]>
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> |
1da177e4 | 132 | |
342709ef PE |
133 | #include "net-sysfs.h" |
134 | ||
d565b0a1 HX |
135 | /* Instead of increasing this, you should create a hash table. */ |
136 | #define MAX_GRO_SKBS 8 | |
137 | ||
5d38a079 HX |
138 | /* This should be increased if a protocol with a bigger head is added. */ |
139 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
140 | ||
1da177e4 | 141 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 142 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
143 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
144 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 145 | static struct list_head offload_base __read_mostly; |
1da177e4 | 146 | |
1da177e4 | 147 | /* |
7562f876 | 148 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
149 | * semaphore. |
150 | * | |
c6d14c84 | 151 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
152 | * |
153 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 154 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
155 | * actual updates. This allows pure readers to access the list even |
156 | * while a writer is preparing to update it. | |
157 | * | |
158 | * To put it another way, dev_base_lock is held for writing only to | |
159 | * protect against pure readers; the rtnl semaphore provides the | |
160 | * protection against other writers. | |
161 | * | |
162 | * See, for example usages, register_netdevice() and | |
163 | * unregister_netdevice(), which must be called with the rtnl | |
164 | * semaphore held. | |
165 | */ | |
1da177e4 | 166 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
167 | EXPORT_SYMBOL(dev_base_lock); |
168 | ||
30e6c9fa | 169 | seqcount_t devnet_rename_seq; |
c91f6df2 | 170 | |
4e985ada TG |
171 | static inline void dev_base_seq_inc(struct net *net) |
172 | { | |
173 | while (++net->dev_base_seq == 0); | |
174 | } | |
175 | ||
881d966b | 176 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 177 | { |
95c96174 ED |
178 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
179 | ||
08e9897d | 180 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
181 | } |
182 | ||
881d966b | 183 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 184 | { |
7c28bd0b | 185 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
186 | } |
187 | ||
e36fa2f7 | 188 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
189 | { |
190 | #ifdef CONFIG_RPS | |
e36fa2f7 | 191 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
192 | #endif |
193 | } | |
194 | ||
e36fa2f7 | 195 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
196 | { |
197 | #ifdef CONFIG_RPS | |
e36fa2f7 | 198 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
199 | #endif |
200 | } | |
201 | ||
ce286d32 EB |
202 | /* Device list insertion */ |
203 | static int list_netdevice(struct net_device *dev) | |
204 | { | |
c346dca1 | 205 | struct net *net = dev_net(dev); |
ce286d32 EB |
206 | |
207 | ASSERT_RTNL(); | |
208 | ||
209 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 210 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 211 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
212 | hlist_add_head_rcu(&dev->index_hlist, |
213 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 214 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
215 | |
216 | dev_base_seq_inc(net); | |
217 | ||
ce286d32 EB |
218 | return 0; |
219 | } | |
220 | ||
fb699dfd ED |
221 | /* Device list removal |
222 | * caller must respect a RCU grace period before freeing/reusing dev | |
223 | */ | |
ce286d32 EB |
224 | static void unlist_netdevice(struct net_device *dev) |
225 | { | |
226 | ASSERT_RTNL(); | |
227 | ||
228 | /* Unlink dev from the device chain */ | |
229 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 230 | list_del_rcu(&dev->dev_list); |
72c9528b | 231 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 232 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 233 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
234 | |
235 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
236 | } |
237 | ||
1da177e4 LT |
238 | /* |
239 | * Our notifier list | |
240 | */ | |
241 | ||
f07d5b94 | 242 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
243 | |
244 | /* | |
245 | * Device drivers call our routines to queue packets here. We empty the | |
246 | * queue in the local softnet handler. | |
247 | */ | |
bea3348e | 248 | |
9958da05 | 249 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 250 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 251 | |
cf508b12 | 252 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 253 | /* |
c773e847 | 254 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
255 | * according to dev->type |
256 | */ | |
257 | static const unsigned short netdev_lock_type[] = | |
258 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
259 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
260 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
261 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
262 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
263 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
264 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
265 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
266 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
267 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
268 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
269 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
270 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
271 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
272 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 273 | |
36cbd3dc | 274 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
275 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
276 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
277 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
278 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
279 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
280 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
281 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
282 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
283 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
284 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
285 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
286 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
287 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
288 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
289 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
290 | |
291 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 292 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
293 | |
294 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
295 | { | |
296 | int i; | |
297 | ||
298 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
299 | if (netdev_lock_type[i] == dev_type) | |
300 | return i; | |
301 | /* the last key is used by default */ | |
302 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
303 | } | |
304 | ||
cf508b12 DM |
305 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
306 | unsigned short dev_type) | |
723e98b7 JP |
307 | { |
308 | int i; | |
309 | ||
310 | i = netdev_lock_pos(dev_type); | |
311 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
312 | netdev_lock_name[i]); | |
313 | } | |
cf508b12 DM |
314 | |
315 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
316 | { | |
317 | int i; | |
318 | ||
319 | i = netdev_lock_pos(dev->type); | |
320 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
321 | &netdev_addr_lock_key[i], | |
322 | netdev_lock_name[i]); | |
323 | } | |
723e98b7 | 324 | #else |
cf508b12 DM |
325 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
326 | unsigned short dev_type) | |
327 | { | |
328 | } | |
329 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
330 | { |
331 | } | |
332 | #endif | |
1da177e4 LT |
333 | |
334 | /******************************************************************************* | |
335 | ||
336 | Protocol management and registration routines | |
337 | ||
338 | *******************************************************************************/ | |
339 | ||
1da177e4 LT |
340 | /* |
341 | * Add a protocol ID to the list. Now that the input handler is | |
342 | * smarter we can dispense with all the messy stuff that used to be | |
343 | * here. | |
344 | * | |
345 | * BEWARE!!! Protocol handlers, mangling input packets, | |
346 | * MUST BE last in hash buckets and checking protocol handlers | |
347 | * MUST start from promiscuous ptype_all chain in net_bh. | |
348 | * It is true now, do not change it. | |
349 | * Explanation follows: if protocol handler, mangling packet, will | |
350 | * be the first on list, it is not able to sense, that packet | |
351 | * is cloned and should be copied-on-write, so that it will | |
352 | * change it and subsequent readers will get broken packet. | |
353 | * --ANK (980803) | |
354 | */ | |
355 | ||
c07b68e8 ED |
356 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
357 | { | |
358 | if (pt->type == htons(ETH_P_ALL)) | |
359 | return &ptype_all; | |
360 | else | |
361 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
362 | } | |
363 | ||
1da177e4 LT |
364 | /** |
365 | * dev_add_pack - add packet handler | |
366 | * @pt: packet type declaration | |
367 | * | |
368 | * Add a protocol handler to the networking stack. The passed &packet_type | |
369 | * is linked into kernel lists and may not be freed until it has been | |
370 | * removed from the kernel lists. | |
371 | * | |
4ec93edb | 372 | * This call does not sleep therefore it can not |
1da177e4 LT |
373 | * guarantee all CPU's that are in middle of receiving packets |
374 | * will see the new packet type (until the next received packet). | |
375 | */ | |
376 | ||
377 | void dev_add_pack(struct packet_type *pt) | |
378 | { | |
c07b68e8 | 379 | struct list_head *head = ptype_head(pt); |
1da177e4 | 380 | |
c07b68e8 ED |
381 | spin_lock(&ptype_lock); |
382 | list_add_rcu(&pt->list, head); | |
383 | spin_unlock(&ptype_lock); | |
1da177e4 | 384 | } |
d1b19dff | 385 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 386 | |
1da177e4 LT |
387 | /** |
388 | * __dev_remove_pack - remove packet handler | |
389 | * @pt: packet type declaration | |
390 | * | |
391 | * Remove a protocol handler that was previously added to the kernel | |
392 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
393 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 394 | * returns. |
1da177e4 LT |
395 | * |
396 | * The packet type might still be in use by receivers | |
397 | * and must not be freed until after all the CPU's have gone | |
398 | * through a quiescent state. | |
399 | */ | |
400 | void __dev_remove_pack(struct packet_type *pt) | |
401 | { | |
c07b68e8 | 402 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
403 | struct packet_type *pt1; |
404 | ||
c07b68e8 | 405 | spin_lock(&ptype_lock); |
1da177e4 LT |
406 | |
407 | list_for_each_entry(pt1, head, list) { | |
408 | if (pt == pt1) { | |
409 | list_del_rcu(&pt->list); | |
410 | goto out; | |
411 | } | |
412 | } | |
413 | ||
7b6cd1ce | 414 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 415 | out: |
c07b68e8 | 416 | spin_unlock(&ptype_lock); |
1da177e4 | 417 | } |
d1b19dff ED |
418 | EXPORT_SYMBOL(__dev_remove_pack); |
419 | ||
1da177e4 LT |
420 | /** |
421 | * dev_remove_pack - remove packet handler | |
422 | * @pt: packet type declaration | |
423 | * | |
424 | * Remove a protocol handler that was previously added to the kernel | |
425 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
426 | * from the kernel lists and can be freed or reused once this function | |
427 | * returns. | |
428 | * | |
429 | * This call sleeps to guarantee that no CPU is looking at the packet | |
430 | * type after return. | |
431 | */ | |
432 | void dev_remove_pack(struct packet_type *pt) | |
433 | { | |
434 | __dev_remove_pack(pt); | |
4ec93edb | 435 | |
1da177e4 LT |
436 | synchronize_net(); |
437 | } | |
d1b19dff | 438 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 439 | |
62532da9 VY |
440 | |
441 | /** | |
442 | * dev_add_offload - register offload handlers | |
443 | * @po: protocol offload declaration | |
444 | * | |
445 | * Add protocol offload handlers to the networking stack. The passed | |
446 | * &proto_offload is linked into kernel lists and may not be freed until | |
447 | * it has been removed from the kernel lists. | |
448 | * | |
449 | * This call does not sleep therefore it can not | |
450 | * guarantee all CPU's that are in middle of receiving packets | |
451 | * will see the new offload handlers (until the next received packet). | |
452 | */ | |
453 | void dev_add_offload(struct packet_offload *po) | |
454 | { | |
455 | struct list_head *head = &offload_base; | |
456 | ||
457 | spin_lock(&offload_lock); | |
458 | list_add_rcu(&po->list, head); | |
459 | spin_unlock(&offload_lock); | |
460 | } | |
461 | EXPORT_SYMBOL(dev_add_offload); | |
462 | ||
463 | /** | |
464 | * __dev_remove_offload - remove offload handler | |
465 | * @po: packet offload declaration | |
466 | * | |
467 | * Remove a protocol offload handler that was previously added to the | |
468 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
469 | * is removed from the kernel lists and can be freed or reused once this | |
470 | * function returns. | |
471 | * | |
472 | * The packet type might still be in use by receivers | |
473 | * and must not be freed until after all the CPU's have gone | |
474 | * through a quiescent state. | |
475 | */ | |
476 | void __dev_remove_offload(struct packet_offload *po) | |
477 | { | |
478 | struct list_head *head = &offload_base; | |
479 | struct packet_offload *po1; | |
480 | ||
c53aa505 | 481 | spin_lock(&offload_lock); |
62532da9 VY |
482 | |
483 | list_for_each_entry(po1, head, list) { | |
484 | if (po == po1) { | |
485 | list_del_rcu(&po->list); | |
486 | goto out; | |
487 | } | |
488 | } | |
489 | ||
490 | pr_warn("dev_remove_offload: %p not found\n", po); | |
491 | out: | |
c53aa505 | 492 | spin_unlock(&offload_lock); |
62532da9 VY |
493 | } |
494 | EXPORT_SYMBOL(__dev_remove_offload); | |
495 | ||
496 | /** | |
497 | * dev_remove_offload - remove packet offload handler | |
498 | * @po: packet offload declaration | |
499 | * | |
500 | * Remove a packet offload handler that was previously added to the kernel | |
501 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
502 | * removed from the kernel lists and can be freed or reused once this | |
503 | * function returns. | |
504 | * | |
505 | * This call sleeps to guarantee that no CPU is looking at the packet | |
506 | * type after return. | |
507 | */ | |
508 | void dev_remove_offload(struct packet_offload *po) | |
509 | { | |
510 | __dev_remove_offload(po); | |
511 | ||
512 | synchronize_net(); | |
513 | } | |
514 | EXPORT_SYMBOL(dev_remove_offload); | |
515 | ||
1da177e4 LT |
516 | /****************************************************************************** |
517 | ||
518 | Device Boot-time Settings Routines | |
519 | ||
520 | *******************************************************************************/ | |
521 | ||
522 | /* Boot time configuration table */ | |
523 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
524 | ||
525 | /** | |
526 | * netdev_boot_setup_add - add new setup entry | |
527 | * @name: name of the device | |
528 | * @map: configured settings for the device | |
529 | * | |
530 | * Adds new setup entry to the dev_boot_setup list. The function | |
531 | * returns 0 on error and 1 on success. This is a generic routine to | |
532 | * all netdevices. | |
533 | */ | |
534 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
535 | { | |
536 | struct netdev_boot_setup *s; | |
537 | int i; | |
538 | ||
539 | s = dev_boot_setup; | |
540 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
541 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
542 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 543 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
544 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
545 | break; | |
546 | } | |
547 | } | |
548 | ||
549 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
550 | } | |
551 | ||
552 | /** | |
553 | * netdev_boot_setup_check - check boot time settings | |
554 | * @dev: the netdevice | |
555 | * | |
556 | * Check boot time settings for the device. | |
557 | * The found settings are set for the device to be used | |
558 | * later in the device probing. | |
559 | * Returns 0 if no settings found, 1 if they are. | |
560 | */ | |
561 | int netdev_boot_setup_check(struct net_device *dev) | |
562 | { | |
563 | struct netdev_boot_setup *s = dev_boot_setup; | |
564 | int i; | |
565 | ||
566 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
567 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 568 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
569 | dev->irq = s[i].map.irq; |
570 | dev->base_addr = s[i].map.base_addr; | |
571 | dev->mem_start = s[i].map.mem_start; | |
572 | dev->mem_end = s[i].map.mem_end; | |
573 | return 1; | |
574 | } | |
575 | } | |
576 | return 0; | |
577 | } | |
d1b19dff | 578 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
579 | |
580 | ||
581 | /** | |
582 | * netdev_boot_base - get address from boot time settings | |
583 | * @prefix: prefix for network device | |
584 | * @unit: id for network device | |
585 | * | |
586 | * Check boot time settings for the base address of device. | |
587 | * The found settings are set for the device to be used | |
588 | * later in the device probing. | |
589 | * Returns 0 if no settings found. | |
590 | */ | |
591 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
592 | { | |
593 | const struct netdev_boot_setup *s = dev_boot_setup; | |
594 | char name[IFNAMSIZ]; | |
595 | int i; | |
596 | ||
597 | sprintf(name, "%s%d", prefix, unit); | |
598 | ||
599 | /* | |
600 | * If device already registered then return base of 1 | |
601 | * to indicate not to probe for this interface | |
602 | */ | |
881d966b | 603 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
604 | return 1; |
605 | ||
606 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
607 | if (!strcmp(name, s[i].name)) | |
608 | return s[i].map.base_addr; | |
609 | return 0; | |
610 | } | |
611 | ||
612 | /* | |
613 | * Saves at boot time configured settings for any netdevice. | |
614 | */ | |
615 | int __init netdev_boot_setup(char *str) | |
616 | { | |
617 | int ints[5]; | |
618 | struct ifmap map; | |
619 | ||
620 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
621 | if (!str || !*str) | |
622 | return 0; | |
623 | ||
624 | /* Save settings */ | |
625 | memset(&map, 0, sizeof(map)); | |
626 | if (ints[0] > 0) | |
627 | map.irq = ints[1]; | |
628 | if (ints[0] > 1) | |
629 | map.base_addr = ints[2]; | |
630 | if (ints[0] > 2) | |
631 | map.mem_start = ints[3]; | |
632 | if (ints[0] > 3) | |
633 | map.mem_end = ints[4]; | |
634 | ||
635 | /* Add new entry to the list */ | |
636 | return netdev_boot_setup_add(str, &map); | |
637 | } | |
638 | ||
639 | __setup("netdev=", netdev_boot_setup); | |
640 | ||
641 | /******************************************************************************* | |
642 | ||
643 | Device Interface Subroutines | |
644 | ||
645 | *******************************************************************************/ | |
646 | ||
647 | /** | |
648 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 649 | * @net: the applicable net namespace |
1da177e4 LT |
650 | * @name: name to find |
651 | * | |
652 | * Find an interface by name. Must be called under RTNL semaphore | |
653 | * or @dev_base_lock. If the name is found a pointer to the device | |
654 | * is returned. If the name is not found then %NULL is returned. The | |
655 | * reference counters are not incremented so the caller must be | |
656 | * careful with locks. | |
657 | */ | |
658 | ||
881d966b | 659 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 660 | { |
0bd8d536 ED |
661 | struct net_device *dev; |
662 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 663 | |
b67bfe0d | 664 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
665 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
666 | return dev; | |
0bd8d536 | 667 | |
1da177e4 LT |
668 | return NULL; |
669 | } | |
d1b19dff | 670 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 671 | |
72c9528b ED |
672 | /** |
673 | * dev_get_by_name_rcu - find a device by its name | |
674 | * @net: the applicable net namespace | |
675 | * @name: name to find | |
676 | * | |
677 | * Find an interface by name. | |
678 | * If the name is found a pointer to the device is returned. | |
679 | * If the name is not found then %NULL is returned. | |
680 | * The reference counters are not incremented so the caller must be | |
681 | * careful with locks. The caller must hold RCU lock. | |
682 | */ | |
683 | ||
684 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
685 | { | |
72c9528b ED |
686 | struct net_device *dev; |
687 | struct hlist_head *head = dev_name_hash(net, name); | |
688 | ||
b67bfe0d | 689 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
690 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
691 | return dev; | |
692 | ||
693 | return NULL; | |
694 | } | |
695 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
696 | ||
1da177e4 LT |
697 | /** |
698 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 699 | * @net: the applicable net namespace |
1da177e4 LT |
700 | * @name: name to find |
701 | * | |
702 | * Find an interface by name. This can be called from any | |
703 | * context and does its own locking. The returned handle has | |
704 | * the usage count incremented and the caller must use dev_put() to | |
705 | * release it when it is no longer needed. %NULL is returned if no | |
706 | * matching device is found. | |
707 | */ | |
708 | ||
881d966b | 709 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
710 | { |
711 | struct net_device *dev; | |
712 | ||
72c9528b ED |
713 | rcu_read_lock(); |
714 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
715 | if (dev) |
716 | dev_hold(dev); | |
72c9528b | 717 | rcu_read_unlock(); |
1da177e4 LT |
718 | return dev; |
719 | } | |
d1b19dff | 720 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
721 | |
722 | /** | |
723 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 724 | * @net: the applicable net namespace |
1da177e4 LT |
725 | * @ifindex: index of device |
726 | * | |
727 | * Search for an interface by index. Returns %NULL if the device | |
728 | * is not found or a pointer to the device. The device has not | |
729 | * had its reference counter increased so the caller must be careful | |
730 | * about locking. The caller must hold either the RTNL semaphore | |
731 | * or @dev_base_lock. | |
732 | */ | |
733 | ||
881d966b | 734 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 735 | { |
0bd8d536 ED |
736 | struct net_device *dev; |
737 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 738 | |
b67bfe0d | 739 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
740 | if (dev->ifindex == ifindex) |
741 | return dev; | |
0bd8d536 | 742 | |
1da177e4 LT |
743 | return NULL; |
744 | } | |
d1b19dff | 745 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 746 | |
fb699dfd ED |
747 | /** |
748 | * dev_get_by_index_rcu - find a device by its ifindex | |
749 | * @net: the applicable net namespace | |
750 | * @ifindex: index of device | |
751 | * | |
752 | * Search for an interface by index. Returns %NULL if the device | |
753 | * is not found or a pointer to the device. The device has not | |
754 | * had its reference counter increased so the caller must be careful | |
755 | * about locking. The caller must hold RCU lock. | |
756 | */ | |
757 | ||
758 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
759 | { | |
fb699dfd ED |
760 | struct net_device *dev; |
761 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
762 | ||
b67bfe0d | 763 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
764 | if (dev->ifindex == ifindex) |
765 | return dev; | |
766 | ||
767 | return NULL; | |
768 | } | |
769 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
770 | ||
1da177e4 LT |
771 | |
772 | /** | |
773 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 774 | * @net: the applicable net namespace |
1da177e4 LT |
775 | * @ifindex: index of device |
776 | * | |
777 | * Search for an interface by index. Returns NULL if the device | |
778 | * is not found or a pointer to the device. The device returned has | |
779 | * had a reference added and the pointer is safe until the user calls | |
780 | * dev_put to indicate they have finished with it. | |
781 | */ | |
782 | ||
881d966b | 783 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
784 | { |
785 | struct net_device *dev; | |
786 | ||
fb699dfd ED |
787 | rcu_read_lock(); |
788 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
789 | if (dev) |
790 | dev_hold(dev); | |
fb699dfd | 791 | rcu_read_unlock(); |
1da177e4 LT |
792 | return dev; |
793 | } | |
d1b19dff | 794 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 LT |
795 | |
796 | /** | |
941666c2 | 797 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 798 | * @net: the applicable net namespace |
1da177e4 LT |
799 | * @type: media type of device |
800 | * @ha: hardware address | |
801 | * | |
802 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
803 | * is not found or a pointer to the device. |
804 | * The caller must hold RCU or RTNL. | |
941666c2 | 805 | * The returned device has not had its ref count increased |
1da177e4 LT |
806 | * and the caller must therefore be careful about locking |
807 | * | |
1da177e4 LT |
808 | */ |
809 | ||
941666c2 ED |
810 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
811 | const char *ha) | |
1da177e4 LT |
812 | { |
813 | struct net_device *dev; | |
814 | ||
941666c2 | 815 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
816 | if (dev->type == type && |
817 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
818 | return dev; |
819 | ||
820 | return NULL; | |
1da177e4 | 821 | } |
941666c2 | 822 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 823 | |
881d966b | 824 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
825 | { |
826 | struct net_device *dev; | |
827 | ||
4e9cac2b | 828 | ASSERT_RTNL(); |
881d966b | 829 | for_each_netdev(net, dev) |
4e9cac2b | 830 | if (dev->type == type) |
7562f876 PE |
831 | return dev; |
832 | ||
833 | return NULL; | |
4e9cac2b | 834 | } |
4e9cac2b PM |
835 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
836 | ||
881d966b | 837 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 838 | { |
99fe3c39 | 839 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 840 | |
99fe3c39 ED |
841 | rcu_read_lock(); |
842 | for_each_netdev_rcu(net, dev) | |
843 | if (dev->type == type) { | |
844 | dev_hold(dev); | |
845 | ret = dev; | |
846 | break; | |
847 | } | |
848 | rcu_read_unlock(); | |
849 | return ret; | |
1da177e4 | 850 | } |
1da177e4 LT |
851 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
852 | ||
853 | /** | |
bb69ae04 | 854 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 855 | * @net: the applicable net namespace |
1da177e4 LT |
856 | * @if_flags: IFF_* values |
857 | * @mask: bitmask of bits in if_flags to check | |
858 | * | |
859 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
860 | * is not found or a pointer to the device. Must be called inside |
861 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
862 | */ |
863 | ||
bb69ae04 | 864 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 865 | unsigned short mask) |
1da177e4 | 866 | { |
7562f876 | 867 | struct net_device *dev, *ret; |
1da177e4 | 868 | |
7562f876 | 869 | ret = NULL; |
c6d14c84 | 870 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 871 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 872 | ret = dev; |
1da177e4 LT |
873 | break; |
874 | } | |
875 | } | |
7562f876 | 876 | return ret; |
1da177e4 | 877 | } |
bb69ae04 | 878 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
879 | |
880 | /** | |
881 | * dev_valid_name - check if name is okay for network device | |
882 | * @name: name string | |
883 | * | |
884 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
885 | * to allow sysfs to work. We also disallow any kind of |
886 | * whitespace. | |
1da177e4 | 887 | */ |
95f050bf | 888 | bool dev_valid_name(const char *name) |
1da177e4 | 889 | { |
c7fa9d18 | 890 | if (*name == '\0') |
95f050bf | 891 | return false; |
b6fe17d6 | 892 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 893 | return false; |
c7fa9d18 | 894 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 895 | return false; |
c7fa9d18 DM |
896 | |
897 | while (*name) { | |
898 | if (*name == '/' || isspace(*name)) | |
95f050bf | 899 | return false; |
c7fa9d18 DM |
900 | name++; |
901 | } | |
95f050bf | 902 | return true; |
1da177e4 | 903 | } |
d1b19dff | 904 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
905 | |
906 | /** | |
b267b179 EB |
907 | * __dev_alloc_name - allocate a name for a device |
908 | * @net: network namespace to allocate the device name in | |
1da177e4 | 909 | * @name: name format string |
b267b179 | 910 | * @buf: scratch buffer and result name string |
1da177e4 LT |
911 | * |
912 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
913 | * id. It scans list of devices to build up a free map, then chooses |
914 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
915 | * while allocating the name and adding the device in order to avoid | |
916 | * duplicates. | |
917 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
918 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
919 | */ |
920 | ||
b267b179 | 921 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
922 | { |
923 | int i = 0; | |
1da177e4 LT |
924 | const char *p; |
925 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 926 | unsigned long *inuse; |
1da177e4 LT |
927 | struct net_device *d; |
928 | ||
929 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
930 | if (p) { | |
931 | /* | |
932 | * Verify the string as this thing may have come from | |
933 | * the user. There must be either one "%d" and no other "%" | |
934 | * characters. | |
935 | */ | |
936 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
937 | return -EINVAL; | |
938 | ||
939 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 940 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
941 | if (!inuse) |
942 | return -ENOMEM; | |
943 | ||
881d966b | 944 | for_each_netdev(net, d) { |
1da177e4 LT |
945 | if (!sscanf(d->name, name, &i)) |
946 | continue; | |
947 | if (i < 0 || i >= max_netdevices) | |
948 | continue; | |
949 | ||
950 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 951 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
952 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
953 | set_bit(i, inuse); | |
954 | } | |
955 | ||
956 | i = find_first_zero_bit(inuse, max_netdevices); | |
957 | free_page((unsigned long) inuse); | |
958 | } | |
959 | ||
d9031024 OP |
960 | if (buf != name) |
961 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 962 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 963 | return i; |
1da177e4 LT |
964 | |
965 | /* It is possible to run out of possible slots | |
966 | * when the name is long and there isn't enough space left | |
967 | * for the digits, or if all bits are used. | |
968 | */ | |
969 | return -ENFILE; | |
970 | } | |
971 | ||
b267b179 EB |
972 | /** |
973 | * dev_alloc_name - allocate a name for a device | |
974 | * @dev: device | |
975 | * @name: name format string | |
976 | * | |
977 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
978 | * id. It scans list of devices to build up a free map, then chooses | |
979 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
980 | * while allocating the name and adding the device in order to avoid | |
981 | * duplicates. | |
982 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
983 | * Returns the number of the unit assigned or a negative errno code. | |
984 | */ | |
985 | ||
986 | int dev_alloc_name(struct net_device *dev, const char *name) | |
987 | { | |
988 | char buf[IFNAMSIZ]; | |
989 | struct net *net; | |
990 | int ret; | |
991 | ||
c346dca1 YH |
992 | BUG_ON(!dev_net(dev)); |
993 | net = dev_net(dev); | |
b267b179 EB |
994 | ret = __dev_alloc_name(net, name, buf); |
995 | if (ret >= 0) | |
996 | strlcpy(dev->name, buf, IFNAMSIZ); | |
997 | return ret; | |
998 | } | |
d1b19dff | 999 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1000 | |
828de4f6 G |
1001 | static int dev_alloc_name_ns(struct net *net, |
1002 | struct net_device *dev, | |
1003 | const char *name) | |
d9031024 | 1004 | { |
828de4f6 G |
1005 | char buf[IFNAMSIZ]; |
1006 | int ret; | |
8ce6cebc | 1007 | |
828de4f6 G |
1008 | ret = __dev_alloc_name(net, name, buf); |
1009 | if (ret >= 0) | |
1010 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1011 | return ret; | |
1012 | } | |
1013 | ||
1014 | static int dev_get_valid_name(struct net *net, | |
1015 | struct net_device *dev, | |
1016 | const char *name) | |
1017 | { | |
1018 | BUG_ON(!net); | |
8ce6cebc | 1019 | |
d9031024 OP |
1020 | if (!dev_valid_name(name)) |
1021 | return -EINVAL; | |
1022 | ||
1c5cae81 | 1023 | if (strchr(name, '%')) |
828de4f6 | 1024 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1025 | else if (__dev_get_by_name(net, name)) |
1026 | return -EEXIST; | |
8ce6cebc DL |
1027 | else if (dev->name != name) |
1028 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1029 | |
1030 | return 0; | |
1031 | } | |
1da177e4 LT |
1032 | |
1033 | /** | |
1034 | * dev_change_name - change name of a device | |
1035 | * @dev: device | |
1036 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1037 | * | |
1038 | * Change name of a device, can pass format strings "eth%d". | |
1039 | * for wildcarding. | |
1040 | */ | |
cf04a4c7 | 1041 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1042 | { |
fcc5a03a | 1043 | char oldname[IFNAMSIZ]; |
1da177e4 | 1044 | int err = 0; |
fcc5a03a | 1045 | int ret; |
881d966b | 1046 | struct net *net; |
1da177e4 LT |
1047 | |
1048 | ASSERT_RTNL(); | |
c346dca1 | 1049 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1050 | |
c346dca1 | 1051 | net = dev_net(dev); |
1da177e4 LT |
1052 | if (dev->flags & IFF_UP) |
1053 | return -EBUSY; | |
1054 | ||
30e6c9fa | 1055 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1056 | |
1057 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1058 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1059 | return 0; |
c91f6df2 | 1060 | } |
c8d90dca | 1061 | |
fcc5a03a HX |
1062 | memcpy(oldname, dev->name, IFNAMSIZ); |
1063 | ||
828de4f6 | 1064 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1065 | if (err < 0) { |
30e6c9fa | 1066 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1067 | return err; |
c91f6df2 | 1068 | } |
1da177e4 | 1069 | |
fcc5a03a | 1070 | rollback: |
a1b3f594 EB |
1071 | ret = device_rename(&dev->dev, dev->name); |
1072 | if (ret) { | |
1073 | memcpy(dev->name, oldname, IFNAMSIZ); | |
30e6c9fa | 1074 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1075 | return ret; |
dcc99773 | 1076 | } |
7f988eab | 1077 | |
30e6c9fa | 1078 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1079 | |
7f988eab | 1080 | write_lock_bh(&dev_base_lock); |
372b2312 | 1081 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1082 | write_unlock_bh(&dev_base_lock); |
1083 | ||
1084 | synchronize_rcu(); | |
1085 | ||
1086 | write_lock_bh(&dev_base_lock); | |
1087 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1088 | write_unlock_bh(&dev_base_lock); |
1089 | ||
056925ab | 1090 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1091 | ret = notifier_to_errno(ret); |
1092 | ||
1093 | if (ret) { | |
91e9c07b ED |
1094 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1095 | if (err >= 0) { | |
fcc5a03a | 1096 | err = ret; |
30e6c9fa | 1097 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a HX |
1098 | memcpy(dev->name, oldname, IFNAMSIZ); |
1099 | goto rollback; | |
91e9c07b | 1100 | } else { |
7b6cd1ce | 1101 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1102 | dev->name, ret); |
fcc5a03a HX |
1103 | } |
1104 | } | |
1da177e4 LT |
1105 | |
1106 | return err; | |
1107 | } | |
1108 | ||
0b815a1a SH |
1109 | /** |
1110 | * dev_set_alias - change ifalias of a device | |
1111 | * @dev: device | |
1112 | * @alias: name up to IFALIASZ | |
f0db275a | 1113 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1114 | * |
1115 | * Set ifalias for a device, | |
1116 | */ | |
1117 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1118 | { | |
7364e445 AK |
1119 | char *new_ifalias; |
1120 | ||
0b815a1a SH |
1121 | ASSERT_RTNL(); |
1122 | ||
1123 | if (len >= IFALIASZ) | |
1124 | return -EINVAL; | |
1125 | ||
96ca4a2c | 1126 | if (!len) { |
388dfc2d SK |
1127 | kfree(dev->ifalias); |
1128 | dev->ifalias = NULL; | |
96ca4a2c OH |
1129 | return 0; |
1130 | } | |
1131 | ||
7364e445 AK |
1132 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1133 | if (!new_ifalias) | |
0b815a1a | 1134 | return -ENOMEM; |
7364e445 | 1135 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1136 | |
1137 | strlcpy(dev->ifalias, alias, len+1); | |
1138 | return len; | |
1139 | } | |
1140 | ||
1141 | ||
d8a33ac4 | 1142 | /** |
3041a069 | 1143 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1144 | * @dev: device to cause notification |
1145 | * | |
1146 | * Called to indicate a device has changed features. | |
1147 | */ | |
1148 | void netdev_features_change(struct net_device *dev) | |
1149 | { | |
056925ab | 1150 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1151 | } |
1152 | EXPORT_SYMBOL(netdev_features_change); | |
1153 | ||
1da177e4 LT |
1154 | /** |
1155 | * netdev_state_change - device changes state | |
1156 | * @dev: device to cause notification | |
1157 | * | |
1158 | * Called to indicate a device has changed state. This function calls | |
1159 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1160 | * to the routing socket. | |
1161 | */ | |
1162 | void netdev_state_change(struct net_device *dev) | |
1163 | { | |
1164 | if (dev->flags & IFF_UP) { | |
056925ab | 1165 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1166 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1167 | } | |
1168 | } | |
d1b19dff | 1169 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1170 | |
ee89bab1 AW |
1171 | /** |
1172 | * netdev_notify_peers - notify network peers about existence of @dev | |
1173 | * @dev: network device | |
1174 | * | |
1175 | * Generate traffic such that interested network peers are aware of | |
1176 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1177 | * a device wants to inform the rest of the network about some sort of | |
1178 | * reconfiguration such as a failover event or virtual machine | |
1179 | * migration. | |
1180 | */ | |
1181 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1182 | { |
ee89bab1 AW |
1183 | rtnl_lock(); |
1184 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1185 | rtnl_unlock(); | |
c1da4ac7 | 1186 | } |
ee89bab1 | 1187 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1188 | |
bd380811 | 1189 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1190 | { |
d314774c | 1191 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1192 | int ret; |
1da177e4 | 1193 | |
e46b66bc BH |
1194 | ASSERT_RTNL(); |
1195 | ||
1da177e4 LT |
1196 | if (!netif_device_present(dev)) |
1197 | return -ENODEV; | |
1198 | ||
ca99ca14 NH |
1199 | /* Block netpoll from trying to do any rx path servicing. |
1200 | * If we don't do this there is a chance ndo_poll_controller | |
1201 | * or ndo_poll may be running while we open the device | |
1202 | */ | |
1203 | ret = netpoll_rx_disable(dev); | |
1204 | if (ret) | |
1205 | return ret; | |
1206 | ||
3b8bcfd5 JB |
1207 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1208 | ret = notifier_to_errno(ret); | |
1209 | if (ret) | |
1210 | return ret; | |
1211 | ||
1da177e4 | 1212 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1213 | |
d314774c SH |
1214 | if (ops->ndo_validate_addr) |
1215 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1216 | |
d314774c SH |
1217 | if (!ret && ops->ndo_open) |
1218 | ret = ops->ndo_open(dev); | |
1da177e4 | 1219 | |
ca99ca14 NH |
1220 | netpoll_rx_enable(dev); |
1221 | ||
bada339b JG |
1222 | if (ret) |
1223 | clear_bit(__LINK_STATE_START, &dev->state); | |
1224 | else { | |
1da177e4 | 1225 | dev->flags |= IFF_UP; |
b4bd07c2 | 1226 | net_dmaengine_get(); |
4417da66 | 1227 | dev_set_rx_mode(dev); |
1da177e4 | 1228 | dev_activate(dev); |
7bf23575 | 1229 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1230 | } |
bada339b | 1231 | |
1da177e4 LT |
1232 | return ret; |
1233 | } | |
1234 | ||
1235 | /** | |
bd380811 PM |
1236 | * dev_open - prepare an interface for use. |
1237 | * @dev: device to open | |
1da177e4 | 1238 | * |
bd380811 PM |
1239 | * Takes a device from down to up state. The device's private open |
1240 | * function is invoked and then the multicast lists are loaded. Finally | |
1241 | * the device is moved into the up state and a %NETDEV_UP message is | |
1242 | * sent to the netdev notifier chain. | |
1243 | * | |
1244 | * Calling this function on an active interface is a nop. On a failure | |
1245 | * a negative errno code is returned. | |
1da177e4 | 1246 | */ |
bd380811 PM |
1247 | int dev_open(struct net_device *dev) |
1248 | { | |
1249 | int ret; | |
1250 | ||
bd380811 PM |
1251 | if (dev->flags & IFF_UP) |
1252 | return 0; | |
1253 | ||
bd380811 PM |
1254 | ret = __dev_open(dev); |
1255 | if (ret < 0) | |
1256 | return ret; | |
1257 | ||
bd380811 PM |
1258 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); |
1259 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1260 | ||
1261 | return ret; | |
1262 | } | |
1263 | EXPORT_SYMBOL(dev_open); | |
1264 | ||
44345724 | 1265 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1266 | { |
44345724 | 1267 | struct net_device *dev; |
e46b66bc | 1268 | |
bd380811 | 1269 | ASSERT_RTNL(); |
9d5010db DM |
1270 | might_sleep(); |
1271 | ||
44345724 | 1272 | list_for_each_entry(dev, head, unreg_list) { |
44345724 | 1273 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1274 | |
44345724 | 1275 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1276 | |
44345724 OP |
1277 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1278 | * can be even on different cpu. So just clear netif_running(). | |
1279 | * | |
1280 | * dev->stop() will invoke napi_disable() on all of it's | |
1281 | * napi_struct instances on this device. | |
1282 | */ | |
1283 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1284 | } | |
1da177e4 | 1285 | |
44345724 | 1286 | dev_deactivate_many(head); |
d8b2a4d2 | 1287 | |
44345724 OP |
1288 | list_for_each_entry(dev, head, unreg_list) { |
1289 | const struct net_device_ops *ops = dev->netdev_ops; | |
1da177e4 | 1290 | |
44345724 OP |
1291 | /* |
1292 | * Call the device specific close. This cannot fail. | |
1293 | * Only if device is UP | |
1294 | * | |
1295 | * We allow it to be called even after a DETACH hot-plug | |
1296 | * event. | |
1297 | */ | |
1298 | if (ops->ndo_stop) | |
1299 | ops->ndo_stop(dev); | |
1300 | ||
44345724 | 1301 | dev->flags &= ~IFF_UP; |
44345724 OP |
1302 | net_dmaengine_put(); |
1303 | } | |
1304 | ||
1305 | return 0; | |
1306 | } | |
1307 | ||
1308 | static int __dev_close(struct net_device *dev) | |
1309 | { | |
f87e6f47 | 1310 | int retval; |
44345724 OP |
1311 | LIST_HEAD(single); |
1312 | ||
ca99ca14 NH |
1313 | /* Temporarily disable netpoll until the interface is down */ |
1314 | retval = netpoll_rx_disable(dev); | |
1315 | if (retval) | |
1316 | return retval; | |
1317 | ||
44345724 | 1318 | list_add(&dev->unreg_list, &single); |
f87e6f47 LT |
1319 | retval = __dev_close_many(&single); |
1320 | list_del(&single); | |
ca99ca14 NH |
1321 | |
1322 | netpoll_rx_enable(dev); | |
f87e6f47 | 1323 | return retval; |
44345724 OP |
1324 | } |
1325 | ||
3fbd8758 | 1326 | static int dev_close_many(struct list_head *head) |
44345724 OP |
1327 | { |
1328 | struct net_device *dev, *tmp; | |
1329 | LIST_HEAD(tmp_list); | |
1da177e4 | 1330 | |
44345724 OP |
1331 | list_for_each_entry_safe(dev, tmp, head, unreg_list) |
1332 | if (!(dev->flags & IFF_UP)) | |
1333 | list_move(&dev->unreg_list, &tmp_list); | |
1334 | ||
1335 | __dev_close_many(head); | |
1da177e4 | 1336 | |
44345724 OP |
1337 | list_for_each_entry(dev, head, unreg_list) { |
1338 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1339 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1340 | } | |
bd380811 | 1341 | |
44345724 OP |
1342 | /* rollback_registered_many needs the complete original list */ |
1343 | list_splice(&tmp_list, head); | |
bd380811 PM |
1344 | return 0; |
1345 | } | |
1346 | ||
1347 | /** | |
1348 | * dev_close - shutdown an interface. | |
1349 | * @dev: device to shutdown | |
1350 | * | |
1351 | * This function moves an active device into down state. A | |
1352 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1353 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1354 | * chain. | |
1355 | */ | |
1356 | int dev_close(struct net_device *dev) | |
1357 | { | |
ca99ca14 | 1358 | int ret = 0; |
e14a5993 ED |
1359 | if (dev->flags & IFF_UP) { |
1360 | LIST_HEAD(single); | |
1da177e4 | 1361 | |
ca99ca14 NH |
1362 | /* Block netpoll rx while the interface is going down */ |
1363 | ret = netpoll_rx_disable(dev); | |
1364 | if (ret) | |
1365 | return ret; | |
1366 | ||
e14a5993 ED |
1367 | list_add(&dev->unreg_list, &single); |
1368 | dev_close_many(&single); | |
1369 | list_del(&single); | |
ca99ca14 NH |
1370 | |
1371 | netpoll_rx_enable(dev); | |
e14a5993 | 1372 | } |
ca99ca14 | 1373 | return ret; |
1da177e4 | 1374 | } |
d1b19dff | 1375 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1376 | |
1377 | ||
0187bdfb BH |
1378 | /** |
1379 | * dev_disable_lro - disable Large Receive Offload on a device | |
1380 | * @dev: device | |
1381 | * | |
1382 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1383 | * called under RTNL. This is needed if received packets may be | |
1384 | * forwarded to another interface. | |
1385 | */ | |
1386 | void dev_disable_lro(struct net_device *dev) | |
1387 | { | |
f11970e3 NH |
1388 | /* |
1389 | * If we're trying to disable lro on a vlan device | |
1390 | * use the underlying physical device instead | |
1391 | */ | |
1392 | if (is_vlan_dev(dev)) | |
1393 | dev = vlan_dev_real_dev(dev); | |
1394 | ||
bc5787c6 MM |
1395 | dev->wanted_features &= ~NETIF_F_LRO; |
1396 | netdev_update_features(dev); | |
27660515 | 1397 | |
22d5969f MM |
1398 | if (unlikely(dev->features & NETIF_F_LRO)) |
1399 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
0187bdfb BH |
1400 | } |
1401 | EXPORT_SYMBOL(dev_disable_lro); | |
1402 | ||
1403 | ||
881d966b EB |
1404 | static int dev_boot_phase = 1; |
1405 | ||
1da177e4 LT |
1406 | /** |
1407 | * register_netdevice_notifier - register a network notifier block | |
1408 | * @nb: notifier | |
1409 | * | |
1410 | * Register a notifier to be called when network device events occur. | |
1411 | * The notifier passed is linked into the kernel structures and must | |
1412 | * not be reused until it has been unregistered. A negative errno code | |
1413 | * is returned on a failure. | |
1414 | * | |
1415 | * When registered all registration and up events are replayed | |
4ec93edb | 1416 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1417 | * view of the network device list. |
1418 | */ | |
1419 | ||
1420 | int register_netdevice_notifier(struct notifier_block *nb) | |
1421 | { | |
1422 | struct net_device *dev; | |
fcc5a03a | 1423 | struct net_device *last; |
881d966b | 1424 | struct net *net; |
1da177e4 LT |
1425 | int err; |
1426 | ||
1427 | rtnl_lock(); | |
f07d5b94 | 1428 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1429 | if (err) |
1430 | goto unlock; | |
881d966b EB |
1431 | if (dev_boot_phase) |
1432 | goto unlock; | |
1433 | for_each_net(net) { | |
1434 | for_each_netdev(net, dev) { | |
1435 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1436 | err = notifier_to_errno(err); | |
1437 | if (err) | |
1438 | goto rollback; | |
1439 | ||
1440 | if (!(dev->flags & IFF_UP)) | |
1441 | continue; | |
1da177e4 | 1442 | |
881d966b EB |
1443 | nb->notifier_call(nb, NETDEV_UP, dev); |
1444 | } | |
1da177e4 | 1445 | } |
fcc5a03a HX |
1446 | |
1447 | unlock: | |
1da177e4 LT |
1448 | rtnl_unlock(); |
1449 | return err; | |
fcc5a03a HX |
1450 | |
1451 | rollback: | |
1452 | last = dev; | |
881d966b EB |
1453 | for_each_net(net) { |
1454 | for_each_netdev(net, dev) { | |
1455 | if (dev == last) | |
8f891489 | 1456 | goto outroll; |
fcc5a03a | 1457 | |
881d966b EB |
1458 | if (dev->flags & IFF_UP) { |
1459 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1460 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1461 | } | |
1462 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
fcc5a03a | 1463 | } |
fcc5a03a | 1464 | } |
c67625a1 | 1465 | |
8f891489 | 1466 | outroll: |
c67625a1 | 1467 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1468 | goto unlock; |
1da177e4 | 1469 | } |
d1b19dff | 1470 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1471 | |
1472 | /** | |
1473 | * unregister_netdevice_notifier - unregister a network notifier block | |
1474 | * @nb: notifier | |
1475 | * | |
1476 | * Unregister a notifier previously registered by | |
1477 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1478 | * kernel structures and may then be reused. A negative errno code | |
1479 | * is returned on a failure. | |
7d3d43da EB |
1480 | * |
1481 | * After unregistering unregister and down device events are synthesized | |
1482 | * for all devices on the device list to the removed notifier to remove | |
1483 | * the need for special case cleanup code. | |
1da177e4 LT |
1484 | */ |
1485 | ||
1486 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1487 | { | |
7d3d43da EB |
1488 | struct net_device *dev; |
1489 | struct net *net; | |
9f514950 HX |
1490 | int err; |
1491 | ||
1492 | rtnl_lock(); | |
f07d5b94 | 1493 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1494 | if (err) |
1495 | goto unlock; | |
1496 | ||
1497 | for_each_net(net) { | |
1498 | for_each_netdev(net, dev) { | |
1499 | if (dev->flags & IFF_UP) { | |
1500 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1501 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1502 | } | |
1503 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
7d3d43da EB |
1504 | } |
1505 | } | |
1506 | unlock: | |
9f514950 HX |
1507 | rtnl_unlock(); |
1508 | return err; | |
1da177e4 | 1509 | } |
d1b19dff | 1510 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 LT |
1511 | |
1512 | /** | |
1513 | * call_netdevice_notifiers - call all network notifier blocks | |
1514 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1515 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1516 | * |
1517 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1518 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1519 | */ |
1520 | ||
ad7379d4 | 1521 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1522 | { |
ab930471 | 1523 | ASSERT_RTNL(); |
ad7379d4 | 1524 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 | 1525 | } |
edf947f1 | 1526 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1527 | |
c5905afb | 1528 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1529 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1530 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1531 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1532 | * static_key_slow_dec() calls. |
b90e5794 ED |
1533 | */ |
1534 | static atomic_t netstamp_needed_deferred; | |
1535 | #endif | |
1da177e4 LT |
1536 | |
1537 | void net_enable_timestamp(void) | |
1538 | { | |
b90e5794 ED |
1539 | #ifdef HAVE_JUMP_LABEL |
1540 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1541 | ||
1542 | if (deferred) { | |
1543 | while (--deferred) | |
c5905afb | 1544 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1545 | return; |
1546 | } | |
1547 | #endif | |
1548 | WARN_ON(in_interrupt()); | |
c5905afb | 1549 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1550 | } |
d1b19dff | 1551 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1552 | |
1553 | void net_disable_timestamp(void) | |
1554 | { | |
b90e5794 ED |
1555 | #ifdef HAVE_JUMP_LABEL |
1556 | if (in_interrupt()) { | |
1557 | atomic_inc(&netstamp_needed_deferred); | |
1558 | return; | |
1559 | } | |
1560 | #endif | |
c5905afb | 1561 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1562 | } |
d1b19dff | 1563 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1564 | |
3b098e2d | 1565 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1566 | { |
588f0330 | 1567 | skb->tstamp.tv64 = 0; |
c5905afb | 1568 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1569 | __net_timestamp(skb); |
1da177e4 LT |
1570 | } |
1571 | ||
588f0330 | 1572 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1573 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1574 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1575 | __net_timestamp(SKB); \ | |
1576 | } \ | |
3b098e2d | 1577 | |
79b569f0 DL |
1578 | static inline bool is_skb_forwardable(struct net_device *dev, |
1579 | struct sk_buff *skb) | |
1580 | { | |
1581 | unsigned int len; | |
1582 | ||
1583 | if (!(dev->flags & IFF_UP)) | |
1584 | return false; | |
1585 | ||
1586 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1587 | if (skb->len <= len) | |
1588 | return true; | |
1589 | ||
1590 | /* if TSO is enabled, we don't care about the length as the packet | |
1591 | * could be forwarded without being segmented before | |
1592 | */ | |
1593 | if (skb_is_gso(skb)) | |
1594 | return true; | |
1595 | ||
1596 | return false; | |
1597 | } | |
1598 | ||
44540960 AB |
1599 | /** |
1600 | * dev_forward_skb - loopback an skb to another netif | |
1601 | * | |
1602 | * @dev: destination network device | |
1603 | * @skb: buffer to forward | |
1604 | * | |
1605 | * return values: | |
1606 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1607 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1608 | * |
1609 | * dev_forward_skb can be used for injecting an skb from the | |
1610 | * start_xmit function of one device into the receive queue | |
1611 | * of another device. | |
1612 | * | |
1613 | * The receiving device may be in another namespace, so | |
1614 | * we have to clear all information in the skb that could | |
1615 | * impact namespace isolation. | |
1616 | */ | |
1617 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1618 | { | |
48c83012 MT |
1619 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { |
1620 | if (skb_copy_ubufs(skb, GFP_ATOMIC)) { | |
1621 | atomic_long_inc(&dev->rx_dropped); | |
1622 | kfree_skb(skb); | |
1623 | return NET_RX_DROP; | |
1624 | } | |
1625 | } | |
1626 | ||
44540960 | 1627 | skb_orphan(skb); |
c736eefa | 1628 | nf_reset(skb); |
44540960 | 1629 | |
79b569f0 | 1630 | if (unlikely(!is_skb_forwardable(dev, skb))) { |
caf586e5 | 1631 | atomic_long_inc(&dev->rx_dropped); |
6ec82562 | 1632 | kfree_skb(skb); |
44540960 | 1633 | return NET_RX_DROP; |
6ec82562 | 1634 | } |
3b9785c6 | 1635 | skb->skb_iif = 0; |
59b9997b DM |
1636 | skb->dev = dev; |
1637 | skb_dst_drop(skb); | |
44540960 AB |
1638 | skb->tstamp.tv64 = 0; |
1639 | skb->pkt_type = PACKET_HOST; | |
1640 | skb->protocol = eth_type_trans(skb, dev); | |
59b9997b DM |
1641 | skb->mark = 0; |
1642 | secpath_reset(skb); | |
1643 | nf_reset(skb); | |
44540960 AB |
1644 | return netif_rx(skb); |
1645 | } | |
1646 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1647 | ||
71d9dec2 CG |
1648 | static inline int deliver_skb(struct sk_buff *skb, |
1649 | struct packet_type *pt_prev, | |
1650 | struct net_device *orig_dev) | |
1651 | { | |
1080e512 MT |
1652 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1653 | return -ENOMEM; | |
71d9dec2 CG |
1654 | atomic_inc(&skb->users); |
1655 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1656 | } | |
1657 | ||
c0de08d0 EL |
1658 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1659 | { | |
a3d744e9 | 1660 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1661 | return false; |
1662 | ||
1663 | if (ptype->id_match) | |
1664 | return ptype->id_match(ptype, skb->sk); | |
1665 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1666 | return true; | |
1667 | ||
1668 | return false; | |
1669 | } | |
1670 | ||
1da177e4 LT |
1671 | /* |
1672 | * Support routine. Sends outgoing frames to any network | |
1673 | * taps currently in use. | |
1674 | */ | |
1675 | ||
f6a78bfc | 1676 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1677 | { |
1678 | struct packet_type *ptype; | |
71d9dec2 CG |
1679 | struct sk_buff *skb2 = NULL; |
1680 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1681 | |
1da177e4 LT |
1682 | rcu_read_lock(); |
1683 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1684 | /* Never send packets back to the socket | |
1685 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1686 | */ | |
1687 | if ((ptype->dev == dev || !ptype->dev) && | |
c0de08d0 | 1688 | (!skb_loop_sk(ptype, skb))) { |
71d9dec2 CG |
1689 | if (pt_prev) { |
1690 | deliver_skb(skb2, pt_prev, skb->dev); | |
1691 | pt_prev = ptype; | |
1692 | continue; | |
1693 | } | |
1694 | ||
1695 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1696 | if (!skb2) |
1697 | break; | |
1698 | ||
70978182 ED |
1699 | net_timestamp_set(skb2); |
1700 | ||
1da177e4 LT |
1701 | /* skb->nh should be correctly |
1702 | set by sender, so that the second statement is | |
1703 | just protection against buggy protocols. | |
1704 | */ | |
459a98ed | 1705 | skb_reset_mac_header(skb2); |
1da177e4 | 1706 | |
d56f90a7 | 1707 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1708 | skb2->network_header > skb2->tail) { |
e87cc472 JP |
1709 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", |
1710 | ntohs(skb2->protocol), | |
1711 | dev->name); | |
c1d2bbe1 | 1712 | skb_reset_network_header(skb2); |
1da177e4 LT |
1713 | } |
1714 | ||
b0e380b1 | 1715 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1716 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1717 | pt_prev = ptype; |
1da177e4 LT |
1718 | } |
1719 | } | |
71d9dec2 CG |
1720 | if (pt_prev) |
1721 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1722 | rcu_read_unlock(); |
1723 | } | |
1724 | ||
2c53040f BH |
1725 | /** |
1726 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1727 | * @dev: Network device |
1728 | * @txq: number of queues available | |
1729 | * | |
1730 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1731 | * valid. To resolve this verify the tc mapping remains valid and if | |
1732 | * not NULL the mapping. With no priorities mapping to this | |
1733 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1734 | * is invalid nothing can be done so disable priority mappings. If is | |
1735 | * expected that drivers will fix this mapping if they can before | |
1736 | * calling netif_set_real_num_tx_queues. | |
1737 | */ | |
bb134d22 | 1738 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1739 | { |
1740 | int i; | |
1741 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1742 | ||
1743 | /* If TC0 is invalidated disable TC mapping */ | |
1744 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1745 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1746 | dev->num_tc = 0; |
1747 | return; | |
1748 | } | |
1749 | ||
1750 | /* Invalidated prio to tc mappings set to TC0 */ | |
1751 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1752 | int q = netdev_get_prio_tc_map(dev, i); | |
1753 | ||
1754 | tc = &dev->tc_to_txq[q]; | |
1755 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1756 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1757 | i, q); | |
4f57c087 JF |
1758 | netdev_set_prio_tc_map(dev, i, 0); |
1759 | } | |
1760 | } | |
1761 | } | |
1762 | ||
537c00de AD |
1763 | #ifdef CONFIG_XPS |
1764 | static DEFINE_MUTEX(xps_map_mutex); | |
1765 | #define xmap_dereference(P) \ | |
1766 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1767 | ||
10cdc3f3 AD |
1768 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1769 | int cpu, u16 index) | |
537c00de | 1770 | { |
10cdc3f3 AD |
1771 | struct xps_map *map = NULL; |
1772 | int pos; | |
537c00de | 1773 | |
10cdc3f3 AD |
1774 | if (dev_maps) |
1775 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1776 | |
10cdc3f3 AD |
1777 | for (pos = 0; map && pos < map->len; pos++) { |
1778 | if (map->queues[pos] == index) { | |
537c00de AD |
1779 | if (map->len > 1) { |
1780 | map->queues[pos] = map->queues[--map->len]; | |
1781 | } else { | |
10cdc3f3 | 1782 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1783 | kfree_rcu(map, rcu); |
1784 | map = NULL; | |
1785 | } | |
10cdc3f3 | 1786 | break; |
537c00de | 1787 | } |
537c00de AD |
1788 | } |
1789 | ||
10cdc3f3 AD |
1790 | return map; |
1791 | } | |
1792 | ||
024e9679 | 1793 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
10cdc3f3 AD |
1794 | { |
1795 | struct xps_dev_maps *dev_maps; | |
024e9679 | 1796 | int cpu, i; |
10cdc3f3 AD |
1797 | bool active = false; |
1798 | ||
1799 | mutex_lock(&xps_map_mutex); | |
1800 | dev_maps = xmap_dereference(dev->xps_maps); | |
1801 | ||
1802 | if (!dev_maps) | |
1803 | goto out_no_maps; | |
1804 | ||
1805 | for_each_possible_cpu(cpu) { | |
024e9679 AD |
1806 | for (i = index; i < dev->num_tx_queues; i++) { |
1807 | if (!remove_xps_queue(dev_maps, cpu, i)) | |
1808 | break; | |
1809 | } | |
1810 | if (i == dev->num_tx_queues) | |
10cdc3f3 AD |
1811 | active = true; |
1812 | } | |
1813 | ||
1814 | if (!active) { | |
537c00de AD |
1815 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
1816 | kfree_rcu(dev_maps, rcu); | |
1817 | } | |
1818 | ||
024e9679 AD |
1819 | for (i = index; i < dev->num_tx_queues; i++) |
1820 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), | |
1821 | NUMA_NO_NODE); | |
1822 | ||
537c00de AD |
1823 | out_no_maps: |
1824 | mutex_unlock(&xps_map_mutex); | |
1825 | } | |
1826 | ||
01c5f864 AD |
1827 | static struct xps_map *expand_xps_map(struct xps_map *map, |
1828 | int cpu, u16 index) | |
1829 | { | |
1830 | struct xps_map *new_map; | |
1831 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
1832 | int i, pos; | |
1833 | ||
1834 | for (pos = 0; map && pos < map->len; pos++) { | |
1835 | if (map->queues[pos] != index) | |
1836 | continue; | |
1837 | return map; | |
1838 | } | |
1839 | ||
1840 | /* Need to add queue to this CPU's existing map */ | |
1841 | if (map) { | |
1842 | if (pos < map->alloc_len) | |
1843 | return map; | |
1844 | ||
1845 | alloc_len = map->alloc_len * 2; | |
1846 | } | |
1847 | ||
1848 | /* Need to allocate new map to store queue on this CPU's map */ | |
1849 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
1850 | cpu_to_node(cpu)); | |
1851 | if (!new_map) | |
1852 | return NULL; | |
1853 | ||
1854 | for (i = 0; i < pos; i++) | |
1855 | new_map->queues[i] = map->queues[i]; | |
1856 | new_map->alloc_len = alloc_len; | |
1857 | new_map->len = pos; | |
1858 | ||
1859 | return new_map; | |
1860 | } | |
1861 | ||
537c00de AD |
1862 | int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index) |
1863 | { | |
01c5f864 | 1864 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 1865 | struct xps_map *map, *new_map; |
537c00de | 1866 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
1867 | int cpu, numa_node_id = -2; |
1868 | bool active = false; | |
537c00de AD |
1869 | |
1870 | mutex_lock(&xps_map_mutex); | |
1871 | ||
1872 | dev_maps = xmap_dereference(dev->xps_maps); | |
1873 | ||
01c5f864 AD |
1874 | /* allocate memory for queue storage */ |
1875 | for_each_online_cpu(cpu) { | |
1876 | if (!cpumask_test_cpu(cpu, mask)) | |
1877 | continue; | |
1878 | ||
1879 | if (!new_dev_maps) | |
1880 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
1881 | if (!new_dev_maps) { |
1882 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 1883 | return -ENOMEM; |
2bb60cb9 | 1884 | } |
01c5f864 AD |
1885 | |
1886 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
1887 | NULL; | |
1888 | ||
1889 | map = expand_xps_map(map, cpu, index); | |
1890 | if (!map) | |
1891 | goto error; | |
1892 | ||
1893 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
1894 | } | |
1895 | ||
1896 | if (!new_dev_maps) | |
1897 | goto out_no_new_maps; | |
1898 | ||
537c00de | 1899 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
1900 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
1901 | /* add queue to CPU maps */ | |
1902 | int pos = 0; | |
1903 | ||
1904 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1905 | while ((pos < map->len) && (map->queues[pos] != index)) | |
1906 | pos++; | |
1907 | ||
1908 | if (pos == map->len) | |
1909 | map->queues[map->len++] = index; | |
537c00de | 1910 | #ifdef CONFIG_NUMA |
537c00de AD |
1911 | if (numa_node_id == -2) |
1912 | numa_node_id = cpu_to_node(cpu); | |
1913 | else if (numa_node_id != cpu_to_node(cpu)) | |
1914 | numa_node_id = -1; | |
537c00de | 1915 | #endif |
01c5f864 AD |
1916 | } else if (dev_maps) { |
1917 | /* fill in the new device map from the old device map */ | |
1918 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
1919 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 1920 | } |
01c5f864 | 1921 | |
537c00de AD |
1922 | } |
1923 | ||
01c5f864 AD |
1924 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
1925 | ||
537c00de | 1926 | /* Cleanup old maps */ |
01c5f864 AD |
1927 | if (dev_maps) { |
1928 | for_each_possible_cpu(cpu) { | |
1929 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1930 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
1931 | if (map && map != new_map) | |
1932 | kfree_rcu(map, rcu); | |
1933 | } | |
537c00de | 1934 | |
01c5f864 | 1935 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
1936 | } |
1937 | ||
01c5f864 AD |
1938 | dev_maps = new_dev_maps; |
1939 | active = true; | |
537c00de | 1940 | |
01c5f864 AD |
1941 | out_no_new_maps: |
1942 | /* update Tx queue numa node */ | |
537c00de AD |
1943 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
1944 | (numa_node_id >= 0) ? numa_node_id : | |
1945 | NUMA_NO_NODE); | |
1946 | ||
01c5f864 AD |
1947 | if (!dev_maps) |
1948 | goto out_no_maps; | |
1949 | ||
1950 | /* removes queue from unused CPUs */ | |
1951 | for_each_possible_cpu(cpu) { | |
1952 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
1953 | continue; | |
1954 | ||
1955 | if (remove_xps_queue(dev_maps, cpu, index)) | |
1956 | active = true; | |
1957 | } | |
1958 | ||
1959 | /* free map if not active */ | |
1960 | if (!active) { | |
1961 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
1962 | kfree_rcu(dev_maps, rcu); | |
1963 | } | |
1964 | ||
1965 | out_no_maps: | |
537c00de AD |
1966 | mutex_unlock(&xps_map_mutex); |
1967 | ||
1968 | return 0; | |
1969 | error: | |
01c5f864 AD |
1970 | /* remove any maps that we added */ |
1971 | for_each_possible_cpu(cpu) { | |
1972 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1973 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
1974 | NULL; | |
1975 | if (new_map && new_map != map) | |
1976 | kfree(new_map); | |
1977 | } | |
1978 | ||
537c00de AD |
1979 | mutex_unlock(&xps_map_mutex); |
1980 | ||
537c00de AD |
1981 | kfree(new_dev_maps); |
1982 | return -ENOMEM; | |
1983 | } | |
1984 | EXPORT_SYMBOL(netif_set_xps_queue); | |
1985 | ||
1986 | #endif | |
f0796d5c JF |
1987 | /* |
1988 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
1989 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
1990 | */ | |
e6484930 | 1991 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 1992 | { |
1d24eb48 TH |
1993 | int rc; |
1994 | ||
e6484930 TH |
1995 | if (txq < 1 || txq > dev->num_tx_queues) |
1996 | return -EINVAL; | |
f0796d5c | 1997 | |
5c56580b BH |
1998 | if (dev->reg_state == NETREG_REGISTERED || |
1999 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2000 | ASSERT_RTNL(); |
2001 | ||
1d24eb48 TH |
2002 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2003 | txq); | |
bf264145 TH |
2004 | if (rc) |
2005 | return rc; | |
2006 | ||
4f57c087 JF |
2007 | if (dev->num_tc) |
2008 | netif_setup_tc(dev, txq); | |
2009 | ||
024e9679 | 2010 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2011 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2012 | #ifdef CONFIG_XPS |
2013 | netif_reset_xps_queues_gt(dev, txq); | |
2014 | #endif | |
2015 | } | |
f0796d5c | 2016 | } |
e6484930 TH |
2017 | |
2018 | dev->real_num_tx_queues = txq; | |
2019 | return 0; | |
f0796d5c JF |
2020 | } |
2021 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2022 | |
62fe0b40 BH |
2023 | #ifdef CONFIG_RPS |
2024 | /** | |
2025 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2026 | * @dev: Network device | |
2027 | * @rxq: Actual number of RX queues | |
2028 | * | |
2029 | * This must be called either with the rtnl_lock held or before | |
2030 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2031 | * negative error code. If called before registration, it always |
2032 | * succeeds. | |
62fe0b40 BH |
2033 | */ |
2034 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2035 | { | |
2036 | int rc; | |
2037 | ||
bd25fa7b TH |
2038 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2039 | return -EINVAL; | |
2040 | ||
62fe0b40 BH |
2041 | if (dev->reg_state == NETREG_REGISTERED) { |
2042 | ASSERT_RTNL(); | |
2043 | ||
62fe0b40 BH |
2044 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2045 | rxq); | |
2046 | if (rc) | |
2047 | return rc; | |
62fe0b40 BH |
2048 | } |
2049 | ||
2050 | dev->real_num_rx_queues = rxq; | |
2051 | return 0; | |
2052 | } | |
2053 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2054 | #endif | |
2055 | ||
2c53040f BH |
2056 | /** |
2057 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2058 | * |
2059 | * This routine should set an upper limit on the number of RSS queues | |
2060 | * used by default by multiqueue devices. | |
2061 | */ | |
a55b138b | 2062 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
2063 | { |
2064 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
2065 | } | |
2066 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2067 | ||
def82a1d | 2068 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 2069 | { |
def82a1d JP |
2070 | struct softnet_data *sd; |
2071 | unsigned long flags; | |
56079431 | 2072 | |
def82a1d JP |
2073 | local_irq_save(flags); |
2074 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
2075 | q->next_sched = NULL; |
2076 | *sd->output_queue_tailp = q; | |
2077 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2078 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2079 | local_irq_restore(flags); | |
2080 | } | |
2081 | ||
2082 | void __netif_schedule(struct Qdisc *q) | |
2083 | { | |
2084 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2085 | __netif_reschedule(q); | |
56079431 DV |
2086 | } |
2087 | EXPORT_SYMBOL(__netif_schedule); | |
2088 | ||
bea3348e | 2089 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 2090 | { |
3578b0c8 | 2091 | if (atomic_dec_and_test(&skb->users)) { |
bea3348e SH |
2092 | struct softnet_data *sd; |
2093 | unsigned long flags; | |
56079431 | 2094 | |
bea3348e SH |
2095 | local_irq_save(flags); |
2096 | sd = &__get_cpu_var(softnet_data); | |
2097 | skb->next = sd->completion_queue; | |
2098 | sd->completion_queue = skb; | |
2099 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2100 | local_irq_restore(flags); | |
2101 | } | |
56079431 | 2102 | } |
bea3348e | 2103 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
2104 | |
2105 | void dev_kfree_skb_any(struct sk_buff *skb) | |
2106 | { | |
2107 | if (in_irq() || irqs_disabled()) | |
2108 | dev_kfree_skb_irq(skb); | |
2109 | else | |
2110 | dev_kfree_skb(skb); | |
2111 | } | |
2112 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
2113 | ||
2114 | ||
bea3348e SH |
2115 | /** |
2116 | * netif_device_detach - mark device as removed | |
2117 | * @dev: network device | |
2118 | * | |
2119 | * Mark device as removed from system and therefore no longer available. | |
2120 | */ | |
56079431 DV |
2121 | void netif_device_detach(struct net_device *dev) |
2122 | { | |
2123 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2124 | netif_running(dev)) { | |
d543103a | 2125 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2126 | } |
2127 | } | |
2128 | EXPORT_SYMBOL(netif_device_detach); | |
2129 | ||
bea3348e SH |
2130 | /** |
2131 | * netif_device_attach - mark device as attached | |
2132 | * @dev: network device | |
2133 | * | |
2134 | * Mark device as attached from system and restart if needed. | |
2135 | */ | |
56079431 DV |
2136 | void netif_device_attach(struct net_device *dev) |
2137 | { | |
2138 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2139 | netif_running(dev)) { | |
d543103a | 2140 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2141 | __netdev_watchdog_up(dev); |
56079431 DV |
2142 | } |
2143 | } | |
2144 | EXPORT_SYMBOL(netif_device_attach); | |
2145 | ||
36c92474 BH |
2146 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2147 | { | |
65e9d2fa | 2148 | static const netdev_features_t null_features = 0; |
36c92474 BH |
2149 | struct net_device *dev = skb->dev; |
2150 | const char *driver = ""; | |
2151 | ||
2152 | if (dev && dev->dev.parent) | |
2153 | driver = dev_driver_string(dev->dev.parent); | |
2154 | ||
2155 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " | |
2156 | "gso_type=%d ip_summed=%d\n", | |
65e9d2fa MM |
2157 | driver, dev ? &dev->features : &null_features, |
2158 | skb->sk ? &skb->sk->sk_route_caps : &null_features, | |
36c92474 BH |
2159 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2160 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2161 | } | |
2162 | ||
1da177e4 LT |
2163 | /* |
2164 | * Invalidate hardware checksum when packet is to be mangled, and | |
2165 | * complete checksum manually on outgoing path. | |
2166 | */ | |
84fa7933 | 2167 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2168 | { |
d3bc23e7 | 2169 | __wsum csum; |
663ead3b | 2170 | int ret = 0, offset; |
1da177e4 | 2171 | |
84fa7933 | 2172 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2173 | goto out_set_summed; |
2174 | ||
2175 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2176 | skb_warn_bad_offload(skb); |
2177 | return -EINVAL; | |
1da177e4 LT |
2178 | } |
2179 | ||
cef401de ED |
2180 | /* Before computing a checksum, we should make sure no frag could |
2181 | * be modified by an external entity : checksum could be wrong. | |
2182 | */ | |
2183 | if (skb_has_shared_frag(skb)) { | |
2184 | ret = __skb_linearize(skb); | |
2185 | if (ret) | |
2186 | goto out; | |
2187 | } | |
2188 | ||
55508d60 | 2189 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2190 | BUG_ON(offset >= skb_headlen(skb)); |
2191 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2192 | ||
2193 | offset += skb->csum_offset; | |
2194 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2195 | ||
2196 | if (skb_cloned(skb) && | |
2197 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2198 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2199 | if (ret) | |
2200 | goto out; | |
2201 | } | |
2202 | ||
a030847e | 2203 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2204 | out_set_summed: |
1da177e4 | 2205 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2206 | out: |
1da177e4 LT |
2207 | return ret; |
2208 | } | |
d1b19dff | 2209 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2210 | |
f6a78bfc | 2211 | /** |
05e8ef4a | 2212 | * skb_mac_gso_segment - mac layer segmentation handler. |
f6a78bfc | 2213 | * @skb: buffer to segment |
576a30eb | 2214 | * @features: features for the output path (see dev->features) |
f6a78bfc | 2215 | */ |
05e8ef4a PS |
2216 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, |
2217 | netdev_features_t features) | |
f6a78bfc HX |
2218 | { |
2219 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
22061d80 | 2220 | struct packet_offload *ptype; |
252e3346 | 2221 | __be16 type = skb->protocol; |
f6a78bfc | 2222 | |
c8d5bcd1 | 2223 | while (type == htons(ETH_P_8021Q)) { |
05e8ef4a | 2224 | int vlan_depth = ETH_HLEN; |
c8d5bcd1 | 2225 | struct vlan_hdr *vh; |
7b9c6090 | 2226 | |
c8d5bcd1 | 2227 | if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN))) |
7b9c6090 JG |
2228 | return ERR_PTR(-EINVAL); |
2229 | ||
c8d5bcd1 JG |
2230 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); |
2231 | type = vh->h_vlan_encapsulated_proto; | |
2232 | vlan_depth += VLAN_HLEN; | |
7b9c6090 JG |
2233 | } |
2234 | ||
f6a78bfc HX |
2235 | __skb_pull(skb, skb->mac_len); |
2236 | ||
2237 | rcu_read_lock(); | |
22061d80 | 2238 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2239 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
84fa7933 | 2240 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
05e8ef4a PS |
2241 | int err; |
2242 | ||
f191a1d1 | 2243 | err = ptype->callbacks.gso_send_check(skb); |
a430a43d HX |
2244 | segs = ERR_PTR(err); |
2245 | if (err || skb_gso_ok(skb, features)) | |
2246 | break; | |
d56f90a7 ACM |
2247 | __skb_push(skb, (skb->data - |
2248 | skb_network_header(skb))); | |
a430a43d | 2249 | } |
f191a1d1 | 2250 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2251 | break; |
2252 | } | |
2253 | } | |
2254 | rcu_read_unlock(); | |
2255 | ||
98e399f8 | 2256 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2257 | |
f6a78bfc HX |
2258 | return segs; |
2259 | } | |
05e8ef4a PS |
2260 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2261 | ||
2262 | ||
2263 | /* openvswitch calls this on rx path, so we need a different check. | |
2264 | */ | |
2265 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2266 | { | |
2267 | if (tx_path) | |
2268 | return skb->ip_summed != CHECKSUM_PARTIAL; | |
2269 | else | |
2270 | return skb->ip_summed == CHECKSUM_NONE; | |
2271 | } | |
2272 | ||
2273 | /** | |
2274 | * __skb_gso_segment - Perform segmentation on skb. | |
2275 | * @skb: buffer to segment | |
2276 | * @features: features for the output path (see dev->features) | |
2277 | * @tx_path: whether it is called in TX path | |
2278 | * | |
2279 | * This function segments the given skb and returns a list of segments. | |
2280 | * | |
2281 | * It may return NULL if the skb requires no segmentation. This is | |
2282 | * only possible when GSO is used for verifying header integrity. | |
2283 | */ | |
2284 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2285 | netdev_features_t features, bool tx_path) | |
2286 | { | |
2287 | if (unlikely(skb_needs_check(skb, tx_path))) { | |
2288 | int err; | |
2289 | ||
2290 | skb_warn_bad_offload(skb); | |
2291 | ||
2292 | if (skb_header_cloned(skb) && | |
2293 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
2294 | return ERR_PTR(err); | |
2295 | } | |
2296 | ||
68c33163 | 2297 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
05e8ef4a PS |
2298 | skb_reset_mac_header(skb); |
2299 | skb_reset_mac_len(skb); | |
2300 | ||
2301 | return skb_mac_gso_segment(skb, features); | |
2302 | } | |
12b0004d | 2303 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2304 | |
fb286bb2 HX |
2305 | /* Take action when hardware reception checksum errors are detected. */ |
2306 | #ifdef CONFIG_BUG | |
2307 | void netdev_rx_csum_fault(struct net_device *dev) | |
2308 | { | |
2309 | if (net_ratelimit()) { | |
7b6cd1ce | 2310 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2311 | dump_stack(); |
2312 | } | |
2313 | } | |
2314 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2315 | #endif | |
2316 | ||
1da177e4 LT |
2317 | /* Actually, we should eliminate this check as soon as we know, that: |
2318 | * 1. IOMMU is present and allows to map all the memory. | |
2319 | * 2. No high memory really exists on this machine. | |
2320 | */ | |
2321 | ||
9092c658 | 2322 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2323 | { |
3d3a8533 | 2324 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2325 | int i; |
5acbbd42 | 2326 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2327 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2328 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2329 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2330 | return 1; |
ea2ab693 | 2331 | } |
5acbbd42 | 2332 | } |
1da177e4 | 2333 | |
5acbbd42 FT |
2334 | if (PCI_DMA_BUS_IS_PHYS) { |
2335 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2336 | |
9092c658 ED |
2337 | if (!pdev) |
2338 | return 0; | |
5acbbd42 | 2339 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2340 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2341 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2342 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2343 | return 1; | |
2344 | } | |
2345 | } | |
3d3a8533 | 2346 | #endif |
1da177e4 LT |
2347 | return 0; |
2348 | } | |
1da177e4 | 2349 | |
f6a78bfc HX |
2350 | struct dev_gso_cb { |
2351 | void (*destructor)(struct sk_buff *skb); | |
2352 | }; | |
2353 | ||
2354 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
2355 | ||
2356 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
2357 | { | |
2358 | struct dev_gso_cb *cb; | |
2359 | ||
2360 | do { | |
2361 | struct sk_buff *nskb = skb->next; | |
2362 | ||
2363 | skb->next = nskb->next; | |
2364 | nskb->next = NULL; | |
2365 | kfree_skb(nskb); | |
2366 | } while (skb->next); | |
2367 | ||
2368 | cb = DEV_GSO_CB(skb); | |
2369 | if (cb->destructor) | |
2370 | cb->destructor(skb); | |
2371 | } | |
2372 | ||
2373 | /** | |
2374 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
2375 | * @skb: buffer to segment | |
91ecb63c | 2376 | * @features: device features as applicable to this skb |
f6a78bfc HX |
2377 | * |
2378 | * This function segments the given skb and stores the list of segments | |
2379 | * in skb->next. | |
2380 | */ | |
c8f44aff | 2381 | static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features) |
f6a78bfc | 2382 | { |
f6a78bfc | 2383 | struct sk_buff *segs; |
576a30eb HX |
2384 | |
2385 | segs = skb_gso_segment(skb, features); | |
2386 | ||
2387 | /* Verifying header integrity only. */ | |
2388 | if (!segs) | |
2389 | return 0; | |
f6a78bfc | 2390 | |
801678c5 | 2391 | if (IS_ERR(segs)) |
f6a78bfc HX |
2392 | return PTR_ERR(segs); |
2393 | ||
2394 | skb->next = segs; | |
2395 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
2396 | skb->destructor = dev_gso_skb_destructor; | |
2397 | ||
2398 | return 0; | |
2399 | } | |
2400 | ||
c8f44aff | 2401 | static bool can_checksum_protocol(netdev_features_t features, __be16 protocol) |
03634668 JG |
2402 | { |
2403 | return ((features & NETIF_F_GEN_CSUM) || | |
2404 | ((features & NETIF_F_V4_CSUM) && | |
2405 | protocol == htons(ETH_P_IP)) || | |
2406 | ((features & NETIF_F_V6_CSUM) && | |
2407 | protocol == htons(ETH_P_IPV6)) || | |
2408 | ((features & NETIF_F_FCOE_CRC) && | |
2409 | protocol == htons(ETH_P_FCOE))); | |
2410 | } | |
2411 | ||
c8f44aff MM |
2412 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
2413 | __be16 protocol, netdev_features_t features) | |
f01a5236 | 2414 | { |
c0d680e5 EC |
2415 | if (skb->ip_summed != CHECKSUM_NONE && |
2416 | !can_checksum_protocol(features, protocol)) { | |
f01a5236 JG |
2417 | features &= ~NETIF_F_ALL_CSUM; |
2418 | features &= ~NETIF_F_SG; | |
2419 | } else if (illegal_highdma(skb->dev, skb)) { | |
2420 | features &= ~NETIF_F_SG; | |
2421 | } | |
2422 | ||
2423 | return features; | |
2424 | } | |
2425 | ||
c8f44aff | 2426 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 JG |
2427 | { |
2428 | __be16 protocol = skb->protocol; | |
c8f44aff | 2429 | netdev_features_t features = skb->dev->features; |
58e998c6 | 2430 | |
30b678d8 BH |
2431 | if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs) |
2432 | features &= ~NETIF_F_GSO_MASK; | |
2433 | ||
58e998c6 JG |
2434 | if (protocol == htons(ETH_P_8021Q)) { |
2435 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
2436 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 JG |
2437 | } else if (!vlan_tx_tag_present(skb)) { |
2438 | return harmonize_features(skb, protocol, features); | |
2439 | } | |
58e998c6 | 2440 | |
6ee400aa | 2441 | features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX); |
f01a5236 JG |
2442 | |
2443 | if (protocol != htons(ETH_P_8021Q)) { | |
2444 | return harmonize_features(skb, protocol, features); | |
2445 | } else { | |
2446 | features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | | |
6ee400aa | 2447 | NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX; |
f01a5236 JG |
2448 | return harmonize_features(skb, protocol, features); |
2449 | } | |
58e998c6 | 2450 | } |
f01a5236 | 2451 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2452 | |
6afff0ca JF |
2453 | /* |
2454 | * Returns true if either: | |
2455 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
d1a53dfd | 2456 | * 2. skb is fragmented and the device does not support SG. |
6afff0ca JF |
2457 | */ |
2458 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
02932ce9 | 2459 | int features) |
6afff0ca | 2460 | { |
02932ce9 JG |
2461 | return skb_is_nonlinear(skb) && |
2462 | ((skb_has_frag_list(skb) && | |
2463 | !(features & NETIF_F_FRAGLIST)) || | |
e1e78db6 | 2464 | (skb_shinfo(skb)->nr_frags && |
02932ce9 | 2465 | !(features & NETIF_F_SG))); |
6afff0ca JF |
2466 | } |
2467 | ||
fd2ea0a7 DM |
2468 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
2469 | struct netdev_queue *txq) | |
f6a78bfc | 2470 | { |
00829823 | 2471 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 2472 | int rc = NETDEV_TX_OK; |
ec764bf0 | 2473 | unsigned int skb_len; |
00829823 | 2474 | |
f6a78bfc | 2475 | if (likely(!skb->next)) { |
c8f44aff | 2476 | netdev_features_t features; |
fc741216 | 2477 | |
93f154b5 | 2478 | /* |
25985edc | 2479 | * If device doesn't need skb->dst, release it right now while |
93f154b5 ED |
2480 | * its hot in this cpu cache |
2481 | */ | |
adf30907 ED |
2482 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
2483 | skb_dst_drop(skb); | |
2484 | ||
fc741216 JG |
2485 | features = netif_skb_features(skb); |
2486 | ||
7b9c6090 | 2487 | if (vlan_tx_tag_present(skb) && |
fc741216 | 2488 | !(features & NETIF_F_HW_VLAN_TX)) { |
7b9c6090 JG |
2489 | skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); |
2490 | if (unlikely(!skb)) | |
2491 | goto out; | |
2492 | ||
2493 | skb->vlan_tci = 0; | |
2494 | } | |
2495 | ||
fc70fb64 AD |
2496 | /* If encapsulation offload request, verify we are testing |
2497 | * hardware encapsulation features instead of standard | |
2498 | * features for the netdev | |
2499 | */ | |
2500 | if (skb->encapsulation) | |
2501 | features &= dev->hw_enc_features; | |
2502 | ||
fc741216 | 2503 | if (netif_needs_gso(skb, features)) { |
91ecb63c | 2504 | if (unlikely(dev_gso_segment(skb, features))) |
9ccb8975 DM |
2505 | goto out_kfree_skb; |
2506 | if (skb->next) | |
2507 | goto gso; | |
6afff0ca | 2508 | } else { |
02932ce9 | 2509 | if (skb_needs_linearize(skb, features) && |
6afff0ca JF |
2510 | __skb_linearize(skb)) |
2511 | goto out_kfree_skb; | |
2512 | ||
2513 | /* If packet is not checksummed and device does not | |
2514 | * support checksumming for this protocol, complete | |
2515 | * checksumming here. | |
2516 | */ | |
2517 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
fc70fb64 AD |
2518 | if (skb->encapsulation) |
2519 | skb_set_inner_transport_header(skb, | |
2520 | skb_checksum_start_offset(skb)); | |
2521 | else | |
2522 | skb_set_transport_header(skb, | |
2523 | skb_checksum_start_offset(skb)); | |
03634668 | 2524 | if (!(features & NETIF_F_ALL_CSUM) && |
6afff0ca JF |
2525 | skb_checksum_help(skb)) |
2526 | goto out_kfree_skb; | |
2527 | } | |
9ccb8975 DM |
2528 | } |
2529 | ||
b40863c6 ED |
2530 | if (!list_empty(&ptype_all)) |
2531 | dev_queue_xmit_nit(skb, dev); | |
2532 | ||
ec764bf0 | 2533 | skb_len = skb->len; |
ac45f602 | 2534 | rc = ops->ndo_start_xmit(skb, dev); |
ec764bf0 | 2535 | trace_net_dev_xmit(skb, rc, dev, skb_len); |
ec634fe3 | 2536 | if (rc == NETDEV_TX_OK) |
08baf561 | 2537 | txq_trans_update(txq); |
ac45f602 | 2538 | return rc; |
f6a78bfc HX |
2539 | } |
2540 | ||
576a30eb | 2541 | gso: |
f6a78bfc HX |
2542 | do { |
2543 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
2544 | |
2545 | skb->next = nskb->next; | |
2546 | nskb->next = NULL; | |
068a2de5 KK |
2547 | |
2548 | /* | |
25985edc | 2549 | * If device doesn't need nskb->dst, release it right now while |
068a2de5 KK |
2550 | * its hot in this cpu cache |
2551 | */ | |
2552 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2553 | skb_dst_drop(nskb); | |
2554 | ||
b40863c6 ED |
2555 | if (!list_empty(&ptype_all)) |
2556 | dev_queue_xmit_nit(nskb, dev); | |
2557 | ||
ec764bf0 | 2558 | skb_len = nskb->len; |
00829823 | 2559 | rc = ops->ndo_start_xmit(nskb, dev); |
ec764bf0 | 2560 | trace_net_dev_xmit(nskb, rc, dev, skb_len); |
ec634fe3 | 2561 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
2562 | if (rc & ~NETDEV_TX_MASK) |
2563 | goto out_kfree_gso_skb; | |
f54d9e8d | 2564 | nskb->next = skb->next; |
f6a78bfc HX |
2565 | skb->next = nskb; |
2566 | return rc; | |
2567 | } | |
08baf561 | 2568 | txq_trans_update(txq); |
73466498 | 2569 | if (unlikely(netif_xmit_stopped(txq) && skb->next)) |
f54d9e8d | 2570 | return NETDEV_TX_BUSY; |
f6a78bfc | 2571 | } while (skb->next); |
4ec93edb | 2572 | |
572a9d7b PM |
2573 | out_kfree_gso_skb: |
2574 | if (likely(skb->next == NULL)) | |
2575 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
f6a78bfc HX |
2576 | out_kfree_skb: |
2577 | kfree_skb(skb); | |
7b9c6090 | 2578 | out: |
572a9d7b | 2579 | return rc; |
f6a78bfc HX |
2580 | } |
2581 | ||
1def9238 ED |
2582 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
2583 | { | |
2584 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2585 | ||
2586 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2587 | ||
2588 | /* To get more precise estimation of bytes sent on wire, | |
2589 | * we add to pkt_len the headers size of all segments | |
2590 | */ | |
2591 | if (shinfo->gso_size) { | |
757b8b1d | 2592 | unsigned int hdr_len; |
1def9238 | 2593 | |
757b8b1d ED |
2594 | /* mac layer + network layer */ |
2595 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
2596 | ||
2597 | /* + transport layer */ | |
1def9238 ED |
2598 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
2599 | hdr_len += tcp_hdrlen(skb); | |
2600 | else | |
2601 | hdr_len += sizeof(struct udphdr); | |
2602 | qdisc_skb_cb(skb)->pkt_len += (shinfo->gso_segs - 1) * hdr_len; | |
2603 | } | |
2604 | } | |
2605 | ||
bbd8a0d3 KK |
2606 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2607 | struct net_device *dev, | |
2608 | struct netdev_queue *txq) | |
2609 | { | |
2610 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2611 | bool contended; |
bbd8a0d3 KK |
2612 | int rc; |
2613 | ||
1def9238 | 2614 | qdisc_pkt_len_init(skb); |
a2da570d | 2615 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
2616 | /* |
2617 | * Heuristic to force contended enqueues to serialize on a | |
2618 | * separate lock before trying to get qdisc main lock. | |
2619 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2620 | * and dequeue packets faster. | |
2621 | */ | |
a2da570d | 2622 | contended = qdisc_is_running(q); |
79640a4c ED |
2623 | if (unlikely(contended)) |
2624 | spin_lock(&q->busylock); | |
2625 | ||
bbd8a0d3 KK |
2626 | spin_lock(root_lock); |
2627 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2628 | kfree_skb(skb); | |
2629 | rc = NET_XMIT_DROP; | |
2630 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2631 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2632 | /* |
2633 | * This is a work-conserving queue; there are no old skbs | |
2634 | * waiting to be sent out; and the qdisc is not running - | |
2635 | * xmit the skb directly. | |
2636 | */ | |
7fee226a ED |
2637 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2638 | skb_dst_force(skb); | |
bfe0d029 | 2639 | |
bfe0d029 ED |
2640 | qdisc_bstats_update(q, skb); |
2641 | ||
79640a4c ED |
2642 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2643 | if (unlikely(contended)) { | |
2644 | spin_unlock(&q->busylock); | |
2645 | contended = false; | |
2646 | } | |
bbd8a0d3 | 2647 | __qdisc_run(q); |
79640a4c | 2648 | } else |
bc135b23 | 2649 | qdisc_run_end(q); |
bbd8a0d3 KK |
2650 | |
2651 | rc = NET_XMIT_SUCCESS; | |
2652 | } else { | |
7fee226a | 2653 | skb_dst_force(skb); |
a2da570d | 2654 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2655 | if (qdisc_run_begin(q)) { |
2656 | if (unlikely(contended)) { | |
2657 | spin_unlock(&q->busylock); | |
2658 | contended = false; | |
2659 | } | |
2660 | __qdisc_run(q); | |
2661 | } | |
bbd8a0d3 KK |
2662 | } |
2663 | spin_unlock(root_lock); | |
79640a4c ED |
2664 | if (unlikely(contended)) |
2665 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2666 | return rc; |
2667 | } | |
2668 | ||
5bc1421e NH |
2669 | #if IS_ENABLED(CONFIG_NETPRIO_CGROUP) |
2670 | static void skb_update_prio(struct sk_buff *skb) | |
2671 | { | |
6977a79d | 2672 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 2673 | |
91c68ce2 ED |
2674 | if (!skb->priority && skb->sk && map) { |
2675 | unsigned int prioidx = skb->sk->sk_cgrp_prioidx; | |
2676 | ||
2677 | if (prioidx < map->priomap_len) | |
2678 | skb->priority = map->priomap[prioidx]; | |
2679 | } | |
5bc1421e NH |
2680 | } |
2681 | #else | |
2682 | #define skb_update_prio(skb) | |
2683 | #endif | |
2684 | ||
745e20f1 | 2685 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2686 | #define RECURSION_LIMIT 10 |
745e20f1 | 2687 | |
95603e22 MM |
2688 | /** |
2689 | * dev_loopback_xmit - loop back @skb | |
2690 | * @skb: buffer to transmit | |
2691 | */ | |
2692 | int dev_loopback_xmit(struct sk_buff *skb) | |
2693 | { | |
2694 | skb_reset_mac_header(skb); | |
2695 | __skb_pull(skb, skb_network_offset(skb)); | |
2696 | skb->pkt_type = PACKET_LOOPBACK; | |
2697 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2698 | WARN_ON(!skb_dst(skb)); | |
2699 | skb_dst_force(skb); | |
2700 | netif_rx_ni(skb); | |
2701 | return 0; | |
2702 | } | |
2703 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2704 | ||
d29f749e DJ |
2705 | /** |
2706 | * dev_queue_xmit - transmit a buffer | |
2707 | * @skb: buffer to transmit | |
2708 | * | |
2709 | * Queue a buffer for transmission to a network device. The caller must | |
2710 | * have set the device and priority and built the buffer before calling | |
2711 | * this function. The function can be called from an interrupt. | |
2712 | * | |
2713 | * A negative errno code is returned on a failure. A success does not | |
2714 | * guarantee the frame will be transmitted as it may be dropped due | |
2715 | * to congestion or traffic shaping. | |
2716 | * | |
2717 | * ----------------------------------------------------------------------------------- | |
2718 | * I notice this method can also return errors from the queue disciplines, | |
2719 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2720 | * be positive. | |
2721 | * | |
2722 | * Regardless of the return value, the skb is consumed, so it is currently | |
2723 | * difficult to retry a send to this method. (You can bump the ref count | |
2724 | * before sending to hold a reference for retry if you are careful.) | |
2725 | * | |
2726 | * When calling this method, interrupts MUST be enabled. This is because | |
2727 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2728 | * --BLG | |
2729 | */ | |
1da177e4 LT |
2730 | int dev_queue_xmit(struct sk_buff *skb) |
2731 | { | |
2732 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2733 | struct netdev_queue *txq; |
1da177e4 LT |
2734 | struct Qdisc *q; |
2735 | int rc = -ENOMEM; | |
2736 | ||
6d1ccff6 ED |
2737 | skb_reset_mac_header(skb); |
2738 | ||
4ec93edb YH |
2739 | /* Disable soft irqs for various locks below. Also |
2740 | * stops preemption for RCU. | |
1da177e4 | 2741 | */ |
4ec93edb | 2742 | rcu_read_lock_bh(); |
1da177e4 | 2743 | |
5bc1421e NH |
2744 | skb_update_prio(skb); |
2745 | ||
8c4c49df | 2746 | txq = netdev_pick_tx(dev, skb); |
a898def2 | 2747 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2748 | |
1da177e4 | 2749 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2750 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2751 | #endif |
cf66ba58 | 2752 | trace_net_dev_queue(skb); |
1da177e4 | 2753 | if (q->enqueue) { |
bbd8a0d3 | 2754 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2755 | goto out; |
1da177e4 LT |
2756 | } |
2757 | ||
2758 | /* The device has no queue. Common case for software devices: | |
2759 | loopback, all the sorts of tunnels... | |
2760 | ||
932ff279 HX |
2761 | Really, it is unlikely that netif_tx_lock protection is necessary |
2762 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2763 | counters.) |
2764 | However, it is possible, that they rely on protection | |
2765 | made by us here. | |
2766 | ||
2767 | Check this and shot the lock. It is not prone from deadlocks. | |
2768 | Either shot noqueue qdisc, it is even simpler 8) | |
2769 | */ | |
2770 | if (dev->flags & IFF_UP) { | |
2771 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2772 | ||
c773e847 | 2773 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2774 | |
745e20f1 ED |
2775 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2776 | goto recursion_alert; | |
2777 | ||
c773e847 | 2778 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2779 | |
73466498 | 2780 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 2781 | __this_cpu_inc(xmit_recursion); |
572a9d7b | 2782 | rc = dev_hard_start_xmit(skb, dev, txq); |
745e20f1 | 2783 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2784 | if (dev_xmit_complete(rc)) { |
c773e847 | 2785 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2786 | goto out; |
2787 | } | |
2788 | } | |
c773e847 | 2789 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
2790 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
2791 | dev->name); | |
1da177e4 LT |
2792 | } else { |
2793 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2794 | * unfortunately |
2795 | */ | |
2796 | recursion_alert: | |
e87cc472 JP |
2797 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
2798 | dev->name); | |
1da177e4 LT |
2799 | } |
2800 | } | |
2801 | ||
2802 | rc = -ENETDOWN; | |
d4828d85 | 2803 | rcu_read_unlock_bh(); |
1da177e4 | 2804 | |
1da177e4 LT |
2805 | kfree_skb(skb); |
2806 | return rc; | |
2807 | out: | |
d4828d85 | 2808 | rcu_read_unlock_bh(); |
1da177e4 LT |
2809 | return rc; |
2810 | } | |
d1b19dff | 2811 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2812 | |
2813 | ||
2814 | /*======================================================================= | |
2815 | Receiver routines | |
2816 | =======================================================================*/ | |
2817 | ||
6b2bedc3 | 2818 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
2819 | EXPORT_SYMBOL(netdev_max_backlog); |
2820 | ||
3b098e2d | 2821 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2822 | int netdev_budget __read_mostly = 300; |
2823 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2824 | |
eecfd7c4 ED |
2825 | /* Called with irq disabled */ |
2826 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2827 | struct napi_struct *napi) | |
2828 | { | |
2829 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2830 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2831 | } | |
2832 | ||
bfb564e7 KK |
2833 | #ifdef CONFIG_RPS |
2834 | ||
2835 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 2836 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
2837 | EXPORT_SYMBOL(rps_sock_flow_table); |
2838 | ||
c5905afb | 2839 | struct static_key rps_needed __read_mostly; |
adc9300e | 2840 | |
c445477d BH |
2841 | static struct rps_dev_flow * |
2842 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2843 | struct rps_dev_flow *rflow, u16 next_cpu) | |
2844 | { | |
09994d1b | 2845 | if (next_cpu != RPS_NO_CPU) { |
c445477d BH |
2846 | #ifdef CONFIG_RFS_ACCEL |
2847 | struct netdev_rx_queue *rxqueue; | |
2848 | struct rps_dev_flow_table *flow_table; | |
2849 | struct rps_dev_flow *old_rflow; | |
2850 | u32 flow_id; | |
2851 | u16 rxq_index; | |
2852 | int rc; | |
2853 | ||
2854 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
2855 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
2856 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
2857 | goto out; |
2858 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
2859 | if (rxq_index == skb_get_rx_queue(skb)) | |
2860 | goto out; | |
2861 | ||
2862 | rxqueue = dev->_rx + rxq_index; | |
2863 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2864 | if (!flow_table) | |
2865 | goto out; | |
2866 | flow_id = skb->rxhash & flow_table->mask; | |
2867 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, | |
2868 | rxq_index, flow_id); | |
2869 | if (rc < 0) | |
2870 | goto out; | |
2871 | old_rflow = rflow; | |
2872 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
2873 | rflow->filter = rc; |
2874 | if (old_rflow->filter == rflow->filter) | |
2875 | old_rflow->filter = RPS_NO_FILTER; | |
2876 | out: | |
2877 | #endif | |
2878 | rflow->last_qtail = | |
09994d1b | 2879 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
2880 | } |
2881 | ||
09994d1b | 2882 | rflow->cpu = next_cpu; |
c445477d BH |
2883 | return rflow; |
2884 | } | |
2885 | ||
bfb564e7 KK |
2886 | /* |
2887 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2888 | * CPU from the RPS map of the receiving queue for a given skb. | |
2889 | * rcu_read_lock must be held on entry. | |
2890 | */ | |
2891 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2892 | struct rps_dev_flow **rflowp) | |
2893 | { | |
2894 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 2895 | struct rps_map *map; |
bfb564e7 KK |
2896 | struct rps_dev_flow_table *flow_table; |
2897 | struct rps_sock_flow_table *sock_flow_table; | |
2898 | int cpu = -1; | |
2899 | u16 tcpu; | |
2900 | ||
2901 | if (skb_rx_queue_recorded(skb)) { | |
2902 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
2903 | if (unlikely(index >= dev->real_num_rx_queues)) { |
2904 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
2905 | "%s received packet on queue %u, but number " | |
2906 | "of RX queues is %u\n", | |
2907 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
2908 | goto done; |
2909 | } | |
2910 | rxqueue = dev->_rx + index; | |
2911 | } else | |
2912 | rxqueue = dev->_rx; | |
2913 | ||
6e3f7faf ED |
2914 | map = rcu_dereference(rxqueue->rps_map); |
2915 | if (map) { | |
85875236 | 2916 | if (map->len == 1 && |
33d480ce | 2917 | !rcu_access_pointer(rxqueue->rps_flow_table)) { |
6febfca9 CG |
2918 | tcpu = map->cpus[0]; |
2919 | if (cpu_online(tcpu)) | |
2920 | cpu = tcpu; | |
2921 | goto done; | |
2922 | } | |
33d480ce | 2923 | } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) { |
bfb564e7 | 2924 | goto done; |
6febfca9 | 2925 | } |
bfb564e7 | 2926 | |
2d47b459 | 2927 | skb_reset_network_header(skb); |
bfb564e7 KK |
2928 | if (!skb_get_rxhash(skb)) |
2929 | goto done; | |
2930 | ||
fec5e652 TH |
2931 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
2932 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
2933 | if (flow_table && sock_flow_table) { | |
2934 | u16 next_cpu; | |
2935 | struct rps_dev_flow *rflow; | |
2936 | ||
2937 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
2938 | tcpu = rflow->cpu; | |
2939 | ||
2940 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
2941 | sock_flow_table->mask]; | |
2942 | ||
2943 | /* | |
2944 | * If the desired CPU (where last recvmsg was done) is | |
2945 | * different from current CPU (one in the rx-queue flow | |
2946 | * table entry), switch if one of the following holds: | |
2947 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
2948 | * - Current CPU is offline. | |
2949 | * - The current CPU's queue tail has advanced beyond the | |
2950 | * last packet that was enqueued using this table entry. | |
2951 | * This guarantees that all previous packets for the flow | |
2952 | * have been dequeued, thus preserving in order delivery. | |
2953 | */ | |
2954 | if (unlikely(tcpu != next_cpu) && | |
2955 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
2956 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
baefa31d TH |
2957 | rflow->last_qtail)) >= 0)) { |
2958 | tcpu = next_cpu; | |
c445477d | 2959 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 2960 | } |
c445477d | 2961 | |
fec5e652 TH |
2962 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
2963 | *rflowp = rflow; | |
2964 | cpu = tcpu; | |
2965 | goto done; | |
2966 | } | |
2967 | } | |
2968 | ||
0a9627f2 | 2969 | if (map) { |
fec5e652 | 2970 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; |
0a9627f2 TH |
2971 | |
2972 | if (cpu_online(tcpu)) { | |
2973 | cpu = tcpu; | |
2974 | goto done; | |
2975 | } | |
2976 | } | |
2977 | ||
2978 | done: | |
0a9627f2 TH |
2979 | return cpu; |
2980 | } | |
2981 | ||
c445477d BH |
2982 | #ifdef CONFIG_RFS_ACCEL |
2983 | ||
2984 | /** | |
2985 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
2986 | * @dev: Device on which the filter was set | |
2987 | * @rxq_index: RX queue index | |
2988 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
2989 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
2990 | * | |
2991 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
2992 | * this function for each installed filter and remove the filters for | |
2993 | * which it returns %true. | |
2994 | */ | |
2995 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
2996 | u32 flow_id, u16 filter_id) | |
2997 | { | |
2998 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
2999 | struct rps_dev_flow_table *flow_table; | |
3000 | struct rps_dev_flow *rflow; | |
3001 | bool expire = true; | |
3002 | int cpu; | |
3003 | ||
3004 | rcu_read_lock(); | |
3005 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3006 | if (flow_table && flow_id <= flow_table->mask) { | |
3007 | rflow = &flow_table->flows[flow_id]; | |
3008 | cpu = ACCESS_ONCE(rflow->cpu); | |
3009 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
3010 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
3011 | rflow->last_qtail) < | |
3012 | (int)(10 * flow_table->mask))) | |
3013 | expire = false; | |
3014 | } | |
3015 | rcu_read_unlock(); | |
3016 | return expire; | |
3017 | } | |
3018 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3019 | ||
3020 | #endif /* CONFIG_RFS_ACCEL */ | |
3021 | ||
0a9627f2 | 3022 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3023 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3024 | { |
e36fa2f7 ED |
3025 | struct softnet_data *sd = data; |
3026 | ||
eecfd7c4 | 3027 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3028 | sd->received_rps++; |
0a9627f2 | 3029 | } |
e36fa2f7 | 3030 | |
fec5e652 | 3031 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3032 | |
e36fa2f7 ED |
3033 | /* |
3034 | * Check if this softnet_data structure is another cpu one | |
3035 | * If yes, queue it to our IPI list and return 1 | |
3036 | * If no, return 0 | |
3037 | */ | |
3038 | static int rps_ipi_queued(struct softnet_data *sd) | |
3039 | { | |
3040 | #ifdef CONFIG_RPS | |
3041 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
3042 | ||
3043 | if (sd != mysd) { | |
3044 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3045 | mysd->rps_ipi_list = sd; | |
3046 | ||
3047 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3048 | return 1; | |
3049 | } | |
3050 | #endif /* CONFIG_RPS */ | |
3051 | return 0; | |
3052 | } | |
3053 | ||
0a9627f2 TH |
3054 | /* |
3055 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3056 | * queue (may be a remote CPU queue). | |
3057 | */ | |
fec5e652 TH |
3058 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3059 | unsigned int *qtail) | |
0a9627f2 | 3060 | { |
e36fa2f7 | 3061 | struct softnet_data *sd; |
0a9627f2 TH |
3062 | unsigned long flags; |
3063 | ||
e36fa2f7 | 3064 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3065 | |
3066 | local_irq_save(flags); | |
0a9627f2 | 3067 | |
e36fa2f7 | 3068 | rps_lock(sd); |
6e7676c1 CG |
3069 | if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) { |
3070 | if (skb_queue_len(&sd->input_pkt_queue)) { | |
0a9627f2 | 3071 | enqueue: |
e36fa2f7 | 3072 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3073 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3074 | rps_unlock(sd); |
152102c7 | 3075 | local_irq_restore(flags); |
0a9627f2 TH |
3076 | return NET_RX_SUCCESS; |
3077 | } | |
3078 | ||
ebda37c2 ED |
3079 | /* Schedule NAPI for backlog device |
3080 | * We can use non atomic operation since we own the queue lock | |
3081 | */ | |
3082 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3083 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3084 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3085 | } |
3086 | goto enqueue; | |
3087 | } | |
3088 | ||
dee42870 | 3089 | sd->dropped++; |
e36fa2f7 | 3090 | rps_unlock(sd); |
0a9627f2 | 3091 | |
0a9627f2 TH |
3092 | local_irq_restore(flags); |
3093 | ||
caf586e5 | 3094 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3095 | kfree_skb(skb); |
3096 | return NET_RX_DROP; | |
3097 | } | |
1da177e4 | 3098 | |
1da177e4 LT |
3099 | /** |
3100 | * netif_rx - post buffer to the network code | |
3101 | * @skb: buffer to post | |
3102 | * | |
3103 | * This function receives a packet from a device driver and queues it for | |
3104 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3105 | * may be dropped during processing for congestion control or by the | |
3106 | * protocol layers. | |
3107 | * | |
3108 | * return values: | |
3109 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
3110 | * NET_RX_DROP (packet was dropped) |
3111 | * | |
3112 | */ | |
3113 | ||
3114 | int netif_rx(struct sk_buff *skb) | |
3115 | { | |
b0e28f1e | 3116 | int ret; |
1da177e4 LT |
3117 | |
3118 | /* if netpoll wants it, pretend we never saw it */ | |
3119 | if (netpoll_rx(skb)) | |
3120 | return NET_RX_DROP; | |
3121 | ||
588f0330 | 3122 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3123 | |
cf66ba58 | 3124 | trace_netif_rx(skb); |
df334545 | 3125 | #ifdef CONFIG_RPS |
c5905afb | 3126 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3127 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3128 | int cpu; |
3129 | ||
cece1945 | 3130 | preempt_disable(); |
b0e28f1e | 3131 | rcu_read_lock(); |
fec5e652 TH |
3132 | |
3133 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3134 | if (cpu < 0) |
3135 | cpu = smp_processor_id(); | |
fec5e652 TH |
3136 | |
3137 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3138 | ||
b0e28f1e | 3139 | rcu_read_unlock(); |
cece1945 | 3140 | preempt_enable(); |
adc9300e ED |
3141 | } else |
3142 | #endif | |
fec5e652 TH |
3143 | { |
3144 | unsigned int qtail; | |
3145 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3146 | put_cpu(); | |
3147 | } | |
b0e28f1e | 3148 | return ret; |
1da177e4 | 3149 | } |
d1b19dff | 3150 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3151 | |
3152 | int netif_rx_ni(struct sk_buff *skb) | |
3153 | { | |
3154 | int err; | |
3155 | ||
3156 | preempt_disable(); | |
3157 | err = netif_rx(skb); | |
3158 | if (local_softirq_pending()) | |
3159 | do_softirq(); | |
3160 | preempt_enable(); | |
3161 | ||
3162 | return err; | |
3163 | } | |
1da177e4 LT |
3164 | EXPORT_SYMBOL(netif_rx_ni); |
3165 | ||
1da177e4 LT |
3166 | static void net_tx_action(struct softirq_action *h) |
3167 | { | |
3168 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
3169 | ||
3170 | if (sd->completion_queue) { | |
3171 | struct sk_buff *clist; | |
3172 | ||
3173 | local_irq_disable(); | |
3174 | clist = sd->completion_queue; | |
3175 | sd->completion_queue = NULL; | |
3176 | local_irq_enable(); | |
3177 | ||
3178 | while (clist) { | |
3179 | struct sk_buff *skb = clist; | |
3180 | clist = clist->next; | |
3181 | ||
547b792c | 3182 | WARN_ON(atomic_read(&skb->users)); |
07dc22e7 | 3183 | trace_kfree_skb(skb, net_tx_action); |
1da177e4 LT |
3184 | __kfree_skb(skb); |
3185 | } | |
3186 | } | |
3187 | ||
3188 | if (sd->output_queue) { | |
37437bb2 | 3189 | struct Qdisc *head; |
1da177e4 LT |
3190 | |
3191 | local_irq_disable(); | |
3192 | head = sd->output_queue; | |
3193 | sd->output_queue = NULL; | |
a9cbd588 | 3194 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3195 | local_irq_enable(); |
3196 | ||
3197 | while (head) { | |
37437bb2 DM |
3198 | struct Qdisc *q = head; |
3199 | spinlock_t *root_lock; | |
3200 | ||
1da177e4 LT |
3201 | head = head->next_sched; |
3202 | ||
5fb66229 | 3203 | root_lock = qdisc_lock(q); |
37437bb2 | 3204 | if (spin_trylock(root_lock)) { |
def82a1d JP |
3205 | smp_mb__before_clear_bit(); |
3206 | clear_bit(__QDISC_STATE_SCHED, | |
3207 | &q->state); | |
37437bb2 DM |
3208 | qdisc_run(q); |
3209 | spin_unlock(root_lock); | |
1da177e4 | 3210 | } else { |
195648bb | 3211 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3212 | &q->state)) { |
195648bb | 3213 | __netif_reschedule(q); |
e8a83e10 JP |
3214 | } else { |
3215 | smp_mb__before_clear_bit(); | |
3216 | clear_bit(__QDISC_STATE_SCHED, | |
3217 | &q->state); | |
3218 | } | |
1da177e4 LT |
3219 | } |
3220 | } | |
3221 | } | |
3222 | } | |
3223 | ||
ab95bfe0 JP |
3224 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3225 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3226 | /* This hook is defined here for ATM LANE */ |
3227 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3228 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3229 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3230 | #endif |
1da177e4 | 3231 | |
1da177e4 LT |
3232 | #ifdef CONFIG_NET_CLS_ACT |
3233 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3234 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3235 | * a compare and 2 stores extra right now if we dont have it on | |
3236 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3237 | * NOTE: This doesn't stop any functionality; if you dont have |
3238 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3239 | * |
3240 | */ | |
24824a09 | 3241 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3242 | { |
1da177e4 | 3243 | struct net_device *dev = skb->dev; |
f697c3e8 | 3244 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3245 | int result = TC_ACT_OK; |
3246 | struct Qdisc *q; | |
4ec93edb | 3247 | |
de384830 | 3248 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
e87cc472 JP |
3249 | net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n", |
3250 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3251 | return TC_ACT_SHOT; |
3252 | } | |
1da177e4 | 3253 | |
f697c3e8 HX |
3254 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3255 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3256 | |
83874000 | 3257 | q = rxq->qdisc; |
8d50b53d | 3258 | if (q != &noop_qdisc) { |
83874000 | 3259 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3260 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3261 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3262 | spin_unlock(qdisc_lock(q)); |
3263 | } | |
f697c3e8 HX |
3264 | |
3265 | return result; | |
3266 | } | |
86e65da9 | 3267 | |
f697c3e8 HX |
3268 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3269 | struct packet_type **pt_prev, | |
3270 | int *ret, struct net_device *orig_dev) | |
3271 | { | |
24824a09 ED |
3272 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3273 | ||
3274 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
f697c3e8 | 3275 | goto out; |
1da177e4 | 3276 | |
f697c3e8 HX |
3277 | if (*pt_prev) { |
3278 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3279 | *pt_prev = NULL; | |
1da177e4 LT |
3280 | } |
3281 | ||
24824a09 | 3282 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3283 | case TC_ACT_SHOT: |
3284 | case TC_ACT_STOLEN: | |
3285 | kfree_skb(skb); | |
3286 | return NULL; | |
3287 | } | |
3288 | ||
3289 | out: | |
3290 | skb->tc_verd = 0; | |
3291 | return skb; | |
1da177e4 LT |
3292 | } |
3293 | #endif | |
3294 | ||
ab95bfe0 JP |
3295 | /** |
3296 | * netdev_rx_handler_register - register receive handler | |
3297 | * @dev: device to register a handler for | |
3298 | * @rx_handler: receive handler to register | |
93e2c32b | 3299 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 JP |
3300 | * |
3301 | * Register a receive hander for a device. This handler will then be | |
3302 | * called from __netif_receive_skb. A negative errno code is returned | |
3303 | * on a failure. | |
3304 | * | |
3305 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3306 | * |
3307 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3308 | */ |
3309 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3310 | rx_handler_func_t *rx_handler, |
3311 | void *rx_handler_data) | |
ab95bfe0 JP |
3312 | { |
3313 | ASSERT_RTNL(); | |
3314 | ||
3315 | if (dev->rx_handler) | |
3316 | return -EBUSY; | |
3317 | ||
93e2c32b | 3318 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3319 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3320 | ||
3321 | return 0; | |
3322 | } | |
3323 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3324 | ||
3325 | /** | |
3326 | * netdev_rx_handler_unregister - unregister receive handler | |
3327 | * @dev: device to unregister a handler from | |
3328 | * | |
3329 | * Unregister a receive hander from a device. | |
3330 | * | |
3331 | * The caller must hold the rtnl_mutex. | |
3332 | */ | |
3333 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3334 | { | |
3335 | ||
3336 | ASSERT_RTNL(); | |
a9b3cd7f SH |
3337 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
3338 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); | |
ab95bfe0 JP |
3339 | } |
3340 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3341 | ||
b4b9e355 MG |
3342 | /* |
3343 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3344 | * the special handling of PFMEMALLOC skbs. | |
3345 | */ | |
3346 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3347 | { | |
3348 | switch (skb->protocol) { | |
3349 | case __constant_htons(ETH_P_ARP): | |
3350 | case __constant_htons(ETH_P_IP): | |
3351 | case __constant_htons(ETH_P_IPV6): | |
3352 | case __constant_htons(ETH_P_8021Q): | |
3353 | return true; | |
3354 | default: | |
3355 | return false; | |
3356 | } | |
3357 | } | |
3358 | ||
9754e293 | 3359 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
3360 | { |
3361 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3362 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3363 | struct net_device *orig_dev; |
63d8ea7f | 3364 | struct net_device *null_or_dev; |
8a4eb573 | 3365 | bool deliver_exact = false; |
1da177e4 | 3366 | int ret = NET_RX_DROP; |
252e3346 | 3367 | __be16 type; |
1da177e4 | 3368 | |
588f0330 | 3369 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3370 | |
cf66ba58 | 3371 | trace_netif_receive_skb(skb); |
9b22ea56 | 3372 | |
1da177e4 | 3373 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 3374 | if (netpoll_receive_skb(skb)) |
b4b9e355 | 3375 | goto out; |
1da177e4 | 3376 | |
cc9bd5ce | 3377 | orig_dev = skb->dev; |
8f903c70 | 3378 | |
c1d2bbe1 | 3379 | skb_reset_network_header(skb); |
fda55eca ED |
3380 | if (!skb_transport_header_was_set(skb)) |
3381 | skb_reset_transport_header(skb); | |
0b5c9db1 | 3382 | skb_reset_mac_len(skb); |
1da177e4 LT |
3383 | |
3384 | pt_prev = NULL; | |
3385 | ||
3386 | rcu_read_lock(); | |
3387 | ||
63d8ea7f | 3388 | another_round: |
b6858177 | 3389 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
3390 | |
3391 | __this_cpu_inc(softnet_data.processed); | |
3392 | ||
bcc6d479 JP |
3393 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) { |
3394 | skb = vlan_untag(skb); | |
3395 | if (unlikely(!skb)) | |
b4b9e355 | 3396 | goto unlock; |
bcc6d479 JP |
3397 | } |
3398 | ||
1da177e4 LT |
3399 | #ifdef CONFIG_NET_CLS_ACT |
3400 | if (skb->tc_verd & TC_NCLS) { | |
3401 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3402 | goto ncls; | |
3403 | } | |
3404 | #endif | |
3405 | ||
9754e293 | 3406 | if (pfmemalloc) |
b4b9e355 MG |
3407 | goto skip_taps; |
3408 | ||
1da177e4 | 3409 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
63d8ea7f | 3410 | if (!ptype->dev || ptype->dev == skb->dev) { |
4ec93edb | 3411 | if (pt_prev) |
f2ccd8fa | 3412 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3413 | pt_prev = ptype; |
3414 | } | |
3415 | } | |
3416 | ||
b4b9e355 | 3417 | skip_taps: |
1da177e4 | 3418 | #ifdef CONFIG_NET_CLS_ACT |
f697c3e8 HX |
3419 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3420 | if (!skb) | |
b4b9e355 | 3421 | goto unlock; |
1da177e4 LT |
3422 | ncls: |
3423 | #endif | |
3424 | ||
9754e293 | 3425 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
3426 | goto drop; |
3427 | ||
2425717b JF |
3428 | if (vlan_tx_tag_present(skb)) { |
3429 | if (pt_prev) { | |
3430 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3431 | pt_prev = NULL; | |
3432 | } | |
48cc32d3 | 3433 | if (vlan_do_receive(&skb)) |
2425717b JF |
3434 | goto another_round; |
3435 | else if (unlikely(!skb)) | |
b4b9e355 | 3436 | goto unlock; |
2425717b JF |
3437 | } |
3438 | ||
48cc32d3 | 3439 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
3440 | if (rx_handler) { |
3441 | if (pt_prev) { | |
3442 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3443 | pt_prev = NULL; | |
3444 | } | |
8a4eb573 JP |
3445 | switch (rx_handler(&skb)) { |
3446 | case RX_HANDLER_CONSUMED: | |
b4b9e355 | 3447 | goto unlock; |
8a4eb573 | 3448 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3449 | goto another_round; |
8a4eb573 JP |
3450 | case RX_HANDLER_EXACT: |
3451 | deliver_exact = true; | |
3452 | case RX_HANDLER_PASS: | |
3453 | break; | |
3454 | default: | |
3455 | BUG(); | |
3456 | } | |
ab95bfe0 | 3457 | } |
1da177e4 | 3458 | |
48cc32d3 FZ |
3459 | if (vlan_tx_nonzero_tag_present(skb)) |
3460 | skb->pkt_type = PACKET_OTHERHOST; | |
3461 | ||
63d8ea7f | 3462 | /* deliver only exact match when indicated */ |
8a4eb573 | 3463 | null_or_dev = deliver_exact ? skb->dev : NULL; |
1f3c8804 | 3464 | |
1da177e4 | 3465 | type = skb->protocol; |
82d8a867 PE |
3466 | list_for_each_entry_rcu(ptype, |
3467 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
63d8ea7f | 3468 | if (ptype->type == type && |
e3f48d37 JP |
3469 | (ptype->dev == null_or_dev || ptype->dev == skb->dev || |
3470 | ptype->dev == orig_dev)) { | |
4ec93edb | 3471 | if (pt_prev) |
f2ccd8fa | 3472 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3473 | pt_prev = ptype; |
3474 | } | |
3475 | } | |
3476 | ||
3477 | if (pt_prev) { | |
1080e512 | 3478 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 3479 | goto drop; |
1080e512 MT |
3480 | else |
3481 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 3482 | } else { |
b4b9e355 | 3483 | drop: |
caf586e5 | 3484 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3485 | kfree_skb(skb); |
3486 | /* Jamal, now you will not able to escape explaining | |
3487 | * me how you were going to use this. :-) | |
3488 | */ | |
3489 | ret = NET_RX_DROP; | |
3490 | } | |
3491 | ||
b4b9e355 | 3492 | unlock: |
1da177e4 | 3493 | rcu_read_unlock(); |
b4b9e355 | 3494 | out: |
9754e293 DM |
3495 | return ret; |
3496 | } | |
3497 | ||
3498 | static int __netif_receive_skb(struct sk_buff *skb) | |
3499 | { | |
3500 | int ret; | |
3501 | ||
3502 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
3503 | unsigned long pflags = current->flags; | |
3504 | ||
3505 | /* | |
3506 | * PFMEMALLOC skbs are special, they should | |
3507 | * - be delivered to SOCK_MEMALLOC sockets only | |
3508 | * - stay away from userspace | |
3509 | * - have bounded memory usage | |
3510 | * | |
3511 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
3512 | * context down to all allocation sites. | |
3513 | */ | |
3514 | current->flags |= PF_MEMALLOC; | |
3515 | ret = __netif_receive_skb_core(skb, true); | |
3516 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
3517 | } else | |
3518 | ret = __netif_receive_skb_core(skb, false); | |
3519 | ||
1da177e4 LT |
3520 | return ret; |
3521 | } | |
0a9627f2 TH |
3522 | |
3523 | /** | |
3524 | * netif_receive_skb - process receive buffer from network | |
3525 | * @skb: buffer to process | |
3526 | * | |
3527 | * netif_receive_skb() is the main receive data processing function. | |
3528 | * It always succeeds. The buffer may be dropped during processing | |
3529 | * for congestion control or by the protocol layers. | |
3530 | * | |
3531 | * This function may only be called from softirq context and interrupts | |
3532 | * should be enabled. | |
3533 | * | |
3534 | * Return values (usually ignored): | |
3535 | * NET_RX_SUCCESS: no congestion | |
3536 | * NET_RX_DROP: packet was dropped | |
3537 | */ | |
3538 | int netif_receive_skb(struct sk_buff *skb) | |
3539 | { | |
588f0330 | 3540 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3541 | |
c1f19b51 RC |
3542 | if (skb_defer_rx_timestamp(skb)) |
3543 | return NET_RX_SUCCESS; | |
3544 | ||
df334545 | 3545 | #ifdef CONFIG_RPS |
c5905afb | 3546 | if (static_key_false(&rps_needed)) { |
3b098e2d ED |
3547 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
3548 | int cpu, ret; | |
fec5e652 | 3549 | |
3b098e2d ED |
3550 | rcu_read_lock(); |
3551 | ||
3552 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3553 | |
3b098e2d ED |
3554 | if (cpu >= 0) { |
3555 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3556 | rcu_read_unlock(); | |
adc9300e | 3557 | return ret; |
3b098e2d | 3558 | } |
adc9300e | 3559 | rcu_read_unlock(); |
fec5e652 | 3560 | } |
1e94d72f | 3561 | #endif |
adc9300e | 3562 | return __netif_receive_skb(skb); |
0a9627f2 | 3563 | } |
d1b19dff | 3564 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3565 | |
88751275 ED |
3566 | /* Network device is going away, flush any packets still pending |
3567 | * Called with irqs disabled. | |
3568 | */ | |
152102c7 | 3569 | static void flush_backlog(void *arg) |
6e583ce5 | 3570 | { |
152102c7 | 3571 | struct net_device *dev = arg; |
e36fa2f7 | 3572 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3573 | struct sk_buff *skb, *tmp; |
3574 | ||
e36fa2f7 | 3575 | rps_lock(sd); |
6e7676c1 | 3576 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3577 | if (skb->dev == dev) { |
e36fa2f7 | 3578 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3579 | kfree_skb(skb); |
76cc8b13 | 3580 | input_queue_head_incr(sd); |
6e583ce5 | 3581 | } |
6e7676c1 | 3582 | } |
e36fa2f7 | 3583 | rps_unlock(sd); |
6e7676c1 CG |
3584 | |
3585 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3586 | if (skb->dev == dev) { | |
3587 | __skb_unlink(skb, &sd->process_queue); | |
3588 | kfree_skb(skb); | |
76cc8b13 | 3589 | input_queue_head_incr(sd); |
6e7676c1 CG |
3590 | } |
3591 | } | |
6e583ce5 SH |
3592 | } |
3593 | ||
d565b0a1 HX |
3594 | static int napi_gro_complete(struct sk_buff *skb) |
3595 | { | |
22061d80 | 3596 | struct packet_offload *ptype; |
d565b0a1 | 3597 | __be16 type = skb->protocol; |
22061d80 | 3598 | struct list_head *head = &offload_base; |
d565b0a1 HX |
3599 | int err = -ENOENT; |
3600 | ||
c3c7c254 ED |
3601 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
3602 | ||
fc59f9a3 HX |
3603 | if (NAPI_GRO_CB(skb)->count == 1) { |
3604 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3605 | goto out; |
fc59f9a3 | 3606 | } |
d565b0a1 HX |
3607 | |
3608 | rcu_read_lock(); | |
3609 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3610 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
3611 | continue; |
3612 | ||
f191a1d1 | 3613 | err = ptype->callbacks.gro_complete(skb); |
d565b0a1 HX |
3614 | break; |
3615 | } | |
3616 | rcu_read_unlock(); | |
3617 | ||
3618 | if (err) { | |
3619 | WARN_ON(&ptype->list == head); | |
3620 | kfree_skb(skb); | |
3621 | return NET_RX_SUCCESS; | |
3622 | } | |
3623 | ||
3624 | out: | |
d565b0a1 HX |
3625 | return netif_receive_skb(skb); |
3626 | } | |
3627 | ||
2e71a6f8 ED |
3628 | /* napi->gro_list contains packets ordered by age. |
3629 | * youngest packets at the head of it. | |
3630 | * Complete skbs in reverse order to reduce latencies. | |
3631 | */ | |
3632 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 3633 | { |
2e71a6f8 | 3634 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 3635 | |
2e71a6f8 ED |
3636 | /* scan list and build reverse chain */ |
3637 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
3638 | skb->prev = prev; | |
3639 | prev = skb; | |
3640 | } | |
3641 | ||
3642 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 3643 | skb->next = NULL; |
2e71a6f8 ED |
3644 | |
3645 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
3646 | return; | |
3647 | ||
3648 | prev = skb->prev; | |
d565b0a1 | 3649 | napi_gro_complete(skb); |
2e71a6f8 | 3650 | napi->gro_count--; |
d565b0a1 HX |
3651 | } |
3652 | ||
3653 | napi->gro_list = NULL; | |
3654 | } | |
86cac58b | 3655 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3656 | |
89c5fa33 ED |
3657 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
3658 | { | |
3659 | struct sk_buff *p; | |
3660 | unsigned int maclen = skb->dev->hard_header_len; | |
3661 | ||
3662 | for (p = napi->gro_list; p; p = p->next) { | |
3663 | unsigned long diffs; | |
3664 | ||
3665 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; | |
3666 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
3667 | if (maclen == ETH_HLEN) | |
3668 | diffs |= compare_ether_header(skb_mac_header(p), | |
3669 | skb_gro_mac_header(skb)); | |
3670 | else if (!diffs) | |
3671 | diffs = memcmp(skb_mac_header(p), | |
3672 | skb_gro_mac_header(skb), | |
3673 | maclen); | |
3674 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
3675 | NAPI_GRO_CB(p)->flush = 0; | |
3676 | } | |
3677 | } | |
3678 | ||
bb728820 | 3679 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3680 | { |
3681 | struct sk_buff **pp = NULL; | |
22061d80 | 3682 | struct packet_offload *ptype; |
d565b0a1 | 3683 | __be16 type = skb->protocol; |
22061d80 | 3684 | struct list_head *head = &offload_base; |
0da2afd5 | 3685 | int same_flow; |
5b252f0c | 3686 | enum gro_result ret; |
d565b0a1 | 3687 | |
ce9e76c8 | 3688 | if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb)) |
d565b0a1 HX |
3689 | goto normal; |
3690 | ||
21dc3301 | 3691 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) |
f17f5c91 HX |
3692 | goto normal; |
3693 | ||
89c5fa33 ED |
3694 | gro_list_prepare(napi, skb); |
3695 | ||
d565b0a1 HX |
3696 | rcu_read_lock(); |
3697 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3698 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
3699 | continue; |
3700 | ||
86911732 | 3701 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 3702 | skb_reset_mac_len(skb); |
d565b0a1 HX |
3703 | NAPI_GRO_CB(skb)->same_flow = 0; |
3704 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3705 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 3706 | |
f191a1d1 | 3707 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
3708 | break; |
3709 | } | |
3710 | rcu_read_unlock(); | |
3711 | ||
3712 | if (&ptype->list == head) | |
3713 | goto normal; | |
3714 | ||
0da2afd5 | 3715 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3716 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3717 | |
d565b0a1 HX |
3718 | if (pp) { |
3719 | struct sk_buff *nskb = *pp; | |
3720 | ||
3721 | *pp = nskb->next; | |
3722 | nskb->next = NULL; | |
3723 | napi_gro_complete(nskb); | |
4ae5544f | 3724 | napi->gro_count--; |
d565b0a1 HX |
3725 | } |
3726 | ||
0da2afd5 | 3727 | if (same_flow) |
d565b0a1 HX |
3728 | goto ok; |
3729 | ||
4ae5544f | 3730 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 3731 | goto normal; |
d565b0a1 | 3732 | |
4ae5544f | 3733 | napi->gro_count++; |
d565b0a1 | 3734 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 3735 | NAPI_GRO_CB(skb)->age = jiffies; |
86911732 | 3736 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3737 | skb->next = napi->gro_list; |
3738 | napi->gro_list = skb; | |
5d0d9be8 | 3739 | ret = GRO_HELD; |
d565b0a1 | 3740 | |
ad0f9904 | 3741 | pull: |
cb18978c HX |
3742 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
3743 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3744 | ||
3745 | BUG_ON(skb->end - skb->tail < grow); | |
3746 | ||
3747 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3748 | ||
3749 | skb->tail += grow; | |
3750 | skb->data_len -= grow; | |
3751 | ||
3752 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
9e903e08 | 3753 | skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow); |
cb18978c | 3754 | |
9e903e08 | 3755 | if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) { |
ea2ab693 | 3756 | skb_frag_unref(skb, 0); |
cb18978c HX |
3757 | memmove(skb_shinfo(skb)->frags, |
3758 | skb_shinfo(skb)->frags + 1, | |
e5093aec | 3759 | --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); |
cb18978c | 3760 | } |
ad0f9904 HX |
3761 | } |
3762 | ||
d565b0a1 | 3763 | ok: |
5d0d9be8 | 3764 | return ret; |
d565b0a1 HX |
3765 | |
3766 | normal: | |
ad0f9904 HX |
3767 | ret = GRO_NORMAL; |
3768 | goto pull; | |
5d38a079 | 3769 | } |
96e93eab | 3770 | |
5d38a079 | 3771 | |
bb728820 | 3772 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3773 | { |
5d0d9be8 HX |
3774 | switch (ret) { |
3775 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3776 | if (netif_receive_skb(skb)) |
3777 | ret = GRO_DROP; | |
3778 | break; | |
5d38a079 | 3779 | |
5d0d9be8 | 3780 | case GRO_DROP: |
5d38a079 HX |
3781 | kfree_skb(skb); |
3782 | break; | |
5b252f0c | 3783 | |
daa86548 | 3784 | case GRO_MERGED_FREE: |
d7e8883c ED |
3785 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
3786 | kmem_cache_free(skbuff_head_cache, skb); | |
3787 | else | |
3788 | __kfree_skb(skb); | |
daa86548 ED |
3789 | break; |
3790 | ||
5b252f0c BH |
3791 | case GRO_HELD: |
3792 | case GRO_MERGED: | |
3793 | break; | |
5d38a079 HX |
3794 | } |
3795 | ||
c7c4b3b6 | 3796 | return ret; |
5d0d9be8 | 3797 | } |
5d0d9be8 | 3798 | |
ca07e43e | 3799 | static void skb_gro_reset_offset(struct sk_buff *skb) |
78a478d0 | 3800 | { |
ca07e43e ED |
3801 | const struct skb_shared_info *pinfo = skb_shinfo(skb); |
3802 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
3803 | ||
78a478d0 HX |
3804 | NAPI_GRO_CB(skb)->data_offset = 0; |
3805 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 3806 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 3807 | |
78d3fd0b | 3808 | if (skb->mac_header == skb->tail && |
ca07e43e ED |
3809 | pinfo->nr_frags && |
3810 | !PageHighMem(skb_frag_page(frag0))) { | |
3811 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
3812 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
7489594c | 3813 | } |
78a478d0 | 3814 | } |
78a478d0 | 3815 | |
c7c4b3b6 | 3816 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 3817 | { |
86911732 HX |
3818 | skb_gro_reset_offset(skb); |
3819 | ||
89c5fa33 | 3820 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
3821 | } |
3822 | EXPORT_SYMBOL(napi_gro_receive); | |
3823 | ||
d0c2b0d2 | 3824 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 3825 | { |
96e93eab | 3826 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
3827 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
3828 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 3829 | skb->vlan_tci = 0; |
66c46d74 | 3830 | skb->dev = napi->dev; |
6d152e23 | 3831 | skb->skb_iif = 0; |
96e93eab HX |
3832 | |
3833 | napi->skb = skb; | |
3834 | } | |
96e93eab | 3835 | |
76620aaf | 3836 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 3837 | { |
5d38a079 | 3838 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
3839 | |
3840 | if (!skb) { | |
89d71a66 ED |
3841 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3842 | if (skb) | |
3843 | napi->skb = skb; | |
80595d59 | 3844 | } |
96e93eab HX |
3845 | return skb; |
3846 | } | |
76620aaf | 3847 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3848 | |
bb728820 | 3849 | static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
c7c4b3b6 | 3850 | gro_result_t ret) |
96e93eab | 3851 | { |
5d0d9be8 HX |
3852 | switch (ret) { |
3853 | case GRO_NORMAL: | |
86911732 | 3854 | case GRO_HELD: |
e76b69cc | 3855 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3856 | |
c7c4b3b6 BH |
3857 | if (ret == GRO_HELD) |
3858 | skb_gro_pull(skb, -ETH_HLEN); | |
3859 | else if (netif_receive_skb(skb)) | |
3860 | ret = GRO_DROP; | |
86911732 | 3861 | break; |
5d38a079 | 3862 | |
5d0d9be8 | 3863 | case GRO_DROP: |
5d0d9be8 HX |
3864 | case GRO_MERGED_FREE: |
3865 | napi_reuse_skb(napi, skb); | |
3866 | break; | |
5b252f0c BH |
3867 | |
3868 | case GRO_MERGED: | |
3869 | break; | |
5d0d9be8 | 3870 | } |
5d38a079 | 3871 | |
c7c4b3b6 | 3872 | return ret; |
5d38a079 | 3873 | } |
5d0d9be8 | 3874 | |
4adb9c4a | 3875 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
3876 | { |
3877 | struct sk_buff *skb = napi->skb; | |
3878 | struct ethhdr *eth; | |
a5b1cf28 HX |
3879 | unsigned int hlen; |
3880 | unsigned int off; | |
76620aaf HX |
3881 | |
3882 | napi->skb = NULL; | |
3883 | ||
3884 | skb_reset_mac_header(skb); | |
3885 | skb_gro_reset_offset(skb); | |
3886 | ||
a5b1cf28 HX |
3887 | off = skb_gro_offset(skb); |
3888 | hlen = off + sizeof(*eth); | |
3889 | eth = skb_gro_header_fast(skb, off); | |
3890 | if (skb_gro_header_hard(skb, hlen)) { | |
3891 | eth = skb_gro_header_slow(skb, hlen, off); | |
3892 | if (unlikely(!eth)) { | |
3893 | napi_reuse_skb(napi, skb); | |
3894 | skb = NULL; | |
3895 | goto out; | |
3896 | } | |
76620aaf HX |
3897 | } |
3898 | ||
3899 | skb_gro_pull(skb, sizeof(*eth)); | |
3900 | ||
3901 | /* | |
3902 | * This works because the only protocols we care about don't require | |
3903 | * special handling. We'll fix it up properly at the end. | |
3904 | */ | |
3905 | skb->protocol = eth->h_proto; | |
3906 | ||
3907 | out: | |
3908 | return skb; | |
3909 | } | |
76620aaf | 3910 | |
c7c4b3b6 | 3911 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 3912 | { |
76620aaf | 3913 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
3914 | |
3915 | if (!skb) | |
c7c4b3b6 | 3916 | return GRO_DROP; |
5d0d9be8 | 3917 | |
89c5fa33 | 3918 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 3919 | } |
5d38a079 HX |
3920 | EXPORT_SYMBOL(napi_gro_frags); |
3921 | ||
e326bed2 ED |
3922 | /* |
3923 | * net_rps_action sends any pending IPI's for rps. | |
3924 | * Note: called with local irq disabled, but exits with local irq enabled. | |
3925 | */ | |
3926 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
3927 | { | |
3928 | #ifdef CONFIG_RPS | |
3929 | struct softnet_data *remsd = sd->rps_ipi_list; | |
3930 | ||
3931 | if (remsd) { | |
3932 | sd->rps_ipi_list = NULL; | |
3933 | ||
3934 | local_irq_enable(); | |
3935 | ||
3936 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3937 | while (remsd) { | |
3938 | struct softnet_data *next = remsd->rps_ipi_next; | |
3939 | ||
3940 | if (cpu_online(remsd->cpu)) | |
3941 | __smp_call_function_single(remsd->cpu, | |
3942 | &remsd->csd, 0); | |
3943 | remsd = next; | |
3944 | } | |
3945 | } else | |
3946 | #endif | |
3947 | local_irq_enable(); | |
3948 | } | |
3949 | ||
bea3348e | 3950 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
3951 | { |
3952 | int work = 0; | |
eecfd7c4 | 3953 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 3954 | |
e326bed2 ED |
3955 | #ifdef CONFIG_RPS |
3956 | /* Check if we have pending ipi, its better to send them now, | |
3957 | * not waiting net_rx_action() end. | |
3958 | */ | |
3959 | if (sd->rps_ipi_list) { | |
3960 | local_irq_disable(); | |
3961 | net_rps_action_and_irq_enable(sd); | |
3962 | } | |
3963 | #endif | |
bea3348e | 3964 | napi->weight = weight_p; |
6e7676c1 CG |
3965 | local_irq_disable(); |
3966 | while (work < quota) { | |
1da177e4 | 3967 | struct sk_buff *skb; |
6e7676c1 CG |
3968 | unsigned int qlen; |
3969 | ||
3970 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
3971 | local_irq_enable(); | |
3972 | __netif_receive_skb(skb); | |
6e7676c1 | 3973 | local_irq_disable(); |
76cc8b13 TH |
3974 | input_queue_head_incr(sd); |
3975 | if (++work >= quota) { | |
3976 | local_irq_enable(); | |
3977 | return work; | |
3978 | } | |
6e7676c1 | 3979 | } |
1da177e4 | 3980 | |
e36fa2f7 | 3981 | rps_lock(sd); |
6e7676c1 | 3982 | qlen = skb_queue_len(&sd->input_pkt_queue); |
76cc8b13 | 3983 | if (qlen) |
6e7676c1 CG |
3984 | skb_queue_splice_tail_init(&sd->input_pkt_queue, |
3985 | &sd->process_queue); | |
76cc8b13 | 3986 | |
6e7676c1 | 3987 | if (qlen < quota - work) { |
eecfd7c4 ED |
3988 | /* |
3989 | * Inline a custom version of __napi_complete(). | |
3990 | * only current cpu owns and manipulates this napi, | |
3991 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
3992 | * we can use a plain write instead of clear_bit(), | |
3993 | * and we dont need an smp_mb() memory barrier. | |
3994 | */ | |
3995 | list_del(&napi->poll_list); | |
3996 | napi->state = 0; | |
3997 | ||
6e7676c1 | 3998 | quota = work + qlen; |
bea3348e | 3999 | } |
e36fa2f7 | 4000 | rps_unlock(sd); |
6e7676c1 CG |
4001 | } |
4002 | local_irq_enable(); | |
1da177e4 | 4003 | |
bea3348e SH |
4004 | return work; |
4005 | } | |
1da177e4 | 4006 | |
bea3348e SH |
4007 | /** |
4008 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4009 | * @n: entry to schedule |
bea3348e SH |
4010 | * |
4011 | * The entry's receive function will be scheduled to run | |
4012 | */ | |
b5606c2d | 4013 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4014 | { |
4015 | unsigned long flags; | |
1da177e4 | 4016 | |
bea3348e | 4017 | local_irq_save(flags); |
eecfd7c4 | 4018 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 4019 | local_irq_restore(flags); |
1da177e4 | 4020 | } |
bea3348e SH |
4021 | EXPORT_SYMBOL(__napi_schedule); |
4022 | ||
d565b0a1 HX |
4023 | void __napi_complete(struct napi_struct *n) |
4024 | { | |
4025 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
4026 | BUG_ON(n->gro_list); | |
4027 | ||
4028 | list_del(&n->poll_list); | |
4029 | smp_mb__before_clear_bit(); | |
4030 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
4031 | } | |
4032 | EXPORT_SYMBOL(__napi_complete); | |
4033 | ||
4034 | void napi_complete(struct napi_struct *n) | |
4035 | { | |
4036 | unsigned long flags; | |
4037 | ||
4038 | /* | |
4039 | * don't let napi dequeue from the cpu poll list | |
4040 | * just in case its running on a different cpu | |
4041 | */ | |
4042 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4043 | return; | |
4044 | ||
2e71a6f8 | 4045 | napi_gro_flush(n, false); |
d565b0a1 HX |
4046 | local_irq_save(flags); |
4047 | __napi_complete(n); | |
4048 | local_irq_restore(flags); | |
4049 | } | |
4050 | EXPORT_SYMBOL(napi_complete); | |
4051 | ||
4052 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
4053 | int (*poll)(struct napi_struct *, int), int weight) | |
4054 | { | |
4055 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 4056 | napi->gro_count = 0; |
d565b0a1 | 4057 | napi->gro_list = NULL; |
5d38a079 | 4058 | napi->skb = NULL; |
d565b0a1 HX |
4059 | napi->poll = poll; |
4060 | napi->weight = weight; | |
4061 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 4062 | napi->dev = dev; |
5d38a079 | 4063 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
4064 | spin_lock_init(&napi->poll_lock); |
4065 | napi->poll_owner = -1; | |
4066 | #endif | |
4067 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
4068 | } | |
4069 | EXPORT_SYMBOL(netif_napi_add); | |
4070 | ||
4071 | void netif_napi_del(struct napi_struct *napi) | |
4072 | { | |
4073 | struct sk_buff *skb, *next; | |
4074 | ||
d7b06636 | 4075 | list_del_init(&napi->dev_list); |
76620aaf | 4076 | napi_free_frags(napi); |
d565b0a1 HX |
4077 | |
4078 | for (skb = napi->gro_list; skb; skb = next) { | |
4079 | next = skb->next; | |
4080 | skb->next = NULL; | |
4081 | kfree_skb(skb); | |
4082 | } | |
4083 | ||
4084 | napi->gro_list = NULL; | |
4ae5544f | 4085 | napi->gro_count = 0; |
d565b0a1 HX |
4086 | } |
4087 | EXPORT_SYMBOL(netif_napi_del); | |
4088 | ||
1da177e4 LT |
4089 | static void net_rx_action(struct softirq_action *h) |
4090 | { | |
e326bed2 | 4091 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 4092 | unsigned long time_limit = jiffies + 2; |
51b0bded | 4093 | int budget = netdev_budget; |
53fb95d3 MM |
4094 | void *have; |
4095 | ||
1da177e4 LT |
4096 | local_irq_disable(); |
4097 | ||
e326bed2 | 4098 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
4099 | struct napi_struct *n; |
4100 | int work, weight; | |
1da177e4 | 4101 | |
bea3348e | 4102 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
4103 | * Allow this to run for 2 jiffies since which will allow |
4104 | * an average latency of 1.5/HZ. | |
bea3348e | 4105 | */ |
24f8b238 | 4106 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
4107 | goto softnet_break; |
4108 | ||
4109 | local_irq_enable(); | |
4110 | ||
bea3348e SH |
4111 | /* Even though interrupts have been re-enabled, this |
4112 | * access is safe because interrupts can only add new | |
4113 | * entries to the tail of this list, and only ->poll() | |
4114 | * calls can remove this head entry from the list. | |
4115 | */ | |
e326bed2 | 4116 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 4117 | |
bea3348e SH |
4118 | have = netpoll_poll_lock(n); |
4119 | ||
4120 | weight = n->weight; | |
4121 | ||
0a7606c1 DM |
4122 | /* This NAPI_STATE_SCHED test is for avoiding a race |
4123 | * with netpoll's poll_napi(). Only the entity which | |
4124 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
4125 | * actually make the ->poll() call. Therefore we avoid | |
25985edc | 4126 | * accidentally calling ->poll() when NAPI is not scheduled. |
0a7606c1 DM |
4127 | */ |
4128 | work = 0; | |
4ea7e386 | 4129 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 4130 | work = n->poll(n, weight); |
4ea7e386 NH |
4131 | trace_napi_poll(n); |
4132 | } | |
bea3348e SH |
4133 | |
4134 | WARN_ON_ONCE(work > weight); | |
4135 | ||
4136 | budget -= work; | |
4137 | ||
4138 | local_irq_disable(); | |
4139 | ||
4140 | /* Drivers must not modify the NAPI state if they | |
4141 | * consume the entire weight. In such cases this code | |
4142 | * still "owns" the NAPI instance and therefore can | |
4143 | * move the instance around on the list at-will. | |
4144 | */ | |
fed17f30 | 4145 | if (unlikely(work == weight)) { |
ff780cd8 HX |
4146 | if (unlikely(napi_disable_pending(n))) { |
4147 | local_irq_enable(); | |
4148 | napi_complete(n); | |
4149 | local_irq_disable(); | |
2e71a6f8 ED |
4150 | } else { |
4151 | if (n->gro_list) { | |
4152 | /* flush too old packets | |
4153 | * If HZ < 1000, flush all packets. | |
4154 | */ | |
4155 | local_irq_enable(); | |
4156 | napi_gro_flush(n, HZ >= 1000); | |
4157 | local_irq_disable(); | |
4158 | } | |
e326bed2 | 4159 | list_move_tail(&n->poll_list, &sd->poll_list); |
2e71a6f8 | 4160 | } |
fed17f30 | 4161 | } |
bea3348e SH |
4162 | |
4163 | netpoll_poll_unlock(have); | |
1da177e4 LT |
4164 | } |
4165 | out: | |
e326bed2 | 4166 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 4167 | |
db217334 CL |
4168 | #ifdef CONFIG_NET_DMA |
4169 | /* | |
4170 | * There may not be any more sk_buffs coming right now, so push | |
4171 | * any pending DMA copies to hardware | |
4172 | */ | |
2ba05622 | 4173 | dma_issue_pending_all(); |
db217334 | 4174 | #endif |
bea3348e | 4175 | |
1da177e4 LT |
4176 | return; |
4177 | ||
4178 | softnet_break: | |
dee42870 | 4179 | sd->time_squeeze++; |
1da177e4 LT |
4180 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
4181 | goto out; | |
4182 | } | |
4183 | ||
9ff162a8 JP |
4184 | struct netdev_upper { |
4185 | struct net_device *dev; | |
4186 | bool master; | |
4187 | struct list_head list; | |
4188 | struct rcu_head rcu; | |
4189 | struct list_head search_list; | |
4190 | }; | |
4191 | ||
4192 | static void __append_search_uppers(struct list_head *search_list, | |
4193 | struct net_device *dev) | |
4194 | { | |
4195 | struct netdev_upper *upper; | |
4196 | ||
4197 | list_for_each_entry(upper, &dev->upper_dev_list, list) { | |
4198 | /* check if this upper is not already in search list */ | |
4199 | if (list_empty(&upper->search_list)) | |
4200 | list_add_tail(&upper->search_list, search_list); | |
4201 | } | |
4202 | } | |
4203 | ||
4204 | static bool __netdev_search_upper_dev(struct net_device *dev, | |
4205 | struct net_device *upper_dev) | |
4206 | { | |
4207 | LIST_HEAD(search_list); | |
4208 | struct netdev_upper *upper; | |
4209 | struct netdev_upper *tmp; | |
4210 | bool ret = false; | |
4211 | ||
4212 | __append_search_uppers(&search_list, dev); | |
4213 | list_for_each_entry(upper, &search_list, search_list) { | |
4214 | if (upper->dev == upper_dev) { | |
4215 | ret = true; | |
4216 | break; | |
4217 | } | |
4218 | __append_search_uppers(&search_list, upper->dev); | |
4219 | } | |
4220 | list_for_each_entry_safe(upper, tmp, &search_list, search_list) | |
4221 | INIT_LIST_HEAD(&upper->search_list); | |
4222 | return ret; | |
4223 | } | |
4224 | ||
4225 | static struct netdev_upper *__netdev_find_upper(struct net_device *dev, | |
4226 | struct net_device *upper_dev) | |
4227 | { | |
4228 | struct netdev_upper *upper; | |
4229 | ||
4230 | list_for_each_entry(upper, &dev->upper_dev_list, list) { | |
4231 | if (upper->dev == upper_dev) | |
4232 | return upper; | |
4233 | } | |
4234 | return NULL; | |
4235 | } | |
4236 | ||
4237 | /** | |
4238 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
4239 | * @dev: device | |
4240 | * @upper_dev: upper device to check | |
4241 | * | |
4242 | * Find out if a device is linked to specified upper device and return true | |
4243 | * in case it is. Note that this checks only immediate upper device, | |
4244 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
4245 | */ | |
4246 | bool netdev_has_upper_dev(struct net_device *dev, | |
4247 | struct net_device *upper_dev) | |
4248 | { | |
4249 | ASSERT_RTNL(); | |
4250 | ||
4251 | return __netdev_find_upper(dev, upper_dev); | |
4252 | } | |
4253 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
4254 | ||
4255 | /** | |
4256 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
4257 | * @dev: device | |
4258 | * | |
4259 | * Find out if a device is linked to an upper device and return true in case | |
4260 | * it is. The caller must hold the RTNL lock. | |
4261 | */ | |
4262 | bool netdev_has_any_upper_dev(struct net_device *dev) | |
4263 | { | |
4264 | ASSERT_RTNL(); | |
4265 | ||
4266 | return !list_empty(&dev->upper_dev_list); | |
4267 | } | |
4268 | EXPORT_SYMBOL(netdev_has_any_upper_dev); | |
4269 | ||
4270 | /** | |
4271 | * netdev_master_upper_dev_get - Get master upper device | |
4272 | * @dev: device | |
4273 | * | |
4274 | * Find a master upper device and return pointer to it or NULL in case | |
4275 | * it's not there. The caller must hold the RTNL lock. | |
4276 | */ | |
4277 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
4278 | { | |
4279 | struct netdev_upper *upper; | |
4280 | ||
4281 | ASSERT_RTNL(); | |
4282 | ||
4283 | if (list_empty(&dev->upper_dev_list)) | |
4284 | return NULL; | |
4285 | ||
4286 | upper = list_first_entry(&dev->upper_dev_list, | |
4287 | struct netdev_upper, list); | |
4288 | if (likely(upper->master)) | |
4289 | return upper->dev; | |
4290 | return NULL; | |
4291 | } | |
4292 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
4293 | ||
4294 | /** | |
4295 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
4296 | * @dev: device | |
4297 | * | |
4298 | * Find a master upper device and return pointer to it or NULL in case | |
4299 | * it's not there. The caller must hold the RCU read lock. | |
4300 | */ | |
4301 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
4302 | { | |
4303 | struct netdev_upper *upper; | |
4304 | ||
4305 | upper = list_first_or_null_rcu(&dev->upper_dev_list, | |
4306 | struct netdev_upper, list); | |
4307 | if (upper && likely(upper->master)) | |
4308 | return upper->dev; | |
4309 | return NULL; | |
4310 | } | |
4311 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
4312 | ||
4313 | static int __netdev_upper_dev_link(struct net_device *dev, | |
4314 | struct net_device *upper_dev, bool master) | |
4315 | { | |
4316 | struct netdev_upper *upper; | |
4317 | ||
4318 | ASSERT_RTNL(); | |
4319 | ||
4320 | if (dev == upper_dev) | |
4321 | return -EBUSY; | |
4322 | ||
4323 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
4324 | if (__netdev_search_upper_dev(upper_dev, dev)) | |
4325 | return -EBUSY; | |
4326 | ||
4327 | if (__netdev_find_upper(dev, upper_dev)) | |
4328 | return -EEXIST; | |
4329 | ||
4330 | if (master && netdev_master_upper_dev_get(dev)) | |
4331 | return -EBUSY; | |
4332 | ||
4333 | upper = kmalloc(sizeof(*upper), GFP_KERNEL); | |
4334 | if (!upper) | |
4335 | return -ENOMEM; | |
4336 | ||
4337 | upper->dev = upper_dev; | |
4338 | upper->master = master; | |
4339 | INIT_LIST_HEAD(&upper->search_list); | |
4340 | ||
4341 | /* Ensure that master upper link is always the first item in list. */ | |
4342 | if (master) | |
4343 | list_add_rcu(&upper->list, &dev->upper_dev_list); | |
4344 | else | |
4345 | list_add_tail_rcu(&upper->list, &dev->upper_dev_list); | |
4346 | dev_hold(upper_dev); | |
4347 | ||
4348 | return 0; | |
4349 | } | |
4350 | ||
4351 | /** | |
4352 | * netdev_upper_dev_link - Add a link to the upper device | |
4353 | * @dev: device | |
4354 | * @upper_dev: new upper device | |
4355 | * | |
4356 | * Adds a link to device which is upper to this one. The caller must hold | |
4357 | * the RTNL lock. On a failure a negative errno code is returned. | |
4358 | * On success the reference counts are adjusted and the function | |
4359 | * returns zero. | |
4360 | */ | |
4361 | int netdev_upper_dev_link(struct net_device *dev, | |
4362 | struct net_device *upper_dev) | |
4363 | { | |
4364 | return __netdev_upper_dev_link(dev, upper_dev, false); | |
4365 | } | |
4366 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
4367 | ||
4368 | /** | |
4369 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
4370 | * @dev: device | |
4371 | * @upper_dev: new upper device | |
4372 | * | |
4373 | * Adds a link to device which is upper to this one. In this case, only | |
4374 | * one master upper device can be linked, although other non-master devices | |
4375 | * might be linked as well. The caller must hold the RTNL lock. | |
4376 | * On a failure a negative errno code is returned. On success the reference | |
4377 | * counts are adjusted and the function returns zero. | |
4378 | */ | |
4379 | int netdev_master_upper_dev_link(struct net_device *dev, | |
4380 | struct net_device *upper_dev) | |
4381 | { | |
4382 | return __netdev_upper_dev_link(dev, upper_dev, true); | |
4383 | } | |
4384 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
4385 | ||
4386 | /** | |
4387 | * netdev_upper_dev_unlink - Removes a link to upper device | |
4388 | * @dev: device | |
4389 | * @upper_dev: new upper device | |
4390 | * | |
4391 | * Removes a link to device which is upper to this one. The caller must hold | |
4392 | * the RTNL lock. | |
4393 | */ | |
4394 | void netdev_upper_dev_unlink(struct net_device *dev, | |
4395 | struct net_device *upper_dev) | |
4396 | { | |
4397 | struct netdev_upper *upper; | |
4398 | ||
4399 | ASSERT_RTNL(); | |
4400 | ||
4401 | upper = __netdev_find_upper(dev, upper_dev); | |
4402 | if (!upper) | |
4403 | return; | |
4404 | list_del_rcu(&upper->list); | |
4405 | dev_put(upper_dev); | |
4406 | kfree_rcu(upper, rcu); | |
4407 | } | |
4408 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
4409 | ||
b6c40d68 PM |
4410 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
4411 | { | |
d314774c SH |
4412 | const struct net_device_ops *ops = dev->netdev_ops; |
4413 | ||
4414 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
4415 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
4416 | } |
4417 | ||
dad9b335 | 4418 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 | 4419 | { |
b536db93 | 4420 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
4421 | kuid_t uid; |
4422 | kgid_t gid; | |
1da177e4 | 4423 | |
24023451 PM |
4424 | ASSERT_RTNL(); |
4425 | ||
dad9b335 WC |
4426 | dev->flags |= IFF_PROMISC; |
4427 | dev->promiscuity += inc; | |
4428 | if (dev->promiscuity == 0) { | |
4429 | /* | |
4430 | * Avoid overflow. | |
4431 | * If inc causes overflow, untouch promisc and return error. | |
4432 | */ | |
4433 | if (inc < 0) | |
4434 | dev->flags &= ~IFF_PROMISC; | |
4435 | else { | |
4436 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
4437 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
4438 | dev->name); | |
dad9b335 WC |
4439 | return -EOVERFLOW; |
4440 | } | |
4441 | } | |
52609c0b | 4442 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
4443 | pr_info("device %s %s promiscuous mode\n", |
4444 | dev->name, | |
4445 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
4446 | if (audit_enabled) { |
4447 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
4448 | audit_log(current->audit_context, GFP_ATOMIC, |
4449 | AUDIT_ANOM_PROMISCUOUS, | |
4450 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
4451 | dev->name, (dev->flags & IFF_PROMISC), | |
4452 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 4453 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
4454 | from_kuid(&init_user_ns, uid), |
4455 | from_kgid(&init_user_ns, gid), | |
7759db82 | 4456 | audit_get_sessionid(current)); |
8192b0c4 | 4457 | } |
24023451 | 4458 | |
b6c40d68 | 4459 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 4460 | } |
dad9b335 | 4461 | return 0; |
1da177e4 LT |
4462 | } |
4463 | ||
4417da66 PM |
4464 | /** |
4465 | * dev_set_promiscuity - update promiscuity count on a device | |
4466 | * @dev: device | |
4467 | * @inc: modifier | |
4468 | * | |
4469 | * Add or remove promiscuity from a device. While the count in the device | |
4470 | * remains above zero the interface remains promiscuous. Once it hits zero | |
4471 | * the device reverts back to normal filtering operation. A negative inc | |
4472 | * value is used to drop promiscuity on the device. | |
dad9b335 | 4473 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 4474 | */ |
dad9b335 | 4475 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 4476 | { |
b536db93 | 4477 | unsigned int old_flags = dev->flags; |
dad9b335 | 4478 | int err; |
4417da66 | 4479 | |
dad9b335 | 4480 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 4481 | if (err < 0) |
dad9b335 | 4482 | return err; |
4417da66 PM |
4483 | if (dev->flags != old_flags) |
4484 | dev_set_rx_mode(dev); | |
dad9b335 | 4485 | return err; |
4417da66 | 4486 | } |
d1b19dff | 4487 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 4488 | |
1da177e4 LT |
4489 | /** |
4490 | * dev_set_allmulti - update allmulti count on a device | |
4491 | * @dev: device | |
4492 | * @inc: modifier | |
4493 | * | |
4494 | * Add or remove reception of all multicast frames to a device. While the | |
4495 | * count in the device remains above zero the interface remains listening | |
4496 | * to all interfaces. Once it hits zero the device reverts back to normal | |
4497 | * filtering operation. A negative @inc value is used to drop the counter | |
4498 | * when releasing a resource needing all multicasts. | |
dad9b335 | 4499 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
4500 | */ |
4501 | ||
dad9b335 | 4502 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 | 4503 | { |
b536db93 | 4504 | unsigned int old_flags = dev->flags; |
1da177e4 | 4505 | |
24023451 PM |
4506 | ASSERT_RTNL(); |
4507 | ||
1da177e4 | 4508 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
4509 | dev->allmulti += inc; |
4510 | if (dev->allmulti == 0) { | |
4511 | /* | |
4512 | * Avoid overflow. | |
4513 | * If inc causes overflow, untouch allmulti and return error. | |
4514 | */ | |
4515 | if (inc < 0) | |
4516 | dev->flags &= ~IFF_ALLMULTI; | |
4517 | else { | |
4518 | dev->allmulti -= inc; | |
7b6cd1ce JP |
4519 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
4520 | dev->name); | |
dad9b335 WC |
4521 | return -EOVERFLOW; |
4522 | } | |
4523 | } | |
24023451 | 4524 | if (dev->flags ^ old_flags) { |
b6c40d68 | 4525 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 4526 | dev_set_rx_mode(dev); |
24023451 | 4527 | } |
dad9b335 | 4528 | return 0; |
4417da66 | 4529 | } |
d1b19dff | 4530 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
4531 | |
4532 | /* | |
4533 | * Upload unicast and multicast address lists to device and | |
4534 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 4535 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
4536 | * are present. |
4537 | */ | |
4538 | void __dev_set_rx_mode(struct net_device *dev) | |
4539 | { | |
d314774c SH |
4540 | const struct net_device_ops *ops = dev->netdev_ops; |
4541 | ||
4417da66 PM |
4542 | /* dev_open will call this function so the list will stay sane. */ |
4543 | if (!(dev->flags&IFF_UP)) | |
4544 | return; | |
4545 | ||
4546 | if (!netif_device_present(dev)) | |
40b77c94 | 4547 | return; |
4417da66 | 4548 | |
01789349 | 4549 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
4550 | /* Unicast addresses changes may only happen under the rtnl, |
4551 | * therefore calling __dev_set_promiscuity here is safe. | |
4552 | */ | |
32e7bfc4 | 4553 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 | 4554 | __dev_set_promiscuity(dev, 1); |
2d348d1f | 4555 | dev->uc_promisc = true; |
32e7bfc4 | 4556 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 | 4557 | __dev_set_promiscuity(dev, -1); |
2d348d1f | 4558 | dev->uc_promisc = false; |
4417da66 | 4559 | } |
4417da66 | 4560 | } |
01789349 JP |
4561 | |
4562 | if (ops->ndo_set_rx_mode) | |
4563 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
4564 | } |
4565 | ||
4566 | void dev_set_rx_mode(struct net_device *dev) | |
4567 | { | |
b9e40857 | 4568 | netif_addr_lock_bh(dev); |
4417da66 | 4569 | __dev_set_rx_mode(dev); |
b9e40857 | 4570 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
4571 | } |
4572 | ||
f0db275a SH |
4573 | /** |
4574 | * dev_get_flags - get flags reported to userspace | |
4575 | * @dev: device | |
4576 | * | |
4577 | * Get the combination of flag bits exported through APIs to userspace. | |
4578 | */ | |
95c96174 | 4579 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 4580 | { |
95c96174 | 4581 | unsigned int flags; |
1da177e4 LT |
4582 | |
4583 | flags = (dev->flags & ~(IFF_PROMISC | | |
4584 | IFF_ALLMULTI | | |
b00055aa SR |
4585 | IFF_RUNNING | |
4586 | IFF_LOWER_UP | | |
4587 | IFF_DORMANT)) | | |
1da177e4 LT |
4588 | (dev->gflags & (IFF_PROMISC | |
4589 | IFF_ALLMULTI)); | |
4590 | ||
b00055aa SR |
4591 | if (netif_running(dev)) { |
4592 | if (netif_oper_up(dev)) | |
4593 | flags |= IFF_RUNNING; | |
4594 | if (netif_carrier_ok(dev)) | |
4595 | flags |= IFF_LOWER_UP; | |
4596 | if (netif_dormant(dev)) | |
4597 | flags |= IFF_DORMANT; | |
4598 | } | |
1da177e4 LT |
4599 | |
4600 | return flags; | |
4601 | } | |
d1b19dff | 4602 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4603 | |
bd380811 | 4604 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 4605 | { |
b536db93 | 4606 | unsigned int old_flags = dev->flags; |
bd380811 | 4607 | int ret; |
1da177e4 | 4608 | |
24023451 PM |
4609 | ASSERT_RTNL(); |
4610 | ||
1da177e4 LT |
4611 | /* |
4612 | * Set the flags on our device. | |
4613 | */ | |
4614 | ||
4615 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4616 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4617 | IFF_AUTOMEDIA)) | | |
4618 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4619 | IFF_ALLMULTI)); | |
4620 | ||
4621 | /* | |
4622 | * Load in the correct multicast list now the flags have changed. | |
4623 | */ | |
4624 | ||
b6c40d68 PM |
4625 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4626 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4627 | |
4417da66 | 4628 | dev_set_rx_mode(dev); |
1da177e4 LT |
4629 | |
4630 | /* | |
4631 | * Have we downed the interface. We handle IFF_UP ourselves | |
4632 | * according to user attempts to set it, rather than blindly | |
4633 | * setting it. | |
4634 | */ | |
4635 | ||
4636 | ret = 0; | |
4637 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 4638 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
4639 | |
4640 | if (!ret) | |
4417da66 | 4641 | dev_set_rx_mode(dev); |
1da177e4 LT |
4642 | } |
4643 | ||
1da177e4 | 4644 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
4645 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4646 | ||
1da177e4 LT |
4647 | dev->gflags ^= IFF_PROMISC; |
4648 | dev_set_promiscuity(dev, inc); | |
4649 | } | |
4650 | ||
4651 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4652 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4653 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4654 | */ | |
4655 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4656 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4657 | ||
1da177e4 LT |
4658 | dev->gflags ^= IFF_ALLMULTI; |
4659 | dev_set_allmulti(dev, inc); | |
4660 | } | |
4661 | ||
bd380811 PM |
4662 | return ret; |
4663 | } | |
4664 | ||
4665 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4666 | { | |
4667 | unsigned int changes = dev->flags ^ old_flags; | |
4668 | ||
4669 | if (changes & IFF_UP) { | |
4670 | if (dev->flags & IFF_UP) | |
4671 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4672 | else | |
4673 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4674 | } | |
4675 | ||
4676 | if (dev->flags & IFF_UP && | |
4677 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4678 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4679 | } | |
4680 | ||
4681 | /** | |
4682 | * dev_change_flags - change device settings | |
4683 | * @dev: device | |
4684 | * @flags: device state flags | |
4685 | * | |
4686 | * Change settings on device based state flags. The flags are | |
4687 | * in the userspace exported format. | |
4688 | */ | |
b536db93 | 4689 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 4690 | { |
b536db93 ED |
4691 | int ret; |
4692 | unsigned int changes, old_flags = dev->flags; | |
bd380811 PM |
4693 | |
4694 | ret = __dev_change_flags(dev, flags); | |
4695 | if (ret < 0) | |
4696 | return ret; | |
4697 | ||
4698 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
4699 | if (changes) |
4700 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 4701 | |
bd380811 | 4702 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
4703 | return ret; |
4704 | } | |
d1b19dff | 4705 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4706 | |
f0db275a SH |
4707 | /** |
4708 | * dev_set_mtu - Change maximum transfer unit | |
4709 | * @dev: device | |
4710 | * @new_mtu: new transfer unit | |
4711 | * | |
4712 | * Change the maximum transfer size of the network device. | |
4713 | */ | |
1da177e4 LT |
4714 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4715 | { | |
d314774c | 4716 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4717 | int err; |
4718 | ||
4719 | if (new_mtu == dev->mtu) | |
4720 | return 0; | |
4721 | ||
4722 | /* MTU must be positive. */ | |
4723 | if (new_mtu < 0) | |
4724 | return -EINVAL; | |
4725 | ||
4726 | if (!netif_device_present(dev)) | |
4727 | return -ENODEV; | |
4728 | ||
4729 | err = 0; | |
d314774c SH |
4730 | if (ops->ndo_change_mtu) |
4731 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4732 | else |
4733 | dev->mtu = new_mtu; | |
d314774c | 4734 | |
e3d8fabe | 4735 | if (!err) |
056925ab | 4736 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4737 | return err; |
4738 | } | |
d1b19dff | 4739 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4740 | |
cbda10fa VD |
4741 | /** |
4742 | * dev_set_group - Change group this device belongs to | |
4743 | * @dev: device | |
4744 | * @new_group: group this device should belong to | |
4745 | */ | |
4746 | void dev_set_group(struct net_device *dev, int new_group) | |
4747 | { | |
4748 | dev->group = new_group; | |
4749 | } | |
4750 | EXPORT_SYMBOL(dev_set_group); | |
4751 | ||
f0db275a SH |
4752 | /** |
4753 | * dev_set_mac_address - Change Media Access Control Address | |
4754 | * @dev: device | |
4755 | * @sa: new address | |
4756 | * | |
4757 | * Change the hardware (MAC) address of the device | |
4758 | */ | |
1da177e4 LT |
4759 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4760 | { | |
d314774c | 4761 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4762 | int err; |
4763 | ||
d314774c | 4764 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4765 | return -EOPNOTSUPP; |
4766 | if (sa->sa_family != dev->type) | |
4767 | return -EINVAL; | |
4768 | if (!netif_device_present(dev)) | |
4769 | return -ENODEV; | |
d314774c | 4770 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
4771 | if (err) |
4772 | return err; | |
fbdeca2d | 4773 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 4774 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 4775 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 4776 | return 0; |
1da177e4 | 4777 | } |
d1b19dff | 4778 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 4779 | |
4bf84c35 JP |
4780 | /** |
4781 | * dev_change_carrier - Change device carrier | |
4782 | * @dev: device | |
4783 | * @new_carries: new value | |
4784 | * | |
4785 | * Change device carrier | |
4786 | */ | |
4787 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
4788 | { | |
4789 | const struct net_device_ops *ops = dev->netdev_ops; | |
4790 | ||
4791 | if (!ops->ndo_change_carrier) | |
4792 | return -EOPNOTSUPP; | |
4793 | if (!netif_device_present(dev)) | |
4794 | return -ENODEV; | |
4795 | return ops->ndo_change_carrier(dev, new_carrier); | |
4796 | } | |
4797 | EXPORT_SYMBOL(dev_change_carrier); | |
4798 | ||
1da177e4 LT |
4799 | /** |
4800 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 4801 | * @net: the applicable net namespace |
1da177e4 LT |
4802 | * |
4803 | * Returns a suitable unique value for a new device interface | |
4804 | * number. The caller must hold the rtnl semaphore or the | |
4805 | * dev_base_lock to be sure it remains unique. | |
4806 | */ | |
881d966b | 4807 | static int dev_new_index(struct net *net) |
1da177e4 | 4808 | { |
aa79e66e | 4809 | int ifindex = net->ifindex; |
1da177e4 LT |
4810 | for (;;) { |
4811 | if (++ifindex <= 0) | |
4812 | ifindex = 1; | |
881d966b | 4813 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 4814 | return net->ifindex = ifindex; |
1da177e4 LT |
4815 | } |
4816 | } | |
4817 | ||
1da177e4 | 4818 | /* Delayed registration/unregisteration */ |
3b5b34fd | 4819 | static LIST_HEAD(net_todo_list); |
1da177e4 | 4820 | |
6f05f629 | 4821 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 4822 | { |
1da177e4 | 4823 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
4824 | } |
4825 | ||
9b5e383c | 4826 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 4827 | { |
e93737b0 | 4828 | struct net_device *dev, *tmp; |
9b5e383c | 4829 | |
93ee31f1 DL |
4830 | BUG_ON(dev_boot_phase); |
4831 | ASSERT_RTNL(); | |
4832 | ||
e93737b0 | 4833 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 4834 | /* Some devices call without registering |
e93737b0 KK |
4835 | * for initialization unwind. Remove those |
4836 | * devices and proceed with the remaining. | |
9b5e383c ED |
4837 | */ |
4838 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
4839 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
4840 | dev->name, dev); | |
93ee31f1 | 4841 | |
9b5e383c | 4842 | WARN_ON(1); |
e93737b0 KK |
4843 | list_del(&dev->unreg_list); |
4844 | continue; | |
9b5e383c | 4845 | } |
449f4544 | 4846 | dev->dismantle = true; |
9b5e383c | 4847 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 4848 | } |
93ee31f1 | 4849 | |
44345724 OP |
4850 | /* If device is running, close it first. */ |
4851 | dev_close_many(head); | |
93ee31f1 | 4852 | |
44345724 | 4853 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
4854 | /* And unlink it from device chain. */ |
4855 | unlist_netdevice(dev); | |
93ee31f1 | 4856 | |
9b5e383c ED |
4857 | dev->reg_state = NETREG_UNREGISTERING; |
4858 | } | |
93ee31f1 DL |
4859 | |
4860 | synchronize_net(); | |
4861 | ||
9b5e383c ED |
4862 | list_for_each_entry(dev, head, unreg_list) { |
4863 | /* Shutdown queueing discipline. */ | |
4864 | dev_shutdown(dev); | |
93ee31f1 DL |
4865 | |
4866 | ||
9b5e383c ED |
4867 | /* Notify protocols, that we are about to destroy |
4868 | this device. They should clean all the things. | |
4869 | */ | |
4870 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 4871 | |
a2835763 PM |
4872 | if (!dev->rtnl_link_ops || |
4873 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
4874 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
4875 | ||
9b5e383c ED |
4876 | /* |
4877 | * Flush the unicast and multicast chains | |
4878 | */ | |
a748ee24 | 4879 | dev_uc_flush(dev); |
22bedad3 | 4880 | dev_mc_flush(dev); |
93ee31f1 | 4881 | |
9b5e383c ED |
4882 | if (dev->netdev_ops->ndo_uninit) |
4883 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 4884 | |
9ff162a8 JP |
4885 | /* Notifier chain MUST detach us all upper devices. */ |
4886 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 4887 | |
9b5e383c ED |
4888 | /* Remove entries from kobject tree */ |
4889 | netdev_unregister_kobject(dev); | |
024e9679 AD |
4890 | #ifdef CONFIG_XPS |
4891 | /* Remove XPS queueing entries */ | |
4892 | netif_reset_xps_queues_gt(dev, 0); | |
4893 | #endif | |
9b5e383c | 4894 | } |
93ee31f1 | 4895 | |
850a545b | 4896 | synchronize_net(); |
395264d5 | 4897 | |
a5ee1551 | 4898 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
4899 | dev_put(dev); |
4900 | } | |
4901 | ||
4902 | static void rollback_registered(struct net_device *dev) | |
4903 | { | |
4904 | LIST_HEAD(single); | |
4905 | ||
4906 | list_add(&dev->unreg_list, &single); | |
4907 | rollback_registered_many(&single); | |
ceaaec98 | 4908 | list_del(&single); |
93ee31f1 DL |
4909 | } |
4910 | ||
c8f44aff MM |
4911 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
4912 | netdev_features_t features) | |
b63365a2 | 4913 | { |
57422dc5 MM |
4914 | /* Fix illegal checksum combinations */ |
4915 | if ((features & NETIF_F_HW_CSUM) && | |
4916 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 4917 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
4918 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
4919 | } | |
4920 | ||
b63365a2 HX |
4921 | /* Fix illegal SG+CSUM combinations. */ |
4922 | if ((features & NETIF_F_SG) && | |
4923 | !(features & NETIF_F_ALL_CSUM)) { | |
6f404e44 MM |
4924 | netdev_dbg(dev, |
4925 | "Dropping NETIF_F_SG since no checksum feature.\n"); | |
b63365a2 HX |
4926 | features &= ~NETIF_F_SG; |
4927 | } | |
4928 | ||
4929 | /* TSO requires that SG is present as well. */ | |
ea2d3688 | 4930 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 4931 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 4932 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
4933 | } |
4934 | ||
31d8b9e0 BH |
4935 | /* TSO ECN requires that TSO is present as well. */ |
4936 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
4937 | features &= ~NETIF_F_TSO_ECN; | |
4938 | ||
212b573f MM |
4939 | /* Software GSO depends on SG. */ |
4940 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 4941 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
4942 | features &= ~NETIF_F_GSO; |
4943 | } | |
4944 | ||
acd1130e | 4945 | /* UFO needs SG and checksumming */ |
b63365a2 | 4946 | if (features & NETIF_F_UFO) { |
79032644 MM |
4947 | /* maybe split UFO into V4 and V6? */ |
4948 | if (!((features & NETIF_F_GEN_CSUM) || | |
4949 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
4950 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 4951 | netdev_dbg(dev, |
acd1130e | 4952 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
4953 | features &= ~NETIF_F_UFO; |
4954 | } | |
4955 | ||
4956 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 4957 | netdev_dbg(dev, |
acd1130e | 4958 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
4959 | features &= ~NETIF_F_UFO; |
4960 | } | |
4961 | } | |
4962 | ||
4963 | return features; | |
4964 | } | |
b63365a2 | 4965 | |
6cb6a27c | 4966 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 4967 | { |
c8f44aff | 4968 | netdev_features_t features; |
5455c699 MM |
4969 | int err = 0; |
4970 | ||
87267485 MM |
4971 | ASSERT_RTNL(); |
4972 | ||
5455c699 MM |
4973 | features = netdev_get_wanted_features(dev); |
4974 | ||
4975 | if (dev->netdev_ops->ndo_fix_features) | |
4976 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
4977 | ||
4978 | /* driver might be less strict about feature dependencies */ | |
4979 | features = netdev_fix_features(dev, features); | |
4980 | ||
4981 | if (dev->features == features) | |
6cb6a27c | 4982 | return 0; |
5455c699 | 4983 | |
c8f44aff MM |
4984 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
4985 | &dev->features, &features); | |
5455c699 MM |
4986 | |
4987 | if (dev->netdev_ops->ndo_set_features) | |
4988 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
4989 | ||
6cb6a27c | 4990 | if (unlikely(err < 0)) { |
5455c699 | 4991 | netdev_err(dev, |
c8f44aff MM |
4992 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
4993 | err, &features, &dev->features); | |
6cb6a27c MM |
4994 | return -1; |
4995 | } | |
4996 | ||
4997 | if (!err) | |
4998 | dev->features = features; | |
4999 | ||
5000 | return 1; | |
5001 | } | |
5002 | ||
afe12cc8 MM |
5003 | /** |
5004 | * netdev_update_features - recalculate device features | |
5005 | * @dev: the device to check | |
5006 | * | |
5007 | * Recalculate dev->features set and send notifications if it | |
5008 | * has changed. Should be called after driver or hardware dependent | |
5009 | * conditions might have changed that influence the features. | |
5010 | */ | |
6cb6a27c MM |
5011 | void netdev_update_features(struct net_device *dev) |
5012 | { | |
5013 | if (__netdev_update_features(dev)) | |
5014 | netdev_features_change(dev); | |
5455c699 MM |
5015 | } |
5016 | EXPORT_SYMBOL(netdev_update_features); | |
5017 | ||
afe12cc8 MM |
5018 | /** |
5019 | * netdev_change_features - recalculate device features | |
5020 | * @dev: the device to check | |
5021 | * | |
5022 | * Recalculate dev->features set and send notifications even | |
5023 | * if they have not changed. Should be called instead of | |
5024 | * netdev_update_features() if also dev->vlan_features might | |
5025 | * have changed to allow the changes to be propagated to stacked | |
5026 | * VLAN devices. | |
5027 | */ | |
5028 | void netdev_change_features(struct net_device *dev) | |
5029 | { | |
5030 | __netdev_update_features(dev); | |
5031 | netdev_features_change(dev); | |
5032 | } | |
5033 | EXPORT_SYMBOL(netdev_change_features); | |
5034 | ||
fc4a7489 PM |
5035 | /** |
5036 | * netif_stacked_transfer_operstate - transfer operstate | |
5037 | * @rootdev: the root or lower level device to transfer state from | |
5038 | * @dev: the device to transfer operstate to | |
5039 | * | |
5040 | * Transfer operational state from root to device. This is normally | |
5041 | * called when a stacking relationship exists between the root | |
5042 | * device and the device(a leaf device). | |
5043 | */ | |
5044 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5045 | struct net_device *dev) | |
5046 | { | |
5047 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5048 | netif_dormant_on(dev); | |
5049 | else | |
5050 | netif_dormant_off(dev); | |
5051 | ||
5052 | if (netif_carrier_ok(rootdev)) { | |
5053 | if (!netif_carrier_ok(dev)) | |
5054 | netif_carrier_on(dev); | |
5055 | } else { | |
5056 | if (netif_carrier_ok(dev)) | |
5057 | netif_carrier_off(dev); | |
5058 | } | |
5059 | } | |
5060 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5061 | ||
bf264145 | 5062 | #ifdef CONFIG_RPS |
1b4bf461 ED |
5063 | static int netif_alloc_rx_queues(struct net_device *dev) |
5064 | { | |
1b4bf461 | 5065 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 5066 | struct netdev_rx_queue *rx; |
1b4bf461 | 5067 | |
bd25fa7b | 5068 | BUG_ON(count < 1); |
1b4bf461 | 5069 | |
bd25fa7b | 5070 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
62b5942a | 5071 | if (!rx) |
bd25fa7b | 5072 | return -ENOMEM; |
62b5942a | 5073 | |
bd25fa7b TH |
5074 | dev->_rx = rx; |
5075 | ||
bd25fa7b | 5076 | for (i = 0; i < count; i++) |
fe822240 | 5077 | rx[i].dev = dev; |
1b4bf461 ED |
5078 | return 0; |
5079 | } | |
bf264145 | 5080 | #endif |
1b4bf461 | 5081 | |
aa942104 CG |
5082 | static void netdev_init_one_queue(struct net_device *dev, |
5083 | struct netdev_queue *queue, void *_unused) | |
5084 | { | |
5085 | /* Initialize queue lock */ | |
5086 | spin_lock_init(&queue->_xmit_lock); | |
5087 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
5088 | queue->xmit_lock_owner = -1; | |
b236da69 | 5089 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 5090 | queue->dev = dev; |
114cf580 TH |
5091 | #ifdef CONFIG_BQL |
5092 | dql_init(&queue->dql, HZ); | |
5093 | #endif | |
aa942104 CG |
5094 | } |
5095 | ||
e6484930 TH |
5096 | static int netif_alloc_netdev_queues(struct net_device *dev) |
5097 | { | |
5098 | unsigned int count = dev->num_tx_queues; | |
5099 | struct netdev_queue *tx; | |
5100 | ||
5101 | BUG_ON(count < 1); | |
5102 | ||
5103 | tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL); | |
62b5942a | 5104 | if (!tx) |
e6484930 | 5105 | return -ENOMEM; |
62b5942a | 5106 | |
e6484930 | 5107 | dev->_tx = tx; |
1d24eb48 | 5108 | |
e6484930 TH |
5109 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
5110 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
5111 | |
5112 | return 0; | |
e6484930 TH |
5113 | } |
5114 | ||
1da177e4 LT |
5115 | /** |
5116 | * register_netdevice - register a network device | |
5117 | * @dev: device to register | |
5118 | * | |
5119 | * Take a completed network device structure and add it to the kernel | |
5120 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5121 | * chain. 0 is returned on success. A negative errno code is returned | |
5122 | * on a failure to set up the device, or if the name is a duplicate. | |
5123 | * | |
5124 | * Callers must hold the rtnl semaphore. You may want | |
5125 | * register_netdev() instead of this. | |
5126 | * | |
5127 | * BUGS: | |
5128 | * The locking appears insufficient to guarantee two parallel registers | |
5129 | * will not get the same name. | |
5130 | */ | |
5131 | ||
5132 | int register_netdevice(struct net_device *dev) | |
5133 | { | |
1da177e4 | 5134 | int ret; |
d314774c | 5135 | struct net *net = dev_net(dev); |
1da177e4 LT |
5136 | |
5137 | BUG_ON(dev_boot_phase); | |
5138 | ASSERT_RTNL(); | |
5139 | ||
b17a7c17 SH |
5140 | might_sleep(); |
5141 | ||
1da177e4 LT |
5142 | /* When net_device's are persistent, this will be fatal. */ |
5143 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 5144 | BUG_ON(!net); |
1da177e4 | 5145 | |
f1f28aa3 | 5146 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 5147 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 5148 | |
1da177e4 LT |
5149 | dev->iflink = -1; |
5150 | ||
828de4f6 | 5151 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
5152 | if (ret < 0) |
5153 | goto out; | |
5154 | ||
1da177e4 | 5155 | /* Init, if this function is available */ |
d314774c SH |
5156 | if (dev->netdev_ops->ndo_init) { |
5157 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
5158 | if (ret) { |
5159 | if (ret > 0) | |
5160 | ret = -EIO; | |
90833aa4 | 5161 | goto out; |
1da177e4 LT |
5162 | } |
5163 | } | |
4ec93edb | 5164 | |
d2ed273d MM |
5165 | if (((dev->hw_features | dev->features) & NETIF_F_HW_VLAN_FILTER) && |
5166 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || | |
5167 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
5168 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
5169 | ret = -EINVAL; | |
5170 | goto err_uninit; | |
5171 | } | |
5172 | ||
9c7dafbf PE |
5173 | ret = -EBUSY; |
5174 | if (!dev->ifindex) | |
5175 | dev->ifindex = dev_new_index(net); | |
5176 | else if (__dev_get_by_index(net, dev->ifindex)) | |
5177 | goto err_uninit; | |
5178 | ||
1da177e4 LT |
5179 | if (dev->iflink == -1) |
5180 | dev->iflink = dev->ifindex; | |
5181 | ||
5455c699 MM |
5182 | /* Transfer changeable features to wanted_features and enable |
5183 | * software offloads (GSO and GRO). | |
5184 | */ | |
5185 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
5186 | dev->features |= NETIF_F_SOFT_FEATURES; |
5187 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 5188 | |
c6e1a0d1 | 5189 | /* Turn on no cache copy if HW is doing checksum */ |
34324dc2 MM |
5190 | if (!(dev->flags & IFF_LOOPBACK)) { |
5191 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
5192 | if (dev->features & NETIF_F_ALL_CSUM) { | |
5193 | dev->wanted_features |= NETIF_F_NOCACHE_COPY; | |
5194 | dev->features |= NETIF_F_NOCACHE_COPY; | |
5195 | } | |
c6e1a0d1 TH |
5196 | } |
5197 | ||
1180e7d6 | 5198 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 5199 | */ |
1180e7d6 | 5200 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 5201 | |
7ffbe3fd JB |
5202 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
5203 | ret = notifier_to_errno(ret); | |
5204 | if (ret) | |
5205 | goto err_uninit; | |
5206 | ||
8b41d188 | 5207 | ret = netdev_register_kobject(dev); |
b17a7c17 | 5208 | if (ret) |
7ce1b0ed | 5209 | goto err_uninit; |
b17a7c17 SH |
5210 | dev->reg_state = NETREG_REGISTERED; |
5211 | ||
6cb6a27c | 5212 | __netdev_update_features(dev); |
8e9b59b2 | 5213 | |
1da177e4 LT |
5214 | /* |
5215 | * Default initial state at registry is that the | |
5216 | * device is present. | |
5217 | */ | |
5218 | ||
5219 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5220 | ||
8f4cccbb BH |
5221 | linkwatch_init_dev(dev); |
5222 | ||
1da177e4 | 5223 | dev_init_scheduler(dev); |
1da177e4 | 5224 | dev_hold(dev); |
ce286d32 | 5225 | list_netdevice(dev); |
7bf23575 | 5226 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 5227 | |
948b337e JP |
5228 | /* If the device has permanent device address, driver should |
5229 | * set dev_addr and also addr_assign_type should be set to | |
5230 | * NET_ADDR_PERM (default value). | |
5231 | */ | |
5232 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
5233 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
5234 | ||
1da177e4 | 5235 | /* Notify protocols, that a new device appeared. */ |
056925ab | 5236 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 5237 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
5238 | if (ret) { |
5239 | rollback_registered(dev); | |
5240 | dev->reg_state = NETREG_UNREGISTERED; | |
5241 | } | |
d90a909e EB |
5242 | /* |
5243 | * Prevent userspace races by waiting until the network | |
5244 | * device is fully setup before sending notifications. | |
5245 | */ | |
a2835763 PM |
5246 | if (!dev->rtnl_link_ops || |
5247 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5248 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
5249 | |
5250 | out: | |
5251 | return ret; | |
7ce1b0ed HX |
5252 | |
5253 | err_uninit: | |
d314774c SH |
5254 | if (dev->netdev_ops->ndo_uninit) |
5255 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 5256 | goto out; |
1da177e4 | 5257 | } |
d1b19dff | 5258 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 5259 | |
937f1ba5 BH |
5260 | /** |
5261 | * init_dummy_netdev - init a dummy network device for NAPI | |
5262 | * @dev: device to init | |
5263 | * | |
5264 | * This takes a network device structure and initialize the minimum | |
5265 | * amount of fields so it can be used to schedule NAPI polls without | |
5266 | * registering a full blown interface. This is to be used by drivers | |
5267 | * that need to tie several hardware interfaces to a single NAPI | |
5268 | * poll scheduler due to HW limitations. | |
5269 | */ | |
5270 | int init_dummy_netdev(struct net_device *dev) | |
5271 | { | |
5272 | /* Clear everything. Note we don't initialize spinlocks | |
5273 | * are they aren't supposed to be taken by any of the | |
5274 | * NAPI code and this dummy netdev is supposed to be | |
5275 | * only ever used for NAPI polls | |
5276 | */ | |
5277 | memset(dev, 0, sizeof(struct net_device)); | |
5278 | ||
5279 | /* make sure we BUG if trying to hit standard | |
5280 | * register/unregister code path | |
5281 | */ | |
5282 | dev->reg_state = NETREG_DUMMY; | |
5283 | ||
937f1ba5 BH |
5284 | /* NAPI wants this */ |
5285 | INIT_LIST_HEAD(&dev->napi_list); | |
5286 | ||
5287 | /* a dummy interface is started by default */ | |
5288 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5289 | set_bit(__LINK_STATE_START, &dev->state); | |
5290 | ||
29b4433d ED |
5291 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
5292 | * because users of this 'device' dont need to change | |
5293 | * its refcount. | |
5294 | */ | |
5295 | ||
937f1ba5 BH |
5296 | return 0; |
5297 | } | |
5298 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5299 | ||
5300 | ||
1da177e4 LT |
5301 | /** |
5302 | * register_netdev - register a network device | |
5303 | * @dev: device to register | |
5304 | * | |
5305 | * Take a completed network device structure and add it to the kernel | |
5306 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5307 | * chain. 0 is returned on success. A negative errno code is returned | |
5308 | * on a failure to set up the device, or if the name is a duplicate. | |
5309 | * | |
38b4da38 | 5310 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5311 | * and expands the device name if you passed a format string to |
5312 | * alloc_netdev. | |
5313 | */ | |
5314 | int register_netdev(struct net_device *dev) | |
5315 | { | |
5316 | int err; | |
5317 | ||
5318 | rtnl_lock(); | |
1da177e4 | 5319 | err = register_netdevice(dev); |
1da177e4 LT |
5320 | rtnl_unlock(); |
5321 | return err; | |
5322 | } | |
5323 | EXPORT_SYMBOL(register_netdev); | |
5324 | ||
29b4433d ED |
5325 | int netdev_refcnt_read(const struct net_device *dev) |
5326 | { | |
5327 | int i, refcnt = 0; | |
5328 | ||
5329 | for_each_possible_cpu(i) | |
5330 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
5331 | return refcnt; | |
5332 | } | |
5333 | EXPORT_SYMBOL(netdev_refcnt_read); | |
5334 | ||
2c53040f | 5335 | /** |
1da177e4 | 5336 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 5337 | * @dev: target net_device |
1da177e4 LT |
5338 | * |
5339 | * This is called when unregistering network devices. | |
5340 | * | |
5341 | * Any protocol or device that holds a reference should register | |
5342 | * for netdevice notification, and cleanup and put back the | |
5343 | * reference if they receive an UNREGISTER event. | |
5344 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5345 | * call dev_put. |
1da177e4 LT |
5346 | */ |
5347 | static void netdev_wait_allrefs(struct net_device *dev) | |
5348 | { | |
5349 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 5350 | int refcnt; |
1da177e4 | 5351 | |
e014debe ED |
5352 | linkwatch_forget_dev(dev); |
5353 | ||
1da177e4 | 5354 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
5355 | refcnt = netdev_refcnt_read(dev); |
5356 | ||
5357 | while (refcnt != 0) { | |
1da177e4 | 5358 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 5359 | rtnl_lock(); |
1da177e4 LT |
5360 | |
5361 | /* Rebroadcast unregister notification */ | |
056925ab | 5362 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 5363 | |
748e2d93 | 5364 | __rtnl_unlock(); |
0115e8e3 | 5365 | rcu_barrier(); |
748e2d93 ED |
5366 | rtnl_lock(); |
5367 | ||
0115e8e3 | 5368 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
5369 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
5370 | &dev->state)) { | |
5371 | /* We must not have linkwatch events | |
5372 | * pending on unregister. If this | |
5373 | * happens, we simply run the queue | |
5374 | * unscheduled, resulting in a noop | |
5375 | * for this device. | |
5376 | */ | |
5377 | linkwatch_run_queue(); | |
5378 | } | |
5379 | ||
6756ae4b | 5380 | __rtnl_unlock(); |
1da177e4 LT |
5381 | |
5382 | rebroadcast_time = jiffies; | |
5383 | } | |
5384 | ||
5385 | msleep(250); | |
5386 | ||
29b4433d ED |
5387 | refcnt = netdev_refcnt_read(dev); |
5388 | ||
1da177e4 | 5389 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
5390 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
5391 | dev->name, refcnt); | |
1da177e4 LT |
5392 | warning_time = jiffies; |
5393 | } | |
5394 | } | |
5395 | } | |
5396 | ||
5397 | /* The sequence is: | |
5398 | * | |
5399 | * rtnl_lock(); | |
5400 | * ... | |
5401 | * register_netdevice(x1); | |
5402 | * register_netdevice(x2); | |
5403 | * ... | |
5404 | * unregister_netdevice(y1); | |
5405 | * unregister_netdevice(y2); | |
5406 | * ... | |
5407 | * rtnl_unlock(); | |
5408 | * free_netdev(y1); | |
5409 | * free_netdev(y2); | |
5410 | * | |
58ec3b4d | 5411 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5412 | * This allows us to deal with problems: |
b17a7c17 | 5413 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5414 | * without deadlocking with linkwatch via keventd. |
5415 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5416 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5417 | * |
5418 | * We must not return until all unregister events added during | |
5419 | * the interval the lock was held have been completed. | |
1da177e4 | 5420 | */ |
1da177e4 LT |
5421 | void netdev_run_todo(void) |
5422 | { | |
626ab0e6 | 5423 | struct list_head list; |
1da177e4 | 5424 | |
1da177e4 | 5425 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5426 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5427 | |
5428 | __rtnl_unlock(); | |
626ab0e6 | 5429 | |
0115e8e3 ED |
5430 | |
5431 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
5432 | if (!list_empty(&list)) |
5433 | rcu_barrier(); | |
5434 | ||
1da177e4 LT |
5435 | while (!list_empty(&list)) { |
5436 | struct net_device *dev | |
e5e26d75 | 5437 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
5438 | list_del(&dev->todo_list); |
5439 | ||
748e2d93 | 5440 | rtnl_lock(); |
0115e8e3 | 5441 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 5442 | __rtnl_unlock(); |
0115e8e3 | 5443 | |
b17a7c17 | 5444 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 5445 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
5446 | dev->name, dev->reg_state); |
5447 | dump_stack(); | |
5448 | continue; | |
5449 | } | |
1da177e4 | 5450 | |
b17a7c17 | 5451 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5452 | |
152102c7 | 5453 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 5454 | |
b17a7c17 | 5455 | netdev_wait_allrefs(dev); |
1da177e4 | 5456 | |
b17a7c17 | 5457 | /* paranoia */ |
29b4433d | 5458 | BUG_ON(netdev_refcnt_read(dev)); |
33d480ce ED |
5459 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
5460 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 5461 | WARN_ON(dev->dn_ptr); |
1da177e4 | 5462 | |
b17a7c17 SH |
5463 | if (dev->destructor) |
5464 | dev->destructor(dev); | |
9093bbb2 SH |
5465 | |
5466 | /* Free network device */ | |
5467 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5468 | } |
1da177e4 LT |
5469 | } |
5470 | ||
3cfde79c BH |
5471 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
5472 | * fields in the same order, with only the type differing. | |
5473 | */ | |
77a1abf5 ED |
5474 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
5475 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
5476 | { |
5477 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
5478 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
5479 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
5480 | #else |
5481 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
5482 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
5483 | u64 *dst = (u64 *)stats64; | |
5484 | ||
5485 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
5486 | sizeof(*stats64) / sizeof(u64)); | |
5487 | for (i = 0; i < n; i++) | |
5488 | dst[i] = src[i]; | |
5489 | #endif | |
5490 | } | |
77a1abf5 | 5491 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 5492 | |
eeda3fd6 SH |
5493 | /** |
5494 | * dev_get_stats - get network device statistics | |
5495 | * @dev: device to get statistics from | |
28172739 | 5496 | * @storage: place to store stats |
eeda3fd6 | 5497 | * |
d7753516 BH |
5498 | * Get network statistics from device. Return @storage. |
5499 | * The device driver may provide its own method by setting | |
5500 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
5501 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 5502 | */ |
d7753516 BH |
5503 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
5504 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 5505 | { |
eeda3fd6 SH |
5506 | const struct net_device_ops *ops = dev->netdev_ops; |
5507 | ||
28172739 ED |
5508 | if (ops->ndo_get_stats64) { |
5509 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
5510 | ops->ndo_get_stats64(dev, storage); |
5511 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 5512 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
5513 | } else { |
5514 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 5515 | } |
caf586e5 | 5516 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
28172739 | 5517 | return storage; |
c45d286e | 5518 | } |
eeda3fd6 | 5519 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5520 | |
24824a09 | 5521 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 5522 | { |
24824a09 | 5523 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 5524 | |
24824a09 ED |
5525 | #ifdef CONFIG_NET_CLS_ACT |
5526 | if (queue) | |
5527 | return queue; | |
5528 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
5529 | if (!queue) | |
5530 | return NULL; | |
5531 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
5532 | queue->qdisc = &noop_qdisc; |
5533 | queue->qdisc_sleeping = &noop_qdisc; | |
5534 | rcu_assign_pointer(dev->ingress_queue, queue); | |
5535 | #endif | |
5536 | return queue; | |
bb949fbd DM |
5537 | } |
5538 | ||
2c60db03 ED |
5539 | static const struct ethtool_ops default_ethtool_ops; |
5540 | ||
d07d7507 SG |
5541 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
5542 | const struct ethtool_ops *ops) | |
5543 | { | |
5544 | if (dev->ethtool_ops == &default_ethtool_ops) | |
5545 | dev->ethtool_ops = ops; | |
5546 | } | |
5547 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
5548 | ||
1da177e4 | 5549 | /** |
36909ea4 | 5550 | * alloc_netdev_mqs - allocate network device |
1da177e4 LT |
5551 | * @sizeof_priv: size of private data to allocate space for |
5552 | * @name: device name format string | |
5553 | * @setup: callback to initialize device | |
36909ea4 TH |
5554 | * @txqs: the number of TX subqueues to allocate |
5555 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
5556 | * |
5557 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 | 5558 | * and performs basic initialization. Also allocates subquue structs |
36909ea4 | 5559 | * for each queue on the device. |
1da177e4 | 5560 | */ |
36909ea4 TH |
5561 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
5562 | void (*setup)(struct net_device *), | |
5563 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 5564 | { |
1da177e4 | 5565 | struct net_device *dev; |
7943986c | 5566 | size_t alloc_size; |
1ce8e7b5 | 5567 | struct net_device *p; |
1da177e4 | 5568 | |
b6fe17d6 SH |
5569 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
5570 | ||
36909ea4 | 5571 | if (txqs < 1) { |
7b6cd1ce | 5572 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
5573 | return NULL; |
5574 | } | |
5575 | ||
36909ea4 TH |
5576 | #ifdef CONFIG_RPS |
5577 | if (rxqs < 1) { | |
7b6cd1ce | 5578 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
5579 | return NULL; |
5580 | } | |
5581 | #endif | |
5582 | ||
fd2ea0a7 | 5583 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
5584 | if (sizeof_priv) { |
5585 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 5586 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
5587 | alloc_size += sizeof_priv; |
5588 | } | |
5589 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 5590 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 5591 | |
31380de9 | 5592 | p = kzalloc(alloc_size, GFP_KERNEL); |
62b5942a | 5593 | if (!p) |
1da177e4 | 5594 | return NULL; |
1da177e4 | 5595 | |
1ce8e7b5 | 5596 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 5597 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 5598 | |
29b4433d ED |
5599 | dev->pcpu_refcnt = alloc_percpu(int); |
5600 | if (!dev->pcpu_refcnt) | |
e6484930 | 5601 | goto free_p; |
ab9c73cc | 5602 | |
ab9c73cc | 5603 | if (dev_addr_init(dev)) |
29b4433d | 5604 | goto free_pcpu; |
ab9c73cc | 5605 | |
22bedad3 | 5606 | dev_mc_init(dev); |
a748ee24 | 5607 | dev_uc_init(dev); |
ccffad25 | 5608 | |
c346dca1 | 5609 | dev_net_set(dev, &init_net); |
1da177e4 | 5610 | |
8d3bdbd5 | 5611 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 5612 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 5613 | |
8d3bdbd5 DM |
5614 | INIT_LIST_HEAD(&dev->napi_list); |
5615 | INIT_LIST_HEAD(&dev->unreg_list); | |
5616 | INIT_LIST_HEAD(&dev->link_watch_list); | |
9ff162a8 | 5617 | INIT_LIST_HEAD(&dev->upper_dev_list); |
8d3bdbd5 DM |
5618 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
5619 | setup(dev); | |
5620 | ||
36909ea4 TH |
5621 | dev->num_tx_queues = txqs; |
5622 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 5623 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 5624 | goto free_all; |
e8a0464c | 5625 | |
df334545 | 5626 | #ifdef CONFIG_RPS |
36909ea4 TH |
5627 | dev->num_rx_queues = rxqs; |
5628 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 5629 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 5630 | goto free_all; |
df334545 | 5631 | #endif |
0a9627f2 | 5632 | |
1da177e4 | 5633 | strcpy(dev->name, name); |
cbda10fa | 5634 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
5635 | if (!dev->ethtool_ops) |
5636 | dev->ethtool_ops = &default_ethtool_ops; | |
1da177e4 | 5637 | return dev; |
ab9c73cc | 5638 | |
8d3bdbd5 DM |
5639 | free_all: |
5640 | free_netdev(dev); | |
5641 | return NULL; | |
5642 | ||
29b4433d ED |
5643 | free_pcpu: |
5644 | free_percpu(dev->pcpu_refcnt); | |
ed9af2e8 | 5645 | kfree(dev->_tx); |
fe822240 TH |
5646 | #ifdef CONFIG_RPS |
5647 | kfree(dev->_rx); | |
5648 | #endif | |
5649 | ||
ab9c73cc JP |
5650 | free_p: |
5651 | kfree(p); | |
5652 | return NULL; | |
1da177e4 | 5653 | } |
36909ea4 | 5654 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
5655 | |
5656 | /** | |
5657 | * free_netdev - free network device | |
5658 | * @dev: device | |
5659 | * | |
4ec93edb YH |
5660 | * This function does the last stage of destroying an allocated device |
5661 | * interface. The reference to the device object is released. | |
1da177e4 LT |
5662 | * If this is the last reference then it will be freed. |
5663 | */ | |
5664 | void free_netdev(struct net_device *dev) | |
5665 | { | |
d565b0a1 HX |
5666 | struct napi_struct *p, *n; |
5667 | ||
f3005d7f DL |
5668 | release_net(dev_net(dev)); |
5669 | ||
e8a0464c | 5670 | kfree(dev->_tx); |
fe822240 TH |
5671 | #ifdef CONFIG_RPS |
5672 | kfree(dev->_rx); | |
5673 | #endif | |
e8a0464c | 5674 | |
33d480ce | 5675 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 5676 | |
f001fde5 JP |
5677 | /* Flush device addresses */ |
5678 | dev_addr_flush(dev); | |
5679 | ||
d565b0a1 HX |
5680 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
5681 | netif_napi_del(p); | |
5682 | ||
29b4433d ED |
5683 | free_percpu(dev->pcpu_refcnt); |
5684 | dev->pcpu_refcnt = NULL; | |
5685 | ||
3041a069 | 5686 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
5687 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
5688 | kfree((char *)dev - dev->padded); | |
5689 | return; | |
5690 | } | |
5691 | ||
5692 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
5693 | dev->reg_state = NETREG_RELEASED; | |
5694 | ||
43cb76d9 GKH |
5695 | /* will free via device release */ |
5696 | put_device(&dev->dev); | |
1da177e4 | 5697 | } |
d1b19dff | 5698 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 5699 | |
f0db275a SH |
5700 | /** |
5701 | * synchronize_net - Synchronize with packet receive processing | |
5702 | * | |
5703 | * Wait for packets currently being received to be done. | |
5704 | * Does not block later packets from starting. | |
5705 | */ | |
4ec93edb | 5706 | void synchronize_net(void) |
1da177e4 LT |
5707 | { |
5708 | might_sleep(); | |
be3fc413 ED |
5709 | if (rtnl_is_locked()) |
5710 | synchronize_rcu_expedited(); | |
5711 | else | |
5712 | synchronize_rcu(); | |
1da177e4 | 5713 | } |
d1b19dff | 5714 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
5715 | |
5716 | /** | |
44a0873d | 5717 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 5718 | * @dev: device |
44a0873d | 5719 | * @head: list |
6ebfbc06 | 5720 | * |
1da177e4 | 5721 | * This function shuts down a device interface and removes it |
d59b54b1 | 5722 | * from the kernel tables. |
44a0873d | 5723 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
5724 | * |
5725 | * Callers must hold the rtnl semaphore. You may want | |
5726 | * unregister_netdev() instead of this. | |
5727 | */ | |
5728 | ||
44a0873d | 5729 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 5730 | { |
a6620712 HX |
5731 | ASSERT_RTNL(); |
5732 | ||
44a0873d | 5733 | if (head) { |
9fdce099 | 5734 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
5735 | } else { |
5736 | rollback_registered(dev); | |
5737 | /* Finish processing unregister after unlock */ | |
5738 | net_set_todo(dev); | |
5739 | } | |
1da177e4 | 5740 | } |
44a0873d | 5741 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 5742 | |
9b5e383c ED |
5743 | /** |
5744 | * unregister_netdevice_many - unregister many devices | |
5745 | * @head: list of devices | |
9b5e383c ED |
5746 | */ |
5747 | void unregister_netdevice_many(struct list_head *head) | |
5748 | { | |
5749 | struct net_device *dev; | |
5750 | ||
5751 | if (!list_empty(head)) { | |
5752 | rollback_registered_many(head); | |
5753 | list_for_each_entry(dev, head, unreg_list) | |
5754 | net_set_todo(dev); | |
5755 | } | |
5756 | } | |
63c8099d | 5757 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 5758 | |
1da177e4 LT |
5759 | /** |
5760 | * unregister_netdev - remove device from the kernel | |
5761 | * @dev: device | |
5762 | * | |
5763 | * This function shuts down a device interface and removes it | |
d59b54b1 | 5764 | * from the kernel tables. |
1da177e4 LT |
5765 | * |
5766 | * This is just a wrapper for unregister_netdevice that takes | |
5767 | * the rtnl semaphore. In general you want to use this and not | |
5768 | * unregister_netdevice. | |
5769 | */ | |
5770 | void unregister_netdev(struct net_device *dev) | |
5771 | { | |
5772 | rtnl_lock(); | |
5773 | unregister_netdevice(dev); | |
5774 | rtnl_unlock(); | |
5775 | } | |
1da177e4 LT |
5776 | EXPORT_SYMBOL(unregister_netdev); |
5777 | ||
ce286d32 EB |
5778 | /** |
5779 | * dev_change_net_namespace - move device to different nethost namespace | |
5780 | * @dev: device | |
5781 | * @net: network namespace | |
5782 | * @pat: If not NULL name pattern to try if the current device name | |
5783 | * is already taken in the destination network namespace. | |
5784 | * | |
5785 | * This function shuts down a device interface and moves it | |
5786 | * to a new network namespace. On success 0 is returned, on | |
5787 | * a failure a netagive errno code is returned. | |
5788 | * | |
5789 | * Callers must hold the rtnl semaphore. | |
5790 | */ | |
5791 | ||
5792 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
5793 | { | |
ce286d32 EB |
5794 | int err; |
5795 | ||
5796 | ASSERT_RTNL(); | |
5797 | ||
5798 | /* Don't allow namespace local devices to be moved. */ | |
5799 | err = -EINVAL; | |
5800 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5801 | goto out; | |
5802 | ||
5803 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
5804 | if (dev->reg_state != NETREG_REGISTERED) |
5805 | goto out; | |
5806 | ||
5807 | /* Get out if there is nothing todo */ | |
5808 | err = 0; | |
878628fb | 5809 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
5810 | goto out; |
5811 | ||
5812 | /* Pick the destination device name, and ensure | |
5813 | * we can use it in the destination network namespace. | |
5814 | */ | |
5815 | err = -EEXIST; | |
d9031024 | 5816 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
5817 | /* We get here if we can't use the current device name */ |
5818 | if (!pat) | |
5819 | goto out; | |
828de4f6 | 5820 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
5821 | goto out; |
5822 | } | |
5823 | ||
5824 | /* | |
5825 | * And now a mini version of register_netdevice unregister_netdevice. | |
5826 | */ | |
5827 | ||
5828 | /* If device is running close it first. */ | |
9b772652 | 5829 | dev_close(dev); |
ce286d32 EB |
5830 | |
5831 | /* And unlink it from device chain */ | |
5832 | err = -ENODEV; | |
5833 | unlist_netdevice(dev); | |
5834 | ||
5835 | synchronize_net(); | |
5836 | ||
5837 | /* Shutdown queueing discipline. */ | |
5838 | dev_shutdown(dev); | |
5839 | ||
5840 | /* Notify protocols, that we are about to destroy | |
5841 | this device. They should clean all the things. | |
3b27e105 DL |
5842 | |
5843 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
5844 | This is wanted because this way 8021q and macvlan know | |
5845 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
5846 | */ |
5847 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
5848 | rcu_barrier(); |
5849 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
d2237d35 | 5850 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); |
ce286d32 EB |
5851 | |
5852 | /* | |
5853 | * Flush the unicast and multicast chains | |
5854 | */ | |
a748ee24 | 5855 | dev_uc_flush(dev); |
22bedad3 | 5856 | dev_mc_flush(dev); |
ce286d32 | 5857 | |
4e66ae2e SH |
5858 | /* Send a netdev-removed uevent to the old namespace */ |
5859 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
5860 | ||
ce286d32 | 5861 | /* Actually switch the network namespace */ |
c346dca1 | 5862 | dev_net_set(dev, net); |
ce286d32 | 5863 | |
ce286d32 EB |
5864 | /* If there is an ifindex conflict assign a new one */ |
5865 | if (__dev_get_by_index(net, dev->ifindex)) { | |
5866 | int iflink = (dev->iflink == dev->ifindex); | |
5867 | dev->ifindex = dev_new_index(net); | |
5868 | if (iflink) | |
5869 | dev->iflink = dev->ifindex; | |
5870 | } | |
5871 | ||
4e66ae2e SH |
5872 | /* Send a netdev-add uevent to the new namespace */ |
5873 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
5874 | ||
8b41d188 | 5875 | /* Fixup kobjects */ |
a1b3f594 | 5876 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 5877 | WARN_ON(err); |
ce286d32 EB |
5878 | |
5879 | /* Add the device back in the hashes */ | |
5880 | list_netdevice(dev); | |
5881 | ||
5882 | /* Notify protocols, that a new device appeared. */ | |
5883 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5884 | ||
d90a909e EB |
5885 | /* |
5886 | * Prevent userspace races by waiting until the network | |
5887 | * device is fully setup before sending notifications. | |
5888 | */ | |
5889 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
5890 | ||
ce286d32 EB |
5891 | synchronize_net(); |
5892 | err = 0; | |
5893 | out: | |
5894 | return err; | |
5895 | } | |
463d0183 | 5896 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 5897 | |
1da177e4 LT |
5898 | static int dev_cpu_callback(struct notifier_block *nfb, |
5899 | unsigned long action, | |
5900 | void *ocpu) | |
5901 | { | |
5902 | struct sk_buff **list_skb; | |
1da177e4 LT |
5903 | struct sk_buff *skb; |
5904 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
5905 | struct softnet_data *sd, *oldsd; | |
5906 | ||
8bb78442 | 5907 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
5908 | return NOTIFY_OK; |
5909 | ||
5910 | local_irq_disable(); | |
5911 | cpu = smp_processor_id(); | |
5912 | sd = &per_cpu(softnet_data, cpu); | |
5913 | oldsd = &per_cpu(softnet_data, oldcpu); | |
5914 | ||
5915 | /* Find end of our completion_queue. */ | |
5916 | list_skb = &sd->completion_queue; | |
5917 | while (*list_skb) | |
5918 | list_skb = &(*list_skb)->next; | |
5919 | /* Append completion queue from offline CPU. */ | |
5920 | *list_skb = oldsd->completion_queue; | |
5921 | oldsd->completion_queue = NULL; | |
5922 | ||
1da177e4 | 5923 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
5924 | if (oldsd->output_queue) { |
5925 | *sd->output_queue_tailp = oldsd->output_queue; | |
5926 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
5927 | oldsd->output_queue = NULL; | |
5928 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
5929 | } | |
264524d5 HC |
5930 | /* Append NAPI poll list from offline CPU. */ |
5931 | if (!list_empty(&oldsd->poll_list)) { | |
5932 | list_splice_init(&oldsd->poll_list, &sd->poll_list); | |
5933 | raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
5934 | } | |
1da177e4 LT |
5935 | |
5936 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
5937 | local_irq_enable(); | |
5938 | ||
5939 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 5940 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
1da177e4 | 5941 | netif_rx(skb); |
76cc8b13 | 5942 | input_queue_head_incr(oldsd); |
fec5e652 | 5943 | } |
76cc8b13 | 5944 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
6e7676c1 | 5945 | netif_rx(skb); |
76cc8b13 TH |
5946 | input_queue_head_incr(oldsd); |
5947 | } | |
1da177e4 LT |
5948 | |
5949 | return NOTIFY_OK; | |
5950 | } | |
1da177e4 LT |
5951 | |
5952 | ||
7f353bf2 | 5953 | /** |
b63365a2 HX |
5954 | * netdev_increment_features - increment feature set by one |
5955 | * @all: current feature set | |
5956 | * @one: new feature set | |
5957 | * @mask: mask feature set | |
7f353bf2 HX |
5958 | * |
5959 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
5960 | * @one to the master device with current feature set @all. Will not |
5961 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 5962 | */ |
c8f44aff MM |
5963 | netdev_features_t netdev_increment_features(netdev_features_t all, |
5964 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 5965 | { |
1742f183 MM |
5966 | if (mask & NETIF_F_GEN_CSUM) |
5967 | mask |= NETIF_F_ALL_CSUM; | |
5968 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 5969 | |
1742f183 MM |
5970 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
5971 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 5972 | |
1742f183 MM |
5973 | /* If one device supports hw checksumming, set for all. */ |
5974 | if (all & NETIF_F_GEN_CSUM) | |
5975 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
5976 | |
5977 | return all; | |
5978 | } | |
b63365a2 | 5979 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 5980 | |
30d97d35 PE |
5981 | static struct hlist_head *netdev_create_hash(void) |
5982 | { | |
5983 | int i; | |
5984 | struct hlist_head *hash; | |
5985 | ||
5986 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
5987 | if (hash != NULL) | |
5988 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
5989 | INIT_HLIST_HEAD(&hash[i]); | |
5990 | ||
5991 | return hash; | |
5992 | } | |
5993 | ||
881d966b | 5994 | /* Initialize per network namespace state */ |
4665079c | 5995 | static int __net_init netdev_init(struct net *net) |
881d966b | 5996 | { |
734b6541 RM |
5997 | if (net != &init_net) |
5998 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 5999 | |
30d97d35 PE |
6000 | net->dev_name_head = netdev_create_hash(); |
6001 | if (net->dev_name_head == NULL) | |
6002 | goto err_name; | |
881d966b | 6003 | |
30d97d35 PE |
6004 | net->dev_index_head = netdev_create_hash(); |
6005 | if (net->dev_index_head == NULL) | |
6006 | goto err_idx; | |
881d966b EB |
6007 | |
6008 | return 0; | |
30d97d35 PE |
6009 | |
6010 | err_idx: | |
6011 | kfree(net->dev_name_head); | |
6012 | err_name: | |
6013 | return -ENOMEM; | |
881d966b EB |
6014 | } |
6015 | ||
f0db275a SH |
6016 | /** |
6017 | * netdev_drivername - network driver for the device | |
6018 | * @dev: network device | |
f0db275a SH |
6019 | * |
6020 | * Determine network driver for device. | |
6021 | */ | |
3019de12 | 6022 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 6023 | { |
cf04a4c7 SH |
6024 | const struct device_driver *driver; |
6025 | const struct device *parent; | |
3019de12 | 6026 | const char *empty = ""; |
6579e57b AV |
6027 | |
6028 | parent = dev->dev.parent; | |
6579e57b | 6029 | if (!parent) |
3019de12 | 6030 | return empty; |
6579e57b AV |
6031 | |
6032 | driver = parent->driver; | |
6033 | if (driver && driver->name) | |
3019de12 DM |
6034 | return driver->name; |
6035 | return empty; | |
6579e57b AV |
6036 | } |
6037 | ||
b004ff49 | 6038 | static int __netdev_printk(const char *level, const struct net_device *dev, |
256df2f3 JP |
6039 | struct va_format *vaf) |
6040 | { | |
6041 | int r; | |
6042 | ||
b004ff49 | 6043 | if (dev && dev->dev.parent) { |
666f355f JP |
6044 | r = dev_printk_emit(level[1] - '0', |
6045 | dev->dev.parent, | |
6046 | "%s %s %s: %pV", | |
6047 | dev_driver_string(dev->dev.parent), | |
6048 | dev_name(dev->dev.parent), | |
6049 | netdev_name(dev), vaf); | |
b004ff49 | 6050 | } else if (dev) { |
256df2f3 | 6051 | r = printk("%s%s: %pV", level, netdev_name(dev), vaf); |
b004ff49 | 6052 | } else { |
256df2f3 | 6053 | r = printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 6054 | } |
256df2f3 JP |
6055 | |
6056 | return r; | |
6057 | } | |
6058 | ||
6059 | int netdev_printk(const char *level, const struct net_device *dev, | |
6060 | const char *format, ...) | |
6061 | { | |
6062 | struct va_format vaf; | |
6063 | va_list args; | |
6064 | int r; | |
6065 | ||
6066 | va_start(args, format); | |
6067 | ||
6068 | vaf.fmt = format; | |
6069 | vaf.va = &args; | |
6070 | ||
6071 | r = __netdev_printk(level, dev, &vaf); | |
b004ff49 | 6072 | |
256df2f3 JP |
6073 | va_end(args); |
6074 | ||
6075 | return r; | |
6076 | } | |
6077 | EXPORT_SYMBOL(netdev_printk); | |
6078 | ||
6079 | #define define_netdev_printk_level(func, level) \ | |
6080 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
6081 | { \ | |
6082 | int r; \ | |
6083 | struct va_format vaf; \ | |
6084 | va_list args; \ | |
6085 | \ | |
6086 | va_start(args, fmt); \ | |
6087 | \ | |
6088 | vaf.fmt = fmt; \ | |
6089 | vaf.va = &args; \ | |
6090 | \ | |
6091 | r = __netdev_printk(level, dev, &vaf); \ | |
b004ff49 | 6092 | \ |
256df2f3 JP |
6093 | va_end(args); \ |
6094 | \ | |
6095 | return r; \ | |
6096 | } \ | |
6097 | EXPORT_SYMBOL(func); | |
6098 | ||
6099 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
6100 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
6101 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
6102 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
6103 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
6104 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
6105 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
6106 | ||
4665079c | 6107 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
6108 | { |
6109 | kfree(net->dev_name_head); | |
6110 | kfree(net->dev_index_head); | |
6111 | } | |
6112 | ||
022cbae6 | 6113 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
6114 | .init = netdev_init, |
6115 | .exit = netdev_exit, | |
6116 | }; | |
6117 | ||
4665079c | 6118 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 6119 | { |
e008b5fc | 6120 | struct net_device *dev, *aux; |
ce286d32 | 6121 | /* |
e008b5fc | 6122 | * Push all migratable network devices back to the |
ce286d32 EB |
6123 | * initial network namespace |
6124 | */ | |
6125 | rtnl_lock(); | |
e008b5fc | 6126 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 6127 | int err; |
aca51397 | 6128 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
6129 | |
6130 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6131 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6132 | continue; | |
6133 | ||
e008b5fc EB |
6134 | /* Leave virtual devices for the generic cleanup */ |
6135 | if (dev->rtnl_link_ops) | |
6136 | continue; | |
d0c082ce | 6137 | |
25985edc | 6138 | /* Push remaining network devices to init_net */ |
aca51397 PE |
6139 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
6140 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 6141 | if (err) { |
7b6cd1ce JP |
6142 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
6143 | __func__, dev->name, err); | |
aca51397 | 6144 | BUG(); |
ce286d32 EB |
6145 | } |
6146 | } | |
6147 | rtnl_unlock(); | |
6148 | } | |
6149 | ||
04dc7f6b EB |
6150 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
6151 | { | |
6152 | /* At exit all network devices most be removed from a network | |
b595076a | 6153 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
6154 | * Do this across as many network namespaces as possible to |
6155 | * improve batching efficiency. | |
6156 | */ | |
6157 | struct net_device *dev; | |
6158 | struct net *net; | |
6159 | LIST_HEAD(dev_kill_list); | |
6160 | ||
6161 | rtnl_lock(); | |
6162 | list_for_each_entry(net, net_list, exit_list) { | |
6163 | for_each_netdev_reverse(net, dev) { | |
6164 | if (dev->rtnl_link_ops) | |
6165 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
6166 | else | |
6167 | unregister_netdevice_queue(dev, &dev_kill_list); | |
6168 | } | |
6169 | } | |
6170 | unregister_netdevice_many(&dev_kill_list); | |
ceaaec98 | 6171 | list_del(&dev_kill_list); |
04dc7f6b EB |
6172 | rtnl_unlock(); |
6173 | } | |
6174 | ||
022cbae6 | 6175 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 6176 | .exit = default_device_exit, |
04dc7f6b | 6177 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
6178 | }; |
6179 | ||
1da177e4 LT |
6180 | /* |
6181 | * Initialize the DEV module. At boot time this walks the device list and | |
6182 | * unhooks any devices that fail to initialise (normally hardware not | |
6183 | * present) and leaves us with a valid list of present and active devices. | |
6184 | * | |
6185 | */ | |
6186 | ||
6187 | /* | |
6188 | * This is called single threaded during boot, so no need | |
6189 | * to take the rtnl semaphore. | |
6190 | */ | |
6191 | static int __init net_dev_init(void) | |
6192 | { | |
6193 | int i, rc = -ENOMEM; | |
6194 | ||
6195 | BUG_ON(!dev_boot_phase); | |
6196 | ||
1da177e4 LT |
6197 | if (dev_proc_init()) |
6198 | goto out; | |
6199 | ||
8b41d188 | 6200 | if (netdev_kobject_init()) |
1da177e4 LT |
6201 | goto out; |
6202 | ||
6203 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 6204 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
6205 | INIT_LIST_HEAD(&ptype_base[i]); |
6206 | ||
62532da9 VY |
6207 | INIT_LIST_HEAD(&offload_base); |
6208 | ||
881d966b EB |
6209 | if (register_pernet_subsys(&netdev_net_ops)) |
6210 | goto out; | |
1da177e4 LT |
6211 | |
6212 | /* | |
6213 | * Initialise the packet receive queues. | |
6214 | */ | |
6215 | ||
6f912042 | 6216 | for_each_possible_cpu(i) { |
e36fa2f7 | 6217 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 6218 | |
dee42870 | 6219 | memset(sd, 0, sizeof(*sd)); |
e36fa2f7 | 6220 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 6221 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 ED |
6222 | sd->completion_queue = NULL; |
6223 | INIT_LIST_HEAD(&sd->poll_list); | |
a9cbd588 CG |
6224 | sd->output_queue = NULL; |
6225 | sd->output_queue_tailp = &sd->output_queue; | |
df334545 | 6226 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
6227 | sd->csd.func = rps_trigger_softirq; |
6228 | sd->csd.info = sd; | |
6229 | sd->csd.flags = 0; | |
6230 | sd->cpu = i; | |
1e94d72f | 6231 | #endif |
0a9627f2 | 6232 | |
e36fa2f7 ED |
6233 | sd->backlog.poll = process_backlog; |
6234 | sd->backlog.weight = weight_p; | |
6235 | sd->backlog.gro_list = NULL; | |
6236 | sd->backlog.gro_count = 0; | |
1da177e4 LT |
6237 | } |
6238 | ||
1da177e4 LT |
6239 | dev_boot_phase = 0; |
6240 | ||
505d4f73 EB |
6241 | /* The loopback device is special if any other network devices |
6242 | * is present in a network namespace the loopback device must | |
6243 | * be present. Since we now dynamically allocate and free the | |
6244 | * loopback device ensure this invariant is maintained by | |
6245 | * keeping the loopback device as the first device on the | |
6246 | * list of network devices. Ensuring the loopback devices | |
6247 | * is the first device that appears and the last network device | |
6248 | * that disappears. | |
6249 | */ | |
6250 | if (register_pernet_device(&loopback_net_ops)) | |
6251 | goto out; | |
6252 | ||
6253 | if (register_pernet_device(&default_device_ops)) | |
6254 | goto out; | |
6255 | ||
962cf36c CM |
6256 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
6257 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
6258 | |
6259 | hotcpu_notifier(dev_cpu_callback, 0); | |
6260 | dst_init(); | |
1da177e4 LT |
6261 | rc = 0; |
6262 | out: | |
6263 | return rc; | |
6264 | } | |
6265 | ||
6266 | subsys_initcall(net_dev_init); |