Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * NET3 Protocol independent device support routines. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Derived from the non IP parts of dev.c 1.0.19 | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * | |
14 | * Additional Authors: | |
15 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
16 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
17 | * David Hinds <dahinds@users.sourceforge.net> | |
18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
19 | * Adam Sulmicki <adam@cfar.umd.edu> | |
20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
21 | * | |
22 | * Changes: | |
23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
24 | * to 2 if register_netdev gets called | |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
27 | * Alan Cox : device private ioctl copies fields back. | |
28 | * Alan Cox : Transmit queue code does relevant | |
29 | * stunts to keep the queue safe. | |
30 | * Alan Cox : Fixed double lock. | |
31 | * Alan Cox : Fixed promisc NULL pointer trap | |
32 | * ???????? : Support the full private ioctl range | |
33 | * Alan Cox : Moved ioctl permission check into | |
34 | * drivers | |
35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
36 | * Alan Cox : 100 backlog just doesn't cut it when | |
37 | * you start doing multicast video 8) | |
38 | * Alan Cox : Rewrote net_bh and list manager. | |
39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. | |
40 | * Alan Cox : Took out transmit every packet pass | |
41 | * Saved a few bytes in the ioctl handler | |
42 | * Alan Cox : Network driver sets packet type before | |
43 | * calling netif_rx. Saves a function | |
44 | * call a packet. | |
45 | * Alan Cox : Hashed net_bh() | |
46 | * Richard Kooijman: Timestamp fixes. | |
47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
48 | * Alan Cox : Device lock protection. | |
49 | * Alan Cox : Fixed nasty side effect of device close | |
50 | * changes. | |
51 | * Rudi Cilibrasi : Pass the right thing to | |
52 | * set_mac_address() | |
53 | * Dave Miller : 32bit quantity for the device lock to | |
54 | * make it work out on a Sparc. | |
55 | * Bjorn Ekwall : Added KERNELD hack. | |
56 | * Alan Cox : Cleaned up the backlog initialise. | |
57 | * Craig Metz : SIOCGIFCONF fix if space for under | |
58 | * 1 device. | |
59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
60 | * is no device open function. | |
61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
63 | * Cyrus Durgin : Cleaned for KMOD | |
64 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
65 | * A network device unload needs to purge | |
66 | * the backlog queue. | |
67 | * Paul Rusty Russell : SIOCSIFNAME | |
68 | * Pekka Riikonen : Netdev boot-time settings code | |
69 | * Andrew Morton : Make unregister_netdevice wait | |
70 | * indefinitely on dev->refcnt | |
71 | * J Hadi Salim : - Backlog queue sampling | |
72 | * - netif_rx() feedback | |
73 | */ | |
74 | ||
75 | #include <asm/uaccess.h> | |
1da177e4 | 76 | #include <linux/bitops.h> |
4fc268d2 | 77 | #include <linux/capability.h> |
1da177e4 LT |
78 | #include <linux/cpu.h> |
79 | #include <linux/types.h> | |
80 | #include <linux/kernel.h> | |
08e9897d | 81 | #include <linux/hash.h> |
5a0e3ad6 | 82 | #include <linux/slab.h> |
1da177e4 | 83 | #include <linux/sched.h> |
4a3e2f71 | 84 | #include <linux/mutex.h> |
1da177e4 LT |
85 | #include <linux/string.h> |
86 | #include <linux/mm.h> | |
87 | #include <linux/socket.h> | |
88 | #include <linux/sockios.h> | |
89 | #include <linux/errno.h> | |
90 | #include <linux/interrupt.h> | |
91 | #include <linux/if_ether.h> | |
92 | #include <linux/netdevice.h> | |
93 | #include <linux/etherdevice.h> | |
0187bdfb | 94 | #include <linux/ethtool.h> |
1da177e4 LT |
95 | #include <linux/notifier.h> |
96 | #include <linux/skbuff.h> | |
457c4cbc | 97 | #include <net/net_namespace.h> |
1da177e4 LT |
98 | #include <net/sock.h> |
99 | #include <linux/rtnetlink.h> | |
1da177e4 | 100 | #include <linux/stat.h> |
1da177e4 LT |
101 | #include <net/dst.h> |
102 | #include <net/pkt_sched.h> | |
103 | #include <net/checksum.h> | |
44540960 | 104 | #include <net/xfrm.h> |
1da177e4 LT |
105 | #include <linux/highmem.h> |
106 | #include <linux/init.h> | |
1da177e4 | 107 | #include <linux/module.h> |
1da177e4 LT |
108 | #include <linux/netpoll.h> |
109 | #include <linux/rcupdate.h> | |
110 | #include <linux/delay.h> | |
1da177e4 | 111 | #include <net/iw_handler.h> |
1da177e4 | 112 | #include <asm/current.h> |
5bdb9886 | 113 | #include <linux/audit.h> |
db217334 | 114 | #include <linux/dmaengine.h> |
f6a78bfc | 115 | #include <linux/err.h> |
c7fa9d18 | 116 | #include <linux/ctype.h> |
723e98b7 | 117 | #include <linux/if_arp.h> |
6de329e2 | 118 | #include <linux/if_vlan.h> |
8f0f2223 | 119 | #include <linux/ip.h> |
ad55dcaf | 120 | #include <net/ip.h> |
8f0f2223 DM |
121 | #include <linux/ipv6.h> |
122 | #include <linux/in.h> | |
b6b2fed1 DM |
123 | #include <linux/jhash.h> |
124 | #include <linux/random.h> | |
9cbc1cb8 | 125 | #include <trace/events/napi.h> |
cf66ba58 | 126 | #include <trace/events/net.h> |
07dc22e7 | 127 | #include <trace/events/skb.h> |
5acbbd42 | 128 | #include <linux/pci.h> |
caeda9b9 | 129 | #include <linux/inetdevice.h> |
c445477d | 130 | #include <linux/cpu_rmap.h> |
c5905afb | 131 | #include <linux/static_key.h> |
af12fa6e | 132 | #include <linux/hashtable.h> |
1da177e4 | 133 | |
342709ef PE |
134 | #include "net-sysfs.h" |
135 | ||
d565b0a1 HX |
136 | /* Instead of increasing this, you should create a hash table. */ |
137 | #define MAX_GRO_SKBS 8 | |
138 | ||
5d38a079 HX |
139 | /* This should be increased if a protocol with a bigger head is added. */ |
140 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
141 | ||
1da177e4 | 142 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 143 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
144 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
145 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 146 | static struct list_head offload_base __read_mostly; |
1da177e4 | 147 | |
1da177e4 | 148 | /* |
7562f876 | 149 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
150 | * semaphore. |
151 | * | |
c6d14c84 | 152 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
153 | * |
154 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 155 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
156 | * actual updates. This allows pure readers to access the list even |
157 | * while a writer is preparing to update it. | |
158 | * | |
159 | * To put it another way, dev_base_lock is held for writing only to | |
160 | * protect against pure readers; the rtnl semaphore provides the | |
161 | * protection against other writers. | |
162 | * | |
163 | * See, for example usages, register_netdevice() and | |
164 | * unregister_netdevice(), which must be called with the rtnl | |
165 | * semaphore held. | |
166 | */ | |
1da177e4 | 167 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
168 | EXPORT_SYMBOL(dev_base_lock); |
169 | ||
af12fa6e ET |
170 | /* protects napi_hash addition/deletion and napi_gen_id */ |
171 | static DEFINE_SPINLOCK(napi_hash_lock); | |
172 | ||
173 | static unsigned int napi_gen_id; | |
174 | static DEFINE_HASHTABLE(napi_hash, 8); | |
175 | ||
30e6c9fa | 176 | seqcount_t devnet_rename_seq; |
c91f6df2 | 177 | |
4e985ada TG |
178 | static inline void dev_base_seq_inc(struct net *net) |
179 | { | |
180 | while (++net->dev_base_seq == 0); | |
181 | } | |
182 | ||
881d966b | 183 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 184 | { |
95c96174 ED |
185 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
186 | ||
08e9897d | 187 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
188 | } |
189 | ||
881d966b | 190 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 191 | { |
7c28bd0b | 192 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
193 | } |
194 | ||
e36fa2f7 | 195 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
196 | { |
197 | #ifdef CONFIG_RPS | |
e36fa2f7 | 198 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
199 | #endif |
200 | } | |
201 | ||
e36fa2f7 | 202 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
203 | { |
204 | #ifdef CONFIG_RPS | |
e36fa2f7 | 205 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
206 | #endif |
207 | } | |
208 | ||
ce286d32 | 209 | /* Device list insertion */ |
53759be9 | 210 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 211 | { |
c346dca1 | 212 | struct net *net = dev_net(dev); |
ce286d32 EB |
213 | |
214 | ASSERT_RTNL(); | |
215 | ||
216 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 217 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 218 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
219 | hlist_add_head_rcu(&dev->index_hlist, |
220 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 221 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
222 | |
223 | dev_base_seq_inc(net); | |
ce286d32 EB |
224 | } |
225 | ||
fb699dfd ED |
226 | /* Device list removal |
227 | * caller must respect a RCU grace period before freeing/reusing dev | |
228 | */ | |
ce286d32 EB |
229 | static void unlist_netdevice(struct net_device *dev) |
230 | { | |
231 | ASSERT_RTNL(); | |
232 | ||
233 | /* Unlink dev from the device chain */ | |
234 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 235 | list_del_rcu(&dev->dev_list); |
72c9528b | 236 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 237 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 238 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
239 | |
240 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
241 | } |
242 | ||
1da177e4 LT |
243 | /* |
244 | * Our notifier list | |
245 | */ | |
246 | ||
f07d5b94 | 247 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
248 | |
249 | /* | |
250 | * Device drivers call our routines to queue packets here. We empty the | |
251 | * queue in the local softnet handler. | |
252 | */ | |
bea3348e | 253 | |
9958da05 | 254 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 255 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 256 | |
cf508b12 | 257 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 258 | /* |
c773e847 | 259 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
260 | * according to dev->type |
261 | */ | |
262 | static const unsigned short netdev_lock_type[] = | |
263 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
264 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
265 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
266 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
267 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
268 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
269 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
270 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
271 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
272 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
273 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
274 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
275 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
276 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
277 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 278 | |
36cbd3dc | 279 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
280 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
281 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
282 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
283 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
284 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
285 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
286 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
287 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
288 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
289 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
290 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
291 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
292 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
293 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
294 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
295 | |
296 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 297 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
298 | |
299 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
300 | { | |
301 | int i; | |
302 | ||
303 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
304 | if (netdev_lock_type[i] == dev_type) | |
305 | return i; | |
306 | /* the last key is used by default */ | |
307 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
308 | } | |
309 | ||
cf508b12 DM |
310 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
311 | unsigned short dev_type) | |
723e98b7 JP |
312 | { |
313 | int i; | |
314 | ||
315 | i = netdev_lock_pos(dev_type); | |
316 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
317 | netdev_lock_name[i]); | |
318 | } | |
cf508b12 DM |
319 | |
320 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
321 | { | |
322 | int i; | |
323 | ||
324 | i = netdev_lock_pos(dev->type); | |
325 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
326 | &netdev_addr_lock_key[i], | |
327 | netdev_lock_name[i]); | |
328 | } | |
723e98b7 | 329 | #else |
cf508b12 DM |
330 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
331 | unsigned short dev_type) | |
332 | { | |
333 | } | |
334 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
335 | { |
336 | } | |
337 | #endif | |
1da177e4 LT |
338 | |
339 | /******************************************************************************* | |
340 | ||
341 | Protocol management and registration routines | |
342 | ||
343 | *******************************************************************************/ | |
344 | ||
1da177e4 LT |
345 | /* |
346 | * Add a protocol ID to the list. Now that the input handler is | |
347 | * smarter we can dispense with all the messy stuff that used to be | |
348 | * here. | |
349 | * | |
350 | * BEWARE!!! Protocol handlers, mangling input packets, | |
351 | * MUST BE last in hash buckets and checking protocol handlers | |
352 | * MUST start from promiscuous ptype_all chain in net_bh. | |
353 | * It is true now, do not change it. | |
354 | * Explanation follows: if protocol handler, mangling packet, will | |
355 | * be the first on list, it is not able to sense, that packet | |
356 | * is cloned and should be copied-on-write, so that it will | |
357 | * change it and subsequent readers will get broken packet. | |
358 | * --ANK (980803) | |
359 | */ | |
360 | ||
c07b68e8 ED |
361 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
362 | { | |
363 | if (pt->type == htons(ETH_P_ALL)) | |
364 | return &ptype_all; | |
365 | else | |
366 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
367 | } | |
368 | ||
1da177e4 LT |
369 | /** |
370 | * dev_add_pack - add packet handler | |
371 | * @pt: packet type declaration | |
372 | * | |
373 | * Add a protocol handler to the networking stack. The passed &packet_type | |
374 | * is linked into kernel lists and may not be freed until it has been | |
375 | * removed from the kernel lists. | |
376 | * | |
4ec93edb | 377 | * This call does not sleep therefore it can not |
1da177e4 LT |
378 | * guarantee all CPU's that are in middle of receiving packets |
379 | * will see the new packet type (until the next received packet). | |
380 | */ | |
381 | ||
382 | void dev_add_pack(struct packet_type *pt) | |
383 | { | |
c07b68e8 | 384 | struct list_head *head = ptype_head(pt); |
1da177e4 | 385 | |
c07b68e8 ED |
386 | spin_lock(&ptype_lock); |
387 | list_add_rcu(&pt->list, head); | |
388 | spin_unlock(&ptype_lock); | |
1da177e4 | 389 | } |
d1b19dff | 390 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 391 | |
1da177e4 LT |
392 | /** |
393 | * __dev_remove_pack - remove packet handler | |
394 | * @pt: packet type declaration | |
395 | * | |
396 | * Remove a protocol handler that was previously added to the kernel | |
397 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
398 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 399 | * returns. |
1da177e4 LT |
400 | * |
401 | * The packet type might still be in use by receivers | |
402 | * and must not be freed until after all the CPU's have gone | |
403 | * through a quiescent state. | |
404 | */ | |
405 | void __dev_remove_pack(struct packet_type *pt) | |
406 | { | |
c07b68e8 | 407 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
408 | struct packet_type *pt1; |
409 | ||
c07b68e8 | 410 | spin_lock(&ptype_lock); |
1da177e4 LT |
411 | |
412 | list_for_each_entry(pt1, head, list) { | |
413 | if (pt == pt1) { | |
414 | list_del_rcu(&pt->list); | |
415 | goto out; | |
416 | } | |
417 | } | |
418 | ||
7b6cd1ce | 419 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 420 | out: |
c07b68e8 | 421 | spin_unlock(&ptype_lock); |
1da177e4 | 422 | } |
d1b19dff ED |
423 | EXPORT_SYMBOL(__dev_remove_pack); |
424 | ||
1da177e4 LT |
425 | /** |
426 | * dev_remove_pack - remove packet handler | |
427 | * @pt: packet type declaration | |
428 | * | |
429 | * Remove a protocol handler that was previously added to the kernel | |
430 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
431 | * from the kernel lists and can be freed or reused once this function | |
432 | * returns. | |
433 | * | |
434 | * This call sleeps to guarantee that no CPU is looking at the packet | |
435 | * type after return. | |
436 | */ | |
437 | void dev_remove_pack(struct packet_type *pt) | |
438 | { | |
439 | __dev_remove_pack(pt); | |
4ec93edb | 440 | |
1da177e4 LT |
441 | synchronize_net(); |
442 | } | |
d1b19dff | 443 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 444 | |
62532da9 VY |
445 | |
446 | /** | |
447 | * dev_add_offload - register offload handlers | |
448 | * @po: protocol offload declaration | |
449 | * | |
450 | * Add protocol offload handlers to the networking stack. The passed | |
451 | * &proto_offload is linked into kernel lists and may not be freed until | |
452 | * it has been removed from the kernel lists. | |
453 | * | |
454 | * This call does not sleep therefore it can not | |
455 | * guarantee all CPU's that are in middle of receiving packets | |
456 | * will see the new offload handlers (until the next received packet). | |
457 | */ | |
458 | void dev_add_offload(struct packet_offload *po) | |
459 | { | |
460 | struct list_head *head = &offload_base; | |
461 | ||
462 | spin_lock(&offload_lock); | |
463 | list_add_rcu(&po->list, head); | |
464 | spin_unlock(&offload_lock); | |
465 | } | |
466 | EXPORT_SYMBOL(dev_add_offload); | |
467 | ||
468 | /** | |
469 | * __dev_remove_offload - remove offload handler | |
470 | * @po: packet offload declaration | |
471 | * | |
472 | * Remove a protocol offload handler that was previously added to the | |
473 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
474 | * is removed from the kernel lists and can be freed or reused once this | |
475 | * function returns. | |
476 | * | |
477 | * The packet type might still be in use by receivers | |
478 | * and must not be freed until after all the CPU's have gone | |
479 | * through a quiescent state. | |
480 | */ | |
481 | void __dev_remove_offload(struct packet_offload *po) | |
482 | { | |
483 | struct list_head *head = &offload_base; | |
484 | struct packet_offload *po1; | |
485 | ||
c53aa505 | 486 | spin_lock(&offload_lock); |
62532da9 VY |
487 | |
488 | list_for_each_entry(po1, head, list) { | |
489 | if (po == po1) { | |
490 | list_del_rcu(&po->list); | |
491 | goto out; | |
492 | } | |
493 | } | |
494 | ||
495 | pr_warn("dev_remove_offload: %p not found\n", po); | |
496 | out: | |
c53aa505 | 497 | spin_unlock(&offload_lock); |
62532da9 VY |
498 | } |
499 | EXPORT_SYMBOL(__dev_remove_offload); | |
500 | ||
501 | /** | |
502 | * dev_remove_offload - remove packet offload handler | |
503 | * @po: packet offload declaration | |
504 | * | |
505 | * Remove a packet offload handler that was previously added to the kernel | |
506 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
507 | * removed from the kernel lists and can be freed or reused once this | |
508 | * function returns. | |
509 | * | |
510 | * This call sleeps to guarantee that no CPU is looking at the packet | |
511 | * type after return. | |
512 | */ | |
513 | void dev_remove_offload(struct packet_offload *po) | |
514 | { | |
515 | __dev_remove_offload(po); | |
516 | ||
517 | synchronize_net(); | |
518 | } | |
519 | EXPORT_SYMBOL(dev_remove_offload); | |
520 | ||
1da177e4 LT |
521 | /****************************************************************************** |
522 | ||
523 | Device Boot-time Settings Routines | |
524 | ||
525 | *******************************************************************************/ | |
526 | ||
527 | /* Boot time configuration table */ | |
528 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
529 | ||
530 | /** | |
531 | * netdev_boot_setup_add - add new setup entry | |
532 | * @name: name of the device | |
533 | * @map: configured settings for the device | |
534 | * | |
535 | * Adds new setup entry to the dev_boot_setup list. The function | |
536 | * returns 0 on error and 1 on success. This is a generic routine to | |
537 | * all netdevices. | |
538 | */ | |
539 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
540 | { | |
541 | struct netdev_boot_setup *s; | |
542 | int i; | |
543 | ||
544 | s = dev_boot_setup; | |
545 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
546 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
547 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 548 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
549 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
550 | break; | |
551 | } | |
552 | } | |
553 | ||
554 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
555 | } | |
556 | ||
557 | /** | |
558 | * netdev_boot_setup_check - check boot time settings | |
559 | * @dev: the netdevice | |
560 | * | |
561 | * Check boot time settings for the device. | |
562 | * The found settings are set for the device to be used | |
563 | * later in the device probing. | |
564 | * Returns 0 if no settings found, 1 if they are. | |
565 | */ | |
566 | int netdev_boot_setup_check(struct net_device *dev) | |
567 | { | |
568 | struct netdev_boot_setup *s = dev_boot_setup; | |
569 | int i; | |
570 | ||
571 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
572 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 573 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
574 | dev->irq = s[i].map.irq; |
575 | dev->base_addr = s[i].map.base_addr; | |
576 | dev->mem_start = s[i].map.mem_start; | |
577 | dev->mem_end = s[i].map.mem_end; | |
578 | return 1; | |
579 | } | |
580 | } | |
581 | return 0; | |
582 | } | |
d1b19dff | 583 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
584 | |
585 | ||
586 | /** | |
587 | * netdev_boot_base - get address from boot time settings | |
588 | * @prefix: prefix for network device | |
589 | * @unit: id for network device | |
590 | * | |
591 | * Check boot time settings for the base address of device. | |
592 | * The found settings are set for the device to be used | |
593 | * later in the device probing. | |
594 | * Returns 0 if no settings found. | |
595 | */ | |
596 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
597 | { | |
598 | const struct netdev_boot_setup *s = dev_boot_setup; | |
599 | char name[IFNAMSIZ]; | |
600 | int i; | |
601 | ||
602 | sprintf(name, "%s%d", prefix, unit); | |
603 | ||
604 | /* | |
605 | * If device already registered then return base of 1 | |
606 | * to indicate not to probe for this interface | |
607 | */ | |
881d966b | 608 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
609 | return 1; |
610 | ||
611 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
612 | if (!strcmp(name, s[i].name)) | |
613 | return s[i].map.base_addr; | |
614 | return 0; | |
615 | } | |
616 | ||
617 | /* | |
618 | * Saves at boot time configured settings for any netdevice. | |
619 | */ | |
620 | int __init netdev_boot_setup(char *str) | |
621 | { | |
622 | int ints[5]; | |
623 | struct ifmap map; | |
624 | ||
625 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
626 | if (!str || !*str) | |
627 | return 0; | |
628 | ||
629 | /* Save settings */ | |
630 | memset(&map, 0, sizeof(map)); | |
631 | if (ints[0] > 0) | |
632 | map.irq = ints[1]; | |
633 | if (ints[0] > 1) | |
634 | map.base_addr = ints[2]; | |
635 | if (ints[0] > 2) | |
636 | map.mem_start = ints[3]; | |
637 | if (ints[0] > 3) | |
638 | map.mem_end = ints[4]; | |
639 | ||
640 | /* Add new entry to the list */ | |
641 | return netdev_boot_setup_add(str, &map); | |
642 | } | |
643 | ||
644 | __setup("netdev=", netdev_boot_setup); | |
645 | ||
646 | /******************************************************************************* | |
647 | ||
648 | Device Interface Subroutines | |
649 | ||
650 | *******************************************************************************/ | |
651 | ||
652 | /** | |
653 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 654 | * @net: the applicable net namespace |
1da177e4 LT |
655 | * @name: name to find |
656 | * | |
657 | * Find an interface by name. Must be called under RTNL semaphore | |
658 | * or @dev_base_lock. If the name is found a pointer to the device | |
659 | * is returned. If the name is not found then %NULL is returned. The | |
660 | * reference counters are not incremented so the caller must be | |
661 | * careful with locks. | |
662 | */ | |
663 | ||
881d966b | 664 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 665 | { |
0bd8d536 ED |
666 | struct net_device *dev; |
667 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 668 | |
b67bfe0d | 669 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
670 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
671 | return dev; | |
0bd8d536 | 672 | |
1da177e4 LT |
673 | return NULL; |
674 | } | |
d1b19dff | 675 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 676 | |
72c9528b ED |
677 | /** |
678 | * dev_get_by_name_rcu - find a device by its name | |
679 | * @net: the applicable net namespace | |
680 | * @name: name to find | |
681 | * | |
682 | * Find an interface by name. | |
683 | * If the name is found a pointer to the device is returned. | |
684 | * If the name is not found then %NULL is returned. | |
685 | * The reference counters are not incremented so the caller must be | |
686 | * careful with locks. The caller must hold RCU lock. | |
687 | */ | |
688 | ||
689 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
690 | { | |
72c9528b ED |
691 | struct net_device *dev; |
692 | struct hlist_head *head = dev_name_hash(net, name); | |
693 | ||
b67bfe0d | 694 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
695 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
696 | return dev; | |
697 | ||
698 | return NULL; | |
699 | } | |
700 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
701 | ||
1da177e4 LT |
702 | /** |
703 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 704 | * @net: the applicable net namespace |
1da177e4 LT |
705 | * @name: name to find |
706 | * | |
707 | * Find an interface by name. This can be called from any | |
708 | * context and does its own locking. The returned handle has | |
709 | * the usage count incremented and the caller must use dev_put() to | |
710 | * release it when it is no longer needed. %NULL is returned if no | |
711 | * matching device is found. | |
712 | */ | |
713 | ||
881d966b | 714 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
715 | { |
716 | struct net_device *dev; | |
717 | ||
72c9528b ED |
718 | rcu_read_lock(); |
719 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
720 | if (dev) |
721 | dev_hold(dev); | |
72c9528b | 722 | rcu_read_unlock(); |
1da177e4 LT |
723 | return dev; |
724 | } | |
d1b19dff | 725 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
726 | |
727 | /** | |
728 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 729 | * @net: the applicable net namespace |
1da177e4 LT |
730 | * @ifindex: index of device |
731 | * | |
732 | * Search for an interface by index. Returns %NULL if the device | |
733 | * is not found or a pointer to the device. The device has not | |
734 | * had its reference counter increased so the caller must be careful | |
735 | * about locking. The caller must hold either the RTNL semaphore | |
736 | * or @dev_base_lock. | |
737 | */ | |
738 | ||
881d966b | 739 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 740 | { |
0bd8d536 ED |
741 | struct net_device *dev; |
742 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 743 | |
b67bfe0d | 744 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
745 | if (dev->ifindex == ifindex) |
746 | return dev; | |
0bd8d536 | 747 | |
1da177e4 LT |
748 | return NULL; |
749 | } | |
d1b19dff | 750 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 751 | |
fb699dfd ED |
752 | /** |
753 | * dev_get_by_index_rcu - find a device by its ifindex | |
754 | * @net: the applicable net namespace | |
755 | * @ifindex: index of device | |
756 | * | |
757 | * Search for an interface by index. Returns %NULL if the device | |
758 | * is not found or a pointer to the device. The device has not | |
759 | * had its reference counter increased so the caller must be careful | |
760 | * about locking. The caller must hold RCU lock. | |
761 | */ | |
762 | ||
763 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
764 | { | |
fb699dfd ED |
765 | struct net_device *dev; |
766 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
767 | ||
b67bfe0d | 768 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
769 | if (dev->ifindex == ifindex) |
770 | return dev; | |
771 | ||
772 | return NULL; | |
773 | } | |
774 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
775 | ||
1da177e4 LT |
776 | |
777 | /** | |
778 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 779 | * @net: the applicable net namespace |
1da177e4 LT |
780 | * @ifindex: index of device |
781 | * | |
782 | * Search for an interface by index. Returns NULL if the device | |
783 | * is not found or a pointer to the device. The device returned has | |
784 | * had a reference added and the pointer is safe until the user calls | |
785 | * dev_put to indicate they have finished with it. | |
786 | */ | |
787 | ||
881d966b | 788 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
789 | { |
790 | struct net_device *dev; | |
791 | ||
fb699dfd ED |
792 | rcu_read_lock(); |
793 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
794 | if (dev) |
795 | dev_hold(dev); | |
fb699dfd | 796 | rcu_read_unlock(); |
1da177e4 LT |
797 | return dev; |
798 | } | |
d1b19dff | 799 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 LT |
800 | |
801 | /** | |
941666c2 | 802 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 803 | * @net: the applicable net namespace |
1da177e4 LT |
804 | * @type: media type of device |
805 | * @ha: hardware address | |
806 | * | |
807 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
808 | * is not found or a pointer to the device. |
809 | * The caller must hold RCU or RTNL. | |
941666c2 | 810 | * The returned device has not had its ref count increased |
1da177e4 LT |
811 | * and the caller must therefore be careful about locking |
812 | * | |
1da177e4 LT |
813 | */ |
814 | ||
941666c2 ED |
815 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
816 | const char *ha) | |
1da177e4 LT |
817 | { |
818 | struct net_device *dev; | |
819 | ||
941666c2 | 820 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
821 | if (dev->type == type && |
822 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
823 | return dev; |
824 | ||
825 | return NULL; | |
1da177e4 | 826 | } |
941666c2 | 827 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 828 | |
881d966b | 829 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
830 | { |
831 | struct net_device *dev; | |
832 | ||
4e9cac2b | 833 | ASSERT_RTNL(); |
881d966b | 834 | for_each_netdev(net, dev) |
4e9cac2b | 835 | if (dev->type == type) |
7562f876 PE |
836 | return dev; |
837 | ||
838 | return NULL; | |
4e9cac2b | 839 | } |
4e9cac2b PM |
840 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
841 | ||
881d966b | 842 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 843 | { |
99fe3c39 | 844 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 845 | |
99fe3c39 ED |
846 | rcu_read_lock(); |
847 | for_each_netdev_rcu(net, dev) | |
848 | if (dev->type == type) { | |
849 | dev_hold(dev); | |
850 | ret = dev; | |
851 | break; | |
852 | } | |
853 | rcu_read_unlock(); | |
854 | return ret; | |
1da177e4 | 855 | } |
1da177e4 LT |
856 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
857 | ||
858 | /** | |
bb69ae04 | 859 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 860 | * @net: the applicable net namespace |
1da177e4 LT |
861 | * @if_flags: IFF_* values |
862 | * @mask: bitmask of bits in if_flags to check | |
863 | * | |
864 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
865 | * is not found or a pointer to the device. Must be called inside |
866 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
867 | */ |
868 | ||
bb69ae04 | 869 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 870 | unsigned short mask) |
1da177e4 | 871 | { |
7562f876 | 872 | struct net_device *dev, *ret; |
1da177e4 | 873 | |
7562f876 | 874 | ret = NULL; |
c6d14c84 | 875 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 876 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 877 | ret = dev; |
1da177e4 LT |
878 | break; |
879 | } | |
880 | } | |
7562f876 | 881 | return ret; |
1da177e4 | 882 | } |
bb69ae04 | 883 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
884 | |
885 | /** | |
886 | * dev_valid_name - check if name is okay for network device | |
887 | * @name: name string | |
888 | * | |
889 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
890 | * to allow sysfs to work. We also disallow any kind of |
891 | * whitespace. | |
1da177e4 | 892 | */ |
95f050bf | 893 | bool dev_valid_name(const char *name) |
1da177e4 | 894 | { |
c7fa9d18 | 895 | if (*name == '\0') |
95f050bf | 896 | return false; |
b6fe17d6 | 897 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 898 | return false; |
c7fa9d18 | 899 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 900 | return false; |
c7fa9d18 DM |
901 | |
902 | while (*name) { | |
903 | if (*name == '/' || isspace(*name)) | |
95f050bf | 904 | return false; |
c7fa9d18 DM |
905 | name++; |
906 | } | |
95f050bf | 907 | return true; |
1da177e4 | 908 | } |
d1b19dff | 909 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
910 | |
911 | /** | |
b267b179 EB |
912 | * __dev_alloc_name - allocate a name for a device |
913 | * @net: network namespace to allocate the device name in | |
1da177e4 | 914 | * @name: name format string |
b267b179 | 915 | * @buf: scratch buffer and result name string |
1da177e4 LT |
916 | * |
917 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
918 | * id. It scans list of devices to build up a free map, then chooses |
919 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
920 | * while allocating the name and adding the device in order to avoid | |
921 | * duplicates. | |
922 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
923 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
924 | */ |
925 | ||
b267b179 | 926 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
927 | { |
928 | int i = 0; | |
1da177e4 LT |
929 | const char *p; |
930 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 931 | unsigned long *inuse; |
1da177e4 LT |
932 | struct net_device *d; |
933 | ||
934 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
935 | if (p) { | |
936 | /* | |
937 | * Verify the string as this thing may have come from | |
938 | * the user. There must be either one "%d" and no other "%" | |
939 | * characters. | |
940 | */ | |
941 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
942 | return -EINVAL; | |
943 | ||
944 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 945 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
946 | if (!inuse) |
947 | return -ENOMEM; | |
948 | ||
881d966b | 949 | for_each_netdev(net, d) { |
1da177e4 LT |
950 | if (!sscanf(d->name, name, &i)) |
951 | continue; | |
952 | if (i < 0 || i >= max_netdevices) | |
953 | continue; | |
954 | ||
955 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 956 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
957 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
958 | set_bit(i, inuse); | |
959 | } | |
960 | ||
961 | i = find_first_zero_bit(inuse, max_netdevices); | |
962 | free_page((unsigned long) inuse); | |
963 | } | |
964 | ||
d9031024 OP |
965 | if (buf != name) |
966 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 967 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 968 | return i; |
1da177e4 LT |
969 | |
970 | /* It is possible to run out of possible slots | |
971 | * when the name is long and there isn't enough space left | |
972 | * for the digits, or if all bits are used. | |
973 | */ | |
974 | return -ENFILE; | |
975 | } | |
976 | ||
b267b179 EB |
977 | /** |
978 | * dev_alloc_name - allocate a name for a device | |
979 | * @dev: device | |
980 | * @name: name format string | |
981 | * | |
982 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
983 | * id. It scans list of devices to build up a free map, then chooses | |
984 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
985 | * while allocating the name and adding the device in order to avoid | |
986 | * duplicates. | |
987 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
988 | * Returns the number of the unit assigned or a negative errno code. | |
989 | */ | |
990 | ||
991 | int dev_alloc_name(struct net_device *dev, const char *name) | |
992 | { | |
993 | char buf[IFNAMSIZ]; | |
994 | struct net *net; | |
995 | int ret; | |
996 | ||
c346dca1 YH |
997 | BUG_ON(!dev_net(dev)); |
998 | net = dev_net(dev); | |
b267b179 EB |
999 | ret = __dev_alloc_name(net, name, buf); |
1000 | if (ret >= 0) | |
1001 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1002 | return ret; | |
1003 | } | |
d1b19dff | 1004 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1005 | |
828de4f6 G |
1006 | static int dev_alloc_name_ns(struct net *net, |
1007 | struct net_device *dev, | |
1008 | const char *name) | |
d9031024 | 1009 | { |
828de4f6 G |
1010 | char buf[IFNAMSIZ]; |
1011 | int ret; | |
8ce6cebc | 1012 | |
828de4f6 G |
1013 | ret = __dev_alloc_name(net, name, buf); |
1014 | if (ret >= 0) | |
1015 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1016 | return ret; | |
1017 | } | |
1018 | ||
1019 | static int dev_get_valid_name(struct net *net, | |
1020 | struct net_device *dev, | |
1021 | const char *name) | |
1022 | { | |
1023 | BUG_ON(!net); | |
8ce6cebc | 1024 | |
d9031024 OP |
1025 | if (!dev_valid_name(name)) |
1026 | return -EINVAL; | |
1027 | ||
1c5cae81 | 1028 | if (strchr(name, '%')) |
828de4f6 | 1029 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1030 | else if (__dev_get_by_name(net, name)) |
1031 | return -EEXIST; | |
8ce6cebc DL |
1032 | else if (dev->name != name) |
1033 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1034 | |
1035 | return 0; | |
1036 | } | |
1da177e4 LT |
1037 | |
1038 | /** | |
1039 | * dev_change_name - change name of a device | |
1040 | * @dev: device | |
1041 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1042 | * | |
1043 | * Change name of a device, can pass format strings "eth%d". | |
1044 | * for wildcarding. | |
1045 | */ | |
cf04a4c7 | 1046 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1047 | { |
fcc5a03a | 1048 | char oldname[IFNAMSIZ]; |
1da177e4 | 1049 | int err = 0; |
fcc5a03a | 1050 | int ret; |
881d966b | 1051 | struct net *net; |
1da177e4 LT |
1052 | |
1053 | ASSERT_RTNL(); | |
c346dca1 | 1054 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1055 | |
c346dca1 | 1056 | net = dev_net(dev); |
1da177e4 LT |
1057 | if (dev->flags & IFF_UP) |
1058 | return -EBUSY; | |
1059 | ||
30e6c9fa | 1060 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1061 | |
1062 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1063 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1064 | return 0; |
c91f6df2 | 1065 | } |
c8d90dca | 1066 | |
fcc5a03a HX |
1067 | memcpy(oldname, dev->name, IFNAMSIZ); |
1068 | ||
828de4f6 | 1069 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1070 | if (err < 0) { |
30e6c9fa | 1071 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1072 | return err; |
c91f6df2 | 1073 | } |
1da177e4 | 1074 | |
fcc5a03a | 1075 | rollback: |
a1b3f594 EB |
1076 | ret = device_rename(&dev->dev, dev->name); |
1077 | if (ret) { | |
1078 | memcpy(dev->name, oldname, IFNAMSIZ); | |
30e6c9fa | 1079 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1080 | return ret; |
dcc99773 | 1081 | } |
7f988eab | 1082 | |
30e6c9fa | 1083 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1084 | |
7f988eab | 1085 | write_lock_bh(&dev_base_lock); |
372b2312 | 1086 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1087 | write_unlock_bh(&dev_base_lock); |
1088 | ||
1089 | synchronize_rcu(); | |
1090 | ||
1091 | write_lock_bh(&dev_base_lock); | |
1092 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1093 | write_unlock_bh(&dev_base_lock); |
1094 | ||
056925ab | 1095 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1096 | ret = notifier_to_errno(ret); |
1097 | ||
1098 | if (ret) { | |
91e9c07b ED |
1099 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1100 | if (err >= 0) { | |
fcc5a03a | 1101 | err = ret; |
30e6c9fa | 1102 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a HX |
1103 | memcpy(dev->name, oldname, IFNAMSIZ); |
1104 | goto rollback; | |
91e9c07b | 1105 | } else { |
7b6cd1ce | 1106 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1107 | dev->name, ret); |
fcc5a03a HX |
1108 | } |
1109 | } | |
1da177e4 LT |
1110 | |
1111 | return err; | |
1112 | } | |
1113 | ||
0b815a1a SH |
1114 | /** |
1115 | * dev_set_alias - change ifalias of a device | |
1116 | * @dev: device | |
1117 | * @alias: name up to IFALIASZ | |
f0db275a | 1118 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1119 | * |
1120 | * Set ifalias for a device, | |
1121 | */ | |
1122 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1123 | { | |
7364e445 AK |
1124 | char *new_ifalias; |
1125 | ||
0b815a1a SH |
1126 | ASSERT_RTNL(); |
1127 | ||
1128 | if (len >= IFALIASZ) | |
1129 | return -EINVAL; | |
1130 | ||
96ca4a2c | 1131 | if (!len) { |
388dfc2d SK |
1132 | kfree(dev->ifalias); |
1133 | dev->ifalias = NULL; | |
96ca4a2c OH |
1134 | return 0; |
1135 | } | |
1136 | ||
7364e445 AK |
1137 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1138 | if (!new_ifalias) | |
0b815a1a | 1139 | return -ENOMEM; |
7364e445 | 1140 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1141 | |
1142 | strlcpy(dev->ifalias, alias, len+1); | |
1143 | return len; | |
1144 | } | |
1145 | ||
1146 | ||
d8a33ac4 | 1147 | /** |
3041a069 | 1148 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1149 | * @dev: device to cause notification |
1150 | * | |
1151 | * Called to indicate a device has changed features. | |
1152 | */ | |
1153 | void netdev_features_change(struct net_device *dev) | |
1154 | { | |
056925ab | 1155 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1156 | } |
1157 | EXPORT_SYMBOL(netdev_features_change); | |
1158 | ||
1da177e4 LT |
1159 | /** |
1160 | * netdev_state_change - device changes state | |
1161 | * @dev: device to cause notification | |
1162 | * | |
1163 | * Called to indicate a device has changed state. This function calls | |
1164 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1165 | * to the routing socket. | |
1166 | */ | |
1167 | void netdev_state_change(struct net_device *dev) | |
1168 | { | |
1169 | if (dev->flags & IFF_UP) { | |
056925ab | 1170 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1171 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1172 | } | |
1173 | } | |
d1b19dff | 1174 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1175 | |
ee89bab1 AW |
1176 | /** |
1177 | * netdev_notify_peers - notify network peers about existence of @dev | |
1178 | * @dev: network device | |
1179 | * | |
1180 | * Generate traffic such that interested network peers are aware of | |
1181 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1182 | * a device wants to inform the rest of the network about some sort of | |
1183 | * reconfiguration such as a failover event or virtual machine | |
1184 | * migration. | |
1185 | */ | |
1186 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1187 | { |
ee89bab1 AW |
1188 | rtnl_lock(); |
1189 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1190 | rtnl_unlock(); | |
c1da4ac7 | 1191 | } |
ee89bab1 | 1192 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1193 | |
bd380811 | 1194 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1195 | { |
d314774c | 1196 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1197 | int ret; |
1da177e4 | 1198 | |
e46b66bc BH |
1199 | ASSERT_RTNL(); |
1200 | ||
1da177e4 LT |
1201 | if (!netif_device_present(dev)) |
1202 | return -ENODEV; | |
1203 | ||
ca99ca14 NH |
1204 | /* Block netpoll from trying to do any rx path servicing. |
1205 | * If we don't do this there is a chance ndo_poll_controller | |
1206 | * or ndo_poll may be running while we open the device | |
1207 | */ | |
da6e378b | 1208 | netpoll_rx_disable(dev); |
ca99ca14 | 1209 | |
3b8bcfd5 JB |
1210 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1211 | ret = notifier_to_errno(ret); | |
1212 | if (ret) | |
1213 | return ret; | |
1214 | ||
1da177e4 | 1215 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1216 | |
d314774c SH |
1217 | if (ops->ndo_validate_addr) |
1218 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1219 | |
d314774c SH |
1220 | if (!ret && ops->ndo_open) |
1221 | ret = ops->ndo_open(dev); | |
1da177e4 | 1222 | |
ca99ca14 NH |
1223 | netpoll_rx_enable(dev); |
1224 | ||
bada339b JG |
1225 | if (ret) |
1226 | clear_bit(__LINK_STATE_START, &dev->state); | |
1227 | else { | |
1da177e4 | 1228 | dev->flags |= IFF_UP; |
b4bd07c2 | 1229 | net_dmaengine_get(); |
4417da66 | 1230 | dev_set_rx_mode(dev); |
1da177e4 | 1231 | dev_activate(dev); |
7bf23575 | 1232 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1233 | } |
bada339b | 1234 | |
1da177e4 LT |
1235 | return ret; |
1236 | } | |
1237 | ||
1238 | /** | |
bd380811 PM |
1239 | * dev_open - prepare an interface for use. |
1240 | * @dev: device to open | |
1da177e4 | 1241 | * |
bd380811 PM |
1242 | * Takes a device from down to up state. The device's private open |
1243 | * function is invoked and then the multicast lists are loaded. Finally | |
1244 | * the device is moved into the up state and a %NETDEV_UP message is | |
1245 | * sent to the netdev notifier chain. | |
1246 | * | |
1247 | * Calling this function on an active interface is a nop. On a failure | |
1248 | * a negative errno code is returned. | |
1da177e4 | 1249 | */ |
bd380811 PM |
1250 | int dev_open(struct net_device *dev) |
1251 | { | |
1252 | int ret; | |
1253 | ||
bd380811 PM |
1254 | if (dev->flags & IFF_UP) |
1255 | return 0; | |
1256 | ||
bd380811 PM |
1257 | ret = __dev_open(dev); |
1258 | if (ret < 0) | |
1259 | return ret; | |
1260 | ||
bd380811 PM |
1261 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); |
1262 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1263 | ||
1264 | return ret; | |
1265 | } | |
1266 | EXPORT_SYMBOL(dev_open); | |
1267 | ||
44345724 | 1268 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1269 | { |
44345724 | 1270 | struct net_device *dev; |
e46b66bc | 1271 | |
bd380811 | 1272 | ASSERT_RTNL(); |
9d5010db DM |
1273 | might_sleep(); |
1274 | ||
44345724 | 1275 | list_for_each_entry(dev, head, unreg_list) { |
44345724 | 1276 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1277 | |
44345724 | 1278 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1279 | |
44345724 OP |
1280 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1281 | * can be even on different cpu. So just clear netif_running(). | |
1282 | * | |
1283 | * dev->stop() will invoke napi_disable() on all of it's | |
1284 | * napi_struct instances on this device. | |
1285 | */ | |
1286 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1287 | } | |
1da177e4 | 1288 | |
44345724 | 1289 | dev_deactivate_many(head); |
d8b2a4d2 | 1290 | |
44345724 OP |
1291 | list_for_each_entry(dev, head, unreg_list) { |
1292 | const struct net_device_ops *ops = dev->netdev_ops; | |
1da177e4 | 1293 | |
44345724 OP |
1294 | /* |
1295 | * Call the device specific close. This cannot fail. | |
1296 | * Only if device is UP | |
1297 | * | |
1298 | * We allow it to be called even after a DETACH hot-plug | |
1299 | * event. | |
1300 | */ | |
1301 | if (ops->ndo_stop) | |
1302 | ops->ndo_stop(dev); | |
1303 | ||
44345724 | 1304 | dev->flags &= ~IFF_UP; |
44345724 OP |
1305 | net_dmaengine_put(); |
1306 | } | |
1307 | ||
1308 | return 0; | |
1309 | } | |
1310 | ||
1311 | static int __dev_close(struct net_device *dev) | |
1312 | { | |
f87e6f47 | 1313 | int retval; |
44345724 OP |
1314 | LIST_HEAD(single); |
1315 | ||
ca99ca14 | 1316 | /* Temporarily disable netpoll until the interface is down */ |
da6e378b | 1317 | netpoll_rx_disable(dev); |
ca99ca14 | 1318 | |
44345724 | 1319 | list_add(&dev->unreg_list, &single); |
f87e6f47 LT |
1320 | retval = __dev_close_many(&single); |
1321 | list_del(&single); | |
ca99ca14 NH |
1322 | |
1323 | netpoll_rx_enable(dev); | |
f87e6f47 | 1324 | return retval; |
44345724 OP |
1325 | } |
1326 | ||
3fbd8758 | 1327 | static int dev_close_many(struct list_head *head) |
44345724 OP |
1328 | { |
1329 | struct net_device *dev, *tmp; | |
1330 | LIST_HEAD(tmp_list); | |
1da177e4 | 1331 | |
44345724 OP |
1332 | list_for_each_entry_safe(dev, tmp, head, unreg_list) |
1333 | if (!(dev->flags & IFF_UP)) | |
1334 | list_move(&dev->unreg_list, &tmp_list); | |
1335 | ||
1336 | __dev_close_many(head); | |
1da177e4 | 1337 | |
44345724 OP |
1338 | list_for_each_entry(dev, head, unreg_list) { |
1339 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1340 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1341 | } | |
bd380811 | 1342 | |
44345724 OP |
1343 | /* rollback_registered_many needs the complete original list */ |
1344 | list_splice(&tmp_list, head); | |
bd380811 PM |
1345 | return 0; |
1346 | } | |
1347 | ||
1348 | /** | |
1349 | * dev_close - shutdown an interface. | |
1350 | * @dev: device to shutdown | |
1351 | * | |
1352 | * This function moves an active device into down state. A | |
1353 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1354 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1355 | * chain. | |
1356 | */ | |
1357 | int dev_close(struct net_device *dev) | |
1358 | { | |
e14a5993 ED |
1359 | if (dev->flags & IFF_UP) { |
1360 | LIST_HEAD(single); | |
1da177e4 | 1361 | |
ca99ca14 | 1362 | /* Block netpoll rx while the interface is going down */ |
da6e378b | 1363 | netpoll_rx_disable(dev); |
ca99ca14 | 1364 | |
e14a5993 ED |
1365 | list_add(&dev->unreg_list, &single); |
1366 | dev_close_many(&single); | |
1367 | list_del(&single); | |
ca99ca14 NH |
1368 | |
1369 | netpoll_rx_enable(dev); | |
e14a5993 | 1370 | } |
da6e378b | 1371 | return 0; |
1da177e4 | 1372 | } |
d1b19dff | 1373 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1374 | |
1375 | ||
0187bdfb BH |
1376 | /** |
1377 | * dev_disable_lro - disable Large Receive Offload on a device | |
1378 | * @dev: device | |
1379 | * | |
1380 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1381 | * called under RTNL. This is needed if received packets may be | |
1382 | * forwarded to another interface. | |
1383 | */ | |
1384 | void dev_disable_lro(struct net_device *dev) | |
1385 | { | |
f11970e3 NH |
1386 | /* |
1387 | * If we're trying to disable lro on a vlan device | |
1388 | * use the underlying physical device instead | |
1389 | */ | |
1390 | if (is_vlan_dev(dev)) | |
1391 | dev = vlan_dev_real_dev(dev); | |
1392 | ||
bc5787c6 MM |
1393 | dev->wanted_features &= ~NETIF_F_LRO; |
1394 | netdev_update_features(dev); | |
27660515 | 1395 | |
22d5969f MM |
1396 | if (unlikely(dev->features & NETIF_F_LRO)) |
1397 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
0187bdfb BH |
1398 | } |
1399 | EXPORT_SYMBOL(dev_disable_lro); | |
1400 | ||
351638e7 JP |
1401 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1402 | struct net_device *dev) | |
1403 | { | |
1404 | struct netdev_notifier_info info; | |
1405 | ||
1406 | netdev_notifier_info_init(&info, dev); | |
1407 | return nb->notifier_call(nb, val, &info); | |
1408 | } | |
0187bdfb | 1409 | |
881d966b EB |
1410 | static int dev_boot_phase = 1; |
1411 | ||
1da177e4 LT |
1412 | /** |
1413 | * register_netdevice_notifier - register a network notifier block | |
1414 | * @nb: notifier | |
1415 | * | |
1416 | * Register a notifier to be called when network device events occur. | |
1417 | * The notifier passed is linked into the kernel structures and must | |
1418 | * not be reused until it has been unregistered. A negative errno code | |
1419 | * is returned on a failure. | |
1420 | * | |
1421 | * When registered all registration and up events are replayed | |
4ec93edb | 1422 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1423 | * view of the network device list. |
1424 | */ | |
1425 | ||
1426 | int register_netdevice_notifier(struct notifier_block *nb) | |
1427 | { | |
1428 | struct net_device *dev; | |
fcc5a03a | 1429 | struct net_device *last; |
881d966b | 1430 | struct net *net; |
1da177e4 LT |
1431 | int err; |
1432 | ||
1433 | rtnl_lock(); | |
f07d5b94 | 1434 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1435 | if (err) |
1436 | goto unlock; | |
881d966b EB |
1437 | if (dev_boot_phase) |
1438 | goto unlock; | |
1439 | for_each_net(net) { | |
1440 | for_each_netdev(net, dev) { | |
351638e7 | 1441 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1442 | err = notifier_to_errno(err); |
1443 | if (err) | |
1444 | goto rollback; | |
1445 | ||
1446 | if (!(dev->flags & IFF_UP)) | |
1447 | continue; | |
1da177e4 | 1448 | |
351638e7 | 1449 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1450 | } |
1da177e4 | 1451 | } |
fcc5a03a HX |
1452 | |
1453 | unlock: | |
1da177e4 LT |
1454 | rtnl_unlock(); |
1455 | return err; | |
fcc5a03a HX |
1456 | |
1457 | rollback: | |
1458 | last = dev; | |
881d966b EB |
1459 | for_each_net(net) { |
1460 | for_each_netdev(net, dev) { | |
1461 | if (dev == last) | |
8f891489 | 1462 | goto outroll; |
fcc5a03a | 1463 | |
881d966b | 1464 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1465 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1466 | dev); | |
1467 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1468 | } |
351638e7 | 1469 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1470 | } |
fcc5a03a | 1471 | } |
c67625a1 | 1472 | |
8f891489 | 1473 | outroll: |
c67625a1 | 1474 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1475 | goto unlock; |
1da177e4 | 1476 | } |
d1b19dff | 1477 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1478 | |
1479 | /** | |
1480 | * unregister_netdevice_notifier - unregister a network notifier block | |
1481 | * @nb: notifier | |
1482 | * | |
1483 | * Unregister a notifier previously registered by | |
1484 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1485 | * kernel structures and may then be reused. A negative errno code | |
1486 | * is returned on a failure. | |
7d3d43da EB |
1487 | * |
1488 | * After unregistering unregister and down device events are synthesized | |
1489 | * for all devices on the device list to the removed notifier to remove | |
1490 | * the need for special case cleanup code. | |
1da177e4 LT |
1491 | */ |
1492 | ||
1493 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1494 | { | |
7d3d43da EB |
1495 | struct net_device *dev; |
1496 | struct net *net; | |
9f514950 HX |
1497 | int err; |
1498 | ||
1499 | rtnl_lock(); | |
f07d5b94 | 1500 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1501 | if (err) |
1502 | goto unlock; | |
1503 | ||
1504 | for_each_net(net) { | |
1505 | for_each_netdev(net, dev) { | |
1506 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1507 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1508 | dev); | |
1509 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1510 | } |
351638e7 | 1511 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1512 | } |
1513 | } | |
1514 | unlock: | |
9f514950 HX |
1515 | rtnl_unlock(); |
1516 | return err; | |
1da177e4 | 1517 | } |
d1b19dff | 1518 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1519 | |
351638e7 JP |
1520 | /** |
1521 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1522 | * @val: value passed unmodified to notifier function | |
1523 | * @dev: net_device pointer passed unmodified to notifier function | |
1524 | * @info: notifier information data | |
1525 | * | |
1526 | * Call all network notifier blocks. Parameters and return value | |
1527 | * are as for raw_notifier_call_chain(). | |
1528 | */ | |
1529 | ||
1530 | int call_netdevice_notifiers_info(unsigned long val, struct net_device *dev, | |
1531 | struct netdev_notifier_info *info) | |
1532 | { | |
1533 | ASSERT_RTNL(); | |
1534 | netdev_notifier_info_init(info, dev); | |
1535 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1536 | } | |
1537 | EXPORT_SYMBOL(call_netdevice_notifiers_info); | |
1538 | ||
1da177e4 LT |
1539 | /** |
1540 | * call_netdevice_notifiers - call all network notifier blocks | |
1541 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1542 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1543 | * |
1544 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1545 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1546 | */ |
1547 | ||
ad7379d4 | 1548 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1549 | { |
351638e7 JP |
1550 | struct netdev_notifier_info info; |
1551 | ||
1552 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1553 | } |
edf947f1 | 1554 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1555 | |
c5905afb | 1556 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1557 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1558 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1559 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1560 | * static_key_slow_dec() calls. |
b90e5794 ED |
1561 | */ |
1562 | static atomic_t netstamp_needed_deferred; | |
1563 | #endif | |
1da177e4 LT |
1564 | |
1565 | void net_enable_timestamp(void) | |
1566 | { | |
b90e5794 ED |
1567 | #ifdef HAVE_JUMP_LABEL |
1568 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1569 | ||
1570 | if (deferred) { | |
1571 | while (--deferred) | |
c5905afb | 1572 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1573 | return; |
1574 | } | |
1575 | #endif | |
c5905afb | 1576 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1577 | } |
d1b19dff | 1578 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1579 | |
1580 | void net_disable_timestamp(void) | |
1581 | { | |
b90e5794 ED |
1582 | #ifdef HAVE_JUMP_LABEL |
1583 | if (in_interrupt()) { | |
1584 | atomic_inc(&netstamp_needed_deferred); | |
1585 | return; | |
1586 | } | |
1587 | #endif | |
c5905afb | 1588 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1589 | } |
d1b19dff | 1590 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1591 | |
3b098e2d | 1592 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1593 | { |
588f0330 | 1594 | skb->tstamp.tv64 = 0; |
c5905afb | 1595 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1596 | __net_timestamp(skb); |
1da177e4 LT |
1597 | } |
1598 | ||
588f0330 | 1599 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1600 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1601 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1602 | __net_timestamp(SKB); \ | |
1603 | } \ | |
3b098e2d | 1604 | |
79b569f0 DL |
1605 | static inline bool is_skb_forwardable(struct net_device *dev, |
1606 | struct sk_buff *skb) | |
1607 | { | |
1608 | unsigned int len; | |
1609 | ||
1610 | if (!(dev->flags & IFF_UP)) | |
1611 | return false; | |
1612 | ||
1613 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1614 | if (skb->len <= len) | |
1615 | return true; | |
1616 | ||
1617 | /* if TSO is enabled, we don't care about the length as the packet | |
1618 | * could be forwarded without being segmented before | |
1619 | */ | |
1620 | if (skb_is_gso(skb)) | |
1621 | return true; | |
1622 | ||
1623 | return false; | |
1624 | } | |
1625 | ||
44540960 AB |
1626 | /** |
1627 | * dev_forward_skb - loopback an skb to another netif | |
1628 | * | |
1629 | * @dev: destination network device | |
1630 | * @skb: buffer to forward | |
1631 | * | |
1632 | * return values: | |
1633 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1634 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1635 | * |
1636 | * dev_forward_skb can be used for injecting an skb from the | |
1637 | * start_xmit function of one device into the receive queue | |
1638 | * of another device. | |
1639 | * | |
1640 | * The receiving device may be in another namespace, so | |
1641 | * we have to clear all information in the skb that could | |
1642 | * impact namespace isolation. | |
1643 | */ | |
1644 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1645 | { | |
48c83012 MT |
1646 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { |
1647 | if (skb_copy_ubufs(skb, GFP_ATOMIC)) { | |
1648 | atomic_long_inc(&dev->rx_dropped); | |
1649 | kfree_skb(skb); | |
1650 | return NET_RX_DROP; | |
1651 | } | |
1652 | } | |
1653 | ||
44540960 AB |
1654 | skb_orphan(skb); |
1655 | ||
79b569f0 | 1656 | if (unlikely(!is_skb_forwardable(dev, skb))) { |
caf586e5 | 1657 | atomic_long_inc(&dev->rx_dropped); |
6ec82562 | 1658 | kfree_skb(skb); |
44540960 | 1659 | return NET_RX_DROP; |
6ec82562 | 1660 | } |
3b9785c6 | 1661 | skb->skb_iif = 0; |
59b9997b | 1662 | skb_dst_drop(skb); |
44540960 AB |
1663 | skb->tstamp.tv64 = 0; |
1664 | skb->pkt_type = PACKET_HOST; | |
1665 | skb->protocol = eth_type_trans(skb, dev); | |
59b9997b DM |
1666 | skb->mark = 0; |
1667 | secpath_reset(skb); | |
1668 | nf_reset(skb); | |
124dff01 | 1669 | nf_reset_trace(skb); |
44540960 AB |
1670 | return netif_rx(skb); |
1671 | } | |
1672 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1673 | ||
71d9dec2 CG |
1674 | static inline int deliver_skb(struct sk_buff *skb, |
1675 | struct packet_type *pt_prev, | |
1676 | struct net_device *orig_dev) | |
1677 | { | |
1080e512 MT |
1678 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1679 | return -ENOMEM; | |
71d9dec2 CG |
1680 | atomic_inc(&skb->users); |
1681 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1682 | } | |
1683 | ||
c0de08d0 EL |
1684 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1685 | { | |
a3d744e9 | 1686 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1687 | return false; |
1688 | ||
1689 | if (ptype->id_match) | |
1690 | return ptype->id_match(ptype, skb->sk); | |
1691 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1692 | return true; | |
1693 | ||
1694 | return false; | |
1695 | } | |
1696 | ||
1da177e4 LT |
1697 | /* |
1698 | * Support routine. Sends outgoing frames to any network | |
1699 | * taps currently in use. | |
1700 | */ | |
1701 | ||
f6a78bfc | 1702 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1703 | { |
1704 | struct packet_type *ptype; | |
71d9dec2 CG |
1705 | struct sk_buff *skb2 = NULL; |
1706 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1707 | |
1da177e4 LT |
1708 | rcu_read_lock(); |
1709 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1710 | /* Never send packets back to the socket | |
1711 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1712 | */ | |
1713 | if ((ptype->dev == dev || !ptype->dev) && | |
c0de08d0 | 1714 | (!skb_loop_sk(ptype, skb))) { |
71d9dec2 CG |
1715 | if (pt_prev) { |
1716 | deliver_skb(skb2, pt_prev, skb->dev); | |
1717 | pt_prev = ptype; | |
1718 | continue; | |
1719 | } | |
1720 | ||
1721 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1722 | if (!skb2) |
1723 | break; | |
1724 | ||
70978182 ED |
1725 | net_timestamp_set(skb2); |
1726 | ||
1da177e4 LT |
1727 | /* skb->nh should be correctly |
1728 | set by sender, so that the second statement is | |
1729 | just protection against buggy protocols. | |
1730 | */ | |
459a98ed | 1731 | skb_reset_mac_header(skb2); |
1da177e4 | 1732 | |
d56f90a7 | 1733 | if (skb_network_header(skb2) < skb2->data || |
ced14f68 | 1734 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { |
e87cc472 JP |
1735 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", |
1736 | ntohs(skb2->protocol), | |
1737 | dev->name); | |
c1d2bbe1 | 1738 | skb_reset_network_header(skb2); |
1da177e4 LT |
1739 | } |
1740 | ||
b0e380b1 | 1741 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1742 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1743 | pt_prev = ptype; |
1da177e4 LT |
1744 | } |
1745 | } | |
71d9dec2 CG |
1746 | if (pt_prev) |
1747 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1748 | rcu_read_unlock(); |
1749 | } | |
1750 | ||
2c53040f BH |
1751 | /** |
1752 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1753 | * @dev: Network device |
1754 | * @txq: number of queues available | |
1755 | * | |
1756 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1757 | * valid. To resolve this verify the tc mapping remains valid and if | |
1758 | * not NULL the mapping. With no priorities mapping to this | |
1759 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1760 | * is invalid nothing can be done so disable priority mappings. If is | |
1761 | * expected that drivers will fix this mapping if they can before | |
1762 | * calling netif_set_real_num_tx_queues. | |
1763 | */ | |
bb134d22 | 1764 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1765 | { |
1766 | int i; | |
1767 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1768 | ||
1769 | /* If TC0 is invalidated disable TC mapping */ | |
1770 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1771 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1772 | dev->num_tc = 0; |
1773 | return; | |
1774 | } | |
1775 | ||
1776 | /* Invalidated prio to tc mappings set to TC0 */ | |
1777 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1778 | int q = netdev_get_prio_tc_map(dev, i); | |
1779 | ||
1780 | tc = &dev->tc_to_txq[q]; | |
1781 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1782 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1783 | i, q); | |
4f57c087 JF |
1784 | netdev_set_prio_tc_map(dev, i, 0); |
1785 | } | |
1786 | } | |
1787 | } | |
1788 | ||
537c00de AD |
1789 | #ifdef CONFIG_XPS |
1790 | static DEFINE_MUTEX(xps_map_mutex); | |
1791 | #define xmap_dereference(P) \ | |
1792 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1793 | ||
10cdc3f3 AD |
1794 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1795 | int cpu, u16 index) | |
537c00de | 1796 | { |
10cdc3f3 AD |
1797 | struct xps_map *map = NULL; |
1798 | int pos; | |
537c00de | 1799 | |
10cdc3f3 AD |
1800 | if (dev_maps) |
1801 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1802 | |
10cdc3f3 AD |
1803 | for (pos = 0; map && pos < map->len; pos++) { |
1804 | if (map->queues[pos] == index) { | |
537c00de AD |
1805 | if (map->len > 1) { |
1806 | map->queues[pos] = map->queues[--map->len]; | |
1807 | } else { | |
10cdc3f3 | 1808 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1809 | kfree_rcu(map, rcu); |
1810 | map = NULL; | |
1811 | } | |
10cdc3f3 | 1812 | break; |
537c00de | 1813 | } |
537c00de AD |
1814 | } |
1815 | ||
10cdc3f3 AD |
1816 | return map; |
1817 | } | |
1818 | ||
024e9679 | 1819 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
10cdc3f3 AD |
1820 | { |
1821 | struct xps_dev_maps *dev_maps; | |
024e9679 | 1822 | int cpu, i; |
10cdc3f3 AD |
1823 | bool active = false; |
1824 | ||
1825 | mutex_lock(&xps_map_mutex); | |
1826 | dev_maps = xmap_dereference(dev->xps_maps); | |
1827 | ||
1828 | if (!dev_maps) | |
1829 | goto out_no_maps; | |
1830 | ||
1831 | for_each_possible_cpu(cpu) { | |
024e9679 AD |
1832 | for (i = index; i < dev->num_tx_queues; i++) { |
1833 | if (!remove_xps_queue(dev_maps, cpu, i)) | |
1834 | break; | |
1835 | } | |
1836 | if (i == dev->num_tx_queues) | |
10cdc3f3 AD |
1837 | active = true; |
1838 | } | |
1839 | ||
1840 | if (!active) { | |
537c00de AD |
1841 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
1842 | kfree_rcu(dev_maps, rcu); | |
1843 | } | |
1844 | ||
024e9679 AD |
1845 | for (i = index; i < dev->num_tx_queues; i++) |
1846 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), | |
1847 | NUMA_NO_NODE); | |
1848 | ||
537c00de AD |
1849 | out_no_maps: |
1850 | mutex_unlock(&xps_map_mutex); | |
1851 | } | |
1852 | ||
01c5f864 AD |
1853 | static struct xps_map *expand_xps_map(struct xps_map *map, |
1854 | int cpu, u16 index) | |
1855 | { | |
1856 | struct xps_map *new_map; | |
1857 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
1858 | int i, pos; | |
1859 | ||
1860 | for (pos = 0; map && pos < map->len; pos++) { | |
1861 | if (map->queues[pos] != index) | |
1862 | continue; | |
1863 | return map; | |
1864 | } | |
1865 | ||
1866 | /* Need to add queue to this CPU's existing map */ | |
1867 | if (map) { | |
1868 | if (pos < map->alloc_len) | |
1869 | return map; | |
1870 | ||
1871 | alloc_len = map->alloc_len * 2; | |
1872 | } | |
1873 | ||
1874 | /* Need to allocate new map to store queue on this CPU's map */ | |
1875 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
1876 | cpu_to_node(cpu)); | |
1877 | if (!new_map) | |
1878 | return NULL; | |
1879 | ||
1880 | for (i = 0; i < pos; i++) | |
1881 | new_map->queues[i] = map->queues[i]; | |
1882 | new_map->alloc_len = alloc_len; | |
1883 | new_map->len = pos; | |
1884 | ||
1885 | return new_map; | |
1886 | } | |
1887 | ||
537c00de AD |
1888 | int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index) |
1889 | { | |
01c5f864 | 1890 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 1891 | struct xps_map *map, *new_map; |
537c00de | 1892 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
1893 | int cpu, numa_node_id = -2; |
1894 | bool active = false; | |
537c00de AD |
1895 | |
1896 | mutex_lock(&xps_map_mutex); | |
1897 | ||
1898 | dev_maps = xmap_dereference(dev->xps_maps); | |
1899 | ||
01c5f864 AD |
1900 | /* allocate memory for queue storage */ |
1901 | for_each_online_cpu(cpu) { | |
1902 | if (!cpumask_test_cpu(cpu, mask)) | |
1903 | continue; | |
1904 | ||
1905 | if (!new_dev_maps) | |
1906 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
1907 | if (!new_dev_maps) { |
1908 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 1909 | return -ENOMEM; |
2bb60cb9 | 1910 | } |
01c5f864 AD |
1911 | |
1912 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
1913 | NULL; | |
1914 | ||
1915 | map = expand_xps_map(map, cpu, index); | |
1916 | if (!map) | |
1917 | goto error; | |
1918 | ||
1919 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
1920 | } | |
1921 | ||
1922 | if (!new_dev_maps) | |
1923 | goto out_no_new_maps; | |
1924 | ||
537c00de | 1925 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
1926 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
1927 | /* add queue to CPU maps */ | |
1928 | int pos = 0; | |
1929 | ||
1930 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1931 | while ((pos < map->len) && (map->queues[pos] != index)) | |
1932 | pos++; | |
1933 | ||
1934 | if (pos == map->len) | |
1935 | map->queues[map->len++] = index; | |
537c00de | 1936 | #ifdef CONFIG_NUMA |
537c00de AD |
1937 | if (numa_node_id == -2) |
1938 | numa_node_id = cpu_to_node(cpu); | |
1939 | else if (numa_node_id != cpu_to_node(cpu)) | |
1940 | numa_node_id = -1; | |
537c00de | 1941 | #endif |
01c5f864 AD |
1942 | } else if (dev_maps) { |
1943 | /* fill in the new device map from the old device map */ | |
1944 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
1945 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 1946 | } |
01c5f864 | 1947 | |
537c00de AD |
1948 | } |
1949 | ||
01c5f864 AD |
1950 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
1951 | ||
537c00de | 1952 | /* Cleanup old maps */ |
01c5f864 AD |
1953 | if (dev_maps) { |
1954 | for_each_possible_cpu(cpu) { | |
1955 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1956 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
1957 | if (map && map != new_map) | |
1958 | kfree_rcu(map, rcu); | |
1959 | } | |
537c00de | 1960 | |
01c5f864 | 1961 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
1962 | } |
1963 | ||
01c5f864 AD |
1964 | dev_maps = new_dev_maps; |
1965 | active = true; | |
537c00de | 1966 | |
01c5f864 AD |
1967 | out_no_new_maps: |
1968 | /* update Tx queue numa node */ | |
537c00de AD |
1969 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
1970 | (numa_node_id >= 0) ? numa_node_id : | |
1971 | NUMA_NO_NODE); | |
1972 | ||
01c5f864 AD |
1973 | if (!dev_maps) |
1974 | goto out_no_maps; | |
1975 | ||
1976 | /* removes queue from unused CPUs */ | |
1977 | for_each_possible_cpu(cpu) { | |
1978 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
1979 | continue; | |
1980 | ||
1981 | if (remove_xps_queue(dev_maps, cpu, index)) | |
1982 | active = true; | |
1983 | } | |
1984 | ||
1985 | /* free map if not active */ | |
1986 | if (!active) { | |
1987 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
1988 | kfree_rcu(dev_maps, rcu); | |
1989 | } | |
1990 | ||
1991 | out_no_maps: | |
537c00de AD |
1992 | mutex_unlock(&xps_map_mutex); |
1993 | ||
1994 | return 0; | |
1995 | error: | |
01c5f864 AD |
1996 | /* remove any maps that we added */ |
1997 | for_each_possible_cpu(cpu) { | |
1998 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1999 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2000 | NULL; | |
2001 | if (new_map && new_map != map) | |
2002 | kfree(new_map); | |
2003 | } | |
2004 | ||
537c00de AD |
2005 | mutex_unlock(&xps_map_mutex); |
2006 | ||
537c00de AD |
2007 | kfree(new_dev_maps); |
2008 | return -ENOMEM; | |
2009 | } | |
2010 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2011 | ||
2012 | #endif | |
f0796d5c JF |
2013 | /* |
2014 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2015 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2016 | */ | |
e6484930 | 2017 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2018 | { |
1d24eb48 TH |
2019 | int rc; |
2020 | ||
e6484930 TH |
2021 | if (txq < 1 || txq > dev->num_tx_queues) |
2022 | return -EINVAL; | |
f0796d5c | 2023 | |
5c56580b BH |
2024 | if (dev->reg_state == NETREG_REGISTERED || |
2025 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2026 | ASSERT_RTNL(); |
2027 | ||
1d24eb48 TH |
2028 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2029 | txq); | |
bf264145 TH |
2030 | if (rc) |
2031 | return rc; | |
2032 | ||
4f57c087 JF |
2033 | if (dev->num_tc) |
2034 | netif_setup_tc(dev, txq); | |
2035 | ||
024e9679 | 2036 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2037 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2038 | #ifdef CONFIG_XPS |
2039 | netif_reset_xps_queues_gt(dev, txq); | |
2040 | #endif | |
2041 | } | |
f0796d5c | 2042 | } |
e6484930 TH |
2043 | |
2044 | dev->real_num_tx_queues = txq; | |
2045 | return 0; | |
f0796d5c JF |
2046 | } |
2047 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2048 | |
62fe0b40 BH |
2049 | #ifdef CONFIG_RPS |
2050 | /** | |
2051 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2052 | * @dev: Network device | |
2053 | * @rxq: Actual number of RX queues | |
2054 | * | |
2055 | * This must be called either with the rtnl_lock held or before | |
2056 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2057 | * negative error code. If called before registration, it always |
2058 | * succeeds. | |
62fe0b40 BH |
2059 | */ |
2060 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2061 | { | |
2062 | int rc; | |
2063 | ||
bd25fa7b TH |
2064 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2065 | return -EINVAL; | |
2066 | ||
62fe0b40 BH |
2067 | if (dev->reg_state == NETREG_REGISTERED) { |
2068 | ASSERT_RTNL(); | |
2069 | ||
62fe0b40 BH |
2070 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2071 | rxq); | |
2072 | if (rc) | |
2073 | return rc; | |
62fe0b40 BH |
2074 | } |
2075 | ||
2076 | dev->real_num_rx_queues = rxq; | |
2077 | return 0; | |
2078 | } | |
2079 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2080 | #endif | |
2081 | ||
2c53040f BH |
2082 | /** |
2083 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2084 | * |
2085 | * This routine should set an upper limit on the number of RSS queues | |
2086 | * used by default by multiqueue devices. | |
2087 | */ | |
a55b138b | 2088 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
2089 | { |
2090 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
2091 | } | |
2092 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2093 | ||
def82a1d | 2094 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 2095 | { |
def82a1d JP |
2096 | struct softnet_data *sd; |
2097 | unsigned long flags; | |
56079431 | 2098 | |
def82a1d JP |
2099 | local_irq_save(flags); |
2100 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
2101 | q->next_sched = NULL; |
2102 | *sd->output_queue_tailp = q; | |
2103 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2104 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2105 | local_irq_restore(flags); | |
2106 | } | |
2107 | ||
2108 | void __netif_schedule(struct Qdisc *q) | |
2109 | { | |
2110 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2111 | __netif_reschedule(q); | |
56079431 DV |
2112 | } |
2113 | EXPORT_SYMBOL(__netif_schedule); | |
2114 | ||
bea3348e | 2115 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 2116 | { |
3578b0c8 | 2117 | if (atomic_dec_and_test(&skb->users)) { |
bea3348e SH |
2118 | struct softnet_data *sd; |
2119 | unsigned long flags; | |
56079431 | 2120 | |
bea3348e SH |
2121 | local_irq_save(flags); |
2122 | sd = &__get_cpu_var(softnet_data); | |
2123 | skb->next = sd->completion_queue; | |
2124 | sd->completion_queue = skb; | |
2125 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2126 | local_irq_restore(flags); | |
2127 | } | |
56079431 | 2128 | } |
bea3348e | 2129 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
2130 | |
2131 | void dev_kfree_skb_any(struct sk_buff *skb) | |
2132 | { | |
2133 | if (in_irq() || irqs_disabled()) | |
2134 | dev_kfree_skb_irq(skb); | |
2135 | else | |
2136 | dev_kfree_skb(skb); | |
2137 | } | |
2138 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
2139 | ||
2140 | ||
bea3348e SH |
2141 | /** |
2142 | * netif_device_detach - mark device as removed | |
2143 | * @dev: network device | |
2144 | * | |
2145 | * Mark device as removed from system and therefore no longer available. | |
2146 | */ | |
56079431 DV |
2147 | void netif_device_detach(struct net_device *dev) |
2148 | { | |
2149 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2150 | netif_running(dev)) { | |
d543103a | 2151 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2152 | } |
2153 | } | |
2154 | EXPORT_SYMBOL(netif_device_detach); | |
2155 | ||
bea3348e SH |
2156 | /** |
2157 | * netif_device_attach - mark device as attached | |
2158 | * @dev: network device | |
2159 | * | |
2160 | * Mark device as attached from system and restart if needed. | |
2161 | */ | |
56079431 DV |
2162 | void netif_device_attach(struct net_device *dev) |
2163 | { | |
2164 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2165 | netif_running(dev)) { | |
d543103a | 2166 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2167 | __netdev_watchdog_up(dev); |
56079431 DV |
2168 | } |
2169 | } | |
2170 | EXPORT_SYMBOL(netif_device_attach); | |
2171 | ||
36c92474 BH |
2172 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2173 | { | |
65e9d2fa | 2174 | static const netdev_features_t null_features = 0; |
36c92474 BH |
2175 | struct net_device *dev = skb->dev; |
2176 | const char *driver = ""; | |
2177 | ||
c846ad9b BG |
2178 | if (!net_ratelimit()) |
2179 | return; | |
2180 | ||
36c92474 BH |
2181 | if (dev && dev->dev.parent) |
2182 | driver = dev_driver_string(dev->dev.parent); | |
2183 | ||
2184 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " | |
2185 | "gso_type=%d ip_summed=%d\n", | |
65e9d2fa MM |
2186 | driver, dev ? &dev->features : &null_features, |
2187 | skb->sk ? &skb->sk->sk_route_caps : &null_features, | |
36c92474 BH |
2188 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2189 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2190 | } | |
2191 | ||
1da177e4 LT |
2192 | /* |
2193 | * Invalidate hardware checksum when packet is to be mangled, and | |
2194 | * complete checksum manually on outgoing path. | |
2195 | */ | |
84fa7933 | 2196 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2197 | { |
d3bc23e7 | 2198 | __wsum csum; |
663ead3b | 2199 | int ret = 0, offset; |
1da177e4 | 2200 | |
84fa7933 | 2201 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2202 | goto out_set_summed; |
2203 | ||
2204 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2205 | skb_warn_bad_offload(skb); |
2206 | return -EINVAL; | |
1da177e4 LT |
2207 | } |
2208 | ||
cef401de ED |
2209 | /* Before computing a checksum, we should make sure no frag could |
2210 | * be modified by an external entity : checksum could be wrong. | |
2211 | */ | |
2212 | if (skb_has_shared_frag(skb)) { | |
2213 | ret = __skb_linearize(skb); | |
2214 | if (ret) | |
2215 | goto out; | |
2216 | } | |
2217 | ||
55508d60 | 2218 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2219 | BUG_ON(offset >= skb_headlen(skb)); |
2220 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2221 | ||
2222 | offset += skb->csum_offset; | |
2223 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2224 | ||
2225 | if (skb_cloned(skb) && | |
2226 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2227 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2228 | if (ret) | |
2229 | goto out; | |
2230 | } | |
2231 | ||
a030847e | 2232 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2233 | out_set_summed: |
1da177e4 | 2234 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2235 | out: |
1da177e4 LT |
2236 | return ret; |
2237 | } | |
d1b19dff | 2238 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2239 | |
ec5f0615 | 2240 | __be16 skb_network_protocol(struct sk_buff *skb) |
f6a78bfc | 2241 | { |
252e3346 | 2242 | __be16 type = skb->protocol; |
c80a8512 | 2243 | int vlan_depth = ETH_HLEN; |
f6a78bfc | 2244 | |
19acc327 PS |
2245 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2246 | if (type == htons(ETH_P_TEB)) { | |
2247 | struct ethhdr *eth; | |
2248 | ||
2249 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2250 | return 0; | |
2251 | ||
2252 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2253 | type = eth->h_proto; | |
2254 | } | |
2255 | ||
8ad227ff | 2256 | while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) { |
c8d5bcd1 | 2257 | struct vlan_hdr *vh; |
7b9c6090 | 2258 | |
c8d5bcd1 | 2259 | if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN))) |
ec5f0615 | 2260 | return 0; |
7b9c6090 | 2261 | |
c8d5bcd1 JG |
2262 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); |
2263 | type = vh->h_vlan_encapsulated_proto; | |
2264 | vlan_depth += VLAN_HLEN; | |
7b9c6090 JG |
2265 | } |
2266 | ||
ec5f0615 PS |
2267 | return type; |
2268 | } | |
2269 | ||
2270 | /** | |
2271 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2272 | * @skb: buffer to segment | |
2273 | * @features: features for the output path (see dev->features) | |
2274 | */ | |
2275 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2276 | netdev_features_t features) | |
2277 | { | |
2278 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2279 | struct packet_offload *ptype; | |
2280 | __be16 type = skb_network_protocol(skb); | |
2281 | ||
2282 | if (unlikely(!type)) | |
2283 | return ERR_PTR(-EINVAL); | |
2284 | ||
f6a78bfc HX |
2285 | __skb_pull(skb, skb->mac_len); |
2286 | ||
2287 | rcu_read_lock(); | |
22061d80 | 2288 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2289 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
84fa7933 | 2290 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
05e8ef4a PS |
2291 | int err; |
2292 | ||
f191a1d1 | 2293 | err = ptype->callbacks.gso_send_check(skb); |
a430a43d HX |
2294 | segs = ERR_PTR(err); |
2295 | if (err || skb_gso_ok(skb, features)) | |
2296 | break; | |
d56f90a7 ACM |
2297 | __skb_push(skb, (skb->data - |
2298 | skb_network_header(skb))); | |
a430a43d | 2299 | } |
f191a1d1 | 2300 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2301 | break; |
2302 | } | |
2303 | } | |
2304 | rcu_read_unlock(); | |
2305 | ||
98e399f8 | 2306 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2307 | |
f6a78bfc HX |
2308 | return segs; |
2309 | } | |
05e8ef4a PS |
2310 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2311 | ||
2312 | ||
2313 | /* openvswitch calls this on rx path, so we need a different check. | |
2314 | */ | |
2315 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2316 | { | |
2317 | if (tx_path) | |
2318 | return skb->ip_summed != CHECKSUM_PARTIAL; | |
2319 | else | |
2320 | return skb->ip_summed == CHECKSUM_NONE; | |
2321 | } | |
2322 | ||
2323 | /** | |
2324 | * __skb_gso_segment - Perform segmentation on skb. | |
2325 | * @skb: buffer to segment | |
2326 | * @features: features for the output path (see dev->features) | |
2327 | * @tx_path: whether it is called in TX path | |
2328 | * | |
2329 | * This function segments the given skb and returns a list of segments. | |
2330 | * | |
2331 | * It may return NULL if the skb requires no segmentation. This is | |
2332 | * only possible when GSO is used for verifying header integrity. | |
2333 | */ | |
2334 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2335 | netdev_features_t features, bool tx_path) | |
2336 | { | |
2337 | if (unlikely(skb_needs_check(skb, tx_path))) { | |
2338 | int err; | |
2339 | ||
2340 | skb_warn_bad_offload(skb); | |
2341 | ||
2342 | if (skb_header_cloned(skb) && | |
2343 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
2344 | return ERR_PTR(err); | |
2345 | } | |
2346 | ||
68c33163 | 2347 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
05e8ef4a PS |
2348 | skb_reset_mac_header(skb); |
2349 | skb_reset_mac_len(skb); | |
2350 | ||
2351 | return skb_mac_gso_segment(skb, features); | |
2352 | } | |
12b0004d | 2353 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2354 | |
fb286bb2 HX |
2355 | /* Take action when hardware reception checksum errors are detected. */ |
2356 | #ifdef CONFIG_BUG | |
2357 | void netdev_rx_csum_fault(struct net_device *dev) | |
2358 | { | |
2359 | if (net_ratelimit()) { | |
7b6cd1ce | 2360 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2361 | dump_stack(); |
2362 | } | |
2363 | } | |
2364 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2365 | #endif | |
2366 | ||
1da177e4 LT |
2367 | /* Actually, we should eliminate this check as soon as we know, that: |
2368 | * 1. IOMMU is present and allows to map all the memory. | |
2369 | * 2. No high memory really exists on this machine. | |
2370 | */ | |
2371 | ||
9092c658 | 2372 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2373 | { |
3d3a8533 | 2374 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2375 | int i; |
5acbbd42 | 2376 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2377 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2378 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2379 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2380 | return 1; |
ea2ab693 | 2381 | } |
5acbbd42 | 2382 | } |
1da177e4 | 2383 | |
5acbbd42 FT |
2384 | if (PCI_DMA_BUS_IS_PHYS) { |
2385 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2386 | |
9092c658 ED |
2387 | if (!pdev) |
2388 | return 0; | |
5acbbd42 | 2389 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2390 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2391 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2392 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2393 | return 1; | |
2394 | } | |
2395 | } | |
3d3a8533 | 2396 | #endif |
1da177e4 LT |
2397 | return 0; |
2398 | } | |
1da177e4 | 2399 | |
f6a78bfc HX |
2400 | struct dev_gso_cb { |
2401 | void (*destructor)(struct sk_buff *skb); | |
2402 | }; | |
2403 | ||
2404 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
2405 | ||
2406 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
2407 | { | |
2408 | struct dev_gso_cb *cb; | |
2409 | ||
2410 | do { | |
2411 | struct sk_buff *nskb = skb->next; | |
2412 | ||
2413 | skb->next = nskb->next; | |
2414 | nskb->next = NULL; | |
2415 | kfree_skb(nskb); | |
2416 | } while (skb->next); | |
2417 | ||
2418 | cb = DEV_GSO_CB(skb); | |
2419 | if (cb->destructor) | |
2420 | cb->destructor(skb); | |
2421 | } | |
2422 | ||
2423 | /** | |
2424 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
2425 | * @skb: buffer to segment | |
91ecb63c | 2426 | * @features: device features as applicable to this skb |
f6a78bfc HX |
2427 | * |
2428 | * This function segments the given skb and stores the list of segments | |
2429 | * in skb->next. | |
2430 | */ | |
c8f44aff | 2431 | static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features) |
f6a78bfc | 2432 | { |
f6a78bfc | 2433 | struct sk_buff *segs; |
576a30eb HX |
2434 | |
2435 | segs = skb_gso_segment(skb, features); | |
2436 | ||
2437 | /* Verifying header integrity only. */ | |
2438 | if (!segs) | |
2439 | return 0; | |
f6a78bfc | 2440 | |
801678c5 | 2441 | if (IS_ERR(segs)) |
f6a78bfc HX |
2442 | return PTR_ERR(segs); |
2443 | ||
2444 | skb->next = segs; | |
2445 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
2446 | skb->destructor = dev_gso_skb_destructor; | |
2447 | ||
2448 | return 0; | |
2449 | } | |
2450 | ||
c8f44aff MM |
2451 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
2452 | __be16 protocol, netdev_features_t features) | |
f01a5236 | 2453 | { |
c0d680e5 EC |
2454 | if (skb->ip_summed != CHECKSUM_NONE && |
2455 | !can_checksum_protocol(features, protocol)) { | |
f01a5236 | 2456 | features &= ~NETIF_F_ALL_CSUM; |
f01a5236 JG |
2457 | } else if (illegal_highdma(skb->dev, skb)) { |
2458 | features &= ~NETIF_F_SG; | |
2459 | } | |
2460 | ||
2461 | return features; | |
2462 | } | |
2463 | ||
c8f44aff | 2464 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 JG |
2465 | { |
2466 | __be16 protocol = skb->protocol; | |
c8f44aff | 2467 | netdev_features_t features = skb->dev->features; |
58e998c6 | 2468 | |
30b678d8 BH |
2469 | if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs) |
2470 | features &= ~NETIF_F_GSO_MASK; | |
2471 | ||
8ad227ff | 2472 | if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) { |
58e998c6 JG |
2473 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; |
2474 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 JG |
2475 | } else if (!vlan_tx_tag_present(skb)) { |
2476 | return harmonize_features(skb, protocol, features); | |
2477 | } | |
58e998c6 | 2478 | |
8ad227ff PM |
2479 | features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX | |
2480 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2481 | |
8ad227ff | 2482 | if (protocol != htons(ETH_P_8021Q) && protocol != htons(ETH_P_8021AD)) { |
f01a5236 JG |
2483 | return harmonize_features(skb, protocol, features); |
2484 | } else { | |
2485 | features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | | |
8ad227ff PM |
2486 | NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX | |
2487 | NETIF_F_HW_VLAN_STAG_TX; | |
f01a5236 JG |
2488 | return harmonize_features(skb, protocol, features); |
2489 | } | |
58e998c6 | 2490 | } |
f01a5236 | 2491 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2492 | |
6afff0ca JF |
2493 | /* |
2494 | * Returns true if either: | |
2495 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
d1a53dfd | 2496 | * 2. skb is fragmented and the device does not support SG. |
6afff0ca JF |
2497 | */ |
2498 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
6708c9e5 | 2499 | netdev_features_t features) |
6afff0ca | 2500 | { |
02932ce9 JG |
2501 | return skb_is_nonlinear(skb) && |
2502 | ((skb_has_frag_list(skb) && | |
2503 | !(features & NETIF_F_FRAGLIST)) || | |
e1e78db6 | 2504 | (skb_shinfo(skb)->nr_frags && |
02932ce9 | 2505 | !(features & NETIF_F_SG))); |
6afff0ca JF |
2506 | } |
2507 | ||
fd2ea0a7 DM |
2508 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
2509 | struct netdev_queue *txq) | |
f6a78bfc | 2510 | { |
00829823 | 2511 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 2512 | int rc = NETDEV_TX_OK; |
ec764bf0 | 2513 | unsigned int skb_len; |
00829823 | 2514 | |
f6a78bfc | 2515 | if (likely(!skb->next)) { |
c8f44aff | 2516 | netdev_features_t features; |
fc741216 | 2517 | |
93f154b5 | 2518 | /* |
25985edc | 2519 | * If device doesn't need skb->dst, release it right now while |
93f154b5 ED |
2520 | * its hot in this cpu cache |
2521 | */ | |
adf30907 ED |
2522 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
2523 | skb_dst_drop(skb); | |
2524 | ||
fc741216 JG |
2525 | features = netif_skb_features(skb); |
2526 | ||
7b9c6090 | 2527 | if (vlan_tx_tag_present(skb) && |
86a9bad3 PM |
2528 | !vlan_hw_offload_capable(features, skb->vlan_proto)) { |
2529 | skb = __vlan_put_tag(skb, skb->vlan_proto, | |
2530 | vlan_tx_tag_get(skb)); | |
7b9c6090 JG |
2531 | if (unlikely(!skb)) |
2532 | goto out; | |
2533 | ||
2534 | skb->vlan_tci = 0; | |
2535 | } | |
2536 | ||
fc70fb64 AD |
2537 | /* If encapsulation offload request, verify we are testing |
2538 | * hardware encapsulation features instead of standard | |
2539 | * features for the netdev | |
2540 | */ | |
2541 | if (skb->encapsulation) | |
2542 | features &= dev->hw_enc_features; | |
2543 | ||
fc741216 | 2544 | if (netif_needs_gso(skb, features)) { |
91ecb63c | 2545 | if (unlikely(dev_gso_segment(skb, features))) |
9ccb8975 DM |
2546 | goto out_kfree_skb; |
2547 | if (skb->next) | |
2548 | goto gso; | |
6afff0ca | 2549 | } else { |
02932ce9 | 2550 | if (skb_needs_linearize(skb, features) && |
6afff0ca JF |
2551 | __skb_linearize(skb)) |
2552 | goto out_kfree_skb; | |
2553 | ||
2554 | /* If packet is not checksummed and device does not | |
2555 | * support checksumming for this protocol, complete | |
2556 | * checksumming here. | |
2557 | */ | |
2558 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
fc70fb64 AD |
2559 | if (skb->encapsulation) |
2560 | skb_set_inner_transport_header(skb, | |
2561 | skb_checksum_start_offset(skb)); | |
2562 | else | |
2563 | skb_set_transport_header(skb, | |
2564 | skb_checksum_start_offset(skb)); | |
03634668 | 2565 | if (!(features & NETIF_F_ALL_CSUM) && |
6afff0ca JF |
2566 | skb_checksum_help(skb)) |
2567 | goto out_kfree_skb; | |
2568 | } | |
9ccb8975 DM |
2569 | } |
2570 | ||
b40863c6 ED |
2571 | if (!list_empty(&ptype_all)) |
2572 | dev_queue_xmit_nit(skb, dev); | |
2573 | ||
ec764bf0 | 2574 | skb_len = skb->len; |
ac45f602 | 2575 | rc = ops->ndo_start_xmit(skb, dev); |
ec764bf0 | 2576 | trace_net_dev_xmit(skb, rc, dev, skb_len); |
ec634fe3 | 2577 | if (rc == NETDEV_TX_OK) |
08baf561 | 2578 | txq_trans_update(txq); |
ac45f602 | 2579 | return rc; |
f6a78bfc HX |
2580 | } |
2581 | ||
576a30eb | 2582 | gso: |
f6a78bfc HX |
2583 | do { |
2584 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
2585 | |
2586 | skb->next = nskb->next; | |
2587 | nskb->next = NULL; | |
068a2de5 | 2588 | |
b40863c6 ED |
2589 | if (!list_empty(&ptype_all)) |
2590 | dev_queue_xmit_nit(nskb, dev); | |
2591 | ||
ec764bf0 | 2592 | skb_len = nskb->len; |
00829823 | 2593 | rc = ops->ndo_start_xmit(nskb, dev); |
ec764bf0 | 2594 | trace_net_dev_xmit(nskb, rc, dev, skb_len); |
ec634fe3 | 2595 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
2596 | if (rc & ~NETDEV_TX_MASK) |
2597 | goto out_kfree_gso_skb; | |
f54d9e8d | 2598 | nskb->next = skb->next; |
f6a78bfc HX |
2599 | skb->next = nskb; |
2600 | return rc; | |
2601 | } | |
08baf561 | 2602 | txq_trans_update(txq); |
73466498 | 2603 | if (unlikely(netif_xmit_stopped(txq) && skb->next)) |
f54d9e8d | 2604 | return NETDEV_TX_BUSY; |
f6a78bfc | 2605 | } while (skb->next); |
4ec93edb | 2606 | |
572a9d7b | 2607 | out_kfree_gso_skb: |
0c772159 | 2608 | if (likely(skb->next == NULL)) { |
572a9d7b | 2609 | skb->destructor = DEV_GSO_CB(skb)->destructor; |
0c772159 SS |
2610 | consume_skb(skb); |
2611 | return rc; | |
2612 | } | |
f6a78bfc HX |
2613 | out_kfree_skb: |
2614 | kfree_skb(skb); | |
7b9c6090 | 2615 | out: |
572a9d7b | 2616 | return rc; |
f6a78bfc HX |
2617 | } |
2618 | ||
1def9238 ED |
2619 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
2620 | { | |
2621 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2622 | ||
2623 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2624 | ||
2625 | /* To get more precise estimation of bytes sent on wire, | |
2626 | * we add to pkt_len the headers size of all segments | |
2627 | */ | |
2628 | if (shinfo->gso_size) { | |
757b8b1d | 2629 | unsigned int hdr_len; |
15e5a030 | 2630 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 2631 | |
757b8b1d ED |
2632 | /* mac layer + network layer */ |
2633 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
2634 | ||
2635 | /* + transport layer */ | |
1def9238 ED |
2636 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
2637 | hdr_len += tcp_hdrlen(skb); | |
2638 | else | |
2639 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
2640 | |
2641 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
2642 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
2643 | shinfo->gso_size); | |
2644 | ||
2645 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
2646 | } |
2647 | } | |
2648 | ||
bbd8a0d3 KK |
2649 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2650 | struct net_device *dev, | |
2651 | struct netdev_queue *txq) | |
2652 | { | |
2653 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2654 | bool contended; |
bbd8a0d3 KK |
2655 | int rc; |
2656 | ||
1def9238 | 2657 | qdisc_pkt_len_init(skb); |
a2da570d | 2658 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
2659 | /* |
2660 | * Heuristic to force contended enqueues to serialize on a | |
2661 | * separate lock before trying to get qdisc main lock. | |
2662 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2663 | * and dequeue packets faster. | |
2664 | */ | |
a2da570d | 2665 | contended = qdisc_is_running(q); |
79640a4c ED |
2666 | if (unlikely(contended)) |
2667 | spin_lock(&q->busylock); | |
2668 | ||
bbd8a0d3 KK |
2669 | spin_lock(root_lock); |
2670 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2671 | kfree_skb(skb); | |
2672 | rc = NET_XMIT_DROP; | |
2673 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2674 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2675 | /* |
2676 | * This is a work-conserving queue; there are no old skbs | |
2677 | * waiting to be sent out; and the qdisc is not running - | |
2678 | * xmit the skb directly. | |
2679 | */ | |
7fee226a ED |
2680 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2681 | skb_dst_force(skb); | |
bfe0d029 | 2682 | |
bfe0d029 ED |
2683 | qdisc_bstats_update(q, skb); |
2684 | ||
79640a4c ED |
2685 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2686 | if (unlikely(contended)) { | |
2687 | spin_unlock(&q->busylock); | |
2688 | contended = false; | |
2689 | } | |
bbd8a0d3 | 2690 | __qdisc_run(q); |
79640a4c | 2691 | } else |
bc135b23 | 2692 | qdisc_run_end(q); |
bbd8a0d3 KK |
2693 | |
2694 | rc = NET_XMIT_SUCCESS; | |
2695 | } else { | |
7fee226a | 2696 | skb_dst_force(skb); |
a2da570d | 2697 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2698 | if (qdisc_run_begin(q)) { |
2699 | if (unlikely(contended)) { | |
2700 | spin_unlock(&q->busylock); | |
2701 | contended = false; | |
2702 | } | |
2703 | __qdisc_run(q); | |
2704 | } | |
bbd8a0d3 KK |
2705 | } |
2706 | spin_unlock(root_lock); | |
79640a4c ED |
2707 | if (unlikely(contended)) |
2708 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2709 | return rc; |
2710 | } | |
2711 | ||
5bc1421e NH |
2712 | #if IS_ENABLED(CONFIG_NETPRIO_CGROUP) |
2713 | static void skb_update_prio(struct sk_buff *skb) | |
2714 | { | |
6977a79d | 2715 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 2716 | |
91c68ce2 ED |
2717 | if (!skb->priority && skb->sk && map) { |
2718 | unsigned int prioidx = skb->sk->sk_cgrp_prioidx; | |
2719 | ||
2720 | if (prioidx < map->priomap_len) | |
2721 | skb->priority = map->priomap[prioidx]; | |
2722 | } | |
5bc1421e NH |
2723 | } |
2724 | #else | |
2725 | #define skb_update_prio(skb) | |
2726 | #endif | |
2727 | ||
745e20f1 | 2728 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2729 | #define RECURSION_LIMIT 10 |
745e20f1 | 2730 | |
95603e22 MM |
2731 | /** |
2732 | * dev_loopback_xmit - loop back @skb | |
2733 | * @skb: buffer to transmit | |
2734 | */ | |
2735 | int dev_loopback_xmit(struct sk_buff *skb) | |
2736 | { | |
2737 | skb_reset_mac_header(skb); | |
2738 | __skb_pull(skb, skb_network_offset(skb)); | |
2739 | skb->pkt_type = PACKET_LOOPBACK; | |
2740 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2741 | WARN_ON(!skb_dst(skb)); | |
2742 | skb_dst_force(skb); | |
2743 | netif_rx_ni(skb); | |
2744 | return 0; | |
2745 | } | |
2746 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2747 | ||
d29f749e DJ |
2748 | /** |
2749 | * dev_queue_xmit - transmit a buffer | |
2750 | * @skb: buffer to transmit | |
2751 | * | |
2752 | * Queue a buffer for transmission to a network device. The caller must | |
2753 | * have set the device and priority and built the buffer before calling | |
2754 | * this function. The function can be called from an interrupt. | |
2755 | * | |
2756 | * A negative errno code is returned on a failure. A success does not | |
2757 | * guarantee the frame will be transmitted as it may be dropped due | |
2758 | * to congestion or traffic shaping. | |
2759 | * | |
2760 | * ----------------------------------------------------------------------------------- | |
2761 | * I notice this method can also return errors from the queue disciplines, | |
2762 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2763 | * be positive. | |
2764 | * | |
2765 | * Regardless of the return value, the skb is consumed, so it is currently | |
2766 | * difficult to retry a send to this method. (You can bump the ref count | |
2767 | * before sending to hold a reference for retry if you are careful.) | |
2768 | * | |
2769 | * When calling this method, interrupts MUST be enabled. This is because | |
2770 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2771 | * --BLG | |
2772 | */ | |
1da177e4 LT |
2773 | int dev_queue_xmit(struct sk_buff *skb) |
2774 | { | |
2775 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2776 | struct netdev_queue *txq; |
1da177e4 LT |
2777 | struct Qdisc *q; |
2778 | int rc = -ENOMEM; | |
2779 | ||
6d1ccff6 ED |
2780 | skb_reset_mac_header(skb); |
2781 | ||
4ec93edb YH |
2782 | /* Disable soft irqs for various locks below. Also |
2783 | * stops preemption for RCU. | |
1da177e4 | 2784 | */ |
4ec93edb | 2785 | rcu_read_lock_bh(); |
1da177e4 | 2786 | |
5bc1421e NH |
2787 | skb_update_prio(skb); |
2788 | ||
8c4c49df | 2789 | txq = netdev_pick_tx(dev, skb); |
a898def2 | 2790 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2791 | |
1da177e4 | 2792 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2793 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2794 | #endif |
cf66ba58 | 2795 | trace_net_dev_queue(skb); |
1da177e4 | 2796 | if (q->enqueue) { |
bbd8a0d3 | 2797 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2798 | goto out; |
1da177e4 LT |
2799 | } |
2800 | ||
2801 | /* The device has no queue. Common case for software devices: | |
2802 | loopback, all the sorts of tunnels... | |
2803 | ||
932ff279 HX |
2804 | Really, it is unlikely that netif_tx_lock protection is necessary |
2805 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2806 | counters.) |
2807 | However, it is possible, that they rely on protection | |
2808 | made by us here. | |
2809 | ||
2810 | Check this and shot the lock. It is not prone from deadlocks. | |
2811 | Either shot noqueue qdisc, it is even simpler 8) | |
2812 | */ | |
2813 | if (dev->flags & IFF_UP) { | |
2814 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2815 | ||
c773e847 | 2816 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2817 | |
745e20f1 ED |
2818 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2819 | goto recursion_alert; | |
2820 | ||
c773e847 | 2821 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2822 | |
73466498 | 2823 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 2824 | __this_cpu_inc(xmit_recursion); |
572a9d7b | 2825 | rc = dev_hard_start_xmit(skb, dev, txq); |
745e20f1 | 2826 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2827 | if (dev_xmit_complete(rc)) { |
c773e847 | 2828 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2829 | goto out; |
2830 | } | |
2831 | } | |
c773e847 | 2832 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
2833 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
2834 | dev->name); | |
1da177e4 LT |
2835 | } else { |
2836 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2837 | * unfortunately |
2838 | */ | |
2839 | recursion_alert: | |
e87cc472 JP |
2840 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
2841 | dev->name); | |
1da177e4 LT |
2842 | } |
2843 | } | |
2844 | ||
2845 | rc = -ENETDOWN; | |
d4828d85 | 2846 | rcu_read_unlock_bh(); |
1da177e4 | 2847 | |
1da177e4 LT |
2848 | kfree_skb(skb); |
2849 | return rc; | |
2850 | out: | |
d4828d85 | 2851 | rcu_read_unlock_bh(); |
1da177e4 LT |
2852 | return rc; |
2853 | } | |
d1b19dff | 2854 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2855 | |
2856 | ||
2857 | /*======================================================================= | |
2858 | Receiver routines | |
2859 | =======================================================================*/ | |
2860 | ||
6b2bedc3 | 2861 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
2862 | EXPORT_SYMBOL(netdev_max_backlog); |
2863 | ||
3b098e2d | 2864 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2865 | int netdev_budget __read_mostly = 300; |
2866 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2867 | |
eecfd7c4 ED |
2868 | /* Called with irq disabled */ |
2869 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2870 | struct napi_struct *napi) | |
2871 | { | |
2872 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2873 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2874 | } | |
2875 | ||
bfb564e7 KK |
2876 | #ifdef CONFIG_RPS |
2877 | ||
2878 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 2879 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
2880 | EXPORT_SYMBOL(rps_sock_flow_table); |
2881 | ||
c5905afb | 2882 | struct static_key rps_needed __read_mostly; |
adc9300e | 2883 | |
c445477d BH |
2884 | static struct rps_dev_flow * |
2885 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2886 | struct rps_dev_flow *rflow, u16 next_cpu) | |
2887 | { | |
09994d1b | 2888 | if (next_cpu != RPS_NO_CPU) { |
c445477d BH |
2889 | #ifdef CONFIG_RFS_ACCEL |
2890 | struct netdev_rx_queue *rxqueue; | |
2891 | struct rps_dev_flow_table *flow_table; | |
2892 | struct rps_dev_flow *old_rflow; | |
2893 | u32 flow_id; | |
2894 | u16 rxq_index; | |
2895 | int rc; | |
2896 | ||
2897 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
2898 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
2899 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
2900 | goto out; |
2901 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
2902 | if (rxq_index == skb_get_rx_queue(skb)) | |
2903 | goto out; | |
2904 | ||
2905 | rxqueue = dev->_rx + rxq_index; | |
2906 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2907 | if (!flow_table) | |
2908 | goto out; | |
2909 | flow_id = skb->rxhash & flow_table->mask; | |
2910 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, | |
2911 | rxq_index, flow_id); | |
2912 | if (rc < 0) | |
2913 | goto out; | |
2914 | old_rflow = rflow; | |
2915 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
2916 | rflow->filter = rc; |
2917 | if (old_rflow->filter == rflow->filter) | |
2918 | old_rflow->filter = RPS_NO_FILTER; | |
2919 | out: | |
2920 | #endif | |
2921 | rflow->last_qtail = | |
09994d1b | 2922 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
2923 | } |
2924 | ||
09994d1b | 2925 | rflow->cpu = next_cpu; |
c445477d BH |
2926 | return rflow; |
2927 | } | |
2928 | ||
bfb564e7 KK |
2929 | /* |
2930 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2931 | * CPU from the RPS map of the receiving queue for a given skb. | |
2932 | * rcu_read_lock must be held on entry. | |
2933 | */ | |
2934 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2935 | struct rps_dev_flow **rflowp) | |
2936 | { | |
2937 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 2938 | struct rps_map *map; |
bfb564e7 KK |
2939 | struct rps_dev_flow_table *flow_table; |
2940 | struct rps_sock_flow_table *sock_flow_table; | |
2941 | int cpu = -1; | |
2942 | u16 tcpu; | |
2943 | ||
2944 | if (skb_rx_queue_recorded(skb)) { | |
2945 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
2946 | if (unlikely(index >= dev->real_num_rx_queues)) { |
2947 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
2948 | "%s received packet on queue %u, but number " | |
2949 | "of RX queues is %u\n", | |
2950 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
2951 | goto done; |
2952 | } | |
2953 | rxqueue = dev->_rx + index; | |
2954 | } else | |
2955 | rxqueue = dev->_rx; | |
2956 | ||
6e3f7faf ED |
2957 | map = rcu_dereference(rxqueue->rps_map); |
2958 | if (map) { | |
85875236 | 2959 | if (map->len == 1 && |
33d480ce | 2960 | !rcu_access_pointer(rxqueue->rps_flow_table)) { |
6febfca9 CG |
2961 | tcpu = map->cpus[0]; |
2962 | if (cpu_online(tcpu)) | |
2963 | cpu = tcpu; | |
2964 | goto done; | |
2965 | } | |
33d480ce | 2966 | } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) { |
bfb564e7 | 2967 | goto done; |
6febfca9 | 2968 | } |
bfb564e7 | 2969 | |
2d47b459 | 2970 | skb_reset_network_header(skb); |
bfb564e7 KK |
2971 | if (!skb_get_rxhash(skb)) |
2972 | goto done; | |
2973 | ||
fec5e652 TH |
2974 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
2975 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
2976 | if (flow_table && sock_flow_table) { | |
2977 | u16 next_cpu; | |
2978 | struct rps_dev_flow *rflow; | |
2979 | ||
2980 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
2981 | tcpu = rflow->cpu; | |
2982 | ||
2983 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
2984 | sock_flow_table->mask]; | |
2985 | ||
2986 | /* | |
2987 | * If the desired CPU (where last recvmsg was done) is | |
2988 | * different from current CPU (one in the rx-queue flow | |
2989 | * table entry), switch if one of the following holds: | |
2990 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
2991 | * - Current CPU is offline. | |
2992 | * - The current CPU's queue tail has advanced beyond the | |
2993 | * last packet that was enqueued using this table entry. | |
2994 | * This guarantees that all previous packets for the flow | |
2995 | * have been dequeued, thus preserving in order delivery. | |
2996 | */ | |
2997 | if (unlikely(tcpu != next_cpu) && | |
2998 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
2999 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
baefa31d TH |
3000 | rflow->last_qtail)) >= 0)) { |
3001 | tcpu = next_cpu; | |
c445477d | 3002 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3003 | } |
c445477d | 3004 | |
fec5e652 TH |
3005 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
3006 | *rflowp = rflow; | |
3007 | cpu = tcpu; | |
3008 | goto done; | |
3009 | } | |
3010 | } | |
3011 | ||
0a9627f2 | 3012 | if (map) { |
fec5e652 | 3013 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; |
0a9627f2 TH |
3014 | |
3015 | if (cpu_online(tcpu)) { | |
3016 | cpu = tcpu; | |
3017 | goto done; | |
3018 | } | |
3019 | } | |
3020 | ||
3021 | done: | |
0a9627f2 TH |
3022 | return cpu; |
3023 | } | |
3024 | ||
c445477d BH |
3025 | #ifdef CONFIG_RFS_ACCEL |
3026 | ||
3027 | /** | |
3028 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3029 | * @dev: Device on which the filter was set | |
3030 | * @rxq_index: RX queue index | |
3031 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3032 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3033 | * | |
3034 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3035 | * this function for each installed filter and remove the filters for | |
3036 | * which it returns %true. | |
3037 | */ | |
3038 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3039 | u32 flow_id, u16 filter_id) | |
3040 | { | |
3041 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3042 | struct rps_dev_flow_table *flow_table; | |
3043 | struct rps_dev_flow *rflow; | |
3044 | bool expire = true; | |
3045 | int cpu; | |
3046 | ||
3047 | rcu_read_lock(); | |
3048 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3049 | if (flow_table && flow_id <= flow_table->mask) { | |
3050 | rflow = &flow_table->flows[flow_id]; | |
3051 | cpu = ACCESS_ONCE(rflow->cpu); | |
3052 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
3053 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
3054 | rflow->last_qtail) < | |
3055 | (int)(10 * flow_table->mask))) | |
3056 | expire = false; | |
3057 | } | |
3058 | rcu_read_unlock(); | |
3059 | return expire; | |
3060 | } | |
3061 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3062 | ||
3063 | #endif /* CONFIG_RFS_ACCEL */ | |
3064 | ||
0a9627f2 | 3065 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3066 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3067 | { |
e36fa2f7 ED |
3068 | struct softnet_data *sd = data; |
3069 | ||
eecfd7c4 | 3070 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3071 | sd->received_rps++; |
0a9627f2 | 3072 | } |
e36fa2f7 | 3073 | |
fec5e652 | 3074 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3075 | |
e36fa2f7 ED |
3076 | /* |
3077 | * Check if this softnet_data structure is another cpu one | |
3078 | * If yes, queue it to our IPI list and return 1 | |
3079 | * If no, return 0 | |
3080 | */ | |
3081 | static int rps_ipi_queued(struct softnet_data *sd) | |
3082 | { | |
3083 | #ifdef CONFIG_RPS | |
3084 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
3085 | ||
3086 | if (sd != mysd) { | |
3087 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3088 | mysd->rps_ipi_list = sd; | |
3089 | ||
3090 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3091 | return 1; | |
3092 | } | |
3093 | #endif /* CONFIG_RPS */ | |
3094 | return 0; | |
3095 | } | |
3096 | ||
99bbc707 WB |
3097 | #ifdef CONFIG_NET_FLOW_LIMIT |
3098 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3099 | #endif | |
3100 | ||
3101 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3102 | { | |
3103 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3104 | struct sd_flow_limit *fl; | |
3105 | struct softnet_data *sd; | |
3106 | unsigned int old_flow, new_flow; | |
3107 | ||
3108 | if (qlen < (netdev_max_backlog >> 1)) | |
3109 | return false; | |
3110 | ||
3111 | sd = &__get_cpu_var(softnet_data); | |
3112 | ||
3113 | rcu_read_lock(); | |
3114 | fl = rcu_dereference(sd->flow_limit); | |
3115 | if (fl) { | |
3116 | new_flow = skb_get_rxhash(skb) & (fl->num_buckets - 1); | |
3117 | old_flow = fl->history[fl->history_head]; | |
3118 | fl->history[fl->history_head] = new_flow; | |
3119 | ||
3120 | fl->history_head++; | |
3121 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3122 | ||
3123 | if (likely(fl->buckets[old_flow])) | |
3124 | fl->buckets[old_flow]--; | |
3125 | ||
3126 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3127 | fl->count++; | |
3128 | rcu_read_unlock(); | |
3129 | return true; | |
3130 | } | |
3131 | } | |
3132 | rcu_read_unlock(); | |
3133 | #endif | |
3134 | return false; | |
3135 | } | |
3136 | ||
0a9627f2 TH |
3137 | /* |
3138 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3139 | * queue (may be a remote CPU queue). | |
3140 | */ | |
fec5e652 TH |
3141 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3142 | unsigned int *qtail) | |
0a9627f2 | 3143 | { |
e36fa2f7 | 3144 | struct softnet_data *sd; |
0a9627f2 | 3145 | unsigned long flags; |
99bbc707 | 3146 | unsigned int qlen; |
0a9627f2 | 3147 | |
e36fa2f7 | 3148 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3149 | |
3150 | local_irq_save(flags); | |
0a9627f2 | 3151 | |
e36fa2f7 | 3152 | rps_lock(sd); |
99bbc707 WB |
3153 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3154 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
6e7676c1 | 3155 | if (skb_queue_len(&sd->input_pkt_queue)) { |
0a9627f2 | 3156 | enqueue: |
e36fa2f7 | 3157 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3158 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3159 | rps_unlock(sd); |
152102c7 | 3160 | local_irq_restore(flags); |
0a9627f2 TH |
3161 | return NET_RX_SUCCESS; |
3162 | } | |
3163 | ||
ebda37c2 ED |
3164 | /* Schedule NAPI for backlog device |
3165 | * We can use non atomic operation since we own the queue lock | |
3166 | */ | |
3167 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3168 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3169 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3170 | } |
3171 | goto enqueue; | |
3172 | } | |
3173 | ||
dee42870 | 3174 | sd->dropped++; |
e36fa2f7 | 3175 | rps_unlock(sd); |
0a9627f2 | 3176 | |
0a9627f2 TH |
3177 | local_irq_restore(flags); |
3178 | ||
caf586e5 | 3179 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3180 | kfree_skb(skb); |
3181 | return NET_RX_DROP; | |
3182 | } | |
1da177e4 | 3183 | |
1da177e4 LT |
3184 | /** |
3185 | * netif_rx - post buffer to the network code | |
3186 | * @skb: buffer to post | |
3187 | * | |
3188 | * This function receives a packet from a device driver and queues it for | |
3189 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3190 | * may be dropped during processing for congestion control or by the | |
3191 | * protocol layers. | |
3192 | * | |
3193 | * return values: | |
3194 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
3195 | * NET_RX_DROP (packet was dropped) |
3196 | * | |
3197 | */ | |
3198 | ||
3199 | int netif_rx(struct sk_buff *skb) | |
3200 | { | |
b0e28f1e | 3201 | int ret; |
1da177e4 LT |
3202 | |
3203 | /* if netpoll wants it, pretend we never saw it */ | |
3204 | if (netpoll_rx(skb)) | |
3205 | return NET_RX_DROP; | |
3206 | ||
588f0330 | 3207 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3208 | |
cf66ba58 | 3209 | trace_netif_rx(skb); |
df334545 | 3210 | #ifdef CONFIG_RPS |
c5905afb | 3211 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3212 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3213 | int cpu; |
3214 | ||
cece1945 | 3215 | preempt_disable(); |
b0e28f1e | 3216 | rcu_read_lock(); |
fec5e652 TH |
3217 | |
3218 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3219 | if (cpu < 0) |
3220 | cpu = smp_processor_id(); | |
fec5e652 TH |
3221 | |
3222 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3223 | ||
b0e28f1e | 3224 | rcu_read_unlock(); |
cece1945 | 3225 | preempt_enable(); |
adc9300e ED |
3226 | } else |
3227 | #endif | |
fec5e652 TH |
3228 | { |
3229 | unsigned int qtail; | |
3230 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3231 | put_cpu(); | |
3232 | } | |
b0e28f1e | 3233 | return ret; |
1da177e4 | 3234 | } |
d1b19dff | 3235 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3236 | |
3237 | int netif_rx_ni(struct sk_buff *skb) | |
3238 | { | |
3239 | int err; | |
3240 | ||
3241 | preempt_disable(); | |
3242 | err = netif_rx(skb); | |
3243 | if (local_softirq_pending()) | |
3244 | do_softirq(); | |
3245 | preempt_enable(); | |
3246 | ||
3247 | return err; | |
3248 | } | |
1da177e4 LT |
3249 | EXPORT_SYMBOL(netif_rx_ni); |
3250 | ||
1da177e4 LT |
3251 | static void net_tx_action(struct softirq_action *h) |
3252 | { | |
3253 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
3254 | ||
3255 | if (sd->completion_queue) { | |
3256 | struct sk_buff *clist; | |
3257 | ||
3258 | local_irq_disable(); | |
3259 | clist = sd->completion_queue; | |
3260 | sd->completion_queue = NULL; | |
3261 | local_irq_enable(); | |
3262 | ||
3263 | while (clist) { | |
3264 | struct sk_buff *skb = clist; | |
3265 | clist = clist->next; | |
3266 | ||
547b792c | 3267 | WARN_ON(atomic_read(&skb->users)); |
07dc22e7 | 3268 | trace_kfree_skb(skb, net_tx_action); |
1da177e4 LT |
3269 | __kfree_skb(skb); |
3270 | } | |
3271 | } | |
3272 | ||
3273 | if (sd->output_queue) { | |
37437bb2 | 3274 | struct Qdisc *head; |
1da177e4 LT |
3275 | |
3276 | local_irq_disable(); | |
3277 | head = sd->output_queue; | |
3278 | sd->output_queue = NULL; | |
a9cbd588 | 3279 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3280 | local_irq_enable(); |
3281 | ||
3282 | while (head) { | |
37437bb2 DM |
3283 | struct Qdisc *q = head; |
3284 | spinlock_t *root_lock; | |
3285 | ||
1da177e4 LT |
3286 | head = head->next_sched; |
3287 | ||
5fb66229 | 3288 | root_lock = qdisc_lock(q); |
37437bb2 | 3289 | if (spin_trylock(root_lock)) { |
def82a1d JP |
3290 | smp_mb__before_clear_bit(); |
3291 | clear_bit(__QDISC_STATE_SCHED, | |
3292 | &q->state); | |
37437bb2 DM |
3293 | qdisc_run(q); |
3294 | spin_unlock(root_lock); | |
1da177e4 | 3295 | } else { |
195648bb | 3296 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3297 | &q->state)) { |
195648bb | 3298 | __netif_reschedule(q); |
e8a83e10 JP |
3299 | } else { |
3300 | smp_mb__before_clear_bit(); | |
3301 | clear_bit(__QDISC_STATE_SCHED, | |
3302 | &q->state); | |
3303 | } | |
1da177e4 LT |
3304 | } |
3305 | } | |
3306 | } | |
3307 | } | |
3308 | ||
ab95bfe0 JP |
3309 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3310 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3311 | /* This hook is defined here for ATM LANE */ |
3312 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3313 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3314 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3315 | #endif |
1da177e4 | 3316 | |
1da177e4 LT |
3317 | #ifdef CONFIG_NET_CLS_ACT |
3318 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3319 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3320 | * a compare and 2 stores extra right now if we dont have it on | |
3321 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3322 | * NOTE: This doesn't stop any functionality; if you dont have |
3323 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3324 | * |
3325 | */ | |
24824a09 | 3326 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3327 | { |
1da177e4 | 3328 | struct net_device *dev = skb->dev; |
f697c3e8 | 3329 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3330 | int result = TC_ACT_OK; |
3331 | struct Qdisc *q; | |
4ec93edb | 3332 | |
de384830 | 3333 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
e87cc472 JP |
3334 | net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n", |
3335 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3336 | return TC_ACT_SHOT; |
3337 | } | |
1da177e4 | 3338 | |
f697c3e8 HX |
3339 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3340 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3341 | |
83874000 | 3342 | q = rxq->qdisc; |
8d50b53d | 3343 | if (q != &noop_qdisc) { |
83874000 | 3344 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3345 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3346 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3347 | spin_unlock(qdisc_lock(q)); |
3348 | } | |
f697c3e8 HX |
3349 | |
3350 | return result; | |
3351 | } | |
86e65da9 | 3352 | |
f697c3e8 HX |
3353 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3354 | struct packet_type **pt_prev, | |
3355 | int *ret, struct net_device *orig_dev) | |
3356 | { | |
24824a09 ED |
3357 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3358 | ||
3359 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
f697c3e8 | 3360 | goto out; |
1da177e4 | 3361 | |
f697c3e8 HX |
3362 | if (*pt_prev) { |
3363 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3364 | *pt_prev = NULL; | |
1da177e4 LT |
3365 | } |
3366 | ||
24824a09 | 3367 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3368 | case TC_ACT_SHOT: |
3369 | case TC_ACT_STOLEN: | |
3370 | kfree_skb(skb); | |
3371 | return NULL; | |
3372 | } | |
3373 | ||
3374 | out: | |
3375 | skb->tc_verd = 0; | |
3376 | return skb; | |
1da177e4 LT |
3377 | } |
3378 | #endif | |
3379 | ||
ab95bfe0 JP |
3380 | /** |
3381 | * netdev_rx_handler_register - register receive handler | |
3382 | * @dev: device to register a handler for | |
3383 | * @rx_handler: receive handler to register | |
93e2c32b | 3384 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 JP |
3385 | * |
3386 | * Register a receive hander for a device. This handler will then be | |
3387 | * called from __netif_receive_skb. A negative errno code is returned | |
3388 | * on a failure. | |
3389 | * | |
3390 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3391 | * |
3392 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3393 | */ |
3394 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3395 | rx_handler_func_t *rx_handler, |
3396 | void *rx_handler_data) | |
ab95bfe0 JP |
3397 | { |
3398 | ASSERT_RTNL(); | |
3399 | ||
3400 | if (dev->rx_handler) | |
3401 | return -EBUSY; | |
3402 | ||
00cfec37 | 3403 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 3404 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3405 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3406 | ||
3407 | return 0; | |
3408 | } | |
3409 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3410 | ||
3411 | /** | |
3412 | * netdev_rx_handler_unregister - unregister receive handler | |
3413 | * @dev: device to unregister a handler from | |
3414 | * | |
166ec369 | 3415 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
3416 | * |
3417 | * The caller must hold the rtnl_mutex. | |
3418 | */ | |
3419 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3420 | { | |
3421 | ||
3422 | ASSERT_RTNL(); | |
a9b3cd7f | 3423 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
3424 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
3425 | * section has a guarantee to see a non NULL rx_handler_data | |
3426 | * as well. | |
3427 | */ | |
3428 | synchronize_net(); | |
a9b3cd7f | 3429 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
3430 | } |
3431 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3432 | ||
b4b9e355 MG |
3433 | /* |
3434 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3435 | * the special handling of PFMEMALLOC skbs. | |
3436 | */ | |
3437 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3438 | { | |
3439 | switch (skb->protocol) { | |
3440 | case __constant_htons(ETH_P_ARP): | |
3441 | case __constant_htons(ETH_P_IP): | |
3442 | case __constant_htons(ETH_P_IPV6): | |
3443 | case __constant_htons(ETH_P_8021Q): | |
8ad227ff | 3444 | case __constant_htons(ETH_P_8021AD): |
b4b9e355 MG |
3445 | return true; |
3446 | default: | |
3447 | return false; | |
3448 | } | |
3449 | } | |
3450 | ||
9754e293 | 3451 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
3452 | { |
3453 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3454 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3455 | struct net_device *orig_dev; |
63d8ea7f | 3456 | struct net_device *null_or_dev; |
8a4eb573 | 3457 | bool deliver_exact = false; |
1da177e4 | 3458 | int ret = NET_RX_DROP; |
252e3346 | 3459 | __be16 type; |
1da177e4 | 3460 | |
588f0330 | 3461 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3462 | |
cf66ba58 | 3463 | trace_netif_receive_skb(skb); |
9b22ea56 | 3464 | |
1da177e4 | 3465 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 3466 | if (netpoll_receive_skb(skb)) |
b4b9e355 | 3467 | goto out; |
1da177e4 | 3468 | |
cc9bd5ce | 3469 | orig_dev = skb->dev; |
8f903c70 | 3470 | |
c1d2bbe1 | 3471 | skb_reset_network_header(skb); |
fda55eca ED |
3472 | if (!skb_transport_header_was_set(skb)) |
3473 | skb_reset_transport_header(skb); | |
0b5c9db1 | 3474 | skb_reset_mac_len(skb); |
1da177e4 LT |
3475 | |
3476 | pt_prev = NULL; | |
3477 | ||
3478 | rcu_read_lock(); | |
3479 | ||
63d8ea7f | 3480 | another_round: |
b6858177 | 3481 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
3482 | |
3483 | __this_cpu_inc(softnet_data.processed); | |
3484 | ||
8ad227ff PM |
3485 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
3486 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
bcc6d479 JP |
3487 | skb = vlan_untag(skb); |
3488 | if (unlikely(!skb)) | |
b4b9e355 | 3489 | goto unlock; |
bcc6d479 JP |
3490 | } |
3491 | ||
1da177e4 LT |
3492 | #ifdef CONFIG_NET_CLS_ACT |
3493 | if (skb->tc_verd & TC_NCLS) { | |
3494 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3495 | goto ncls; | |
3496 | } | |
3497 | #endif | |
3498 | ||
9754e293 | 3499 | if (pfmemalloc) |
b4b9e355 MG |
3500 | goto skip_taps; |
3501 | ||
1da177e4 | 3502 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
63d8ea7f | 3503 | if (!ptype->dev || ptype->dev == skb->dev) { |
4ec93edb | 3504 | if (pt_prev) |
f2ccd8fa | 3505 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3506 | pt_prev = ptype; |
3507 | } | |
3508 | } | |
3509 | ||
b4b9e355 | 3510 | skip_taps: |
1da177e4 | 3511 | #ifdef CONFIG_NET_CLS_ACT |
f697c3e8 HX |
3512 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3513 | if (!skb) | |
b4b9e355 | 3514 | goto unlock; |
1da177e4 LT |
3515 | ncls: |
3516 | #endif | |
3517 | ||
9754e293 | 3518 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
3519 | goto drop; |
3520 | ||
2425717b JF |
3521 | if (vlan_tx_tag_present(skb)) { |
3522 | if (pt_prev) { | |
3523 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3524 | pt_prev = NULL; | |
3525 | } | |
48cc32d3 | 3526 | if (vlan_do_receive(&skb)) |
2425717b JF |
3527 | goto another_round; |
3528 | else if (unlikely(!skb)) | |
b4b9e355 | 3529 | goto unlock; |
2425717b JF |
3530 | } |
3531 | ||
48cc32d3 | 3532 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
3533 | if (rx_handler) { |
3534 | if (pt_prev) { | |
3535 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3536 | pt_prev = NULL; | |
3537 | } | |
8a4eb573 JP |
3538 | switch (rx_handler(&skb)) { |
3539 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 3540 | ret = NET_RX_SUCCESS; |
b4b9e355 | 3541 | goto unlock; |
8a4eb573 | 3542 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3543 | goto another_round; |
8a4eb573 JP |
3544 | case RX_HANDLER_EXACT: |
3545 | deliver_exact = true; | |
3546 | case RX_HANDLER_PASS: | |
3547 | break; | |
3548 | default: | |
3549 | BUG(); | |
3550 | } | |
ab95bfe0 | 3551 | } |
1da177e4 | 3552 | |
48cc32d3 FZ |
3553 | if (vlan_tx_nonzero_tag_present(skb)) |
3554 | skb->pkt_type = PACKET_OTHERHOST; | |
3555 | ||
63d8ea7f | 3556 | /* deliver only exact match when indicated */ |
8a4eb573 | 3557 | null_or_dev = deliver_exact ? skb->dev : NULL; |
1f3c8804 | 3558 | |
1da177e4 | 3559 | type = skb->protocol; |
82d8a867 PE |
3560 | list_for_each_entry_rcu(ptype, |
3561 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
63d8ea7f | 3562 | if (ptype->type == type && |
e3f48d37 JP |
3563 | (ptype->dev == null_or_dev || ptype->dev == skb->dev || |
3564 | ptype->dev == orig_dev)) { | |
4ec93edb | 3565 | if (pt_prev) |
f2ccd8fa | 3566 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3567 | pt_prev = ptype; |
3568 | } | |
3569 | } | |
3570 | ||
3571 | if (pt_prev) { | |
1080e512 | 3572 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 3573 | goto drop; |
1080e512 MT |
3574 | else |
3575 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 3576 | } else { |
b4b9e355 | 3577 | drop: |
caf586e5 | 3578 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3579 | kfree_skb(skb); |
3580 | /* Jamal, now you will not able to escape explaining | |
3581 | * me how you were going to use this. :-) | |
3582 | */ | |
3583 | ret = NET_RX_DROP; | |
3584 | } | |
3585 | ||
b4b9e355 | 3586 | unlock: |
1da177e4 | 3587 | rcu_read_unlock(); |
b4b9e355 | 3588 | out: |
9754e293 DM |
3589 | return ret; |
3590 | } | |
3591 | ||
3592 | static int __netif_receive_skb(struct sk_buff *skb) | |
3593 | { | |
3594 | int ret; | |
3595 | ||
3596 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
3597 | unsigned long pflags = current->flags; | |
3598 | ||
3599 | /* | |
3600 | * PFMEMALLOC skbs are special, they should | |
3601 | * - be delivered to SOCK_MEMALLOC sockets only | |
3602 | * - stay away from userspace | |
3603 | * - have bounded memory usage | |
3604 | * | |
3605 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
3606 | * context down to all allocation sites. | |
3607 | */ | |
3608 | current->flags |= PF_MEMALLOC; | |
3609 | ret = __netif_receive_skb_core(skb, true); | |
3610 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
3611 | } else | |
3612 | ret = __netif_receive_skb_core(skb, false); | |
3613 | ||
1da177e4 LT |
3614 | return ret; |
3615 | } | |
0a9627f2 TH |
3616 | |
3617 | /** | |
3618 | * netif_receive_skb - process receive buffer from network | |
3619 | * @skb: buffer to process | |
3620 | * | |
3621 | * netif_receive_skb() is the main receive data processing function. | |
3622 | * It always succeeds. The buffer may be dropped during processing | |
3623 | * for congestion control or by the protocol layers. | |
3624 | * | |
3625 | * This function may only be called from softirq context and interrupts | |
3626 | * should be enabled. | |
3627 | * | |
3628 | * Return values (usually ignored): | |
3629 | * NET_RX_SUCCESS: no congestion | |
3630 | * NET_RX_DROP: packet was dropped | |
3631 | */ | |
3632 | int netif_receive_skb(struct sk_buff *skb) | |
3633 | { | |
588f0330 | 3634 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3635 | |
c1f19b51 RC |
3636 | if (skb_defer_rx_timestamp(skb)) |
3637 | return NET_RX_SUCCESS; | |
3638 | ||
df334545 | 3639 | #ifdef CONFIG_RPS |
c5905afb | 3640 | if (static_key_false(&rps_needed)) { |
3b098e2d ED |
3641 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
3642 | int cpu, ret; | |
fec5e652 | 3643 | |
3b098e2d ED |
3644 | rcu_read_lock(); |
3645 | ||
3646 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3647 | |
3b098e2d ED |
3648 | if (cpu >= 0) { |
3649 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3650 | rcu_read_unlock(); | |
adc9300e | 3651 | return ret; |
3b098e2d | 3652 | } |
adc9300e | 3653 | rcu_read_unlock(); |
fec5e652 | 3654 | } |
1e94d72f | 3655 | #endif |
adc9300e | 3656 | return __netif_receive_skb(skb); |
0a9627f2 | 3657 | } |
d1b19dff | 3658 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3659 | |
88751275 ED |
3660 | /* Network device is going away, flush any packets still pending |
3661 | * Called with irqs disabled. | |
3662 | */ | |
152102c7 | 3663 | static void flush_backlog(void *arg) |
6e583ce5 | 3664 | { |
152102c7 | 3665 | struct net_device *dev = arg; |
e36fa2f7 | 3666 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3667 | struct sk_buff *skb, *tmp; |
3668 | ||
e36fa2f7 | 3669 | rps_lock(sd); |
6e7676c1 | 3670 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3671 | if (skb->dev == dev) { |
e36fa2f7 | 3672 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3673 | kfree_skb(skb); |
76cc8b13 | 3674 | input_queue_head_incr(sd); |
6e583ce5 | 3675 | } |
6e7676c1 | 3676 | } |
e36fa2f7 | 3677 | rps_unlock(sd); |
6e7676c1 CG |
3678 | |
3679 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3680 | if (skb->dev == dev) { | |
3681 | __skb_unlink(skb, &sd->process_queue); | |
3682 | kfree_skb(skb); | |
76cc8b13 | 3683 | input_queue_head_incr(sd); |
6e7676c1 CG |
3684 | } |
3685 | } | |
6e583ce5 SH |
3686 | } |
3687 | ||
d565b0a1 HX |
3688 | static int napi_gro_complete(struct sk_buff *skb) |
3689 | { | |
22061d80 | 3690 | struct packet_offload *ptype; |
d565b0a1 | 3691 | __be16 type = skb->protocol; |
22061d80 | 3692 | struct list_head *head = &offload_base; |
d565b0a1 HX |
3693 | int err = -ENOENT; |
3694 | ||
c3c7c254 ED |
3695 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
3696 | ||
fc59f9a3 HX |
3697 | if (NAPI_GRO_CB(skb)->count == 1) { |
3698 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3699 | goto out; |
fc59f9a3 | 3700 | } |
d565b0a1 HX |
3701 | |
3702 | rcu_read_lock(); | |
3703 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3704 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
3705 | continue; |
3706 | ||
f191a1d1 | 3707 | err = ptype->callbacks.gro_complete(skb); |
d565b0a1 HX |
3708 | break; |
3709 | } | |
3710 | rcu_read_unlock(); | |
3711 | ||
3712 | if (err) { | |
3713 | WARN_ON(&ptype->list == head); | |
3714 | kfree_skb(skb); | |
3715 | return NET_RX_SUCCESS; | |
3716 | } | |
3717 | ||
3718 | out: | |
d565b0a1 HX |
3719 | return netif_receive_skb(skb); |
3720 | } | |
3721 | ||
2e71a6f8 ED |
3722 | /* napi->gro_list contains packets ordered by age. |
3723 | * youngest packets at the head of it. | |
3724 | * Complete skbs in reverse order to reduce latencies. | |
3725 | */ | |
3726 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 3727 | { |
2e71a6f8 | 3728 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 3729 | |
2e71a6f8 ED |
3730 | /* scan list and build reverse chain */ |
3731 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
3732 | skb->prev = prev; | |
3733 | prev = skb; | |
3734 | } | |
3735 | ||
3736 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 3737 | skb->next = NULL; |
2e71a6f8 ED |
3738 | |
3739 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
3740 | return; | |
3741 | ||
3742 | prev = skb->prev; | |
d565b0a1 | 3743 | napi_gro_complete(skb); |
2e71a6f8 | 3744 | napi->gro_count--; |
d565b0a1 HX |
3745 | } |
3746 | ||
3747 | napi->gro_list = NULL; | |
3748 | } | |
86cac58b | 3749 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3750 | |
89c5fa33 ED |
3751 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
3752 | { | |
3753 | struct sk_buff *p; | |
3754 | unsigned int maclen = skb->dev->hard_header_len; | |
3755 | ||
3756 | for (p = napi->gro_list; p; p = p->next) { | |
3757 | unsigned long diffs; | |
3758 | ||
3759 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; | |
3760 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
3761 | if (maclen == ETH_HLEN) | |
3762 | diffs |= compare_ether_header(skb_mac_header(p), | |
3763 | skb_gro_mac_header(skb)); | |
3764 | else if (!diffs) | |
3765 | diffs = memcmp(skb_mac_header(p), | |
3766 | skb_gro_mac_header(skb), | |
3767 | maclen); | |
3768 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
3769 | NAPI_GRO_CB(p)->flush = 0; | |
3770 | } | |
3771 | } | |
3772 | ||
bb728820 | 3773 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3774 | { |
3775 | struct sk_buff **pp = NULL; | |
22061d80 | 3776 | struct packet_offload *ptype; |
d565b0a1 | 3777 | __be16 type = skb->protocol; |
22061d80 | 3778 | struct list_head *head = &offload_base; |
0da2afd5 | 3779 | int same_flow; |
5b252f0c | 3780 | enum gro_result ret; |
d565b0a1 | 3781 | |
ce9e76c8 | 3782 | if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb)) |
d565b0a1 HX |
3783 | goto normal; |
3784 | ||
21dc3301 | 3785 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) |
f17f5c91 HX |
3786 | goto normal; |
3787 | ||
89c5fa33 ED |
3788 | gro_list_prepare(napi, skb); |
3789 | ||
d565b0a1 HX |
3790 | rcu_read_lock(); |
3791 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3792 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
3793 | continue; |
3794 | ||
86911732 | 3795 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 3796 | skb_reset_mac_len(skb); |
d565b0a1 HX |
3797 | NAPI_GRO_CB(skb)->same_flow = 0; |
3798 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3799 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 3800 | |
f191a1d1 | 3801 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
3802 | break; |
3803 | } | |
3804 | rcu_read_unlock(); | |
3805 | ||
3806 | if (&ptype->list == head) | |
3807 | goto normal; | |
3808 | ||
0da2afd5 | 3809 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3810 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3811 | |
d565b0a1 HX |
3812 | if (pp) { |
3813 | struct sk_buff *nskb = *pp; | |
3814 | ||
3815 | *pp = nskb->next; | |
3816 | nskb->next = NULL; | |
3817 | napi_gro_complete(nskb); | |
4ae5544f | 3818 | napi->gro_count--; |
d565b0a1 HX |
3819 | } |
3820 | ||
0da2afd5 | 3821 | if (same_flow) |
d565b0a1 HX |
3822 | goto ok; |
3823 | ||
4ae5544f | 3824 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 3825 | goto normal; |
d565b0a1 | 3826 | |
4ae5544f | 3827 | napi->gro_count++; |
d565b0a1 | 3828 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 3829 | NAPI_GRO_CB(skb)->age = jiffies; |
86911732 | 3830 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3831 | skb->next = napi->gro_list; |
3832 | napi->gro_list = skb; | |
5d0d9be8 | 3833 | ret = GRO_HELD; |
d565b0a1 | 3834 | |
ad0f9904 | 3835 | pull: |
cb18978c HX |
3836 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
3837 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3838 | ||
3839 | BUG_ON(skb->end - skb->tail < grow); | |
3840 | ||
3841 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3842 | ||
3843 | skb->tail += grow; | |
3844 | skb->data_len -= grow; | |
3845 | ||
3846 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
9e903e08 | 3847 | skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow); |
cb18978c | 3848 | |
9e903e08 | 3849 | if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) { |
ea2ab693 | 3850 | skb_frag_unref(skb, 0); |
cb18978c HX |
3851 | memmove(skb_shinfo(skb)->frags, |
3852 | skb_shinfo(skb)->frags + 1, | |
e5093aec | 3853 | --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); |
cb18978c | 3854 | } |
ad0f9904 HX |
3855 | } |
3856 | ||
d565b0a1 | 3857 | ok: |
5d0d9be8 | 3858 | return ret; |
d565b0a1 HX |
3859 | |
3860 | normal: | |
ad0f9904 HX |
3861 | ret = GRO_NORMAL; |
3862 | goto pull; | |
5d38a079 | 3863 | } |
96e93eab | 3864 | |
5d38a079 | 3865 | |
bb728820 | 3866 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3867 | { |
5d0d9be8 HX |
3868 | switch (ret) { |
3869 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3870 | if (netif_receive_skb(skb)) |
3871 | ret = GRO_DROP; | |
3872 | break; | |
5d38a079 | 3873 | |
5d0d9be8 | 3874 | case GRO_DROP: |
5d38a079 HX |
3875 | kfree_skb(skb); |
3876 | break; | |
5b252f0c | 3877 | |
daa86548 | 3878 | case GRO_MERGED_FREE: |
d7e8883c ED |
3879 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
3880 | kmem_cache_free(skbuff_head_cache, skb); | |
3881 | else | |
3882 | __kfree_skb(skb); | |
daa86548 ED |
3883 | break; |
3884 | ||
5b252f0c BH |
3885 | case GRO_HELD: |
3886 | case GRO_MERGED: | |
3887 | break; | |
5d38a079 HX |
3888 | } |
3889 | ||
c7c4b3b6 | 3890 | return ret; |
5d0d9be8 | 3891 | } |
5d0d9be8 | 3892 | |
ca07e43e | 3893 | static void skb_gro_reset_offset(struct sk_buff *skb) |
78a478d0 | 3894 | { |
ca07e43e ED |
3895 | const struct skb_shared_info *pinfo = skb_shinfo(skb); |
3896 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
3897 | ||
78a478d0 HX |
3898 | NAPI_GRO_CB(skb)->data_offset = 0; |
3899 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 3900 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 3901 | |
ced14f68 | 3902 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && |
ca07e43e ED |
3903 | pinfo->nr_frags && |
3904 | !PageHighMem(skb_frag_page(frag0))) { | |
3905 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
3906 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
7489594c | 3907 | } |
78a478d0 | 3908 | } |
78a478d0 | 3909 | |
c7c4b3b6 | 3910 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 3911 | { |
86911732 HX |
3912 | skb_gro_reset_offset(skb); |
3913 | ||
89c5fa33 | 3914 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
3915 | } |
3916 | EXPORT_SYMBOL(napi_gro_receive); | |
3917 | ||
d0c2b0d2 | 3918 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 3919 | { |
96e93eab | 3920 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
3921 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
3922 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 3923 | skb->vlan_tci = 0; |
66c46d74 | 3924 | skb->dev = napi->dev; |
6d152e23 | 3925 | skb->skb_iif = 0; |
96e93eab HX |
3926 | |
3927 | napi->skb = skb; | |
3928 | } | |
96e93eab | 3929 | |
76620aaf | 3930 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 3931 | { |
5d38a079 | 3932 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
3933 | |
3934 | if (!skb) { | |
89d71a66 ED |
3935 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3936 | if (skb) | |
3937 | napi->skb = skb; | |
80595d59 | 3938 | } |
96e93eab HX |
3939 | return skb; |
3940 | } | |
76620aaf | 3941 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3942 | |
bb728820 | 3943 | static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
c7c4b3b6 | 3944 | gro_result_t ret) |
96e93eab | 3945 | { |
5d0d9be8 HX |
3946 | switch (ret) { |
3947 | case GRO_NORMAL: | |
86911732 | 3948 | case GRO_HELD: |
e76b69cc | 3949 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3950 | |
c7c4b3b6 BH |
3951 | if (ret == GRO_HELD) |
3952 | skb_gro_pull(skb, -ETH_HLEN); | |
3953 | else if (netif_receive_skb(skb)) | |
3954 | ret = GRO_DROP; | |
86911732 | 3955 | break; |
5d38a079 | 3956 | |
5d0d9be8 | 3957 | case GRO_DROP: |
5d0d9be8 HX |
3958 | case GRO_MERGED_FREE: |
3959 | napi_reuse_skb(napi, skb); | |
3960 | break; | |
5b252f0c BH |
3961 | |
3962 | case GRO_MERGED: | |
3963 | break; | |
5d0d9be8 | 3964 | } |
5d38a079 | 3965 | |
c7c4b3b6 | 3966 | return ret; |
5d38a079 | 3967 | } |
5d0d9be8 | 3968 | |
4adb9c4a | 3969 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
3970 | { |
3971 | struct sk_buff *skb = napi->skb; | |
3972 | struct ethhdr *eth; | |
a5b1cf28 HX |
3973 | unsigned int hlen; |
3974 | unsigned int off; | |
76620aaf HX |
3975 | |
3976 | napi->skb = NULL; | |
3977 | ||
3978 | skb_reset_mac_header(skb); | |
3979 | skb_gro_reset_offset(skb); | |
3980 | ||
a5b1cf28 HX |
3981 | off = skb_gro_offset(skb); |
3982 | hlen = off + sizeof(*eth); | |
3983 | eth = skb_gro_header_fast(skb, off); | |
3984 | if (skb_gro_header_hard(skb, hlen)) { | |
3985 | eth = skb_gro_header_slow(skb, hlen, off); | |
3986 | if (unlikely(!eth)) { | |
3987 | napi_reuse_skb(napi, skb); | |
3988 | skb = NULL; | |
3989 | goto out; | |
3990 | } | |
76620aaf HX |
3991 | } |
3992 | ||
3993 | skb_gro_pull(skb, sizeof(*eth)); | |
3994 | ||
3995 | /* | |
3996 | * This works because the only protocols we care about don't require | |
3997 | * special handling. We'll fix it up properly at the end. | |
3998 | */ | |
3999 | skb->protocol = eth->h_proto; | |
4000 | ||
4001 | out: | |
4002 | return skb; | |
4003 | } | |
76620aaf | 4004 | |
c7c4b3b6 | 4005 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4006 | { |
76620aaf | 4007 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4008 | |
4009 | if (!skb) | |
c7c4b3b6 | 4010 | return GRO_DROP; |
5d0d9be8 | 4011 | |
89c5fa33 | 4012 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4013 | } |
5d38a079 HX |
4014 | EXPORT_SYMBOL(napi_gro_frags); |
4015 | ||
e326bed2 ED |
4016 | /* |
4017 | * net_rps_action sends any pending IPI's for rps. | |
4018 | * Note: called with local irq disabled, but exits with local irq enabled. | |
4019 | */ | |
4020 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4021 | { | |
4022 | #ifdef CONFIG_RPS | |
4023 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4024 | ||
4025 | if (remsd) { | |
4026 | sd->rps_ipi_list = NULL; | |
4027 | ||
4028 | local_irq_enable(); | |
4029 | ||
4030 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
4031 | while (remsd) { | |
4032 | struct softnet_data *next = remsd->rps_ipi_next; | |
4033 | ||
4034 | if (cpu_online(remsd->cpu)) | |
4035 | __smp_call_function_single(remsd->cpu, | |
4036 | &remsd->csd, 0); | |
4037 | remsd = next; | |
4038 | } | |
4039 | } else | |
4040 | #endif | |
4041 | local_irq_enable(); | |
4042 | } | |
4043 | ||
bea3348e | 4044 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
4045 | { |
4046 | int work = 0; | |
eecfd7c4 | 4047 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 4048 | |
e326bed2 ED |
4049 | #ifdef CONFIG_RPS |
4050 | /* Check if we have pending ipi, its better to send them now, | |
4051 | * not waiting net_rx_action() end. | |
4052 | */ | |
4053 | if (sd->rps_ipi_list) { | |
4054 | local_irq_disable(); | |
4055 | net_rps_action_and_irq_enable(sd); | |
4056 | } | |
4057 | #endif | |
bea3348e | 4058 | napi->weight = weight_p; |
6e7676c1 CG |
4059 | local_irq_disable(); |
4060 | while (work < quota) { | |
1da177e4 | 4061 | struct sk_buff *skb; |
6e7676c1 CG |
4062 | unsigned int qlen; |
4063 | ||
4064 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
4065 | local_irq_enable(); | |
4066 | __netif_receive_skb(skb); | |
6e7676c1 | 4067 | local_irq_disable(); |
76cc8b13 TH |
4068 | input_queue_head_incr(sd); |
4069 | if (++work >= quota) { | |
4070 | local_irq_enable(); | |
4071 | return work; | |
4072 | } | |
6e7676c1 | 4073 | } |
1da177e4 | 4074 | |
e36fa2f7 | 4075 | rps_lock(sd); |
6e7676c1 | 4076 | qlen = skb_queue_len(&sd->input_pkt_queue); |
76cc8b13 | 4077 | if (qlen) |
6e7676c1 CG |
4078 | skb_queue_splice_tail_init(&sd->input_pkt_queue, |
4079 | &sd->process_queue); | |
76cc8b13 | 4080 | |
6e7676c1 | 4081 | if (qlen < quota - work) { |
eecfd7c4 ED |
4082 | /* |
4083 | * Inline a custom version of __napi_complete(). | |
4084 | * only current cpu owns and manipulates this napi, | |
4085 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
4086 | * we can use a plain write instead of clear_bit(), | |
4087 | * and we dont need an smp_mb() memory barrier. | |
4088 | */ | |
4089 | list_del(&napi->poll_list); | |
4090 | napi->state = 0; | |
4091 | ||
6e7676c1 | 4092 | quota = work + qlen; |
bea3348e | 4093 | } |
e36fa2f7 | 4094 | rps_unlock(sd); |
6e7676c1 CG |
4095 | } |
4096 | local_irq_enable(); | |
1da177e4 | 4097 | |
bea3348e SH |
4098 | return work; |
4099 | } | |
1da177e4 | 4100 | |
bea3348e SH |
4101 | /** |
4102 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4103 | * @n: entry to schedule |
bea3348e SH |
4104 | * |
4105 | * The entry's receive function will be scheduled to run | |
4106 | */ | |
b5606c2d | 4107 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4108 | { |
4109 | unsigned long flags; | |
1da177e4 | 4110 | |
bea3348e | 4111 | local_irq_save(flags); |
eecfd7c4 | 4112 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 4113 | local_irq_restore(flags); |
1da177e4 | 4114 | } |
bea3348e SH |
4115 | EXPORT_SYMBOL(__napi_schedule); |
4116 | ||
d565b0a1 HX |
4117 | void __napi_complete(struct napi_struct *n) |
4118 | { | |
4119 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
4120 | BUG_ON(n->gro_list); | |
4121 | ||
4122 | list_del(&n->poll_list); | |
4123 | smp_mb__before_clear_bit(); | |
4124 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
4125 | } | |
4126 | EXPORT_SYMBOL(__napi_complete); | |
4127 | ||
4128 | void napi_complete(struct napi_struct *n) | |
4129 | { | |
4130 | unsigned long flags; | |
4131 | ||
4132 | /* | |
4133 | * don't let napi dequeue from the cpu poll list | |
4134 | * just in case its running on a different cpu | |
4135 | */ | |
4136 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4137 | return; | |
4138 | ||
2e71a6f8 | 4139 | napi_gro_flush(n, false); |
d565b0a1 HX |
4140 | local_irq_save(flags); |
4141 | __napi_complete(n); | |
4142 | local_irq_restore(flags); | |
4143 | } | |
4144 | EXPORT_SYMBOL(napi_complete); | |
4145 | ||
af12fa6e ET |
4146 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
4147 | struct napi_struct *napi_by_id(unsigned int napi_id) | |
4148 | { | |
4149 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
4150 | struct napi_struct *napi; | |
4151 | ||
4152 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
4153 | if (napi->napi_id == napi_id) | |
4154 | return napi; | |
4155 | ||
4156 | return NULL; | |
4157 | } | |
4158 | EXPORT_SYMBOL_GPL(napi_by_id); | |
4159 | ||
4160 | void napi_hash_add(struct napi_struct *napi) | |
4161 | { | |
4162 | if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) { | |
4163 | ||
4164 | spin_lock(&napi_hash_lock); | |
4165 | ||
4166 | /* 0 is not a valid id, we also skip an id that is taken | |
4167 | * we expect both events to be extremely rare | |
4168 | */ | |
4169 | napi->napi_id = 0; | |
4170 | while (!napi->napi_id) { | |
4171 | napi->napi_id = ++napi_gen_id; | |
4172 | if (napi_by_id(napi->napi_id)) | |
4173 | napi->napi_id = 0; | |
4174 | } | |
4175 | ||
4176 | hlist_add_head_rcu(&napi->napi_hash_node, | |
4177 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
4178 | ||
4179 | spin_unlock(&napi_hash_lock); | |
4180 | } | |
4181 | } | |
4182 | EXPORT_SYMBOL_GPL(napi_hash_add); | |
4183 | ||
4184 | /* Warning : caller is responsible to make sure rcu grace period | |
4185 | * is respected before freeing memory containing @napi | |
4186 | */ | |
4187 | void napi_hash_del(struct napi_struct *napi) | |
4188 | { | |
4189 | spin_lock(&napi_hash_lock); | |
4190 | ||
4191 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) | |
4192 | hlist_del_rcu(&napi->napi_hash_node); | |
4193 | ||
4194 | spin_unlock(&napi_hash_lock); | |
4195 | } | |
4196 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
4197 | ||
d565b0a1 HX |
4198 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
4199 | int (*poll)(struct napi_struct *, int), int weight) | |
4200 | { | |
4201 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 4202 | napi->gro_count = 0; |
d565b0a1 | 4203 | napi->gro_list = NULL; |
5d38a079 | 4204 | napi->skb = NULL; |
d565b0a1 | 4205 | napi->poll = poll; |
82dc3c63 ED |
4206 | if (weight > NAPI_POLL_WEIGHT) |
4207 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
4208 | weight, dev->name); | |
d565b0a1 HX |
4209 | napi->weight = weight; |
4210 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 4211 | napi->dev = dev; |
5d38a079 | 4212 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
4213 | spin_lock_init(&napi->poll_lock); |
4214 | napi->poll_owner = -1; | |
4215 | #endif | |
4216 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
4217 | } | |
4218 | EXPORT_SYMBOL(netif_napi_add); | |
4219 | ||
4220 | void netif_napi_del(struct napi_struct *napi) | |
4221 | { | |
4222 | struct sk_buff *skb, *next; | |
4223 | ||
d7b06636 | 4224 | list_del_init(&napi->dev_list); |
76620aaf | 4225 | napi_free_frags(napi); |
d565b0a1 HX |
4226 | |
4227 | for (skb = napi->gro_list; skb; skb = next) { | |
4228 | next = skb->next; | |
4229 | skb->next = NULL; | |
4230 | kfree_skb(skb); | |
4231 | } | |
4232 | ||
4233 | napi->gro_list = NULL; | |
4ae5544f | 4234 | napi->gro_count = 0; |
d565b0a1 HX |
4235 | } |
4236 | EXPORT_SYMBOL(netif_napi_del); | |
4237 | ||
1da177e4 LT |
4238 | static void net_rx_action(struct softirq_action *h) |
4239 | { | |
e326bed2 | 4240 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 4241 | unsigned long time_limit = jiffies + 2; |
51b0bded | 4242 | int budget = netdev_budget; |
53fb95d3 MM |
4243 | void *have; |
4244 | ||
1da177e4 LT |
4245 | local_irq_disable(); |
4246 | ||
e326bed2 | 4247 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
4248 | struct napi_struct *n; |
4249 | int work, weight; | |
1da177e4 | 4250 | |
bea3348e | 4251 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
4252 | * Allow this to run for 2 jiffies since which will allow |
4253 | * an average latency of 1.5/HZ. | |
bea3348e | 4254 | */ |
d1f41b67 | 4255 | if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit))) |
1da177e4 LT |
4256 | goto softnet_break; |
4257 | ||
4258 | local_irq_enable(); | |
4259 | ||
bea3348e SH |
4260 | /* Even though interrupts have been re-enabled, this |
4261 | * access is safe because interrupts can only add new | |
4262 | * entries to the tail of this list, and only ->poll() | |
4263 | * calls can remove this head entry from the list. | |
4264 | */ | |
e326bed2 | 4265 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 4266 | |
bea3348e SH |
4267 | have = netpoll_poll_lock(n); |
4268 | ||
4269 | weight = n->weight; | |
4270 | ||
0a7606c1 DM |
4271 | /* This NAPI_STATE_SCHED test is for avoiding a race |
4272 | * with netpoll's poll_napi(). Only the entity which | |
4273 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
4274 | * actually make the ->poll() call. Therefore we avoid | |
25985edc | 4275 | * accidentally calling ->poll() when NAPI is not scheduled. |
0a7606c1 DM |
4276 | */ |
4277 | work = 0; | |
4ea7e386 | 4278 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 4279 | work = n->poll(n, weight); |
4ea7e386 NH |
4280 | trace_napi_poll(n); |
4281 | } | |
bea3348e SH |
4282 | |
4283 | WARN_ON_ONCE(work > weight); | |
4284 | ||
4285 | budget -= work; | |
4286 | ||
4287 | local_irq_disable(); | |
4288 | ||
4289 | /* Drivers must not modify the NAPI state if they | |
4290 | * consume the entire weight. In such cases this code | |
4291 | * still "owns" the NAPI instance and therefore can | |
4292 | * move the instance around on the list at-will. | |
4293 | */ | |
fed17f30 | 4294 | if (unlikely(work == weight)) { |
ff780cd8 HX |
4295 | if (unlikely(napi_disable_pending(n))) { |
4296 | local_irq_enable(); | |
4297 | napi_complete(n); | |
4298 | local_irq_disable(); | |
2e71a6f8 ED |
4299 | } else { |
4300 | if (n->gro_list) { | |
4301 | /* flush too old packets | |
4302 | * If HZ < 1000, flush all packets. | |
4303 | */ | |
4304 | local_irq_enable(); | |
4305 | napi_gro_flush(n, HZ >= 1000); | |
4306 | local_irq_disable(); | |
4307 | } | |
e326bed2 | 4308 | list_move_tail(&n->poll_list, &sd->poll_list); |
2e71a6f8 | 4309 | } |
fed17f30 | 4310 | } |
bea3348e SH |
4311 | |
4312 | netpoll_poll_unlock(have); | |
1da177e4 LT |
4313 | } |
4314 | out: | |
e326bed2 | 4315 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 4316 | |
db217334 CL |
4317 | #ifdef CONFIG_NET_DMA |
4318 | /* | |
4319 | * There may not be any more sk_buffs coming right now, so push | |
4320 | * any pending DMA copies to hardware | |
4321 | */ | |
2ba05622 | 4322 | dma_issue_pending_all(); |
db217334 | 4323 | #endif |
bea3348e | 4324 | |
1da177e4 LT |
4325 | return; |
4326 | ||
4327 | softnet_break: | |
dee42870 | 4328 | sd->time_squeeze++; |
1da177e4 LT |
4329 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
4330 | goto out; | |
4331 | } | |
4332 | ||
9ff162a8 JP |
4333 | struct netdev_upper { |
4334 | struct net_device *dev; | |
4335 | bool master; | |
4336 | struct list_head list; | |
4337 | struct rcu_head rcu; | |
4338 | struct list_head search_list; | |
4339 | }; | |
4340 | ||
4341 | static void __append_search_uppers(struct list_head *search_list, | |
4342 | struct net_device *dev) | |
4343 | { | |
4344 | struct netdev_upper *upper; | |
4345 | ||
4346 | list_for_each_entry(upper, &dev->upper_dev_list, list) { | |
4347 | /* check if this upper is not already in search list */ | |
4348 | if (list_empty(&upper->search_list)) | |
4349 | list_add_tail(&upper->search_list, search_list); | |
4350 | } | |
4351 | } | |
4352 | ||
4353 | static bool __netdev_search_upper_dev(struct net_device *dev, | |
4354 | struct net_device *upper_dev) | |
4355 | { | |
4356 | LIST_HEAD(search_list); | |
4357 | struct netdev_upper *upper; | |
4358 | struct netdev_upper *tmp; | |
4359 | bool ret = false; | |
4360 | ||
4361 | __append_search_uppers(&search_list, dev); | |
4362 | list_for_each_entry(upper, &search_list, search_list) { | |
4363 | if (upper->dev == upper_dev) { | |
4364 | ret = true; | |
4365 | break; | |
4366 | } | |
4367 | __append_search_uppers(&search_list, upper->dev); | |
4368 | } | |
4369 | list_for_each_entry_safe(upper, tmp, &search_list, search_list) | |
4370 | INIT_LIST_HEAD(&upper->search_list); | |
4371 | return ret; | |
4372 | } | |
4373 | ||
4374 | static struct netdev_upper *__netdev_find_upper(struct net_device *dev, | |
4375 | struct net_device *upper_dev) | |
4376 | { | |
4377 | struct netdev_upper *upper; | |
4378 | ||
4379 | list_for_each_entry(upper, &dev->upper_dev_list, list) { | |
4380 | if (upper->dev == upper_dev) | |
4381 | return upper; | |
4382 | } | |
4383 | return NULL; | |
4384 | } | |
4385 | ||
4386 | /** | |
4387 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
4388 | * @dev: device | |
4389 | * @upper_dev: upper device to check | |
4390 | * | |
4391 | * Find out if a device is linked to specified upper device and return true | |
4392 | * in case it is. Note that this checks only immediate upper device, | |
4393 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
4394 | */ | |
4395 | bool netdev_has_upper_dev(struct net_device *dev, | |
4396 | struct net_device *upper_dev) | |
4397 | { | |
4398 | ASSERT_RTNL(); | |
4399 | ||
4400 | return __netdev_find_upper(dev, upper_dev); | |
4401 | } | |
4402 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
4403 | ||
4404 | /** | |
4405 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
4406 | * @dev: device | |
4407 | * | |
4408 | * Find out if a device is linked to an upper device and return true in case | |
4409 | * it is. The caller must hold the RTNL lock. | |
4410 | */ | |
4411 | bool netdev_has_any_upper_dev(struct net_device *dev) | |
4412 | { | |
4413 | ASSERT_RTNL(); | |
4414 | ||
4415 | return !list_empty(&dev->upper_dev_list); | |
4416 | } | |
4417 | EXPORT_SYMBOL(netdev_has_any_upper_dev); | |
4418 | ||
4419 | /** | |
4420 | * netdev_master_upper_dev_get - Get master upper device | |
4421 | * @dev: device | |
4422 | * | |
4423 | * Find a master upper device and return pointer to it or NULL in case | |
4424 | * it's not there. The caller must hold the RTNL lock. | |
4425 | */ | |
4426 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
4427 | { | |
4428 | struct netdev_upper *upper; | |
4429 | ||
4430 | ASSERT_RTNL(); | |
4431 | ||
4432 | if (list_empty(&dev->upper_dev_list)) | |
4433 | return NULL; | |
4434 | ||
4435 | upper = list_first_entry(&dev->upper_dev_list, | |
4436 | struct netdev_upper, list); | |
4437 | if (likely(upper->master)) | |
4438 | return upper->dev; | |
4439 | return NULL; | |
4440 | } | |
4441 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
4442 | ||
4443 | /** | |
4444 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
4445 | * @dev: device | |
4446 | * | |
4447 | * Find a master upper device and return pointer to it or NULL in case | |
4448 | * it's not there. The caller must hold the RCU read lock. | |
4449 | */ | |
4450 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
4451 | { | |
4452 | struct netdev_upper *upper; | |
4453 | ||
4454 | upper = list_first_or_null_rcu(&dev->upper_dev_list, | |
4455 | struct netdev_upper, list); | |
4456 | if (upper && likely(upper->master)) | |
4457 | return upper->dev; | |
4458 | return NULL; | |
4459 | } | |
4460 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
4461 | ||
4462 | static int __netdev_upper_dev_link(struct net_device *dev, | |
4463 | struct net_device *upper_dev, bool master) | |
4464 | { | |
4465 | struct netdev_upper *upper; | |
4466 | ||
4467 | ASSERT_RTNL(); | |
4468 | ||
4469 | if (dev == upper_dev) | |
4470 | return -EBUSY; | |
4471 | ||
4472 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
4473 | if (__netdev_search_upper_dev(upper_dev, dev)) | |
4474 | return -EBUSY; | |
4475 | ||
4476 | if (__netdev_find_upper(dev, upper_dev)) | |
4477 | return -EEXIST; | |
4478 | ||
4479 | if (master && netdev_master_upper_dev_get(dev)) | |
4480 | return -EBUSY; | |
4481 | ||
4482 | upper = kmalloc(sizeof(*upper), GFP_KERNEL); | |
4483 | if (!upper) | |
4484 | return -ENOMEM; | |
4485 | ||
4486 | upper->dev = upper_dev; | |
4487 | upper->master = master; | |
4488 | INIT_LIST_HEAD(&upper->search_list); | |
4489 | ||
4490 | /* Ensure that master upper link is always the first item in list. */ | |
4491 | if (master) | |
4492 | list_add_rcu(&upper->list, &dev->upper_dev_list); | |
4493 | else | |
4494 | list_add_tail_rcu(&upper->list, &dev->upper_dev_list); | |
4495 | dev_hold(upper_dev); | |
42e52bf9 | 4496 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 JP |
4497 | return 0; |
4498 | } | |
4499 | ||
4500 | /** | |
4501 | * netdev_upper_dev_link - Add a link to the upper device | |
4502 | * @dev: device | |
4503 | * @upper_dev: new upper device | |
4504 | * | |
4505 | * Adds a link to device which is upper to this one. The caller must hold | |
4506 | * the RTNL lock. On a failure a negative errno code is returned. | |
4507 | * On success the reference counts are adjusted and the function | |
4508 | * returns zero. | |
4509 | */ | |
4510 | int netdev_upper_dev_link(struct net_device *dev, | |
4511 | struct net_device *upper_dev) | |
4512 | { | |
4513 | return __netdev_upper_dev_link(dev, upper_dev, false); | |
4514 | } | |
4515 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
4516 | ||
4517 | /** | |
4518 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
4519 | * @dev: device | |
4520 | * @upper_dev: new upper device | |
4521 | * | |
4522 | * Adds a link to device which is upper to this one. In this case, only | |
4523 | * one master upper device can be linked, although other non-master devices | |
4524 | * might be linked as well. The caller must hold the RTNL lock. | |
4525 | * On a failure a negative errno code is returned. On success the reference | |
4526 | * counts are adjusted and the function returns zero. | |
4527 | */ | |
4528 | int netdev_master_upper_dev_link(struct net_device *dev, | |
4529 | struct net_device *upper_dev) | |
4530 | { | |
4531 | return __netdev_upper_dev_link(dev, upper_dev, true); | |
4532 | } | |
4533 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
4534 | ||
4535 | /** | |
4536 | * netdev_upper_dev_unlink - Removes a link to upper device | |
4537 | * @dev: device | |
4538 | * @upper_dev: new upper device | |
4539 | * | |
4540 | * Removes a link to device which is upper to this one. The caller must hold | |
4541 | * the RTNL lock. | |
4542 | */ | |
4543 | void netdev_upper_dev_unlink(struct net_device *dev, | |
4544 | struct net_device *upper_dev) | |
4545 | { | |
4546 | struct netdev_upper *upper; | |
4547 | ||
4548 | ASSERT_RTNL(); | |
4549 | ||
4550 | upper = __netdev_find_upper(dev, upper_dev); | |
4551 | if (!upper) | |
4552 | return; | |
4553 | list_del_rcu(&upper->list); | |
4554 | dev_put(upper_dev); | |
4555 | kfree_rcu(upper, rcu); | |
42e52bf9 | 4556 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 JP |
4557 | } |
4558 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
4559 | ||
b6c40d68 PM |
4560 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
4561 | { | |
d314774c SH |
4562 | const struct net_device_ops *ops = dev->netdev_ops; |
4563 | ||
4564 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
4565 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
4566 | } |
4567 | ||
dad9b335 | 4568 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 | 4569 | { |
b536db93 | 4570 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
4571 | kuid_t uid; |
4572 | kgid_t gid; | |
1da177e4 | 4573 | |
24023451 PM |
4574 | ASSERT_RTNL(); |
4575 | ||
dad9b335 WC |
4576 | dev->flags |= IFF_PROMISC; |
4577 | dev->promiscuity += inc; | |
4578 | if (dev->promiscuity == 0) { | |
4579 | /* | |
4580 | * Avoid overflow. | |
4581 | * If inc causes overflow, untouch promisc and return error. | |
4582 | */ | |
4583 | if (inc < 0) | |
4584 | dev->flags &= ~IFF_PROMISC; | |
4585 | else { | |
4586 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
4587 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
4588 | dev->name); | |
dad9b335 WC |
4589 | return -EOVERFLOW; |
4590 | } | |
4591 | } | |
52609c0b | 4592 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
4593 | pr_info("device %s %s promiscuous mode\n", |
4594 | dev->name, | |
4595 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
4596 | if (audit_enabled) { |
4597 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
4598 | audit_log(current->audit_context, GFP_ATOMIC, |
4599 | AUDIT_ANOM_PROMISCUOUS, | |
4600 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
4601 | dev->name, (dev->flags & IFF_PROMISC), | |
4602 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 4603 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
4604 | from_kuid(&init_user_ns, uid), |
4605 | from_kgid(&init_user_ns, gid), | |
7759db82 | 4606 | audit_get_sessionid(current)); |
8192b0c4 | 4607 | } |
24023451 | 4608 | |
b6c40d68 | 4609 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 4610 | } |
dad9b335 | 4611 | return 0; |
1da177e4 LT |
4612 | } |
4613 | ||
4417da66 PM |
4614 | /** |
4615 | * dev_set_promiscuity - update promiscuity count on a device | |
4616 | * @dev: device | |
4617 | * @inc: modifier | |
4618 | * | |
4619 | * Add or remove promiscuity from a device. While the count in the device | |
4620 | * remains above zero the interface remains promiscuous. Once it hits zero | |
4621 | * the device reverts back to normal filtering operation. A negative inc | |
4622 | * value is used to drop promiscuity on the device. | |
dad9b335 | 4623 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 4624 | */ |
dad9b335 | 4625 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 4626 | { |
b536db93 | 4627 | unsigned int old_flags = dev->flags; |
dad9b335 | 4628 | int err; |
4417da66 | 4629 | |
dad9b335 | 4630 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 4631 | if (err < 0) |
dad9b335 | 4632 | return err; |
4417da66 PM |
4633 | if (dev->flags != old_flags) |
4634 | dev_set_rx_mode(dev); | |
dad9b335 | 4635 | return err; |
4417da66 | 4636 | } |
d1b19dff | 4637 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 4638 | |
1da177e4 LT |
4639 | /** |
4640 | * dev_set_allmulti - update allmulti count on a device | |
4641 | * @dev: device | |
4642 | * @inc: modifier | |
4643 | * | |
4644 | * Add or remove reception of all multicast frames to a device. While the | |
4645 | * count in the device remains above zero the interface remains listening | |
4646 | * to all interfaces. Once it hits zero the device reverts back to normal | |
4647 | * filtering operation. A negative @inc value is used to drop the counter | |
4648 | * when releasing a resource needing all multicasts. | |
dad9b335 | 4649 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
4650 | */ |
4651 | ||
dad9b335 | 4652 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 | 4653 | { |
b536db93 | 4654 | unsigned int old_flags = dev->flags; |
1da177e4 | 4655 | |
24023451 PM |
4656 | ASSERT_RTNL(); |
4657 | ||
1da177e4 | 4658 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
4659 | dev->allmulti += inc; |
4660 | if (dev->allmulti == 0) { | |
4661 | /* | |
4662 | * Avoid overflow. | |
4663 | * If inc causes overflow, untouch allmulti and return error. | |
4664 | */ | |
4665 | if (inc < 0) | |
4666 | dev->flags &= ~IFF_ALLMULTI; | |
4667 | else { | |
4668 | dev->allmulti -= inc; | |
7b6cd1ce JP |
4669 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
4670 | dev->name); | |
dad9b335 WC |
4671 | return -EOVERFLOW; |
4672 | } | |
4673 | } | |
24023451 | 4674 | if (dev->flags ^ old_flags) { |
b6c40d68 | 4675 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 4676 | dev_set_rx_mode(dev); |
24023451 | 4677 | } |
dad9b335 | 4678 | return 0; |
4417da66 | 4679 | } |
d1b19dff | 4680 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
4681 | |
4682 | /* | |
4683 | * Upload unicast and multicast address lists to device and | |
4684 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 4685 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
4686 | * are present. |
4687 | */ | |
4688 | void __dev_set_rx_mode(struct net_device *dev) | |
4689 | { | |
d314774c SH |
4690 | const struct net_device_ops *ops = dev->netdev_ops; |
4691 | ||
4417da66 PM |
4692 | /* dev_open will call this function so the list will stay sane. */ |
4693 | if (!(dev->flags&IFF_UP)) | |
4694 | return; | |
4695 | ||
4696 | if (!netif_device_present(dev)) | |
40b77c94 | 4697 | return; |
4417da66 | 4698 | |
01789349 | 4699 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
4700 | /* Unicast addresses changes may only happen under the rtnl, |
4701 | * therefore calling __dev_set_promiscuity here is safe. | |
4702 | */ | |
32e7bfc4 | 4703 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 | 4704 | __dev_set_promiscuity(dev, 1); |
2d348d1f | 4705 | dev->uc_promisc = true; |
32e7bfc4 | 4706 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 | 4707 | __dev_set_promiscuity(dev, -1); |
2d348d1f | 4708 | dev->uc_promisc = false; |
4417da66 | 4709 | } |
4417da66 | 4710 | } |
01789349 JP |
4711 | |
4712 | if (ops->ndo_set_rx_mode) | |
4713 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
4714 | } |
4715 | ||
4716 | void dev_set_rx_mode(struct net_device *dev) | |
4717 | { | |
b9e40857 | 4718 | netif_addr_lock_bh(dev); |
4417da66 | 4719 | __dev_set_rx_mode(dev); |
b9e40857 | 4720 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
4721 | } |
4722 | ||
f0db275a SH |
4723 | /** |
4724 | * dev_get_flags - get flags reported to userspace | |
4725 | * @dev: device | |
4726 | * | |
4727 | * Get the combination of flag bits exported through APIs to userspace. | |
4728 | */ | |
95c96174 | 4729 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 4730 | { |
95c96174 | 4731 | unsigned int flags; |
1da177e4 LT |
4732 | |
4733 | flags = (dev->flags & ~(IFF_PROMISC | | |
4734 | IFF_ALLMULTI | | |
b00055aa SR |
4735 | IFF_RUNNING | |
4736 | IFF_LOWER_UP | | |
4737 | IFF_DORMANT)) | | |
1da177e4 LT |
4738 | (dev->gflags & (IFF_PROMISC | |
4739 | IFF_ALLMULTI)); | |
4740 | ||
b00055aa SR |
4741 | if (netif_running(dev)) { |
4742 | if (netif_oper_up(dev)) | |
4743 | flags |= IFF_RUNNING; | |
4744 | if (netif_carrier_ok(dev)) | |
4745 | flags |= IFF_LOWER_UP; | |
4746 | if (netif_dormant(dev)) | |
4747 | flags |= IFF_DORMANT; | |
4748 | } | |
1da177e4 LT |
4749 | |
4750 | return flags; | |
4751 | } | |
d1b19dff | 4752 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4753 | |
bd380811 | 4754 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 4755 | { |
b536db93 | 4756 | unsigned int old_flags = dev->flags; |
bd380811 | 4757 | int ret; |
1da177e4 | 4758 | |
24023451 PM |
4759 | ASSERT_RTNL(); |
4760 | ||
1da177e4 LT |
4761 | /* |
4762 | * Set the flags on our device. | |
4763 | */ | |
4764 | ||
4765 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4766 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4767 | IFF_AUTOMEDIA)) | | |
4768 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4769 | IFF_ALLMULTI)); | |
4770 | ||
4771 | /* | |
4772 | * Load in the correct multicast list now the flags have changed. | |
4773 | */ | |
4774 | ||
b6c40d68 PM |
4775 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4776 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4777 | |
4417da66 | 4778 | dev_set_rx_mode(dev); |
1da177e4 LT |
4779 | |
4780 | /* | |
4781 | * Have we downed the interface. We handle IFF_UP ourselves | |
4782 | * according to user attempts to set it, rather than blindly | |
4783 | * setting it. | |
4784 | */ | |
4785 | ||
4786 | ret = 0; | |
4787 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 4788 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
4789 | |
4790 | if (!ret) | |
4417da66 | 4791 | dev_set_rx_mode(dev); |
1da177e4 LT |
4792 | } |
4793 | ||
1da177e4 | 4794 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
4795 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4796 | ||
1da177e4 LT |
4797 | dev->gflags ^= IFF_PROMISC; |
4798 | dev_set_promiscuity(dev, inc); | |
4799 | } | |
4800 | ||
4801 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4802 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4803 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4804 | */ | |
4805 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4806 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4807 | ||
1da177e4 LT |
4808 | dev->gflags ^= IFF_ALLMULTI; |
4809 | dev_set_allmulti(dev, inc); | |
4810 | } | |
4811 | ||
bd380811 PM |
4812 | return ret; |
4813 | } | |
4814 | ||
4815 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4816 | { | |
4817 | unsigned int changes = dev->flags ^ old_flags; | |
4818 | ||
4819 | if (changes & IFF_UP) { | |
4820 | if (dev->flags & IFF_UP) | |
4821 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4822 | else | |
4823 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4824 | } | |
4825 | ||
4826 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
4827 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
4828 | struct netdev_notifier_change_info change_info; | |
4829 | ||
4830 | change_info.flags_changed = changes; | |
4831 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
4832 | &change_info.info); | |
4833 | } | |
bd380811 PM |
4834 | } |
4835 | ||
4836 | /** | |
4837 | * dev_change_flags - change device settings | |
4838 | * @dev: device | |
4839 | * @flags: device state flags | |
4840 | * | |
4841 | * Change settings on device based state flags. The flags are | |
4842 | * in the userspace exported format. | |
4843 | */ | |
b536db93 | 4844 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 4845 | { |
b536db93 ED |
4846 | int ret; |
4847 | unsigned int changes, old_flags = dev->flags; | |
bd380811 PM |
4848 | |
4849 | ret = __dev_change_flags(dev, flags); | |
4850 | if (ret < 0) | |
4851 | return ret; | |
4852 | ||
4853 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
4854 | if (changes) |
4855 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 4856 | |
bd380811 | 4857 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
4858 | return ret; |
4859 | } | |
d1b19dff | 4860 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4861 | |
f0db275a SH |
4862 | /** |
4863 | * dev_set_mtu - Change maximum transfer unit | |
4864 | * @dev: device | |
4865 | * @new_mtu: new transfer unit | |
4866 | * | |
4867 | * Change the maximum transfer size of the network device. | |
4868 | */ | |
1da177e4 LT |
4869 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4870 | { | |
d314774c | 4871 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4872 | int err; |
4873 | ||
4874 | if (new_mtu == dev->mtu) | |
4875 | return 0; | |
4876 | ||
4877 | /* MTU must be positive. */ | |
4878 | if (new_mtu < 0) | |
4879 | return -EINVAL; | |
4880 | ||
4881 | if (!netif_device_present(dev)) | |
4882 | return -ENODEV; | |
4883 | ||
4884 | err = 0; | |
d314774c SH |
4885 | if (ops->ndo_change_mtu) |
4886 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4887 | else |
4888 | dev->mtu = new_mtu; | |
d314774c | 4889 | |
e3d8fabe | 4890 | if (!err) |
056925ab | 4891 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4892 | return err; |
4893 | } | |
d1b19dff | 4894 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4895 | |
cbda10fa VD |
4896 | /** |
4897 | * dev_set_group - Change group this device belongs to | |
4898 | * @dev: device | |
4899 | * @new_group: group this device should belong to | |
4900 | */ | |
4901 | void dev_set_group(struct net_device *dev, int new_group) | |
4902 | { | |
4903 | dev->group = new_group; | |
4904 | } | |
4905 | EXPORT_SYMBOL(dev_set_group); | |
4906 | ||
f0db275a SH |
4907 | /** |
4908 | * dev_set_mac_address - Change Media Access Control Address | |
4909 | * @dev: device | |
4910 | * @sa: new address | |
4911 | * | |
4912 | * Change the hardware (MAC) address of the device | |
4913 | */ | |
1da177e4 LT |
4914 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4915 | { | |
d314774c | 4916 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4917 | int err; |
4918 | ||
d314774c | 4919 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4920 | return -EOPNOTSUPP; |
4921 | if (sa->sa_family != dev->type) | |
4922 | return -EINVAL; | |
4923 | if (!netif_device_present(dev)) | |
4924 | return -ENODEV; | |
d314774c | 4925 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
4926 | if (err) |
4927 | return err; | |
fbdeca2d | 4928 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 4929 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 4930 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 4931 | return 0; |
1da177e4 | 4932 | } |
d1b19dff | 4933 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 4934 | |
4bf84c35 JP |
4935 | /** |
4936 | * dev_change_carrier - Change device carrier | |
4937 | * @dev: device | |
691b3b7e | 4938 | * @new_carrier: new value |
4bf84c35 JP |
4939 | * |
4940 | * Change device carrier | |
4941 | */ | |
4942 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
4943 | { | |
4944 | const struct net_device_ops *ops = dev->netdev_ops; | |
4945 | ||
4946 | if (!ops->ndo_change_carrier) | |
4947 | return -EOPNOTSUPP; | |
4948 | if (!netif_device_present(dev)) | |
4949 | return -ENODEV; | |
4950 | return ops->ndo_change_carrier(dev, new_carrier); | |
4951 | } | |
4952 | EXPORT_SYMBOL(dev_change_carrier); | |
4953 | ||
1da177e4 LT |
4954 | /** |
4955 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 4956 | * @net: the applicable net namespace |
1da177e4 LT |
4957 | * |
4958 | * Returns a suitable unique value for a new device interface | |
4959 | * number. The caller must hold the rtnl semaphore or the | |
4960 | * dev_base_lock to be sure it remains unique. | |
4961 | */ | |
881d966b | 4962 | static int dev_new_index(struct net *net) |
1da177e4 | 4963 | { |
aa79e66e | 4964 | int ifindex = net->ifindex; |
1da177e4 LT |
4965 | for (;;) { |
4966 | if (++ifindex <= 0) | |
4967 | ifindex = 1; | |
881d966b | 4968 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 4969 | return net->ifindex = ifindex; |
1da177e4 LT |
4970 | } |
4971 | } | |
4972 | ||
1da177e4 | 4973 | /* Delayed registration/unregisteration */ |
3b5b34fd | 4974 | static LIST_HEAD(net_todo_list); |
1da177e4 | 4975 | |
6f05f629 | 4976 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 4977 | { |
1da177e4 | 4978 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
4979 | } |
4980 | ||
9b5e383c | 4981 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 4982 | { |
e93737b0 | 4983 | struct net_device *dev, *tmp; |
9b5e383c | 4984 | |
93ee31f1 DL |
4985 | BUG_ON(dev_boot_phase); |
4986 | ASSERT_RTNL(); | |
4987 | ||
e93737b0 | 4988 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 4989 | /* Some devices call without registering |
e93737b0 KK |
4990 | * for initialization unwind. Remove those |
4991 | * devices and proceed with the remaining. | |
9b5e383c ED |
4992 | */ |
4993 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
4994 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
4995 | dev->name, dev); | |
93ee31f1 | 4996 | |
9b5e383c | 4997 | WARN_ON(1); |
e93737b0 KK |
4998 | list_del(&dev->unreg_list); |
4999 | continue; | |
9b5e383c | 5000 | } |
449f4544 | 5001 | dev->dismantle = true; |
9b5e383c | 5002 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5003 | } |
93ee31f1 | 5004 | |
44345724 OP |
5005 | /* If device is running, close it first. */ |
5006 | dev_close_many(head); | |
93ee31f1 | 5007 | |
44345724 | 5008 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5009 | /* And unlink it from device chain. */ |
5010 | unlist_netdevice(dev); | |
93ee31f1 | 5011 | |
9b5e383c ED |
5012 | dev->reg_state = NETREG_UNREGISTERING; |
5013 | } | |
93ee31f1 DL |
5014 | |
5015 | synchronize_net(); | |
5016 | ||
9b5e383c ED |
5017 | list_for_each_entry(dev, head, unreg_list) { |
5018 | /* Shutdown queueing discipline. */ | |
5019 | dev_shutdown(dev); | |
93ee31f1 DL |
5020 | |
5021 | ||
9b5e383c ED |
5022 | /* Notify protocols, that we are about to destroy |
5023 | this device. They should clean all the things. | |
5024 | */ | |
5025 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5026 | |
a2835763 PM |
5027 | if (!dev->rtnl_link_ops || |
5028 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5029 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
5030 | ||
9b5e383c ED |
5031 | /* |
5032 | * Flush the unicast and multicast chains | |
5033 | */ | |
a748ee24 | 5034 | dev_uc_flush(dev); |
22bedad3 | 5035 | dev_mc_flush(dev); |
93ee31f1 | 5036 | |
9b5e383c ED |
5037 | if (dev->netdev_ops->ndo_uninit) |
5038 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5039 | |
9ff162a8 JP |
5040 | /* Notifier chain MUST detach us all upper devices. */ |
5041 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 5042 | |
9b5e383c ED |
5043 | /* Remove entries from kobject tree */ |
5044 | netdev_unregister_kobject(dev); | |
024e9679 AD |
5045 | #ifdef CONFIG_XPS |
5046 | /* Remove XPS queueing entries */ | |
5047 | netif_reset_xps_queues_gt(dev, 0); | |
5048 | #endif | |
9b5e383c | 5049 | } |
93ee31f1 | 5050 | |
850a545b | 5051 | synchronize_net(); |
395264d5 | 5052 | |
a5ee1551 | 5053 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5054 | dev_put(dev); |
5055 | } | |
5056 | ||
5057 | static void rollback_registered(struct net_device *dev) | |
5058 | { | |
5059 | LIST_HEAD(single); | |
5060 | ||
5061 | list_add(&dev->unreg_list, &single); | |
5062 | rollback_registered_many(&single); | |
ceaaec98 | 5063 | list_del(&single); |
93ee31f1 DL |
5064 | } |
5065 | ||
c8f44aff MM |
5066 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
5067 | netdev_features_t features) | |
b63365a2 | 5068 | { |
57422dc5 MM |
5069 | /* Fix illegal checksum combinations */ |
5070 | if ((features & NETIF_F_HW_CSUM) && | |
5071 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5072 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
5073 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
5074 | } | |
5075 | ||
b63365a2 | 5076 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 5077 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 5078 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 5079 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
5080 | } |
5081 | ||
ec5f0615 PS |
5082 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
5083 | !(features & NETIF_F_IP_CSUM)) { | |
5084 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
5085 | features &= ~NETIF_F_TSO; | |
5086 | features &= ~NETIF_F_TSO_ECN; | |
5087 | } | |
5088 | ||
5089 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
5090 | !(features & NETIF_F_IPV6_CSUM)) { | |
5091 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
5092 | features &= ~NETIF_F_TSO6; | |
5093 | } | |
5094 | ||
31d8b9e0 BH |
5095 | /* TSO ECN requires that TSO is present as well. */ |
5096 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
5097 | features &= ~NETIF_F_TSO_ECN; | |
5098 | ||
212b573f MM |
5099 | /* Software GSO depends on SG. */ |
5100 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 5101 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
5102 | features &= ~NETIF_F_GSO; |
5103 | } | |
5104 | ||
acd1130e | 5105 | /* UFO needs SG and checksumming */ |
b63365a2 | 5106 | if (features & NETIF_F_UFO) { |
79032644 MM |
5107 | /* maybe split UFO into V4 and V6? */ |
5108 | if (!((features & NETIF_F_GEN_CSUM) || | |
5109 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
5110 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5111 | netdev_dbg(dev, |
acd1130e | 5112 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
5113 | features &= ~NETIF_F_UFO; |
5114 | } | |
5115 | ||
5116 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 5117 | netdev_dbg(dev, |
acd1130e | 5118 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
5119 | features &= ~NETIF_F_UFO; |
5120 | } | |
5121 | } | |
5122 | ||
5123 | return features; | |
5124 | } | |
b63365a2 | 5125 | |
6cb6a27c | 5126 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 5127 | { |
c8f44aff | 5128 | netdev_features_t features; |
5455c699 MM |
5129 | int err = 0; |
5130 | ||
87267485 MM |
5131 | ASSERT_RTNL(); |
5132 | ||
5455c699 MM |
5133 | features = netdev_get_wanted_features(dev); |
5134 | ||
5135 | if (dev->netdev_ops->ndo_fix_features) | |
5136 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
5137 | ||
5138 | /* driver might be less strict about feature dependencies */ | |
5139 | features = netdev_fix_features(dev, features); | |
5140 | ||
5141 | if (dev->features == features) | |
6cb6a27c | 5142 | return 0; |
5455c699 | 5143 | |
c8f44aff MM |
5144 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
5145 | &dev->features, &features); | |
5455c699 MM |
5146 | |
5147 | if (dev->netdev_ops->ndo_set_features) | |
5148 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5149 | ||
6cb6a27c | 5150 | if (unlikely(err < 0)) { |
5455c699 | 5151 | netdev_err(dev, |
c8f44aff MM |
5152 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
5153 | err, &features, &dev->features); | |
6cb6a27c MM |
5154 | return -1; |
5155 | } | |
5156 | ||
5157 | if (!err) | |
5158 | dev->features = features; | |
5159 | ||
5160 | return 1; | |
5161 | } | |
5162 | ||
afe12cc8 MM |
5163 | /** |
5164 | * netdev_update_features - recalculate device features | |
5165 | * @dev: the device to check | |
5166 | * | |
5167 | * Recalculate dev->features set and send notifications if it | |
5168 | * has changed. Should be called after driver or hardware dependent | |
5169 | * conditions might have changed that influence the features. | |
5170 | */ | |
6cb6a27c MM |
5171 | void netdev_update_features(struct net_device *dev) |
5172 | { | |
5173 | if (__netdev_update_features(dev)) | |
5174 | netdev_features_change(dev); | |
5455c699 MM |
5175 | } |
5176 | EXPORT_SYMBOL(netdev_update_features); | |
5177 | ||
afe12cc8 MM |
5178 | /** |
5179 | * netdev_change_features - recalculate device features | |
5180 | * @dev: the device to check | |
5181 | * | |
5182 | * Recalculate dev->features set and send notifications even | |
5183 | * if they have not changed. Should be called instead of | |
5184 | * netdev_update_features() if also dev->vlan_features might | |
5185 | * have changed to allow the changes to be propagated to stacked | |
5186 | * VLAN devices. | |
5187 | */ | |
5188 | void netdev_change_features(struct net_device *dev) | |
5189 | { | |
5190 | __netdev_update_features(dev); | |
5191 | netdev_features_change(dev); | |
5192 | } | |
5193 | EXPORT_SYMBOL(netdev_change_features); | |
5194 | ||
fc4a7489 PM |
5195 | /** |
5196 | * netif_stacked_transfer_operstate - transfer operstate | |
5197 | * @rootdev: the root or lower level device to transfer state from | |
5198 | * @dev: the device to transfer operstate to | |
5199 | * | |
5200 | * Transfer operational state from root to device. This is normally | |
5201 | * called when a stacking relationship exists between the root | |
5202 | * device and the device(a leaf device). | |
5203 | */ | |
5204 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5205 | struct net_device *dev) | |
5206 | { | |
5207 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5208 | netif_dormant_on(dev); | |
5209 | else | |
5210 | netif_dormant_off(dev); | |
5211 | ||
5212 | if (netif_carrier_ok(rootdev)) { | |
5213 | if (!netif_carrier_ok(dev)) | |
5214 | netif_carrier_on(dev); | |
5215 | } else { | |
5216 | if (netif_carrier_ok(dev)) | |
5217 | netif_carrier_off(dev); | |
5218 | } | |
5219 | } | |
5220 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5221 | ||
bf264145 | 5222 | #ifdef CONFIG_RPS |
1b4bf461 ED |
5223 | static int netif_alloc_rx_queues(struct net_device *dev) |
5224 | { | |
1b4bf461 | 5225 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 5226 | struct netdev_rx_queue *rx; |
1b4bf461 | 5227 | |
bd25fa7b | 5228 | BUG_ON(count < 1); |
1b4bf461 | 5229 | |
bd25fa7b | 5230 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
62b5942a | 5231 | if (!rx) |
bd25fa7b | 5232 | return -ENOMEM; |
62b5942a | 5233 | |
bd25fa7b TH |
5234 | dev->_rx = rx; |
5235 | ||
bd25fa7b | 5236 | for (i = 0; i < count; i++) |
fe822240 | 5237 | rx[i].dev = dev; |
1b4bf461 ED |
5238 | return 0; |
5239 | } | |
bf264145 | 5240 | #endif |
1b4bf461 | 5241 | |
aa942104 CG |
5242 | static void netdev_init_one_queue(struct net_device *dev, |
5243 | struct netdev_queue *queue, void *_unused) | |
5244 | { | |
5245 | /* Initialize queue lock */ | |
5246 | spin_lock_init(&queue->_xmit_lock); | |
5247 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
5248 | queue->xmit_lock_owner = -1; | |
b236da69 | 5249 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 5250 | queue->dev = dev; |
114cf580 TH |
5251 | #ifdef CONFIG_BQL |
5252 | dql_init(&queue->dql, HZ); | |
5253 | #endif | |
aa942104 CG |
5254 | } |
5255 | ||
e6484930 TH |
5256 | static int netif_alloc_netdev_queues(struct net_device *dev) |
5257 | { | |
5258 | unsigned int count = dev->num_tx_queues; | |
5259 | struct netdev_queue *tx; | |
5260 | ||
5261 | BUG_ON(count < 1); | |
5262 | ||
5263 | tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL); | |
62b5942a | 5264 | if (!tx) |
e6484930 | 5265 | return -ENOMEM; |
62b5942a | 5266 | |
e6484930 | 5267 | dev->_tx = tx; |
1d24eb48 | 5268 | |
e6484930 TH |
5269 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
5270 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
5271 | |
5272 | return 0; | |
e6484930 TH |
5273 | } |
5274 | ||
1da177e4 LT |
5275 | /** |
5276 | * register_netdevice - register a network device | |
5277 | * @dev: device to register | |
5278 | * | |
5279 | * Take a completed network device structure and add it to the kernel | |
5280 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5281 | * chain. 0 is returned on success. A negative errno code is returned | |
5282 | * on a failure to set up the device, or if the name is a duplicate. | |
5283 | * | |
5284 | * Callers must hold the rtnl semaphore. You may want | |
5285 | * register_netdev() instead of this. | |
5286 | * | |
5287 | * BUGS: | |
5288 | * The locking appears insufficient to guarantee two parallel registers | |
5289 | * will not get the same name. | |
5290 | */ | |
5291 | ||
5292 | int register_netdevice(struct net_device *dev) | |
5293 | { | |
1da177e4 | 5294 | int ret; |
d314774c | 5295 | struct net *net = dev_net(dev); |
1da177e4 LT |
5296 | |
5297 | BUG_ON(dev_boot_phase); | |
5298 | ASSERT_RTNL(); | |
5299 | ||
b17a7c17 SH |
5300 | might_sleep(); |
5301 | ||
1da177e4 LT |
5302 | /* When net_device's are persistent, this will be fatal. */ |
5303 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 5304 | BUG_ON(!net); |
1da177e4 | 5305 | |
f1f28aa3 | 5306 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 5307 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 5308 | |
1da177e4 LT |
5309 | dev->iflink = -1; |
5310 | ||
828de4f6 | 5311 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
5312 | if (ret < 0) |
5313 | goto out; | |
5314 | ||
1da177e4 | 5315 | /* Init, if this function is available */ |
d314774c SH |
5316 | if (dev->netdev_ops->ndo_init) { |
5317 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
5318 | if (ret) { |
5319 | if (ret > 0) | |
5320 | ret = -EIO; | |
90833aa4 | 5321 | goto out; |
1da177e4 LT |
5322 | } |
5323 | } | |
4ec93edb | 5324 | |
f646968f PM |
5325 | if (((dev->hw_features | dev->features) & |
5326 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
5327 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
5328 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
5329 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
5330 | ret = -EINVAL; | |
5331 | goto err_uninit; | |
5332 | } | |
5333 | ||
9c7dafbf PE |
5334 | ret = -EBUSY; |
5335 | if (!dev->ifindex) | |
5336 | dev->ifindex = dev_new_index(net); | |
5337 | else if (__dev_get_by_index(net, dev->ifindex)) | |
5338 | goto err_uninit; | |
5339 | ||
1da177e4 LT |
5340 | if (dev->iflink == -1) |
5341 | dev->iflink = dev->ifindex; | |
5342 | ||
5455c699 MM |
5343 | /* Transfer changeable features to wanted_features and enable |
5344 | * software offloads (GSO and GRO). | |
5345 | */ | |
5346 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
5347 | dev->features |= NETIF_F_SOFT_FEATURES; |
5348 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 5349 | |
c6e1a0d1 | 5350 | /* Turn on no cache copy if HW is doing checksum */ |
34324dc2 MM |
5351 | if (!(dev->flags & IFF_LOOPBACK)) { |
5352 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
5353 | if (dev->features & NETIF_F_ALL_CSUM) { | |
5354 | dev->wanted_features |= NETIF_F_NOCACHE_COPY; | |
5355 | dev->features |= NETIF_F_NOCACHE_COPY; | |
5356 | } | |
c6e1a0d1 TH |
5357 | } |
5358 | ||
1180e7d6 | 5359 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 5360 | */ |
1180e7d6 | 5361 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 5362 | |
ee579677 PS |
5363 | /* Make NETIF_F_SG inheritable to tunnel devices. |
5364 | */ | |
5365 | dev->hw_enc_features |= NETIF_F_SG; | |
5366 | ||
0d89d203 SH |
5367 | /* Make NETIF_F_SG inheritable to MPLS. |
5368 | */ | |
5369 | dev->mpls_features |= NETIF_F_SG; | |
5370 | ||
7ffbe3fd JB |
5371 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
5372 | ret = notifier_to_errno(ret); | |
5373 | if (ret) | |
5374 | goto err_uninit; | |
5375 | ||
8b41d188 | 5376 | ret = netdev_register_kobject(dev); |
b17a7c17 | 5377 | if (ret) |
7ce1b0ed | 5378 | goto err_uninit; |
b17a7c17 SH |
5379 | dev->reg_state = NETREG_REGISTERED; |
5380 | ||
6cb6a27c | 5381 | __netdev_update_features(dev); |
8e9b59b2 | 5382 | |
1da177e4 LT |
5383 | /* |
5384 | * Default initial state at registry is that the | |
5385 | * device is present. | |
5386 | */ | |
5387 | ||
5388 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5389 | ||
8f4cccbb BH |
5390 | linkwatch_init_dev(dev); |
5391 | ||
1da177e4 | 5392 | dev_init_scheduler(dev); |
1da177e4 | 5393 | dev_hold(dev); |
ce286d32 | 5394 | list_netdevice(dev); |
7bf23575 | 5395 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 5396 | |
948b337e JP |
5397 | /* If the device has permanent device address, driver should |
5398 | * set dev_addr and also addr_assign_type should be set to | |
5399 | * NET_ADDR_PERM (default value). | |
5400 | */ | |
5401 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
5402 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
5403 | ||
1da177e4 | 5404 | /* Notify protocols, that a new device appeared. */ |
056925ab | 5405 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 5406 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
5407 | if (ret) { |
5408 | rollback_registered(dev); | |
5409 | dev->reg_state = NETREG_UNREGISTERED; | |
5410 | } | |
d90a909e EB |
5411 | /* |
5412 | * Prevent userspace races by waiting until the network | |
5413 | * device is fully setup before sending notifications. | |
5414 | */ | |
a2835763 PM |
5415 | if (!dev->rtnl_link_ops || |
5416 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5417 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
5418 | |
5419 | out: | |
5420 | return ret; | |
7ce1b0ed HX |
5421 | |
5422 | err_uninit: | |
d314774c SH |
5423 | if (dev->netdev_ops->ndo_uninit) |
5424 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 5425 | goto out; |
1da177e4 | 5426 | } |
d1b19dff | 5427 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 5428 | |
937f1ba5 BH |
5429 | /** |
5430 | * init_dummy_netdev - init a dummy network device for NAPI | |
5431 | * @dev: device to init | |
5432 | * | |
5433 | * This takes a network device structure and initialize the minimum | |
5434 | * amount of fields so it can be used to schedule NAPI polls without | |
5435 | * registering a full blown interface. This is to be used by drivers | |
5436 | * that need to tie several hardware interfaces to a single NAPI | |
5437 | * poll scheduler due to HW limitations. | |
5438 | */ | |
5439 | int init_dummy_netdev(struct net_device *dev) | |
5440 | { | |
5441 | /* Clear everything. Note we don't initialize spinlocks | |
5442 | * are they aren't supposed to be taken by any of the | |
5443 | * NAPI code and this dummy netdev is supposed to be | |
5444 | * only ever used for NAPI polls | |
5445 | */ | |
5446 | memset(dev, 0, sizeof(struct net_device)); | |
5447 | ||
5448 | /* make sure we BUG if trying to hit standard | |
5449 | * register/unregister code path | |
5450 | */ | |
5451 | dev->reg_state = NETREG_DUMMY; | |
5452 | ||
937f1ba5 BH |
5453 | /* NAPI wants this */ |
5454 | INIT_LIST_HEAD(&dev->napi_list); | |
5455 | ||
5456 | /* a dummy interface is started by default */ | |
5457 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5458 | set_bit(__LINK_STATE_START, &dev->state); | |
5459 | ||
29b4433d ED |
5460 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
5461 | * because users of this 'device' dont need to change | |
5462 | * its refcount. | |
5463 | */ | |
5464 | ||
937f1ba5 BH |
5465 | return 0; |
5466 | } | |
5467 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5468 | ||
5469 | ||
1da177e4 LT |
5470 | /** |
5471 | * register_netdev - register a network device | |
5472 | * @dev: device to register | |
5473 | * | |
5474 | * Take a completed network device structure and add it to the kernel | |
5475 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5476 | * chain. 0 is returned on success. A negative errno code is returned | |
5477 | * on a failure to set up the device, or if the name is a duplicate. | |
5478 | * | |
38b4da38 | 5479 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5480 | * and expands the device name if you passed a format string to |
5481 | * alloc_netdev. | |
5482 | */ | |
5483 | int register_netdev(struct net_device *dev) | |
5484 | { | |
5485 | int err; | |
5486 | ||
5487 | rtnl_lock(); | |
1da177e4 | 5488 | err = register_netdevice(dev); |
1da177e4 LT |
5489 | rtnl_unlock(); |
5490 | return err; | |
5491 | } | |
5492 | EXPORT_SYMBOL(register_netdev); | |
5493 | ||
29b4433d ED |
5494 | int netdev_refcnt_read(const struct net_device *dev) |
5495 | { | |
5496 | int i, refcnt = 0; | |
5497 | ||
5498 | for_each_possible_cpu(i) | |
5499 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
5500 | return refcnt; | |
5501 | } | |
5502 | EXPORT_SYMBOL(netdev_refcnt_read); | |
5503 | ||
2c53040f | 5504 | /** |
1da177e4 | 5505 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 5506 | * @dev: target net_device |
1da177e4 LT |
5507 | * |
5508 | * This is called when unregistering network devices. | |
5509 | * | |
5510 | * Any protocol or device that holds a reference should register | |
5511 | * for netdevice notification, and cleanup and put back the | |
5512 | * reference if they receive an UNREGISTER event. | |
5513 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5514 | * call dev_put. |
1da177e4 LT |
5515 | */ |
5516 | static void netdev_wait_allrefs(struct net_device *dev) | |
5517 | { | |
5518 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 5519 | int refcnt; |
1da177e4 | 5520 | |
e014debe ED |
5521 | linkwatch_forget_dev(dev); |
5522 | ||
1da177e4 | 5523 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
5524 | refcnt = netdev_refcnt_read(dev); |
5525 | ||
5526 | while (refcnt != 0) { | |
1da177e4 | 5527 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 5528 | rtnl_lock(); |
1da177e4 LT |
5529 | |
5530 | /* Rebroadcast unregister notification */ | |
056925ab | 5531 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 5532 | |
748e2d93 | 5533 | __rtnl_unlock(); |
0115e8e3 | 5534 | rcu_barrier(); |
748e2d93 ED |
5535 | rtnl_lock(); |
5536 | ||
0115e8e3 | 5537 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
5538 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
5539 | &dev->state)) { | |
5540 | /* We must not have linkwatch events | |
5541 | * pending on unregister. If this | |
5542 | * happens, we simply run the queue | |
5543 | * unscheduled, resulting in a noop | |
5544 | * for this device. | |
5545 | */ | |
5546 | linkwatch_run_queue(); | |
5547 | } | |
5548 | ||
6756ae4b | 5549 | __rtnl_unlock(); |
1da177e4 LT |
5550 | |
5551 | rebroadcast_time = jiffies; | |
5552 | } | |
5553 | ||
5554 | msleep(250); | |
5555 | ||
29b4433d ED |
5556 | refcnt = netdev_refcnt_read(dev); |
5557 | ||
1da177e4 | 5558 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
5559 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
5560 | dev->name, refcnt); | |
1da177e4 LT |
5561 | warning_time = jiffies; |
5562 | } | |
5563 | } | |
5564 | } | |
5565 | ||
5566 | /* The sequence is: | |
5567 | * | |
5568 | * rtnl_lock(); | |
5569 | * ... | |
5570 | * register_netdevice(x1); | |
5571 | * register_netdevice(x2); | |
5572 | * ... | |
5573 | * unregister_netdevice(y1); | |
5574 | * unregister_netdevice(y2); | |
5575 | * ... | |
5576 | * rtnl_unlock(); | |
5577 | * free_netdev(y1); | |
5578 | * free_netdev(y2); | |
5579 | * | |
58ec3b4d | 5580 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5581 | * This allows us to deal with problems: |
b17a7c17 | 5582 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5583 | * without deadlocking with linkwatch via keventd. |
5584 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5585 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5586 | * |
5587 | * We must not return until all unregister events added during | |
5588 | * the interval the lock was held have been completed. | |
1da177e4 | 5589 | */ |
1da177e4 LT |
5590 | void netdev_run_todo(void) |
5591 | { | |
626ab0e6 | 5592 | struct list_head list; |
1da177e4 | 5593 | |
1da177e4 | 5594 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5595 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5596 | |
5597 | __rtnl_unlock(); | |
626ab0e6 | 5598 | |
0115e8e3 ED |
5599 | |
5600 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
5601 | if (!list_empty(&list)) |
5602 | rcu_barrier(); | |
5603 | ||
1da177e4 LT |
5604 | while (!list_empty(&list)) { |
5605 | struct net_device *dev | |
e5e26d75 | 5606 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
5607 | list_del(&dev->todo_list); |
5608 | ||
748e2d93 | 5609 | rtnl_lock(); |
0115e8e3 | 5610 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 5611 | __rtnl_unlock(); |
0115e8e3 | 5612 | |
b17a7c17 | 5613 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 5614 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
5615 | dev->name, dev->reg_state); |
5616 | dump_stack(); | |
5617 | continue; | |
5618 | } | |
1da177e4 | 5619 | |
b17a7c17 | 5620 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5621 | |
152102c7 | 5622 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 5623 | |
b17a7c17 | 5624 | netdev_wait_allrefs(dev); |
1da177e4 | 5625 | |
b17a7c17 | 5626 | /* paranoia */ |
29b4433d | 5627 | BUG_ON(netdev_refcnt_read(dev)); |
33d480ce ED |
5628 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
5629 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 5630 | WARN_ON(dev->dn_ptr); |
1da177e4 | 5631 | |
b17a7c17 SH |
5632 | if (dev->destructor) |
5633 | dev->destructor(dev); | |
9093bbb2 SH |
5634 | |
5635 | /* Free network device */ | |
5636 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5637 | } |
1da177e4 LT |
5638 | } |
5639 | ||
3cfde79c BH |
5640 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
5641 | * fields in the same order, with only the type differing. | |
5642 | */ | |
77a1abf5 ED |
5643 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
5644 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
5645 | { |
5646 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
5647 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
5648 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
5649 | #else |
5650 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
5651 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
5652 | u64 *dst = (u64 *)stats64; | |
5653 | ||
5654 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
5655 | sizeof(*stats64) / sizeof(u64)); | |
5656 | for (i = 0; i < n; i++) | |
5657 | dst[i] = src[i]; | |
5658 | #endif | |
5659 | } | |
77a1abf5 | 5660 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 5661 | |
eeda3fd6 SH |
5662 | /** |
5663 | * dev_get_stats - get network device statistics | |
5664 | * @dev: device to get statistics from | |
28172739 | 5665 | * @storage: place to store stats |
eeda3fd6 | 5666 | * |
d7753516 BH |
5667 | * Get network statistics from device. Return @storage. |
5668 | * The device driver may provide its own method by setting | |
5669 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
5670 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 5671 | */ |
d7753516 BH |
5672 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
5673 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 5674 | { |
eeda3fd6 SH |
5675 | const struct net_device_ops *ops = dev->netdev_ops; |
5676 | ||
28172739 ED |
5677 | if (ops->ndo_get_stats64) { |
5678 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
5679 | ops->ndo_get_stats64(dev, storage); |
5680 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 5681 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
5682 | } else { |
5683 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 5684 | } |
caf586e5 | 5685 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
28172739 | 5686 | return storage; |
c45d286e | 5687 | } |
eeda3fd6 | 5688 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5689 | |
24824a09 | 5690 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 5691 | { |
24824a09 | 5692 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 5693 | |
24824a09 ED |
5694 | #ifdef CONFIG_NET_CLS_ACT |
5695 | if (queue) | |
5696 | return queue; | |
5697 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
5698 | if (!queue) | |
5699 | return NULL; | |
5700 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
5701 | queue->qdisc = &noop_qdisc; |
5702 | queue->qdisc_sleeping = &noop_qdisc; | |
5703 | rcu_assign_pointer(dev->ingress_queue, queue); | |
5704 | #endif | |
5705 | return queue; | |
bb949fbd DM |
5706 | } |
5707 | ||
2c60db03 ED |
5708 | static const struct ethtool_ops default_ethtool_ops; |
5709 | ||
d07d7507 SG |
5710 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
5711 | const struct ethtool_ops *ops) | |
5712 | { | |
5713 | if (dev->ethtool_ops == &default_ethtool_ops) | |
5714 | dev->ethtool_ops = ops; | |
5715 | } | |
5716 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
5717 | ||
1da177e4 | 5718 | /** |
36909ea4 | 5719 | * alloc_netdev_mqs - allocate network device |
1da177e4 LT |
5720 | * @sizeof_priv: size of private data to allocate space for |
5721 | * @name: device name format string | |
5722 | * @setup: callback to initialize device | |
36909ea4 TH |
5723 | * @txqs: the number of TX subqueues to allocate |
5724 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
5725 | * |
5726 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 | 5727 | * and performs basic initialization. Also allocates subquue structs |
36909ea4 | 5728 | * for each queue on the device. |
1da177e4 | 5729 | */ |
36909ea4 TH |
5730 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
5731 | void (*setup)(struct net_device *), | |
5732 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 5733 | { |
1da177e4 | 5734 | struct net_device *dev; |
7943986c | 5735 | size_t alloc_size; |
1ce8e7b5 | 5736 | struct net_device *p; |
1da177e4 | 5737 | |
b6fe17d6 SH |
5738 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
5739 | ||
36909ea4 | 5740 | if (txqs < 1) { |
7b6cd1ce | 5741 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
5742 | return NULL; |
5743 | } | |
5744 | ||
36909ea4 TH |
5745 | #ifdef CONFIG_RPS |
5746 | if (rxqs < 1) { | |
7b6cd1ce | 5747 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
5748 | return NULL; |
5749 | } | |
5750 | #endif | |
5751 | ||
fd2ea0a7 | 5752 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
5753 | if (sizeof_priv) { |
5754 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 5755 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
5756 | alloc_size += sizeof_priv; |
5757 | } | |
5758 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 5759 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 5760 | |
31380de9 | 5761 | p = kzalloc(alloc_size, GFP_KERNEL); |
62b5942a | 5762 | if (!p) |
1da177e4 | 5763 | return NULL; |
1da177e4 | 5764 | |
1ce8e7b5 | 5765 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 5766 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 5767 | |
29b4433d ED |
5768 | dev->pcpu_refcnt = alloc_percpu(int); |
5769 | if (!dev->pcpu_refcnt) | |
e6484930 | 5770 | goto free_p; |
ab9c73cc | 5771 | |
ab9c73cc | 5772 | if (dev_addr_init(dev)) |
29b4433d | 5773 | goto free_pcpu; |
ab9c73cc | 5774 | |
22bedad3 | 5775 | dev_mc_init(dev); |
a748ee24 | 5776 | dev_uc_init(dev); |
ccffad25 | 5777 | |
c346dca1 | 5778 | dev_net_set(dev, &init_net); |
1da177e4 | 5779 | |
8d3bdbd5 | 5780 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 5781 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 5782 | |
8d3bdbd5 DM |
5783 | INIT_LIST_HEAD(&dev->napi_list); |
5784 | INIT_LIST_HEAD(&dev->unreg_list); | |
5785 | INIT_LIST_HEAD(&dev->link_watch_list); | |
9ff162a8 | 5786 | INIT_LIST_HEAD(&dev->upper_dev_list); |
8d3bdbd5 DM |
5787 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
5788 | setup(dev); | |
5789 | ||
36909ea4 TH |
5790 | dev->num_tx_queues = txqs; |
5791 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 5792 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 5793 | goto free_all; |
e8a0464c | 5794 | |
df334545 | 5795 | #ifdef CONFIG_RPS |
36909ea4 TH |
5796 | dev->num_rx_queues = rxqs; |
5797 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 5798 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 5799 | goto free_all; |
df334545 | 5800 | #endif |
0a9627f2 | 5801 | |
1da177e4 | 5802 | strcpy(dev->name, name); |
cbda10fa | 5803 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
5804 | if (!dev->ethtool_ops) |
5805 | dev->ethtool_ops = &default_ethtool_ops; | |
1da177e4 | 5806 | return dev; |
ab9c73cc | 5807 | |
8d3bdbd5 DM |
5808 | free_all: |
5809 | free_netdev(dev); | |
5810 | return NULL; | |
5811 | ||
29b4433d ED |
5812 | free_pcpu: |
5813 | free_percpu(dev->pcpu_refcnt); | |
ed9af2e8 | 5814 | kfree(dev->_tx); |
fe822240 TH |
5815 | #ifdef CONFIG_RPS |
5816 | kfree(dev->_rx); | |
5817 | #endif | |
5818 | ||
ab9c73cc JP |
5819 | free_p: |
5820 | kfree(p); | |
5821 | return NULL; | |
1da177e4 | 5822 | } |
36909ea4 | 5823 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
5824 | |
5825 | /** | |
5826 | * free_netdev - free network device | |
5827 | * @dev: device | |
5828 | * | |
4ec93edb YH |
5829 | * This function does the last stage of destroying an allocated device |
5830 | * interface. The reference to the device object is released. | |
1da177e4 LT |
5831 | * If this is the last reference then it will be freed. |
5832 | */ | |
5833 | void free_netdev(struct net_device *dev) | |
5834 | { | |
d565b0a1 HX |
5835 | struct napi_struct *p, *n; |
5836 | ||
f3005d7f DL |
5837 | release_net(dev_net(dev)); |
5838 | ||
e8a0464c | 5839 | kfree(dev->_tx); |
fe822240 TH |
5840 | #ifdef CONFIG_RPS |
5841 | kfree(dev->_rx); | |
5842 | #endif | |
e8a0464c | 5843 | |
33d480ce | 5844 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 5845 | |
f001fde5 JP |
5846 | /* Flush device addresses */ |
5847 | dev_addr_flush(dev); | |
5848 | ||
d565b0a1 HX |
5849 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
5850 | netif_napi_del(p); | |
5851 | ||
29b4433d ED |
5852 | free_percpu(dev->pcpu_refcnt); |
5853 | dev->pcpu_refcnt = NULL; | |
5854 | ||
3041a069 | 5855 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
5856 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
5857 | kfree((char *)dev - dev->padded); | |
5858 | return; | |
5859 | } | |
5860 | ||
5861 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
5862 | dev->reg_state = NETREG_RELEASED; | |
5863 | ||
43cb76d9 GKH |
5864 | /* will free via device release */ |
5865 | put_device(&dev->dev); | |
1da177e4 | 5866 | } |
d1b19dff | 5867 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 5868 | |
f0db275a SH |
5869 | /** |
5870 | * synchronize_net - Synchronize with packet receive processing | |
5871 | * | |
5872 | * Wait for packets currently being received to be done. | |
5873 | * Does not block later packets from starting. | |
5874 | */ | |
4ec93edb | 5875 | void synchronize_net(void) |
1da177e4 LT |
5876 | { |
5877 | might_sleep(); | |
be3fc413 ED |
5878 | if (rtnl_is_locked()) |
5879 | synchronize_rcu_expedited(); | |
5880 | else | |
5881 | synchronize_rcu(); | |
1da177e4 | 5882 | } |
d1b19dff | 5883 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
5884 | |
5885 | /** | |
44a0873d | 5886 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 5887 | * @dev: device |
44a0873d | 5888 | * @head: list |
6ebfbc06 | 5889 | * |
1da177e4 | 5890 | * This function shuts down a device interface and removes it |
d59b54b1 | 5891 | * from the kernel tables. |
44a0873d | 5892 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
5893 | * |
5894 | * Callers must hold the rtnl semaphore. You may want | |
5895 | * unregister_netdev() instead of this. | |
5896 | */ | |
5897 | ||
44a0873d | 5898 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 5899 | { |
a6620712 HX |
5900 | ASSERT_RTNL(); |
5901 | ||
44a0873d | 5902 | if (head) { |
9fdce099 | 5903 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
5904 | } else { |
5905 | rollback_registered(dev); | |
5906 | /* Finish processing unregister after unlock */ | |
5907 | net_set_todo(dev); | |
5908 | } | |
1da177e4 | 5909 | } |
44a0873d | 5910 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 5911 | |
9b5e383c ED |
5912 | /** |
5913 | * unregister_netdevice_many - unregister many devices | |
5914 | * @head: list of devices | |
9b5e383c ED |
5915 | */ |
5916 | void unregister_netdevice_many(struct list_head *head) | |
5917 | { | |
5918 | struct net_device *dev; | |
5919 | ||
5920 | if (!list_empty(head)) { | |
5921 | rollback_registered_many(head); | |
5922 | list_for_each_entry(dev, head, unreg_list) | |
5923 | net_set_todo(dev); | |
5924 | } | |
5925 | } | |
63c8099d | 5926 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 5927 | |
1da177e4 LT |
5928 | /** |
5929 | * unregister_netdev - remove device from the kernel | |
5930 | * @dev: device | |
5931 | * | |
5932 | * This function shuts down a device interface and removes it | |
d59b54b1 | 5933 | * from the kernel tables. |
1da177e4 LT |
5934 | * |
5935 | * This is just a wrapper for unregister_netdevice that takes | |
5936 | * the rtnl semaphore. In general you want to use this and not | |
5937 | * unregister_netdevice. | |
5938 | */ | |
5939 | void unregister_netdev(struct net_device *dev) | |
5940 | { | |
5941 | rtnl_lock(); | |
5942 | unregister_netdevice(dev); | |
5943 | rtnl_unlock(); | |
5944 | } | |
1da177e4 LT |
5945 | EXPORT_SYMBOL(unregister_netdev); |
5946 | ||
ce286d32 EB |
5947 | /** |
5948 | * dev_change_net_namespace - move device to different nethost namespace | |
5949 | * @dev: device | |
5950 | * @net: network namespace | |
5951 | * @pat: If not NULL name pattern to try if the current device name | |
5952 | * is already taken in the destination network namespace. | |
5953 | * | |
5954 | * This function shuts down a device interface and moves it | |
5955 | * to a new network namespace. On success 0 is returned, on | |
5956 | * a failure a netagive errno code is returned. | |
5957 | * | |
5958 | * Callers must hold the rtnl semaphore. | |
5959 | */ | |
5960 | ||
5961 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
5962 | { | |
ce286d32 EB |
5963 | int err; |
5964 | ||
5965 | ASSERT_RTNL(); | |
5966 | ||
5967 | /* Don't allow namespace local devices to be moved. */ | |
5968 | err = -EINVAL; | |
5969 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5970 | goto out; | |
5971 | ||
5972 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
5973 | if (dev->reg_state != NETREG_REGISTERED) |
5974 | goto out; | |
5975 | ||
5976 | /* Get out if there is nothing todo */ | |
5977 | err = 0; | |
878628fb | 5978 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
5979 | goto out; |
5980 | ||
5981 | /* Pick the destination device name, and ensure | |
5982 | * we can use it in the destination network namespace. | |
5983 | */ | |
5984 | err = -EEXIST; | |
d9031024 | 5985 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
5986 | /* We get here if we can't use the current device name */ |
5987 | if (!pat) | |
5988 | goto out; | |
828de4f6 | 5989 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
5990 | goto out; |
5991 | } | |
5992 | ||
5993 | /* | |
5994 | * And now a mini version of register_netdevice unregister_netdevice. | |
5995 | */ | |
5996 | ||
5997 | /* If device is running close it first. */ | |
9b772652 | 5998 | dev_close(dev); |
ce286d32 EB |
5999 | |
6000 | /* And unlink it from device chain */ | |
6001 | err = -ENODEV; | |
6002 | unlist_netdevice(dev); | |
6003 | ||
6004 | synchronize_net(); | |
6005 | ||
6006 | /* Shutdown queueing discipline. */ | |
6007 | dev_shutdown(dev); | |
6008 | ||
6009 | /* Notify protocols, that we are about to destroy | |
6010 | this device. They should clean all the things. | |
3b27e105 DL |
6011 | |
6012 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
6013 | This is wanted because this way 8021q and macvlan know | |
6014 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
6015 | */ |
6016 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
6017 | rcu_barrier(); |
6018 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
d2237d35 | 6019 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); |
ce286d32 EB |
6020 | |
6021 | /* | |
6022 | * Flush the unicast and multicast chains | |
6023 | */ | |
a748ee24 | 6024 | dev_uc_flush(dev); |
22bedad3 | 6025 | dev_mc_flush(dev); |
ce286d32 | 6026 | |
4e66ae2e SH |
6027 | /* Send a netdev-removed uevent to the old namespace */ |
6028 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
6029 | ||
ce286d32 | 6030 | /* Actually switch the network namespace */ |
c346dca1 | 6031 | dev_net_set(dev, net); |
ce286d32 | 6032 | |
ce286d32 EB |
6033 | /* If there is an ifindex conflict assign a new one */ |
6034 | if (__dev_get_by_index(net, dev->ifindex)) { | |
6035 | int iflink = (dev->iflink == dev->ifindex); | |
6036 | dev->ifindex = dev_new_index(net); | |
6037 | if (iflink) | |
6038 | dev->iflink = dev->ifindex; | |
6039 | } | |
6040 | ||
4e66ae2e SH |
6041 | /* Send a netdev-add uevent to the new namespace */ |
6042 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
6043 | ||
8b41d188 | 6044 | /* Fixup kobjects */ |
a1b3f594 | 6045 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 6046 | WARN_ON(err); |
ce286d32 EB |
6047 | |
6048 | /* Add the device back in the hashes */ | |
6049 | list_netdevice(dev); | |
6050 | ||
6051 | /* Notify protocols, that a new device appeared. */ | |
6052 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
6053 | ||
d90a909e EB |
6054 | /* |
6055 | * Prevent userspace races by waiting until the network | |
6056 | * device is fully setup before sending notifications. | |
6057 | */ | |
6058 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
6059 | ||
ce286d32 EB |
6060 | synchronize_net(); |
6061 | err = 0; | |
6062 | out: | |
6063 | return err; | |
6064 | } | |
463d0183 | 6065 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 6066 | |
1da177e4 LT |
6067 | static int dev_cpu_callback(struct notifier_block *nfb, |
6068 | unsigned long action, | |
6069 | void *ocpu) | |
6070 | { | |
6071 | struct sk_buff **list_skb; | |
1da177e4 LT |
6072 | struct sk_buff *skb; |
6073 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
6074 | struct softnet_data *sd, *oldsd; | |
6075 | ||
8bb78442 | 6076 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
6077 | return NOTIFY_OK; |
6078 | ||
6079 | local_irq_disable(); | |
6080 | cpu = smp_processor_id(); | |
6081 | sd = &per_cpu(softnet_data, cpu); | |
6082 | oldsd = &per_cpu(softnet_data, oldcpu); | |
6083 | ||
6084 | /* Find end of our completion_queue. */ | |
6085 | list_skb = &sd->completion_queue; | |
6086 | while (*list_skb) | |
6087 | list_skb = &(*list_skb)->next; | |
6088 | /* Append completion queue from offline CPU. */ | |
6089 | *list_skb = oldsd->completion_queue; | |
6090 | oldsd->completion_queue = NULL; | |
6091 | ||
1da177e4 | 6092 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
6093 | if (oldsd->output_queue) { |
6094 | *sd->output_queue_tailp = oldsd->output_queue; | |
6095 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
6096 | oldsd->output_queue = NULL; | |
6097 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
6098 | } | |
264524d5 HC |
6099 | /* Append NAPI poll list from offline CPU. */ |
6100 | if (!list_empty(&oldsd->poll_list)) { | |
6101 | list_splice_init(&oldsd->poll_list, &sd->poll_list); | |
6102 | raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6103 | } | |
1da177e4 LT |
6104 | |
6105 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
6106 | local_irq_enable(); | |
6107 | ||
6108 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 6109 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
1da177e4 | 6110 | netif_rx(skb); |
76cc8b13 | 6111 | input_queue_head_incr(oldsd); |
fec5e652 | 6112 | } |
76cc8b13 | 6113 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
6e7676c1 | 6114 | netif_rx(skb); |
76cc8b13 TH |
6115 | input_queue_head_incr(oldsd); |
6116 | } | |
1da177e4 LT |
6117 | |
6118 | return NOTIFY_OK; | |
6119 | } | |
1da177e4 LT |
6120 | |
6121 | ||
7f353bf2 | 6122 | /** |
b63365a2 HX |
6123 | * netdev_increment_features - increment feature set by one |
6124 | * @all: current feature set | |
6125 | * @one: new feature set | |
6126 | * @mask: mask feature set | |
7f353bf2 HX |
6127 | * |
6128 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
6129 | * @one to the master device with current feature set @all. Will not |
6130 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 6131 | */ |
c8f44aff MM |
6132 | netdev_features_t netdev_increment_features(netdev_features_t all, |
6133 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 6134 | { |
1742f183 MM |
6135 | if (mask & NETIF_F_GEN_CSUM) |
6136 | mask |= NETIF_F_ALL_CSUM; | |
6137 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 6138 | |
1742f183 MM |
6139 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
6140 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 6141 | |
1742f183 MM |
6142 | /* If one device supports hw checksumming, set for all. */ |
6143 | if (all & NETIF_F_GEN_CSUM) | |
6144 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
6145 | |
6146 | return all; | |
6147 | } | |
b63365a2 | 6148 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 6149 | |
430f03cd | 6150 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
6151 | { |
6152 | int i; | |
6153 | struct hlist_head *hash; | |
6154 | ||
6155 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
6156 | if (hash != NULL) | |
6157 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
6158 | INIT_HLIST_HEAD(&hash[i]); | |
6159 | ||
6160 | return hash; | |
6161 | } | |
6162 | ||
881d966b | 6163 | /* Initialize per network namespace state */ |
4665079c | 6164 | static int __net_init netdev_init(struct net *net) |
881d966b | 6165 | { |
734b6541 RM |
6166 | if (net != &init_net) |
6167 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 6168 | |
30d97d35 PE |
6169 | net->dev_name_head = netdev_create_hash(); |
6170 | if (net->dev_name_head == NULL) | |
6171 | goto err_name; | |
881d966b | 6172 | |
30d97d35 PE |
6173 | net->dev_index_head = netdev_create_hash(); |
6174 | if (net->dev_index_head == NULL) | |
6175 | goto err_idx; | |
881d966b EB |
6176 | |
6177 | return 0; | |
30d97d35 PE |
6178 | |
6179 | err_idx: | |
6180 | kfree(net->dev_name_head); | |
6181 | err_name: | |
6182 | return -ENOMEM; | |
881d966b EB |
6183 | } |
6184 | ||
f0db275a SH |
6185 | /** |
6186 | * netdev_drivername - network driver for the device | |
6187 | * @dev: network device | |
f0db275a SH |
6188 | * |
6189 | * Determine network driver for device. | |
6190 | */ | |
3019de12 | 6191 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 6192 | { |
cf04a4c7 SH |
6193 | const struct device_driver *driver; |
6194 | const struct device *parent; | |
3019de12 | 6195 | const char *empty = ""; |
6579e57b AV |
6196 | |
6197 | parent = dev->dev.parent; | |
6579e57b | 6198 | if (!parent) |
3019de12 | 6199 | return empty; |
6579e57b AV |
6200 | |
6201 | driver = parent->driver; | |
6202 | if (driver && driver->name) | |
3019de12 DM |
6203 | return driver->name; |
6204 | return empty; | |
6579e57b AV |
6205 | } |
6206 | ||
b004ff49 | 6207 | static int __netdev_printk(const char *level, const struct net_device *dev, |
256df2f3 JP |
6208 | struct va_format *vaf) |
6209 | { | |
6210 | int r; | |
6211 | ||
b004ff49 | 6212 | if (dev && dev->dev.parent) { |
666f355f JP |
6213 | r = dev_printk_emit(level[1] - '0', |
6214 | dev->dev.parent, | |
6215 | "%s %s %s: %pV", | |
6216 | dev_driver_string(dev->dev.parent), | |
6217 | dev_name(dev->dev.parent), | |
6218 | netdev_name(dev), vaf); | |
b004ff49 | 6219 | } else if (dev) { |
256df2f3 | 6220 | r = printk("%s%s: %pV", level, netdev_name(dev), vaf); |
b004ff49 | 6221 | } else { |
256df2f3 | 6222 | r = printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 6223 | } |
256df2f3 JP |
6224 | |
6225 | return r; | |
6226 | } | |
6227 | ||
6228 | int netdev_printk(const char *level, const struct net_device *dev, | |
6229 | const char *format, ...) | |
6230 | { | |
6231 | struct va_format vaf; | |
6232 | va_list args; | |
6233 | int r; | |
6234 | ||
6235 | va_start(args, format); | |
6236 | ||
6237 | vaf.fmt = format; | |
6238 | vaf.va = &args; | |
6239 | ||
6240 | r = __netdev_printk(level, dev, &vaf); | |
b004ff49 | 6241 | |
256df2f3 JP |
6242 | va_end(args); |
6243 | ||
6244 | return r; | |
6245 | } | |
6246 | EXPORT_SYMBOL(netdev_printk); | |
6247 | ||
6248 | #define define_netdev_printk_level(func, level) \ | |
6249 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
6250 | { \ | |
6251 | int r; \ | |
6252 | struct va_format vaf; \ | |
6253 | va_list args; \ | |
6254 | \ | |
6255 | va_start(args, fmt); \ | |
6256 | \ | |
6257 | vaf.fmt = fmt; \ | |
6258 | vaf.va = &args; \ | |
6259 | \ | |
6260 | r = __netdev_printk(level, dev, &vaf); \ | |
b004ff49 | 6261 | \ |
256df2f3 JP |
6262 | va_end(args); \ |
6263 | \ | |
6264 | return r; \ | |
6265 | } \ | |
6266 | EXPORT_SYMBOL(func); | |
6267 | ||
6268 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
6269 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
6270 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
6271 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
6272 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
6273 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
6274 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
6275 | ||
4665079c | 6276 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
6277 | { |
6278 | kfree(net->dev_name_head); | |
6279 | kfree(net->dev_index_head); | |
6280 | } | |
6281 | ||
022cbae6 | 6282 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
6283 | .init = netdev_init, |
6284 | .exit = netdev_exit, | |
6285 | }; | |
6286 | ||
4665079c | 6287 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 6288 | { |
e008b5fc | 6289 | struct net_device *dev, *aux; |
ce286d32 | 6290 | /* |
e008b5fc | 6291 | * Push all migratable network devices back to the |
ce286d32 EB |
6292 | * initial network namespace |
6293 | */ | |
6294 | rtnl_lock(); | |
e008b5fc | 6295 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 6296 | int err; |
aca51397 | 6297 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
6298 | |
6299 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6300 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6301 | continue; | |
6302 | ||
e008b5fc EB |
6303 | /* Leave virtual devices for the generic cleanup */ |
6304 | if (dev->rtnl_link_ops) | |
6305 | continue; | |
d0c082ce | 6306 | |
25985edc | 6307 | /* Push remaining network devices to init_net */ |
aca51397 PE |
6308 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
6309 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 6310 | if (err) { |
7b6cd1ce JP |
6311 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
6312 | __func__, dev->name, err); | |
aca51397 | 6313 | BUG(); |
ce286d32 EB |
6314 | } |
6315 | } | |
6316 | rtnl_unlock(); | |
6317 | } | |
6318 | ||
04dc7f6b EB |
6319 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
6320 | { | |
6321 | /* At exit all network devices most be removed from a network | |
b595076a | 6322 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
6323 | * Do this across as many network namespaces as possible to |
6324 | * improve batching efficiency. | |
6325 | */ | |
6326 | struct net_device *dev; | |
6327 | struct net *net; | |
6328 | LIST_HEAD(dev_kill_list); | |
6329 | ||
6330 | rtnl_lock(); | |
6331 | list_for_each_entry(net, net_list, exit_list) { | |
6332 | for_each_netdev_reverse(net, dev) { | |
6333 | if (dev->rtnl_link_ops) | |
6334 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
6335 | else | |
6336 | unregister_netdevice_queue(dev, &dev_kill_list); | |
6337 | } | |
6338 | } | |
6339 | unregister_netdevice_many(&dev_kill_list); | |
ceaaec98 | 6340 | list_del(&dev_kill_list); |
04dc7f6b EB |
6341 | rtnl_unlock(); |
6342 | } | |
6343 | ||
022cbae6 | 6344 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 6345 | .exit = default_device_exit, |
04dc7f6b | 6346 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
6347 | }; |
6348 | ||
1da177e4 LT |
6349 | /* |
6350 | * Initialize the DEV module. At boot time this walks the device list and | |
6351 | * unhooks any devices that fail to initialise (normally hardware not | |
6352 | * present) and leaves us with a valid list of present and active devices. | |
6353 | * | |
6354 | */ | |
6355 | ||
6356 | /* | |
6357 | * This is called single threaded during boot, so no need | |
6358 | * to take the rtnl semaphore. | |
6359 | */ | |
6360 | static int __init net_dev_init(void) | |
6361 | { | |
6362 | int i, rc = -ENOMEM; | |
6363 | ||
6364 | BUG_ON(!dev_boot_phase); | |
6365 | ||
1da177e4 LT |
6366 | if (dev_proc_init()) |
6367 | goto out; | |
6368 | ||
8b41d188 | 6369 | if (netdev_kobject_init()) |
1da177e4 LT |
6370 | goto out; |
6371 | ||
6372 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 6373 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
6374 | INIT_LIST_HEAD(&ptype_base[i]); |
6375 | ||
62532da9 VY |
6376 | INIT_LIST_HEAD(&offload_base); |
6377 | ||
881d966b EB |
6378 | if (register_pernet_subsys(&netdev_net_ops)) |
6379 | goto out; | |
1da177e4 LT |
6380 | |
6381 | /* | |
6382 | * Initialise the packet receive queues. | |
6383 | */ | |
6384 | ||
6f912042 | 6385 | for_each_possible_cpu(i) { |
e36fa2f7 | 6386 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 6387 | |
dee42870 | 6388 | memset(sd, 0, sizeof(*sd)); |
e36fa2f7 | 6389 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 6390 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 ED |
6391 | sd->completion_queue = NULL; |
6392 | INIT_LIST_HEAD(&sd->poll_list); | |
a9cbd588 CG |
6393 | sd->output_queue = NULL; |
6394 | sd->output_queue_tailp = &sd->output_queue; | |
df334545 | 6395 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
6396 | sd->csd.func = rps_trigger_softirq; |
6397 | sd->csd.info = sd; | |
6398 | sd->csd.flags = 0; | |
6399 | sd->cpu = i; | |
1e94d72f | 6400 | #endif |
0a9627f2 | 6401 | |
e36fa2f7 ED |
6402 | sd->backlog.poll = process_backlog; |
6403 | sd->backlog.weight = weight_p; | |
6404 | sd->backlog.gro_list = NULL; | |
6405 | sd->backlog.gro_count = 0; | |
99bbc707 WB |
6406 | |
6407 | #ifdef CONFIG_NET_FLOW_LIMIT | |
6408 | sd->flow_limit = NULL; | |
6409 | #endif | |
1da177e4 LT |
6410 | } |
6411 | ||
1da177e4 LT |
6412 | dev_boot_phase = 0; |
6413 | ||
505d4f73 EB |
6414 | /* The loopback device is special if any other network devices |
6415 | * is present in a network namespace the loopback device must | |
6416 | * be present. Since we now dynamically allocate and free the | |
6417 | * loopback device ensure this invariant is maintained by | |
6418 | * keeping the loopback device as the first device on the | |
6419 | * list of network devices. Ensuring the loopback devices | |
6420 | * is the first device that appears and the last network device | |
6421 | * that disappears. | |
6422 | */ | |
6423 | if (register_pernet_device(&loopback_net_ops)) | |
6424 | goto out; | |
6425 | ||
6426 | if (register_pernet_device(&default_device_ops)) | |
6427 | goto out; | |
6428 | ||
962cf36c CM |
6429 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
6430 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
6431 | |
6432 | hotcpu_notifier(dev_cpu_callback, 0); | |
6433 | dst_init(); | |
1da177e4 LT |
6434 | rc = 0; |
6435 | out: | |
6436 | return rc; | |
6437 | } | |
6438 | ||
6439 | subsys_initcall(net_dev_init); |