55eca ED |
3625 | if (!skb_transport_header_was_set(skb)) |
3626 | skb_reset_transport_header(skb); | |
0b5c9db1 | 3627 | skb_reset_mac_len(skb); |
1da177e4 LT |
3628 | |
3629 | pt_prev = NULL; | |
3630 | ||
3631 | rcu_read_lock(); | |
3632 | ||
63d8ea7f | 3633 | another_round: |
b6858177 | 3634 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
3635 | |
3636 | __this_cpu_inc(softnet_data.processed); | |
3637 | ||
8ad227ff PM |
3638 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
3639 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 3640 | skb = skb_vlan_untag(skb); |
bcc6d479 | 3641 | if (unlikely(!skb)) |
b4b9e355 | 3642 | goto unlock; |
bcc6d479 JP |
3643 | } |
3644 | ||
1da177e4 LT |
3645 | #ifdef CONFIG_NET_CLS_ACT |
3646 | if (skb->tc_verd & TC_NCLS) { | |
3647 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3648 | goto ncls; | |
3649 | } | |
3650 | #endif | |
3651 | ||
9754e293 | 3652 | if (pfmemalloc) |
b4b9e355 MG |
3653 | goto skip_taps; |
3654 | ||
1da177e4 | 3655 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
63d8ea7f | 3656 | if (!ptype->dev || ptype->dev == skb->dev) { |
4ec93edb | 3657 | if (pt_prev) |
f2ccd8fa | 3658 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3659 | pt_prev = ptype; |
3660 | } | |
3661 | } | |
3662 | ||
b4b9e355 | 3663 | skip_taps: |
1da177e4 | 3664 | #ifdef CONFIG_NET_CLS_ACT |
f697c3e8 HX |
3665 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3666 | if (!skb) | |
b4b9e355 | 3667 | goto unlock; |
1da177e4 LT |
3668 | ncls: |
3669 | #endif | |
3670 | ||
9754e293 | 3671 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
3672 | goto drop; |
3673 | ||
2425717b JF |
3674 | if (vlan_tx_tag_present(skb)) { |
3675 | if (pt_prev) { | |
3676 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3677 | pt_prev = NULL; | |
3678 | } | |
48cc32d3 | 3679 | if (vlan_do_receive(&skb)) |
2425717b JF |
3680 | goto another_round; |
3681 | else if (unlikely(!skb)) | |
b4b9e355 | 3682 | goto unlock; |
2425717b JF |
3683 | } |
3684 | ||
48cc32d3 | 3685 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
3686 | if (rx_handler) { |
3687 | if (pt_prev) { | |
3688 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3689 | pt_prev = NULL; | |
3690 | } | |
8a4eb573 JP |
3691 | switch (rx_handler(&skb)) { |
3692 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 3693 | ret = NET_RX_SUCCESS; |
b4b9e355 | 3694 | goto unlock; |
8a4eb573 | 3695 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3696 | goto another_round; |
8a4eb573 JP |
3697 | case RX_HANDLER_EXACT: |
3698 | deliver_exact = true; | |
3699 | case RX_HANDLER_PASS: | |
3700 | break; | |
3701 | default: | |
3702 | BUG(); | |
3703 | } | |
ab95bfe0 | 3704 | } |
1da177e4 | 3705 | |
d4b812de ED |
3706 | if (unlikely(vlan_tx_tag_present(skb))) { |
3707 | if (vlan_tx_tag_get_id(skb)) | |
3708 | skb->pkt_type = PACKET_OTHERHOST; | |
3709 | /* Note: we might in the future use prio bits | |
3710 | * and set skb->priority like in vlan_do_receive() | |
3711 | * For the time being, just ignore Priority Code Point | |
3712 | */ | |
3713 | skb->vlan_tci = 0; | |
3714 | } | |
48cc32d3 | 3715 | |
63d8ea7f | 3716 | /* deliver only exact match when indicated */ |
8a4eb573 | 3717 | null_or_dev = deliver_exact ? skb->dev : NULL; |
1f3c8804 | 3718 | |
1da177e4 | 3719 | type = skb->protocol; |
82d8a867 PE |
3720 | list_for_each_entry_rcu(ptype, |
3721 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
63d8ea7f | 3722 | if (ptype->type == type && |
e3f48d37 JP |
3723 | (ptype->dev == null_or_dev || ptype->dev == skb->dev || |
3724 | ptype->dev == orig_dev)) { | |
4ec93edb | 3725 | if (pt_prev) |
f2ccd8fa | 3726 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3727 | pt_prev = ptype; |
3728 | } | |
3729 | } | |
3730 | ||
3731 | if (pt_prev) { | |
1080e512 | 3732 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 3733 | goto drop; |
1080e512 MT |
3734 | else |
3735 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 3736 | } else { |
b4b9e355 | 3737 | drop: |
caf586e5 | 3738 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3739 | kfree_skb(skb); |
3740 | /* Jamal, now you will not able to escape explaining | |
3741 | * me how you were going to use this. :-) | |
3742 | */ | |
3743 | ret = NET_RX_DROP; | |
3744 | } | |
3745 | ||
b4b9e355 | 3746 | unlock: |
1da177e4 | 3747 | rcu_read_unlock(); |
9754e293 DM |
3748 | return ret; |
3749 | } | |
3750 | ||
3751 | static int __netif_receive_skb(struct sk_buff *skb) | |
3752 | { | |
3753 | int ret; | |
3754 | ||
3755 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
3756 | unsigned long pflags = current->flags; | |
3757 | ||
3758 | /* | |
3759 | * PFMEMALLOC skbs are special, they should | |
3760 | * - be delivered to SOCK_MEMALLOC sockets only | |
3761 | * - stay away from userspace | |
3762 | * - have bounded memory usage | |
3763 | * | |
3764 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
3765 | * context down to all allocation sites. | |
3766 | */ | |
3767 | current->flags |= PF_MEMALLOC; | |
3768 | ret = __netif_receive_skb_core(skb, true); | |
3769 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
3770 | } else | |
3771 | ret = __netif_receive_skb_core(skb, false); | |
3772 | ||
1da177e4 LT |
3773 | return ret; |
3774 | } | |
0a9627f2 | 3775 | |
ae78dbfa | 3776 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 3777 | { |
588f0330 | 3778 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3779 | |
c1f19b51 RC |
3780 | if (skb_defer_rx_timestamp(skb)) |
3781 | return NET_RX_SUCCESS; | |
3782 | ||
df334545 | 3783 | #ifdef CONFIG_RPS |
c5905afb | 3784 | if (static_key_false(&rps_needed)) { |
3b098e2d ED |
3785 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
3786 | int cpu, ret; | |
fec5e652 | 3787 | |
3b098e2d ED |
3788 | rcu_read_lock(); |
3789 | ||
3790 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3791 | |
3b098e2d ED |
3792 | if (cpu >= 0) { |
3793 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3794 | rcu_read_unlock(); | |
adc9300e | 3795 | return ret; |
3b098e2d | 3796 | } |
adc9300e | 3797 | rcu_read_unlock(); |
fec5e652 | 3798 | } |
1e94d72f | 3799 | #endif |
adc9300e | 3800 | return __netif_receive_skb(skb); |
0a9627f2 | 3801 | } |
ae78dbfa BH |
3802 | |
3803 | /** | |
3804 | * netif_receive_skb - process receive buffer from network | |
3805 | * @skb: buffer to process | |
3806 | * | |
3807 | * netif_receive_skb() is the main receive data processing function. | |
3808 | * It always succeeds. The buffer may be dropped during processing | |
3809 | * for congestion control or by the protocol layers. | |
3810 | * | |
3811 | * This function may only be called from softirq context and interrupts | |
3812 | * should be enabled. | |
3813 | * | |
3814 | * Return values (usually ignored): | |
3815 | * NET_RX_SUCCESS: no congestion | |
3816 | * NET_RX_DROP: packet was dropped | |
3817 | */ | |
3818 | int netif_receive_skb(struct sk_buff *skb) | |
3819 | { | |
3820 | trace_netif_receive_skb_entry(skb); | |
3821 | ||
3822 | return netif_receive_skb_internal(skb); | |
3823 | } | |
d1b19dff | 3824 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3825 | |
88751275 ED |
3826 | /* Network device is going away, flush any packets still pending |
3827 | * Called with irqs disabled. | |
3828 | */ | |
152102c7 | 3829 | static void flush_backlog(void *arg) |
6e583ce5 | 3830 | { |
152102c7 | 3831 | struct net_device *dev = arg; |
e36fa2f7 | 3832 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3833 | struct sk_buff *skb, *tmp; |
3834 | ||
e36fa2f7 | 3835 | rps_lock(sd); |
6e7676c1 | 3836 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3837 | if (skb->dev == dev) { |
e36fa2f7 | 3838 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3839 | kfree_skb(skb); |
76cc8b13 | 3840 | input_queue_head_incr(sd); |
6e583ce5 | 3841 | } |
6e7676c1 | 3842 | } |
e36fa2f7 | 3843 | rps_unlock(sd); |
6e7676c1 CG |
3844 | |
3845 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3846 | if (skb->dev == dev) { | |
3847 | __skb_unlink(skb, &sd->process_queue); | |
3848 | kfree_skb(skb); | |
76cc8b13 | 3849 | input_queue_head_incr(sd); |
6e7676c1 CG |
3850 | } |
3851 | } | |
6e583ce5 SH |
3852 | } |
3853 | ||
d565b0a1 HX |
3854 | static int napi_gro_complete(struct sk_buff *skb) |
3855 | { | |
22061d80 | 3856 | struct packet_offload *ptype; |
d565b0a1 | 3857 | __be16 type = skb->protocol; |
22061d80 | 3858 | struct list_head *head = &offload_base; |
d565b0a1 HX |
3859 | int err = -ENOENT; |
3860 | ||
c3c7c254 ED |
3861 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
3862 | ||
fc59f9a3 HX |
3863 | if (NAPI_GRO_CB(skb)->count == 1) { |
3864 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3865 | goto out; |
fc59f9a3 | 3866 | } |
d565b0a1 HX |
3867 | |
3868 | rcu_read_lock(); | |
3869 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3870 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
3871 | continue; |
3872 | ||
299603e8 | 3873 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
3874 | break; |
3875 | } | |
3876 | rcu_read_unlock(); | |
3877 | ||
3878 | if (err) { | |
3879 | WARN_ON(&ptype->list == head); | |
3880 | kfree_skb(skb); | |
3881 | return NET_RX_SUCCESS; | |
3882 | } | |
3883 | ||
3884 | out: | |
ae78dbfa | 3885 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
3886 | } |
3887 | ||
2e71a6f8 ED |
3888 | /* napi->gro_list contains packets ordered by age. |
3889 | * youngest packets at the head of it. | |
3890 | * Complete skbs in reverse order to reduce latencies. | |
3891 | */ | |
3892 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 3893 | { |
2e71a6f8 | 3894 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 3895 | |
2e71a6f8 ED |
3896 | /* scan list and build reverse chain */ |
3897 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
3898 | skb->prev = prev; | |
3899 | prev = skb; | |
3900 | } | |
3901 | ||
3902 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 3903 | skb->next = NULL; |
2e71a6f8 ED |
3904 | |
3905 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
3906 | return; | |
3907 | ||
3908 | prev = skb->prev; | |
d565b0a1 | 3909 | napi_gro_complete(skb); |
2e71a6f8 | 3910 | napi->gro_count--; |
d565b0a1 HX |
3911 | } |
3912 | ||
3913 | napi->gro_list = NULL; | |
3914 | } | |
86cac58b | 3915 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3916 | |
89c5fa33 ED |
3917 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
3918 | { | |
3919 | struct sk_buff *p; | |
3920 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 3921 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
3922 | |
3923 | for (p = napi->gro_list; p; p = p->next) { | |
3924 | unsigned long diffs; | |
3925 | ||
0b4cec8c TH |
3926 | NAPI_GRO_CB(p)->flush = 0; |
3927 | ||
3928 | if (hash != skb_get_hash_raw(p)) { | |
3929 | NAPI_GRO_CB(p)->same_flow = 0; | |
3930 | continue; | |
3931 | } | |
3932 | ||
89c5fa33 ED |
3933 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
3934 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
3935 | if (maclen == ETH_HLEN) | |
3936 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 3937 | skb_mac_header(skb)); |
89c5fa33 ED |
3938 | else if (!diffs) |
3939 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 3940 | skb_mac_header(skb), |
89c5fa33 ED |
3941 | maclen); |
3942 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
3943 | } |
3944 | } | |
3945 | ||
299603e8 JC |
3946 | static void skb_gro_reset_offset(struct sk_buff *skb) |
3947 | { | |
3948 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
3949 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
3950 | ||
3951 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3952 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
3953 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
3954 | ||
3955 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
3956 | pinfo->nr_frags && | |
3957 | !PageHighMem(skb_frag_page(frag0))) { | |
3958 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
3959 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
89c5fa33 ED |
3960 | } |
3961 | } | |
3962 | ||
a50e233c ED |
3963 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
3964 | { | |
3965 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
3966 | ||
3967 | BUG_ON(skb->end - skb->tail < grow); | |
3968 | ||
3969 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3970 | ||
3971 | skb->data_len -= grow; | |
3972 | skb->tail += grow; | |
3973 | ||
3974 | pinfo->frags[0].page_offset += grow; | |
3975 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
3976 | ||
3977 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
3978 | skb_frag_unref(skb, 0); | |
3979 | memmove(pinfo->frags, pinfo->frags + 1, | |
3980 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
3981 | } | |
3982 | } | |
3983 | ||
bb728820 | 3984 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3985 | { |
3986 | struct sk_buff **pp = NULL; | |
22061d80 | 3987 | struct packet_offload *ptype; |
d565b0a1 | 3988 | __be16 type = skb->protocol; |
22061d80 | 3989 | struct list_head *head = &offload_base; |
0da2afd5 | 3990 | int same_flow; |
5b252f0c | 3991 | enum gro_result ret; |
a50e233c | 3992 | int grow; |
d565b0a1 | 3993 | |
9c62a68d | 3994 | if (!(skb->dev->features & NETIF_F_GRO)) |
d565b0a1 HX |
3995 | goto normal; |
3996 | ||
5a212329 | 3997 | if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad) |
f17f5c91 HX |
3998 | goto normal; |
3999 | ||
89c5fa33 | 4000 | gro_list_prepare(napi, skb); |
573e8fca | 4001 | |
d565b0a1 HX |
4002 | rcu_read_lock(); |
4003 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4004 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4005 | continue; |
4006 | ||
86911732 | 4007 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4008 | skb_reset_mac_len(skb); |
d565b0a1 HX |
4009 | NAPI_GRO_CB(skb)->same_flow = 0; |
4010 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 4011 | NAPI_GRO_CB(skb)->free = 0; |
b582ef09 | 4012 | NAPI_GRO_CB(skb)->udp_mark = 0; |
662880f4 TH |
4013 | |
4014 | /* Setup for GRO checksum validation */ | |
4015 | switch (skb->ip_summed) { | |
4016 | case CHECKSUM_COMPLETE: | |
4017 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4018 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4019 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4020 | break; | |
4021 | case CHECKSUM_UNNECESSARY: | |
4022 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4023 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4024 | break; | |
4025 | default: | |
4026 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4027 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4028 | } | |
d565b0a1 | 4029 | |
f191a1d1 | 4030 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4031 | break; |
4032 | } | |
4033 | rcu_read_unlock(); | |
4034 | ||
4035 | if (&ptype->list == head) | |
4036 | goto normal; | |
4037 | ||
0da2afd5 | 4038 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4039 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4040 | |
d565b0a1 HX |
4041 | if (pp) { |
4042 | struct sk_buff *nskb = *pp; | |
4043 | ||
4044 | *pp = nskb->next; | |
4045 | nskb->next = NULL; | |
4046 | napi_gro_complete(nskb); | |
4ae5544f | 4047 | napi->gro_count--; |
d565b0a1 HX |
4048 | } |
4049 | ||
0da2afd5 | 4050 | if (same_flow) |
d565b0a1 HX |
4051 | goto ok; |
4052 | ||
600adc18 | 4053 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4054 | goto normal; |
d565b0a1 | 4055 | |
600adc18 ED |
4056 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4057 | struct sk_buff *nskb = napi->gro_list; | |
4058 | ||
4059 | /* locate the end of the list to select the 'oldest' flow */ | |
4060 | while (nskb->next) { | |
4061 | pp = &nskb->next; | |
4062 | nskb = *pp; | |
4063 | } | |
4064 | *pp = NULL; | |
4065 | nskb->next = NULL; | |
4066 | napi_gro_complete(nskb); | |
4067 | } else { | |
4068 | napi->gro_count++; | |
4069 | } | |
d565b0a1 | 4070 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4071 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4072 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4073 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4074 | skb->next = napi->gro_list; |
4075 | napi->gro_list = skb; | |
5d0d9be8 | 4076 | ret = GRO_HELD; |
d565b0a1 | 4077 | |
ad0f9904 | 4078 | pull: |
a50e233c ED |
4079 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4080 | if (grow > 0) | |
4081 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4082 | ok: |
5d0d9be8 | 4083 | return ret; |
d565b0a1 HX |
4084 | |
4085 | normal: | |
ad0f9904 HX |
4086 | ret = GRO_NORMAL; |
4087 | goto pull; | |
5d38a079 | 4088 | } |
96e93eab | 4089 | |
bf5a755f JC |
4090 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4091 | { | |
4092 | struct list_head *offload_head = &offload_base; | |
4093 | struct packet_offload *ptype; | |
4094 | ||
4095 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4096 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4097 | continue; | |
4098 | return ptype; | |
4099 | } | |
4100 | return NULL; | |
4101 | } | |
e27a2f83 | 4102 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4103 | |
4104 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4105 | { | |
4106 | struct list_head *offload_head = &offload_base; | |
4107 | struct packet_offload *ptype; | |
4108 | ||
4109 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4110 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4111 | continue; | |
4112 | return ptype; | |
4113 | } | |
4114 | return NULL; | |
4115 | } | |
e27a2f83 | 4116 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4117 | |
bb728820 | 4118 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4119 | { |
5d0d9be8 HX |
4120 | switch (ret) { |
4121 | case GRO_NORMAL: | |
ae78dbfa | 4122 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4123 | ret = GRO_DROP; |
4124 | break; | |
5d38a079 | 4125 | |
5d0d9be8 | 4126 | case GRO_DROP: |
5d38a079 HX |
4127 | kfree_skb(skb); |
4128 | break; | |
5b252f0c | 4129 | |
daa86548 | 4130 | case GRO_MERGED_FREE: |
d7e8883c ED |
4131 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
4132 | kmem_cache_free(skbuff_head_cache, skb); | |
4133 | else | |
4134 | __kfree_skb(skb); | |
daa86548 ED |
4135 | break; |
4136 | ||
5b252f0c BH |
4137 | case GRO_HELD: |
4138 | case GRO_MERGED: | |
4139 | break; | |
5d38a079 HX |
4140 | } |
4141 | ||
c7c4b3b6 | 4142 | return ret; |
5d0d9be8 | 4143 | } |
5d0d9be8 | 4144 | |
c7c4b3b6 | 4145 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4146 | { |
ae78dbfa | 4147 | trace_napi_gro_receive_entry(skb); |
86911732 | 4148 | |
a50e233c ED |
4149 | skb_gro_reset_offset(skb); |
4150 | ||
89c5fa33 | 4151 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4152 | } |
4153 | EXPORT_SYMBOL(napi_gro_receive); | |
4154 | ||
d0c2b0d2 | 4155 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4156 | { |
96e93eab | 4157 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4158 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4159 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4160 | skb->vlan_tci = 0; |
66c46d74 | 4161 | skb->dev = napi->dev; |
6d152e23 | 4162 | skb->skb_iif = 0; |
c3caf119 JC |
4163 | skb->encapsulation = 0; |
4164 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4165 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
96e93eab HX |
4166 | |
4167 | napi->skb = skb; | |
4168 | } | |
96e93eab | 4169 | |
76620aaf | 4170 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4171 | { |
5d38a079 | 4172 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4173 | |
4174 | if (!skb) { | |
89d71a66 | 4175 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
84b9cd63 | 4176 | napi->skb = skb; |
80595d59 | 4177 | } |
96e93eab HX |
4178 | return skb; |
4179 | } | |
76620aaf | 4180 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4181 | |
a50e233c ED |
4182 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4183 | struct sk_buff *skb, | |
4184 | gro_result_t ret) | |
96e93eab | 4185 | { |
5d0d9be8 HX |
4186 | switch (ret) { |
4187 | case GRO_NORMAL: | |
a50e233c ED |
4188 | case GRO_HELD: |
4189 | __skb_push(skb, ETH_HLEN); | |
4190 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4191 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4192 | ret = GRO_DROP; |
86911732 | 4193 | break; |
5d38a079 | 4194 | |
5d0d9be8 | 4195 | case GRO_DROP: |
5d0d9be8 HX |
4196 | case GRO_MERGED_FREE: |
4197 | napi_reuse_skb(napi, skb); | |
4198 | break; | |
5b252f0c BH |
4199 | |
4200 | case GRO_MERGED: | |
4201 | break; | |
5d0d9be8 | 4202 | } |
5d38a079 | 4203 | |
c7c4b3b6 | 4204 | return ret; |
5d38a079 | 4205 | } |
5d0d9be8 | 4206 | |
a50e233c ED |
4207 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
4208 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
4209 | * We copy ethernet header into skb->data to have a common layout. | |
4210 | */ | |
4adb9c4a | 4211 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4212 | { |
4213 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
4214 | const struct ethhdr *eth; |
4215 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
4216 | |
4217 | napi->skb = NULL; | |
4218 | ||
a50e233c ED |
4219 | skb_reset_mac_header(skb); |
4220 | skb_gro_reset_offset(skb); | |
4221 | ||
4222 | eth = skb_gro_header_fast(skb, 0); | |
4223 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
4224 | eth = skb_gro_header_slow(skb, hlen, 0); | |
4225 | if (unlikely(!eth)) { | |
4226 | napi_reuse_skb(napi, skb); | |
4227 | return NULL; | |
4228 | } | |
4229 | } else { | |
4230 | gro_pull_from_frag0(skb, hlen); | |
4231 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
4232 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 4233 | } |
a50e233c ED |
4234 | __skb_pull(skb, hlen); |
4235 | ||
4236 | /* | |
4237 | * This works because the only protocols we care about don't require | |
4238 | * special handling. | |
4239 | * We'll fix it up properly in napi_frags_finish() | |
4240 | */ | |
4241 | skb->protocol = eth->h_proto; | |
76620aaf | 4242 | |
76620aaf HX |
4243 | return skb; |
4244 | } | |
76620aaf | 4245 | |
c7c4b3b6 | 4246 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4247 | { |
76620aaf | 4248 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4249 | |
4250 | if (!skb) | |
c7c4b3b6 | 4251 | return GRO_DROP; |
5d0d9be8 | 4252 | |
ae78dbfa BH |
4253 | trace_napi_gro_frags_entry(skb); |
4254 | ||
89c5fa33 | 4255 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4256 | } |
5d38a079 HX |
4257 | EXPORT_SYMBOL(napi_gro_frags); |
4258 | ||
573e8fca TH |
4259 | /* Compute the checksum from gro_offset and return the folded value |
4260 | * after adding in any pseudo checksum. | |
4261 | */ | |
4262 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
4263 | { | |
4264 | __wsum wsum; | |
4265 | __sum16 sum; | |
4266 | ||
4267 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
4268 | ||
4269 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
4270 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
4271 | if (likely(!sum)) { | |
4272 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
4273 | !skb->csum_complete_sw) | |
4274 | netdev_rx_csum_fault(skb->dev); | |
4275 | } | |
4276 | ||
4277 | NAPI_GRO_CB(skb)->csum = wsum; | |
4278 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4279 | ||
4280 | return sum; | |
4281 | } | |
4282 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
4283 | ||
e326bed2 | 4284 | /* |
855abcf0 | 4285 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
4286 | * Note: called with local irq disabled, but exits with local irq enabled. |
4287 | */ | |
4288 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4289 | { | |
4290 | #ifdef CONFIG_RPS | |
4291 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4292 | ||
4293 | if (remsd) { | |
4294 | sd->rps_ipi_list = NULL; | |
4295 | ||
4296 | local_irq_enable(); | |
4297 | ||
4298 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
4299 | while (remsd) { | |
4300 | struct softnet_data *next = remsd->rps_ipi_next; | |
4301 | ||
4302 | if (cpu_online(remsd->cpu)) | |
c46fff2a | 4303 | smp_call_function_single_async(remsd->cpu, |
fce8ad15 | 4304 | &remsd->csd); |
e326bed2 ED |
4305 | remsd = next; |
4306 | } | |
4307 | } else | |
4308 | #endif | |
4309 | local_irq_enable(); | |
4310 | } | |
4311 | ||
bea3348e | 4312 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
4313 | { |
4314 | int work = 0; | |
eecfd7c4 | 4315 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 4316 | |
e326bed2 ED |
4317 | #ifdef CONFIG_RPS |
4318 | /* Check if we have pending ipi, its better to send them now, | |
4319 | * not waiting net_rx_action() end. | |
4320 | */ | |
4321 | if (sd->rps_ipi_list) { | |
4322 | local_irq_disable(); | |
4323 | net_rps_action_and_irq_enable(sd); | |
4324 | } | |
4325 | #endif | |
bea3348e | 4326 | napi->weight = weight_p; |
6e7676c1 | 4327 | local_irq_disable(); |
11ef7a89 | 4328 | while (1) { |
1da177e4 | 4329 | struct sk_buff *skb; |
6e7676c1 CG |
4330 | |
4331 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
4332 | local_irq_enable(); | |
4333 | __netif_receive_skb(skb); | |
6e7676c1 | 4334 | local_irq_disable(); |
76cc8b13 TH |
4335 | input_queue_head_incr(sd); |
4336 | if (++work >= quota) { | |
4337 | local_irq_enable(); | |
4338 | return work; | |
4339 | } | |
6e7676c1 | 4340 | } |
1da177e4 | 4341 | |
e36fa2f7 | 4342 | rps_lock(sd); |
11ef7a89 | 4343 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
4344 | /* |
4345 | * Inline a custom version of __napi_complete(). | |
4346 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
4347 | * and NAPI_STATE_SCHED is the only possible flag set |
4348 | * on backlog. | |
4349 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
4350 | * and we dont need an smp_mb() memory barrier. |
4351 | */ | |
4352 | list_del(&napi->poll_list); | |
4353 | napi->state = 0; | |
11ef7a89 | 4354 | rps_unlock(sd); |
eecfd7c4 | 4355 | |
11ef7a89 | 4356 | break; |
bea3348e | 4357 | } |
11ef7a89 TH |
4358 | |
4359 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
4360 | &sd->process_queue); | |
e36fa2f7 | 4361 | rps_unlock(sd); |
6e7676c1 CG |
4362 | } |
4363 | local_irq_enable(); | |
1da177e4 | 4364 | |
bea3348e SH |
4365 | return work; |
4366 | } | |
1da177e4 | 4367 | |
bea3348e SH |
4368 | /** |
4369 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4370 | * @n: entry to schedule |
bea3348e SH |
4371 | * |
4372 | * The entry's receive function will be scheduled to run | |
4373 | */ | |
b5606c2d | 4374 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4375 | { |
4376 | unsigned long flags; | |
1da177e4 | 4377 | |
bea3348e | 4378 | local_irq_save(flags); |
eecfd7c4 | 4379 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 4380 | local_irq_restore(flags); |
1da177e4 | 4381 | } |
bea3348e SH |
4382 | EXPORT_SYMBOL(__napi_schedule); |
4383 | ||
d565b0a1 HX |
4384 | void __napi_complete(struct napi_struct *n) |
4385 | { | |
4386 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
4387 | BUG_ON(n->gro_list); | |
4388 | ||
4389 | list_del(&n->poll_list); | |
4e857c58 | 4390 | smp_mb__before_atomic(); |
d565b0a1 HX |
4391 | clear_bit(NAPI_STATE_SCHED, &n->state); |
4392 | } | |
4393 | EXPORT_SYMBOL(__napi_complete); | |
4394 | ||
4395 | void napi_complete(struct napi_struct *n) | |
4396 | { | |
4397 | unsigned long flags; | |
4398 | ||
4399 | /* | |
4400 | * don't let napi dequeue from the cpu poll list | |
4401 | * just in case its running on a different cpu | |
4402 | */ | |
4403 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4404 | return; | |
4405 | ||
2e71a6f8 | 4406 | napi_gro_flush(n, false); |
d565b0a1 HX |
4407 | local_irq_save(flags); |
4408 | __napi_complete(n); | |
4409 | local_irq_restore(flags); | |
4410 | } | |
4411 | EXPORT_SYMBOL(napi_complete); | |
4412 | ||
af12fa6e ET |
4413 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
4414 | struct napi_struct *napi_by_id(unsigned int napi_id) | |
4415 | { | |
4416 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
4417 | struct napi_struct *napi; | |
4418 | ||
4419 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
4420 | if (napi->napi_id == napi_id) | |
4421 | return napi; | |
4422 | ||
4423 | return NULL; | |
4424 | } | |
4425 | EXPORT_SYMBOL_GPL(napi_by_id); | |
4426 | ||
4427 | void napi_hash_add(struct napi_struct *napi) | |
4428 | { | |
4429 | if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) { | |
4430 | ||
4431 | spin_lock(&napi_hash_lock); | |
4432 | ||
4433 | /* 0 is not a valid id, we also skip an id that is taken | |
4434 | * we expect both events to be extremely rare | |
4435 | */ | |
4436 | napi->napi_id = 0; | |
4437 | while (!napi->napi_id) { | |
4438 | napi->napi_id = ++napi_gen_id; | |
4439 | if (napi_by_id(napi->napi_id)) | |
4440 | napi->napi_id = 0; | |
4441 | } | |
4442 | ||
4443 | hlist_add_head_rcu(&napi->napi_hash_node, | |
4444 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
4445 | ||
4446 | spin_unlock(&napi_hash_lock); | |
4447 | } | |
4448 | } | |
4449 | EXPORT_SYMBOL_GPL(napi_hash_add); | |
4450 | ||
4451 | /* Warning : caller is responsible to make sure rcu grace period | |
4452 | * is respected before freeing memory containing @napi | |
4453 | */ | |
4454 | void napi_hash_del(struct napi_struct *napi) | |
4455 | { | |
4456 | spin_lock(&napi_hash_lock); | |
4457 | ||
4458 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) | |
4459 | hlist_del_rcu(&napi->napi_hash_node); | |
4460 | ||
4461 | spin_unlock(&napi_hash_lock); | |
4462 | } | |
4463 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
4464 | ||
d565b0a1 HX |
4465 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
4466 | int (*poll)(struct napi_struct *, int), int weight) | |
4467 | { | |
4468 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 4469 | napi->gro_count = 0; |
d565b0a1 | 4470 | napi->gro_list = NULL; |
5d38a079 | 4471 | napi->skb = NULL; |
d565b0a1 | 4472 | napi->poll = poll; |
82dc3c63 ED |
4473 | if (weight > NAPI_POLL_WEIGHT) |
4474 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
4475 | weight, dev->name); | |
d565b0a1 HX |
4476 | napi->weight = weight; |
4477 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 4478 | napi->dev = dev; |
5d38a079 | 4479 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
4480 | spin_lock_init(&napi->poll_lock); |
4481 | napi->poll_owner = -1; | |
4482 | #endif | |
4483 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
4484 | } | |
4485 | EXPORT_SYMBOL(netif_napi_add); | |
4486 | ||
4487 | void netif_napi_del(struct napi_struct *napi) | |
4488 | { | |
d7b06636 | 4489 | list_del_init(&napi->dev_list); |
76620aaf | 4490 | napi_free_frags(napi); |
d565b0a1 | 4491 | |
289dccbe | 4492 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 4493 | napi->gro_list = NULL; |
4ae5544f | 4494 | napi->gro_count = 0; |
d565b0a1 HX |
4495 | } |
4496 | EXPORT_SYMBOL(netif_napi_del); | |
4497 | ||
1da177e4 LT |
4498 | static void net_rx_action(struct softirq_action *h) |
4499 | { | |
e326bed2 | 4500 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 4501 | unsigned long time_limit = jiffies + 2; |
51b0bded | 4502 | int budget = netdev_budget; |
53fb95d3 MM |
4503 | void *have; |
4504 | ||
1da177e4 LT |
4505 | local_irq_disable(); |
4506 | ||
e326bed2 | 4507 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
4508 | struct napi_struct *n; |
4509 | int work, weight; | |
1da177e4 | 4510 | |
bea3348e | 4511 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
4512 | * Allow this to run for 2 jiffies since which will allow |
4513 | * an average latency of 1.5/HZ. | |
bea3348e | 4514 | */ |
d1f41b67 | 4515 | if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit))) |
1da177e4 LT |
4516 | goto softnet_break; |
4517 | ||
4518 | local_irq_enable(); | |
4519 | ||
bea3348e SH |
4520 | /* Even though interrupts have been re-enabled, this |
4521 | * access is safe because interrupts can only add new | |
4522 | * entries to the tail of this list, and only ->poll() | |
4523 | * calls can remove this head entry from the list. | |
4524 | */ | |
e326bed2 | 4525 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 4526 | |
bea3348e SH |
4527 | have = netpoll_poll_lock(n); |
4528 | ||
4529 | weight = n->weight; | |
4530 | ||
0a7606c1 DM |
4531 | /* This NAPI_STATE_SCHED test is for avoiding a race |
4532 | * with netpoll's poll_napi(). Only the entity which | |
4533 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
4534 | * actually make the ->poll() call. Therefore we avoid | |
25985edc | 4535 | * accidentally calling ->poll() when NAPI is not scheduled. |
0a7606c1 DM |
4536 | */ |
4537 | work = 0; | |
4ea7e386 | 4538 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 4539 | work = n->poll(n, weight); |
4ea7e386 NH |
4540 | trace_napi_poll(n); |
4541 | } | |
bea3348e SH |
4542 | |
4543 | WARN_ON_ONCE(work > weight); | |
4544 | ||
4545 | budget -= work; | |
4546 | ||
4547 | local_irq_disable(); | |
4548 | ||
4549 | /* Drivers must not modify the NAPI state if they | |
4550 | * consume the entire weight. In such cases this code | |
4551 | * still "owns" the NAPI instance and therefore can | |
4552 | * move the instance around on the list at-will. | |
4553 | */ | |
fed17f30 | 4554 | if (unlikely(work == weight)) { |
ff780cd8 HX |
4555 | if (unlikely(napi_disable_pending(n))) { |
4556 | local_irq_enable(); | |
4557 | napi_complete(n); | |
4558 | local_irq_disable(); | |
2e71a6f8 ED |
4559 | } else { |
4560 | if (n->gro_list) { | |
4561 | /* flush too old packets | |
4562 | * If HZ < 1000, flush all packets. | |
4563 | */ | |
4564 | local_irq_enable(); | |
4565 | napi_gro_flush(n, HZ >= 1000); | |
4566 | local_irq_disable(); | |
4567 | } | |
e326bed2 | 4568 | list_move_tail(&n->poll_list, &sd->poll_list); |
2e71a6f8 | 4569 | } |
fed17f30 | 4570 | } |
bea3348e SH |
4571 | |
4572 | netpoll_poll_unlock(have); | |
1da177e4 LT |
4573 | } |
4574 | out: | |
e326bed2 | 4575 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 4576 | |
db217334 CL |
4577 | #ifdef CONFIG_NET_DMA |
4578 | /* | |
4579 | * There may not be any more sk_buffs coming right now, so push | |
4580 | * any pending DMA copies to hardware | |
4581 | */ | |
2ba05622 | 4582 | dma_issue_pending_all(); |
db217334 | 4583 | #endif |
bea3348e | 4584 | |
1da177e4 LT |
4585 | return; |
4586 | ||
4587 | softnet_break: | |
dee42870 | 4588 | sd->time_squeeze++; |
1da177e4 LT |
4589 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
4590 | goto out; | |
4591 | } | |
4592 | ||
aa9d8560 | 4593 | struct netdev_adjacent { |
9ff162a8 | 4594 | struct net_device *dev; |
5d261913 VF |
4595 | |
4596 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 4597 | bool master; |
5d261913 | 4598 | |
5d261913 VF |
4599 | /* counter for the number of times this device was added to us */ |
4600 | u16 ref_nr; | |
4601 | ||
402dae96 VF |
4602 | /* private field for the users */ |
4603 | void *private; | |
4604 | ||
9ff162a8 JP |
4605 | struct list_head list; |
4606 | struct rcu_head rcu; | |
9ff162a8 JP |
4607 | }; |
4608 | ||
5d261913 VF |
4609 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev, |
4610 | struct net_device *adj_dev, | |
2f268f12 | 4611 | struct list_head *adj_list) |
9ff162a8 | 4612 | { |
5d261913 | 4613 | struct netdev_adjacent *adj; |
5d261913 | 4614 | |
2f268f12 | 4615 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
4616 | if (adj->dev == adj_dev) |
4617 | return adj; | |
9ff162a8 JP |
4618 | } |
4619 | return NULL; | |
4620 | } | |
4621 | ||
4622 | /** | |
4623 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
4624 | * @dev: device | |
4625 | * @upper_dev: upper device to check | |
4626 | * | |
4627 | * Find out if a device is linked to specified upper device and return true | |
4628 | * in case it is. Note that this checks only immediate upper device, | |
4629 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
4630 | */ | |
4631 | bool netdev_has_upper_dev(struct net_device *dev, | |
4632 | struct net_device *upper_dev) | |
4633 | { | |
4634 | ASSERT_RTNL(); | |
4635 | ||
2f268f12 | 4636 | return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper); |
9ff162a8 JP |
4637 | } |
4638 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
4639 | ||
4640 | /** | |
4641 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
4642 | * @dev: device | |
4643 | * | |
4644 | * Find out if a device is linked to an upper device and return true in case | |
4645 | * it is. The caller must hold the RTNL lock. | |
4646 | */ | |
1d143d9f | 4647 | static bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
4648 | { |
4649 | ASSERT_RTNL(); | |
4650 | ||
2f268f12 | 4651 | return !list_empty(&dev->all_adj_list.upper); |
9ff162a8 | 4652 | } |
9ff162a8 JP |
4653 | |
4654 | /** | |
4655 | * netdev_master_upper_dev_get - Get master upper device | |
4656 | * @dev: device | |
4657 | * | |
4658 | * Find a master upper device and return pointer to it or NULL in case | |
4659 | * it's not there. The caller must hold the RTNL lock. | |
4660 | */ | |
4661 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
4662 | { | |
aa9d8560 | 4663 | struct netdev_adjacent *upper; |
9ff162a8 JP |
4664 | |
4665 | ASSERT_RTNL(); | |
4666 | ||
2f268f12 | 4667 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
4668 | return NULL; |
4669 | ||
2f268f12 | 4670 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 4671 | struct netdev_adjacent, list); |
9ff162a8 JP |
4672 | if (likely(upper->master)) |
4673 | return upper->dev; | |
4674 | return NULL; | |
4675 | } | |
4676 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
4677 | ||
b6ccba4c VF |
4678 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
4679 | { | |
4680 | struct netdev_adjacent *adj; | |
4681 | ||
4682 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
4683 | ||
4684 | return adj->private; | |
4685 | } | |
4686 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
4687 | ||
44a40855 VY |
4688 | /** |
4689 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
4690 | * @dev: device | |
4691 | * @iter: list_head ** of the current position | |
4692 | * | |
4693 | * Gets the next device from the dev's upper list, starting from iter | |
4694 | * position. The caller must hold RCU read lock. | |
4695 | */ | |
4696 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
4697 | struct list_head **iter) | |
4698 | { | |
4699 | struct netdev_adjacent *upper; | |
4700 | ||
4701 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
4702 | ||
4703 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4704 | ||
4705 | if (&upper->list == &dev->adj_list.upper) | |
4706 | return NULL; | |
4707 | ||
4708 | *iter = &upper->list; | |
4709 | ||
4710 | return upper->dev; | |
4711 | } | |
4712 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
4713 | ||
31088a11 VF |
4714 | /** |
4715 | * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list | |
48311f46 VF |
4716 | * @dev: device |
4717 | * @iter: list_head ** of the current position | |
4718 | * | |
4719 | * Gets the next device from the dev's upper list, starting from iter | |
4720 | * position. The caller must hold RCU read lock. | |
4721 | */ | |
2f268f12 VF |
4722 | struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, |
4723 | struct list_head **iter) | |
48311f46 VF |
4724 | { |
4725 | struct netdev_adjacent *upper; | |
4726 | ||
85328240 | 4727 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); |
48311f46 VF |
4728 | |
4729 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4730 | ||
2f268f12 | 4731 | if (&upper->list == &dev->all_adj_list.upper) |
48311f46 VF |
4732 | return NULL; |
4733 | ||
4734 | *iter = &upper->list; | |
4735 | ||
4736 | return upper->dev; | |
4737 | } | |
2f268f12 | 4738 | EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); |
48311f46 | 4739 | |
31088a11 VF |
4740 | /** |
4741 | * netdev_lower_get_next_private - Get the next ->private from the | |
4742 | * lower neighbour list | |
4743 | * @dev: device | |
4744 | * @iter: list_head ** of the current position | |
4745 | * | |
4746 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
4747 | * list, starting from iter position. The caller must hold either hold the | |
4748 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
4749 | * list will remain unchainged. | |
4750 | */ | |
4751 | void *netdev_lower_get_next_private(struct net_device *dev, | |
4752 | struct list_head **iter) | |
4753 | { | |
4754 | struct netdev_adjacent *lower; | |
4755 | ||
4756 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
4757 | ||
4758 | if (&lower->list == &dev->adj_list.lower) | |
4759 | return NULL; | |
4760 | ||
6859e7df | 4761 | *iter = lower->list.next; |
31088a11 VF |
4762 | |
4763 | return lower->private; | |
4764 | } | |
4765 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
4766 | ||
4767 | /** | |
4768 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
4769 | * lower neighbour list, RCU | |
4770 | * variant | |
4771 | * @dev: device | |
4772 | * @iter: list_head ** of the current position | |
4773 | * | |
4774 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
4775 | * list, starting from iter position. The caller must hold RCU read lock. | |
4776 | */ | |
4777 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
4778 | struct list_head **iter) | |
4779 | { | |
4780 | struct netdev_adjacent *lower; | |
4781 | ||
4782 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4783 | ||
4784 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4785 | ||
4786 | if (&lower->list == &dev->adj_list.lower) | |
4787 | return NULL; | |
4788 | ||
6859e7df | 4789 | *iter = &lower->list; |
31088a11 VF |
4790 | |
4791 | return lower->private; | |
4792 | } | |
4793 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
4794 | ||
4085ebe8 VY |
4795 | /** |
4796 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
4797 | * list | |
4798 | * @dev: device | |
4799 | * @iter: list_head ** of the current position | |
4800 | * | |
4801 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
4802 | * list, starting from iter position. The caller must hold RTNL lock or | |
4803 | * its own locking that guarantees that the neighbour lower | |
4804 | * list will remain unchainged. | |
4805 | */ | |
4806 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
4807 | { | |
4808 | struct netdev_adjacent *lower; | |
4809 | ||
4810 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
4811 | ||
4812 | if (&lower->list == &dev->adj_list.lower) | |
4813 | return NULL; | |
4814 | ||
4815 | *iter = &lower->list; | |
4816 | ||
4817 | return lower->dev; | |
4818 | } | |
4819 | EXPORT_SYMBOL(netdev_lower_get_next); | |
4820 | ||
e001bfad | 4821 | /** |
4822 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
4823 | * lower neighbour list, RCU | |
4824 | * variant | |
4825 | * @dev: device | |
4826 | * | |
4827 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
4828 | * list. The caller must hold RCU read lock. | |
4829 | */ | |
4830 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
4831 | { | |
4832 | struct netdev_adjacent *lower; | |
4833 | ||
4834 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
4835 | struct netdev_adjacent, list); | |
4836 | if (lower) | |
4837 | return lower->private; | |
4838 | return NULL; | |
4839 | } | |
4840 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
4841 | ||
9ff162a8 JP |
4842 | /** |
4843 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
4844 | * @dev: device | |
4845 | * | |
4846 | * Find a master upper device and return pointer to it or NULL in case | |
4847 | * it's not there. The caller must hold the RCU read lock. | |
4848 | */ | |
4849 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
4850 | { | |
aa9d8560 | 4851 | struct netdev_adjacent *upper; |
9ff162a8 | 4852 | |
2f268f12 | 4853 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 4854 | struct netdev_adjacent, list); |
9ff162a8 JP |
4855 | if (upper && likely(upper->master)) |
4856 | return upper->dev; | |
4857 | return NULL; | |
4858 | } | |
4859 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
4860 | ||
0a59f3a9 | 4861 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
4862 | struct net_device *adj_dev, |
4863 | struct list_head *dev_list) | |
4864 | { | |
4865 | char linkname[IFNAMSIZ+7]; | |
4866 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
4867 | "upper_%s" : "lower_%s", adj_dev->name); | |
4868 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
4869 | linkname); | |
4870 | } | |
0a59f3a9 | 4871 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
4872 | char *name, |
4873 | struct list_head *dev_list) | |
4874 | { | |
4875 | char linkname[IFNAMSIZ+7]; | |
4876 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
4877 | "upper_%s" : "lower_%s", name); | |
4878 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
4879 | } | |
4880 | ||
7ce64c79 AF |
4881 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
4882 | struct net_device *adj_dev, | |
4883 | struct list_head *dev_list) | |
4884 | { | |
4885 | return (dev_list == &dev->adj_list.upper || | |
4886 | dev_list == &dev->adj_list.lower) && | |
4887 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
4888 | } | |
3ee32707 | 4889 | |
5d261913 VF |
4890 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
4891 | struct net_device *adj_dev, | |
7863c054 | 4892 | struct list_head *dev_list, |
402dae96 | 4893 | void *private, bool master) |
5d261913 VF |
4894 | { |
4895 | struct netdev_adjacent *adj; | |
842d67a7 | 4896 | int ret; |
5d261913 | 4897 | |
7863c054 | 4898 | adj = __netdev_find_adj(dev, adj_dev, dev_list); |
5d261913 VF |
4899 | |
4900 | if (adj) { | |
5d261913 VF |
4901 | adj->ref_nr++; |
4902 | return 0; | |
4903 | } | |
4904 | ||
4905 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
4906 | if (!adj) | |
4907 | return -ENOMEM; | |
4908 | ||
4909 | adj->dev = adj_dev; | |
4910 | adj->master = master; | |
5d261913 | 4911 | adj->ref_nr = 1; |
402dae96 | 4912 | adj->private = private; |
5d261913 | 4913 | dev_hold(adj_dev); |
2f268f12 VF |
4914 | |
4915 | pr_debug("dev_hold for %s, because of link added from %s to %s\n", | |
4916 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 | 4917 | |
7ce64c79 | 4918 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 4919 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
4920 | if (ret) |
4921 | goto free_adj; | |
4922 | } | |
4923 | ||
7863c054 | 4924 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
4925 | if (master) { |
4926 | ret = sysfs_create_link(&(dev->dev.kobj), | |
4927 | &(adj_dev->dev.kobj), "master"); | |
4928 | if (ret) | |
5831d66e | 4929 | goto remove_symlinks; |
842d67a7 | 4930 | |
7863c054 | 4931 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 4932 | } else { |
7863c054 | 4933 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 4934 | } |
5d261913 VF |
4935 | |
4936 | return 0; | |
842d67a7 | 4937 | |
5831d66e | 4938 | remove_symlinks: |
7ce64c79 | 4939 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 4940 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
4941 | free_adj: |
4942 | kfree(adj); | |
974daef7 | 4943 | dev_put(adj_dev); |
842d67a7 VF |
4944 | |
4945 | return ret; | |
5d261913 VF |
4946 | } |
4947 | ||
1d143d9f | 4948 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
4949 | struct net_device *adj_dev, | |
4950 | struct list_head *dev_list) | |
5d261913 VF |
4951 | { |
4952 | struct netdev_adjacent *adj; | |
4953 | ||
7863c054 | 4954 | adj = __netdev_find_adj(dev, adj_dev, dev_list); |
5d261913 | 4955 | |
2f268f12 VF |
4956 | if (!adj) { |
4957 | pr_err("tried to remove device %s from %s\n", | |
4958 | dev->name, adj_dev->name); | |
5d261913 | 4959 | BUG(); |
2f268f12 | 4960 | } |
5d261913 VF |
4961 | |
4962 | if (adj->ref_nr > 1) { | |
2f268f12 VF |
4963 | pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name, |
4964 | adj->ref_nr-1); | |
5d261913 VF |
4965 | adj->ref_nr--; |
4966 | return; | |
4967 | } | |
4968 | ||
842d67a7 VF |
4969 | if (adj->master) |
4970 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
4971 | ||
7ce64c79 | 4972 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 4973 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 4974 | |
5d261913 | 4975 | list_del_rcu(&adj->list); |
2f268f12 VF |
4976 | pr_debug("dev_put for %s, because link removed from %s to %s\n", |
4977 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 VF |
4978 | dev_put(adj_dev); |
4979 | kfree_rcu(adj, rcu); | |
4980 | } | |
4981 | ||
1d143d9f | 4982 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
4983 | struct net_device *upper_dev, | |
4984 | struct list_head *up_list, | |
4985 | struct list_head *down_list, | |
4986 | void *private, bool master) | |
5d261913 VF |
4987 | { |
4988 | int ret; | |
4989 | ||
402dae96 VF |
4990 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private, |
4991 | master); | |
5d261913 VF |
4992 | if (ret) |
4993 | return ret; | |
4994 | ||
402dae96 VF |
4995 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private, |
4996 | false); | |
5d261913 | 4997 | if (ret) { |
2f268f12 | 4998 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5d261913 VF |
4999 | return ret; |
5000 | } | |
5001 | ||
5002 | return 0; | |
5003 | } | |
5004 | ||
1d143d9f | 5005 | static int __netdev_adjacent_dev_link(struct net_device *dev, |
5006 | struct net_device *upper_dev) | |
5d261913 | 5007 | { |
2f268f12 VF |
5008 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
5009 | &dev->all_adj_list.upper, | |
5010 | &upper_dev->all_adj_list.lower, | |
402dae96 | 5011 | NULL, false); |
5d261913 VF |
5012 | } |
5013 | ||
1d143d9f | 5014 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
5015 | struct net_device *upper_dev, | |
5016 | struct list_head *up_list, | |
5017 | struct list_head *down_list) | |
5d261913 | 5018 | { |
2f268f12 VF |
5019 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5020 | __netdev_adjacent_dev_remove(upper_dev, dev, down_list); | |
5d261913 VF |
5021 | } |
5022 | ||
1d143d9f | 5023 | static void __netdev_adjacent_dev_unlink(struct net_device *dev, |
5024 | struct net_device *upper_dev) | |
5d261913 | 5025 | { |
2f268f12 VF |
5026 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, |
5027 | &dev->all_adj_list.upper, | |
5028 | &upper_dev->all_adj_list.lower); | |
5029 | } | |
5030 | ||
1d143d9f | 5031 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
5032 | struct net_device *upper_dev, | |
5033 | void *private, bool master) | |
2f268f12 VF |
5034 | { |
5035 | int ret = __netdev_adjacent_dev_link(dev, upper_dev); | |
5036 | ||
5037 | if (ret) | |
5038 | return ret; | |
5039 | ||
5040 | ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, | |
5041 | &dev->adj_list.upper, | |
5042 | &upper_dev->adj_list.lower, | |
402dae96 | 5043 | private, master); |
2f268f12 VF |
5044 | if (ret) { |
5045 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5046 | return ret; | |
5047 | } | |
5048 | ||
5049 | return 0; | |
5d261913 VF |
5050 | } |
5051 | ||
1d143d9f | 5052 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
5053 | struct net_device *upper_dev) | |
2f268f12 VF |
5054 | { |
5055 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5056 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, | |
5057 | &dev->adj_list.upper, | |
5058 | &upper_dev->adj_list.lower); | |
5059 | } | |
5d261913 | 5060 | |
9ff162a8 | 5061 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 VF |
5062 | struct net_device *upper_dev, bool master, |
5063 | void *private) | |
9ff162a8 | 5064 | { |
5d261913 VF |
5065 | struct netdev_adjacent *i, *j, *to_i, *to_j; |
5066 | int ret = 0; | |
9ff162a8 JP |
5067 | |
5068 | ASSERT_RTNL(); | |
5069 | ||
5070 | if (dev == upper_dev) | |
5071 | return -EBUSY; | |
5072 | ||
5073 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
2f268f12 | 5074 | if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper)) |
9ff162a8 JP |
5075 | return -EBUSY; |
5076 | ||
2f268f12 | 5077 | if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper)) |
9ff162a8 JP |
5078 | return -EEXIST; |
5079 | ||
5080 | if (master && netdev_master_upper_dev_get(dev)) | |
5081 | return -EBUSY; | |
5082 | ||
402dae96 VF |
5083 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private, |
5084 | master); | |
5d261913 VF |
5085 | if (ret) |
5086 | return ret; | |
9ff162a8 | 5087 | |
5d261913 | 5088 | /* Now that we linked these devs, make all the upper_dev's |
2f268f12 | 5089 | * all_adj_list.upper visible to every dev's all_adj_list.lower an |
5d261913 VF |
5090 | * versa, and don't forget the devices itself. All of these |
5091 | * links are non-neighbours. | |
5092 | */ | |
2f268f12 VF |
5093 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5094 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5095 | pr_debug("Interlinking %s with %s, non-neighbour\n", | |
5096 | i->dev->name, j->dev->name); | |
5d261913 VF |
5097 | ret = __netdev_adjacent_dev_link(i->dev, j->dev); |
5098 | if (ret) | |
5099 | goto rollback_mesh; | |
5100 | } | |
5101 | } | |
5102 | ||
5103 | /* add dev to every upper_dev's upper device */ | |
2f268f12 VF |
5104 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5105 | pr_debug("linking %s's upper device %s with %s\n", | |
5106 | upper_dev->name, i->dev->name, dev->name); | |
5d261913 VF |
5107 | ret = __netdev_adjacent_dev_link(dev, i->dev); |
5108 | if (ret) | |
5109 | goto rollback_upper_mesh; | |
5110 | } | |
5111 | ||
5112 | /* add upper_dev to every dev's lower device */ | |
2f268f12 VF |
5113 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5114 | pr_debug("linking %s's lower device %s with %s\n", dev->name, | |
5115 | i->dev->name, upper_dev->name); | |
5d261913 VF |
5116 | ret = __netdev_adjacent_dev_link(i->dev, upper_dev); |
5117 | if (ret) | |
5118 | goto rollback_lower_mesh; | |
5119 | } | |
9ff162a8 | 5120 | |
42e52bf9 | 5121 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 | 5122 | return 0; |
5d261913 VF |
5123 | |
5124 | rollback_lower_mesh: | |
5125 | to_i = i; | |
2f268f12 | 5126 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5d261913 VF |
5127 | if (i == to_i) |
5128 | break; | |
5129 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); | |
5130 | } | |
5131 | ||
5132 | i = NULL; | |
5133 | ||
5134 | rollback_upper_mesh: | |
5135 | to_i = i; | |
2f268f12 | 5136 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5d261913 VF |
5137 | if (i == to_i) |
5138 | break; | |
5139 | __netdev_adjacent_dev_unlink(dev, i->dev); | |
5140 | } | |
5141 | ||
5142 | i = j = NULL; | |
5143 | ||
5144 | rollback_mesh: | |
5145 | to_i = i; | |
5146 | to_j = j; | |
2f268f12 VF |
5147 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5148 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5d261913 VF |
5149 | if (i == to_i && j == to_j) |
5150 | break; | |
5151 | __netdev_adjacent_dev_unlink(i->dev, j->dev); | |
5152 | } | |
5153 | if (i == to_i) | |
5154 | break; | |
5155 | } | |
5156 | ||
2f268f12 | 5157 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5158 | |
5159 | return ret; | |
9ff162a8 JP |
5160 | } |
5161 | ||
5162 | /** | |
5163 | * netdev_upper_dev_link - Add a link to the upper device | |
5164 | * @dev: device | |
5165 | * @upper_dev: new upper device | |
5166 | * | |
5167 | * Adds a link to device which is upper to this one. The caller must hold | |
5168 | * the RTNL lock. On a failure a negative errno code is returned. | |
5169 | * On success the reference counts are adjusted and the function | |
5170 | * returns zero. | |
5171 | */ | |
5172 | int netdev_upper_dev_link(struct net_device *dev, | |
5173 | struct net_device *upper_dev) | |
5174 | { | |
402dae96 | 5175 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL); |
9ff162a8 JP |
5176 | } |
5177 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
5178 | ||
5179 | /** | |
5180 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
5181 | * @dev: device | |
5182 | * @upper_dev: new upper device | |
5183 | * | |
5184 | * Adds a link to device which is upper to this one. In this case, only | |
5185 | * one master upper device can be linked, although other non-master devices | |
5186 | * might be linked as well. The caller must hold the RTNL lock. | |
5187 | * On a failure a negative errno code is returned. On success the reference | |
5188 | * counts are adjusted and the function returns zero. | |
5189 | */ | |
5190 | int netdev_master_upper_dev_link(struct net_device *dev, | |
5191 | struct net_device *upper_dev) | |
5192 | { | |
402dae96 | 5193 | return __netdev_upper_dev_link(dev, upper_dev, true, NULL); |
9ff162a8 JP |
5194 | } |
5195 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
5196 | ||
402dae96 VF |
5197 | int netdev_master_upper_dev_link_private(struct net_device *dev, |
5198 | struct net_device *upper_dev, | |
5199 | void *private) | |
5200 | { | |
5201 | return __netdev_upper_dev_link(dev, upper_dev, true, private); | |
5202 | } | |
5203 | EXPORT_SYMBOL(netdev_master_upper_dev_link_private); | |
5204 | ||
9ff162a8 JP |
5205 | /** |
5206 | * netdev_upper_dev_unlink - Removes a link to upper device | |
5207 | * @dev: device | |
5208 | * @upper_dev: new upper device | |
5209 | * | |
5210 | * Removes a link to device which is upper to this one. The caller must hold | |
5211 | * the RTNL lock. | |
5212 | */ | |
5213 | void netdev_upper_dev_unlink(struct net_device *dev, | |
5214 | struct net_device *upper_dev) | |
5215 | { | |
5d261913 | 5216 | struct netdev_adjacent *i, *j; |
9ff162a8 JP |
5217 | ASSERT_RTNL(); |
5218 | ||
2f268f12 | 5219 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5220 | |
5221 | /* Here is the tricky part. We must remove all dev's lower | |
5222 | * devices from all upper_dev's upper devices and vice | |
5223 | * versa, to maintain the graph relationship. | |
5224 | */ | |
2f268f12 VF |
5225 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5226 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) | |
5d261913 VF |
5227 | __netdev_adjacent_dev_unlink(i->dev, j->dev); |
5228 | ||
5229 | /* remove also the devices itself from lower/upper device | |
5230 | * list | |
5231 | */ | |
2f268f12 | 5232 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5d261913 VF |
5233 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); |
5234 | ||
2f268f12 | 5235 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) |
5d261913 VF |
5236 | __netdev_adjacent_dev_unlink(dev, i->dev); |
5237 | ||
42e52bf9 | 5238 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 JP |
5239 | } |
5240 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
5241 | ||
4c75431a AF |
5242 | void netdev_adjacent_add_links(struct net_device *dev) |
5243 | { | |
5244 | struct netdev_adjacent *iter; | |
5245 | ||
5246 | struct net *net = dev_net(dev); | |
5247 | ||
5248 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
5249 | if (!net_eq(net,dev_net(iter->dev))) | |
5250 | continue; | |
5251 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5252 | &iter->dev->adj_list.lower); | |
5253 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5254 | &dev->adj_list.upper); | |
5255 | } | |
5256 | ||
5257 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
5258 | if (!net_eq(net,dev_net(iter->dev))) | |
5259 | continue; | |
5260 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5261 | &iter->dev->adj_list.upper); | |
5262 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5263 | &dev->adj_list.lower); | |
5264 | } | |
5265 | } | |
5266 | ||
5267 | void netdev_adjacent_del_links(struct net_device *dev) | |
5268 | { | |
5269 | struct netdev_adjacent *iter; | |
5270 | ||
5271 | struct net *net = dev_net(dev); | |
5272 | ||
5273 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
5274 | if (!net_eq(net,dev_net(iter->dev))) | |
5275 | continue; | |
5276 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5277 | &iter->dev->adj_list.lower); | |
5278 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5279 | &dev->adj_list.upper); | |
5280 | } | |
5281 | ||
5282 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
5283 | if (!net_eq(net,dev_net(iter->dev))) | |
5284 | continue; | |
5285 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5286 | &iter->dev->adj_list.upper); | |
5287 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5288 | &dev->adj_list.lower); | |
5289 | } | |
5290 | } | |
5291 | ||
5bb025fa | 5292 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 5293 | { |
5bb025fa | 5294 | struct netdev_adjacent *iter; |
402dae96 | 5295 | |
4c75431a AF |
5296 | struct net *net = dev_net(dev); |
5297 | ||
5bb025fa | 5298 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
4c75431a AF |
5299 | if (!net_eq(net,dev_net(iter->dev))) |
5300 | continue; | |
5bb025fa VF |
5301 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
5302 | &iter->dev->adj_list.lower); | |
5303 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5304 | &iter->dev->adj_list.lower); | |
5305 | } | |
402dae96 | 5306 | |
5bb025fa | 5307 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
4c75431a AF |
5308 | if (!net_eq(net,dev_net(iter->dev))) |
5309 | continue; | |
5bb025fa VF |
5310 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
5311 | &iter->dev->adj_list.upper); | |
5312 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5313 | &iter->dev->adj_list.upper); | |
5314 | } | |
402dae96 | 5315 | } |
402dae96 VF |
5316 | |
5317 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
5318 | struct net_device *lower_dev) | |
5319 | { | |
5320 | struct netdev_adjacent *lower; | |
5321 | ||
5322 | if (!lower_dev) | |
5323 | return NULL; | |
5324 | lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower); | |
5325 | if (!lower) | |
5326 | return NULL; | |
5327 | ||
5328 | return lower->private; | |
5329 | } | |
5330 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
5331 | ||
4085ebe8 VY |
5332 | |
5333 | int dev_get_nest_level(struct net_device *dev, | |
5334 | bool (*type_check)(struct net_device *dev)) | |
5335 | { | |
5336 | struct net_device *lower = NULL; | |
5337 | struct list_head *iter; | |
5338 | int max_nest = -1; | |
5339 | int nest; | |
5340 | ||
5341 | ASSERT_RTNL(); | |
5342 | ||
5343 | netdev_for_each_lower_dev(dev, lower, iter) { | |
5344 | nest = dev_get_nest_level(lower, type_check); | |
5345 | if (max_nest < nest) | |
5346 | max_nest = nest; | |
5347 | } | |
5348 | ||
5349 | if (type_check(dev)) | |
5350 | max_nest++; | |
5351 | ||
5352 | return max_nest; | |
5353 | } | |
5354 | EXPORT_SYMBOL(dev_get_nest_level); | |
5355 | ||
b6c40d68 PM |
5356 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
5357 | { | |
d314774c SH |
5358 | const struct net_device_ops *ops = dev->netdev_ops; |
5359 | ||
d2615bf4 | 5360 | if (ops->ndo_change_rx_flags) |
d314774c | 5361 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
5362 | } |
5363 | ||
991fb3f7 | 5364 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5365 | { |
b536db93 | 5366 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
5367 | kuid_t uid; |
5368 | kgid_t gid; | |
1da177e4 | 5369 | |
24023451 PM |
5370 | ASSERT_RTNL(); |
5371 | ||
dad9b335 WC |
5372 | dev->flags |= IFF_PROMISC; |
5373 | dev->promiscuity += inc; | |
5374 | if (dev->promiscuity == 0) { | |
5375 | /* | |
5376 | * Avoid overflow. | |
5377 | * If inc causes overflow, untouch promisc and return error. | |
5378 | */ | |
5379 | if (inc < 0) | |
5380 | dev->flags &= ~IFF_PROMISC; | |
5381 | else { | |
5382 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
5383 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
5384 | dev->name); | |
dad9b335 WC |
5385 | return -EOVERFLOW; |
5386 | } | |
5387 | } | |
52609c0b | 5388 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
5389 | pr_info("device %s %s promiscuous mode\n", |
5390 | dev->name, | |
5391 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
5392 | if (audit_enabled) { |
5393 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
5394 | audit_log(current->audit_context, GFP_ATOMIC, |
5395 | AUDIT_ANOM_PROMISCUOUS, | |
5396 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
5397 | dev->name, (dev->flags & IFF_PROMISC), | |
5398 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 5399 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
5400 | from_kuid(&init_user_ns, uid), |
5401 | from_kgid(&init_user_ns, gid), | |
7759db82 | 5402 | audit_get_sessionid(current)); |
8192b0c4 | 5403 | } |
24023451 | 5404 | |
b6c40d68 | 5405 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 5406 | } |
991fb3f7 ND |
5407 | if (notify) |
5408 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 5409 | return 0; |
1da177e4 LT |
5410 | } |
5411 | ||
4417da66 PM |
5412 | /** |
5413 | * dev_set_promiscuity - update promiscuity count on a device | |
5414 | * @dev: device | |
5415 | * @inc: modifier | |
5416 | * | |
5417 | * Add or remove promiscuity from a device. While the count in the device | |
5418 | * remains above zero the interface remains promiscuous. Once it hits zero | |
5419 | * the device reverts back to normal filtering operation. A negative inc | |
5420 | * value is used to drop promiscuity on the device. | |
dad9b335 | 5421 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 5422 | */ |
dad9b335 | 5423 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 5424 | { |
b536db93 | 5425 | unsigned int old_flags = dev->flags; |
dad9b335 | 5426 | int err; |
4417da66 | 5427 | |
991fb3f7 | 5428 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 5429 | if (err < 0) |
dad9b335 | 5430 | return err; |
4417da66 PM |
5431 | if (dev->flags != old_flags) |
5432 | dev_set_rx_mode(dev); | |
dad9b335 | 5433 | return err; |
4417da66 | 5434 | } |
d1b19dff | 5435 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 5436 | |
991fb3f7 | 5437 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5438 | { |
991fb3f7 | 5439 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 5440 | |
24023451 PM |
5441 | ASSERT_RTNL(); |
5442 | ||
1da177e4 | 5443 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
5444 | dev->allmulti += inc; |
5445 | if (dev->allmulti == 0) { | |
5446 | /* | |
5447 | * Avoid overflow. | |
5448 | * If inc causes overflow, untouch allmulti and return error. | |
5449 | */ | |
5450 | if (inc < 0) | |
5451 | dev->flags &= ~IFF_ALLMULTI; | |
5452 | else { | |
5453 | dev->allmulti -= inc; | |
7b6cd1ce JP |
5454 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
5455 | dev->name); | |
dad9b335 WC |
5456 | return -EOVERFLOW; |
5457 | } | |
5458 | } | |
24023451 | 5459 | if (dev->flags ^ old_flags) { |
b6c40d68 | 5460 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 5461 | dev_set_rx_mode(dev); |
991fb3f7 ND |
5462 | if (notify) |
5463 | __dev_notify_flags(dev, old_flags, | |
5464 | dev->gflags ^ old_gflags); | |
24023451 | 5465 | } |
dad9b335 | 5466 | return 0; |
4417da66 | 5467 | } |
991fb3f7 ND |
5468 | |
5469 | /** | |
5470 | * dev_set_allmulti - update allmulti count on a device | |
5471 | * @dev: device | |
5472 | * @inc: modifier | |
5473 | * | |
5474 | * Add or remove reception of all multicast frames to a device. While the | |
5475 | * count in the device remains above zero the interface remains listening | |
5476 | * to all interfaces. Once it hits zero the device reverts back to normal | |
5477 | * filtering operation. A negative @inc value is used to drop the counter | |
5478 | * when releasing a resource needing all multicasts. | |
5479 | * Return 0 if successful or a negative errno code on error. | |
5480 | */ | |
5481 | ||
5482 | int dev_set_allmulti(struct net_device *dev, int inc) | |
5483 | { | |
5484 | return __dev_set_allmulti(dev, inc, true); | |
5485 | } | |
d1b19dff | 5486 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
5487 | |
5488 | /* | |
5489 | * Upload unicast and multicast address lists to device and | |
5490 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 5491 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
5492 | * are present. |
5493 | */ | |
5494 | void __dev_set_rx_mode(struct net_device *dev) | |
5495 | { | |
d314774c SH |
5496 | const struct net_device_ops *ops = dev->netdev_ops; |
5497 | ||
4417da66 PM |
5498 | /* dev_open will call this function so the list will stay sane. */ |
5499 | if (!(dev->flags&IFF_UP)) | |
5500 | return; | |
5501 | ||
5502 | if (!netif_device_present(dev)) | |
40b77c94 | 5503 | return; |
4417da66 | 5504 | |
01789349 | 5505 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
5506 | /* Unicast addresses changes may only happen under the rtnl, |
5507 | * therefore calling __dev_set_promiscuity here is safe. | |
5508 | */ | |
32e7bfc4 | 5509 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 5510 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 5511 | dev->uc_promisc = true; |
32e7bfc4 | 5512 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 5513 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 5514 | dev->uc_promisc = false; |
4417da66 | 5515 | } |
4417da66 | 5516 | } |
01789349 JP |
5517 | |
5518 | if (ops->ndo_set_rx_mode) | |
5519 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
5520 | } |
5521 | ||
5522 | void dev_set_rx_mode(struct net_device *dev) | |
5523 | { | |
b9e40857 | 5524 | netif_addr_lock_bh(dev); |
4417da66 | 5525 | __dev_set_rx_mode(dev); |
b9e40857 | 5526 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
5527 | } |
5528 | ||
f0db275a SH |
5529 | /** |
5530 | * dev_get_flags - get flags reported to userspace | |
5531 | * @dev: device | |
5532 | * | |
5533 | * Get the combination of flag bits exported through APIs to userspace. | |
5534 | */ | |
95c96174 | 5535 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 5536 | { |
95c96174 | 5537 | unsigned int flags; |
1da177e4 LT |
5538 | |
5539 | flags = (dev->flags & ~(IFF_PROMISC | | |
5540 | IFF_ALLMULTI | | |
b00055aa SR |
5541 | IFF_RUNNING | |
5542 | IFF_LOWER_UP | | |
5543 | IFF_DORMANT)) | | |
1da177e4 LT |
5544 | (dev->gflags & (IFF_PROMISC | |
5545 | IFF_ALLMULTI)); | |
5546 | ||
b00055aa SR |
5547 | if (netif_running(dev)) { |
5548 | if (netif_oper_up(dev)) | |
5549 | flags |= IFF_RUNNING; | |
5550 | if (netif_carrier_ok(dev)) | |
5551 | flags |= IFF_LOWER_UP; | |
5552 | if (netif_dormant(dev)) | |
5553 | flags |= IFF_DORMANT; | |
5554 | } | |
1da177e4 LT |
5555 | |
5556 | return flags; | |
5557 | } | |
d1b19dff | 5558 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 5559 | |
bd380811 | 5560 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 5561 | { |
b536db93 | 5562 | unsigned int old_flags = dev->flags; |
bd380811 | 5563 | int ret; |
1da177e4 | 5564 | |
24023451 PM |
5565 | ASSERT_RTNL(); |
5566 | ||
1da177e4 LT |
5567 | /* |
5568 | * Set the flags on our device. | |
5569 | */ | |
5570 | ||
5571 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
5572 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
5573 | IFF_AUTOMEDIA)) | | |
5574 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
5575 | IFF_ALLMULTI)); | |
5576 | ||
5577 | /* | |
5578 | * Load in the correct multicast list now the flags have changed. | |
5579 | */ | |
5580 | ||
b6c40d68 PM |
5581 | if ((old_flags ^ flags) & IFF_MULTICAST) |
5582 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 5583 | |
4417da66 | 5584 | dev_set_rx_mode(dev); |
1da177e4 LT |
5585 | |
5586 | /* | |
5587 | * Have we downed the interface. We handle IFF_UP ourselves | |
5588 | * according to user attempts to set it, rather than blindly | |
5589 | * setting it. | |
5590 | */ | |
5591 | ||
5592 | ret = 0; | |
d215d10f | 5593 | if ((old_flags ^ flags) & IFF_UP) |
bd380811 | 5594 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 | 5595 | |
1da177e4 | 5596 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 5597 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 5598 | unsigned int old_flags = dev->flags; |
d1b19dff | 5599 | |
1da177e4 | 5600 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
5601 | |
5602 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
5603 | if (dev->flags != old_flags) | |
5604 | dev_set_rx_mode(dev); | |
1da177e4 LT |
5605 | } |
5606 | ||
5607 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
5608 | is important. Some (broken) drivers set IFF_PROMISC, when | |
5609 | IFF_ALLMULTI is requested not asking us and not reporting. | |
5610 | */ | |
5611 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
5612 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
5613 | ||
1da177e4 | 5614 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 5615 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
5616 | } |
5617 | ||
bd380811 PM |
5618 | return ret; |
5619 | } | |
5620 | ||
a528c219 ND |
5621 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
5622 | unsigned int gchanges) | |
bd380811 PM |
5623 | { |
5624 | unsigned int changes = dev->flags ^ old_flags; | |
5625 | ||
a528c219 | 5626 | if (gchanges) |
7f294054 | 5627 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 5628 | |
bd380811 PM |
5629 | if (changes & IFF_UP) { |
5630 | if (dev->flags & IFF_UP) | |
5631 | call_netdevice_notifiers(NETDEV_UP, dev); | |
5632 | else | |
5633 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
5634 | } | |
5635 | ||
5636 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
5637 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
5638 | struct netdev_notifier_change_info change_info; | |
5639 | ||
5640 | change_info.flags_changed = changes; | |
5641 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
5642 | &change_info.info); | |
5643 | } | |
bd380811 PM |
5644 | } |
5645 | ||
5646 | /** | |
5647 | * dev_change_flags - change device settings | |
5648 | * @dev: device | |
5649 | * @flags: device state flags | |
5650 | * | |
5651 | * Change settings on device based state flags. The flags are | |
5652 | * in the userspace exported format. | |
5653 | */ | |
b536db93 | 5654 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 5655 | { |
b536db93 | 5656 | int ret; |
991fb3f7 | 5657 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
5658 | |
5659 | ret = __dev_change_flags(dev, flags); | |
5660 | if (ret < 0) | |
5661 | return ret; | |
5662 | ||
991fb3f7 | 5663 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 5664 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
5665 | return ret; |
5666 | } | |
d1b19dff | 5667 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 5668 | |
2315dc91 VF |
5669 | static int __dev_set_mtu(struct net_device *dev, int new_mtu) |
5670 | { | |
5671 | const struct net_device_ops *ops = dev->netdev_ops; | |
5672 | ||
5673 | if (ops->ndo_change_mtu) | |
5674 | return ops->ndo_change_mtu(dev, new_mtu); | |
5675 | ||
5676 | dev->mtu = new_mtu; | |
5677 | return 0; | |
5678 | } | |
5679 | ||
f0db275a SH |
5680 | /** |
5681 | * dev_set_mtu - Change maximum transfer unit | |
5682 | * @dev: device | |
5683 | * @new_mtu: new transfer unit | |
5684 | * | |
5685 | * Change the maximum transfer size of the network device. | |
5686 | */ | |
1da177e4 LT |
5687 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
5688 | { | |
2315dc91 | 5689 | int err, orig_mtu; |
1da177e4 LT |
5690 | |
5691 | if (new_mtu == dev->mtu) | |
5692 | return 0; | |
5693 | ||
5694 | /* MTU must be positive. */ | |
5695 | if (new_mtu < 0) | |
5696 | return -EINVAL; | |
5697 | ||
5698 | if (!netif_device_present(dev)) | |
5699 | return -ENODEV; | |
5700 | ||
1d486bfb VF |
5701 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
5702 | err = notifier_to_errno(err); | |
5703 | if (err) | |
5704 | return err; | |
d314774c | 5705 | |
2315dc91 VF |
5706 | orig_mtu = dev->mtu; |
5707 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 5708 | |
2315dc91 VF |
5709 | if (!err) { |
5710 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
5711 | err = notifier_to_errno(err); | |
5712 | if (err) { | |
5713 | /* setting mtu back and notifying everyone again, | |
5714 | * so that they have a chance to revert changes. | |
5715 | */ | |
5716 | __dev_set_mtu(dev, orig_mtu); | |
5717 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
5718 | } | |
5719 | } | |
1da177e4 LT |
5720 | return err; |
5721 | } | |
d1b19dff | 5722 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 5723 | |
cbda10fa VD |
5724 | /** |
5725 | * dev_set_group - Change group this device belongs to | |
5726 | * @dev: device | |
5727 | * @new_group: group this device should belong to | |
5728 | */ | |
5729 | void dev_set_group(struct net_device *dev, int new_group) | |
5730 | { | |
5731 | dev->group = new_group; | |
5732 | } | |
5733 | EXPORT_SYMBOL(dev_set_group); | |
5734 | ||
f0db275a SH |
5735 | /** |
5736 | * dev_set_mac_address - Change Media Access Control Address | |
5737 | * @dev: device | |
5738 | * @sa: new address | |
5739 | * | |
5740 | * Change the hardware (MAC) address of the device | |
5741 | */ | |
1da177e4 LT |
5742 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
5743 | { | |
d314774c | 5744 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
5745 | int err; |
5746 | ||
d314774c | 5747 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
5748 | return -EOPNOTSUPP; |
5749 | if (sa->sa_family != dev->type) | |
5750 | return -EINVAL; | |
5751 | if (!netif_device_present(dev)) | |
5752 | return -ENODEV; | |
d314774c | 5753 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
5754 | if (err) |
5755 | return err; | |
fbdeca2d | 5756 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 5757 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 5758 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 5759 | return 0; |
1da177e4 | 5760 | } |
d1b19dff | 5761 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 5762 | |
4bf84c35 JP |
5763 | /** |
5764 | * dev_change_carrier - Change device carrier | |
5765 | * @dev: device | |
691b3b7e | 5766 | * @new_carrier: new value |
4bf84c35 JP |
5767 | * |
5768 | * Change device carrier | |
5769 | */ | |
5770 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
5771 | { | |
5772 | const struct net_device_ops *ops = dev->netdev_ops; | |
5773 | ||
5774 | if (!ops->ndo_change_carrier) | |
5775 | return -EOPNOTSUPP; | |
5776 | if (!netif_device_present(dev)) | |
5777 | return -ENODEV; | |
5778 | return ops->ndo_change_carrier(dev, new_carrier); | |
5779 | } | |
5780 | EXPORT_SYMBOL(dev_change_carrier); | |
5781 | ||
66b52b0d JP |
5782 | /** |
5783 | * dev_get_phys_port_id - Get device physical port ID | |
5784 | * @dev: device | |
5785 | * @ppid: port ID | |
5786 | * | |
5787 | * Get device physical port ID | |
5788 | */ | |
5789 | int dev_get_phys_port_id(struct net_device *dev, | |
5790 | struct netdev_phys_port_id *ppid) | |
5791 | { | |
5792 | const struct net_device_ops *ops = dev->netdev_ops; | |
5793 | ||
5794 | if (!ops->ndo_get_phys_port_id) | |
5795 | return -EOPNOTSUPP; | |
5796 | return ops->ndo_get_phys_port_id(dev, ppid); | |
5797 | } | |
5798 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
5799 | ||
1da177e4 LT |
5800 | /** |
5801 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5802 | * @net: the applicable net namespace |
1da177e4 LT |
5803 | * |
5804 | * Returns a suitable unique value for a new device interface | |
5805 | * number. The caller must hold the rtnl semaphore or the | |
5806 | * dev_base_lock to be sure it remains unique. | |
5807 | */ | |
881d966b | 5808 | static int dev_new_index(struct net *net) |
1da177e4 | 5809 | { |
aa79e66e | 5810 | int ifindex = net->ifindex; |
1da177e4 LT |
5811 | for (;;) { |
5812 | if (++ifindex <= 0) | |
5813 | ifindex = 1; | |
881d966b | 5814 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 5815 | return net->ifindex = ifindex; |
1da177e4 LT |
5816 | } |
5817 | } | |
5818 | ||
1da177e4 | 5819 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5820 | static LIST_HEAD(net_todo_list); |
200b916f | 5821 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 5822 | |
6f05f629 | 5823 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5824 | { |
1da177e4 | 5825 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 5826 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
5827 | } |
5828 | ||
9b5e383c | 5829 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5830 | { |
e93737b0 | 5831 | struct net_device *dev, *tmp; |
5cde2829 | 5832 | LIST_HEAD(close_head); |
9b5e383c | 5833 | |
93ee31f1 DL |
5834 | BUG_ON(dev_boot_phase); |
5835 | ASSERT_RTNL(); | |
5836 | ||
e93737b0 | 5837 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5838 | /* Some devices call without registering |
e93737b0 KK |
5839 | * for initialization unwind. Remove those |
5840 | * devices and proceed with the remaining. | |
9b5e383c ED |
5841 | */ |
5842 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
5843 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
5844 | dev->name, dev); | |
93ee31f1 | 5845 | |
9b5e383c | 5846 | WARN_ON(1); |
e93737b0 KK |
5847 | list_del(&dev->unreg_list); |
5848 | continue; | |
9b5e383c | 5849 | } |
449f4544 | 5850 | dev->dismantle = true; |
9b5e383c | 5851 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5852 | } |
93ee31f1 | 5853 | |
44345724 | 5854 | /* If device is running, close it first. */ |
5cde2829 EB |
5855 | list_for_each_entry(dev, head, unreg_list) |
5856 | list_add_tail(&dev->close_list, &close_head); | |
5857 | dev_close_many(&close_head); | |
93ee31f1 | 5858 | |
44345724 | 5859 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5860 | /* And unlink it from device chain. */ |
5861 | unlist_netdevice(dev); | |
93ee31f1 | 5862 | |
9b5e383c ED |
5863 | dev->reg_state = NETREG_UNREGISTERING; |
5864 | } | |
93ee31f1 DL |
5865 | |
5866 | synchronize_net(); | |
5867 | ||
9b5e383c ED |
5868 | list_for_each_entry(dev, head, unreg_list) { |
5869 | /* Shutdown queueing discipline. */ | |
5870 | dev_shutdown(dev); | |
93ee31f1 DL |
5871 | |
5872 | ||
9b5e383c ED |
5873 | /* Notify protocols, that we are about to destroy |
5874 | this device. They should clean all the things. | |
5875 | */ | |
5876 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5877 | |
9b5e383c ED |
5878 | /* |
5879 | * Flush the unicast and multicast chains | |
5880 | */ | |
a748ee24 | 5881 | dev_uc_flush(dev); |
22bedad3 | 5882 | dev_mc_flush(dev); |
93ee31f1 | 5883 | |
9b5e383c ED |
5884 | if (dev->netdev_ops->ndo_uninit) |
5885 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5886 | |
56bfa7ee RP |
5887 | if (!dev->rtnl_link_ops || |
5888 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5889 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); | |
5890 | ||
9ff162a8 JP |
5891 | /* Notifier chain MUST detach us all upper devices. */ |
5892 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 5893 | |
9b5e383c ED |
5894 | /* Remove entries from kobject tree */ |
5895 | netdev_unregister_kobject(dev); | |
024e9679 AD |
5896 | #ifdef CONFIG_XPS |
5897 | /* Remove XPS queueing entries */ | |
5898 | netif_reset_xps_queues_gt(dev, 0); | |
5899 | #endif | |
9b5e383c | 5900 | } |
93ee31f1 | 5901 | |
850a545b | 5902 | synchronize_net(); |
395264d5 | 5903 | |
a5ee1551 | 5904 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5905 | dev_put(dev); |
5906 | } | |
5907 | ||
5908 | static void rollback_registered(struct net_device *dev) | |
5909 | { | |
5910 | LIST_HEAD(single); | |
5911 | ||
5912 | list_add(&dev->unreg_list, &single); | |
5913 | rollback_registered_many(&single); | |
ceaaec98 | 5914 | list_del(&single); |
93ee31f1 DL |
5915 | } |
5916 | ||
c8f44aff MM |
5917 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
5918 | netdev_features_t features) | |
b63365a2 | 5919 | { |
57422dc5 MM |
5920 | /* Fix illegal checksum combinations */ |
5921 | if ((features & NETIF_F_HW_CSUM) && | |
5922 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5923 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
5924 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
5925 | } | |
5926 | ||
b63365a2 | 5927 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 5928 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 5929 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 5930 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
5931 | } |
5932 | ||
ec5f0615 PS |
5933 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
5934 | !(features & NETIF_F_IP_CSUM)) { | |
5935 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
5936 | features &= ~NETIF_F_TSO; | |
5937 | features &= ~NETIF_F_TSO_ECN; | |
5938 | } | |
5939 | ||
5940 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
5941 | !(features & NETIF_F_IPV6_CSUM)) { | |
5942 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
5943 | features &= ~NETIF_F_TSO6; | |
5944 | } | |
5945 | ||
31d8b9e0 BH |
5946 | /* TSO ECN requires that TSO is present as well. */ |
5947 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
5948 | features &= ~NETIF_F_TSO_ECN; | |
5949 | ||
212b573f MM |
5950 | /* Software GSO depends on SG. */ |
5951 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 5952 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
5953 | features &= ~NETIF_F_GSO; |
5954 | } | |
5955 | ||
acd1130e | 5956 | /* UFO needs SG and checksumming */ |
b63365a2 | 5957 | if (features & NETIF_F_UFO) { |
79032644 MM |
5958 | /* maybe split UFO into V4 and V6? */ |
5959 | if (!((features & NETIF_F_GEN_CSUM) || | |
5960 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
5961 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5962 | netdev_dbg(dev, |
acd1130e | 5963 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
5964 | features &= ~NETIF_F_UFO; |
5965 | } | |
5966 | ||
5967 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 5968 | netdev_dbg(dev, |
acd1130e | 5969 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
5970 | features &= ~NETIF_F_UFO; |
5971 | } | |
5972 | } | |
5973 | ||
d0290214 JP |
5974 | #ifdef CONFIG_NET_RX_BUSY_POLL |
5975 | if (dev->netdev_ops->ndo_busy_poll) | |
5976 | features |= NETIF_F_BUSY_POLL; | |
5977 | else | |
5978 | #endif | |
5979 | features &= ~NETIF_F_BUSY_POLL; | |
5980 | ||
b63365a2 HX |
5981 | return features; |
5982 | } | |
b63365a2 | 5983 | |
6cb6a27c | 5984 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 5985 | { |
c8f44aff | 5986 | netdev_features_t features; |
5455c699 MM |
5987 | int err = 0; |
5988 | ||
87267485 MM |
5989 | ASSERT_RTNL(); |
5990 | ||
5455c699 MM |
5991 | features = netdev_get_wanted_features(dev); |
5992 | ||
5993 | if (dev->netdev_ops->ndo_fix_features) | |
5994 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
5995 | ||
5996 | /* driver might be less strict about feature dependencies */ | |
5997 | features = netdev_fix_features(dev, features); | |
5998 | ||
5999 | if (dev->features == features) | |
6cb6a27c | 6000 | return 0; |
5455c699 | 6001 | |
c8f44aff MM |
6002 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
6003 | &dev->features, &features); | |
5455c699 MM |
6004 | |
6005 | if (dev->netdev_ops->ndo_set_features) | |
6006 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
6007 | ||
6cb6a27c | 6008 | if (unlikely(err < 0)) { |
5455c699 | 6009 | netdev_err(dev, |
c8f44aff MM |
6010 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
6011 | err, &features, &dev->features); | |
6cb6a27c MM |
6012 | return -1; |
6013 | } | |
6014 | ||
6015 | if (!err) | |
6016 | dev->features = features; | |
6017 | ||
6018 | return 1; | |
6019 | } | |
6020 | ||
afe12cc8 MM |
6021 | /** |
6022 | * netdev_update_features - recalculate device features | |
6023 | * @dev: the device to check | |
6024 | * | |
6025 | * Recalculate dev->features set and send notifications if it | |
6026 | * has changed. Should be called after driver or hardware dependent | |
6027 | * conditions might have changed that influence the features. | |
6028 | */ | |
6cb6a27c MM |
6029 | void netdev_update_features(struct net_device *dev) |
6030 | { | |
6031 | if (__netdev_update_features(dev)) | |
6032 | netdev_features_change(dev); | |
5455c699 MM |
6033 | } |
6034 | EXPORT_SYMBOL(netdev_update_features); | |
6035 | ||
afe12cc8 MM |
6036 | /** |
6037 | * netdev_change_features - recalculate device features | |
6038 | * @dev: the device to check | |
6039 | * | |
6040 | * Recalculate dev->features set and send notifications even | |
6041 | * if they have not changed. Should be called instead of | |
6042 | * netdev_update_features() if also dev->vlan_features might | |
6043 | * have changed to allow the changes to be propagated to stacked | |
6044 | * VLAN devices. | |
6045 | */ | |
6046 | void netdev_change_features(struct net_device *dev) | |
6047 | { | |
6048 | __netdev_update_features(dev); | |
6049 | netdev_features_change(dev); | |
6050 | } | |
6051 | EXPORT_SYMBOL(netdev_change_features); | |
6052 | ||
fc4a7489 PM |
6053 | /** |
6054 | * netif_stacked_transfer_operstate - transfer operstate | |
6055 | * @rootdev: the root or lower level device to transfer state from | |
6056 | * @dev: the device to transfer operstate to | |
6057 | * | |
6058 | * Transfer operational state from root to device. This is normally | |
6059 | * called when a stacking relationship exists between the root | |
6060 | * device and the device(a leaf device). | |
6061 | */ | |
6062 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
6063 | struct net_device *dev) | |
6064 | { | |
6065 | if (rootdev->operstate == IF_OPER_DORMANT) | |
6066 | netif_dormant_on(dev); | |
6067 | else | |
6068 | netif_dormant_off(dev); | |
6069 | ||
6070 | if (netif_carrier_ok(rootdev)) { | |
6071 | if (!netif_carrier_ok(dev)) | |
6072 | netif_carrier_on(dev); | |
6073 | } else { | |
6074 | if (netif_carrier_ok(dev)) | |
6075 | netif_carrier_off(dev); | |
6076 | } | |
6077 | } | |
6078 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
6079 | ||
a953be53 | 6080 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
6081 | static int netif_alloc_rx_queues(struct net_device *dev) |
6082 | { | |
1b4bf461 | 6083 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 6084 | struct netdev_rx_queue *rx; |
1b4bf461 | 6085 | |
bd25fa7b | 6086 | BUG_ON(count < 1); |
1b4bf461 | 6087 | |
bd25fa7b | 6088 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
62b5942a | 6089 | if (!rx) |
bd25fa7b | 6090 | return -ENOMEM; |
62b5942a | 6091 | |
bd25fa7b TH |
6092 | dev->_rx = rx; |
6093 | ||
bd25fa7b | 6094 | for (i = 0; i < count; i++) |
fe822240 | 6095 | rx[i].dev = dev; |
1b4bf461 ED |
6096 | return 0; |
6097 | } | |
bf264145 | 6098 | #endif |
1b4bf461 | 6099 | |
aa942104 CG |
6100 | static void netdev_init_one_queue(struct net_device *dev, |
6101 | struct netdev_queue *queue, void *_unused) | |
6102 | { | |
6103 | /* Initialize queue lock */ | |
6104 | spin_lock_init(&queue->_xmit_lock); | |
6105 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
6106 | queue->xmit_lock_owner = -1; | |
b236da69 | 6107 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 6108 | queue->dev = dev; |
114cf580 TH |
6109 | #ifdef CONFIG_BQL |
6110 | dql_init(&queue->dql, HZ); | |
6111 | #endif | |
aa942104 CG |
6112 | } |
6113 | ||
60877a32 ED |
6114 | static void netif_free_tx_queues(struct net_device *dev) |
6115 | { | |
4cb28970 | 6116 | kvfree(dev->_tx); |
60877a32 ED |
6117 | } |
6118 | ||
e6484930 TH |
6119 | static int netif_alloc_netdev_queues(struct net_device *dev) |
6120 | { | |
6121 | unsigned int count = dev->num_tx_queues; | |
6122 | struct netdev_queue *tx; | |
60877a32 | 6123 | size_t sz = count * sizeof(*tx); |
e6484930 | 6124 | |
60877a32 | 6125 | BUG_ON(count < 1 || count > 0xffff); |
62b5942a | 6126 | |
60877a32 ED |
6127 | tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6128 | if (!tx) { | |
6129 | tx = vzalloc(sz); | |
6130 | if (!tx) | |
6131 | return -ENOMEM; | |
6132 | } | |
e6484930 | 6133 | dev->_tx = tx; |
1d24eb48 | 6134 | |
e6484930 TH |
6135 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
6136 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
6137 | |
6138 | return 0; | |
e6484930 TH |
6139 | } |
6140 | ||
1da177e4 LT |
6141 | /** |
6142 | * register_netdevice - register a network device | |
6143 | * @dev: device to register | |
6144 | * | |
6145 | * Take a completed network device structure and add it to the kernel | |
6146 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6147 | * chain. 0 is returned on success. A negative errno code is returned | |
6148 | * on a failure to set up the device, or if the name is a duplicate. | |
6149 | * | |
6150 | * Callers must hold the rtnl semaphore. You may want | |
6151 | * register_netdev() instead of this. | |
6152 | * | |
6153 | * BUGS: | |
6154 | * The locking appears insufficient to guarantee two parallel registers | |
6155 | * will not get the same name. | |
6156 | */ | |
6157 | ||
6158 | int register_netdevice(struct net_device *dev) | |
6159 | { | |
1da177e4 | 6160 | int ret; |
d314774c | 6161 | struct net *net = dev_net(dev); |
1da177e4 LT |
6162 | |
6163 | BUG_ON(dev_boot_phase); | |
6164 | ASSERT_RTNL(); | |
6165 | ||
b17a7c17 SH |
6166 | might_sleep(); |
6167 | ||
1da177e4 LT |
6168 | /* When net_device's are persistent, this will be fatal. */ |
6169 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 6170 | BUG_ON(!net); |
1da177e4 | 6171 | |
f1f28aa3 | 6172 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 6173 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 6174 | |
1da177e4 LT |
6175 | dev->iflink = -1; |
6176 | ||
828de4f6 | 6177 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
6178 | if (ret < 0) |
6179 | goto out; | |
6180 | ||
1da177e4 | 6181 | /* Init, if this function is available */ |
d314774c SH |
6182 | if (dev->netdev_ops->ndo_init) { |
6183 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
6184 | if (ret) { |
6185 | if (ret > 0) | |
6186 | ret = -EIO; | |
90833aa4 | 6187 | goto out; |
1da177e4 LT |
6188 | } |
6189 | } | |
4ec93edb | 6190 | |
f646968f PM |
6191 | if (((dev->hw_features | dev->features) & |
6192 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
6193 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
6194 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
6195 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
6196 | ret = -EINVAL; | |
6197 | goto err_uninit; | |
6198 | } | |
6199 | ||
9c7dafbf PE |
6200 | ret = -EBUSY; |
6201 | if (!dev->ifindex) | |
6202 | dev->ifindex = dev_new_index(net); | |
6203 | else if (__dev_get_by_index(net, dev->ifindex)) | |
6204 | goto err_uninit; | |
6205 | ||
1da177e4 LT |
6206 | if (dev->iflink == -1) |
6207 | dev->iflink = dev->ifindex; | |
6208 | ||
5455c699 MM |
6209 | /* Transfer changeable features to wanted_features and enable |
6210 | * software offloads (GSO and GRO). | |
6211 | */ | |
6212 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
6213 | dev->features |= NETIF_F_SOFT_FEATURES; |
6214 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 6215 | |
34324dc2 MM |
6216 | if (!(dev->flags & IFF_LOOPBACK)) { |
6217 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
c6e1a0d1 TH |
6218 | } |
6219 | ||
1180e7d6 | 6220 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 6221 | */ |
1180e7d6 | 6222 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 6223 | |
ee579677 PS |
6224 | /* Make NETIF_F_SG inheritable to tunnel devices. |
6225 | */ | |
6226 | dev->hw_enc_features |= NETIF_F_SG; | |
6227 | ||
0d89d203 SH |
6228 | /* Make NETIF_F_SG inheritable to MPLS. |
6229 | */ | |
6230 | dev->mpls_features |= NETIF_F_SG; | |
6231 | ||
7ffbe3fd JB |
6232 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
6233 | ret = notifier_to_errno(ret); | |
6234 | if (ret) | |
6235 | goto err_uninit; | |
6236 | ||
8b41d188 | 6237 | ret = netdev_register_kobject(dev); |
b17a7c17 | 6238 | if (ret) |
7ce1b0ed | 6239 | goto err_uninit; |
b17a7c17 SH |
6240 | dev->reg_state = NETREG_REGISTERED; |
6241 | ||
6cb6a27c | 6242 | __netdev_update_features(dev); |
8e9b59b2 | 6243 | |
1da177e4 LT |
6244 | /* |
6245 | * Default initial state at registry is that the | |
6246 | * device is present. | |
6247 | */ | |
6248 | ||
6249 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6250 | ||
8f4cccbb BH |
6251 | linkwatch_init_dev(dev); |
6252 | ||
1da177e4 | 6253 | dev_init_scheduler(dev); |
1da177e4 | 6254 | dev_hold(dev); |
ce286d32 | 6255 | list_netdevice(dev); |
7bf23575 | 6256 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 6257 | |
948b337e JP |
6258 | /* If the device has permanent device address, driver should |
6259 | * set dev_addr and also addr_assign_type should be set to | |
6260 | * NET_ADDR_PERM (default value). | |
6261 | */ | |
6262 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
6263 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
6264 | ||
1da177e4 | 6265 | /* Notify protocols, that a new device appeared. */ |
056925ab | 6266 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 6267 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
6268 | if (ret) { |
6269 | rollback_registered(dev); | |
6270 | dev->reg_state = NETREG_UNREGISTERED; | |
6271 | } | |
d90a909e EB |
6272 | /* |
6273 | * Prevent userspace races by waiting until the network | |
6274 | * device is fully setup before sending notifications. | |
6275 | */ | |
a2835763 PM |
6276 | if (!dev->rtnl_link_ops || |
6277 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 6278 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
6279 | |
6280 | out: | |
6281 | return ret; | |
7ce1b0ed HX |
6282 | |
6283 | err_uninit: | |
d314774c SH |
6284 | if (dev->netdev_ops->ndo_uninit) |
6285 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 6286 | goto out; |
1da177e4 | 6287 | } |
d1b19dff | 6288 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 6289 | |
937f1ba5 BH |
6290 | /** |
6291 | * init_dummy_netdev - init a dummy network device for NAPI | |
6292 | * @dev: device to init | |
6293 | * | |
6294 | * This takes a network device structure and initialize the minimum | |
6295 | * amount of fields so it can be used to schedule NAPI polls without | |
6296 | * registering a full blown interface. This is to be used by drivers | |
6297 | * that need to tie several hardware interfaces to a single NAPI | |
6298 | * poll scheduler due to HW limitations. | |
6299 | */ | |
6300 | int init_dummy_netdev(struct net_device *dev) | |
6301 | { | |
6302 | /* Clear everything. Note we don't initialize spinlocks | |
6303 | * are they aren't supposed to be taken by any of the | |
6304 | * NAPI code and this dummy netdev is supposed to be | |
6305 | * only ever used for NAPI polls | |
6306 | */ | |
6307 | memset(dev, 0, sizeof(struct net_device)); | |
6308 | ||
6309 | /* make sure we BUG if trying to hit standard | |
6310 | * register/unregister code path | |
6311 | */ | |
6312 | dev->reg_state = NETREG_DUMMY; | |
6313 | ||
937f1ba5 BH |
6314 | /* NAPI wants this */ |
6315 | INIT_LIST_HEAD(&dev->napi_list); | |
6316 | ||
6317 | /* a dummy interface is started by default */ | |
6318 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6319 | set_bit(__LINK_STATE_START, &dev->state); | |
6320 | ||
29b4433d ED |
6321 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
6322 | * because users of this 'device' dont need to change | |
6323 | * its refcount. | |
6324 | */ | |
6325 | ||
937f1ba5 BH |
6326 | return 0; |
6327 | } | |
6328 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
6329 | ||
6330 | ||
1da177e4 LT |
6331 | /** |
6332 | * register_netdev - register a network device | |
6333 | * @dev: device to register | |
6334 | * | |
6335 | * Take a completed network device structure and add it to the kernel | |
6336 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6337 | * chain. 0 is returned on success. A negative errno code is returned | |
6338 | * on a failure to set up the device, or if the name is a duplicate. | |
6339 | * | |
38b4da38 | 6340 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
6341 | * and expands the device name if you passed a format string to |
6342 | * alloc_netdev. | |
6343 | */ | |
6344 | int register_netdev(struct net_device *dev) | |
6345 | { | |
6346 | int err; | |
6347 | ||
6348 | rtnl_lock(); | |
1da177e4 | 6349 | err = register_netdevice(dev); |
1da177e4 LT |
6350 | rtnl_unlock(); |
6351 | return err; | |
6352 | } | |
6353 | EXPORT_SYMBOL(register_netdev); | |
6354 | ||
29b4433d ED |
6355 | int netdev_refcnt_read(const struct net_device *dev) |
6356 | { | |
6357 | int i, refcnt = 0; | |
6358 | ||
6359 | for_each_possible_cpu(i) | |
6360 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
6361 | return refcnt; | |
6362 | } | |
6363 | EXPORT_SYMBOL(netdev_refcnt_read); | |
6364 | ||
2c53040f | 6365 | /** |
1da177e4 | 6366 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 6367 | * @dev: target net_device |
1da177e4 LT |
6368 | * |
6369 | * This is called when unregistering network devices. | |
6370 | * | |
6371 | * Any protocol or device that holds a reference should register | |
6372 | * for netdevice notification, and cleanup and put back the | |
6373 | * reference if they receive an UNREGISTER event. | |
6374 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 6375 | * call dev_put. |
1da177e4 LT |
6376 | */ |
6377 | static void netdev_wait_allrefs(struct net_device *dev) | |
6378 | { | |
6379 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 6380 | int refcnt; |
1da177e4 | 6381 | |
e014debe ED |
6382 | linkwatch_forget_dev(dev); |
6383 | ||
1da177e4 | 6384 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
6385 | refcnt = netdev_refcnt_read(dev); |
6386 | ||
6387 | while (refcnt != 0) { | |
1da177e4 | 6388 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 6389 | rtnl_lock(); |
1da177e4 LT |
6390 | |
6391 | /* Rebroadcast unregister notification */ | |
056925ab | 6392 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 6393 | |
748e2d93 | 6394 | __rtnl_unlock(); |
0115e8e3 | 6395 | rcu_barrier(); |
748e2d93 ED |
6396 | rtnl_lock(); |
6397 | ||
0115e8e3 | 6398 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
6399 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
6400 | &dev->state)) { | |
6401 | /* We must not have linkwatch events | |
6402 | * pending on unregister. If this | |
6403 | * happens, we simply run the queue | |
6404 | * unscheduled, resulting in a noop | |
6405 | * for this device. | |
6406 | */ | |
6407 | linkwatch_run_queue(); | |
6408 | } | |
6409 | ||
6756ae4b | 6410 | __rtnl_unlock(); |
1da177e4 LT |
6411 | |
6412 | rebroadcast_time = jiffies; | |
6413 | } | |
6414 | ||
6415 | msleep(250); | |
6416 | ||
29b4433d ED |
6417 | refcnt = netdev_refcnt_read(dev); |
6418 | ||
1da177e4 | 6419 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
6420 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
6421 | dev->name, refcnt); | |
1da177e4 LT |
6422 | warning_time = jiffies; |
6423 | } | |
6424 | } | |
6425 | } | |
6426 | ||
6427 | /* The sequence is: | |
6428 | * | |
6429 | * rtnl_lock(); | |
6430 | * ... | |
6431 | * register_netdevice(x1); | |
6432 | * register_netdevice(x2); | |
6433 | * ... | |
6434 | * unregister_netdevice(y1); | |
6435 | * unregister_netdevice(y2); | |
6436 | * ... | |
6437 | * rtnl_unlock(); | |
6438 | * free_netdev(y1); | |
6439 | * free_netdev(y2); | |
6440 | * | |
58ec3b4d | 6441 | * We are invoked by rtnl_unlock(). |
1da177e4 | 6442 | * This allows us to deal with problems: |
b17a7c17 | 6443 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
6444 | * without deadlocking with linkwatch via keventd. |
6445 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
6446 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
6447 | * |
6448 | * We must not return until all unregister events added during | |
6449 | * the interval the lock was held have been completed. | |
1da177e4 | 6450 | */ |
1da177e4 LT |
6451 | void netdev_run_todo(void) |
6452 | { | |
626ab0e6 | 6453 | struct list_head list; |
1da177e4 | 6454 | |
1da177e4 | 6455 | /* Snapshot list, allow later requests */ |
626ab0e6 | 6456 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
6457 | |
6458 | __rtnl_unlock(); | |
626ab0e6 | 6459 | |
0115e8e3 ED |
6460 | |
6461 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
6462 | if (!list_empty(&list)) |
6463 | rcu_barrier(); | |
6464 | ||
1da177e4 LT |
6465 | while (!list_empty(&list)) { |
6466 | struct net_device *dev | |
e5e26d75 | 6467 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
6468 | list_del(&dev->todo_list); |
6469 | ||
748e2d93 | 6470 | rtnl_lock(); |
0115e8e3 | 6471 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 6472 | __rtnl_unlock(); |
0115e8e3 | 6473 | |
b17a7c17 | 6474 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 6475 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
6476 | dev->name, dev->reg_state); |
6477 | dump_stack(); | |
6478 | continue; | |
6479 | } | |
1da177e4 | 6480 | |
b17a7c17 | 6481 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 6482 | |
152102c7 | 6483 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 6484 | |
b17a7c17 | 6485 | netdev_wait_allrefs(dev); |
1da177e4 | 6486 | |
b17a7c17 | 6487 | /* paranoia */ |
29b4433d | 6488 | BUG_ON(netdev_refcnt_read(dev)); |
33d480ce ED |
6489 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
6490 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 6491 | WARN_ON(dev->dn_ptr); |
1da177e4 | 6492 | |
b17a7c17 SH |
6493 | if (dev->destructor) |
6494 | dev->destructor(dev); | |
9093bbb2 | 6495 | |
50624c93 EB |
6496 | /* Report a network device has been unregistered */ |
6497 | rtnl_lock(); | |
6498 | dev_net(dev)->dev_unreg_count--; | |
6499 | __rtnl_unlock(); | |
6500 | wake_up(&netdev_unregistering_wq); | |
6501 | ||
9093bbb2 SH |
6502 | /* Free network device */ |
6503 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 6504 | } |
1da177e4 LT |
6505 | } |
6506 | ||
3cfde79c BH |
6507 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
6508 | * fields in the same order, with only the type differing. | |
6509 | */ | |
77a1abf5 ED |
6510 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
6511 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
6512 | { |
6513 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
6514 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
6515 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
6516 | #else |
6517 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
6518 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
6519 | u64 *dst = (u64 *)stats64; | |
6520 | ||
6521 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
6522 | sizeof(*stats64) / sizeof(u64)); | |
6523 | for (i = 0; i < n; i++) | |
6524 | dst[i] = src[i]; | |
6525 | #endif | |
6526 | } | |
77a1abf5 | 6527 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 6528 | |
eeda3fd6 SH |
6529 | /** |
6530 | * dev_get_stats - get network device statistics | |
6531 | * @dev: device to get statistics from | |
28172739 | 6532 | * @storage: place to store stats |
eeda3fd6 | 6533 | * |
d7753516 BH |
6534 | * Get network statistics from device. Return @storage. |
6535 | * The device driver may provide its own method by setting | |
6536 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
6537 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 6538 | */ |
d7753516 BH |
6539 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
6540 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 6541 | { |
eeda3fd6 SH |
6542 | const struct net_device_ops *ops = dev->netdev_ops; |
6543 | ||
28172739 ED |
6544 | if (ops->ndo_get_stats64) { |
6545 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
6546 | ops->ndo_get_stats64(dev, storage); |
6547 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 6548 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
6549 | } else { |
6550 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 6551 | } |
caf586e5 | 6552 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
015f0688 | 6553 | storage->tx_dropped += atomic_long_read(&dev->tx_dropped); |
28172739 | 6554 | return storage; |
c45d286e | 6555 | } |
eeda3fd6 | 6556 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 6557 | |
24824a09 | 6558 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 6559 | { |
24824a09 | 6560 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 6561 | |
24824a09 ED |
6562 | #ifdef CONFIG_NET_CLS_ACT |
6563 | if (queue) | |
6564 | return queue; | |
6565 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
6566 | if (!queue) | |
6567 | return NULL; | |
6568 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
6569 | queue->qdisc = &noop_qdisc; |
6570 | queue->qdisc_sleeping = &noop_qdisc; | |
6571 | rcu_assign_pointer(dev->ingress_queue, queue); | |
6572 | #endif | |
6573 | return queue; | |
bb949fbd DM |
6574 | } |
6575 | ||
2c60db03 ED |
6576 | static const struct ethtool_ops default_ethtool_ops; |
6577 | ||
d07d7507 SG |
6578 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
6579 | const struct ethtool_ops *ops) | |
6580 | { | |
6581 | if (dev->ethtool_ops == &default_ethtool_ops) | |
6582 | dev->ethtool_ops = ops; | |
6583 | } | |
6584 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
6585 | ||
74d332c1 ED |
6586 | void netdev_freemem(struct net_device *dev) |
6587 | { | |
6588 | char *addr = (char *)dev - dev->padded; | |
6589 | ||
4cb28970 | 6590 | kvfree(addr); |
74d332c1 ED |
6591 | } |
6592 | ||
1da177e4 | 6593 | /** |
36909ea4 | 6594 | * alloc_netdev_mqs - allocate network device |
c835a677 TG |
6595 | * @sizeof_priv: size of private data to allocate space for |
6596 | * @name: device name format string | |
6597 | * @name_assign_type: origin of device name | |
6598 | * @setup: callback to initialize device | |
6599 | * @txqs: the number of TX subqueues to allocate | |
6600 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
6601 | * |
6602 | * Allocates a struct net_device with private data area for driver use | |
90e51adf | 6603 | * and performs basic initialization. Also allocates subqueue structs |
36909ea4 | 6604 | * for each queue on the device. |
1da177e4 | 6605 | */ |
36909ea4 | 6606 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 6607 | unsigned char name_assign_type, |
36909ea4 TH |
6608 | void (*setup)(struct net_device *), |
6609 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 6610 | { |
1da177e4 | 6611 | struct net_device *dev; |
7943986c | 6612 | size_t alloc_size; |
1ce8e7b5 | 6613 | struct net_device *p; |
1da177e4 | 6614 | |
b6fe17d6 SH |
6615 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
6616 | ||
36909ea4 | 6617 | if (txqs < 1) { |
7b6cd1ce | 6618 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
6619 | return NULL; |
6620 | } | |
6621 | ||
a953be53 | 6622 | #ifdef CONFIG_SYSFS |
36909ea4 | 6623 | if (rxqs < 1) { |
7b6cd1ce | 6624 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
6625 | return NULL; |
6626 | } | |
6627 | #endif | |
6628 | ||
fd2ea0a7 | 6629 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
6630 | if (sizeof_priv) { |
6631 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 6632 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
6633 | alloc_size += sizeof_priv; |
6634 | } | |
6635 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 6636 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 6637 | |
74d332c1 ED |
6638 | p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6639 | if (!p) | |
6640 | p = vzalloc(alloc_size); | |
62b5942a | 6641 | if (!p) |
1da177e4 | 6642 | return NULL; |
1da177e4 | 6643 | |
1ce8e7b5 | 6644 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 6645 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 6646 | |
29b4433d ED |
6647 | dev->pcpu_refcnt = alloc_percpu(int); |
6648 | if (!dev->pcpu_refcnt) | |
74d332c1 | 6649 | goto free_dev; |
ab9c73cc | 6650 | |
ab9c73cc | 6651 | if (dev_addr_init(dev)) |
29b4433d | 6652 | goto free_pcpu; |
ab9c73cc | 6653 | |
22bedad3 | 6654 | dev_mc_init(dev); |
a748ee24 | 6655 | dev_uc_init(dev); |
ccffad25 | 6656 | |
c346dca1 | 6657 | dev_net_set(dev, &init_net); |
1da177e4 | 6658 | |
8d3bdbd5 | 6659 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 6660 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 6661 | |
8d3bdbd5 DM |
6662 | INIT_LIST_HEAD(&dev->napi_list); |
6663 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 6664 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 6665 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
6666 | INIT_LIST_HEAD(&dev->adj_list.upper); |
6667 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
6668 | INIT_LIST_HEAD(&dev->all_adj_list.upper); | |
6669 | INIT_LIST_HEAD(&dev->all_adj_list.lower); | |
8d3bdbd5 DM |
6670 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
6671 | setup(dev); | |
6672 | ||
36909ea4 TH |
6673 | dev->num_tx_queues = txqs; |
6674 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 6675 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 6676 | goto free_all; |
e8a0464c | 6677 | |
a953be53 | 6678 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
6679 | dev->num_rx_queues = rxqs; |
6680 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 6681 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 6682 | goto free_all; |
df334545 | 6683 | #endif |
0a9627f2 | 6684 | |
1da177e4 | 6685 | strcpy(dev->name, name); |
c835a677 | 6686 | dev->name_assign_type = name_assign_type; |
cbda10fa | 6687 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
6688 | if (!dev->ethtool_ops) |
6689 | dev->ethtool_ops = &default_ethtool_ops; | |
1da177e4 | 6690 | return dev; |
ab9c73cc | 6691 | |
8d3bdbd5 DM |
6692 | free_all: |
6693 | free_netdev(dev); | |
6694 | return NULL; | |
6695 | ||
29b4433d ED |
6696 | free_pcpu: |
6697 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
6698 | free_dev: |
6699 | netdev_freemem(dev); | |
ab9c73cc | 6700 | return NULL; |
1da177e4 | 6701 | } |
36909ea4 | 6702 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
6703 | |
6704 | /** | |
6705 | * free_netdev - free network device | |
6706 | * @dev: device | |
6707 | * | |
4ec93edb YH |
6708 | * This function does the last stage of destroying an allocated device |
6709 | * interface. The reference to the device object is released. | |
1da177e4 LT |
6710 | * If this is the last reference then it will be freed. |
6711 | */ | |
6712 | void free_netdev(struct net_device *dev) | |
6713 | { | |
d565b0a1 HX |
6714 | struct napi_struct *p, *n; |
6715 | ||
f3005d7f DL |
6716 | release_net(dev_net(dev)); |
6717 | ||
60877a32 | 6718 | netif_free_tx_queues(dev); |
a953be53 | 6719 | #ifdef CONFIG_SYSFS |
fe822240 TH |
6720 | kfree(dev->_rx); |
6721 | #endif | |
e8a0464c | 6722 | |
33d480ce | 6723 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 6724 | |
f001fde5 JP |
6725 | /* Flush device addresses */ |
6726 | dev_addr_flush(dev); | |
6727 | ||
d565b0a1 HX |
6728 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
6729 | netif_napi_del(p); | |
6730 | ||
29b4433d ED |
6731 | free_percpu(dev->pcpu_refcnt); |
6732 | dev->pcpu_refcnt = NULL; | |
6733 | ||
3041a069 | 6734 | /* Compatibility with error handling in drivers */ |
1da177e4 | 6735 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 6736 | netdev_freemem(dev); |
1da177e4 LT |
6737 | return; |
6738 | } | |
6739 | ||
6740 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
6741 | dev->reg_state = NETREG_RELEASED; | |
6742 | ||
43cb76d9 GKH |
6743 | /* will free via device release */ |
6744 | put_device(&dev->dev); | |
1da177e4 | 6745 | } |
d1b19dff | 6746 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 6747 | |
f0db275a SH |
6748 | /** |
6749 | * synchronize_net - Synchronize with packet receive processing | |
6750 | * | |
6751 | * Wait for packets currently being received to be done. | |
6752 | * Does not block later packets from starting. | |
6753 | */ | |
4ec93edb | 6754 | void synchronize_net(void) |
1da177e4 LT |
6755 | { |
6756 | might_sleep(); | |
be3fc413 ED |
6757 | if (rtnl_is_locked()) |
6758 | synchronize_rcu_expedited(); | |
6759 | else | |
6760 | synchronize_rcu(); | |
1da177e4 | 6761 | } |
d1b19dff | 6762 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
6763 | |
6764 | /** | |
44a0873d | 6765 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 6766 | * @dev: device |
44a0873d | 6767 | * @head: list |
6ebfbc06 | 6768 | * |
1da177e4 | 6769 | * This function shuts down a device interface and removes it |
d59b54b1 | 6770 | * from the kernel tables. |
44a0873d | 6771 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
6772 | * |
6773 | * Callers must hold the rtnl semaphore. You may want | |
6774 | * unregister_netdev() instead of this. | |
6775 | */ | |
6776 | ||
44a0873d | 6777 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 6778 | { |
a6620712 HX |
6779 | ASSERT_RTNL(); |
6780 | ||
44a0873d | 6781 | if (head) { |
9fdce099 | 6782 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
6783 | } else { |
6784 | rollback_registered(dev); | |
6785 | /* Finish processing unregister after unlock */ | |
6786 | net_set_todo(dev); | |
6787 | } | |
1da177e4 | 6788 | } |
44a0873d | 6789 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 6790 | |
9b5e383c ED |
6791 | /** |
6792 | * unregister_netdevice_many - unregister many devices | |
6793 | * @head: list of devices | |
87757a91 ED |
6794 | * |
6795 | * Note: As most callers use a stack allocated list_head, | |
6796 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
6797 | */ |
6798 | void unregister_netdevice_many(struct list_head *head) | |
6799 | { | |
6800 | struct net_device *dev; | |
6801 | ||
6802 | if (!list_empty(head)) { | |
6803 | rollback_registered_many(head); | |
6804 | list_for_each_entry(dev, head, unreg_list) | |
6805 | net_set_todo(dev); | |
87757a91 | 6806 | list_del(head); |
9b5e383c ED |
6807 | } |
6808 | } | |
63c8099d | 6809 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 6810 | |
1da177e4 LT |
6811 | /** |
6812 | * unregister_netdev - remove device from the kernel | |
6813 | * @dev: device | |
6814 | * | |
6815 | * This function shuts down a device interface and removes it | |
d59b54b1 | 6816 | * from the kernel tables. |
1da177e4 LT |
6817 | * |
6818 | * This is just a wrapper for unregister_netdevice that takes | |
6819 | * the rtnl semaphore. In general you want to use this and not | |
6820 | * unregister_netdevice. | |
6821 | */ | |
6822 | void unregister_netdev(struct net_device *dev) | |
6823 | { | |
6824 | rtnl_lock(); | |
6825 | unregister_netdevice(dev); | |
6826 | rtnl_unlock(); | |
6827 | } | |
1da177e4 LT |
6828 | EXPORT_SYMBOL(unregister_netdev); |
6829 | ||
ce286d32 EB |
6830 | /** |
6831 | * dev_change_net_namespace - move device to different nethost namespace | |
6832 | * @dev: device | |
6833 | * @net: network namespace | |
6834 | * @pat: If not NULL name pattern to try if the current device name | |
6835 | * is already taken in the destination network namespace. | |
6836 | * | |
6837 | * This function shuts down a device interface and moves it | |
6838 | * to a new network namespace. On success 0 is returned, on | |
6839 | * a failure a netagive errno code is returned. | |
6840 | * | |
6841 | * Callers must hold the rtnl semaphore. | |
6842 | */ | |
6843 | ||
6844 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
6845 | { | |
ce286d32 EB |
6846 | int err; |
6847 | ||
6848 | ASSERT_RTNL(); | |
6849 | ||
6850 | /* Don't allow namespace local devices to be moved. */ | |
6851 | err = -EINVAL; | |
6852 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6853 | goto out; | |
6854 | ||
6855 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
6856 | if (dev->reg_state != NETREG_REGISTERED) |
6857 | goto out; | |
6858 | ||
6859 | /* Get out if there is nothing todo */ | |
6860 | err = 0; | |
878628fb | 6861 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
6862 | goto out; |
6863 | ||
6864 | /* Pick the destination device name, and ensure | |
6865 | * we can use it in the destination network namespace. | |
6866 | */ | |
6867 | err = -EEXIST; | |
d9031024 | 6868 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
6869 | /* We get here if we can't use the current device name */ |
6870 | if (!pat) | |
6871 | goto out; | |
828de4f6 | 6872 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
6873 | goto out; |
6874 | } | |
6875 | ||
6876 | /* | |
6877 | * And now a mini version of register_netdevice unregister_netdevice. | |
6878 | */ | |
6879 | ||
6880 | /* If device is running close it first. */ | |
9b772652 | 6881 | dev_close(dev); |
ce286d32 EB |
6882 | |
6883 | /* And unlink it from device chain */ | |
6884 | err = -ENODEV; | |
6885 | unlist_netdevice(dev); | |
6886 | ||
6887 | synchronize_net(); | |
6888 | ||
6889 | /* Shutdown queueing discipline. */ | |
6890 | dev_shutdown(dev); | |
6891 | ||
6892 | /* Notify protocols, that we are about to destroy | |
6893 | this device. They should clean all the things. | |
3b27e105 DL |
6894 | |
6895 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
6896 | This is wanted because this way 8021q and macvlan know | |
6897 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
6898 | */ |
6899 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
6900 | rcu_barrier(); |
6901 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 6902 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
6903 | |
6904 | /* | |
6905 | * Flush the unicast and multicast chains | |
6906 | */ | |
a748ee24 | 6907 | dev_uc_flush(dev); |
22bedad3 | 6908 | dev_mc_flush(dev); |
ce286d32 | 6909 | |
4e66ae2e SH |
6910 | /* Send a netdev-removed uevent to the old namespace */ |
6911 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 6912 | netdev_adjacent_del_links(dev); |
4e66ae2e | 6913 | |
ce286d32 | 6914 | /* Actually switch the network namespace */ |
c346dca1 | 6915 | dev_net_set(dev, net); |
ce286d32 | 6916 | |
ce286d32 EB |
6917 | /* If there is an ifindex conflict assign a new one */ |
6918 | if (__dev_get_by_index(net, dev->ifindex)) { | |
6919 | int iflink = (dev->iflink == dev->ifindex); | |
6920 | dev->ifindex = dev_new_index(net); | |
6921 | if (iflink) | |
6922 | dev->iflink = dev->ifindex; | |
6923 | } | |
6924 | ||
4e66ae2e SH |
6925 | /* Send a netdev-add uevent to the new namespace */ |
6926 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 6927 | netdev_adjacent_add_links(dev); |
4e66ae2e | 6928 | |
8b41d188 | 6929 | /* Fixup kobjects */ |
a1b3f594 | 6930 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 6931 | WARN_ON(err); |
ce286d32 EB |
6932 | |
6933 | /* Add the device back in the hashes */ | |
6934 | list_netdevice(dev); | |
6935 | ||
6936 | /* Notify protocols, that a new device appeared. */ | |
6937 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
6938 | ||
d90a909e EB |
6939 | /* |
6940 | * Prevent userspace races by waiting until the network | |
6941 | * device is fully setup before sending notifications. | |
6942 | */ | |
7f294054 | 6943 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 6944 | |
ce286d32 EB |
6945 | synchronize_net(); |
6946 | err = 0; | |
6947 | out: | |
6948 | return err; | |
6949 | } | |
463d0183 | 6950 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 6951 | |
1da177e4 LT |
6952 | static int dev_cpu_callback(struct notifier_block *nfb, |
6953 | unsigned long action, | |
6954 | void *ocpu) | |
6955 | { | |
6956 | struct sk_buff **list_skb; | |
1da177e4 LT |
6957 | struct sk_buff *skb; |
6958 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
6959 | struct softnet_data *sd, *oldsd; | |
6960 | ||
8bb78442 | 6961 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
6962 | return NOTIFY_OK; |
6963 | ||
6964 | local_irq_disable(); | |
6965 | cpu = smp_processor_id(); | |
6966 | sd = &per_cpu(softnet_data, cpu); | |
6967 | oldsd = &per_cpu(softnet_data, oldcpu); | |
6968 | ||
6969 | /* Find end of our completion_queue. */ | |
6970 | list_skb = &sd->completion_queue; | |
6971 | while (*list_skb) | |
6972 | list_skb = &(*list_skb)->next; | |
6973 | /* Append completion queue from offline CPU. */ | |
6974 | *list_skb = oldsd->completion_queue; | |
6975 | oldsd->completion_queue = NULL; | |
6976 | ||
1da177e4 | 6977 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
6978 | if (oldsd->output_queue) { |
6979 | *sd->output_queue_tailp = oldsd->output_queue; | |
6980 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
6981 | oldsd->output_queue = NULL; | |
6982 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
6983 | } | |
264524d5 HC |
6984 | /* Append NAPI poll list from offline CPU. */ |
6985 | if (!list_empty(&oldsd->poll_list)) { | |
6986 | list_splice_init(&oldsd->poll_list, &sd->poll_list); | |
6987 | raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6988 | } | |
1da177e4 LT |
6989 | |
6990 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
6991 | local_irq_enable(); | |
6992 | ||
6993 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 6994 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
ae78dbfa | 6995 | netif_rx_internal(skb); |
76cc8b13 | 6996 | input_queue_head_incr(oldsd); |
fec5e652 | 6997 | } |
76cc8b13 | 6998 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
ae78dbfa | 6999 | netif_rx_internal(skb); |
76cc8b13 TH |
7000 | input_queue_head_incr(oldsd); |
7001 | } | |
1da177e4 LT |
7002 | |
7003 | return NOTIFY_OK; | |
7004 | } | |
1da177e4 LT |
7005 | |
7006 | ||
7f353bf2 | 7007 | /** |
b63365a2 HX |
7008 | * netdev_increment_features - increment feature set by one |
7009 | * @all: current feature set | |
7010 | * @one: new feature set | |
7011 | * @mask: mask feature set | |
7f353bf2 HX |
7012 | * |
7013 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
7014 | * @one to the master device with current feature set @all. Will not |
7015 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 7016 | */ |
c8f44aff MM |
7017 | netdev_features_t netdev_increment_features(netdev_features_t all, |
7018 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 7019 | { |
1742f183 MM |
7020 | if (mask & NETIF_F_GEN_CSUM) |
7021 | mask |= NETIF_F_ALL_CSUM; | |
7022 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 7023 | |
1742f183 MM |
7024 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
7025 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 7026 | |
1742f183 MM |
7027 | /* If one device supports hw checksumming, set for all. */ |
7028 | if (all & NETIF_F_GEN_CSUM) | |
7029 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
7030 | |
7031 | return all; | |
7032 | } | |
b63365a2 | 7033 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 7034 | |
430f03cd | 7035 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
7036 | { |
7037 | int i; | |
7038 | struct hlist_head *hash; | |
7039 | ||
7040 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
7041 | if (hash != NULL) | |
7042 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
7043 | INIT_HLIST_HEAD(&hash[i]); | |
7044 | ||
7045 | return hash; | |
7046 | } | |
7047 | ||
881d966b | 7048 | /* Initialize per network namespace state */ |
4665079c | 7049 | static int __net_init netdev_init(struct net *net) |
881d966b | 7050 | { |
734b6541 RM |
7051 | if (net != &init_net) |
7052 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 7053 | |
30d97d35 PE |
7054 | net->dev_name_head = netdev_create_hash(); |
7055 | if (net->dev_name_head == NULL) | |
7056 | goto err_name; | |
881d966b | 7057 | |
30d97d35 PE |
7058 | net->dev_index_head = netdev_create_hash(); |
7059 | if (net->dev_index_head == NULL) | |
7060 | goto err_idx; | |
881d966b EB |
7061 | |
7062 | return 0; | |
30d97d35 PE |
7063 | |
7064 | err_idx: | |
7065 | kfree(net->dev_name_head); | |
7066 | err_name: | |
7067 | return -ENOMEM; | |
881d966b EB |
7068 | } |
7069 | ||
f0db275a SH |
7070 | /** |
7071 | * netdev_drivername - network driver for the device | |
7072 | * @dev: network device | |
f0db275a SH |
7073 | * |
7074 | * Determine network driver for device. | |
7075 | */ | |
3019de12 | 7076 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 7077 | { |
cf04a4c7 SH |
7078 | const struct device_driver *driver; |
7079 | const struct device *parent; | |
3019de12 | 7080 | const char *empty = ""; |
6579e57b AV |
7081 | |
7082 | parent = dev->dev.parent; | |
6579e57b | 7083 | if (!parent) |
3019de12 | 7084 | return empty; |
6579e57b AV |
7085 | |
7086 | driver = parent->driver; | |
7087 | if (driver && driver->name) | |
3019de12 DM |
7088 | return driver->name; |
7089 | return empty; | |
6579e57b AV |
7090 | } |
7091 | ||
6ea754eb JP |
7092 | static void __netdev_printk(const char *level, const struct net_device *dev, |
7093 | struct va_format *vaf) | |
256df2f3 | 7094 | { |
b004ff49 | 7095 | if (dev && dev->dev.parent) { |
6ea754eb JP |
7096 | dev_printk_emit(level[1] - '0', |
7097 | dev->dev.parent, | |
7098 | "%s %s %s%s: %pV", | |
7099 | dev_driver_string(dev->dev.parent), | |
7100 | dev_name(dev->dev.parent), | |
7101 | netdev_name(dev), netdev_reg_state(dev), | |
7102 | vaf); | |
b004ff49 | 7103 | } else if (dev) { |
6ea754eb JP |
7104 | printk("%s%s%s: %pV", |
7105 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 7106 | } else { |
6ea754eb | 7107 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 7108 | } |
256df2f3 JP |
7109 | } |
7110 | ||
6ea754eb JP |
7111 | void netdev_printk(const char *level, const struct net_device *dev, |
7112 | const char *format, ...) | |
256df2f3 JP |
7113 | { |
7114 | struct va_format vaf; | |
7115 | va_list args; | |
256df2f3 JP |
7116 | |
7117 | va_start(args, format); | |
7118 | ||
7119 | vaf.fmt = format; | |
7120 | vaf.va = &args; | |
7121 | ||
6ea754eb | 7122 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 7123 | |
256df2f3 | 7124 | va_end(args); |
256df2f3 JP |
7125 | } |
7126 | EXPORT_SYMBOL(netdev_printk); | |
7127 | ||
7128 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 7129 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 7130 | { \ |
256df2f3 JP |
7131 | struct va_format vaf; \ |
7132 | va_list args; \ | |
7133 | \ | |
7134 | va_start(args, fmt); \ | |
7135 | \ | |
7136 | vaf.fmt = fmt; \ | |
7137 | vaf.va = &args; \ | |
7138 | \ | |
6ea754eb | 7139 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 7140 | \ |
256df2f3 | 7141 | va_end(args); \ |
256df2f3 JP |
7142 | } \ |
7143 | EXPORT_SYMBOL(func); | |
7144 | ||
7145 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
7146 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
7147 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
7148 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
7149 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
7150 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
7151 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
7152 | ||
4665079c | 7153 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
7154 | { |
7155 | kfree(net->dev_name_head); | |
7156 | kfree(net->dev_index_head); | |
7157 | } | |
7158 | ||
022cbae6 | 7159 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
7160 | .init = netdev_init, |
7161 | .exit = netdev_exit, | |
7162 | }; | |
7163 | ||
4665079c | 7164 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 7165 | { |
e008b5fc | 7166 | struct net_device *dev, *aux; |
ce286d32 | 7167 | /* |
e008b5fc | 7168 | * Push all migratable network devices back to the |
ce286d32 EB |
7169 | * initial network namespace |
7170 | */ | |
7171 | rtnl_lock(); | |
e008b5fc | 7172 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 7173 | int err; |
aca51397 | 7174 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
7175 | |
7176 | /* Ignore unmoveable devices (i.e. loopback) */ | |
7177 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7178 | continue; | |
7179 | ||
e008b5fc EB |
7180 | /* Leave virtual devices for the generic cleanup */ |
7181 | if (dev->rtnl_link_ops) | |
7182 | continue; | |
d0c082ce | 7183 | |
25985edc | 7184 | /* Push remaining network devices to init_net */ |
aca51397 PE |
7185 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
7186 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 7187 | if (err) { |
7b6cd1ce JP |
7188 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
7189 | __func__, dev->name, err); | |
aca51397 | 7190 | BUG(); |
ce286d32 EB |
7191 | } |
7192 | } | |
7193 | rtnl_unlock(); | |
7194 | } | |
7195 | ||
50624c93 EB |
7196 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
7197 | { | |
7198 | /* Return with the rtnl_lock held when there are no network | |
7199 | * devices unregistering in any network namespace in net_list. | |
7200 | */ | |
7201 | struct net *net; | |
7202 | bool unregistering; | |
7203 | DEFINE_WAIT(wait); | |
7204 | ||
7205 | for (;;) { | |
7206 | prepare_to_wait(&netdev_unregistering_wq, &wait, | |
7207 | TASK_UNINTERRUPTIBLE); | |
7208 | unregistering = false; | |
7209 | rtnl_lock(); | |
7210 | list_for_each_entry(net, net_list, exit_list) { | |
7211 | if (net->dev_unreg_count > 0) { | |
7212 | unregistering = true; | |
7213 | break; | |
7214 | } | |
7215 | } | |
7216 | if (!unregistering) | |
7217 | break; | |
7218 | __rtnl_unlock(); | |
7219 | schedule(); | |
7220 | } | |
7221 | finish_wait(&netdev_unregistering_wq, &wait); | |
7222 | } | |
7223 | ||
04dc7f6b EB |
7224 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
7225 | { | |
7226 | /* At exit all network devices most be removed from a network | |
b595076a | 7227 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
7228 | * Do this across as many network namespaces as possible to |
7229 | * improve batching efficiency. | |
7230 | */ | |
7231 | struct net_device *dev; | |
7232 | struct net *net; | |
7233 | LIST_HEAD(dev_kill_list); | |
7234 | ||
50624c93 EB |
7235 | /* To prevent network device cleanup code from dereferencing |
7236 | * loopback devices or network devices that have been freed | |
7237 | * wait here for all pending unregistrations to complete, | |
7238 | * before unregistring the loopback device and allowing the | |
7239 | * network namespace be freed. | |
7240 | * | |
7241 | * The netdev todo list containing all network devices | |
7242 | * unregistrations that happen in default_device_exit_batch | |
7243 | * will run in the rtnl_unlock() at the end of | |
7244 | * default_device_exit_batch. | |
7245 | */ | |
7246 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
7247 | list_for_each_entry(net, net_list, exit_list) { |
7248 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 7249 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
7250 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
7251 | else | |
7252 | unregister_netdevice_queue(dev, &dev_kill_list); | |
7253 | } | |
7254 | } | |
7255 | unregister_netdevice_many(&dev_kill_list); | |
7256 | rtnl_unlock(); | |
7257 | } | |
7258 | ||
022cbae6 | 7259 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 7260 | .exit = default_device_exit, |
04dc7f6b | 7261 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
7262 | }; |
7263 | ||
1da177e4 LT |
7264 | /* |
7265 | * Initialize the DEV module. At boot time this walks the device list and | |
7266 | * unhooks any devices that fail to initialise (normally hardware not | |
7267 | * present) and leaves us with a valid list of present and active devices. | |
7268 | * | |
7269 | */ | |
7270 | ||
7271 | /* | |
7272 | * This is called single threaded during boot, so no need | |
7273 | * to take the rtnl semaphore. | |
7274 | */ | |
7275 | static int __init net_dev_init(void) | |
7276 | { | |
7277 | int i, rc = -ENOMEM; | |
7278 | ||
7279 | BUG_ON(!dev_boot_phase); | |
7280 | ||
1da177e4 LT |
7281 | if (dev_proc_init()) |
7282 | goto out; | |
7283 | ||
8b41d188 | 7284 | if (netdev_kobject_init()) |
1da177e4 LT |
7285 | goto out; |
7286 | ||
7287 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 7288 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
7289 | INIT_LIST_HEAD(&ptype_base[i]); |
7290 | ||
62532da9 VY |
7291 | INIT_LIST_HEAD(&offload_base); |
7292 | ||
881d966b EB |
7293 | if (register_pernet_subsys(&netdev_net_ops)) |
7294 | goto out; | |
1da177e4 LT |
7295 | |
7296 | /* | |
7297 | * Initialise the packet receive queues. | |
7298 | */ | |
7299 | ||
6f912042 | 7300 | for_each_possible_cpu(i) { |
e36fa2f7 | 7301 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 7302 | |
e36fa2f7 | 7303 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 7304 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 7305 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 7306 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 7307 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
7308 | sd->csd.func = rps_trigger_softirq; |
7309 | sd->csd.info = sd; | |
e36fa2f7 | 7310 | sd->cpu = i; |
1e94d72f | 7311 | #endif |
0a9627f2 | 7312 | |
e36fa2f7 ED |
7313 | sd->backlog.poll = process_backlog; |
7314 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
7315 | } |
7316 | ||
1da177e4 LT |
7317 | dev_boot_phase = 0; |
7318 | ||
505d4f73 EB |
7319 | /* The loopback device is special if any other network devices |
7320 | * is present in a network namespace the loopback device must | |
7321 | * be present. Since we now dynamically allocate and free the | |
7322 | * loopback device ensure this invariant is maintained by | |
7323 | * keeping the loopback device as the first device on the | |
7324 | * list of network devices. Ensuring the loopback devices | |
7325 | * is the first device that appears and the last network device | |
7326 | * that disappears. | |
7327 | */ | |
7328 | if (register_pernet_device(&loopback_net_ops)) | |
7329 | goto out; | |
7330 | ||
7331 | if (register_pernet_device(&default_device_ops)) | |
7332 | goto out; | |
7333 | ||
962cf36c CM |
7334 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
7335 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
7336 | |
7337 | hotcpu_notifier(dev_cpu_callback, 0); | |
7338 | dst_init(); | |
1da177e4 LT |
7339 | rc = 0; |
7340 | out: | |
7341 | return rc; | |
7342 | } | |
7343 | ||
7344 | subsys_initcall(net_dev_init); |