Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * NET3 Protocol independent device support routines. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Derived from the non IP parts of dev.c 1.0.19 | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * | |
14 | * Additional Authors: | |
15 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
16 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
17 | * David Hinds <dahinds@users.sourceforge.net> | |
18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
19 | * Adam Sulmicki <adam@cfar.umd.edu> | |
20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
21 | * | |
22 | * Changes: | |
23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
24 | * to 2 if register_netdev gets called | |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
27 | * Alan Cox : device private ioctl copies fields back. | |
28 | * Alan Cox : Transmit queue code does relevant | |
29 | * stunts to keep the queue safe. | |
30 | * Alan Cox : Fixed double lock. | |
31 | * Alan Cox : Fixed promisc NULL pointer trap | |
32 | * ???????? : Support the full private ioctl range | |
33 | * Alan Cox : Moved ioctl permission check into | |
34 | * drivers | |
35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
36 | * Alan Cox : 100 backlog just doesn't cut it when | |
37 | * you start doing multicast video 8) | |
38 | * Alan Cox : Rewrote net_bh and list manager. | |
39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. | |
40 | * Alan Cox : Took out transmit every packet pass | |
41 | * Saved a few bytes in the ioctl handler | |
42 | * Alan Cox : Network driver sets packet type before | |
43 | * calling netif_rx. Saves a function | |
44 | * call a packet. | |
45 | * Alan Cox : Hashed net_bh() | |
46 | * Richard Kooijman: Timestamp fixes. | |
47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
48 | * Alan Cox : Device lock protection. | |
49 | * Alan Cox : Fixed nasty side effect of device close | |
50 | * changes. | |
51 | * Rudi Cilibrasi : Pass the right thing to | |
52 | * set_mac_address() | |
53 | * Dave Miller : 32bit quantity for the device lock to | |
54 | * make it work out on a Sparc. | |
55 | * Bjorn Ekwall : Added KERNELD hack. | |
56 | * Alan Cox : Cleaned up the backlog initialise. | |
57 | * Craig Metz : SIOCGIFCONF fix if space for under | |
58 | * 1 device. | |
59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
60 | * is no device open function. | |
61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
63 | * Cyrus Durgin : Cleaned for KMOD | |
64 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
65 | * A network device unload needs to purge | |
66 | * the backlog queue. | |
67 | * Paul Rusty Russell : SIOCSIFNAME | |
68 | * Pekka Riikonen : Netdev boot-time settings code | |
69 | * Andrew Morton : Make unregister_netdevice wait | |
70 | * indefinitely on dev->refcnt | |
71 | * J Hadi Salim : - Backlog queue sampling | |
72 | * - netif_rx() feedback | |
73 | */ | |
74 | ||
75 | #include <asm/uaccess.h> | |
1da177e4 | 76 | #include <linux/bitops.h> |
4fc268d2 | 77 | #include <linux/capability.h> |
1da177e4 LT |
78 | #include <linux/cpu.h> |
79 | #include <linux/types.h> | |
80 | #include <linux/kernel.h> | |
08e9897d | 81 | #include <linux/hash.h> |
5a0e3ad6 | 82 | #include <linux/slab.h> |
1da177e4 | 83 | #include <linux/sched.h> |
4a3e2f71 | 84 | #include <linux/mutex.h> |
1da177e4 LT |
85 | #include <linux/string.h> |
86 | #include <linux/mm.h> | |
87 | #include <linux/socket.h> | |
88 | #include <linux/sockios.h> | |
89 | #include <linux/errno.h> | |
90 | #include <linux/interrupt.h> | |
91 | #include <linux/if_ether.h> | |
92 | #include <linux/netdevice.h> | |
93 | #include <linux/etherdevice.h> | |
0187bdfb | 94 | #include <linux/ethtool.h> |
1da177e4 LT |
95 | #include <linux/notifier.h> |
96 | #include <linux/skbuff.h> | |
457c4cbc | 97 | #include <net/net_namespace.h> |
1da177e4 LT |
98 | #include <net/sock.h> |
99 | #include <linux/rtnetlink.h> | |
1da177e4 | 100 | #include <linux/stat.h> |
1da177e4 LT |
101 | #include <net/dst.h> |
102 | #include <net/pkt_sched.h> | |
103 | #include <net/checksum.h> | |
44540960 | 104 | #include <net/xfrm.h> |
1da177e4 LT |
105 | #include <linux/highmem.h> |
106 | #include <linux/init.h> | |
1da177e4 | 107 | #include <linux/module.h> |
1da177e4 LT |
108 | #include <linux/netpoll.h> |
109 | #include <linux/rcupdate.h> | |
110 | #include <linux/delay.h> | |
1da177e4 | 111 | #include <net/iw_handler.h> |
1da177e4 | 112 | #include <asm/current.h> |
5bdb9886 | 113 | #include <linux/audit.h> |
db217334 | 114 | #include <linux/dmaengine.h> |
f6a78bfc | 115 | #include <linux/err.h> |
c7fa9d18 | 116 | #include <linux/ctype.h> |
723e98b7 | 117 | #include <linux/if_arp.h> |
6de329e2 | 118 | #include <linux/if_vlan.h> |
8f0f2223 | 119 | #include <linux/ip.h> |
ad55dcaf | 120 | #include <net/ip.h> |
8f0f2223 DM |
121 | #include <linux/ipv6.h> |
122 | #include <linux/in.h> | |
b6b2fed1 DM |
123 | #include <linux/jhash.h> |
124 | #include <linux/random.h> | |
9cbc1cb8 | 125 | #include <trace/events/napi.h> |
cf66ba58 | 126 | #include <trace/events/net.h> |
07dc22e7 | 127 | #include <trace/events/skb.h> |
5acbbd42 | 128 | #include <linux/pci.h> |
caeda9b9 | 129 | #include <linux/inetdevice.h> |
c445477d | 130 | #include <linux/cpu_rmap.h> |
c5905afb | 131 | #include <linux/static_key.h> |
af12fa6e | 132 | #include <linux/hashtable.h> |
60877a32 | 133 | #include <linux/vmalloc.h> |
529d0489 | 134 | #include <linux/if_macvlan.h> |
e7fd2885 | 135 | #include <linux/errqueue.h> |
1da177e4 | 136 | |
342709ef PE |
137 | #include "net-sysfs.h" |
138 | ||
d565b0a1 HX |
139 | /* Instead of increasing this, you should create a hash table. */ |
140 | #define MAX_GRO_SKBS 8 | |
141 | ||
5d38a079 HX |
142 | /* This should be increased if a protocol with a bigger head is added. */ |
143 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
144 | ||
1da177e4 | 145 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 146 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
147 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
148 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 149 | static struct list_head offload_base __read_mostly; |
1da177e4 | 150 | |
ae78dbfa | 151 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
152 | static int call_netdevice_notifiers_info(unsigned long val, |
153 | struct net_device *dev, | |
154 | struct netdev_notifier_info *info); | |
ae78dbfa | 155 | |
1da177e4 | 156 | /* |
7562f876 | 157 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
158 | * semaphore. |
159 | * | |
c6d14c84 | 160 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
161 | * |
162 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 163 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
164 | * actual updates. This allows pure readers to access the list even |
165 | * while a writer is preparing to update it. | |
166 | * | |
167 | * To put it another way, dev_base_lock is held for writing only to | |
168 | * protect against pure readers; the rtnl semaphore provides the | |
169 | * protection against other writers. | |
170 | * | |
171 | * See, for example usages, register_netdevice() and | |
172 | * unregister_netdevice(), which must be called with the rtnl | |
173 | * semaphore held. | |
174 | */ | |
1da177e4 | 175 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
176 | EXPORT_SYMBOL(dev_base_lock); |
177 | ||
af12fa6e ET |
178 | /* protects napi_hash addition/deletion and napi_gen_id */ |
179 | static DEFINE_SPINLOCK(napi_hash_lock); | |
180 | ||
181 | static unsigned int napi_gen_id; | |
182 | static DEFINE_HASHTABLE(napi_hash, 8); | |
183 | ||
18afa4b0 | 184 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 185 | |
4e985ada TG |
186 | static inline void dev_base_seq_inc(struct net *net) |
187 | { | |
188 | while (++net->dev_base_seq == 0); | |
189 | } | |
190 | ||
881d966b | 191 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 192 | { |
95c96174 ED |
193 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
194 | ||
08e9897d | 195 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
196 | } |
197 | ||
881d966b | 198 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 199 | { |
7c28bd0b | 200 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
201 | } |
202 | ||
e36fa2f7 | 203 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
204 | { |
205 | #ifdef CONFIG_RPS | |
e36fa2f7 | 206 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
207 | #endif |
208 | } | |
209 | ||
e36fa2f7 | 210 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
211 | { |
212 | #ifdef CONFIG_RPS | |
e36fa2f7 | 213 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
214 | #endif |
215 | } | |
216 | ||
ce286d32 | 217 | /* Device list insertion */ |
53759be9 | 218 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 219 | { |
c346dca1 | 220 | struct net *net = dev_net(dev); |
ce286d32 EB |
221 | |
222 | ASSERT_RTNL(); | |
223 | ||
224 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 225 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 226 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
227 | hlist_add_head_rcu(&dev->index_hlist, |
228 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 229 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
230 | |
231 | dev_base_seq_inc(net); | |
ce286d32 EB |
232 | } |
233 | ||
fb699dfd ED |
234 | /* Device list removal |
235 | * caller must respect a RCU grace period before freeing/reusing dev | |
236 | */ | |
ce286d32 EB |
237 | static void unlist_netdevice(struct net_device *dev) |
238 | { | |
239 | ASSERT_RTNL(); | |
240 | ||
241 | /* Unlink dev from the device chain */ | |
242 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 243 | list_del_rcu(&dev->dev_list); |
72c9528b | 244 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 245 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 246 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
247 | |
248 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
249 | } |
250 | ||
1da177e4 LT |
251 | /* |
252 | * Our notifier list | |
253 | */ | |
254 | ||
f07d5b94 | 255 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
256 | |
257 | /* | |
258 | * Device drivers call our routines to queue packets here. We empty the | |
259 | * queue in the local softnet handler. | |
260 | */ | |
bea3348e | 261 | |
9958da05 | 262 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 263 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 264 | |
cf508b12 | 265 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 266 | /* |
c773e847 | 267 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
268 | * according to dev->type |
269 | */ | |
270 | static const unsigned short netdev_lock_type[] = | |
271 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
272 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
273 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
274 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
275 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
276 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
277 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
278 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
279 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
280 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
281 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
282 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
283 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
284 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
285 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 286 | |
36cbd3dc | 287 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
288 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
289 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
290 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
291 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
292 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
293 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
294 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
295 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
296 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
297 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
298 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
299 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
300 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
301 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
302 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
303 | |
304 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 305 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
306 | |
307 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
308 | { | |
309 | int i; | |
310 | ||
311 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
312 | if (netdev_lock_type[i] == dev_type) | |
313 | return i; | |
314 | /* the last key is used by default */ | |
315 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
316 | } | |
317 | ||
cf508b12 DM |
318 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
319 | unsigned short dev_type) | |
723e98b7 JP |
320 | { |
321 | int i; | |
322 | ||
323 | i = netdev_lock_pos(dev_type); | |
324 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
325 | netdev_lock_name[i]); | |
326 | } | |
cf508b12 DM |
327 | |
328 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
329 | { | |
330 | int i; | |
331 | ||
332 | i = netdev_lock_pos(dev->type); | |
333 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
334 | &netdev_addr_lock_key[i], | |
335 | netdev_lock_name[i]); | |
336 | } | |
723e98b7 | 337 | #else |
cf508b12 DM |
338 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
339 | unsigned short dev_type) | |
340 | { | |
341 | } | |
342 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
343 | { |
344 | } | |
345 | #endif | |
1da177e4 LT |
346 | |
347 | /******************************************************************************* | |
348 | ||
349 | Protocol management and registration routines | |
350 | ||
351 | *******************************************************************************/ | |
352 | ||
1da177e4 LT |
353 | /* |
354 | * Add a protocol ID to the list. Now that the input handler is | |
355 | * smarter we can dispense with all the messy stuff that used to be | |
356 | * here. | |
357 | * | |
358 | * BEWARE!!! Protocol handlers, mangling input packets, | |
359 | * MUST BE last in hash buckets and checking protocol handlers | |
360 | * MUST start from promiscuous ptype_all chain in net_bh. | |
361 | * It is true now, do not change it. | |
362 | * Explanation follows: if protocol handler, mangling packet, will | |
363 | * be the first on list, it is not able to sense, that packet | |
364 | * is cloned and should be copied-on-write, so that it will | |
365 | * change it and subsequent readers will get broken packet. | |
366 | * --ANK (980803) | |
367 | */ | |
368 | ||
c07b68e8 ED |
369 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
370 | { | |
371 | if (pt->type == htons(ETH_P_ALL)) | |
372 | return &ptype_all; | |
373 | else | |
374 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
375 | } | |
376 | ||
1da177e4 LT |
377 | /** |
378 | * dev_add_pack - add packet handler | |
379 | * @pt: packet type declaration | |
380 | * | |
381 | * Add a protocol handler to the networking stack. The passed &packet_type | |
382 | * is linked into kernel lists and may not be freed until it has been | |
383 | * removed from the kernel lists. | |
384 | * | |
4ec93edb | 385 | * This call does not sleep therefore it can not |
1da177e4 LT |
386 | * guarantee all CPU's that are in middle of receiving packets |
387 | * will see the new packet type (until the next received packet). | |
388 | */ | |
389 | ||
390 | void dev_add_pack(struct packet_type *pt) | |
391 | { | |
c07b68e8 | 392 | struct list_head *head = ptype_head(pt); |
1da177e4 | 393 | |
c07b68e8 ED |
394 | spin_lock(&ptype_lock); |
395 | list_add_rcu(&pt->list, head); | |
396 | spin_unlock(&ptype_lock); | |
1da177e4 | 397 | } |
d1b19dff | 398 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 399 | |
1da177e4 LT |
400 | /** |
401 | * __dev_remove_pack - remove packet handler | |
402 | * @pt: packet type declaration | |
403 | * | |
404 | * Remove a protocol handler that was previously added to the kernel | |
405 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
406 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 407 | * returns. |
1da177e4 LT |
408 | * |
409 | * The packet type might still be in use by receivers | |
410 | * and must not be freed until after all the CPU's have gone | |
411 | * through a quiescent state. | |
412 | */ | |
413 | void __dev_remove_pack(struct packet_type *pt) | |
414 | { | |
c07b68e8 | 415 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
416 | struct packet_type *pt1; |
417 | ||
c07b68e8 | 418 | spin_lock(&ptype_lock); |
1da177e4 LT |
419 | |
420 | list_for_each_entry(pt1, head, list) { | |
421 | if (pt == pt1) { | |
422 | list_del_rcu(&pt->list); | |
423 | goto out; | |
424 | } | |
425 | } | |
426 | ||
7b6cd1ce | 427 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 428 | out: |
c07b68e8 | 429 | spin_unlock(&ptype_lock); |
1da177e4 | 430 | } |
d1b19dff ED |
431 | EXPORT_SYMBOL(__dev_remove_pack); |
432 | ||
1da177e4 LT |
433 | /** |
434 | * dev_remove_pack - remove packet handler | |
435 | * @pt: packet type declaration | |
436 | * | |
437 | * Remove a protocol handler that was previously added to the kernel | |
438 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
439 | * from the kernel lists and can be freed or reused once this function | |
440 | * returns. | |
441 | * | |
442 | * This call sleeps to guarantee that no CPU is looking at the packet | |
443 | * type after return. | |
444 | */ | |
445 | void dev_remove_pack(struct packet_type *pt) | |
446 | { | |
447 | __dev_remove_pack(pt); | |
4ec93edb | 448 | |
1da177e4 LT |
449 | synchronize_net(); |
450 | } | |
d1b19dff | 451 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 452 | |
62532da9 VY |
453 | |
454 | /** | |
455 | * dev_add_offload - register offload handlers | |
456 | * @po: protocol offload declaration | |
457 | * | |
458 | * Add protocol offload handlers to the networking stack. The passed | |
459 | * &proto_offload is linked into kernel lists and may not be freed until | |
460 | * it has been removed from the kernel lists. | |
461 | * | |
462 | * This call does not sleep therefore it can not | |
463 | * guarantee all CPU's that are in middle of receiving packets | |
464 | * will see the new offload handlers (until the next received packet). | |
465 | */ | |
466 | void dev_add_offload(struct packet_offload *po) | |
467 | { | |
468 | struct list_head *head = &offload_base; | |
469 | ||
470 | spin_lock(&offload_lock); | |
471 | list_add_rcu(&po->list, head); | |
472 | spin_unlock(&offload_lock); | |
473 | } | |
474 | EXPORT_SYMBOL(dev_add_offload); | |
475 | ||
476 | /** | |
477 | * __dev_remove_offload - remove offload handler | |
478 | * @po: packet offload declaration | |
479 | * | |
480 | * Remove a protocol offload handler that was previously added to the | |
481 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
482 | * is removed from the kernel lists and can be freed or reused once this | |
483 | * function returns. | |
484 | * | |
485 | * The packet type might still be in use by receivers | |
486 | * and must not be freed until after all the CPU's have gone | |
487 | * through a quiescent state. | |
488 | */ | |
1d143d9f | 489 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
490 | { |
491 | struct list_head *head = &offload_base; | |
492 | struct packet_offload *po1; | |
493 | ||
c53aa505 | 494 | spin_lock(&offload_lock); |
62532da9 VY |
495 | |
496 | list_for_each_entry(po1, head, list) { | |
497 | if (po == po1) { | |
498 | list_del_rcu(&po->list); | |
499 | goto out; | |
500 | } | |
501 | } | |
502 | ||
503 | pr_warn("dev_remove_offload: %p not found\n", po); | |
504 | out: | |
c53aa505 | 505 | spin_unlock(&offload_lock); |
62532da9 | 506 | } |
62532da9 VY |
507 | |
508 | /** | |
509 | * dev_remove_offload - remove packet offload handler | |
510 | * @po: packet offload declaration | |
511 | * | |
512 | * Remove a packet offload handler that was previously added to the kernel | |
513 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
514 | * removed from the kernel lists and can be freed or reused once this | |
515 | * function returns. | |
516 | * | |
517 | * This call sleeps to guarantee that no CPU is looking at the packet | |
518 | * type after return. | |
519 | */ | |
520 | void dev_remove_offload(struct packet_offload *po) | |
521 | { | |
522 | __dev_remove_offload(po); | |
523 | ||
524 | synchronize_net(); | |
525 | } | |
526 | EXPORT_SYMBOL(dev_remove_offload); | |
527 | ||
1da177e4 LT |
528 | /****************************************************************************** |
529 | ||
530 | Device Boot-time Settings Routines | |
531 | ||
532 | *******************************************************************************/ | |
533 | ||
534 | /* Boot time configuration table */ | |
535 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
536 | ||
537 | /** | |
538 | * netdev_boot_setup_add - add new setup entry | |
539 | * @name: name of the device | |
540 | * @map: configured settings for the device | |
541 | * | |
542 | * Adds new setup entry to the dev_boot_setup list. The function | |
543 | * returns 0 on error and 1 on success. This is a generic routine to | |
544 | * all netdevices. | |
545 | */ | |
546 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
547 | { | |
548 | struct netdev_boot_setup *s; | |
549 | int i; | |
550 | ||
551 | s = dev_boot_setup; | |
552 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
553 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
554 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 555 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
556 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
557 | break; | |
558 | } | |
559 | } | |
560 | ||
561 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
562 | } | |
563 | ||
564 | /** | |
565 | * netdev_boot_setup_check - check boot time settings | |
566 | * @dev: the netdevice | |
567 | * | |
568 | * Check boot time settings for the device. | |
569 | * The found settings are set for the device to be used | |
570 | * later in the device probing. | |
571 | * Returns 0 if no settings found, 1 if they are. | |
572 | */ | |
573 | int netdev_boot_setup_check(struct net_device *dev) | |
574 | { | |
575 | struct netdev_boot_setup *s = dev_boot_setup; | |
576 | int i; | |
577 | ||
578 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
579 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 580 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
581 | dev->irq = s[i].map.irq; |
582 | dev->base_addr = s[i].map.base_addr; | |
583 | dev->mem_start = s[i].map.mem_start; | |
584 | dev->mem_end = s[i].map.mem_end; | |
585 | return 1; | |
586 | } | |
587 | } | |
588 | return 0; | |
589 | } | |
d1b19dff | 590 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
591 | |
592 | ||
593 | /** | |
594 | * netdev_boot_base - get address from boot time settings | |
595 | * @prefix: prefix for network device | |
596 | * @unit: id for network device | |
597 | * | |
598 | * Check boot time settings for the base address of device. | |
599 | * The found settings are set for the device to be used | |
600 | * later in the device probing. | |
601 | * Returns 0 if no settings found. | |
602 | */ | |
603 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
604 | { | |
605 | const struct netdev_boot_setup *s = dev_boot_setup; | |
606 | char name[IFNAMSIZ]; | |
607 | int i; | |
608 | ||
609 | sprintf(name, "%s%d", prefix, unit); | |
610 | ||
611 | /* | |
612 | * If device already registered then return base of 1 | |
613 | * to indicate not to probe for this interface | |
614 | */ | |
881d966b | 615 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
616 | return 1; |
617 | ||
618 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
619 | if (!strcmp(name, s[i].name)) | |
620 | return s[i].map.base_addr; | |
621 | return 0; | |
622 | } | |
623 | ||
624 | /* | |
625 | * Saves at boot time configured settings for any netdevice. | |
626 | */ | |
627 | int __init netdev_boot_setup(char *str) | |
628 | { | |
629 | int ints[5]; | |
630 | struct ifmap map; | |
631 | ||
632 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
633 | if (!str || !*str) | |
634 | return 0; | |
635 | ||
636 | /* Save settings */ | |
637 | memset(&map, 0, sizeof(map)); | |
638 | if (ints[0] > 0) | |
639 | map.irq = ints[1]; | |
640 | if (ints[0] > 1) | |
641 | map.base_addr = ints[2]; | |
642 | if (ints[0] > 2) | |
643 | map.mem_start = ints[3]; | |
644 | if (ints[0] > 3) | |
645 | map.mem_end = ints[4]; | |
646 | ||
647 | /* Add new entry to the list */ | |
648 | return netdev_boot_setup_add(str, &map); | |
649 | } | |
650 | ||
651 | __setup("netdev=", netdev_boot_setup); | |
652 | ||
653 | /******************************************************************************* | |
654 | ||
655 | Device Interface Subroutines | |
656 | ||
657 | *******************************************************************************/ | |
658 | ||
659 | /** | |
660 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 661 | * @net: the applicable net namespace |
1da177e4 LT |
662 | * @name: name to find |
663 | * | |
664 | * Find an interface by name. Must be called under RTNL semaphore | |
665 | * or @dev_base_lock. If the name is found a pointer to the device | |
666 | * is returned. If the name is not found then %NULL is returned. The | |
667 | * reference counters are not incremented so the caller must be | |
668 | * careful with locks. | |
669 | */ | |
670 | ||
881d966b | 671 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 672 | { |
0bd8d536 ED |
673 | struct net_device *dev; |
674 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 675 | |
b67bfe0d | 676 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
677 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
678 | return dev; | |
0bd8d536 | 679 | |
1da177e4 LT |
680 | return NULL; |
681 | } | |
d1b19dff | 682 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 683 | |
72c9528b ED |
684 | /** |
685 | * dev_get_by_name_rcu - find a device by its name | |
686 | * @net: the applicable net namespace | |
687 | * @name: name to find | |
688 | * | |
689 | * Find an interface by name. | |
690 | * If the name is found a pointer to the device is returned. | |
691 | * If the name is not found then %NULL is returned. | |
692 | * The reference counters are not incremented so the caller must be | |
693 | * careful with locks. The caller must hold RCU lock. | |
694 | */ | |
695 | ||
696 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
697 | { | |
72c9528b ED |
698 | struct net_device *dev; |
699 | struct hlist_head *head = dev_name_hash(net, name); | |
700 | ||
b67bfe0d | 701 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
702 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
703 | return dev; | |
704 | ||
705 | return NULL; | |
706 | } | |
707 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
708 | ||
1da177e4 LT |
709 | /** |
710 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 711 | * @net: the applicable net namespace |
1da177e4 LT |
712 | * @name: name to find |
713 | * | |
714 | * Find an interface by name. This can be called from any | |
715 | * context and does its own locking. The returned handle has | |
716 | * the usage count incremented and the caller must use dev_put() to | |
717 | * release it when it is no longer needed. %NULL is returned if no | |
718 | * matching device is found. | |
719 | */ | |
720 | ||
881d966b | 721 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
722 | { |
723 | struct net_device *dev; | |
724 | ||
72c9528b ED |
725 | rcu_read_lock(); |
726 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
727 | if (dev) |
728 | dev_hold(dev); | |
72c9528b | 729 | rcu_read_unlock(); |
1da177e4 LT |
730 | return dev; |
731 | } | |
d1b19dff | 732 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
733 | |
734 | /** | |
735 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 736 | * @net: the applicable net namespace |
1da177e4 LT |
737 | * @ifindex: index of device |
738 | * | |
739 | * Search for an interface by index. Returns %NULL if the device | |
740 | * is not found or a pointer to the device. The device has not | |
741 | * had its reference counter increased so the caller must be careful | |
742 | * about locking. The caller must hold either the RTNL semaphore | |
743 | * or @dev_base_lock. | |
744 | */ | |
745 | ||
881d966b | 746 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 747 | { |
0bd8d536 ED |
748 | struct net_device *dev; |
749 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 750 | |
b67bfe0d | 751 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
752 | if (dev->ifindex == ifindex) |
753 | return dev; | |
0bd8d536 | 754 | |
1da177e4 LT |
755 | return NULL; |
756 | } | |
d1b19dff | 757 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 758 | |
fb699dfd ED |
759 | /** |
760 | * dev_get_by_index_rcu - find a device by its ifindex | |
761 | * @net: the applicable net namespace | |
762 | * @ifindex: index of device | |
763 | * | |
764 | * Search for an interface by index. Returns %NULL if the device | |
765 | * is not found or a pointer to the device. The device has not | |
766 | * had its reference counter increased so the caller must be careful | |
767 | * about locking. The caller must hold RCU lock. | |
768 | */ | |
769 | ||
770 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
771 | { | |
fb699dfd ED |
772 | struct net_device *dev; |
773 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
774 | ||
b67bfe0d | 775 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
776 | if (dev->ifindex == ifindex) |
777 | return dev; | |
778 | ||
779 | return NULL; | |
780 | } | |
781 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
782 | ||
1da177e4 LT |
783 | |
784 | /** | |
785 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 786 | * @net: the applicable net namespace |
1da177e4 LT |
787 | * @ifindex: index of device |
788 | * | |
789 | * Search for an interface by index. Returns NULL if the device | |
790 | * is not found or a pointer to the device. The device returned has | |
791 | * had a reference added and the pointer is safe until the user calls | |
792 | * dev_put to indicate they have finished with it. | |
793 | */ | |
794 | ||
881d966b | 795 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
796 | { |
797 | struct net_device *dev; | |
798 | ||
fb699dfd ED |
799 | rcu_read_lock(); |
800 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
801 | if (dev) |
802 | dev_hold(dev); | |
fb699dfd | 803 | rcu_read_unlock(); |
1da177e4 LT |
804 | return dev; |
805 | } | |
d1b19dff | 806 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 807 | |
5dbe7c17 NS |
808 | /** |
809 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
810 | * @net: network namespace | |
811 | * @name: a pointer to the buffer where the name will be stored. | |
812 | * @ifindex: the ifindex of the interface to get the name from. | |
813 | * | |
814 | * The use of raw_seqcount_begin() and cond_resched() before | |
815 | * retrying is required as we want to give the writers a chance | |
816 | * to complete when CONFIG_PREEMPT is not set. | |
817 | */ | |
818 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
819 | { | |
820 | struct net_device *dev; | |
821 | unsigned int seq; | |
822 | ||
823 | retry: | |
824 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
825 | rcu_read_lock(); | |
826 | dev = dev_get_by_index_rcu(net, ifindex); | |
827 | if (!dev) { | |
828 | rcu_read_unlock(); | |
829 | return -ENODEV; | |
830 | } | |
831 | ||
832 | strcpy(name, dev->name); | |
833 | rcu_read_unlock(); | |
834 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
835 | cond_resched(); | |
836 | goto retry; | |
837 | } | |
838 | ||
839 | return 0; | |
840 | } | |
841 | ||
1da177e4 | 842 | /** |
941666c2 | 843 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 844 | * @net: the applicable net namespace |
1da177e4 LT |
845 | * @type: media type of device |
846 | * @ha: hardware address | |
847 | * | |
848 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
849 | * is not found or a pointer to the device. |
850 | * The caller must hold RCU or RTNL. | |
941666c2 | 851 | * The returned device has not had its ref count increased |
1da177e4 LT |
852 | * and the caller must therefore be careful about locking |
853 | * | |
1da177e4 LT |
854 | */ |
855 | ||
941666c2 ED |
856 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
857 | const char *ha) | |
1da177e4 LT |
858 | { |
859 | struct net_device *dev; | |
860 | ||
941666c2 | 861 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
862 | if (dev->type == type && |
863 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
864 | return dev; |
865 | ||
866 | return NULL; | |
1da177e4 | 867 | } |
941666c2 | 868 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 869 | |
881d966b | 870 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
871 | { |
872 | struct net_device *dev; | |
873 | ||
4e9cac2b | 874 | ASSERT_RTNL(); |
881d966b | 875 | for_each_netdev(net, dev) |
4e9cac2b | 876 | if (dev->type == type) |
7562f876 PE |
877 | return dev; |
878 | ||
879 | return NULL; | |
4e9cac2b | 880 | } |
4e9cac2b PM |
881 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
882 | ||
881d966b | 883 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 884 | { |
99fe3c39 | 885 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 886 | |
99fe3c39 ED |
887 | rcu_read_lock(); |
888 | for_each_netdev_rcu(net, dev) | |
889 | if (dev->type == type) { | |
890 | dev_hold(dev); | |
891 | ret = dev; | |
892 | break; | |
893 | } | |
894 | rcu_read_unlock(); | |
895 | return ret; | |
1da177e4 | 896 | } |
1da177e4 LT |
897 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
898 | ||
899 | /** | |
bb69ae04 | 900 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 901 | * @net: the applicable net namespace |
1da177e4 LT |
902 | * @if_flags: IFF_* values |
903 | * @mask: bitmask of bits in if_flags to check | |
904 | * | |
905 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
906 | * is not found or a pointer to the device. Must be called inside |
907 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
908 | */ |
909 | ||
bb69ae04 | 910 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 911 | unsigned short mask) |
1da177e4 | 912 | { |
7562f876 | 913 | struct net_device *dev, *ret; |
1da177e4 | 914 | |
7562f876 | 915 | ret = NULL; |
c6d14c84 | 916 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 917 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 918 | ret = dev; |
1da177e4 LT |
919 | break; |
920 | } | |
921 | } | |
7562f876 | 922 | return ret; |
1da177e4 | 923 | } |
bb69ae04 | 924 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
925 | |
926 | /** | |
927 | * dev_valid_name - check if name is okay for network device | |
928 | * @name: name string | |
929 | * | |
930 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
931 | * to allow sysfs to work. We also disallow any kind of |
932 | * whitespace. | |
1da177e4 | 933 | */ |
95f050bf | 934 | bool dev_valid_name(const char *name) |
1da177e4 | 935 | { |
c7fa9d18 | 936 | if (*name == '\0') |
95f050bf | 937 | return false; |
b6fe17d6 | 938 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 939 | return false; |
c7fa9d18 | 940 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 941 | return false; |
c7fa9d18 DM |
942 | |
943 | while (*name) { | |
944 | if (*name == '/' || isspace(*name)) | |
95f050bf | 945 | return false; |
c7fa9d18 DM |
946 | name++; |
947 | } | |
95f050bf | 948 | return true; |
1da177e4 | 949 | } |
d1b19dff | 950 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
951 | |
952 | /** | |
b267b179 EB |
953 | * __dev_alloc_name - allocate a name for a device |
954 | * @net: network namespace to allocate the device name in | |
1da177e4 | 955 | * @name: name format string |
b267b179 | 956 | * @buf: scratch buffer and result name string |
1da177e4 LT |
957 | * |
958 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
959 | * id. It scans list of devices to build up a free map, then chooses |
960 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
961 | * while allocating the name and adding the device in order to avoid | |
962 | * duplicates. | |
963 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
964 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
965 | */ |
966 | ||
b267b179 | 967 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
968 | { |
969 | int i = 0; | |
1da177e4 LT |
970 | const char *p; |
971 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 972 | unsigned long *inuse; |
1da177e4 LT |
973 | struct net_device *d; |
974 | ||
975 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
976 | if (p) { | |
977 | /* | |
978 | * Verify the string as this thing may have come from | |
979 | * the user. There must be either one "%d" and no other "%" | |
980 | * characters. | |
981 | */ | |
982 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
983 | return -EINVAL; | |
984 | ||
985 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 986 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
987 | if (!inuse) |
988 | return -ENOMEM; | |
989 | ||
881d966b | 990 | for_each_netdev(net, d) { |
1da177e4 LT |
991 | if (!sscanf(d->name, name, &i)) |
992 | continue; | |
993 | if (i < 0 || i >= max_netdevices) | |
994 | continue; | |
995 | ||
996 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 997 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
998 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
999 | set_bit(i, inuse); | |
1000 | } | |
1001 | ||
1002 | i = find_first_zero_bit(inuse, max_netdevices); | |
1003 | free_page((unsigned long) inuse); | |
1004 | } | |
1005 | ||
d9031024 OP |
1006 | if (buf != name) |
1007 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1008 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1009 | return i; |
1da177e4 LT |
1010 | |
1011 | /* It is possible to run out of possible slots | |
1012 | * when the name is long and there isn't enough space left | |
1013 | * for the digits, or if all bits are used. | |
1014 | */ | |
1015 | return -ENFILE; | |
1016 | } | |
1017 | ||
b267b179 EB |
1018 | /** |
1019 | * dev_alloc_name - allocate a name for a device | |
1020 | * @dev: device | |
1021 | * @name: name format string | |
1022 | * | |
1023 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1024 | * id. It scans list of devices to build up a free map, then chooses | |
1025 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1026 | * while allocating the name and adding the device in order to avoid | |
1027 | * duplicates. | |
1028 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1029 | * Returns the number of the unit assigned or a negative errno code. | |
1030 | */ | |
1031 | ||
1032 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1033 | { | |
1034 | char buf[IFNAMSIZ]; | |
1035 | struct net *net; | |
1036 | int ret; | |
1037 | ||
c346dca1 YH |
1038 | BUG_ON(!dev_net(dev)); |
1039 | net = dev_net(dev); | |
b267b179 EB |
1040 | ret = __dev_alloc_name(net, name, buf); |
1041 | if (ret >= 0) | |
1042 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1043 | return ret; | |
1044 | } | |
d1b19dff | 1045 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1046 | |
828de4f6 G |
1047 | static int dev_alloc_name_ns(struct net *net, |
1048 | struct net_device *dev, | |
1049 | const char *name) | |
d9031024 | 1050 | { |
828de4f6 G |
1051 | char buf[IFNAMSIZ]; |
1052 | int ret; | |
8ce6cebc | 1053 | |
828de4f6 G |
1054 | ret = __dev_alloc_name(net, name, buf); |
1055 | if (ret >= 0) | |
1056 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1057 | return ret; | |
1058 | } | |
1059 | ||
1060 | static int dev_get_valid_name(struct net *net, | |
1061 | struct net_device *dev, | |
1062 | const char *name) | |
1063 | { | |
1064 | BUG_ON(!net); | |
8ce6cebc | 1065 | |
d9031024 OP |
1066 | if (!dev_valid_name(name)) |
1067 | return -EINVAL; | |
1068 | ||
1c5cae81 | 1069 | if (strchr(name, '%')) |
828de4f6 | 1070 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1071 | else if (__dev_get_by_name(net, name)) |
1072 | return -EEXIST; | |
8ce6cebc DL |
1073 | else if (dev->name != name) |
1074 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1075 | |
1076 | return 0; | |
1077 | } | |
1da177e4 LT |
1078 | |
1079 | /** | |
1080 | * dev_change_name - change name of a device | |
1081 | * @dev: device | |
1082 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1083 | * | |
1084 | * Change name of a device, can pass format strings "eth%d". | |
1085 | * for wildcarding. | |
1086 | */ | |
cf04a4c7 | 1087 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1088 | { |
238fa362 | 1089 | unsigned char old_assign_type; |
fcc5a03a | 1090 | char oldname[IFNAMSIZ]; |
1da177e4 | 1091 | int err = 0; |
fcc5a03a | 1092 | int ret; |
881d966b | 1093 | struct net *net; |
1da177e4 LT |
1094 | |
1095 | ASSERT_RTNL(); | |
c346dca1 | 1096 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1097 | |
c346dca1 | 1098 | net = dev_net(dev); |
1da177e4 LT |
1099 | if (dev->flags & IFF_UP) |
1100 | return -EBUSY; | |
1101 | ||
30e6c9fa | 1102 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1103 | |
1104 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1105 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1106 | return 0; |
c91f6df2 | 1107 | } |
c8d90dca | 1108 | |
fcc5a03a HX |
1109 | memcpy(oldname, dev->name, IFNAMSIZ); |
1110 | ||
828de4f6 | 1111 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1112 | if (err < 0) { |
30e6c9fa | 1113 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1114 | return err; |
c91f6df2 | 1115 | } |
1da177e4 | 1116 | |
6fe82a39 VF |
1117 | if (oldname[0] && !strchr(oldname, '%')) |
1118 | netdev_info(dev, "renamed from %s\n", oldname); | |
1119 | ||
238fa362 TG |
1120 | old_assign_type = dev->name_assign_type; |
1121 | dev->name_assign_type = NET_NAME_RENAMED; | |
1122 | ||
fcc5a03a | 1123 | rollback: |
a1b3f594 EB |
1124 | ret = device_rename(&dev->dev, dev->name); |
1125 | if (ret) { | |
1126 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1127 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1128 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1129 | return ret; |
dcc99773 | 1130 | } |
7f988eab | 1131 | |
30e6c9fa | 1132 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1133 | |
5bb025fa VF |
1134 | netdev_adjacent_rename_links(dev, oldname); |
1135 | ||
7f988eab | 1136 | write_lock_bh(&dev_base_lock); |
372b2312 | 1137 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1138 | write_unlock_bh(&dev_base_lock); |
1139 | ||
1140 | synchronize_rcu(); | |
1141 | ||
1142 | write_lock_bh(&dev_base_lock); | |
1143 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1144 | write_unlock_bh(&dev_base_lock); |
1145 | ||
056925ab | 1146 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1147 | ret = notifier_to_errno(ret); |
1148 | ||
1149 | if (ret) { | |
91e9c07b ED |
1150 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1151 | if (err >= 0) { | |
fcc5a03a | 1152 | err = ret; |
30e6c9fa | 1153 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1154 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1155 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1156 | dev->name_assign_type = old_assign_type; |
1157 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1158 | goto rollback; |
91e9c07b | 1159 | } else { |
7b6cd1ce | 1160 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1161 | dev->name, ret); |
fcc5a03a HX |
1162 | } |
1163 | } | |
1da177e4 LT |
1164 | |
1165 | return err; | |
1166 | } | |
1167 | ||
0b815a1a SH |
1168 | /** |
1169 | * dev_set_alias - change ifalias of a device | |
1170 | * @dev: device | |
1171 | * @alias: name up to IFALIASZ | |
f0db275a | 1172 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1173 | * |
1174 | * Set ifalias for a device, | |
1175 | */ | |
1176 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1177 | { | |
7364e445 AK |
1178 | char *new_ifalias; |
1179 | ||
0b815a1a SH |
1180 | ASSERT_RTNL(); |
1181 | ||
1182 | if (len >= IFALIASZ) | |
1183 | return -EINVAL; | |
1184 | ||
96ca4a2c | 1185 | if (!len) { |
388dfc2d SK |
1186 | kfree(dev->ifalias); |
1187 | dev->ifalias = NULL; | |
96ca4a2c OH |
1188 | return 0; |
1189 | } | |
1190 | ||
7364e445 AK |
1191 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1192 | if (!new_ifalias) | |
0b815a1a | 1193 | return -ENOMEM; |
7364e445 | 1194 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1195 | |
1196 | strlcpy(dev->ifalias, alias, len+1); | |
1197 | return len; | |
1198 | } | |
1199 | ||
1200 | ||
d8a33ac4 | 1201 | /** |
3041a069 | 1202 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1203 | * @dev: device to cause notification |
1204 | * | |
1205 | * Called to indicate a device has changed features. | |
1206 | */ | |
1207 | void netdev_features_change(struct net_device *dev) | |
1208 | { | |
056925ab | 1209 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1210 | } |
1211 | EXPORT_SYMBOL(netdev_features_change); | |
1212 | ||
1da177e4 LT |
1213 | /** |
1214 | * netdev_state_change - device changes state | |
1215 | * @dev: device to cause notification | |
1216 | * | |
1217 | * Called to indicate a device has changed state. This function calls | |
1218 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1219 | * to the routing socket. | |
1220 | */ | |
1221 | void netdev_state_change(struct net_device *dev) | |
1222 | { | |
1223 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1224 | struct netdev_notifier_change_info change_info; |
1225 | ||
1226 | change_info.flags_changed = 0; | |
1227 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1228 | &change_info.info); | |
7f294054 | 1229 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1230 | } |
1231 | } | |
d1b19dff | 1232 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1233 | |
ee89bab1 AW |
1234 | /** |
1235 | * netdev_notify_peers - notify network peers about existence of @dev | |
1236 | * @dev: network device | |
1237 | * | |
1238 | * Generate traffic such that interested network peers are aware of | |
1239 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1240 | * a device wants to inform the rest of the network about some sort of | |
1241 | * reconfiguration such as a failover event or virtual machine | |
1242 | * migration. | |
1243 | */ | |
1244 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1245 | { |
ee89bab1 AW |
1246 | rtnl_lock(); |
1247 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1248 | rtnl_unlock(); | |
c1da4ac7 | 1249 | } |
ee89bab1 | 1250 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1251 | |
bd380811 | 1252 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1253 | { |
d314774c | 1254 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1255 | int ret; |
1da177e4 | 1256 | |
e46b66bc BH |
1257 | ASSERT_RTNL(); |
1258 | ||
1da177e4 LT |
1259 | if (!netif_device_present(dev)) |
1260 | return -ENODEV; | |
1261 | ||
ca99ca14 NH |
1262 | /* Block netpoll from trying to do any rx path servicing. |
1263 | * If we don't do this there is a chance ndo_poll_controller | |
1264 | * or ndo_poll may be running while we open the device | |
1265 | */ | |
66b5552f | 1266 | netpoll_poll_disable(dev); |
ca99ca14 | 1267 | |
3b8bcfd5 JB |
1268 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1269 | ret = notifier_to_errno(ret); | |
1270 | if (ret) | |
1271 | return ret; | |
1272 | ||
1da177e4 | 1273 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1274 | |
d314774c SH |
1275 | if (ops->ndo_validate_addr) |
1276 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1277 | |
d314774c SH |
1278 | if (!ret && ops->ndo_open) |
1279 | ret = ops->ndo_open(dev); | |
1da177e4 | 1280 | |
66b5552f | 1281 | netpoll_poll_enable(dev); |
ca99ca14 | 1282 | |
bada339b JG |
1283 | if (ret) |
1284 | clear_bit(__LINK_STATE_START, &dev->state); | |
1285 | else { | |
1da177e4 | 1286 | dev->flags |= IFF_UP; |
b4bd07c2 | 1287 | net_dmaengine_get(); |
4417da66 | 1288 | dev_set_rx_mode(dev); |
1da177e4 | 1289 | dev_activate(dev); |
7bf23575 | 1290 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1291 | } |
bada339b | 1292 | |
1da177e4 LT |
1293 | return ret; |
1294 | } | |
1295 | ||
1296 | /** | |
bd380811 PM |
1297 | * dev_open - prepare an interface for use. |
1298 | * @dev: device to open | |
1da177e4 | 1299 | * |
bd380811 PM |
1300 | * Takes a device from down to up state. The device's private open |
1301 | * function is invoked and then the multicast lists are loaded. Finally | |
1302 | * the device is moved into the up state and a %NETDEV_UP message is | |
1303 | * sent to the netdev notifier chain. | |
1304 | * | |
1305 | * Calling this function on an active interface is a nop. On a failure | |
1306 | * a negative errno code is returned. | |
1da177e4 | 1307 | */ |
bd380811 PM |
1308 | int dev_open(struct net_device *dev) |
1309 | { | |
1310 | int ret; | |
1311 | ||
bd380811 PM |
1312 | if (dev->flags & IFF_UP) |
1313 | return 0; | |
1314 | ||
bd380811 PM |
1315 | ret = __dev_open(dev); |
1316 | if (ret < 0) | |
1317 | return ret; | |
1318 | ||
7f294054 | 1319 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1320 | call_netdevice_notifiers(NETDEV_UP, dev); |
1321 | ||
1322 | return ret; | |
1323 | } | |
1324 | EXPORT_SYMBOL(dev_open); | |
1325 | ||
44345724 | 1326 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1327 | { |
44345724 | 1328 | struct net_device *dev; |
e46b66bc | 1329 | |
bd380811 | 1330 | ASSERT_RTNL(); |
9d5010db DM |
1331 | might_sleep(); |
1332 | ||
5cde2829 | 1333 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1334 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1335 | netpoll_poll_disable(dev); |
3f4df206 | 1336 | |
44345724 | 1337 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1338 | |
44345724 | 1339 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1340 | |
44345724 OP |
1341 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1342 | * can be even on different cpu. So just clear netif_running(). | |
1343 | * | |
1344 | * dev->stop() will invoke napi_disable() on all of it's | |
1345 | * napi_struct instances on this device. | |
1346 | */ | |
4e857c58 | 1347 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1348 | } |
1da177e4 | 1349 | |
44345724 | 1350 | dev_deactivate_many(head); |
d8b2a4d2 | 1351 | |
5cde2829 | 1352 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1353 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1354 | |
44345724 OP |
1355 | /* |
1356 | * Call the device specific close. This cannot fail. | |
1357 | * Only if device is UP | |
1358 | * | |
1359 | * We allow it to be called even after a DETACH hot-plug | |
1360 | * event. | |
1361 | */ | |
1362 | if (ops->ndo_stop) | |
1363 | ops->ndo_stop(dev); | |
1364 | ||
44345724 | 1365 | dev->flags &= ~IFF_UP; |
44345724 | 1366 | net_dmaengine_put(); |
66b5552f | 1367 | netpoll_poll_enable(dev); |
44345724 OP |
1368 | } |
1369 | ||
1370 | return 0; | |
1371 | } | |
1372 | ||
1373 | static int __dev_close(struct net_device *dev) | |
1374 | { | |
f87e6f47 | 1375 | int retval; |
44345724 OP |
1376 | LIST_HEAD(single); |
1377 | ||
5cde2829 | 1378 | list_add(&dev->close_list, &single); |
f87e6f47 LT |
1379 | retval = __dev_close_many(&single); |
1380 | list_del(&single); | |
ca99ca14 | 1381 | |
f87e6f47 | 1382 | return retval; |
44345724 OP |
1383 | } |
1384 | ||
3fbd8758 | 1385 | static int dev_close_many(struct list_head *head) |
44345724 OP |
1386 | { |
1387 | struct net_device *dev, *tmp; | |
1da177e4 | 1388 | |
5cde2829 EB |
1389 | /* Remove the devices that don't need to be closed */ |
1390 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1391 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1392 | list_del_init(&dev->close_list); |
44345724 OP |
1393 | |
1394 | __dev_close_many(head); | |
1da177e4 | 1395 | |
5cde2829 | 1396 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1397 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1398 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
5cde2829 | 1399 | list_del_init(&dev->close_list); |
44345724 | 1400 | } |
bd380811 PM |
1401 | |
1402 | return 0; | |
1403 | } | |
1404 | ||
1405 | /** | |
1406 | * dev_close - shutdown an interface. | |
1407 | * @dev: device to shutdown | |
1408 | * | |
1409 | * This function moves an active device into down state. A | |
1410 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1411 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1412 | * chain. | |
1413 | */ | |
1414 | int dev_close(struct net_device *dev) | |
1415 | { | |
e14a5993 ED |
1416 | if (dev->flags & IFF_UP) { |
1417 | LIST_HEAD(single); | |
1da177e4 | 1418 | |
5cde2829 | 1419 | list_add(&dev->close_list, &single); |
e14a5993 ED |
1420 | dev_close_many(&single); |
1421 | list_del(&single); | |
1422 | } | |
da6e378b | 1423 | return 0; |
1da177e4 | 1424 | } |
d1b19dff | 1425 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1426 | |
1427 | ||
0187bdfb BH |
1428 | /** |
1429 | * dev_disable_lro - disable Large Receive Offload on a device | |
1430 | * @dev: device | |
1431 | * | |
1432 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1433 | * called under RTNL. This is needed if received packets may be | |
1434 | * forwarded to another interface. | |
1435 | */ | |
1436 | void dev_disable_lro(struct net_device *dev) | |
1437 | { | |
f11970e3 NH |
1438 | /* |
1439 | * If we're trying to disable lro on a vlan device | |
1440 | * use the underlying physical device instead | |
1441 | */ | |
1442 | if (is_vlan_dev(dev)) | |
1443 | dev = vlan_dev_real_dev(dev); | |
1444 | ||
529d0489 MK |
1445 | /* the same for macvlan devices */ |
1446 | if (netif_is_macvlan(dev)) | |
1447 | dev = macvlan_dev_real_dev(dev); | |
1448 | ||
bc5787c6 MM |
1449 | dev->wanted_features &= ~NETIF_F_LRO; |
1450 | netdev_update_features(dev); | |
27660515 | 1451 | |
22d5969f MM |
1452 | if (unlikely(dev->features & NETIF_F_LRO)) |
1453 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
0187bdfb BH |
1454 | } |
1455 | EXPORT_SYMBOL(dev_disable_lro); | |
1456 | ||
351638e7 JP |
1457 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1458 | struct net_device *dev) | |
1459 | { | |
1460 | struct netdev_notifier_info info; | |
1461 | ||
1462 | netdev_notifier_info_init(&info, dev); | |
1463 | return nb->notifier_call(nb, val, &info); | |
1464 | } | |
0187bdfb | 1465 | |
881d966b EB |
1466 | static int dev_boot_phase = 1; |
1467 | ||
1da177e4 LT |
1468 | /** |
1469 | * register_netdevice_notifier - register a network notifier block | |
1470 | * @nb: notifier | |
1471 | * | |
1472 | * Register a notifier to be called when network device events occur. | |
1473 | * The notifier passed is linked into the kernel structures and must | |
1474 | * not be reused until it has been unregistered. A negative errno code | |
1475 | * is returned on a failure. | |
1476 | * | |
1477 | * When registered all registration and up events are replayed | |
4ec93edb | 1478 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1479 | * view of the network device list. |
1480 | */ | |
1481 | ||
1482 | int register_netdevice_notifier(struct notifier_block *nb) | |
1483 | { | |
1484 | struct net_device *dev; | |
fcc5a03a | 1485 | struct net_device *last; |
881d966b | 1486 | struct net *net; |
1da177e4 LT |
1487 | int err; |
1488 | ||
1489 | rtnl_lock(); | |
f07d5b94 | 1490 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1491 | if (err) |
1492 | goto unlock; | |
881d966b EB |
1493 | if (dev_boot_phase) |
1494 | goto unlock; | |
1495 | for_each_net(net) { | |
1496 | for_each_netdev(net, dev) { | |
351638e7 | 1497 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1498 | err = notifier_to_errno(err); |
1499 | if (err) | |
1500 | goto rollback; | |
1501 | ||
1502 | if (!(dev->flags & IFF_UP)) | |
1503 | continue; | |
1da177e4 | 1504 | |
351638e7 | 1505 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1506 | } |
1da177e4 | 1507 | } |
fcc5a03a HX |
1508 | |
1509 | unlock: | |
1da177e4 LT |
1510 | rtnl_unlock(); |
1511 | return err; | |
fcc5a03a HX |
1512 | |
1513 | rollback: | |
1514 | last = dev; | |
881d966b EB |
1515 | for_each_net(net) { |
1516 | for_each_netdev(net, dev) { | |
1517 | if (dev == last) | |
8f891489 | 1518 | goto outroll; |
fcc5a03a | 1519 | |
881d966b | 1520 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1521 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1522 | dev); | |
1523 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1524 | } |
351638e7 | 1525 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1526 | } |
fcc5a03a | 1527 | } |
c67625a1 | 1528 | |
8f891489 | 1529 | outroll: |
c67625a1 | 1530 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1531 | goto unlock; |
1da177e4 | 1532 | } |
d1b19dff | 1533 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1534 | |
1535 | /** | |
1536 | * unregister_netdevice_notifier - unregister a network notifier block | |
1537 | * @nb: notifier | |
1538 | * | |
1539 | * Unregister a notifier previously registered by | |
1540 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1541 | * kernel structures and may then be reused. A negative errno code | |
1542 | * is returned on a failure. | |
7d3d43da EB |
1543 | * |
1544 | * After unregistering unregister and down device events are synthesized | |
1545 | * for all devices on the device list to the removed notifier to remove | |
1546 | * the need for special case cleanup code. | |
1da177e4 LT |
1547 | */ |
1548 | ||
1549 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1550 | { | |
7d3d43da EB |
1551 | struct net_device *dev; |
1552 | struct net *net; | |
9f514950 HX |
1553 | int err; |
1554 | ||
1555 | rtnl_lock(); | |
f07d5b94 | 1556 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1557 | if (err) |
1558 | goto unlock; | |
1559 | ||
1560 | for_each_net(net) { | |
1561 | for_each_netdev(net, dev) { | |
1562 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1563 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1564 | dev); | |
1565 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1566 | } |
351638e7 | 1567 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1568 | } |
1569 | } | |
1570 | unlock: | |
9f514950 HX |
1571 | rtnl_unlock(); |
1572 | return err; | |
1da177e4 | 1573 | } |
d1b19dff | 1574 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1575 | |
351638e7 JP |
1576 | /** |
1577 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1578 | * @val: value passed unmodified to notifier function | |
1579 | * @dev: net_device pointer passed unmodified to notifier function | |
1580 | * @info: notifier information data | |
1581 | * | |
1582 | * Call all network notifier blocks. Parameters and return value | |
1583 | * are as for raw_notifier_call_chain(). | |
1584 | */ | |
1585 | ||
1d143d9f | 1586 | static int call_netdevice_notifiers_info(unsigned long val, |
1587 | struct net_device *dev, | |
1588 | struct netdev_notifier_info *info) | |
351638e7 JP |
1589 | { |
1590 | ASSERT_RTNL(); | |
1591 | netdev_notifier_info_init(info, dev); | |
1592 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1593 | } | |
351638e7 | 1594 | |
1da177e4 LT |
1595 | /** |
1596 | * call_netdevice_notifiers - call all network notifier blocks | |
1597 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1598 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1599 | * |
1600 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1601 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1602 | */ |
1603 | ||
ad7379d4 | 1604 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1605 | { |
351638e7 JP |
1606 | struct netdev_notifier_info info; |
1607 | ||
1608 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1609 | } |
edf947f1 | 1610 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1611 | |
c5905afb | 1612 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1613 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1614 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1615 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1616 | * static_key_slow_dec() calls. |
b90e5794 ED |
1617 | */ |
1618 | static atomic_t netstamp_needed_deferred; | |
1619 | #endif | |
1da177e4 LT |
1620 | |
1621 | void net_enable_timestamp(void) | |
1622 | { | |
b90e5794 ED |
1623 | #ifdef HAVE_JUMP_LABEL |
1624 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1625 | ||
1626 | if (deferred) { | |
1627 | while (--deferred) | |
c5905afb | 1628 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1629 | return; |
1630 | } | |
1631 | #endif | |
c5905afb | 1632 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1633 | } |
d1b19dff | 1634 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1635 | |
1636 | void net_disable_timestamp(void) | |
1637 | { | |
b90e5794 ED |
1638 | #ifdef HAVE_JUMP_LABEL |
1639 | if (in_interrupt()) { | |
1640 | atomic_inc(&netstamp_needed_deferred); | |
1641 | return; | |
1642 | } | |
1643 | #endif | |
c5905afb | 1644 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1645 | } |
d1b19dff | 1646 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1647 | |
3b098e2d | 1648 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1649 | { |
588f0330 | 1650 | skb->tstamp.tv64 = 0; |
c5905afb | 1651 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1652 | __net_timestamp(skb); |
1da177e4 LT |
1653 | } |
1654 | ||
588f0330 | 1655 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1656 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1657 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1658 | __net_timestamp(SKB); \ | |
1659 | } \ | |
3b098e2d | 1660 | |
1ee481fb | 1661 | bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb) |
79b569f0 DL |
1662 | { |
1663 | unsigned int len; | |
1664 | ||
1665 | if (!(dev->flags & IFF_UP)) | |
1666 | return false; | |
1667 | ||
1668 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1669 | if (skb->len <= len) | |
1670 | return true; | |
1671 | ||
1672 | /* if TSO is enabled, we don't care about the length as the packet | |
1673 | * could be forwarded without being segmented before | |
1674 | */ | |
1675 | if (skb_is_gso(skb)) | |
1676 | return true; | |
1677 | ||
1678 | return false; | |
1679 | } | |
1ee481fb | 1680 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1681 | |
a0265d28 HX |
1682 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1683 | { | |
1684 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { | |
1685 | if (skb_copy_ubufs(skb, GFP_ATOMIC)) { | |
1686 | atomic_long_inc(&dev->rx_dropped); | |
1687 | kfree_skb(skb); | |
1688 | return NET_RX_DROP; | |
1689 | } | |
1690 | } | |
1691 | ||
1692 | if (unlikely(!is_skb_forwardable(dev, skb))) { | |
1693 | atomic_long_inc(&dev->rx_dropped); | |
1694 | kfree_skb(skb); | |
1695 | return NET_RX_DROP; | |
1696 | } | |
1697 | ||
1698 | skb_scrub_packet(skb, true); | |
1699 | skb->protocol = eth_type_trans(skb, dev); | |
1700 | ||
1701 | return 0; | |
1702 | } | |
1703 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1704 | ||
44540960 AB |
1705 | /** |
1706 | * dev_forward_skb - loopback an skb to another netif | |
1707 | * | |
1708 | * @dev: destination network device | |
1709 | * @skb: buffer to forward | |
1710 | * | |
1711 | * return values: | |
1712 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1713 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1714 | * |
1715 | * dev_forward_skb can be used for injecting an skb from the | |
1716 | * start_xmit function of one device into the receive queue | |
1717 | * of another device. | |
1718 | * | |
1719 | * The receiving device may be in another namespace, so | |
1720 | * we have to clear all information in the skb that could | |
1721 | * impact namespace isolation. | |
1722 | */ | |
1723 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1724 | { | |
a0265d28 | 1725 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1726 | } |
1727 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1728 | ||
71d9dec2 CG |
1729 | static inline int deliver_skb(struct sk_buff *skb, |
1730 | struct packet_type *pt_prev, | |
1731 | struct net_device *orig_dev) | |
1732 | { | |
1080e512 MT |
1733 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1734 | return -ENOMEM; | |
71d9dec2 CG |
1735 | atomic_inc(&skb->users); |
1736 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1737 | } | |
1738 | ||
c0de08d0 EL |
1739 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1740 | { | |
a3d744e9 | 1741 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1742 | return false; |
1743 | ||
1744 | if (ptype->id_match) | |
1745 | return ptype->id_match(ptype, skb->sk); | |
1746 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1747 | return true; | |
1748 | ||
1749 | return false; | |
1750 | } | |
1751 | ||
1da177e4 LT |
1752 | /* |
1753 | * Support routine. Sends outgoing frames to any network | |
1754 | * taps currently in use. | |
1755 | */ | |
1756 | ||
f6a78bfc | 1757 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1758 | { |
1759 | struct packet_type *ptype; | |
71d9dec2 CG |
1760 | struct sk_buff *skb2 = NULL; |
1761 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1762 | |
1da177e4 LT |
1763 | rcu_read_lock(); |
1764 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1765 | /* Never send packets back to the socket | |
1766 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1767 | */ | |
1768 | if ((ptype->dev == dev || !ptype->dev) && | |
c0de08d0 | 1769 | (!skb_loop_sk(ptype, skb))) { |
71d9dec2 CG |
1770 | if (pt_prev) { |
1771 | deliver_skb(skb2, pt_prev, skb->dev); | |
1772 | pt_prev = ptype; | |
1773 | continue; | |
1774 | } | |
1775 | ||
1776 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1777 | if (!skb2) |
1778 | break; | |
1779 | ||
70978182 ED |
1780 | net_timestamp_set(skb2); |
1781 | ||
1da177e4 LT |
1782 | /* skb->nh should be correctly |
1783 | set by sender, so that the second statement is | |
1784 | just protection against buggy protocols. | |
1785 | */ | |
459a98ed | 1786 | skb_reset_mac_header(skb2); |
1da177e4 | 1787 | |
d56f90a7 | 1788 | if (skb_network_header(skb2) < skb2->data || |
ced14f68 | 1789 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { |
e87cc472 JP |
1790 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", |
1791 | ntohs(skb2->protocol), | |
1792 | dev->name); | |
c1d2bbe1 | 1793 | skb_reset_network_header(skb2); |
1da177e4 LT |
1794 | } |
1795 | ||
b0e380b1 | 1796 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1797 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1798 | pt_prev = ptype; |
1da177e4 LT |
1799 | } |
1800 | } | |
71d9dec2 CG |
1801 | if (pt_prev) |
1802 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1803 | rcu_read_unlock(); |
1804 | } | |
1805 | ||
2c53040f BH |
1806 | /** |
1807 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1808 | * @dev: Network device |
1809 | * @txq: number of queues available | |
1810 | * | |
1811 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1812 | * valid. To resolve this verify the tc mapping remains valid and if | |
1813 | * not NULL the mapping. With no priorities mapping to this | |
1814 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1815 | * is invalid nothing can be done so disable priority mappings. If is | |
1816 | * expected that drivers will fix this mapping if they can before | |
1817 | * calling netif_set_real_num_tx_queues. | |
1818 | */ | |
bb134d22 | 1819 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1820 | { |
1821 | int i; | |
1822 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1823 | ||
1824 | /* If TC0 is invalidated disable TC mapping */ | |
1825 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1826 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1827 | dev->num_tc = 0; |
1828 | return; | |
1829 | } | |
1830 | ||
1831 | /* Invalidated prio to tc mappings set to TC0 */ | |
1832 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1833 | int q = netdev_get_prio_tc_map(dev, i); | |
1834 | ||
1835 | tc = &dev->tc_to_txq[q]; | |
1836 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1837 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1838 | i, q); | |
4f57c087 JF |
1839 | netdev_set_prio_tc_map(dev, i, 0); |
1840 | } | |
1841 | } | |
1842 | } | |
1843 | ||
537c00de AD |
1844 | #ifdef CONFIG_XPS |
1845 | static DEFINE_MUTEX(xps_map_mutex); | |
1846 | #define xmap_dereference(P) \ | |
1847 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1848 | ||
10cdc3f3 AD |
1849 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1850 | int cpu, u16 index) | |
537c00de | 1851 | { |
10cdc3f3 AD |
1852 | struct xps_map *map = NULL; |
1853 | int pos; | |
537c00de | 1854 | |
10cdc3f3 AD |
1855 | if (dev_maps) |
1856 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1857 | |
10cdc3f3 AD |
1858 | for (pos = 0; map && pos < map->len; pos++) { |
1859 | if (map->queues[pos] == index) { | |
537c00de AD |
1860 | if (map->len > 1) { |
1861 | map->queues[pos] = map->queues[--map->len]; | |
1862 | } else { | |
10cdc3f3 | 1863 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1864 | kfree_rcu(map, rcu); |
1865 | map = NULL; | |
1866 | } | |
10cdc3f3 | 1867 | break; |
537c00de | 1868 | } |
537c00de AD |
1869 | } |
1870 | ||
10cdc3f3 AD |
1871 | return map; |
1872 | } | |
1873 | ||
024e9679 | 1874 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
10cdc3f3 AD |
1875 | { |
1876 | struct xps_dev_maps *dev_maps; | |
024e9679 | 1877 | int cpu, i; |
10cdc3f3 AD |
1878 | bool active = false; |
1879 | ||
1880 | mutex_lock(&xps_map_mutex); | |
1881 | dev_maps = xmap_dereference(dev->xps_maps); | |
1882 | ||
1883 | if (!dev_maps) | |
1884 | goto out_no_maps; | |
1885 | ||
1886 | for_each_possible_cpu(cpu) { | |
024e9679 AD |
1887 | for (i = index; i < dev->num_tx_queues; i++) { |
1888 | if (!remove_xps_queue(dev_maps, cpu, i)) | |
1889 | break; | |
1890 | } | |
1891 | if (i == dev->num_tx_queues) | |
10cdc3f3 AD |
1892 | active = true; |
1893 | } | |
1894 | ||
1895 | if (!active) { | |
537c00de AD |
1896 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
1897 | kfree_rcu(dev_maps, rcu); | |
1898 | } | |
1899 | ||
024e9679 AD |
1900 | for (i = index; i < dev->num_tx_queues; i++) |
1901 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), | |
1902 | NUMA_NO_NODE); | |
1903 | ||
537c00de AD |
1904 | out_no_maps: |
1905 | mutex_unlock(&xps_map_mutex); | |
1906 | } | |
1907 | ||
01c5f864 AD |
1908 | static struct xps_map *expand_xps_map(struct xps_map *map, |
1909 | int cpu, u16 index) | |
1910 | { | |
1911 | struct xps_map *new_map; | |
1912 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
1913 | int i, pos; | |
1914 | ||
1915 | for (pos = 0; map && pos < map->len; pos++) { | |
1916 | if (map->queues[pos] != index) | |
1917 | continue; | |
1918 | return map; | |
1919 | } | |
1920 | ||
1921 | /* Need to add queue to this CPU's existing map */ | |
1922 | if (map) { | |
1923 | if (pos < map->alloc_len) | |
1924 | return map; | |
1925 | ||
1926 | alloc_len = map->alloc_len * 2; | |
1927 | } | |
1928 | ||
1929 | /* Need to allocate new map to store queue on this CPU's map */ | |
1930 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
1931 | cpu_to_node(cpu)); | |
1932 | if (!new_map) | |
1933 | return NULL; | |
1934 | ||
1935 | for (i = 0; i < pos; i++) | |
1936 | new_map->queues[i] = map->queues[i]; | |
1937 | new_map->alloc_len = alloc_len; | |
1938 | new_map->len = pos; | |
1939 | ||
1940 | return new_map; | |
1941 | } | |
1942 | ||
3573540c MT |
1943 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
1944 | u16 index) | |
537c00de | 1945 | { |
01c5f864 | 1946 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 1947 | struct xps_map *map, *new_map; |
537c00de | 1948 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
1949 | int cpu, numa_node_id = -2; |
1950 | bool active = false; | |
537c00de AD |
1951 | |
1952 | mutex_lock(&xps_map_mutex); | |
1953 | ||
1954 | dev_maps = xmap_dereference(dev->xps_maps); | |
1955 | ||
01c5f864 AD |
1956 | /* allocate memory for queue storage */ |
1957 | for_each_online_cpu(cpu) { | |
1958 | if (!cpumask_test_cpu(cpu, mask)) | |
1959 | continue; | |
1960 | ||
1961 | if (!new_dev_maps) | |
1962 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
1963 | if (!new_dev_maps) { |
1964 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 1965 | return -ENOMEM; |
2bb60cb9 | 1966 | } |
01c5f864 AD |
1967 | |
1968 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
1969 | NULL; | |
1970 | ||
1971 | map = expand_xps_map(map, cpu, index); | |
1972 | if (!map) | |
1973 | goto error; | |
1974 | ||
1975 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
1976 | } | |
1977 | ||
1978 | if (!new_dev_maps) | |
1979 | goto out_no_new_maps; | |
1980 | ||
537c00de | 1981 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
1982 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
1983 | /* add queue to CPU maps */ | |
1984 | int pos = 0; | |
1985 | ||
1986 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1987 | while ((pos < map->len) && (map->queues[pos] != index)) | |
1988 | pos++; | |
1989 | ||
1990 | if (pos == map->len) | |
1991 | map->queues[map->len++] = index; | |
537c00de | 1992 | #ifdef CONFIG_NUMA |
537c00de AD |
1993 | if (numa_node_id == -2) |
1994 | numa_node_id = cpu_to_node(cpu); | |
1995 | else if (numa_node_id != cpu_to_node(cpu)) | |
1996 | numa_node_id = -1; | |
537c00de | 1997 | #endif |
01c5f864 AD |
1998 | } else if (dev_maps) { |
1999 | /* fill in the new device map from the old device map */ | |
2000 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2001 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 2002 | } |
01c5f864 | 2003 | |
537c00de AD |
2004 | } |
2005 | ||
01c5f864 AD |
2006 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2007 | ||
537c00de | 2008 | /* Cleanup old maps */ |
01c5f864 AD |
2009 | if (dev_maps) { |
2010 | for_each_possible_cpu(cpu) { | |
2011 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2012 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2013 | if (map && map != new_map) | |
2014 | kfree_rcu(map, rcu); | |
2015 | } | |
537c00de | 2016 | |
01c5f864 | 2017 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
2018 | } |
2019 | ||
01c5f864 AD |
2020 | dev_maps = new_dev_maps; |
2021 | active = true; | |
537c00de | 2022 | |
01c5f864 AD |
2023 | out_no_new_maps: |
2024 | /* update Tx queue numa node */ | |
537c00de AD |
2025 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2026 | (numa_node_id >= 0) ? numa_node_id : | |
2027 | NUMA_NO_NODE); | |
2028 | ||
01c5f864 AD |
2029 | if (!dev_maps) |
2030 | goto out_no_maps; | |
2031 | ||
2032 | /* removes queue from unused CPUs */ | |
2033 | for_each_possible_cpu(cpu) { | |
2034 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
2035 | continue; | |
2036 | ||
2037 | if (remove_xps_queue(dev_maps, cpu, index)) | |
2038 | active = true; | |
2039 | } | |
2040 | ||
2041 | /* free map if not active */ | |
2042 | if (!active) { | |
2043 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2044 | kfree_rcu(dev_maps, rcu); | |
2045 | } | |
2046 | ||
2047 | out_no_maps: | |
537c00de AD |
2048 | mutex_unlock(&xps_map_mutex); |
2049 | ||
2050 | return 0; | |
2051 | error: | |
01c5f864 AD |
2052 | /* remove any maps that we added */ |
2053 | for_each_possible_cpu(cpu) { | |
2054 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2055 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2056 | NULL; | |
2057 | if (new_map && new_map != map) | |
2058 | kfree(new_map); | |
2059 | } | |
2060 | ||
537c00de AD |
2061 | mutex_unlock(&xps_map_mutex); |
2062 | ||
537c00de AD |
2063 | kfree(new_dev_maps); |
2064 | return -ENOMEM; | |
2065 | } | |
2066 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2067 | ||
2068 | #endif | |
f0796d5c JF |
2069 | /* |
2070 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2071 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2072 | */ | |
e6484930 | 2073 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2074 | { |
1d24eb48 TH |
2075 | int rc; |
2076 | ||
e6484930 TH |
2077 | if (txq < 1 || txq > dev->num_tx_queues) |
2078 | return -EINVAL; | |
f0796d5c | 2079 | |
5c56580b BH |
2080 | if (dev->reg_state == NETREG_REGISTERED || |
2081 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2082 | ASSERT_RTNL(); |
2083 | ||
1d24eb48 TH |
2084 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2085 | txq); | |
bf264145 TH |
2086 | if (rc) |
2087 | return rc; | |
2088 | ||
4f57c087 JF |
2089 | if (dev->num_tc) |
2090 | netif_setup_tc(dev, txq); | |
2091 | ||
024e9679 | 2092 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2093 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2094 | #ifdef CONFIG_XPS |
2095 | netif_reset_xps_queues_gt(dev, txq); | |
2096 | #endif | |
2097 | } | |
f0796d5c | 2098 | } |
e6484930 TH |
2099 | |
2100 | dev->real_num_tx_queues = txq; | |
2101 | return 0; | |
f0796d5c JF |
2102 | } |
2103 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2104 | |
a953be53 | 2105 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2106 | /** |
2107 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2108 | * @dev: Network device | |
2109 | * @rxq: Actual number of RX queues | |
2110 | * | |
2111 | * This must be called either with the rtnl_lock held or before | |
2112 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2113 | * negative error code. If called before registration, it always |
2114 | * succeeds. | |
62fe0b40 BH |
2115 | */ |
2116 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2117 | { | |
2118 | int rc; | |
2119 | ||
bd25fa7b TH |
2120 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2121 | return -EINVAL; | |
2122 | ||
62fe0b40 BH |
2123 | if (dev->reg_state == NETREG_REGISTERED) { |
2124 | ASSERT_RTNL(); | |
2125 | ||
62fe0b40 BH |
2126 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2127 | rxq); | |
2128 | if (rc) | |
2129 | return rc; | |
62fe0b40 BH |
2130 | } |
2131 | ||
2132 | dev->real_num_rx_queues = rxq; | |
2133 | return 0; | |
2134 | } | |
2135 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2136 | #endif | |
2137 | ||
2c53040f BH |
2138 | /** |
2139 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2140 | * |
2141 | * This routine should set an upper limit on the number of RSS queues | |
2142 | * used by default by multiqueue devices. | |
2143 | */ | |
a55b138b | 2144 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
2145 | { |
2146 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
2147 | } | |
2148 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2149 | ||
def82a1d | 2150 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 2151 | { |
def82a1d JP |
2152 | struct softnet_data *sd; |
2153 | unsigned long flags; | |
56079431 | 2154 | |
def82a1d JP |
2155 | local_irq_save(flags); |
2156 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
2157 | q->next_sched = NULL; |
2158 | *sd->output_queue_tailp = q; | |
2159 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2160 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2161 | local_irq_restore(flags); | |
2162 | } | |
2163 | ||
2164 | void __netif_schedule(struct Qdisc *q) | |
2165 | { | |
2166 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2167 | __netif_reschedule(q); | |
56079431 DV |
2168 | } |
2169 | EXPORT_SYMBOL(__netif_schedule); | |
2170 | ||
e6247027 ED |
2171 | struct dev_kfree_skb_cb { |
2172 | enum skb_free_reason reason; | |
2173 | }; | |
2174 | ||
2175 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2176 | { |
e6247027 ED |
2177 | return (struct dev_kfree_skb_cb *)skb->cb; |
2178 | } | |
2179 | ||
2180 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) | |
56079431 | 2181 | { |
e6247027 | 2182 | unsigned long flags; |
56079431 | 2183 | |
e6247027 ED |
2184 | if (likely(atomic_read(&skb->users) == 1)) { |
2185 | smp_rmb(); | |
2186 | atomic_set(&skb->users, 0); | |
2187 | } else if (likely(!atomic_dec_and_test(&skb->users))) { | |
2188 | return; | |
bea3348e | 2189 | } |
e6247027 ED |
2190 | get_kfree_skb_cb(skb)->reason = reason; |
2191 | local_irq_save(flags); | |
2192 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2193 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2194 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2195 | local_irq_restore(flags); | |
56079431 | 2196 | } |
e6247027 | 2197 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2198 | |
e6247027 | 2199 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2200 | { |
2201 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2202 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2203 | else |
2204 | dev_kfree_skb(skb); | |
2205 | } | |
e6247027 | 2206 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2207 | |
2208 | ||
bea3348e SH |
2209 | /** |
2210 | * netif_device_detach - mark device as removed | |
2211 | * @dev: network device | |
2212 | * | |
2213 | * Mark device as removed from system and therefore no longer available. | |
2214 | */ | |
56079431 DV |
2215 | void netif_device_detach(struct net_device *dev) |
2216 | { | |
2217 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2218 | netif_running(dev)) { | |
d543103a | 2219 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2220 | } |
2221 | } | |
2222 | EXPORT_SYMBOL(netif_device_detach); | |
2223 | ||
bea3348e SH |
2224 | /** |
2225 | * netif_device_attach - mark device as attached | |
2226 | * @dev: network device | |
2227 | * | |
2228 | * Mark device as attached from system and restart if needed. | |
2229 | */ | |
56079431 DV |
2230 | void netif_device_attach(struct net_device *dev) |
2231 | { | |
2232 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2233 | netif_running(dev)) { | |
d543103a | 2234 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2235 | __netdev_watchdog_up(dev); |
56079431 DV |
2236 | } |
2237 | } | |
2238 | EXPORT_SYMBOL(netif_device_attach); | |
2239 | ||
36c92474 BH |
2240 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2241 | { | |
65e9d2fa | 2242 | static const netdev_features_t null_features = 0; |
36c92474 BH |
2243 | struct net_device *dev = skb->dev; |
2244 | const char *driver = ""; | |
2245 | ||
c846ad9b BG |
2246 | if (!net_ratelimit()) |
2247 | return; | |
2248 | ||
36c92474 BH |
2249 | if (dev && dev->dev.parent) |
2250 | driver = dev_driver_string(dev->dev.parent); | |
2251 | ||
2252 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " | |
2253 | "gso_type=%d ip_summed=%d\n", | |
65e9d2fa MM |
2254 | driver, dev ? &dev->features : &null_features, |
2255 | skb->sk ? &skb->sk->sk_route_caps : &null_features, | |
36c92474 BH |
2256 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2257 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2258 | } | |
2259 | ||
1da177e4 LT |
2260 | /* |
2261 | * Invalidate hardware checksum when packet is to be mangled, and | |
2262 | * complete checksum manually on outgoing path. | |
2263 | */ | |
84fa7933 | 2264 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2265 | { |
d3bc23e7 | 2266 | __wsum csum; |
663ead3b | 2267 | int ret = 0, offset; |
1da177e4 | 2268 | |
84fa7933 | 2269 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2270 | goto out_set_summed; |
2271 | ||
2272 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2273 | skb_warn_bad_offload(skb); |
2274 | return -EINVAL; | |
1da177e4 LT |
2275 | } |
2276 | ||
cef401de ED |
2277 | /* Before computing a checksum, we should make sure no frag could |
2278 | * be modified by an external entity : checksum could be wrong. | |
2279 | */ | |
2280 | if (skb_has_shared_frag(skb)) { | |
2281 | ret = __skb_linearize(skb); | |
2282 | if (ret) | |
2283 | goto out; | |
2284 | } | |
2285 | ||
55508d60 | 2286 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2287 | BUG_ON(offset >= skb_headlen(skb)); |
2288 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2289 | ||
2290 | offset += skb->csum_offset; | |
2291 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2292 | ||
2293 | if (skb_cloned(skb) && | |
2294 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2295 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2296 | if (ret) | |
2297 | goto out; | |
2298 | } | |
2299 | ||
a030847e | 2300 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2301 | out_set_summed: |
1da177e4 | 2302 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2303 | out: |
1da177e4 LT |
2304 | return ret; |
2305 | } | |
d1b19dff | 2306 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2307 | |
53d6471c | 2308 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2309 | { |
4b9b1cdf | 2310 | unsigned int vlan_depth = skb->mac_len; |
252e3346 | 2311 | __be16 type = skb->protocol; |
f6a78bfc | 2312 | |
19acc327 PS |
2313 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2314 | if (type == htons(ETH_P_TEB)) { | |
2315 | struct ethhdr *eth; | |
2316 | ||
2317 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2318 | return 0; | |
2319 | ||
2320 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2321 | type = eth->h_proto; | |
2322 | } | |
2323 | ||
4b9b1cdf NA |
2324 | /* if skb->protocol is 802.1Q/AD then the header should already be |
2325 | * present at mac_len - VLAN_HLEN (if mac_len > 0), or at | |
2326 | * ETH_HLEN otherwise | |
2327 | */ | |
2328 | if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) { | |
2329 | if (vlan_depth) { | |
80019d31 | 2330 | if (WARN_ON(vlan_depth < VLAN_HLEN)) |
4b9b1cdf NA |
2331 | return 0; |
2332 | vlan_depth -= VLAN_HLEN; | |
2333 | } else { | |
2334 | vlan_depth = ETH_HLEN; | |
2335 | } | |
2336 | do { | |
2337 | struct vlan_hdr *vh; | |
2338 | ||
2339 | if (unlikely(!pskb_may_pull(skb, | |
2340 | vlan_depth + VLAN_HLEN))) | |
2341 | return 0; | |
2342 | ||
2343 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); | |
2344 | type = vh->h_vlan_encapsulated_proto; | |
2345 | vlan_depth += VLAN_HLEN; | |
2346 | } while (type == htons(ETH_P_8021Q) || | |
2347 | type == htons(ETH_P_8021AD)); | |
7b9c6090 JG |
2348 | } |
2349 | ||
53d6471c VY |
2350 | *depth = vlan_depth; |
2351 | ||
ec5f0615 PS |
2352 | return type; |
2353 | } | |
2354 | ||
2355 | /** | |
2356 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2357 | * @skb: buffer to segment | |
2358 | * @features: features for the output path (see dev->features) | |
2359 | */ | |
2360 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2361 | netdev_features_t features) | |
2362 | { | |
2363 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2364 | struct packet_offload *ptype; | |
53d6471c VY |
2365 | int vlan_depth = skb->mac_len; |
2366 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2367 | |
2368 | if (unlikely(!type)) | |
2369 | return ERR_PTR(-EINVAL); | |
2370 | ||
53d6471c | 2371 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2372 | |
2373 | rcu_read_lock(); | |
22061d80 | 2374 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2375 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
84fa7933 | 2376 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
05e8ef4a PS |
2377 | int err; |
2378 | ||
f191a1d1 | 2379 | err = ptype->callbacks.gso_send_check(skb); |
a430a43d HX |
2380 | segs = ERR_PTR(err); |
2381 | if (err || skb_gso_ok(skb, features)) | |
2382 | break; | |
d56f90a7 ACM |
2383 | __skb_push(skb, (skb->data - |
2384 | skb_network_header(skb))); | |
a430a43d | 2385 | } |
f191a1d1 | 2386 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2387 | break; |
2388 | } | |
2389 | } | |
2390 | rcu_read_unlock(); | |
2391 | ||
98e399f8 | 2392 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2393 | |
f6a78bfc HX |
2394 | return segs; |
2395 | } | |
05e8ef4a PS |
2396 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2397 | ||
2398 | ||
2399 | /* openvswitch calls this on rx path, so we need a different check. | |
2400 | */ | |
2401 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2402 | { | |
2403 | if (tx_path) | |
2404 | return skb->ip_summed != CHECKSUM_PARTIAL; | |
2405 | else | |
2406 | return skb->ip_summed == CHECKSUM_NONE; | |
2407 | } | |
2408 | ||
2409 | /** | |
2410 | * __skb_gso_segment - Perform segmentation on skb. | |
2411 | * @skb: buffer to segment | |
2412 | * @features: features for the output path (see dev->features) | |
2413 | * @tx_path: whether it is called in TX path | |
2414 | * | |
2415 | * This function segments the given skb and returns a list of segments. | |
2416 | * | |
2417 | * It may return NULL if the skb requires no segmentation. This is | |
2418 | * only possible when GSO is used for verifying header integrity. | |
2419 | */ | |
2420 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2421 | netdev_features_t features, bool tx_path) | |
2422 | { | |
2423 | if (unlikely(skb_needs_check(skb, tx_path))) { | |
2424 | int err; | |
2425 | ||
2426 | skb_warn_bad_offload(skb); | |
2427 | ||
a40e0a66 | 2428 | err = skb_cow_head(skb, 0); |
2429 | if (err < 0) | |
05e8ef4a PS |
2430 | return ERR_PTR(err); |
2431 | } | |
2432 | ||
68c33163 | 2433 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2434 | SKB_GSO_CB(skb)->encap_level = 0; |
2435 | ||
05e8ef4a PS |
2436 | skb_reset_mac_header(skb); |
2437 | skb_reset_mac_len(skb); | |
2438 | ||
2439 | return skb_mac_gso_segment(skb, features); | |
2440 | } | |
12b0004d | 2441 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2442 | |
fb286bb2 HX |
2443 | /* Take action when hardware reception checksum errors are detected. */ |
2444 | #ifdef CONFIG_BUG | |
2445 | void netdev_rx_csum_fault(struct net_device *dev) | |
2446 | { | |
2447 | if (net_ratelimit()) { | |
7b6cd1ce | 2448 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2449 | dump_stack(); |
2450 | } | |
2451 | } | |
2452 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2453 | #endif | |
2454 | ||
1da177e4 LT |
2455 | /* Actually, we should eliminate this check as soon as we know, that: |
2456 | * 1. IOMMU is present and allows to map all the memory. | |
2457 | * 2. No high memory really exists on this machine. | |
2458 | */ | |
2459 | ||
c1e756bf | 2460 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2461 | { |
3d3a8533 | 2462 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2463 | int i; |
5acbbd42 | 2464 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2465 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2466 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2467 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2468 | return 1; |
ea2ab693 | 2469 | } |
5acbbd42 | 2470 | } |
1da177e4 | 2471 | |
5acbbd42 FT |
2472 | if (PCI_DMA_BUS_IS_PHYS) { |
2473 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2474 | |
9092c658 ED |
2475 | if (!pdev) |
2476 | return 0; | |
5acbbd42 | 2477 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2478 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2479 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2480 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2481 | return 1; | |
2482 | } | |
2483 | } | |
3d3a8533 | 2484 | #endif |
1da177e4 LT |
2485 | return 0; |
2486 | } | |
1da177e4 | 2487 | |
f6a78bfc HX |
2488 | struct dev_gso_cb { |
2489 | void (*destructor)(struct sk_buff *skb); | |
2490 | }; | |
2491 | ||
2492 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
2493 | ||
2494 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
2495 | { | |
2496 | struct dev_gso_cb *cb; | |
2497 | ||
289dccbe ED |
2498 | kfree_skb_list(skb->next); |
2499 | skb->next = NULL; | |
f6a78bfc HX |
2500 | |
2501 | cb = DEV_GSO_CB(skb); | |
2502 | if (cb->destructor) | |
2503 | cb->destructor(skb); | |
2504 | } | |
2505 | ||
2506 | /** | |
2507 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
2508 | * @skb: buffer to segment | |
91ecb63c | 2509 | * @features: device features as applicable to this skb |
f6a78bfc HX |
2510 | * |
2511 | * This function segments the given skb and stores the list of segments | |
2512 | * in skb->next. | |
2513 | */ | |
c8f44aff | 2514 | static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features) |
f6a78bfc | 2515 | { |
f6a78bfc | 2516 | struct sk_buff *segs; |
576a30eb HX |
2517 | |
2518 | segs = skb_gso_segment(skb, features); | |
2519 | ||
2520 | /* Verifying header integrity only. */ | |
2521 | if (!segs) | |
2522 | return 0; | |
f6a78bfc | 2523 | |
801678c5 | 2524 | if (IS_ERR(segs)) |
f6a78bfc HX |
2525 | return PTR_ERR(segs); |
2526 | ||
2527 | skb->next = segs; | |
2528 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
2529 | skb->destructor = dev_gso_skb_destructor; | |
2530 | ||
2531 | return 0; | |
2532 | } | |
2533 | ||
3b392ddb SH |
2534 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2535 | * instead of standard features for the netdev. | |
2536 | */ | |
2537 | #ifdef CONFIG_NET_MPLS_GSO | |
2538 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2539 | netdev_features_t features, | |
2540 | __be16 type) | |
2541 | { | |
2542 | if (type == htons(ETH_P_MPLS_UC) || type == htons(ETH_P_MPLS_MC)) | |
2543 | features &= skb->dev->mpls_features; | |
2544 | ||
2545 | return features; | |
2546 | } | |
2547 | #else | |
2548 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2549 | netdev_features_t features, | |
2550 | __be16 type) | |
2551 | { | |
2552 | return features; | |
2553 | } | |
2554 | #endif | |
2555 | ||
c8f44aff | 2556 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2557 | netdev_features_t features) |
f01a5236 | 2558 | { |
53d6471c | 2559 | int tmp; |
3b392ddb SH |
2560 | __be16 type; |
2561 | ||
2562 | type = skb_network_protocol(skb, &tmp); | |
2563 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2564 | |
c0d680e5 | 2565 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2566 | !can_checksum_protocol(features, type)) { |
f01a5236 | 2567 | features &= ~NETIF_F_ALL_CSUM; |
c1e756bf | 2568 | } else if (illegal_highdma(skb->dev, skb)) { |
f01a5236 JG |
2569 | features &= ~NETIF_F_SG; |
2570 | } | |
2571 | ||
2572 | return features; | |
2573 | } | |
2574 | ||
c1e756bf | 2575 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 JG |
2576 | { |
2577 | __be16 protocol = skb->protocol; | |
c1e756bf | 2578 | netdev_features_t features = skb->dev->features; |
58e998c6 | 2579 | |
c1e756bf | 2580 | if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs) |
30b678d8 BH |
2581 | features &= ~NETIF_F_GSO_MASK; |
2582 | ||
8ad227ff | 2583 | if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) { |
58e998c6 JG |
2584 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; |
2585 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 | 2586 | } else if (!vlan_tx_tag_present(skb)) { |
c1e756bf | 2587 | return harmonize_features(skb, features); |
f01a5236 | 2588 | } |
58e998c6 | 2589 | |
c1e756bf | 2590 | features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX | |
8ad227ff | 2591 | NETIF_F_HW_VLAN_STAG_TX); |
f01a5236 | 2592 | |
cdbaa0bb | 2593 | if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) |
f01a5236 | 2594 | features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | |
8ad227ff PM |
2595 | NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX | |
2596 | NETIF_F_HW_VLAN_STAG_TX; | |
cdbaa0bb | 2597 | |
c1e756bf | 2598 | return harmonize_features(skb, features); |
58e998c6 | 2599 | } |
c1e756bf | 2600 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2601 | |
2ea25513 DM |
2602 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
2603 | struct netdev_queue *txq) | |
2604 | { | |
2605 | unsigned int len; | |
2606 | int rc; | |
2607 | ||
2608 | if (!list_empty(&ptype_all)) | |
2609 | dev_queue_xmit_nit(skb, dev); | |
2610 | ||
2611 | len = skb->len; | |
2612 | trace_net_dev_start_xmit(skb, dev); | |
fa2dbdc2 | 2613 | rc = netdev_start_xmit(skb, dev, txq, false); |
2ea25513 DM |
2614 | trace_net_dev_xmit(skb, rc, dev, len); |
2615 | ||
2616 | return rc; | |
2617 | } | |
2618 | ||
7f2e870f DM |
2619 | static struct sk_buff *xmit_list(struct sk_buff *first, struct net_device *dev, |
2620 | struct netdev_queue *txq, int *ret) | |
2621 | { | |
2622 | struct sk_buff *skb = first; | |
2623 | int rc = NETDEV_TX_OK; | |
2624 | ||
2625 | while (skb) { | |
2626 | struct sk_buff *next = skb->next; | |
2627 | ||
2628 | skb->next = NULL; | |
2629 | rc = xmit_one(skb, dev, txq); | |
2630 | if (unlikely(!dev_xmit_complete(rc))) { | |
2631 | skb->next = next; | |
2632 | goto out; | |
2633 | } | |
2634 | ||
2635 | skb = next; | |
2636 | if (netif_xmit_stopped(txq) && skb) { | |
2637 | rc = NETDEV_TX_BUSY; | |
2638 | break; | |
2639 | } | |
2640 | } | |
2641 | ||
2642 | out: | |
2643 | *ret = rc; | |
2644 | return skb; | |
2645 | } | |
2646 | ||
fd2ea0a7 | 2647 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
f663dd9a | 2648 | struct netdev_queue *txq) |
f6a78bfc | 2649 | { |
572a9d7b | 2650 | int rc = NETDEV_TX_OK; |
00829823 | 2651 | |
f6a78bfc | 2652 | if (likely(!skb->next)) { |
c8f44aff | 2653 | netdev_features_t features; |
fc741216 | 2654 | |
93f154b5 | 2655 | /* |
25985edc | 2656 | * If device doesn't need skb->dst, release it right now while |
93f154b5 ED |
2657 | * its hot in this cpu cache |
2658 | */ | |
adf30907 ED |
2659 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
2660 | skb_dst_drop(skb); | |
2661 | ||
fc741216 JG |
2662 | features = netif_skb_features(skb); |
2663 | ||
7b9c6090 | 2664 | if (vlan_tx_tag_present(skb) && |
86a9bad3 PM |
2665 | !vlan_hw_offload_capable(features, skb->vlan_proto)) { |
2666 | skb = __vlan_put_tag(skb, skb->vlan_proto, | |
2667 | vlan_tx_tag_get(skb)); | |
7b9c6090 JG |
2668 | if (unlikely(!skb)) |
2669 | goto out; | |
2670 | ||
2671 | skb->vlan_tci = 0; | |
2672 | } | |
2673 | ||
fc70fb64 AD |
2674 | /* If encapsulation offload request, verify we are testing |
2675 | * hardware encapsulation features instead of standard | |
2676 | * features for the netdev | |
2677 | */ | |
2678 | if (skb->encapsulation) | |
2679 | features &= dev->hw_enc_features; | |
2680 | ||
fc741216 | 2681 | if (netif_needs_gso(skb, features)) { |
91ecb63c | 2682 | if (unlikely(dev_gso_segment(skb, features))) |
9ccb8975 DM |
2683 | goto out_kfree_skb; |
2684 | if (skb->next) | |
2685 | goto gso; | |
6afff0ca | 2686 | } else { |
02932ce9 | 2687 | if (skb_needs_linearize(skb, features) && |
6afff0ca JF |
2688 | __skb_linearize(skb)) |
2689 | goto out_kfree_skb; | |
2690 | ||
2691 | /* If packet is not checksummed and device does not | |
2692 | * support checksumming for this protocol, complete | |
2693 | * checksumming here. | |
2694 | */ | |
2695 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
fc70fb64 AD |
2696 | if (skb->encapsulation) |
2697 | skb_set_inner_transport_header(skb, | |
2698 | skb_checksum_start_offset(skb)); | |
2699 | else | |
2700 | skb_set_transport_header(skb, | |
2701 | skb_checksum_start_offset(skb)); | |
03634668 | 2702 | if (!(features & NETIF_F_ALL_CSUM) && |
6afff0ca JF |
2703 | skb_checksum_help(skb)) |
2704 | goto out_kfree_skb; | |
2705 | } | |
9ccb8975 DM |
2706 | } |
2707 | ||
2ea25513 | 2708 | return xmit_one(skb, dev, txq); |
f6a78bfc HX |
2709 | } |
2710 | ||
576a30eb | 2711 | gso: |
7f2e870f | 2712 | skb->next = xmit_list(skb->next, dev, txq, &rc); |
0c772159 | 2713 | if (likely(skb->next == NULL)) { |
572a9d7b | 2714 | skb->destructor = DEV_GSO_CB(skb)->destructor; |
0c772159 SS |
2715 | consume_skb(skb); |
2716 | return rc; | |
2717 | } | |
f6a78bfc HX |
2718 | out_kfree_skb: |
2719 | kfree_skb(skb); | |
7b9c6090 | 2720 | out: |
572a9d7b | 2721 | return rc; |
f6a78bfc | 2722 | } |
a6cc0cfa | 2723 | EXPORT_SYMBOL_GPL(dev_hard_start_xmit); |
f6a78bfc | 2724 | |
1def9238 ED |
2725 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
2726 | { | |
2727 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2728 | ||
2729 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2730 | ||
2731 | /* To get more precise estimation of bytes sent on wire, | |
2732 | * we add to pkt_len the headers size of all segments | |
2733 | */ | |
2734 | if (shinfo->gso_size) { | |
757b8b1d | 2735 | unsigned int hdr_len; |
15e5a030 | 2736 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 2737 | |
757b8b1d ED |
2738 | /* mac layer + network layer */ |
2739 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
2740 | ||
2741 | /* + transport layer */ | |
1def9238 ED |
2742 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
2743 | hdr_len += tcp_hdrlen(skb); | |
2744 | else | |
2745 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
2746 | |
2747 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
2748 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
2749 | shinfo->gso_size); | |
2750 | ||
2751 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
2752 | } |
2753 | } | |
2754 | ||
bbd8a0d3 KK |
2755 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2756 | struct net_device *dev, | |
2757 | struct netdev_queue *txq) | |
2758 | { | |
2759 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2760 | bool contended; |
bbd8a0d3 KK |
2761 | int rc; |
2762 | ||
1def9238 | 2763 | qdisc_pkt_len_init(skb); |
a2da570d | 2764 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
2765 | /* |
2766 | * Heuristic to force contended enqueues to serialize on a | |
2767 | * separate lock before trying to get qdisc main lock. | |
9bf2b8c2 YX |
2768 | * This permits __QDISC___STATE_RUNNING owner to get the lock more |
2769 | * often and dequeue packets faster. | |
79640a4c | 2770 | */ |
a2da570d | 2771 | contended = qdisc_is_running(q); |
79640a4c ED |
2772 | if (unlikely(contended)) |
2773 | spin_lock(&q->busylock); | |
2774 | ||
bbd8a0d3 KK |
2775 | spin_lock(root_lock); |
2776 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2777 | kfree_skb(skb); | |
2778 | rc = NET_XMIT_DROP; | |
2779 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2780 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2781 | /* |
2782 | * This is a work-conserving queue; there are no old skbs | |
2783 | * waiting to be sent out; and the qdisc is not running - | |
2784 | * xmit the skb directly. | |
2785 | */ | |
7fee226a ED |
2786 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2787 | skb_dst_force(skb); | |
bfe0d029 | 2788 | |
bfe0d029 ED |
2789 | qdisc_bstats_update(q, skb); |
2790 | ||
79640a4c ED |
2791 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2792 | if (unlikely(contended)) { | |
2793 | spin_unlock(&q->busylock); | |
2794 | contended = false; | |
2795 | } | |
bbd8a0d3 | 2796 | __qdisc_run(q); |
79640a4c | 2797 | } else |
bc135b23 | 2798 | qdisc_run_end(q); |
bbd8a0d3 KK |
2799 | |
2800 | rc = NET_XMIT_SUCCESS; | |
2801 | } else { | |
7fee226a | 2802 | skb_dst_force(skb); |
a2da570d | 2803 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2804 | if (qdisc_run_begin(q)) { |
2805 | if (unlikely(contended)) { | |
2806 | spin_unlock(&q->busylock); | |
2807 | contended = false; | |
2808 | } | |
2809 | __qdisc_run(q); | |
2810 | } | |
bbd8a0d3 KK |
2811 | } |
2812 | spin_unlock(root_lock); | |
79640a4c ED |
2813 | if (unlikely(contended)) |
2814 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2815 | return rc; |
2816 | } | |
2817 | ||
86f8515f | 2818 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
2819 | static void skb_update_prio(struct sk_buff *skb) |
2820 | { | |
6977a79d | 2821 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 2822 | |
91c68ce2 ED |
2823 | if (!skb->priority && skb->sk && map) { |
2824 | unsigned int prioidx = skb->sk->sk_cgrp_prioidx; | |
2825 | ||
2826 | if (prioidx < map->priomap_len) | |
2827 | skb->priority = map->priomap[prioidx]; | |
2828 | } | |
5bc1421e NH |
2829 | } |
2830 | #else | |
2831 | #define skb_update_prio(skb) | |
2832 | #endif | |
2833 | ||
745e20f1 | 2834 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2835 | #define RECURSION_LIMIT 10 |
745e20f1 | 2836 | |
95603e22 MM |
2837 | /** |
2838 | * dev_loopback_xmit - loop back @skb | |
2839 | * @skb: buffer to transmit | |
2840 | */ | |
2841 | int dev_loopback_xmit(struct sk_buff *skb) | |
2842 | { | |
2843 | skb_reset_mac_header(skb); | |
2844 | __skb_pull(skb, skb_network_offset(skb)); | |
2845 | skb->pkt_type = PACKET_LOOPBACK; | |
2846 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2847 | WARN_ON(!skb_dst(skb)); | |
2848 | skb_dst_force(skb); | |
2849 | netif_rx_ni(skb); | |
2850 | return 0; | |
2851 | } | |
2852 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2853 | ||
d29f749e | 2854 | /** |
9d08dd3d | 2855 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 2856 | * @skb: buffer to transmit |
9d08dd3d | 2857 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
2858 | * |
2859 | * Queue a buffer for transmission to a network device. The caller must | |
2860 | * have set the device and priority and built the buffer before calling | |
2861 | * this function. The function can be called from an interrupt. | |
2862 | * | |
2863 | * A negative errno code is returned on a failure. A success does not | |
2864 | * guarantee the frame will be transmitted as it may be dropped due | |
2865 | * to congestion or traffic shaping. | |
2866 | * | |
2867 | * ----------------------------------------------------------------------------------- | |
2868 | * I notice this method can also return errors from the queue disciplines, | |
2869 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2870 | * be positive. | |
2871 | * | |
2872 | * Regardless of the return value, the skb is consumed, so it is currently | |
2873 | * difficult to retry a send to this method. (You can bump the ref count | |
2874 | * before sending to hold a reference for retry if you are careful.) | |
2875 | * | |
2876 | * When calling this method, interrupts MUST be enabled. This is because | |
2877 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2878 | * --BLG | |
2879 | */ | |
0a59f3a9 | 2880 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
2881 | { |
2882 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2883 | struct netdev_queue *txq; |
1da177e4 LT |
2884 | struct Qdisc *q; |
2885 | int rc = -ENOMEM; | |
2886 | ||
6d1ccff6 ED |
2887 | skb_reset_mac_header(skb); |
2888 | ||
e7fd2885 WB |
2889 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
2890 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
2891 | ||
4ec93edb YH |
2892 | /* Disable soft irqs for various locks below. Also |
2893 | * stops preemption for RCU. | |
1da177e4 | 2894 | */ |
4ec93edb | 2895 | rcu_read_lock_bh(); |
1da177e4 | 2896 | |
5bc1421e NH |
2897 | skb_update_prio(skb); |
2898 | ||
f663dd9a | 2899 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 2900 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2901 | |
1da177e4 | 2902 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2903 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2904 | #endif |
cf66ba58 | 2905 | trace_net_dev_queue(skb); |
1da177e4 | 2906 | if (q->enqueue) { |
bbd8a0d3 | 2907 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2908 | goto out; |
1da177e4 LT |
2909 | } |
2910 | ||
2911 | /* The device has no queue. Common case for software devices: | |
2912 | loopback, all the sorts of tunnels... | |
2913 | ||
932ff279 HX |
2914 | Really, it is unlikely that netif_tx_lock protection is necessary |
2915 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2916 | counters.) |
2917 | However, it is possible, that they rely on protection | |
2918 | made by us here. | |
2919 | ||
2920 | Check this and shot the lock. It is not prone from deadlocks. | |
2921 | Either shot noqueue qdisc, it is even simpler 8) | |
2922 | */ | |
2923 | if (dev->flags & IFF_UP) { | |
2924 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2925 | ||
c773e847 | 2926 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2927 | |
745e20f1 ED |
2928 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2929 | goto recursion_alert; | |
2930 | ||
c773e847 | 2931 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2932 | |
73466498 | 2933 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 2934 | __this_cpu_inc(xmit_recursion); |
f663dd9a | 2935 | rc = dev_hard_start_xmit(skb, dev, txq); |
745e20f1 | 2936 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2937 | if (dev_xmit_complete(rc)) { |
c773e847 | 2938 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2939 | goto out; |
2940 | } | |
2941 | } | |
c773e847 | 2942 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
2943 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
2944 | dev->name); | |
1da177e4 LT |
2945 | } else { |
2946 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2947 | * unfortunately |
2948 | */ | |
2949 | recursion_alert: | |
e87cc472 JP |
2950 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
2951 | dev->name); | |
1da177e4 LT |
2952 | } |
2953 | } | |
2954 | ||
2955 | rc = -ENETDOWN; | |
d4828d85 | 2956 | rcu_read_unlock_bh(); |
1da177e4 | 2957 | |
015f0688 | 2958 | atomic_long_inc(&dev->tx_dropped); |
1da177e4 LT |
2959 | kfree_skb(skb); |
2960 | return rc; | |
2961 | out: | |
d4828d85 | 2962 | rcu_read_unlock_bh(); |
1da177e4 LT |
2963 | return rc; |
2964 | } | |
f663dd9a JW |
2965 | |
2966 | int dev_queue_xmit(struct sk_buff *skb) | |
2967 | { | |
2968 | return __dev_queue_xmit(skb, NULL); | |
2969 | } | |
d1b19dff | 2970 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 2971 | |
f663dd9a JW |
2972 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
2973 | { | |
2974 | return __dev_queue_xmit(skb, accel_priv); | |
2975 | } | |
2976 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
2977 | ||
1da177e4 LT |
2978 | |
2979 | /*======================================================================= | |
2980 | Receiver routines | |
2981 | =======================================================================*/ | |
2982 | ||
6b2bedc3 | 2983 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
2984 | EXPORT_SYMBOL(netdev_max_backlog); |
2985 | ||
3b098e2d | 2986 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2987 | int netdev_budget __read_mostly = 300; |
2988 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2989 | |
eecfd7c4 ED |
2990 | /* Called with irq disabled */ |
2991 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2992 | struct napi_struct *napi) | |
2993 | { | |
2994 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2995 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2996 | } | |
2997 | ||
bfb564e7 KK |
2998 | #ifdef CONFIG_RPS |
2999 | ||
3000 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3001 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
3002 | EXPORT_SYMBOL(rps_sock_flow_table); |
3003 | ||
c5905afb | 3004 | struct static_key rps_needed __read_mostly; |
adc9300e | 3005 | |
c445477d BH |
3006 | static struct rps_dev_flow * |
3007 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3008 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3009 | { | |
09994d1b | 3010 | if (next_cpu != RPS_NO_CPU) { |
c445477d BH |
3011 | #ifdef CONFIG_RFS_ACCEL |
3012 | struct netdev_rx_queue *rxqueue; | |
3013 | struct rps_dev_flow_table *flow_table; | |
3014 | struct rps_dev_flow *old_rflow; | |
3015 | u32 flow_id; | |
3016 | u16 rxq_index; | |
3017 | int rc; | |
3018 | ||
3019 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3020 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3021 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3022 | goto out; |
3023 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3024 | if (rxq_index == skb_get_rx_queue(skb)) | |
3025 | goto out; | |
3026 | ||
3027 | rxqueue = dev->_rx + rxq_index; | |
3028 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3029 | if (!flow_table) | |
3030 | goto out; | |
61b905da | 3031 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3032 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3033 | rxq_index, flow_id); | |
3034 | if (rc < 0) | |
3035 | goto out; | |
3036 | old_rflow = rflow; | |
3037 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3038 | rflow->filter = rc; |
3039 | if (old_rflow->filter == rflow->filter) | |
3040 | old_rflow->filter = RPS_NO_FILTER; | |
3041 | out: | |
3042 | #endif | |
3043 | rflow->last_qtail = | |
09994d1b | 3044 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3045 | } |
3046 | ||
09994d1b | 3047 | rflow->cpu = next_cpu; |
c445477d BH |
3048 | return rflow; |
3049 | } | |
3050 | ||
bfb564e7 KK |
3051 | /* |
3052 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3053 | * CPU from the RPS map of the receiving queue for a given skb. | |
3054 | * rcu_read_lock must be held on entry. | |
3055 | */ | |
3056 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3057 | struct rps_dev_flow **rflowp) | |
3058 | { | |
3059 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 3060 | struct rps_map *map; |
bfb564e7 KK |
3061 | struct rps_dev_flow_table *flow_table; |
3062 | struct rps_sock_flow_table *sock_flow_table; | |
3063 | int cpu = -1; | |
3064 | u16 tcpu; | |
61b905da | 3065 | u32 hash; |
bfb564e7 KK |
3066 | |
3067 | if (skb_rx_queue_recorded(skb)) { | |
3068 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
3069 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3070 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3071 | "%s received packet on queue %u, but number " | |
3072 | "of RX queues is %u\n", | |
3073 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3074 | goto done; |
3075 | } | |
3076 | rxqueue = dev->_rx + index; | |
3077 | } else | |
3078 | rxqueue = dev->_rx; | |
3079 | ||
6e3f7faf ED |
3080 | map = rcu_dereference(rxqueue->rps_map); |
3081 | if (map) { | |
85875236 | 3082 | if (map->len == 1 && |
33d480ce | 3083 | !rcu_access_pointer(rxqueue->rps_flow_table)) { |
6febfca9 CG |
3084 | tcpu = map->cpus[0]; |
3085 | if (cpu_online(tcpu)) | |
3086 | cpu = tcpu; | |
3087 | goto done; | |
3088 | } | |
33d480ce | 3089 | } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) { |
bfb564e7 | 3090 | goto done; |
6febfca9 | 3091 | } |
bfb564e7 | 3092 | |
2d47b459 | 3093 | skb_reset_network_header(skb); |
61b905da TH |
3094 | hash = skb_get_hash(skb); |
3095 | if (!hash) | |
bfb564e7 KK |
3096 | goto done; |
3097 | ||
fec5e652 TH |
3098 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
3099 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
3100 | if (flow_table && sock_flow_table) { | |
3101 | u16 next_cpu; | |
3102 | struct rps_dev_flow *rflow; | |
3103 | ||
61b905da | 3104 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3105 | tcpu = rflow->cpu; |
3106 | ||
61b905da | 3107 | next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask]; |
fec5e652 TH |
3108 | |
3109 | /* | |
3110 | * If the desired CPU (where last recvmsg was done) is | |
3111 | * different from current CPU (one in the rx-queue flow | |
3112 | * table entry), switch if one of the following holds: | |
3113 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
3114 | * - Current CPU is offline. | |
3115 | * - The current CPU's queue tail has advanced beyond the | |
3116 | * last packet that was enqueued using this table entry. | |
3117 | * This guarantees that all previous packets for the flow | |
3118 | * have been dequeued, thus preserving in order delivery. | |
3119 | */ | |
3120 | if (unlikely(tcpu != next_cpu) && | |
3121 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
3122 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
baefa31d TH |
3123 | rflow->last_qtail)) >= 0)) { |
3124 | tcpu = next_cpu; | |
c445477d | 3125 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3126 | } |
c445477d | 3127 | |
fec5e652 TH |
3128 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
3129 | *rflowp = rflow; | |
3130 | cpu = tcpu; | |
3131 | goto done; | |
3132 | } | |
3133 | } | |
3134 | ||
0a9627f2 | 3135 | if (map) { |
8fc54f68 | 3136 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3137 | if (cpu_online(tcpu)) { |
3138 | cpu = tcpu; | |
3139 | goto done; | |
3140 | } | |
3141 | } | |
3142 | ||
3143 | done: | |
0a9627f2 TH |
3144 | return cpu; |
3145 | } | |
3146 | ||
c445477d BH |
3147 | #ifdef CONFIG_RFS_ACCEL |
3148 | ||
3149 | /** | |
3150 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3151 | * @dev: Device on which the filter was set | |
3152 | * @rxq_index: RX queue index | |
3153 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3154 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3155 | * | |
3156 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3157 | * this function for each installed filter and remove the filters for | |
3158 | * which it returns %true. | |
3159 | */ | |
3160 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3161 | u32 flow_id, u16 filter_id) | |
3162 | { | |
3163 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3164 | struct rps_dev_flow_table *flow_table; | |
3165 | struct rps_dev_flow *rflow; | |
3166 | bool expire = true; | |
3167 | int cpu; | |
3168 | ||
3169 | rcu_read_lock(); | |
3170 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3171 | if (flow_table && flow_id <= flow_table->mask) { | |
3172 | rflow = &flow_table->flows[flow_id]; | |
3173 | cpu = ACCESS_ONCE(rflow->cpu); | |
3174 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
3175 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
3176 | rflow->last_qtail) < | |
3177 | (int)(10 * flow_table->mask))) | |
3178 | expire = false; | |
3179 | } | |
3180 | rcu_read_unlock(); | |
3181 | return expire; | |
3182 | } | |
3183 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3184 | ||
3185 | #endif /* CONFIG_RFS_ACCEL */ | |
3186 | ||
0a9627f2 | 3187 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3188 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3189 | { |
e36fa2f7 ED |
3190 | struct softnet_data *sd = data; |
3191 | ||
eecfd7c4 | 3192 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3193 | sd->received_rps++; |
0a9627f2 | 3194 | } |
e36fa2f7 | 3195 | |
fec5e652 | 3196 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3197 | |
e36fa2f7 ED |
3198 | /* |
3199 | * Check if this softnet_data structure is another cpu one | |
3200 | * If yes, queue it to our IPI list and return 1 | |
3201 | * If no, return 0 | |
3202 | */ | |
3203 | static int rps_ipi_queued(struct softnet_data *sd) | |
3204 | { | |
3205 | #ifdef CONFIG_RPS | |
3206 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
3207 | ||
3208 | if (sd != mysd) { | |
3209 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3210 | mysd->rps_ipi_list = sd; | |
3211 | ||
3212 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3213 | return 1; | |
3214 | } | |
3215 | #endif /* CONFIG_RPS */ | |
3216 | return 0; | |
3217 | } | |
3218 | ||
99bbc707 WB |
3219 | #ifdef CONFIG_NET_FLOW_LIMIT |
3220 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3221 | #endif | |
3222 | ||
3223 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3224 | { | |
3225 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3226 | struct sd_flow_limit *fl; | |
3227 | struct softnet_data *sd; | |
3228 | unsigned int old_flow, new_flow; | |
3229 | ||
3230 | if (qlen < (netdev_max_backlog >> 1)) | |
3231 | return false; | |
3232 | ||
3233 | sd = &__get_cpu_var(softnet_data); | |
3234 | ||
3235 | rcu_read_lock(); | |
3236 | fl = rcu_dereference(sd->flow_limit); | |
3237 | if (fl) { | |
3958afa1 | 3238 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3239 | old_flow = fl->history[fl->history_head]; |
3240 | fl->history[fl->history_head] = new_flow; | |
3241 | ||
3242 | fl->history_head++; | |
3243 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3244 | ||
3245 | if (likely(fl->buckets[old_flow])) | |
3246 | fl->buckets[old_flow]--; | |
3247 | ||
3248 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3249 | fl->count++; | |
3250 | rcu_read_unlock(); | |
3251 | return true; | |
3252 | } | |
3253 | } | |
3254 | rcu_read_unlock(); | |
3255 | #endif | |
3256 | return false; | |
3257 | } | |
3258 | ||
0a9627f2 TH |
3259 | /* |
3260 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3261 | * queue (may be a remote CPU queue). | |
3262 | */ | |
fec5e652 TH |
3263 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3264 | unsigned int *qtail) | |
0a9627f2 | 3265 | { |
e36fa2f7 | 3266 | struct softnet_data *sd; |
0a9627f2 | 3267 | unsigned long flags; |
99bbc707 | 3268 | unsigned int qlen; |
0a9627f2 | 3269 | |
e36fa2f7 | 3270 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3271 | |
3272 | local_irq_save(flags); | |
0a9627f2 | 3273 | |
e36fa2f7 | 3274 | rps_lock(sd); |
99bbc707 WB |
3275 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3276 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
6e7676c1 | 3277 | if (skb_queue_len(&sd->input_pkt_queue)) { |
0a9627f2 | 3278 | enqueue: |
e36fa2f7 | 3279 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3280 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3281 | rps_unlock(sd); |
152102c7 | 3282 | local_irq_restore(flags); |
0a9627f2 TH |
3283 | return NET_RX_SUCCESS; |
3284 | } | |
3285 | ||
ebda37c2 ED |
3286 | /* Schedule NAPI for backlog device |
3287 | * We can use non atomic operation since we own the queue lock | |
3288 | */ | |
3289 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3290 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3291 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3292 | } |
3293 | goto enqueue; | |
3294 | } | |
3295 | ||
dee42870 | 3296 | sd->dropped++; |
e36fa2f7 | 3297 | rps_unlock(sd); |
0a9627f2 | 3298 | |
0a9627f2 TH |
3299 | local_irq_restore(flags); |
3300 | ||
caf586e5 | 3301 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3302 | kfree_skb(skb); |
3303 | return NET_RX_DROP; | |
3304 | } | |
1da177e4 | 3305 | |
ae78dbfa | 3306 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3307 | { |
b0e28f1e | 3308 | int ret; |
1da177e4 | 3309 | |
588f0330 | 3310 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3311 | |
cf66ba58 | 3312 | trace_netif_rx(skb); |
df334545 | 3313 | #ifdef CONFIG_RPS |
c5905afb | 3314 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3315 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3316 | int cpu; |
3317 | ||
cece1945 | 3318 | preempt_disable(); |
b0e28f1e | 3319 | rcu_read_lock(); |
fec5e652 TH |
3320 | |
3321 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3322 | if (cpu < 0) |
3323 | cpu = smp_processor_id(); | |
fec5e652 TH |
3324 | |
3325 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3326 | ||
b0e28f1e | 3327 | rcu_read_unlock(); |
cece1945 | 3328 | preempt_enable(); |
adc9300e ED |
3329 | } else |
3330 | #endif | |
fec5e652 TH |
3331 | { |
3332 | unsigned int qtail; | |
3333 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3334 | put_cpu(); | |
3335 | } | |
b0e28f1e | 3336 | return ret; |
1da177e4 | 3337 | } |
ae78dbfa BH |
3338 | |
3339 | /** | |
3340 | * netif_rx - post buffer to the network code | |
3341 | * @skb: buffer to post | |
3342 | * | |
3343 | * This function receives a packet from a device driver and queues it for | |
3344 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3345 | * may be dropped during processing for congestion control or by the | |
3346 | * protocol layers. | |
3347 | * | |
3348 | * return values: | |
3349 | * NET_RX_SUCCESS (no congestion) | |
3350 | * NET_RX_DROP (packet was dropped) | |
3351 | * | |
3352 | */ | |
3353 | ||
3354 | int netif_rx(struct sk_buff *skb) | |
3355 | { | |
3356 | trace_netif_rx_entry(skb); | |
3357 | ||
3358 | return netif_rx_internal(skb); | |
3359 | } | |
d1b19dff | 3360 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3361 | |
3362 | int netif_rx_ni(struct sk_buff *skb) | |
3363 | { | |
3364 | int err; | |
3365 | ||
ae78dbfa BH |
3366 | trace_netif_rx_ni_entry(skb); |
3367 | ||
1da177e4 | 3368 | preempt_disable(); |
ae78dbfa | 3369 | err = netif_rx_internal(skb); |
1da177e4 LT |
3370 | if (local_softirq_pending()) |
3371 | do_softirq(); | |
3372 | preempt_enable(); | |
3373 | ||
3374 | return err; | |
3375 | } | |
1da177e4 LT |
3376 | EXPORT_SYMBOL(netif_rx_ni); |
3377 | ||
1da177e4 LT |
3378 | static void net_tx_action(struct softirq_action *h) |
3379 | { | |
3380 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
3381 | ||
3382 | if (sd->completion_queue) { | |
3383 | struct sk_buff *clist; | |
3384 | ||
3385 | local_irq_disable(); | |
3386 | clist = sd->completion_queue; | |
3387 | sd->completion_queue = NULL; | |
3388 | local_irq_enable(); | |
3389 | ||
3390 | while (clist) { | |
3391 | struct sk_buff *skb = clist; | |
3392 | clist = clist->next; | |
3393 | ||
547b792c | 3394 | WARN_ON(atomic_read(&skb->users)); |
e6247027 ED |
3395 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
3396 | trace_consume_skb(skb); | |
3397 | else | |
3398 | trace_kfree_skb(skb, net_tx_action); | |
1da177e4 LT |
3399 | __kfree_skb(skb); |
3400 | } | |
3401 | } | |
3402 | ||
3403 | if (sd->output_queue) { | |
37437bb2 | 3404 | struct Qdisc *head; |
1da177e4 LT |
3405 | |
3406 | local_irq_disable(); | |
3407 | head = sd->output_queue; | |
3408 | sd->output_queue = NULL; | |
a9cbd588 | 3409 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3410 | local_irq_enable(); |
3411 | ||
3412 | while (head) { | |
37437bb2 DM |
3413 | struct Qdisc *q = head; |
3414 | spinlock_t *root_lock; | |
3415 | ||
1da177e4 LT |
3416 | head = head->next_sched; |
3417 | ||
5fb66229 | 3418 | root_lock = qdisc_lock(q); |
37437bb2 | 3419 | if (spin_trylock(root_lock)) { |
4e857c58 | 3420 | smp_mb__before_atomic(); |
def82a1d JP |
3421 | clear_bit(__QDISC_STATE_SCHED, |
3422 | &q->state); | |
37437bb2 DM |
3423 | qdisc_run(q); |
3424 | spin_unlock(root_lock); | |
1da177e4 | 3425 | } else { |
195648bb | 3426 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3427 | &q->state)) { |
195648bb | 3428 | __netif_reschedule(q); |
e8a83e10 | 3429 | } else { |
4e857c58 | 3430 | smp_mb__before_atomic(); |
e8a83e10 JP |
3431 | clear_bit(__QDISC_STATE_SCHED, |
3432 | &q->state); | |
3433 | } | |
1da177e4 LT |
3434 | } |
3435 | } | |
3436 | } | |
3437 | } | |
3438 | ||
ab95bfe0 JP |
3439 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3440 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3441 | /* This hook is defined here for ATM LANE */ |
3442 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3443 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3444 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3445 | #endif |
1da177e4 | 3446 | |
1da177e4 LT |
3447 | #ifdef CONFIG_NET_CLS_ACT |
3448 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3449 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3450 | * a compare and 2 stores extra right now if we dont have it on | |
3451 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3452 | * NOTE: This doesn't stop any functionality; if you dont have |
3453 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3454 | * |
3455 | */ | |
24824a09 | 3456 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3457 | { |
1da177e4 | 3458 | struct net_device *dev = skb->dev; |
f697c3e8 | 3459 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3460 | int result = TC_ACT_OK; |
3461 | struct Qdisc *q; | |
4ec93edb | 3462 | |
de384830 | 3463 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
e87cc472 JP |
3464 | net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n", |
3465 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3466 | return TC_ACT_SHOT; |
3467 | } | |
1da177e4 | 3468 | |
f697c3e8 HX |
3469 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3470 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3471 | |
83874000 | 3472 | q = rxq->qdisc; |
8d50b53d | 3473 | if (q != &noop_qdisc) { |
83874000 | 3474 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3475 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3476 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3477 | spin_unlock(qdisc_lock(q)); |
3478 | } | |
f697c3e8 HX |
3479 | |
3480 | return result; | |
3481 | } | |
86e65da9 | 3482 | |
f697c3e8 HX |
3483 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3484 | struct packet_type **pt_prev, | |
3485 | int *ret, struct net_device *orig_dev) | |
3486 | { | |
24824a09 ED |
3487 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3488 | ||
3489 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
f697c3e8 | 3490 | goto out; |
1da177e4 | 3491 | |
f697c3e8 HX |
3492 | if (*pt_prev) { |
3493 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3494 | *pt_prev = NULL; | |
1da177e4 LT |
3495 | } |
3496 | ||
24824a09 | 3497 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3498 | case TC_ACT_SHOT: |
3499 | case TC_ACT_STOLEN: | |
3500 | kfree_skb(skb); | |
3501 | return NULL; | |
3502 | } | |
3503 | ||
3504 | out: | |
3505 | skb->tc_verd = 0; | |
3506 | return skb; | |
1da177e4 LT |
3507 | } |
3508 | #endif | |
3509 | ||
ab95bfe0 JP |
3510 | /** |
3511 | * netdev_rx_handler_register - register receive handler | |
3512 | * @dev: device to register a handler for | |
3513 | * @rx_handler: receive handler to register | |
93e2c32b | 3514 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 3515 | * |
e227867f | 3516 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
3517 | * called from __netif_receive_skb. A negative errno code is returned |
3518 | * on a failure. | |
3519 | * | |
3520 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3521 | * |
3522 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3523 | */ |
3524 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3525 | rx_handler_func_t *rx_handler, |
3526 | void *rx_handler_data) | |
ab95bfe0 JP |
3527 | { |
3528 | ASSERT_RTNL(); | |
3529 | ||
3530 | if (dev->rx_handler) | |
3531 | return -EBUSY; | |
3532 | ||
00cfec37 | 3533 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 3534 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3535 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3536 | ||
3537 | return 0; | |
3538 | } | |
3539 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3540 | ||
3541 | /** | |
3542 | * netdev_rx_handler_unregister - unregister receive handler | |
3543 | * @dev: device to unregister a handler from | |
3544 | * | |
166ec369 | 3545 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
3546 | * |
3547 | * The caller must hold the rtnl_mutex. | |
3548 | */ | |
3549 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3550 | { | |
3551 | ||
3552 | ASSERT_RTNL(); | |
a9b3cd7f | 3553 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
3554 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
3555 | * section has a guarantee to see a non NULL rx_handler_data | |
3556 | * as well. | |
3557 | */ | |
3558 | synchronize_net(); | |
a9b3cd7f | 3559 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
3560 | } |
3561 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3562 | ||
b4b9e355 MG |
3563 | /* |
3564 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3565 | * the special handling of PFMEMALLOC skbs. | |
3566 | */ | |
3567 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3568 | { | |
3569 | switch (skb->protocol) { | |
2b8837ae JP |
3570 | case htons(ETH_P_ARP): |
3571 | case htons(ETH_P_IP): | |
3572 | case htons(ETH_P_IPV6): | |
3573 | case htons(ETH_P_8021Q): | |
3574 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
3575 | return true; |
3576 | default: | |
3577 | return false; | |
3578 | } | |
3579 | } | |
3580 | ||
9754e293 | 3581 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
3582 | { |
3583 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3584 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3585 | struct net_device *orig_dev; |
63d8ea7f | 3586 | struct net_device *null_or_dev; |
8a4eb573 | 3587 | bool deliver_exact = false; |
1da177e4 | 3588 | int ret = NET_RX_DROP; |
252e3346 | 3589 | __be16 type; |
1da177e4 | 3590 | |
588f0330 | 3591 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3592 | |
cf66ba58 | 3593 | trace_netif_receive_skb(skb); |
9b22ea56 | 3594 | |
cc9bd5ce | 3595 | orig_dev = skb->dev; |
8f903c70 | 3596 | |
c1d2bbe1 | 3597 | skb_reset_network_header(skb); |
fda55eca ED |
3598 | if (!skb_transport_header_was_set(skb)) |
3599 | skb_reset_transport_header(skb); | |
0b5c9db1 | 3600 | skb_reset_mac_len(skb); |
1da177e4 LT |
3601 | |
3602 | pt_prev = NULL; | |
3603 | ||
3604 | rcu_read_lock(); | |
3605 | ||
63d8ea7f | 3606 | another_round: |
b6858177 | 3607 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
3608 | |
3609 | __this_cpu_inc(softnet_data.processed); | |
3610 | ||
8ad227ff PM |
3611 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
3612 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 3613 | skb = skb_vlan_untag(skb); |
bcc6d479 | 3614 | if (unlikely(!skb)) |
b4b9e355 | 3615 | goto unlock; |
bcc6d479 JP |
3616 | } |
3617 | ||
1da177e4 LT |
3618 | #ifdef CONFIG_NET_CLS_ACT |
3619 | if (skb->tc_verd & TC_NCLS) { | |
3620 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3621 | goto ncls; | |
3622 | } | |
3623 | #endif | |
3624 | ||
9754e293 | 3625 | if (pfmemalloc) |
b4b9e355 MG |
3626 | goto skip_taps; |
3627 | ||
1da177e4 | 3628 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
63d8ea7f | 3629 | if (!ptype->dev || ptype->dev == skb->dev) { |
4ec93edb | 3630 | if (pt_prev) |
f2ccd8fa | 3631 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3632 | pt_prev = ptype; |
3633 | } | |
3634 | } | |
3635 | ||
b4b9e355 | 3636 | skip_taps: |
1da177e4 | 3637 | #ifdef CONFIG_NET_CLS_ACT |
f697c3e8 HX |
3638 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3639 | if (!skb) | |
b4b9e355 | 3640 | goto unlock; |
1da177e4 LT |
3641 | ncls: |
3642 | #endif | |
3643 | ||
9754e293 | 3644 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
3645 | goto drop; |
3646 | ||
2425717b JF |
3647 | if (vlan_tx_tag_present(skb)) { |
3648 | if (pt_prev) { | |
3649 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3650 | pt_prev = NULL; | |
3651 | } | |
48cc32d3 | 3652 | if (vlan_do_receive(&skb)) |
2425717b JF |
3653 | goto another_round; |
3654 | else if (unlikely(!skb)) | |
b4b9e355 | 3655 | goto unlock; |
2425717b JF |
3656 | } |
3657 | ||
48cc32d3 | 3658 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
3659 | if (rx_handler) { |
3660 | if (pt_prev) { | |
3661 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3662 | pt_prev = NULL; | |
3663 | } | |
8a4eb573 JP |
3664 | switch (rx_handler(&skb)) { |
3665 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 3666 | ret = NET_RX_SUCCESS; |
b4b9e355 | 3667 | goto unlock; |
8a4eb573 | 3668 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3669 | goto another_round; |
8a4eb573 JP |
3670 | case RX_HANDLER_EXACT: |
3671 | deliver_exact = true; | |
3672 | case RX_HANDLER_PASS: | |
3673 | break; | |
3674 | default: | |
3675 | BUG(); | |
3676 | } | |
ab95bfe0 | 3677 | } |
1da177e4 | 3678 | |
d4b812de ED |
3679 | if (unlikely(vlan_tx_tag_present(skb))) { |
3680 | if (vlan_tx_tag_get_id(skb)) | |
3681 | skb->pkt_type = PACKET_OTHERHOST; | |
3682 | /* Note: we might in the future use prio bits | |
3683 | * and set skb->priority like in vlan_do_receive() | |
3684 | * For the time being, just ignore Priority Code Point | |
3685 | */ | |
3686 | skb->vlan_tci = 0; | |
3687 | } | |
48cc32d3 | 3688 | |
63d8ea7f | 3689 | /* deliver only exact match when indicated */ |
8a4eb573 | 3690 | null_or_dev = deliver_exact ? skb->dev : NULL; |
1f3c8804 | 3691 | |
1da177e4 | 3692 | type = skb->protocol; |
82d8a867 PE |
3693 | list_for_each_entry_rcu(ptype, |
3694 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
63d8ea7f | 3695 | if (ptype->type == type && |
e3f48d37 JP |
3696 | (ptype->dev == null_or_dev || ptype->dev == skb->dev || |
3697 | ptype->dev == orig_dev)) { | |
4ec93edb | 3698 | if (pt_prev) |
f2ccd8fa | 3699 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3700 | pt_prev = ptype; |
3701 | } | |
3702 | } | |
3703 | ||
3704 | if (pt_prev) { | |
1080e512 | 3705 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 3706 | goto drop; |
1080e512 MT |
3707 | else |
3708 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 3709 | } else { |
b4b9e355 | 3710 | drop: |
caf586e5 | 3711 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3712 | kfree_skb(skb); |
3713 | /* Jamal, now you will not able to escape explaining | |
3714 | * me how you were going to use this. :-) | |
3715 | */ | |
3716 | ret = NET_RX_DROP; | |
3717 | } | |
3718 | ||
b4b9e355 | 3719 | unlock: |
1da177e4 | 3720 | rcu_read_unlock(); |
9754e293 DM |
3721 | return ret; |
3722 | } | |
3723 | ||
3724 | static int __netif_receive_skb(struct sk_buff *skb) | |
3725 | { | |
3726 | int ret; | |
3727 | ||
3728 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
3729 | unsigned long pflags = current->flags; | |
3730 | ||
3731 | /* | |
3732 | * PFMEMALLOC skbs are special, they should | |
3733 | * - be delivered to SOCK_MEMALLOC sockets only | |
3734 | * - stay away from userspace | |
3735 | * - have bounded memory usage | |
3736 | * | |
3737 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
3738 | * context down to all allocation sites. | |
3739 | */ | |
3740 | current->flags |= PF_MEMALLOC; | |
3741 | ret = __netif_receive_skb_core(skb, true); | |
3742 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
3743 | } else | |
3744 | ret = __netif_receive_skb_core(skb, false); | |
3745 | ||
1da177e4 LT |
3746 | return ret; |
3747 | } | |
0a9627f2 | 3748 | |
ae78dbfa | 3749 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 3750 | { |
588f0330 | 3751 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3752 | |
c1f19b51 RC |
3753 | if (skb_defer_rx_timestamp(skb)) |
3754 | return NET_RX_SUCCESS; | |
3755 | ||
df334545 | 3756 | #ifdef CONFIG_RPS |
c5905afb | 3757 | if (static_key_false(&rps_needed)) { |
3b098e2d ED |
3758 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
3759 | int cpu, ret; | |
fec5e652 | 3760 | |
3b098e2d ED |
3761 | rcu_read_lock(); |
3762 | ||
3763 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3764 | |
3b098e2d ED |
3765 | if (cpu >= 0) { |
3766 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3767 | rcu_read_unlock(); | |
adc9300e | 3768 | return ret; |
3b098e2d | 3769 | } |
adc9300e | 3770 | rcu_read_unlock(); |
fec5e652 | 3771 | } |
1e94d72f | 3772 | #endif |
adc9300e | 3773 | return __netif_receive_skb(skb); |
0a9627f2 | 3774 | } |
ae78dbfa BH |
3775 | |
3776 | /** | |
3777 | * netif_receive_skb - process receive buffer from network | |
3778 | * @skb: buffer to process | |
3779 | * | |
3780 | * netif_receive_skb() is the main receive data processing function. | |
3781 | * It always succeeds. The buffer may be dropped during processing | |
3782 | * for congestion control or by the protocol layers. | |
3783 | * | |
3784 | * This function may only be called from softirq context and interrupts | |
3785 | * should be enabled. | |
3786 | * | |
3787 | * Return values (usually ignored): | |
3788 | * NET_RX_SUCCESS: no congestion | |
3789 | * NET_RX_DROP: packet was dropped | |
3790 | */ | |
3791 | int netif_receive_skb(struct sk_buff *skb) | |
3792 | { | |
3793 | trace_netif_receive_skb_entry(skb); | |
3794 | ||
3795 | return netif_receive_skb_internal(skb); | |
3796 | } | |
d1b19dff | 3797 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3798 | |
88751275 ED |
3799 | /* Network device is going away, flush any packets still pending |
3800 | * Called with irqs disabled. | |
3801 | */ | |
152102c7 | 3802 | static void flush_backlog(void *arg) |
6e583ce5 | 3803 | { |
152102c7 | 3804 | struct net_device *dev = arg; |
e36fa2f7 | 3805 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3806 | struct sk_buff *skb, *tmp; |
3807 | ||
e36fa2f7 | 3808 | rps_lock(sd); |
6e7676c1 | 3809 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3810 | if (skb->dev == dev) { |
e36fa2f7 | 3811 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3812 | kfree_skb(skb); |
76cc8b13 | 3813 | input_queue_head_incr(sd); |
6e583ce5 | 3814 | } |
6e7676c1 | 3815 | } |
e36fa2f7 | 3816 | rps_unlock(sd); |
6e7676c1 CG |
3817 | |
3818 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3819 | if (skb->dev == dev) { | |
3820 | __skb_unlink(skb, &sd->process_queue); | |
3821 | kfree_skb(skb); | |
76cc8b13 | 3822 | input_queue_head_incr(sd); |
6e7676c1 CG |
3823 | } |
3824 | } | |
6e583ce5 SH |
3825 | } |
3826 | ||
d565b0a1 HX |
3827 | static int napi_gro_complete(struct sk_buff *skb) |
3828 | { | |
22061d80 | 3829 | struct packet_offload *ptype; |
d565b0a1 | 3830 | __be16 type = skb->protocol; |
22061d80 | 3831 | struct list_head *head = &offload_base; |
d565b0a1 HX |
3832 | int err = -ENOENT; |
3833 | ||
c3c7c254 ED |
3834 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
3835 | ||
fc59f9a3 HX |
3836 | if (NAPI_GRO_CB(skb)->count == 1) { |
3837 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3838 | goto out; |
fc59f9a3 | 3839 | } |
d565b0a1 HX |
3840 | |
3841 | rcu_read_lock(); | |
3842 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3843 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
3844 | continue; |
3845 | ||
299603e8 | 3846 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
3847 | break; |
3848 | } | |
3849 | rcu_read_unlock(); | |
3850 | ||
3851 | if (err) { | |
3852 | WARN_ON(&ptype->list == head); | |
3853 | kfree_skb(skb); | |
3854 | return NET_RX_SUCCESS; | |
3855 | } | |
3856 | ||
3857 | out: | |
ae78dbfa | 3858 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
3859 | } |
3860 | ||
2e71a6f8 ED |
3861 | /* napi->gro_list contains packets ordered by age. |
3862 | * youngest packets at the head of it. | |
3863 | * Complete skbs in reverse order to reduce latencies. | |
3864 | */ | |
3865 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 3866 | { |
2e71a6f8 | 3867 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 3868 | |
2e71a6f8 ED |
3869 | /* scan list and build reverse chain */ |
3870 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
3871 | skb->prev = prev; | |
3872 | prev = skb; | |
3873 | } | |
3874 | ||
3875 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 3876 | skb->next = NULL; |
2e71a6f8 ED |
3877 | |
3878 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
3879 | return; | |
3880 | ||
3881 | prev = skb->prev; | |
d565b0a1 | 3882 | napi_gro_complete(skb); |
2e71a6f8 | 3883 | napi->gro_count--; |
d565b0a1 HX |
3884 | } |
3885 | ||
3886 | napi->gro_list = NULL; | |
3887 | } | |
86cac58b | 3888 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3889 | |
89c5fa33 ED |
3890 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
3891 | { | |
3892 | struct sk_buff *p; | |
3893 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 3894 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
3895 | |
3896 | for (p = napi->gro_list; p; p = p->next) { | |
3897 | unsigned long diffs; | |
3898 | ||
0b4cec8c TH |
3899 | NAPI_GRO_CB(p)->flush = 0; |
3900 | ||
3901 | if (hash != skb_get_hash_raw(p)) { | |
3902 | NAPI_GRO_CB(p)->same_flow = 0; | |
3903 | continue; | |
3904 | } | |
3905 | ||
89c5fa33 ED |
3906 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
3907 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
3908 | if (maclen == ETH_HLEN) | |
3909 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 3910 | skb_mac_header(skb)); |
89c5fa33 ED |
3911 | else if (!diffs) |
3912 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 3913 | skb_mac_header(skb), |
89c5fa33 ED |
3914 | maclen); |
3915 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
3916 | } |
3917 | } | |
3918 | ||
299603e8 JC |
3919 | static void skb_gro_reset_offset(struct sk_buff *skb) |
3920 | { | |
3921 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
3922 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
3923 | ||
3924 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3925 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
3926 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
3927 | ||
3928 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
3929 | pinfo->nr_frags && | |
3930 | !PageHighMem(skb_frag_page(frag0))) { | |
3931 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
3932 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
89c5fa33 ED |
3933 | } |
3934 | } | |
3935 | ||
a50e233c ED |
3936 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
3937 | { | |
3938 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
3939 | ||
3940 | BUG_ON(skb->end - skb->tail < grow); | |
3941 | ||
3942 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3943 | ||
3944 | skb->data_len -= grow; | |
3945 | skb->tail += grow; | |
3946 | ||
3947 | pinfo->frags[0].page_offset += grow; | |
3948 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
3949 | ||
3950 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
3951 | skb_frag_unref(skb, 0); | |
3952 | memmove(pinfo->frags, pinfo->frags + 1, | |
3953 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
3954 | } | |
3955 | } | |
3956 | ||
bb728820 | 3957 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3958 | { |
3959 | struct sk_buff **pp = NULL; | |
22061d80 | 3960 | struct packet_offload *ptype; |
d565b0a1 | 3961 | __be16 type = skb->protocol; |
22061d80 | 3962 | struct list_head *head = &offload_base; |
0da2afd5 | 3963 | int same_flow; |
5b252f0c | 3964 | enum gro_result ret; |
a50e233c | 3965 | int grow; |
d565b0a1 | 3966 | |
9c62a68d | 3967 | if (!(skb->dev->features & NETIF_F_GRO)) |
d565b0a1 HX |
3968 | goto normal; |
3969 | ||
21dc3301 | 3970 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) |
f17f5c91 HX |
3971 | goto normal; |
3972 | ||
89c5fa33 | 3973 | gro_list_prepare(napi, skb); |
573e8fca | 3974 | |
d565b0a1 HX |
3975 | rcu_read_lock(); |
3976 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3977 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
3978 | continue; |
3979 | ||
86911732 | 3980 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 3981 | skb_reset_mac_len(skb); |
d565b0a1 HX |
3982 | NAPI_GRO_CB(skb)->same_flow = 0; |
3983 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3984 | NAPI_GRO_CB(skb)->free = 0; |
b582ef09 | 3985 | NAPI_GRO_CB(skb)->udp_mark = 0; |
662880f4 TH |
3986 | |
3987 | /* Setup for GRO checksum validation */ | |
3988 | switch (skb->ip_summed) { | |
3989 | case CHECKSUM_COMPLETE: | |
3990 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
3991 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
3992 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
3993 | break; | |
3994 | case CHECKSUM_UNNECESSARY: | |
3995 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
3996 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
3997 | break; | |
3998 | default: | |
3999 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4000 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4001 | } | |
d565b0a1 | 4002 | |
f191a1d1 | 4003 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4004 | break; |
4005 | } | |
4006 | rcu_read_unlock(); | |
4007 | ||
4008 | if (&ptype->list == head) | |
4009 | goto normal; | |
4010 | ||
0da2afd5 | 4011 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4012 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4013 | |
d565b0a1 HX |
4014 | if (pp) { |
4015 | struct sk_buff *nskb = *pp; | |
4016 | ||
4017 | *pp = nskb->next; | |
4018 | nskb->next = NULL; | |
4019 | napi_gro_complete(nskb); | |
4ae5544f | 4020 | napi->gro_count--; |
d565b0a1 HX |
4021 | } |
4022 | ||
0da2afd5 | 4023 | if (same_flow) |
d565b0a1 HX |
4024 | goto ok; |
4025 | ||
600adc18 | 4026 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4027 | goto normal; |
d565b0a1 | 4028 | |
600adc18 ED |
4029 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4030 | struct sk_buff *nskb = napi->gro_list; | |
4031 | ||
4032 | /* locate the end of the list to select the 'oldest' flow */ | |
4033 | while (nskb->next) { | |
4034 | pp = &nskb->next; | |
4035 | nskb = *pp; | |
4036 | } | |
4037 | *pp = NULL; | |
4038 | nskb->next = NULL; | |
4039 | napi_gro_complete(nskb); | |
4040 | } else { | |
4041 | napi->gro_count++; | |
4042 | } | |
d565b0a1 | 4043 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4044 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4045 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4046 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4047 | skb->next = napi->gro_list; |
4048 | napi->gro_list = skb; | |
5d0d9be8 | 4049 | ret = GRO_HELD; |
d565b0a1 | 4050 | |
ad0f9904 | 4051 | pull: |
a50e233c ED |
4052 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4053 | if (grow > 0) | |
4054 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4055 | ok: |
5d0d9be8 | 4056 | return ret; |
d565b0a1 HX |
4057 | |
4058 | normal: | |
ad0f9904 HX |
4059 | ret = GRO_NORMAL; |
4060 | goto pull; | |
5d38a079 | 4061 | } |
96e93eab | 4062 | |
bf5a755f JC |
4063 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4064 | { | |
4065 | struct list_head *offload_head = &offload_base; | |
4066 | struct packet_offload *ptype; | |
4067 | ||
4068 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4069 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4070 | continue; | |
4071 | return ptype; | |
4072 | } | |
4073 | return NULL; | |
4074 | } | |
e27a2f83 | 4075 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4076 | |
4077 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4078 | { | |
4079 | struct list_head *offload_head = &offload_base; | |
4080 | struct packet_offload *ptype; | |
4081 | ||
4082 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4083 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4084 | continue; | |
4085 | return ptype; | |
4086 | } | |
4087 | return NULL; | |
4088 | } | |
e27a2f83 | 4089 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4090 | |
bb728820 | 4091 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4092 | { |
5d0d9be8 HX |
4093 | switch (ret) { |
4094 | case GRO_NORMAL: | |
ae78dbfa | 4095 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4096 | ret = GRO_DROP; |
4097 | break; | |
5d38a079 | 4098 | |
5d0d9be8 | 4099 | case GRO_DROP: |
5d38a079 HX |
4100 | kfree_skb(skb); |
4101 | break; | |
5b252f0c | 4102 | |
daa86548 | 4103 | case GRO_MERGED_FREE: |
d7e8883c ED |
4104 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
4105 | kmem_cache_free(skbuff_head_cache, skb); | |
4106 | else | |
4107 | __kfree_skb(skb); | |
daa86548 ED |
4108 | break; |
4109 | ||
5b252f0c BH |
4110 | case GRO_HELD: |
4111 | case GRO_MERGED: | |
4112 | break; | |
5d38a079 HX |
4113 | } |
4114 | ||
c7c4b3b6 | 4115 | return ret; |
5d0d9be8 | 4116 | } |
5d0d9be8 | 4117 | |
c7c4b3b6 | 4118 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4119 | { |
ae78dbfa | 4120 | trace_napi_gro_receive_entry(skb); |
86911732 | 4121 | |
a50e233c ED |
4122 | skb_gro_reset_offset(skb); |
4123 | ||
89c5fa33 | 4124 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4125 | } |
4126 | EXPORT_SYMBOL(napi_gro_receive); | |
4127 | ||
d0c2b0d2 | 4128 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4129 | { |
96e93eab | 4130 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4131 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4132 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4133 | skb->vlan_tci = 0; |
66c46d74 | 4134 | skb->dev = napi->dev; |
6d152e23 | 4135 | skb->skb_iif = 0; |
c3caf119 JC |
4136 | skb->encapsulation = 0; |
4137 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4138 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
96e93eab HX |
4139 | |
4140 | napi->skb = skb; | |
4141 | } | |
96e93eab | 4142 | |
76620aaf | 4143 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4144 | { |
5d38a079 | 4145 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4146 | |
4147 | if (!skb) { | |
89d71a66 | 4148 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
84b9cd63 | 4149 | napi->skb = skb; |
80595d59 | 4150 | } |
96e93eab HX |
4151 | return skb; |
4152 | } | |
76620aaf | 4153 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4154 | |
a50e233c ED |
4155 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4156 | struct sk_buff *skb, | |
4157 | gro_result_t ret) | |
96e93eab | 4158 | { |
5d0d9be8 HX |
4159 | switch (ret) { |
4160 | case GRO_NORMAL: | |
a50e233c ED |
4161 | case GRO_HELD: |
4162 | __skb_push(skb, ETH_HLEN); | |
4163 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4164 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4165 | ret = GRO_DROP; |
86911732 | 4166 | break; |
5d38a079 | 4167 | |
5d0d9be8 | 4168 | case GRO_DROP: |
5d0d9be8 HX |
4169 | case GRO_MERGED_FREE: |
4170 | napi_reuse_skb(napi, skb); | |
4171 | break; | |
5b252f0c BH |
4172 | |
4173 | case GRO_MERGED: | |
4174 | break; | |
5d0d9be8 | 4175 | } |
5d38a079 | 4176 | |
c7c4b3b6 | 4177 | return ret; |
5d38a079 | 4178 | } |
5d0d9be8 | 4179 | |
a50e233c ED |
4180 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
4181 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
4182 | * We copy ethernet header into skb->data to have a common layout. | |
4183 | */ | |
4adb9c4a | 4184 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4185 | { |
4186 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
4187 | const struct ethhdr *eth; |
4188 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
4189 | |
4190 | napi->skb = NULL; | |
4191 | ||
a50e233c ED |
4192 | skb_reset_mac_header(skb); |
4193 | skb_gro_reset_offset(skb); | |
4194 | ||
4195 | eth = skb_gro_header_fast(skb, 0); | |
4196 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
4197 | eth = skb_gro_header_slow(skb, hlen, 0); | |
4198 | if (unlikely(!eth)) { | |
4199 | napi_reuse_skb(napi, skb); | |
4200 | return NULL; | |
4201 | } | |
4202 | } else { | |
4203 | gro_pull_from_frag0(skb, hlen); | |
4204 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
4205 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 4206 | } |
a50e233c ED |
4207 | __skb_pull(skb, hlen); |
4208 | ||
4209 | /* | |
4210 | * This works because the only protocols we care about don't require | |
4211 | * special handling. | |
4212 | * We'll fix it up properly in napi_frags_finish() | |
4213 | */ | |
4214 | skb->protocol = eth->h_proto; | |
76620aaf | 4215 | |
76620aaf HX |
4216 | return skb; |
4217 | } | |
76620aaf | 4218 | |
c7c4b3b6 | 4219 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4220 | { |
76620aaf | 4221 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4222 | |
4223 | if (!skb) | |
c7c4b3b6 | 4224 | return GRO_DROP; |
5d0d9be8 | 4225 | |
ae78dbfa BH |
4226 | trace_napi_gro_frags_entry(skb); |
4227 | ||
89c5fa33 | 4228 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4229 | } |
5d38a079 HX |
4230 | EXPORT_SYMBOL(napi_gro_frags); |
4231 | ||
573e8fca TH |
4232 | /* Compute the checksum from gro_offset and return the folded value |
4233 | * after adding in any pseudo checksum. | |
4234 | */ | |
4235 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
4236 | { | |
4237 | __wsum wsum; | |
4238 | __sum16 sum; | |
4239 | ||
4240 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
4241 | ||
4242 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
4243 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
4244 | if (likely(!sum)) { | |
4245 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
4246 | !skb->csum_complete_sw) | |
4247 | netdev_rx_csum_fault(skb->dev); | |
4248 | } | |
4249 | ||
4250 | NAPI_GRO_CB(skb)->csum = wsum; | |
4251 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4252 | ||
4253 | return sum; | |
4254 | } | |
4255 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
4256 | ||
e326bed2 | 4257 | /* |
855abcf0 | 4258 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
4259 | * Note: called with local irq disabled, but exits with local irq enabled. |
4260 | */ | |
4261 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4262 | { | |
4263 | #ifdef CONFIG_RPS | |
4264 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4265 | ||
4266 | if (remsd) { | |
4267 | sd->rps_ipi_list = NULL; | |
4268 | ||
4269 | local_irq_enable(); | |
4270 | ||
4271 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
4272 | while (remsd) { | |
4273 | struct softnet_data *next = remsd->rps_ipi_next; | |
4274 | ||
4275 | if (cpu_online(remsd->cpu)) | |
c46fff2a | 4276 | smp_call_function_single_async(remsd->cpu, |
fce8ad15 | 4277 | &remsd->csd); |
e326bed2 ED |
4278 | remsd = next; |
4279 | } | |
4280 | } else | |
4281 | #endif | |
4282 | local_irq_enable(); | |
4283 | } | |
4284 | ||
bea3348e | 4285 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
4286 | { |
4287 | int work = 0; | |
eecfd7c4 | 4288 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 4289 | |
e326bed2 ED |
4290 | #ifdef CONFIG_RPS |
4291 | /* Check if we have pending ipi, its better to send them now, | |
4292 | * not waiting net_rx_action() end. | |
4293 | */ | |
4294 | if (sd->rps_ipi_list) { | |
4295 | local_irq_disable(); | |
4296 | net_rps_action_and_irq_enable(sd); | |
4297 | } | |
4298 | #endif | |
bea3348e | 4299 | napi->weight = weight_p; |
6e7676c1 | 4300 | local_irq_disable(); |
11ef7a89 | 4301 | while (1) { |
1da177e4 | 4302 | struct sk_buff *skb; |
6e7676c1 CG |
4303 | |
4304 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
4305 | local_irq_enable(); | |
4306 | __netif_receive_skb(skb); | |
6e7676c1 | 4307 | local_irq_disable(); |
76cc8b13 TH |
4308 | input_queue_head_incr(sd); |
4309 | if (++work >= quota) { | |
4310 | local_irq_enable(); | |
4311 | return work; | |
4312 | } | |
6e7676c1 | 4313 | } |
1da177e4 | 4314 | |
e36fa2f7 | 4315 | rps_lock(sd); |
11ef7a89 | 4316 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
4317 | /* |
4318 | * Inline a custom version of __napi_complete(). | |
4319 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
4320 | * and NAPI_STATE_SCHED is the only possible flag set |
4321 | * on backlog. | |
4322 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
4323 | * and we dont need an smp_mb() memory barrier. |
4324 | */ | |
4325 | list_del(&napi->poll_list); | |
4326 | napi->state = 0; | |
11ef7a89 | 4327 | rps_unlock(sd); |
eecfd7c4 | 4328 | |
11ef7a89 | 4329 | break; |
bea3348e | 4330 | } |
11ef7a89 TH |
4331 | |
4332 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
4333 | &sd->process_queue); | |
e36fa2f7 | 4334 | rps_unlock(sd); |
6e7676c1 CG |
4335 | } |
4336 | local_irq_enable(); | |
1da177e4 | 4337 | |
bea3348e SH |
4338 | return work; |
4339 | } | |
1da177e4 | 4340 | |
bea3348e SH |
4341 | /** |
4342 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4343 | * @n: entry to schedule |
bea3348e SH |
4344 | * |
4345 | * The entry's receive function will be scheduled to run | |
4346 | */ | |
b5606c2d | 4347 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4348 | { |
4349 | unsigned long flags; | |
1da177e4 | 4350 | |
bea3348e | 4351 | local_irq_save(flags); |
eecfd7c4 | 4352 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 4353 | local_irq_restore(flags); |
1da177e4 | 4354 | } |
bea3348e SH |
4355 | EXPORT_SYMBOL(__napi_schedule); |
4356 | ||
d565b0a1 HX |
4357 | void __napi_complete(struct napi_struct *n) |
4358 | { | |
4359 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
4360 | BUG_ON(n->gro_list); | |
4361 | ||
4362 | list_del(&n->poll_list); | |
4e857c58 | 4363 | smp_mb__before_atomic(); |
d565b0a1 HX |
4364 | clear_bit(NAPI_STATE_SCHED, &n->state); |
4365 | } | |
4366 | EXPORT_SYMBOL(__napi_complete); | |
4367 | ||
4368 | void napi_complete(struct napi_struct *n) | |
4369 | { | |
4370 | unsigned long flags; | |
4371 | ||
4372 | /* | |
4373 | * don't let napi dequeue from the cpu poll list | |
4374 | * just in case its running on a different cpu | |
4375 | */ | |
4376 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4377 | return; | |
4378 | ||
2e71a6f8 | 4379 | napi_gro_flush(n, false); |
d565b0a1 HX |
4380 | local_irq_save(flags); |
4381 | __napi_complete(n); | |
4382 | local_irq_restore(flags); | |
4383 | } | |
4384 | EXPORT_SYMBOL(napi_complete); | |
4385 | ||
af12fa6e ET |
4386 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
4387 | struct napi_struct *napi_by_id(unsigned int napi_id) | |
4388 | { | |
4389 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
4390 | struct napi_struct *napi; | |
4391 | ||
4392 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
4393 | if (napi->napi_id == napi_id) | |
4394 | return napi; | |
4395 | ||
4396 | return NULL; | |
4397 | } | |
4398 | EXPORT_SYMBOL_GPL(napi_by_id); | |
4399 | ||
4400 | void napi_hash_add(struct napi_struct *napi) | |
4401 | { | |
4402 | if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) { | |
4403 | ||
4404 | spin_lock(&napi_hash_lock); | |
4405 | ||
4406 | /* 0 is not a valid id, we also skip an id that is taken | |
4407 | * we expect both events to be extremely rare | |
4408 | */ | |
4409 | napi->napi_id = 0; | |
4410 | while (!napi->napi_id) { | |
4411 | napi->napi_id = ++napi_gen_id; | |
4412 | if (napi_by_id(napi->napi_id)) | |
4413 | napi->napi_id = 0; | |
4414 | } | |
4415 | ||
4416 | hlist_add_head_rcu(&napi->napi_hash_node, | |
4417 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
4418 | ||
4419 | spin_unlock(&napi_hash_lock); | |
4420 | } | |
4421 | } | |
4422 | EXPORT_SYMBOL_GPL(napi_hash_add); | |
4423 | ||
4424 | /* Warning : caller is responsible to make sure rcu grace period | |
4425 | * is respected before freeing memory containing @napi | |
4426 | */ | |
4427 | void napi_hash_del(struct napi_struct *napi) | |
4428 | { | |
4429 | spin_lock(&napi_hash_lock); | |
4430 | ||
4431 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) | |
4432 | hlist_del_rcu(&napi->napi_hash_node); | |
4433 | ||
4434 | spin_unlock(&napi_hash_lock); | |
4435 | } | |
4436 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
4437 | ||
d565b0a1 HX |
4438 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
4439 | int (*poll)(struct napi_struct *, int), int weight) | |
4440 | { | |
4441 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 4442 | napi->gro_count = 0; |
d565b0a1 | 4443 | napi->gro_list = NULL; |
5d38a079 | 4444 | napi->skb = NULL; |
d565b0a1 | 4445 | napi->poll = poll; |
82dc3c63 ED |
4446 | if (weight > NAPI_POLL_WEIGHT) |
4447 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
4448 | weight, dev->name); | |
d565b0a1 HX |
4449 | napi->weight = weight; |
4450 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 4451 | napi->dev = dev; |
5d38a079 | 4452 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
4453 | spin_lock_init(&napi->poll_lock); |
4454 | napi->poll_owner = -1; | |
4455 | #endif | |
4456 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
4457 | } | |
4458 | EXPORT_SYMBOL(netif_napi_add); | |
4459 | ||
4460 | void netif_napi_del(struct napi_struct *napi) | |
4461 | { | |
d7b06636 | 4462 | list_del_init(&napi->dev_list); |
76620aaf | 4463 | napi_free_frags(napi); |
d565b0a1 | 4464 | |
289dccbe | 4465 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 4466 | napi->gro_list = NULL; |
4ae5544f | 4467 | napi->gro_count = 0; |
d565b0a1 HX |
4468 | } |
4469 | EXPORT_SYMBOL(netif_napi_del); | |
4470 | ||
1da177e4 LT |
4471 | static void net_rx_action(struct softirq_action *h) |
4472 | { | |
e326bed2 | 4473 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 4474 | unsigned long time_limit = jiffies + 2; |
51b0bded | 4475 | int budget = netdev_budget; |
53fb95d3 MM |
4476 | void *have; |
4477 | ||
1da177e4 LT |
4478 | local_irq_disable(); |
4479 | ||
e326bed2 | 4480 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
4481 | struct napi_struct *n; |
4482 | int work, weight; | |
1da177e4 | 4483 | |
bea3348e | 4484 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
4485 | * Allow this to run for 2 jiffies since which will allow |
4486 | * an average latency of 1.5/HZ. | |
bea3348e | 4487 | */ |
d1f41b67 | 4488 | if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit))) |
1da177e4 LT |
4489 | goto softnet_break; |
4490 | ||
4491 | local_irq_enable(); | |
4492 | ||
bea3348e SH |
4493 | /* Even though interrupts have been re-enabled, this |
4494 | * access is safe because interrupts can only add new | |
4495 | * entries to the tail of this list, and only ->poll() | |
4496 | * calls can remove this head entry from the list. | |
4497 | */ | |
e326bed2 | 4498 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 4499 | |
bea3348e SH |
4500 | have = netpoll_poll_lock(n); |
4501 | ||
4502 | weight = n->weight; | |
4503 | ||
0a7606c1 DM |
4504 | /* This NAPI_STATE_SCHED test is for avoiding a race |
4505 | * with netpoll's poll_napi(). Only the entity which | |
4506 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
4507 | * actually make the ->poll() call. Therefore we avoid | |
25985edc | 4508 | * accidentally calling ->poll() when NAPI is not scheduled. |
0a7606c1 DM |
4509 | */ |
4510 | work = 0; | |
4ea7e386 | 4511 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 4512 | work = n->poll(n, weight); |
4ea7e386 NH |
4513 | trace_napi_poll(n); |
4514 | } | |
bea3348e SH |
4515 | |
4516 | WARN_ON_ONCE(work > weight); | |
4517 | ||
4518 | budget -= work; | |
4519 | ||
4520 | local_irq_disable(); | |
4521 | ||
4522 | /* Drivers must not modify the NAPI state if they | |
4523 | * consume the entire weight. In such cases this code | |
4524 | * still "owns" the NAPI instance and therefore can | |
4525 | * move the instance around on the list at-will. | |
4526 | */ | |
fed17f30 | 4527 | if (unlikely(work == weight)) { |
ff780cd8 HX |
4528 | if (unlikely(napi_disable_pending(n))) { |
4529 | local_irq_enable(); | |
4530 | napi_complete(n); | |
4531 | local_irq_disable(); | |
2e71a6f8 ED |
4532 | } else { |
4533 | if (n->gro_list) { | |
4534 | /* flush too old packets | |
4535 | * If HZ < 1000, flush all packets. | |
4536 | */ | |
4537 | local_irq_enable(); | |
4538 | napi_gro_flush(n, HZ >= 1000); | |
4539 | local_irq_disable(); | |
4540 | } | |
e326bed2 | 4541 | list_move_tail(&n->poll_list, &sd->poll_list); |
2e71a6f8 | 4542 | } |
fed17f30 | 4543 | } |
bea3348e SH |
4544 | |
4545 | netpoll_poll_unlock(have); | |
1da177e4 LT |
4546 | } |
4547 | out: | |
e326bed2 | 4548 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 4549 | |
db217334 CL |
4550 | #ifdef CONFIG_NET_DMA |
4551 | /* | |
4552 | * There may not be any more sk_buffs coming right now, so push | |
4553 | * any pending DMA copies to hardware | |
4554 | */ | |
2ba05622 | 4555 | dma_issue_pending_all(); |
db217334 | 4556 | #endif |
bea3348e | 4557 | |
1da177e4 LT |
4558 | return; |
4559 | ||
4560 | softnet_break: | |
dee42870 | 4561 | sd->time_squeeze++; |
1da177e4 LT |
4562 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
4563 | goto out; | |
4564 | } | |
4565 | ||
aa9d8560 | 4566 | struct netdev_adjacent { |
9ff162a8 | 4567 | struct net_device *dev; |
5d261913 VF |
4568 | |
4569 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 4570 | bool master; |
5d261913 | 4571 | |
5d261913 VF |
4572 | /* counter for the number of times this device was added to us */ |
4573 | u16 ref_nr; | |
4574 | ||
402dae96 VF |
4575 | /* private field for the users */ |
4576 | void *private; | |
4577 | ||
9ff162a8 JP |
4578 | struct list_head list; |
4579 | struct rcu_head rcu; | |
9ff162a8 JP |
4580 | }; |
4581 | ||
5d261913 VF |
4582 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev, |
4583 | struct net_device *adj_dev, | |
2f268f12 | 4584 | struct list_head *adj_list) |
9ff162a8 | 4585 | { |
5d261913 | 4586 | struct netdev_adjacent *adj; |
5d261913 | 4587 | |
2f268f12 | 4588 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
4589 | if (adj->dev == adj_dev) |
4590 | return adj; | |
9ff162a8 JP |
4591 | } |
4592 | return NULL; | |
4593 | } | |
4594 | ||
4595 | /** | |
4596 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
4597 | * @dev: device | |
4598 | * @upper_dev: upper device to check | |
4599 | * | |
4600 | * Find out if a device is linked to specified upper device and return true | |
4601 | * in case it is. Note that this checks only immediate upper device, | |
4602 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
4603 | */ | |
4604 | bool netdev_has_upper_dev(struct net_device *dev, | |
4605 | struct net_device *upper_dev) | |
4606 | { | |
4607 | ASSERT_RTNL(); | |
4608 | ||
2f268f12 | 4609 | return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper); |
9ff162a8 JP |
4610 | } |
4611 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
4612 | ||
4613 | /** | |
4614 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
4615 | * @dev: device | |
4616 | * | |
4617 | * Find out if a device is linked to an upper device and return true in case | |
4618 | * it is. The caller must hold the RTNL lock. | |
4619 | */ | |
1d143d9f | 4620 | static bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
4621 | { |
4622 | ASSERT_RTNL(); | |
4623 | ||
2f268f12 | 4624 | return !list_empty(&dev->all_adj_list.upper); |
9ff162a8 | 4625 | } |
9ff162a8 JP |
4626 | |
4627 | /** | |
4628 | * netdev_master_upper_dev_get - Get master upper device | |
4629 | * @dev: device | |
4630 | * | |
4631 | * Find a master upper device and return pointer to it or NULL in case | |
4632 | * it's not there. The caller must hold the RTNL lock. | |
4633 | */ | |
4634 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
4635 | { | |
aa9d8560 | 4636 | struct netdev_adjacent *upper; |
9ff162a8 JP |
4637 | |
4638 | ASSERT_RTNL(); | |
4639 | ||
2f268f12 | 4640 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
4641 | return NULL; |
4642 | ||
2f268f12 | 4643 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 4644 | struct netdev_adjacent, list); |
9ff162a8 JP |
4645 | if (likely(upper->master)) |
4646 | return upper->dev; | |
4647 | return NULL; | |
4648 | } | |
4649 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
4650 | ||
b6ccba4c VF |
4651 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
4652 | { | |
4653 | struct netdev_adjacent *adj; | |
4654 | ||
4655 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
4656 | ||
4657 | return adj->private; | |
4658 | } | |
4659 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
4660 | ||
44a40855 VY |
4661 | /** |
4662 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
4663 | * @dev: device | |
4664 | * @iter: list_head ** of the current position | |
4665 | * | |
4666 | * Gets the next device from the dev's upper list, starting from iter | |
4667 | * position. The caller must hold RCU read lock. | |
4668 | */ | |
4669 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
4670 | struct list_head **iter) | |
4671 | { | |
4672 | struct netdev_adjacent *upper; | |
4673 | ||
4674 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
4675 | ||
4676 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4677 | ||
4678 | if (&upper->list == &dev->adj_list.upper) | |
4679 | return NULL; | |
4680 | ||
4681 | *iter = &upper->list; | |
4682 | ||
4683 | return upper->dev; | |
4684 | } | |
4685 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
4686 | ||
31088a11 VF |
4687 | /** |
4688 | * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list | |
48311f46 VF |
4689 | * @dev: device |
4690 | * @iter: list_head ** of the current position | |
4691 | * | |
4692 | * Gets the next device from the dev's upper list, starting from iter | |
4693 | * position. The caller must hold RCU read lock. | |
4694 | */ | |
2f268f12 VF |
4695 | struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, |
4696 | struct list_head **iter) | |
48311f46 VF |
4697 | { |
4698 | struct netdev_adjacent *upper; | |
4699 | ||
85328240 | 4700 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); |
48311f46 VF |
4701 | |
4702 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4703 | ||
2f268f12 | 4704 | if (&upper->list == &dev->all_adj_list.upper) |
48311f46 VF |
4705 | return NULL; |
4706 | ||
4707 | *iter = &upper->list; | |
4708 | ||
4709 | return upper->dev; | |
4710 | } | |
2f268f12 | 4711 | EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); |
48311f46 | 4712 | |
31088a11 VF |
4713 | /** |
4714 | * netdev_lower_get_next_private - Get the next ->private from the | |
4715 | * lower neighbour list | |
4716 | * @dev: device | |
4717 | * @iter: list_head ** of the current position | |
4718 | * | |
4719 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
4720 | * list, starting from iter position. The caller must hold either hold the | |
4721 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
4722 | * list will remain unchainged. | |
4723 | */ | |
4724 | void *netdev_lower_get_next_private(struct net_device *dev, | |
4725 | struct list_head **iter) | |
4726 | { | |
4727 | struct netdev_adjacent *lower; | |
4728 | ||
4729 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
4730 | ||
4731 | if (&lower->list == &dev->adj_list.lower) | |
4732 | return NULL; | |
4733 | ||
6859e7df | 4734 | *iter = lower->list.next; |
31088a11 VF |
4735 | |
4736 | return lower->private; | |
4737 | } | |
4738 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
4739 | ||
4740 | /** | |
4741 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
4742 | * lower neighbour list, RCU | |
4743 | * variant | |
4744 | * @dev: device | |
4745 | * @iter: list_head ** of the current position | |
4746 | * | |
4747 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
4748 | * list, starting from iter position. The caller must hold RCU read lock. | |
4749 | */ | |
4750 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
4751 | struct list_head **iter) | |
4752 | { | |
4753 | struct netdev_adjacent *lower; | |
4754 | ||
4755 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4756 | ||
4757 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4758 | ||
4759 | if (&lower->list == &dev->adj_list.lower) | |
4760 | return NULL; | |
4761 | ||
6859e7df | 4762 | *iter = &lower->list; |
31088a11 VF |
4763 | |
4764 | return lower->private; | |
4765 | } | |
4766 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
4767 | ||
4085ebe8 VY |
4768 | /** |
4769 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
4770 | * list | |
4771 | * @dev: device | |
4772 | * @iter: list_head ** of the current position | |
4773 | * | |
4774 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
4775 | * list, starting from iter position. The caller must hold RTNL lock or | |
4776 | * its own locking that guarantees that the neighbour lower | |
4777 | * list will remain unchainged. | |
4778 | */ | |
4779 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
4780 | { | |
4781 | struct netdev_adjacent *lower; | |
4782 | ||
4783 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
4784 | ||
4785 | if (&lower->list == &dev->adj_list.lower) | |
4786 | return NULL; | |
4787 | ||
4788 | *iter = &lower->list; | |
4789 | ||
4790 | return lower->dev; | |
4791 | } | |
4792 | EXPORT_SYMBOL(netdev_lower_get_next); | |
4793 | ||
e001bfad | 4794 | /** |
4795 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
4796 | * lower neighbour list, RCU | |
4797 | * variant | |
4798 | * @dev: device | |
4799 | * | |
4800 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
4801 | * list. The caller must hold RCU read lock. | |
4802 | */ | |
4803 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
4804 | { | |
4805 | struct netdev_adjacent *lower; | |
4806 | ||
4807 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
4808 | struct netdev_adjacent, list); | |
4809 | if (lower) | |
4810 | return lower->private; | |
4811 | return NULL; | |
4812 | } | |
4813 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
4814 | ||
9ff162a8 JP |
4815 | /** |
4816 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
4817 | * @dev: device | |
4818 | * | |
4819 | * Find a master upper device and return pointer to it or NULL in case | |
4820 | * it's not there. The caller must hold the RCU read lock. | |
4821 | */ | |
4822 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
4823 | { | |
aa9d8560 | 4824 | struct netdev_adjacent *upper; |
9ff162a8 | 4825 | |
2f268f12 | 4826 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 4827 | struct netdev_adjacent, list); |
9ff162a8 JP |
4828 | if (upper && likely(upper->master)) |
4829 | return upper->dev; | |
4830 | return NULL; | |
4831 | } | |
4832 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
4833 | ||
0a59f3a9 | 4834 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
4835 | struct net_device *adj_dev, |
4836 | struct list_head *dev_list) | |
4837 | { | |
4838 | char linkname[IFNAMSIZ+7]; | |
4839 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
4840 | "upper_%s" : "lower_%s", adj_dev->name); | |
4841 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
4842 | linkname); | |
4843 | } | |
0a59f3a9 | 4844 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
4845 | char *name, |
4846 | struct list_head *dev_list) | |
4847 | { | |
4848 | char linkname[IFNAMSIZ+7]; | |
4849 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
4850 | "upper_%s" : "lower_%s", name); | |
4851 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
4852 | } | |
4853 | ||
4854 | #define netdev_adjacent_is_neigh_list(dev, dev_list) \ | |
4855 | (dev_list == &dev->adj_list.upper || \ | |
4856 | dev_list == &dev->adj_list.lower) | |
4857 | ||
5d261913 VF |
4858 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
4859 | struct net_device *adj_dev, | |
7863c054 | 4860 | struct list_head *dev_list, |
402dae96 | 4861 | void *private, bool master) |
5d261913 VF |
4862 | { |
4863 | struct netdev_adjacent *adj; | |
842d67a7 | 4864 | int ret; |
5d261913 | 4865 | |
7863c054 | 4866 | adj = __netdev_find_adj(dev, adj_dev, dev_list); |
5d261913 VF |
4867 | |
4868 | if (adj) { | |
5d261913 VF |
4869 | adj->ref_nr++; |
4870 | return 0; | |
4871 | } | |
4872 | ||
4873 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
4874 | if (!adj) | |
4875 | return -ENOMEM; | |
4876 | ||
4877 | adj->dev = adj_dev; | |
4878 | adj->master = master; | |
5d261913 | 4879 | adj->ref_nr = 1; |
402dae96 | 4880 | adj->private = private; |
5d261913 | 4881 | dev_hold(adj_dev); |
2f268f12 VF |
4882 | |
4883 | pr_debug("dev_hold for %s, because of link added from %s to %s\n", | |
4884 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 | 4885 | |
3ee32707 VF |
4886 | if (netdev_adjacent_is_neigh_list(dev, dev_list)) { |
4887 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); | |
5831d66e VF |
4888 | if (ret) |
4889 | goto free_adj; | |
4890 | } | |
4891 | ||
7863c054 | 4892 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
4893 | if (master) { |
4894 | ret = sysfs_create_link(&(dev->dev.kobj), | |
4895 | &(adj_dev->dev.kobj), "master"); | |
4896 | if (ret) | |
5831d66e | 4897 | goto remove_symlinks; |
842d67a7 | 4898 | |
7863c054 | 4899 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 4900 | } else { |
7863c054 | 4901 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 4902 | } |
5d261913 VF |
4903 | |
4904 | return 0; | |
842d67a7 | 4905 | |
5831d66e | 4906 | remove_symlinks: |
3ee32707 VF |
4907 | if (netdev_adjacent_is_neigh_list(dev, dev_list)) |
4908 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); | |
842d67a7 VF |
4909 | free_adj: |
4910 | kfree(adj); | |
974daef7 | 4911 | dev_put(adj_dev); |
842d67a7 VF |
4912 | |
4913 | return ret; | |
5d261913 VF |
4914 | } |
4915 | ||
1d143d9f | 4916 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
4917 | struct net_device *adj_dev, | |
4918 | struct list_head *dev_list) | |
5d261913 VF |
4919 | { |
4920 | struct netdev_adjacent *adj; | |
4921 | ||
7863c054 | 4922 | adj = __netdev_find_adj(dev, adj_dev, dev_list); |
5d261913 | 4923 | |
2f268f12 VF |
4924 | if (!adj) { |
4925 | pr_err("tried to remove device %s from %s\n", | |
4926 | dev->name, adj_dev->name); | |
5d261913 | 4927 | BUG(); |
2f268f12 | 4928 | } |
5d261913 VF |
4929 | |
4930 | if (adj->ref_nr > 1) { | |
2f268f12 VF |
4931 | pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name, |
4932 | adj->ref_nr-1); | |
5d261913 VF |
4933 | adj->ref_nr--; |
4934 | return; | |
4935 | } | |
4936 | ||
842d67a7 VF |
4937 | if (adj->master) |
4938 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
4939 | ||
3ee32707 VF |
4940 | if (netdev_adjacent_is_neigh_list(dev, dev_list)) |
4941 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); | |
5831d66e | 4942 | |
5d261913 | 4943 | list_del_rcu(&adj->list); |
2f268f12 VF |
4944 | pr_debug("dev_put for %s, because link removed from %s to %s\n", |
4945 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 VF |
4946 | dev_put(adj_dev); |
4947 | kfree_rcu(adj, rcu); | |
4948 | } | |
4949 | ||
1d143d9f | 4950 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
4951 | struct net_device *upper_dev, | |
4952 | struct list_head *up_list, | |
4953 | struct list_head *down_list, | |
4954 | void *private, bool master) | |
5d261913 VF |
4955 | { |
4956 | int ret; | |
4957 | ||
402dae96 VF |
4958 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private, |
4959 | master); | |
5d261913 VF |
4960 | if (ret) |
4961 | return ret; | |
4962 | ||
402dae96 VF |
4963 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private, |
4964 | false); | |
5d261913 | 4965 | if (ret) { |
2f268f12 | 4966 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5d261913 VF |
4967 | return ret; |
4968 | } | |
4969 | ||
4970 | return 0; | |
4971 | } | |
4972 | ||
1d143d9f | 4973 | static int __netdev_adjacent_dev_link(struct net_device *dev, |
4974 | struct net_device *upper_dev) | |
5d261913 | 4975 | { |
2f268f12 VF |
4976 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
4977 | &dev->all_adj_list.upper, | |
4978 | &upper_dev->all_adj_list.lower, | |
402dae96 | 4979 | NULL, false); |
5d261913 VF |
4980 | } |
4981 | ||
1d143d9f | 4982 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
4983 | struct net_device *upper_dev, | |
4984 | struct list_head *up_list, | |
4985 | struct list_head *down_list) | |
5d261913 | 4986 | { |
2f268f12 VF |
4987 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
4988 | __netdev_adjacent_dev_remove(upper_dev, dev, down_list); | |
5d261913 VF |
4989 | } |
4990 | ||
1d143d9f | 4991 | static void __netdev_adjacent_dev_unlink(struct net_device *dev, |
4992 | struct net_device *upper_dev) | |
5d261913 | 4993 | { |
2f268f12 VF |
4994 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, |
4995 | &dev->all_adj_list.upper, | |
4996 | &upper_dev->all_adj_list.lower); | |
4997 | } | |
4998 | ||
1d143d9f | 4999 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
5000 | struct net_device *upper_dev, | |
5001 | void *private, bool master) | |
2f268f12 VF |
5002 | { |
5003 | int ret = __netdev_adjacent_dev_link(dev, upper_dev); | |
5004 | ||
5005 | if (ret) | |
5006 | return ret; | |
5007 | ||
5008 | ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, | |
5009 | &dev->adj_list.upper, | |
5010 | &upper_dev->adj_list.lower, | |
402dae96 | 5011 | private, master); |
2f268f12 VF |
5012 | if (ret) { |
5013 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5014 | return ret; | |
5015 | } | |
5016 | ||
5017 | return 0; | |
5d261913 VF |
5018 | } |
5019 | ||
1d143d9f | 5020 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
5021 | struct net_device *upper_dev) | |
2f268f12 VF |
5022 | { |
5023 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5024 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, | |
5025 | &dev->adj_list.upper, | |
5026 | &upper_dev->adj_list.lower); | |
5027 | } | |
5d261913 | 5028 | |
9ff162a8 | 5029 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 VF |
5030 | struct net_device *upper_dev, bool master, |
5031 | void *private) | |
9ff162a8 | 5032 | { |
5d261913 VF |
5033 | struct netdev_adjacent *i, *j, *to_i, *to_j; |
5034 | int ret = 0; | |
9ff162a8 JP |
5035 | |
5036 | ASSERT_RTNL(); | |
5037 | ||
5038 | if (dev == upper_dev) | |
5039 | return -EBUSY; | |
5040 | ||
5041 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
2f268f12 | 5042 | if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper)) |
9ff162a8 JP |
5043 | return -EBUSY; |
5044 | ||
2f268f12 | 5045 | if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper)) |
9ff162a8 JP |
5046 | return -EEXIST; |
5047 | ||
5048 | if (master && netdev_master_upper_dev_get(dev)) | |
5049 | return -EBUSY; | |
5050 | ||
402dae96 VF |
5051 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private, |
5052 | master); | |
5d261913 VF |
5053 | if (ret) |
5054 | return ret; | |
9ff162a8 | 5055 | |
5d261913 | 5056 | /* Now that we linked these devs, make all the upper_dev's |
2f268f12 | 5057 | * all_adj_list.upper visible to every dev's all_adj_list.lower an |
5d261913 VF |
5058 | * versa, and don't forget the devices itself. All of these |
5059 | * links are non-neighbours. | |
5060 | */ | |
2f268f12 VF |
5061 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5062 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5063 | pr_debug("Interlinking %s with %s, non-neighbour\n", | |
5064 | i->dev->name, j->dev->name); | |
5d261913 VF |
5065 | ret = __netdev_adjacent_dev_link(i->dev, j->dev); |
5066 | if (ret) | |
5067 | goto rollback_mesh; | |
5068 | } | |
5069 | } | |
5070 | ||
5071 | /* add dev to every upper_dev's upper device */ | |
2f268f12 VF |
5072 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5073 | pr_debug("linking %s's upper device %s with %s\n", | |
5074 | upper_dev->name, i->dev->name, dev->name); | |
5d261913 VF |
5075 | ret = __netdev_adjacent_dev_link(dev, i->dev); |
5076 | if (ret) | |
5077 | goto rollback_upper_mesh; | |
5078 | } | |
5079 | ||
5080 | /* add upper_dev to every dev's lower device */ | |
2f268f12 VF |
5081 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5082 | pr_debug("linking %s's lower device %s with %s\n", dev->name, | |
5083 | i->dev->name, upper_dev->name); | |
5d261913 VF |
5084 | ret = __netdev_adjacent_dev_link(i->dev, upper_dev); |
5085 | if (ret) | |
5086 | goto rollback_lower_mesh; | |
5087 | } | |
9ff162a8 | 5088 | |
42e52bf9 | 5089 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 | 5090 | return 0; |
5d261913 VF |
5091 | |
5092 | rollback_lower_mesh: | |
5093 | to_i = i; | |
2f268f12 | 5094 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5d261913 VF |
5095 | if (i == to_i) |
5096 | break; | |
5097 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); | |
5098 | } | |
5099 | ||
5100 | i = NULL; | |
5101 | ||
5102 | rollback_upper_mesh: | |
5103 | to_i = i; | |
2f268f12 | 5104 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5d261913 VF |
5105 | if (i == to_i) |
5106 | break; | |
5107 | __netdev_adjacent_dev_unlink(dev, i->dev); | |
5108 | } | |
5109 | ||
5110 | i = j = NULL; | |
5111 | ||
5112 | rollback_mesh: | |
5113 | to_i = i; | |
5114 | to_j = j; | |
2f268f12 VF |
5115 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5116 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5d261913 VF |
5117 | if (i == to_i && j == to_j) |
5118 | break; | |
5119 | __netdev_adjacent_dev_unlink(i->dev, j->dev); | |
5120 | } | |
5121 | if (i == to_i) | |
5122 | break; | |
5123 | } | |
5124 | ||
2f268f12 | 5125 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5126 | |
5127 | return ret; | |
9ff162a8 JP |
5128 | } |
5129 | ||
5130 | /** | |
5131 | * netdev_upper_dev_link - Add a link to the upper device | |
5132 | * @dev: device | |
5133 | * @upper_dev: new upper device | |
5134 | * | |
5135 | * Adds a link to device which is upper to this one. The caller must hold | |
5136 | * the RTNL lock. On a failure a negative errno code is returned. | |
5137 | * On success the reference counts are adjusted and the function | |
5138 | * returns zero. | |
5139 | */ | |
5140 | int netdev_upper_dev_link(struct net_device *dev, | |
5141 | struct net_device *upper_dev) | |
5142 | { | |
402dae96 | 5143 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL); |
9ff162a8 JP |
5144 | } |
5145 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
5146 | ||
5147 | /** | |
5148 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
5149 | * @dev: device | |
5150 | * @upper_dev: new upper device | |
5151 | * | |
5152 | * Adds a link to device which is upper to this one. In this case, only | |
5153 | * one master upper device can be linked, although other non-master devices | |
5154 | * might be linked as well. The caller must hold the RTNL lock. | |
5155 | * On a failure a negative errno code is returned. On success the reference | |
5156 | * counts are adjusted and the function returns zero. | |
5157 | */ | |
5158 | int netdev_master_upper_dev_link(struct net_device *dev, | |
5159 | struct net_device *upper_dev) | |
5160 | { | |
402dae96 | 5161 | return __netdev_upper_dev_link(dev, upper_dev, true, NULL); |
9ff162a8 JP |
5162 | } |
5163 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
5164 | ||
402dae96 VF |
5165 | int netdev_master_upper_dev_link_private(struct net_device *dev, |
5166 | struct net_device *upper_dev, | |
5167 | void *private) | |
5168 | { | |
5169 | return __netdev_upper_dev_link(dev, upper_dev, true, private); | |
5170 | } | |
5171 | EXPORT_SYMBOL(netdev_master_upper_dev_link_private); | |
5172 | ||
9ff162a8 JP |
5173 | /** |
5174 | * netdev_upper_dev_unlink - Removes a link to upper device | |
5175 | * @dev: device | |
5176 | * @upper_dev: new upper device | |
5177 | * | |
5178 | * Removes a link to device which is upper to this one. The caller must hold | |
5179 | * the RTNL lock. | |
5180 | */ | |
5181 | void netdev_upper_dev_unlink(struct net_device *dev, | |
5182 | struct net_device *upper_dev) | |
5183 | { | |
5d261913 | 5184 | struct netdev_adjacent *i, *j; |
9ff162a8 JP |
5185 | ASSERT_RTNL(); |
5186 | ||
2f268f12 | 5187 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5188 | |
5189 | /* Here is the tricky part. We must remove all dev's lower | |
5190 | * devices from all upper_dev's upper devices and vice | |
5191 | * versa, to maintain the graph relationship. | |
5192 | */ | |
2f268f12 VF |
5193 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5194 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) | |
5d261913 VF |
5195 | __netdev_adjacent_dev_unlink(i->dev, j->dev); |
5196 | ||
5197 | /* remove also the devices itself from lower/upper device | |
5198 | * list | |
5199 | */ | |
2f268f12 | 5200 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5d261913 VF |
5201 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); |
5202 | ||
2f268f12 | 5203 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) |
5d261913 VF |
5204 | __netdev_adjacent_dev_unlink(dev, i->dev); |
5205 | ||
42e52bf9 | 5206 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 JP |
5207 | } |
5208 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
5209 | ||
5bb025fa | 5210 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 5211 | { |
5bb025fa | 5212 | struct netdev_adjacent *iter; |
402dae96 | 5213 | |
5bb025fa VF |
5214 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
5215 | netdev_adjacent_sysfs_del(iter->dev, oldname, | |
5216 | &iter->dev->adj_list.lower); | |
5217 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5218 | &iter->dev->adj_list.lower); | |
5219 | } | |
402dae96 | 5220 | |
5bb025fa VF |
5221 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
5222 | netdev_adjacent_sysfs_del(iter->dev, oldname, | |
5223 | &iter->dev->adj_list.upper); | |
5224 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5225 | &iter->dev->adj_list.upper); | |
5226 | } | |
402dae96 | 5227 | } |
402dae96 VF |
5228 | |
5229 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
5230 | struct net_device *lower_dev) | |
5231 | { | |
5232 | struct netdev_adjacent *lower; | |
5233 | ||
5234 | if (!lower_dev) | |
5235 | return NULL; | |
5236 | lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower); | |
5237 | if (!lower) | |
5238 | return NULL; | |
5239 | ||
5240 | return lower->private; | |
5241 | } | |
5242 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
5243 | ||
4085ebe8 VY |
5244 | |
5245 | int dev_get_nest_level(struct net_device *dev, | |
5246 | bool (*type_check)(struct net_device *dev)) | |
5247 | { | |
5248 | struct net_device *lower = NULL; | |
5249 | struct list_head *iter; | |
5250 | int max_nest = -1; | |
5251 | int nest; | |
5252 | ||
5253 | ASSERT_RTNL(); | |
5254 | ||
5255 | netdev_for_each_lower_dev(dev, lower, iter) { | |
5256 | nest = dev_get_nest_level(lower, type_check); | |
5257 | if (max_nest < nest) | |
5258 | max_nest = nest; | |
5259 | } | |
5260 | ||
5261 | if (type_check(dev)) | |
5262 | max_nest++; | |
5263 | ||
5264 | return max_nest; | |
5265 | } | |
5266 | EXPORT_SYMBOL(dev_get_nest_level); | |
5267 | ||
b6c40d68 PM |
5268 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
5269 | { | |
d314774c SH |
5270 | const struct net_device_ops *ops = dev->netdev_ops; |
5271 | ||
d2615bf4 | 5272 | if (ops->ndo_change_rx_flags) |
d314774c | 5273 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
5274 | } |
5275 | ||
991fb3f7 | 5276 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5277 | { |
b536db93 | 5278 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
5279 | kuid_t uid; |
5280 | kgid_t gid; | |
1da177e4 | 5281 | |
24023451 PM |
5282 | ASSERT_RTNL(); |
5283 | ||
dad9b335 WC |
5284 | dev->flags |= IFF_PROMISC; |
5285 | dev->promiscuity += inc; | |
5286 | if (dev->promiscuity == 0) { | |
5287 | /* | |
5288 | * Avoid overflow. | |
5289 | * If inc causes overflow, untouch promisc and return error. | |
5290 | */ | |
5291 | if (inc < 0) | |
5292 | dev->flags &= ~IFF_PROMISC; | |
5293 | else { | |
5294 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
5295 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
5296 | dev->name); | |
dad9b335 WC |
5297 | return -EOVERFLOW; |
5298 | } | |
5299 | } | |
52609c0b | 5300 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
5301 | pr_info("device %s %s promiscuous mode\n", |
5302 | dev->name, | |
5303 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
5304 | if (audit_enabled) { |
5305 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
5306 | audit_log(current->audit_context, GFP_ATOMIC, |
5307 | AUDIT_ANOM_PROMISCUOUS, | |
5308 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
5309 | dev->name, (dev->flags & IFF_PROMISC), | |
5310 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 5311 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
5312 | from_kuid(&init_user_ns, uid), |
5313 | from_kgid(&init_user_ns, gid), | |
7759db82 | 5314 | audit_get_sessionid(current)); |
8192b0c4 | 5315 | } |
24023451 | 5316 | |
b6c40d68 | 5317 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 5318 | } |
991fb3f7 ND |
5319 | if (notify) |
5320 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 5321 | return 0; |
1da177e4 LT |
5322 | } |
5323 | ||
4417da66 PM |
5324 | /** |
5325 | * dev_set_promiscuity - update promiscuity count on a device | |
5326 | * @dev: device | |
5327 | * @inc: modifier | |
5328 | * | |
5329 | * Add or remove promiscuity from a device. While the count in the device | |
5330 | * remains above zero the interface remains promiscuous. Once it hits zero | |
5331 | * the device reverts back to normal filtering operation. A negative inc | |
5332 | * value is used to drop promiscuity on the device. | |
dad9b335 | 5333 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 5334 | */ |
dad9b335 | 5335 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 5336 | { |
b536db93 | 5337 | unsigned int old_flags = dev->flags; |
dad9b335 | 5338 | int err; |
4417da66 | 5339 | |
991fb3f7 | 5340 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 5341 | if (err < 0) |
dad9b335 | 5342 | return err; |
4417da66 PM |
5343 | if (dev->flags != old_flags) |
5344 | dev_set_rx_mode(dev); | |
dad9b335 | 5345 | return err; |
4417da66 | 5346 | } |
d1b19dff | 5347 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 5348 | |
991fb3f7 | 5349 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5350 | { |
991fb3f7 | 5351 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 5352 | |
24023451 PM |
5353 | ASSERT_RTNL(); |
5354 | ||
1da177e4 | 5355 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
5356 | dev->allmulti += inc; |
5357 | if (dev->allmulti == 0) { | |
5358 | /* | |
5359 | * Avoid overflow. | |
5360 | * If inc causes overflow, untouch allmulti and return error. | |
5361 | */ | |
5362 | if (inc < 0) | |
5363 | dev->flags &= ~IFF_ALLMULTI; | |
5364 | else { | |
5365 | dev->allmulti -= inc; | |
7b6cd1ce JP |
5366 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
5367 | dev->name); | |
dad9b335 WC |
5368 | return -EOVERFLOW; |
5369 | } | |
5370 | } | |
24023451 | 5371 | if (dev->flags ^ old_flags) { |
b6c40d68 | 5372 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 5373 | dev_set_rx_mode(dev); |
991fb3f7 ND |
5374 | if (notify) |
5375 | __dev_notify_flags(dev, old_flags, | |
5376 | dev->gflags ^ old_gflags); | |
24023451 | 5377 | } |
dad9b335 | 5378 | return 0; |
4417da66 | 5379 | } |
991fb3f7 ND |
5380 | |
5381 | /** | |
5382 | * dev_set_allmulti - update allmulti count on a device | |
5383 | * @dev: device | |
5384 | * @inc: modifier | |
5385 | * | |
5386 | * Add or remove reception of all multicast frames to a device. While the | |
5387 | * count in the device remains above zero the interface remains listening | |
5388 | * to all interfaces. Once it hits zero the device reverts back to normal | |
5389 | * filtering operation. A negative @inc value is used to drop the counter | |
5390 | * when releasing a resource needing all multicasts. | |
5391 | * Return 0 if successful or a negative errno code on error. | |
5392 | */ | |
5393 | ||
5394 | int dev_set_allmulti(struct net_device *dev, int inc) | |
5395 | { | |
5396 | return __dev_set_allmulti(dev, inc, true); | |
5397 | } | |
d1b19dff | 5398 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
5399 | |
5400 | /* | |
5401 | * Upload unicast and multicast address lists to device and | |
5402 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 5403 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
5404 | * are present. |
5405 | */ | |
5406 | void __dev_set_rx_mode(struct net_device *dev) | |
5407 | { | |
d314774c SH |
5408 | const struct net_device_ops *ops = dev->netdev_ops; |
5409 | ||
4417da66 PM |
5410 | /* dev_open will call this function so the list will stay sane. */ |
5411 | if (!(dev->flags&IFF_UP)) | |
5412 | return; | |
5413 | ||
5414 | if (!netif_device_present(dev)) | |
40b77c94 | 5415 | return; |
4417da66 | 5416 | |
01789349 | 5417 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
5418 | /* Unicast addresses changes may only happen under the rtnl, |
5419 | * therefore calling __dev_set_promiscuity here is safe. | |
5420 | */ | |
32e7bfc4 | 5421 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 5422 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 5423 | dev->uc_promisc = true; |
32e7bfc4 | 5424 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 5425 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 5426 | dev->uc_promisc = false; |
4417da66 | 5427 | } |
4417da66 | 5428 | } |
01789349 JP |
5429 | |
5430 | if (ops->ndo_set_rx_mode) | |
5431 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
5432 | } |
5433 | ||
5434 | void dev_set_rx_mode(struct net_device *dev) | |
5435 | { | |
b9e40857 | 5436 | netif_addr_lock_bh(dev); |
4417da66 | 5437 | __dev_set_rx_mode(dev); |
b9e40857 | 5438 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
5439 | } |
5440 | ||
f0db275a SH |
5441 | /** |
5442 | * dev_get_flags - get flags reported to userspace | |
5443 | * @dev: device | |
5444 | * | |
5445 | * Get the combination of flag bits exported through APIs to userspace. | |
5446 | */ | |
95c96174 | 5447 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 5448 | { |
95c96174 | 5449 | unsigned int flags; |
1da177e4 LT |
5450 | |
5451 | flags = (dev->flags & ~(IFF_PROMISC | | |
5452 | IFF_ALLMULTI | | |
b00055aa SR |
5453 | IFF_RUNNING | |
5454 | IFF_LOWER_UP | | |
5455 | IFF_DORMANT)) | | |
1da177e4 LT |
5456 | (dev->gflags & (IFF_PROMISC | |
5457 | IFF_ALLMULTI)); | |
5458 | ||
b00055aa SR |
5459 | if (netif_running(dev)) { |
5460 | if (netif_oper_up(dev)) | |
5461 | flags |= IFF_RUNNING; | |
5462 | if (netif_carrier_ok(dev)) | |
5463 | flags |= IFF_LOWER_UP; | |
5464 | if (netif_dormant(dev)) | |
5465 | flags |= IFF_DORMANT; | |
5466 | } | |
1da177e4 LT |
5467 | |
5468 | return flags; | |
5469 | } | |
d1b19dff | 5470 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 5471 | |
bd380811 | 5472 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 5473 | { |
b536db93 | 5474 | unsigned int old_flags = dev->flags; |
bd380811 | 5475 | int ret; |
1da177e4 | 5476 | |
24023451 PM |
5477 | ASSERT_RTNL(); |
5478 | ||
1da177e4 LT |
5479 | /* |
5480 | * Set the flags on our device. | |
5481 | */ | |
5482 | ||
5483 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
5484 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
5485 | IFF_AUTOMEDIA)) | | |
5486 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
5487 | IFF_ALLMULTI)); | |
5488 | ||
5489 | /* | |
5490 | * Load in the correct multicast list now the flags have changed. | |
5491 | */ | |
5492 | ||
b6c40d68 PM |
5493 | if ((old_flags ^ flags) & IFF_MULTICAST) |
5494 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 5495 | |
4417da66 | 5496 | dev_set_rx_mode(dev); |
1da177e4 LT |
5497 | |
5498 | /* | |
5499 | * Have we downed the interface. We handle IFF_UP ourselves | |
5500 | * according to user attempts to set it, rather than blindly | |
5501 | * setting it. | |
5502 | */ | |
5503 | ||
5504 | ret = 0; | |
d215d10f | 5505 | if ((old_flags ^ flags) & IFF_UP) |
bd380811 | 5506 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 | 5507 | |
1da177e4 | 5508 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 5509 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 5510 | unsigned int old_flags = dev->flags; |
d1b19dff | 5511 | |
1da177e4 | 5512 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
5513 | |
5514 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
5515 | if (dev->flags != old_flags) | |
5516 | dev_set_rx_mode(dev); | |
1da177e4 LT |
5517 | } |
5518 | ||
5519 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
5520 | is important. Some (broken) drivers set IFF_PROMISC, when | |
5521 | IFF_ALLMULTI is requested not asking us and not reporting. | |
5522 | */ | |
5523 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
5524 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
5525 | ||
1da177e4 | 5526 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 5527 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
5528 | } |
5529 | ||
bd380811 PM |
5530 | return ret; |
5531 | } | |
5532 | ||
a528c219 ND |
5533 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
5534 | unsigned int gchanges) | |
bd380811 PM |
5535 | { |
5536 | unsigned int changes = dev->flags ^ old_flags; | |
5537 | ||
a528c219 | 5538 | if (gchanges) |
7f294054 | 5539 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 5540 | |
bd380811 PM |
5541 | if (changes & IFF_UP) { |
5542 | if (dev->flags & IFF_UP) | |
5543 | call_netdevice_notifiers(NETDEV_UP, dev); | |
5544 | else | |
5545 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
5546 | } | |
5547 | ||
5548 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
5549 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
5550 | struct netdev_notifier_change_info change_info; | |
5551 | ||
5552 | change_info.flags_changed = changes; | |
5553 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
5554 | &change_info.info); | |
5555 | } | |
bd380811 PM |
5556 | } |
5557 | ||
5558 | /** | |
5559 | * dev_change_flags - change device settings | |
5560 | * @dev: device | |
5561 | * @flags: device state flags | |
5562 | * | |
5563 | * Change settings on device based state flags. The flags are | |
5564 | * in the userspace exported format. | |
5565 | */ | |
b536db93 | 5566 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 5567 | { |
b536db93 | 5568 | int ret; |
991fb3f7 | 5569 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
5570 | |
5571 | ret = __dev_change_flags(dev, flags); | |
5572 | if (ret < 0) | |
5573 | return ret; | |
5574 | ||
991fb3f7 | 5575 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 5576 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
5577 | return ret; |
5578 | } | |
d1b19dff | 5579 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 5580 | |
2315dc91 VF |
5581 | static int __dev_set_mtu(struct net_device *dev, int new_mtu) |
5582 | { | |
5583 | const struct net_device_ops *ops = dev->netdev_ops; | |
5584 | ||
5585 | if (ops->ndo_change_mtu) | |
5586 | return ops->ndo_change_mtu(dev, new_mtu); | |
5587 | ||
5588 | dev->mtu = new_mtu; | |
5589 | return 0; | |
5590 | } | |
5591 | ||
f0db275a SH |
5592 | /** |
5593 | * dev_set_mtu - Change maximum transfer unit | |
5594 | * @dev: device | |
5595 | * @new_mtu: new transfer unit | |
5596 | * | |
5597 | * Change the maximum transfer size of the network device. | |
5598 | */ | |
1da177e4 LT |
5599 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
5600 | { | |
2315dc91 | 5601 | int err, orig_mtu; |
1da177e4 LT |
5602 | |
5603 | if (new_mtu == dev->mtu) | |
5604 | return 0; | |
5605 | ||
5606 | /* MTU must be positive. */ | |
5607 | if (new_mtu < 0) | |
5608 | return -EINVAL; | |
5609 | ||
5610 | if (!netif_device_present(dev)) | |
5611 | return -ENODEV; | |
5612 | ||
1d486bfb VF |
5613 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
5614 | err = notifier_to_errno(err); | |
5615 | if (err) | |
5616 | return err; | |
d314774c | 5617 | |
2315dc91 VF |
5618 | orig_mtu = dev->mtu; |
5619 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 5620 | |
2315dc91 VF |
5621 | if (!err) { |
5622 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
5623 | err = notifier_to_errno(err); | |
5624 | if (err) { | |
5625 | /* setting mtu back and notifying everyone again, | |
5626 | * so that they have a chance to revert changes. | |
5627 | */ | |
5628 | __dev_set_mtu(dev, orig_mtu); | |
5629 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
5630 | } | |
5631 | } | |
1da177e4 LT |
5632 | return err; |
5633 | } | |
d1b19dff | 5634 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 5635 | |
cbda10fa VD |
5636 | /** |
5637 | * dev_set_group - Change group this device belongs to | |
5638 | * @dev: device | |
5639 | * @new_group: group this device should belong to | |
5640 | */ | |
5641 | void dev_set_group(struct net_device *dev, int new_group) | |
5642 | { | |
5643 | dev->group = new_group; | |
5644 | } | |
5645 | EXPORT_SYMBOL(dev_set_group); | |
5646 | ||
f0db275a SH |
5647 | /** |
5648 | * dev_set_mac_address - Change Media Access Control Address | |
5649 | * @dev: device | |
5650 | * @sa: new address | |
5651 | * | |
5652 | * Change the hardware (MAC) address of the device | |
5653 | */ | |
1da177e4 LT |
5654 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
5655 | { | |
d314774c | 5656 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
5657 | int err; |
5658 | ||
d314774c | 5659 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
5660 | return -EOPNOTSUPP; |
5661 | if (sa->sa_family != dev->type) | |
5662 | return -EINVAL; | |
5663 | if (!netif_device_present(dev)) | |
5664 | return -ENODEV; | |
d314774c | 5665 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
5666 | if (err) |
5667 | return err; | |
fbdeca2d | 5668 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 5669 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 5670 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 5671 | return 0; |
1da177e4 | 5672 | } |
d1b19dff | 5673 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 5674 | |
4bf84c35 JP |
5675 | /** |
5676 | * dev_change_carrier - Change device carrier | |
5677 | * @dev: device | |
691b3b7e | 5678 | * @new_carrier: new value |
4bf84c35 JP |
5679 | * |
5680 | * Change device carrier | |
5681 | */ | |
5682 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
5683 | { | |
5684 | const struct net_device_ops *ops = dev->netdev_ops; | |
5685 | ||
5686 | if (!ops->ndo_change_carrier) | |
5687 | return -EOPNOTSUPP; | |
5688 | if (!netif_device_present(dev)) | |
5689 | return -ENODEV; | |
5690 | return ops->ndo_change_carrier(dev, new_carrier); | |
5691 | } | |
5692 | EXPORT_SYMBOL(dev_change_carrier); | |
5693 | ||
66b52b0d JP |
5694 | /** |
5695 | * dev_get_phys_port_id - Get device physical port ID | |
5696 | * @dev: device | |
5697 | * @ppid: port ID | |
5698 | * | |
5699 | * Get device physical port ID | |
5700 | */ | |
5701 | int dev_get_phys_port_id(struct net_device *dev, | |
5702 | struct netdev_phys_port_id *ppid) | |
5703 | { | |
5704 | const struct net_device_ops *ops = dev->netdev_ops; | |
5705 | ||
5706 | if (!ops->ndo_get_phys_port_id) | |
5707 | return -EOPNOTSUPP; | |
5708 | return ops->ndo_get_phys_port_id(dev, ppid); | |
5709 | } | |
5710 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
5711 | ||
1da177e4 LT |
5712 | /** |
5713 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5714 | * @net: the applicable net namespace |
1da177e4 LT |
5715 | * |
5716 | * Returns a suitable unique value for a new device interface | |
5717 | * number. The caller must hold the rtnl semaphore or the | |
5718 | * dev_base_lock to be sure it remains unique. | |
5719 | */ | |
881d966b | 5720 | static int dev_new_index(struct net *net) |
1da177e4 | 5721 | { |
aa79e66e | 5722 | int ifindex = net->ifindex; |
1da177e4 LT |
5723 | for (;;) { |
5724 | if (++ifindex <= 0) | |
5725 | ifindex = 1; | |
881d966b | 5726 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 5727 | return net->ifindex = ifindex; |
1da177e4 LT |
5728 | } |
5729 | } | |
5730 | ||
1da177e4 | 5731 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5732 | static LIST_HEAD(net_todo_list); |
200b916f | 5733 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 5734 | |
6f05f629 | 5735 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5736 | { |
1da177e4 | 5737 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 5738 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
5739 | } |
5740 | ||
9b5e383c | 5741 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5742 | { |
e93737b0 | 5743 | struct net_device *dev, *tmp; |
5cde2829 | 5744 | LIST_HEAD(close_head); |
9b5e383c | 5745 | |
93ee31f1 DL |
5746 | BUG_ON(dev_boot_phase); |
5747 | ASSERT_RTNL(); | |
5748 | ||
e93737b0 | 5749 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5750 | /* Some devices call without registering |
e93737b0 KK |
5751 | * for initialization unwind. Remove those |
5752 | * devices and proceed with the remaining. | |
9b5e383c ED |
5753 | */ |
5754 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
5755 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
5756 | dev->name, dev); | |
93ee31f1 | 5757 | |
9b5e383c | 5758 | WARN_ON(1); |
e93737b0 KK |
5759 | list_del(&dev->unreg_list); |
5760 | continue; | |
9b5e383c | 5761 | } |
449f4544 | 5762 | dev->dismantle = true; |
9b5e383c | 5763 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5764 | } |
93ee31f1 | 5765 | |
44345724 | 5766 | /* If device is running, close it first. */ |
5cde2829 EB |
5767 | list_for_each_entry(dev, head, unreg_list) |
5768 | list_add_tail(&dev->close_list, &close_head); | |
5769 | dev_close_many(&close_head); | |
93ee31f1 | 5770 | |
44345724 | 5771 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5772 | /* And unlink it from device chain. */ |
5773 | unlist_netdevice(dev); | |
93ee31f1 | 5774 | |
9b5e383c ED |
5775 | dev->reg_state = NETREG_UNREGISTERING; |
5776 | } | |
93ee31f1 DL |
5777 | |
5778 | synchronize_net(); | |
5779 | ||
9b5e383c ED |
5780 | list_for_each_entry(dev, head, unreg_list) { |
5781 | /* Shutdown queueing discipline. */ | |
5782 | dev_shutdown(dev); | |
93ee31f1 DL |
5783 | |
5784 | ||
9b5e383c ED |
5785 | /* Notify protocols, that we are about to destroy |
5786 | this device. They should clean all the things. | |
5787 | */ | |
5788 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5789 | |
9b5e383c ED |
5790 | /* |
5791 | * Flush the unicast and multicast chains | |
5792 | */ | |
a748ee24 | 5793 | dev_uc_flush(dev); |
22bedad3 | 5794 | dev_mc_flush(dev); |
93ee31f1 | 5795 | |
9b5e383c ED |
5796 | if (dev->netdev_ops->ndo_uninit) |
5797 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5798 | |
56bfa7ee RP |
5799 | if (!dev->rtnl_link_ops || |
5800 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5801 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); | |
5802 | ||
9ff162a8 JP |
5803 | /* Notifier chain MUST detach us all upper devices. */ |
5804 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 5805 | |
9b5e383c ED |
5806 | /* Remove entries from kobject tree */ |
5807 | netdev_unregister_kobject(dev); | |
024e9679 AD |
5808 | #ifdef CONFIG_XPS |
5809 | /* Remove XPS queueing entries */ | |
5810 | netif_reset_xps_queues_gt(dev, 0); | |
5811 | #endif | |
9b5e383c | 5812 | } |
93ee31f1 | 5813 | |
850a545b | 5814 | synchronize_net(); |
395264d5 | 5815 | |
a5ee1551 | 5816 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5817 | dev_put(dev); |
5818 | } | |
5819 | ||
5820 | static void rollback_registered(struct net_device *dev) | |
5821 | { | |
5822 | LIST_HEAD(single); | |
5823 | ||
5824 | list_add(&dev->unreg_list, &single); | |
5825 | rollback_registered_many(&single); | |
ceaaec98 | 5826 | list_del(&single); |
93ee31f1 DL |
5827 | } |
5828 | ||
c8f44aff MM |
5829 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
5830 | netdev_features_t features) | |
b63365a2 | 5831 | { |
57422dc5 MM |
5832 | /* Fix illegal checksum combinations */ |
5833 | if ((features & NETIF_F_HW_CSUM) && | |
5834 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5835 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
5836 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
5837 | } | |
5838 | ||
b63365a2 | 5839 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 5840 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 5841 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 5842 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
5843 | } |
5844 | ||
ec5f0615 PS |
5845 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
5846 | !(features & NETIF_F_IP_CSUM)) { | |
5847 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
5848 | features &= ~NETIF_F_TSO; | |
5849 | features &= ~NETIF_F_TSO_ECN; | |
5850 | } | |
5851 | ||
5852 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
5853 | !(features & NETIF_F_IPV6_CSUM)) { | |
5854 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
5855 | features &= ~NETIF_F_TSO6; | |
5856 | } | |
5857 | ||
31d8b9e0 BH |
5858 | /* TSO ECN requires that TSO is present as well. */ |
5859 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
5860 | features &= ~NETIF_F_TSO_ECN; | |
5861 | ||
212b573f MM |
5862 | /* Software GSO depends on SG. */ |
5863 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 5864 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
5865 | features &= ~NETIF_F_GSO; |
5866 | } | |
5867 | ||
acd1130e | 5868 | /* UFO needs SG and checksumming */ |
b63365a2 | 5869 | if (features & NETIF_F_UFO) { |
79032644 MM |
5870 | /* maybe split UFO into V4 and V6? */ |
5871 | if (!((features & NETIF_F_GEN_CSUM) || | |
5872 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
5873 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5874 | netdev_dbg(dev, |
acd1130e | 5875 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
5876 | features &= ~NETIF_F_UFO; |
5877 | } | |
5878 | ||
5879 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 5880 | netdev_dbg(dev, |
acd1130e | 5881 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
5882 | features &= ~NETIF_F_UFO; |
5883 | } | |
5884 | } | |
5885 | ||
d0290214 JP |
5886 | #ifdef CONFIG_NET_RX_BUSY_POLL |
5887 | if (dev->netdev_ops->ndo_busy_poll) | |
5888 | features |= NETIF_F_BUSY_POLL; | |
5889 | else | |
5890 | #endif | |
5891 | features &= ~NETIF_F_BUSY_POLL; | |
5892 | ||
b63365a2 HX |
5893 | return features; |
5894 | } | |
b63365a2 | 5895 | |
6cb6a27c | 5896 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 5897 | { |
c8f44aff | 5898 | netdev_features_t features; |
5455c699 MM |
5899 | int err = 0; |
5900 | ||
87267485 MM |
5901 | ASSERT_RTNL(); |
5902 | ||
5455c699 MM |
5903 | features = netdev_get_wanted_features(dev); |
5904 | ||
5905 | if (dev->netdev_ops->ndo_fix_features) | |
5906 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
5907 | ||
5908 | /* driver might be less strict about feature dependencies */ | |
5909 | features = netdev_fix_features(dev, features); | |
5910 | ||
5911 | if (dev->features == features) | |
6cb6a27c | 5912 | return 0; |
5455c699 | 5913 | |
c8f44aff MM |
5914 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
5915 | &dev->features, &features); | |
5455c699 MM |
5916 | |
5917 | if (dev->netdev_ops->ndo_set_features) | |
5918 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5919 | ||
6cb6a27c | 5920 | if (unlikely(err < 0)) { |
5455c699 | 5921 | netdev_err(dev, |
c8f44aff MM |
5922 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
5923 | err, &features, &dev->features); | |
6cb6a27c MM |
5924 | return -1; |
5925 | } | |
5926 | ||
5927 | if (!err) | |
5928 | dev->features = features; | |
5929 | ||
5930 | return 1; | |
5931 | } | |
5932 | ||
afe12cc8 MM |
5933 | /** |
5934 | * netdev_update_features - recalculate device features | |
5935 | * @dev: the device to check | |
5936 | * | |
5937 | * Recalculate dev->features set and send notifications if it | |
5938 | * has changed. Should be called after driver or hardware dependent | |
5939 | * conditions might have changed that influence the features. | |
5940 | */ | |
6cb6a27c MM |
5941 | void netdev_update_features(struct net_device *dev) |
5942 | { | |
5943 | if (__netdev_update_features(dev)) | |
5944 | netdev_features_change(dev); | |
5455c699 MM |
5945 | } |
5946 | EXPORT_SYMBOL(netdev_update_features); | |
5947 | ||
afe12cc8 MM |
5948 | /** |
5949 | * netdev_change_features - recalculate device features | |
5950 | * @dev: the device to check | |
5951 | * | |
5952 | * Recalculate dev->features set and send notifications even | |
5953 | * if they have not changed. Should be called instead of | |
5954 | * netdev_update_features() if also dev->vlan_features might | |
5955 | * have changed to allow the changes to be propagated to stacked | |
5956 | * VLAN devices. | |
5957 | */ | |
5958 | void netdev_change_features(struct net_device *dev) | |
5959 | { | |
5960 | __netdev_update_features(dev); | |
5961 | netdev_features_change(dev); | |
5962 | } | |
5963 | EXPORT_SYMBOL(netdev_change_features); | |
5964 | ||
fc4a7489 PM |
5965 | /** |
5966 | * netif_stacked_transfer_operstate - transfer operstate | |
5967 | * @rootdev: the root or lower level device to transfer state from | |
5968 | * @dev: the device to transfer operstate to | |
5969 | * | |
5970 | * Transfer operational state from root to device. This is normally | |
5971 | * called when a stacking relationship exists between the root | |
5972 | * device and the device(a leaf device). | |
5973 | */ | |
5974 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5975 | struct net_device *dev) | |
5976 | { | |
5977 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5978 | netif_dormant_on(dev); | |
5979 | else | |
5980 | netif_dormant_off(dev); | |
5981 | ||
5982 | if (netif_carrier_ok(rootdev)) { | |
5983 | if (!netif_carrier_ok(dev)) | |
5984 | netif_carrier_on(dev); | |
5985 | } else { | |
5986 | if (netif_carrier_ok(dev)) | |
5987 | netif_carrier_off(dev); | |
5988 | } | |
5989 | } | |
5990 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5991 | ||
a953be53 | 5992 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
5993 | static int netif_alloc_rx_queues(struct net_device *dev) |
5994 | { | |
1b4bf461 | 5995 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 5996 | struct netdev_rx_queue *rx; |
1b4bf461 | 5997 | |
bd25fa7b | 5998 | BUG_ON(count < 1); |
1b4bf461 | 5999 | |
bd25fa7b | 6000 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
62b5942a | 6001 | if (!rx) |
bd25fa7b | 6002 | return -ENOMEM; |
62b5942a | 6003 | |
bd25fa7b TH |
6004 | dev->_rx = rx; |
6005 | ||
bd25fa7b | 6006 | for (i = 0; i < count; i++) |
fe822240 | 6007 | rx[i].dev = dev; |
1b4bf461 ED |
6008 | return 0; |
6009 | } | |
bf264145 | 6010 | #endif |
1b4bf461 | 6011 | |
aa942104 CG |
6012 | static void netdev_init_one_queue(struct net_device *dev, |
6013 | struct netdev_queue *queue, void *_unused) | |
6014 | { | |
6015 | /* Initialize queue lock */ | |
6016 | spin_lock_init(&queue->_xmit_lock); | |
6017 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
6018 | queue->xmit_lock_owner = -1; | |
b236da69 | 6019 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 6020 | queue->dev = dev; |
114cf580 TH |
6021 | #ifdef CONFIG_BQL |
6022 | dql_init(&queue->dql, HZ); | |
6023 | #endif | |
aa942104 CG |
6024 | } |
6025 | ||
60877a32 ED |
6026 | static void netif_free_tx_queues(struct net_device *dev) |
6027 | { | |
4cb28970 | 6028 | kvfree(dev->_tx); |
60877a32 ED |
6029 | } |
6030 | ||
e6484930 TH |
6031 | static int netif_alloc_netdev_queues(struct net_device *dev) |
6032 | { | |
6033 | unsigned int count = dev->num_tx_queues; | |
6034 | struct netdev_queue *tx; | |
60877a32 | 6035 | size_t sz = count * sizeof(*tx); |
e6484930 | 6036 | |
60877a32 | 6037 | BUG_ON(count < 1 || count > 0xffff); |
62b5942a | 6038 | |
60877a32 ED |
6039 | tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6040 | if (!tx) { | |
6041 | tx = vzalloc(sz); | |
6042 | if (!tx) | |
6043 | return -ENOMEM; | |
6044 | } | |
e6484930 | 6045 | dev->_tx = tx; |
1d24eb48 | 6046 | |
e6484930 TH |
6047 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
6048 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
6049 | |
6050 | return 0; | |
e6484930 TH |
6051 | } |
6052 | ||
1da177e4 LT |
6053 | /** |
6054 | * register_netdevice - register a network device | |
6055 | * @dev: device to register | |
6056 | * | |
6057 | * Take a completed network device structure and add it to the kernel | |
6058 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6059 | * chain. 0 is returned on success. A negative errno code is returned | |
6060 | * on a failure to set up the device, or if the name is a duplicate. | |
6061 | * | |
6062 | * Callers must hold the rtnl semaphore. You may want | |
6063 | * register_netdev() instead of this. | |
6064 | * | |
6065 | * BUGS: | |
6066 | * The locking appears insufficient to guarantee two parallel registers | |
6067 | * will not get the same name. | |
6068 | */ | |
6069 | ||
6070 | int register_netdevice(struct net_device *dev) | |
6071 | { | |
1da177e4 | 6072 | int ret; |
d314774c | 6073 | struct net *net = dev_net(dev); |
1da177e4 LT |
6074 | |
6075 | BUG_ON(dev_boot_phase); | |
6076 | ASSERT_RTNL(); | |
6077 | ||
b17a7c17 SH |
6078 | might_sleep(); |
6079 | ||
1da177e4 LT |
6080 | /* When net_device's are persistent, this will be fatal. */ |
6081 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 6082 | BUG_ON(!net); |
1da177e4 | 6083 | |
f1f28aa3 | 6084 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 6085 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 6086 | |
1da177e4 LT |
6087 | dev->iflink = -1; |
6088 | ||
828de4f6 | 6089 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
6090 | if (ret < 0) |
6091 | goto out; | |
6092 | ||
1da177e4 | 6093 | /* Init, if this function is available */ |
d314774c SH |
6094 | if (dev->netdev_ops->ndo_init) { |
6095 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
6096 | if (ret) { |
6097 | if (ret > 0) | |
6098 | ret = -EIO; | |
90833aa4 | 6099 | goto out; |
1da177e4 LT |
6100 | } |
6101 | } | |
4ec93edb | 6102 | |
f646968f PM |
6103 | if (((dev->hw_features | dev->features) & |
6104 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
6105 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
6106 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
6107 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
6108 | ret = -EINVAL; | |
6109 | goto err_uninit; | |
6110 | } | |
6111 | ||
9c7dafbf PE |
6112 | ret = -EBUSY; |
6113 | if (!dev->ifindex) | |
6114 | dev->ifindex = dev_new_index(net); | |
6115 | else if (__dev_get_by_index(net, dev->ifindex)) | |
6116 | goto err_uninit; | |
6117 | ||
1da177e4 LT |
6118 | if (dev->iflink == -1) |
6119 | dev->iflink = dev->ifindex; | |
6120 | ||
5455c699 MM |
6121 | /* Transfer changeable features to wanted_features and enable |
6122 | * software offloads (GSO and GRO). | |
6123 | */ | |
6124 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
6125 | dev->features |= NETIF_F_SOFT_FEATURES; |
6126 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 6127 | |
34324dc2 MM |
6128 | if (!(dev->flags & IFF_LOOPBACK)) { |
6129 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
c6e1a0d1 TH |
6130 | } |
6131 | ||
1180e7d6 | 6132 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 6133 | */ |
1180e7d6 | 6134 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 6135 | |
ee579677 PS |
6136 | /* Make NETIF_F_SG inheritable to tunnel devices. |
6137 | */ | |
6138 | dev->hw_enc_features |= NETIF_F_SG; | |
6139 | ||
0d89d203 SH |
6140 | /* Make NETIF_F_SG inheritable to MPLS. |
6141 | */ | |
6142 | dev->mpls_features |= NETIF_F_SG; | |
6143 | ||
7ffbe3fd JB |
6144 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
6145 | ret = notifier_to_errno(ret); | |
6146 | if (ret) | |
6147 | goto err_uninit; | |
6148 | ||
8b41d188 | 6149 | ret = netdev_register_kobject(dev); |
b17a7c17 | 6150 | if (ret) |
7ce1b0ed | 6151 | goto err_uninit; |
b17a7c17 SH |
6152 | dev->reg_state = NETREG_REGISTERED; |
6153 | ||
6cb6a27c | 6154 | __netdev_update_features(dev); |
8e9b59b2 | 6155 | |
1da177e4 LT |
6156 | /* |
6157 | * Default initial state at registry is that the | |
6158 | * device is present. | |
6159 | */ | |
6160 | ||
6161 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6162 | ||
8f4cccbb BH |
6163 | linkwatch_init_dev(dev); |
6164 | ||
1da177e4 | 6165 | dev_init_scheduler(dev); |
1da177e4 | 6166 | dev_hold(dev); |
ce286d32 | 6167 | list_netdevice(dev); |
7bf23575 | 6168 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 6169 | |
948b337e JP |
6170 | /* If the device has permanent device address, driver should |
6171 | * set dev_addr and also addr_assign_type should be set to | |
6172 | * NET_ADDR_PERM (default value). | |
6173 | */ | |
6174 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
6175 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
6176 | ||
1da177e4 | 6177 | /* Notify protocols, that a new device appeared. */ |
056925ab | 6178 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 6179 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
6180 | if (ret) { |
6181 | rollback_registered(dev); | |
6182 | dev->reg_state = NETREG_UNREGISTERED; | |
6183 | } | |
d90a909e EB |
6184 | /* |
6185 | * Prevent userspace races by waiting until the network | |
6186 | * device is fully setup before sending notifications. | |
6187 | */ | |
a2835763 PM |
6188 | if (!dev->rtnl_link_ops || |
6189 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 6190 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
6191 | |
6192 | out: | |
6193 | return ret; | |
7ce1b0ed HX |
6194 | |
6195 | err_uninit: | |
d314774c SH |
6196 | if (dev->netdev_ops->ndo_uninit) |
6197 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 6198 | goto out; |
1da177e4 | 6199 | } |
d1b19dff | 6200 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 6201 | |
937f1ba5 BH |
6202 | /** |
6203 | * init_dummy_netdev - init a dummy network device for NAPI | |
6204 | * @dev: device to init | |
6205 | * | |
6206 | * This takes a network device structure and initialize the minimum | |
6207 | * amount of fields so it can be used to schedule NAPI polls without | |
6208 | * registering a full blown interface. This is to be used by drivers | |
6209 | * that need to tie several hardware interfaces to a single NAPI | |
6210 | * poll scheduler due to HW limitations. | |
6211 | */ | |
6212 | int init_dummy_netdev(struct net_device *dev) | |
6213 | { | |
6214 | /* Clear everything. Note we don't initialize spinlocks | |
6215 | * are they aren't supposed to be taken by any of the | |
6216 | * NAPI code and this dummy netdev is supposed to be | |
6217 | * only ever used for NAPI polls | |
6218 | */ | |
6219 | memset(dev, 0, sizeof(struct net_device)); | |
6220 | ||
6221 | /* make sure we BUG if trying to hit standard | |
6222 | * register/unregister code path | |
6223 | */ | |
6224 | dev->reg_state = NETREG_DUMMY; | |
6225 | ||
937f1ba5 BH |
6226 | /* NAPI wants this */ |
6227 | INIT_LIST_HEAD(&dev->napi_list); | |
6228 | ||
6229 | /* a dummy interface is started by default */ | |
6230 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6231 | set_bit(__LINK_STATE_START, &dev->state); | |
6232 | ||
29b4433d ED |
6233 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
6234 | * because users of this 'device' dont need to change | |
6235 | * its refcount. | |
6236 | */ | |
6237 | ||
937f1ba5 BH |
6238 | return 0; |
6239 | } | |
6240 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
6241 | ||
6242 | ||
1da177e4 LT |
6243 | /** |
6244 | * register_netdev - register a network device | |
6245 | * @dev: device to register | |
6246 | * | |
6247 | * Take a completed network device structure and add it to the kernel | |
6248 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6249 | * chain. 0 is returned on success. A negative errno code is returned | |
6250 | * on a failure to set up the device, or if the name is a duplicate. | |
6251 | * | |
38b4da38 | 6252 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
6253 | * and expands the device name if you passed a format string to |
6254 | * alloc_netdev. | |
6255 | */ | |
6256 | int register_netdev(struct net_device *dev) | |
6257 | { | |
6258 | int err; | |
6259 | ||
6260 | rtnl_lock(); | |
1da177e4 | 6261 | err = register_netdevice(dev); |
1da177e4 LT |
6262 | rtnl_unlock(); |
6263 | return err; | |
6264 | } | |
6265 | EXPORT_SYMBOL(register_netdev); | |
6266 | ||
29b4433d ED |
6267 | int netdev_refcnt_read(const struct net_device *dev) |
6268 | { | |
6269 | int i, refcnt = 0; | |
6270 | ||
6271 | for_each_possible_cpu(i) | |
6272 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
6273 | return refcnt; | |
6274 | } | |
6275 | EXPORT_SYMBOL(netdev_refcnt_read); | |
6276 | ||
2c53040f | 6277 | /** |
1da177e4 | 6278 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 6279 | * @dev: target net_device |
1da177e4 LT |
6280 | * |
6281 | * This is called when unregistering network devices. | |
6282 | * | |
6283 | * Any protocol or device that holds a reference should register | |
6284 | * for netdevice notification, and cleanup and put back the | |
6285 | * reference if they receive an UNREGISTER event. | |
6286 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 6287 | * call dev_put. |
1da177e4 LT |
6288 | */ |
6289 | static void netdev_wait_allrefs(struct net_device *dev) | |
6290 | { | |
6291 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 6292 | int refcnt; |
1da177e4 | 6293 | |
e014debe ED |
6294 | linkwatch_forget_dev(dev); |
6295 | ||
1da177e4 | 6296 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
6297 | refcnt = netdev_refcnt_read(dev); |
6298 | ||
6299 | while (refcnt != 0) { | |
1da177e4 | 6300 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 6301 | rtnl_lock(); |
1da177e4 LT |
6302 | |
6303 | /* Rebroadcast unregister notification */ | |
056925ab | 6304 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 6305 | |
748e2d93 | 6306 | __rtnl_unlock(); |
0115e8e3 | 6307 | rcu_barrier(); |
748e2d93 ED |
6308 | rtnl_lock(); |
6309 | ||
0115e8e3 | 6310 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
6311 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
6312 | &dev->state)) { | |
6313 | /* We must not have linkwatch events | |
6314 | * pending on unregister. If this | |
6315 | * happens, we simply run the queue | |
6316 | * unscheduled, resulting in a noop | |
6317 | * for this device. | |
6318 | */ | |
6319 | linkwatch_run_queue(); | |
6320 | } | |
6321 | ||
6756ae4b | 6322 | __rtnl_unlock(); |
1da177e4 LT |
6323 | |
6324 | rebroadcast_time = jiffies; | |
6325 | } | |
6326 | ||
6327 | msleep(250); | |
6328 | ||
29b4433d ED |
6329 | refcnt = netdev_refcnt_read(dev); |
6330 | ||
1da177e4 | 6331 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
6332 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
6333 | dev->name, refcnt); | |
1da177e4 LT |
6334 | warning_time = jiffies; |
6335 | } | |
6336 | } | |
6337 | } | |
6338 | ||
6339 | /* The sequence is: | |
6340 | * | |
6341 | * rtnl_lock(); | |
6342 | * ... | |
6343 | * register_netdevice(x1); | |
6344 | * register_netdevice(x2); | |
6345 | * ... | |
6346 | * unregister_netdevice(y1); | |
6347 | * unregister_netdevice(y2); | |
6348 | * ... | |
6349 | * rtnl_unlock(); | |
6350 | * free_netdev(y1); | |
6351 | * free_netdev(y2); | |
6352 | * | |
58ec3b4d | 6353 | * We are invoked by rtnl_unlock(). |
1da177e4 | 6354 | * This allows us to deal with problems: |
b17a7c17 | 6355 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
6356 | * without deadlocking with linkwatch via keventd. |
6357 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
6358 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
6359 | * |
6360 | * We must not return until all unregister events added during | |
6361 | * the interval the lock was held have been completed. | |
1da177e4 | 6362 | */ |
1da177e4 LT |
6363 | void netdev_run_todo(void) |
6364 | { | |
626ab0e6 | 6365 | struct list_head list; |
1da177e4 | 6366 | |
1da177e4 | 6367 | /* Snapshot list, allow later requests */ |
626ab0e6 | 6368 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
6369 | |
6370 | __rtnl_unlock(); | |
626ab0e6 | 6371 | |
0115e8e3 ED |
6372 | |
6373 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
6374 | if (!list_empty(&list)) |
6375 | rcu_barrier(); | |
6376 | ||
1da177e4 LT |
6377 | while (!list_empty(&list)) { |
6378 | struct net_device *dev | |
e5e26d75 | 6379 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
6380 | list_del(&dev->todo_list); |
6381 | ||
748e2d93 | 6382 | rtnl_lock(); |
0115e8e3 | 6383 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 6384 | __rtnl_unlock(); |
0115e8e3 | 6385 | |
b17a7c17 | 6386 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 6387 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
6388 | dev->name, dev->reg_state); |
6389 | dump_stack(); | |
6390 | continue; | |
6391 | } | |
1da177e4 | 6392 | |
b17a7c17 | 6393 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 6394 | |
152102c7 | 6395 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 6396 | |
b17a7c17 | 6397 | netdev_wait_allrefs(dev); |
1da177e4 | 6398 | |
b17a7c17 | 6399 | /* paranoia */ |
29b4433d | 6400 | BUG_ON(netdev_refcnt_read(dev)); |
33d480ce ED |
6401 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
6402 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 6403 | WARN_ON(dev->dn_ptr); |
1da177e4 | 6404 | |
b17a7c17 SH |
6405 | if (dev->destructor) |
6406 | dev->destructor(dev); | |
9093bbb2 | 6407 | |
50624c93 EB |
6408 | /* Report a network device has been unregistered */ |
6409 | rtnl_lock(); | |
6410 | dev_net(dev)->dev_unreg_count--; | |
6411 | __rtnl_unlock(); | |
6412 | wake_up(&netdev_unregistering_wq); | |
6413 | ||
9093bbb2 SH |
6414 | /* Free network device */ |
6415 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 6416 | } |
1da177e4 LT |
6417 | } |
6418 | ||
3cfde79c BH |
6419 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
6420 | * fields in the same order, with only the type differing. | |
6421 | */ | |
77a1abf5 ED |
6422 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
6423 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
6424 | { |
6425 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
6426 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
6427 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
6428 | #else |
6429 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
6430 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
6431 | u64 *dst = (u64 *)stats64; | |
6432 | ||
6433 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
6434 | sizeof(*stats64) / sizeof(u64)); | |
6435 | for (i = 0; i < n; i++) | |
6436 | dst[i] = src[i]; | |
6437 | #endif | |
6438 | } | |
77a1abf5 | 6439 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 6440 | |
eeda3fd6 SH |
6441 | /** |
6442 | * dev_get_stats - get network device statistics | |
6443 | * @dev: device to get statistics from | |
28172739 | 6444 | * @storage: place to store stats |
eeda3fd6 | 6445 | * |
d7753516 BH |
6446 | * Get network statistics from device. Return @storage. |
6447 | * The device driver may provide its own method by setting | |
6448 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
6449 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 6450 | */ |
d7753516 BH |
6451 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
6452 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 6453 | { |
eeda3fd6 SH |
6454 | const struct net_device_ops *ops = dev->netdev_ops; |
6455 | ||
28172739 ED |
6456 | if (ops->ndo_get_stats64) { |
6457 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
6458 | ops->ndo_get_stats64(dev, storage); |
6459 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 6460 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
6461 | } else { |
6462 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 6463 | } |
caf586e5 | 6464 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
015f0688 | 6465 | storage->tx_dropped += atomic_long_read(&dev->tx_dropped); |
28172739 | 6466 | return storage; |
c45d286e | 6467 | } |
eeda3fd6 | 6468 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 6469 | |
24824a09 | 6470 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 6471 | { |
24824a09 | 6472 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 6473 | |
24824a09 ED |
6474 | #ifdef CONFIG_NET_CLS_ACT |
6475 | if (queue) | |
6476 | return queue; | |
6477 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
6478 | if (!queue) | |
6479 | return NULL; | |
6480 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
6481 | queue->qdisc = &noop_qdisc; |
6482 | queue->qdisc_sleeping = &noop_qdisc; | |
6483 | rcu_assign_pointer(dev->ingress_queue, queue); | |
6484 | #endif | |
6485 | return queue; | |
bb949fbd DM |
6486 | } |
6487 | ||
2c60db03 ED |
6488 | static const struct ethtool_ops default_ethtool_ops; |
6489 | ||
d07d7507 SG |
6490 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
6491 | const struct ethtool_ops *ops) | |
6492 | { | |
6493 | if (dev->ethtool_ops == &default_ethtool_ops) | |
6494 | dev->ethtool_ops = ops; | |
6495 | } | |
6496 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
6497 | ||
74d332c1 ED |
6498 | void netdev_freemem(struct net_device *dev) |
6499 | { | |
6500 | char *addr = (char *)dev - dev->padded; | |
6501 | ||
4cb28970 | 6502 | kvfree(addr); |
74d332c1 ED |
6503 | } |
6504 | ||
1da177e4 | 6505 | /** |
36909ea4 | 6506 | * alloc_netdev_mqs - allocate network device |
c835a677 TG |
6507 | * @sizeof_priv: size of private data to allocate space for |
6508 | * @name: device name format string | |
6509 | * @name_assign_type: origin of device name | |
6510 | * @setup: callback to initialize device | |
6511 | * @txqs: the number of TX subqueues to allocate | |
6512 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
6513 | * |
6514 | * Allocates a struct net_device with private data area for driver use | |
90e51adf | 6515 | * and performs basic initialization. Also allocates subqueue structs |
36909ea4 | 6516 | * for each queue on the device. |
1da177e4 | 6517 | */ |
36909ea4 | 6518 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 6519 | unsigned char name_assign_type, |
36909ea4 TH |
6520 | void (*setup)(struct net_device *), |
6521 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 6522 | { |
1da177e4 | 6523 | struct net_device *dev; |
7943986c | 6524 | size_t alloc_size; |
1ce8e7b5 | 6525 | struct net_device *p; |
1da177e4 | 6526 | |
b6fe17d6 SH |
6527 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
6528 | ||
36909ea4 | 6529 | if (txqs < 1) { |
7b6cd1ce | 6530 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
6531 | return NULL; |
6532 | } | |
6533 | ||
a953be53 | 6534 | #ifdef CONFIG_SYSFS |
36909ea4 | 6535 | if (rxqs < 1) { |
7b6cd1ce | 6536 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
6537 | return NULL; |
6538 | } | |
6539 | #endif | |
6540 | ||
fd2ea0a7 | 6541 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
6542 | if (sizeof_priv) { |
6543 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 6544 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
6545 | alloc_size += sizeof_priv; |
6546 | } | |
6547 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 6548 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 6549 | |
74d332c1 ED |
6550 | p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6551 | if (!p) | |
6552 | p = vzalloc(alloc_size); | |
62b5942a | 6553 | if (!p) |
1da177e4 | 6554 | return NULL; |
1da177e4 | 6555 | |
1ce8e7b5 | 6556 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 6557 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 6558 | |
29b4433d ED |
6559 | dev->pcpu_refcnt = alloc_percpu(int); |
6560 | if (!dev->pcpu_refcnt) | |
74d332c1 | 6561 | goto free_dev; |
ab9c73cc | 6562 | |
ab9c73cc | 6563 | if (dev_addr_init(dev)) |
29b4433d | 6564 | goto free_pcpu; |
ab9c73cc | 6565 | |
22bedad3 | 6566 | dev_mc_init(dev); |
a748ee24 | 6567 | dev_uc_init(dev); |
ccffad25 | 6568 | |
c346dca1 | 6569 | dev_net_set(dev, &init_net); |
1da177e4 | 6570 | |
8d3bdbd5 | 6571 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 6572 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 6573 | |
8d3bdbd5 DM |
6574 | INIT_LIST_HEAD(&dev->napi_list); |
6575 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 6576 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 6577 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
6578 | INIT_LIST_HEAD(&dev->adj_list.upper); |
6579 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
6580 | INIT_LIST_HEAD(&dev->all_adj_list.upper); | |
6581 | INIT_LIST_HEAD(&dev->all_adj_list.lower); | |
8d3bdbd5 DM |
6582 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
6583 | setup(dev); | |
6584 | ||
36909ea4 TH |
6585 | dev->num_tx_queues = txqs; |
6586 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 6587 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 6588 | goto free_all; |
e8a0464c | 6589 | |
a953be53 | 6590 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
6591 | dev->num_rx_queues = rxqs; |
6592 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 6593 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 6594 | goto free_all; |
df334545 | 6595 | #endif |
0a9627f2 | 6596 | |
1da177e4 | 6597 | strcpy(dev->name, name); |
c835a677 | 6598 | dev->name_assign_type = name_assign_type; |
cbda10fa | 6599 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
6600 | if (!dev->ethtool_ops) |
6601 | dev->ethtool_ops = &default_ethtool_ops; | |
1da177e4 | 6602 | return dev; |
ab9c73cc | 6603 | |
8d3bdbd5 DM |
6604 | free_all: |
6605 | free_netdev(dev); | |
6606 | return NULL; | |
6607 | ||
29b4433d ED |
6608 | free_pcpu: |
6609 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
6610 | free_dev: |
6611 | netdev_freemem(dev); | |
ab9c73cc | 6612 | return NULL; |
1da177e4 | 6613 | } |
36909ea4 | 6614 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
6615 | |
6616 | /** | |
6617 | * free_netdev - free network device | |
6618 | * @dev: device | |
6619 | * | |
4ec93edb YH |
6620 | * This function does the last stage of destroying an allocated device |
6621 | * interface. The reference to the device object is released. | |
1da177e4 LT |
6622 | * If this is the last reference then it will be freed. |
6623 | */ | |
6624 | void free_netdev(struct net_device *dev) | |
6625 | { | |
d565b0a1 HX |
6626 | struct napi_struct *p, *n; |
6627 | ||
f3005d7f DL |
6628 | release_net(dev_net(dev)); |
6629 | ||
60877a32 | 6630 | netif_free_tx_queues(dev); |
a953be53 | 6631 | #ifdef CONFIG_SYSFS |
fe822240 TH |
6632 | kfree(dev->_rx); |
6633 | #endif | |
e8a0464c | 6634 | |
33d480ce | 6635 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 6636 | |
f001fde5 JP |
6637 | /* Flush device addresses */ |
6638 | dev_addr_flush(dev); | |
6639 | ||
d565b0a1 HX |
6640 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
6641 | netif_napi_del(p); | |
6642 | ||
29b4433d ED |
6643 | free_percpu(dev->pcpu_refcnt); |
6644 | dev->pcpu_refcnt = NULL; | |
6645 | ||
3041a069 | 6646 | /* Compatibility with error handling in drivers */ |
1da177e4 | 6647 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 6648 | netdev_freemem(dev); |
1da177e4 LT |
6649 | return; |
6650 | } | |
6651 | ||
6652 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
6653 | dev->reg_state = NETREG_RELEASED; | |
6654 | ||
43cb76d9 GKH |
6655 | /* will free via device release */ |
6656 | put_device(&dev->dev); | |
1da177e4 | 6657 | } |
d1b19dff | 6658 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 6659 | |
f0db275a SH |
6660 | /** |
6661 | * synchronize_net - Synchronize with packet receive processing | |
6662 | * | |
6663 | * Wait for packets currently being received to be done. | |
6664 | * Does not block later packets from starting. | |
6665 | */ | |
4ec93edb | 6666 | void synchronize_net(void) |
1da177e4 LT |
6667 | { |
6668 | might_sleep(); | |
be3fc413 ED |
6669 | if (rtnl_is_locked()) |
6670 | synchronize_rcu_expedited(); | |
6671 | else | |
6672 | synchronize_rcu(); | |
1da177e4 | 6673 | } |
d1b19dff | 6674 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
6675 | |
6676 | /** | |
44a0873d | 6677 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 6678 | * @dev: device |
44a0873d | 6679 | * @head: list |
6ebfbc06 | 6680 | * |
1da177e4 | 6681 | * This function shuts down a device interface and removes it |
d59b54b1 | 6682 | * from the kernel tables. |
44a0873d | 6683 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
6684 | * |
6685 | * Callers must hold the rtnl semaphore. You may want | |
6686 | * unregister_netdev() instead of this. | |
6687 | */ | |
6688 | ||
44a0873d | 6689 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 6690 | { |
a6620712 HX |
6691 | ASSERT_RTNL(); |
6692 | ||
44a0873d | 6693 | if (head) { |
9fdce099 | 6694 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
6695 | } else { |
6696 | rollback_registered(dev); | |
6697 | /* Finish processing unregister after unlock */ | |
6698 | net_set_todo(dev); | |
6699 | } | |
1da177e4 | 6700 | } |
44a0873d | 6701 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 6702 | |
9b5e383c ED |
6703 | /** |
6704 | * unregister_netdevice_many - unregister many devices | |
6705 | * @head: list of devices | |
87757a91 ED |
6706 | * |
6707 | * Note: As most callers use a stack allocated list_head, | |
6708 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
6709 | */ |
6710 | void unregister_netdevice_many(struct list_head *head) | |
6711 | { | |
6712 | struct net_device *dev; | |
6713 | ||
6714 | if (!list_empty(head)) { | |
6715 | rollback_registered_many(head); | |
6716 | list_for_each_entry(dev, head, unreg_list) | |
6717 | net_set_todo(dev); | |
87757a91 | 6718 | list_del(head); |
9b5e383c ED |
6719 | } |
6720 | } | |
63c8099d | 6721 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 6722 | |
1da177e4 LT |
6723 | /** |
6724 | * unregister_netdev - remove device from the kernel | |
6725 | * @dev: device | |
6726 | * | |
6727 | * This function shuts down a device interface and removes it | |
d59b54b1 | 6728 | * from the kernel tables. |
1da177e4 LT |
6729 | * |
6730 | * This is just a wrapper for unregister_netdevice that takes | |
6731 | * the rtnl semaphore. In general you want to use this and not | |
6732 | * unregister_netdevice. | |
6733 | */ | |
6734 | void unregister_netdev(struct net_device *dev) | |
6735 | { | |
6736 | rtnl_lock(); | |
6737 | unregister_netdevice(dev); | |
6738 | rtnl_unlock(); | |
6739 | } | |
1da177e4 LT |
6740 | EXPORT_SYMBOL(unregister_netdev); |
6741 | ||
ce286d32 EB |
6742 | /** |
6743 | * dev_change_net_namespace - move device to different nethost namespace | |
6744 | * @dev: device | |
6745 | * @net: network namespace | |
6746 | * @pat: If not NULL name pattern to try if the current device name | |
6747 | * is already taken in the destination network namespace. | |
6748 | * | |
6749 | * This function shuts down a device interface and moves it | |
6750 | * to a new network namespace. On success 0 is returned, on | |
6751 | * a failure a netagive errno code is returned. | |
6752 | * | |
6753 | * Callers must hold the rtnl semaphore. | |
6754 | */ | |
6755 | ||
6756 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
6757 | { | |
ce286d32 EB |
6758 | int err; |
6759 | ||
6760 | ASSERT_RTNL(); | |
6761 | ||
6762 | /* Don't allow namespace local devices to be moved. */ | |
6763 | err = -EINVAL; | |
6764 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6765 | goto out; | |
6766 | ||
6767 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
6768 | if (dev->reg_state != NETREG_REGISTERED) |
6769 | goto out; | |
6770 | ||
6771 | /* Get out if there is nothing todo */ | |
6772 | err = 0; | |
878628fb | 6773 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
6774 | goto out; |
6775 | ||
6776 | /* Pick the destination device name, and ensure | |
6777 | * we can use it in the destination network namespace. | |
6778 | */ | |
6779 | err = -EEXIST; | |
d9031024 | 6780 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
6781 | /* We get here if we can't use the current device name */ |
6782 | if (!pat) | |
6783 | goto out; | |
828de4f6 | 6784 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
6785 | goto out; |
6786 | } | |
6787 | ||
6788 | /* | |
6789 | * And now a mini version of register_netdevice unregister_netdevice. | |
6790 | */ | |
6791 | ||
6792 | /* If device is running close it first. */ | |
9b772652 | 6793 | dev_close(dev); |
ce286d32 EB |
6794 | |
6795 | /* And unlink it from device chain */ | |
6796 | err = -ENODEV; | |
6797 | unlist_netdevice(dev); | |
6798 | ||
6799 | synchronize_net(); | |
6800 | ||
6801 | /* Shutdown queueing discipline. */ | |
6802 | dev_shutdown(dev); | |
6803 | ||
6804 | /* Notify protocols, that we are about to destroy | |
6805 | this device. They should clean all the things. | |
3b27e105 DL |
6806 | |
6807 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
6808 | This is wanted because this way 8021q and macvlan know | |
6809 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
6810 | */ |
6811 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
6812 | rcu_barrier(); |
6813 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 6814 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
6815 | |
6816 | /* | |
6817 | * Flush the unicast and multicast chains | |
6818 | */ | |
a748ee24 | 6819 | dev_uc_flush(dev); |
22bedad3 | 6820 | dev_mc_flush(dev); |
ce286d32 | 6821 | |
4e66ae2e SH |
6822 | /* Send a netdev-removed uevent to the old namespace */ |
6823 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
6824 | ||
ce286d32 | 6825 | /* Actually switch the network namespace */ |
c346dca1 | 6826 | dev_net_set(dev, net); |
ce286d32 | 6827 | |
ce286d32 EB |
6828 | /* If there is an ifindex conflict assign a new one */ |
6829 | if (__dev_get_by_index(net, dev->ifindex)) { | |
6830 | int iflink = (dev->iflink == dev->ifindex); | |
6831 | dev->ifindex = dev_new_index(net); | |
6832 | if (iflink) | |
6833 | dev->iflink = dev->ifindex; | |
6834 | } | |
6835 | ||
4e66ae2e SH |
6836 | /* Send a netdev-add uevent to the new namespace */ |
6837 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
6838 | ||
8b41d188 | 6839 | /* Fixup kobjects */ |
a1b3f594 | 6840 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 6841 | WARN_ON(err); |
ce286d32 EB |
6842 | |
6843 | /* Add the device back in the hashes */ | |
6844 | list_netdevice(dev); | |
6845 | ||
6846 | /* Notify protocols, that a new device appeared. */ | |
6847 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
6848 | ||
d90a909e EB |
6849 | /* |
6850 | * Prevent userspace races by waiting until the network | |
6851 | * device is fully setup before sending notifications. | |
6852 | */ | |
7f294054 | 6853 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 6854 | |
ce286d32 EB |
6855 | synchronize_net(); |
6856 | err = 0; | |
6857 | out: | |
6858 | return err; | |
6859 | } | |
463d0183 | 6860 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 6861 | |
1da177e4 LT |
6862 | static int dev_cpu_callback(struct notifier_block *nfb, |
6863 | unsigned long action, | |
6864 | void *ocpu) | |
6865 | { | |
6866 | struct sk_buff **list_skb; | |
1da177e4 LT |
6867 | struct sk_buff *skb; |
6868 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
6869 | struct softnet_data *sd, *oldsd; | |
6870 | ||
8bb78442 | 6871 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
6872 | return NOTIFY_OK; |
6873 | ||
6874 | local_irq_disable(); | |
6875 | cpu = smp_processor_id(); | |
6876 | sd = &per_cpu(softnet_data, cpu); | |
6877 | oldsd = &per_cpu(softnet_data, oldcpu); | |
6878 | ||
6879 | /* Find end of our completion_queue. */ | |
6880 | list_skb = &sd->completion_queue; | |
6881 | while (*list_skb) | |
6882 | list_skb = &(*list_skb)->next; | |
6883 | /* Append completion queue from offline CPU. */ | |
6884 | *list_skb = oldsd->completion_queue; | |
6885 | oldsd->completion_queue = NULL; | |
6886 | ||
1da177e4 | 6887 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
6888 | if (oldsd->output_queue) { |
6889 | *sd->output_queue_tailp = oldsd->output_queue; | |
6890 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
6891 | oldsd->output_queue = NULL; | |
6892 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
6893 | } | |
264524d5 HC |
6894 | /* Append NAPI poll list from offline CPU. */ |
6895 | if (!list_empty(&oldsd->poll_list)) { | |
6896 | list_splice_init(&oldsd->poll_list, &sd->poll_list); | |
6897 | raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6898 | } | |
1da177e4 LT |
6899 | |
6900 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
6901 | local_irq_enable(); | |
6902 | ||
6903 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 6904 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
ae78dbfa | 6905 | netif_rx_internal(skb); |
76cc8b13 | 6906 | input_queue_head_incr(oldsd); |
fec5e652 | 6907 | } |
76cc8b13 | 6908 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
ae78dbfa | 6909 | netif_rx_internal(skb); |
76cc8b13 TH |
6910 | input_queue_head_incr(oldsd); |
6911 | } | |
1da177e4 LT |
6912 | |
6913 | return NOTIFY_OK; | |
6914 | } | |
1da177e4 LT |
6915 | |
6916 | ||
7f353bf2 | 6917 | /** |
b63365a2 HX |
6918 | * netdev_increment_features - increment feature set by one |
6919 | * @all: current feature set | |
6920 | * @one: new feature set | |
6921 | * @mask: mask feature set | |
7f353bf2 HX |
6922 | * |
6923 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
6924 | * @one to the master device with current feature set @all. Will not |
6925 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 6926 | */ |
c8f44aff MM |
6927 | netdev_features_t netdev_increment_features(netdev_features_t all, |
6928 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 6929 | { |
1742f183 MM |
6930 | if (mask & NETIF_F_GEN_CSUM) |
6931 | mask |= NETIF_F_ALL_CSUM; | |
6932 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 6933 | |
1742f183 MM |
6934 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
6935 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 6936 | |
1742f183 MM |
6937 | /* If one device supports hw checksumming, set for all. */ |
6938 | if (all & NETIF_F_GEN_CSUM) | |
6939 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
6940 | |
6941 | return all; | |
6942 | } | |
b63365a2 | 6943 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 6944 | |
430f03cd | 6945 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
6946 | { |
6947 | int i; | |
6948 | struct hlist_head *hash; | |
6949 | ||
6950 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
6951 | if (hash != NULL) | |
6952 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
6953 | INIT_HLIST_HEAD(&hash[i]); | |
6954 | ||
6955 | return hash; | |
6956 | } | |
6957 | ||
881d966b | 6958 | /* Initialize per network namespace state */ |
4665079c | 6959 | static int __net_init netdev_init(struct net *net) |
881d966b | 6960 | { |
734b6541 RM |
6961 | if (net != &init_net) |
6962 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 6963 | |
30d97d35 PE |
6964 | net->dev_name_head = netdev_create_hash(); |
6965 | if (net->dev_name_head == NULL) | |
6966 | goto err_name; | |
881d966b | 6967 | |
30d97d35 PE |
6968 | net->dev_index_head = netdev_create_hash(); |
6969 | if (net->dev_index_head == NULL) | |
6970 | goto err_idx; | |
881d966b EB |
6971 | |
6972 | return 0; | |
30d97d35 PE |
6973 | |
6974 | err_idx: | |
6975 | kfree(net->dev_name_head); | |
6976 | err_name: | |
6977 | return -ENOMEM; | |
881d966b EB |
6978 | } |
6979 | ||
f0db275a SH |
6980 | /** |
6981 | * netdev_drivername - network driver for the device | |
6982 | * @dev: network device | |
f0db275a SH |
6983 | * |
6984 | * Determine network driver for device. | |
6985 | */ | |
3019de12 | 6986 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 6987 | { |
cf04a4c7 SH |
6988 | const struct device_driver *driver; |
6989 | const struct device *parent; | |
3019de12 | 6990 | const char *empty = ""; |
6579e57b AV |
6991 | |
6992 | parent = dev->dev.parent; | |
6579e57b | 6993 | if (!parent) |
3019de12 | 6994 | return empty; |
6579e57b AV |
6995 | |
6996 | driver = parent->driver; | |
6997 | if (driver && driver->name) | |
3019de12 DM |
6998 | return driver->name; |
6999 | return empty; | |
6579e57b AV |
7000 | } |
7001 | ||
b004ff49 | 7002 | static int __netdev_printk(const char *level, const struct net_device *dev, |
256df2f3 JP |
7003 | struct va_format *vaf) |
7004 | { | |
7005 | int r; | |
7006 | ||
b004ff49 | 7007 | if (dev && dev->dev.parent) { |
666f355f JP |
7008 | r = dev_printk_emit(level[1] - '0', |
7009 | dev->dev.parent, | |
ccc7f496 | 7010 | "%s %s %s%s: %pV", |
666f355f JP |
7011 | dev_driver_string(dev->dev.parent), |
7012 | dev_name(dev->dev.parent), | |
ccc7f496 VF |
7013 | netdev_name(dev), netdev_reg_state(dev), |
7014 | vaf); | |
b004ff49 | 7015 | } else if (dev) { |
ccc7f496 VF |
7016 | r = printk("%s%s%s: %pV", level, netdev_name(dev), |
7017 | netdev_reg_state(dev), vaf); | |
b004ff49 | 7018 | } else { |
256df2f3 | 7019 | r = printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 7020 | } |
256df2f3 JP |
7021 | |
7022 | return r; | |
7023 | } | |
7024 | ||
7025 | int netdev_printk(const char *level, const struct net_device *dev, | |
7026 | const char *format, ...) | |
7027 | { | |
7028 | struct va_format vaf; | |
7029 | va_list args; | |
7030 | int r; | |
7031 | ||
7032 | va_start(args, format); | |
7033 | ||
7034 | vaf.fmt = format; | |
7035 | vaf.va = &args; | |
7036 | ||
7037 | r = __netdev_printk(level, dev, &vaf); | |
b004ff49 | 7038 | |
256df2f3 JP |
7039 | va_end(args); |
7040 | ||
7041 | return r; | |
7042 | } | |
7043 | EXPORT_SYMBOL(netdev_printk); | |
7044 | ||
7045 | #define define_netdev_printk_level(func, level) \ | |
7046 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
7047 | { \ | |
7048 | int r; \ | |
7049 | struct va_format vaf; \ | |
7050 | va_list args; \ | |
7051 | \ | |
7052 | va_start(args, fmt); \ | |
7053 | \ | |
7054 | vaf.fmt = fmt; \ | |
7055 | vaf.va = &args; \ | |
7056 | \ | |
7057 | r = __netdev_printk(level, dev, &vaf); \ | |
b004ff49 | 7058 | \ |
256df2f3 JP |
7059 | va_end(args); \ |
7060 | \ | |
7061 | return r; \ | |
7062 | } \ | |
7063 | EXPORT_SYMBOL(func); | |
7064 | ||
7065 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
7066 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
7067 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
7068 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
7069 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
7070 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
7071 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
7072 | ||
4665079c | 7073 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
7074 | { |
7075 | kfree(net->dev_name_head); | |
7076 | kfree(net->dev_index_head); | |
7077 | } | |
7078 | ||
022cbae6 | 7079 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
7080 | .init = netdev_init, |
7081 | .exit = netdev_exit, | |
7082 | }; | |
7083 | ||
4665079c | 7084 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 7085 | { |
e008b5fc | 7086 | struct net_device *dev, *aux; |
ce286d32 | 7087 | /* |
e008b5fc | 7088 | * Push all migratable network devices back to the |
ce286d32 EB |
7089 | * initial network namespace |
7090 | */ | |
7091 | rtnl_lock(); | |
e008b5fc | 7092 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 7093 | int err; |
aca51397 | 7094 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
7095 | |
7096 | /* Ignore unmoveable devices (i.e. loopback) */ | |
7097 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7098 | continue; | |
7099 | ||
e008b5fc EB |
7100 | /* Leave virtual devices for the generic cleanup */ |
7101 | if (dev->rtnl_link_ops) | |
7102 | continue; | |
d0c082ce | 7103 | |
25985edc | 7104 | /* Push remaining network devices to init_net */ |
aca51397 PE |
7105 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
7106 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 7107 | if (err) { |
7b6cd1ce JP |
7108 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
7109 | __func__, dev->name, err); | |
aca51397 | 7110 | BUG(); |
ce286d32 EB |
7111 | } |
7112 | } | |
7113 | rtnl_unlock(); | |
7114 | } | |
7115 | ||
50624c93 EB |
7116 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
7117 | { | |
7118 | /* Return with the rtnl_lock held when there are no network | |
7119 | * devices unregistering in any network namespace in net_list. | |
7120 | */ | |
7121 | struct net *net; | |
7122 | bool unregistering; | |
7123 | DEFINE_WAIT(wait); | |
7124 | ||
7125 | for (;;) { | |
7126 | prepare_to_wait(&netdev_unregistering_wq, &wait, | |
7127 | TASK_UNINTERRUPTIBLE); | |
7128 | unregistering = false; | |
7129 | rtnl_lock(); | |
7130 | list_for_each_entry(net, net_list, exit_list) { | |
7131 | if (net->dev_unreg_count > 0) { | |
7132 | unregistering = true; | |
7133 | break; | |
7134 | } | |
7135 | } | |
7136 | if (!unregistering) | |
7137 | break; | |
7138 | __rtnl_unlock(); | |
7139 | schedule(); | |
7140 | } | |
7141 | finish_wait(&netdev_unregistering_wq, &wait); | |
7142 | } | |
7143 | ||
04dc7f6b EB |
7144 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
7145 | { | |
7146 | /* At exit all network devices most be removed from a network | |
b595076a | 7147 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
7148 | * Do this across as many network namespaces as possible to |
7149 | * improve batching efficiency. | |
7150 | */ | |
7151 | struct net_device *dev; | |
7152 | struct net *net; | |
7153 | LIST_HEAD(dev_kill_list); | |
7154 | ||
50624c93 EB |
7155 | /* To prevent network device cleanup code from dereferencing |
7156 | * loopback devices or network devices that have been freed | |
7157 | * wait here for all pending unregistrations to complete, | |
7158 | * before unregistring the loopback device and allowing the | |
7159 | * network namespace be freed. | |
7160 | * | |
7161 | * The netdev todo list containing all network devices | |
7162 | * unregistrations that happen in default_device_exit_batch | |
7163 | * will run in the rtnl_unlock() at the end of | |
7164 | * default_device_exit_batch. | |
7165 | */ | |
7166 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
7167 | list_for_each_entry(net, net_list, exit_list) { |
7168 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 7169 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
7170 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
7171 | else | |
7172 | unregister_netdevice_queue(dev, &dev_kill_list); | |
7173 | } | |
7174 | } | |
7175 | unregister_netdevice_many(&dev_kill_list); | |
7176 | rtnl_unlock(); | |
7177 | } | |
7178 | ||
022cbae6 | 7179 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 7180 | .exit = default_device_exit, |
04dc7f6b | 7181 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
7182 | }; |
7183 | ||
1da177e4 LT |
7184 | /* |
7185 | * Initialize the DEV module. At boot time this walks the device list and | |
7186 | * unhooks any devices that fail to initialise (normally hardware not | |
7187 | * present) and leaves us with a valid list of present and active devices. | |
7188 | * | |
7189 | */ | |
7190 | ||
7191 | /* | |
7192 | * This is called single threaded during boot, so no need | |
7193 | * to take the rtnl semaphore. | |
7194 | */ | |
7195 | static int __init net_dev_init(void) | |
7196 | { | |
7197 | int i, rc = -ENOMEM; | |
7198 | ||
7199 | BUG_ON(!dev_boot_phase); | |
7200 | ||
1da177e4 LT |
7201 | if (dev_proc_init()) |
7202 | goto out; | |
7203 | ||
8b41d188 | 7204 | if (netdev_kobject_init()) |
1da177e4 LT |
7205 | goto out; |
7206 | ||
7207 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 7208 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
7209 | INIT_LIST_HEAD(&ptype_base[i]); |
7210 | ||
62532da9 VY |
7211 | INIT_LIST_HEAD(&offload_base); |
7212 | ||
881d966b EB |
7213 | if (register_pernet_subsys(&netdev_net_ops)) |
7214 | goto out; | |
1da177e4 LT |
7215 | |
7216 | /* | |
7217 | * Initialise the packet receive queues. | |
7218 | */ | |
7219 | ||
6f912042 | 7220 | for_each_possible_cpu(i) { |
e36fa2f7 | 7221 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 7222 | |
e36fa2f7 | 7223 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 7224 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 7225 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 7226 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 7227 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
7228 | sd->csd.func = rps_trigger_softirq; |
7229 | sd->csd.info = sd; | |
e36fa2f7 | 7230 | sd->cpu = i; |
1e94d72f | 7231 | #endif |
0a9627f2 | 7232 | |
e36fa2f7 ED |
7233 | sd->backlog.poll = process_backlog; |
7234 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
7235 | } |
7236 | ||
1da177e4 LT |
7237 | dev_boot_phase = 0; |
7238 | ||
505d4f73 EB |
7239 | /* The loopback device is special if any other network devices |
7240 | * is present in a network namespace the loopback device must | |
7241 | * be present. Since we now dynamically allocate and free the | |
7242 | * loopback device ensure this invariant is maintained by | |
7243 | * keeping the loopback device as the first device on the | |
7244 | * list of network devices. Ensuring the loopback devices | |
7245 | * is the first device that appears and the last network device | |
7246 | * that disappears. | |
7247 | */ | |
7248 | if (register_pernet_device(&loopback_net_ops)) | |
7249 | goto out; | |
7250 | ||
7251 | if (register_pernet_device(&default_device_ops)) | |
7252 | goto out; | |
7253 | ||
962cf36c CM |
7254 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
7255 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
7256 | |
7257 | hotcpu_notifier(dev_cpu_callback, 0); | |
7258 | dst_init(); | |
1da177e4 LT |
7259 | rc = 0; |
7260 | out: | |
7261 | return rc; | |
7262 | } | |
7263 | ||
7264 | subsys_initcall(net_dev_init); |