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> |
1da177e4 | 135 | |
342709ef PE |
136 | #include "net-sysfs.h" |
137 | ||
d565b0a1 HX |
138 | /* Instead of increasing this, you should create a hash table. */ |
139 | #define MAX_GRO_SKBS 8 | |
140 | ||
5d38a079 HX |
141 | /* This should be increased if a protocol with a bigger head is added. */ |
142 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
143 | ||
1da177e4 | 144 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 145 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
146 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
147 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 148 | static struct list_head offload_base __read_mostly; |
1da177e4 | 149 | |
ae78dbfa BH |
150 | static int netif_rx_internal(struct sk_buff *skb); |
151 | ||
1da177e4 | 152 | /* |
7562f876 | 153 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
154 | * semaphore. |
155 | * | |
c6d14c84 | 156 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
157 | * |
158 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 159 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
160 | * actual updates. This allows pure readers to access the list even |
161 | * while a writer is preparing to update it. | |
162 | * | |
163 | * To put it another way, dev_base_lock is held for writing only to | |
164 | * protect against pure readers; the rtnl semaphore provides the | |
165 | * protection against other writers. | |
166 | * | |
167 | * See, for example usages, register_netdevice() and | |
168 | * unregister_netdevice(), which must be called with the rtnl | |
169 | * semaphore held. | |
170 | */ | |
1da177e4 | 171 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
172 | EXPORT_SYMBOL(dev_base_lock); |
173 | ||
af12fa6e ET |
174 | /* protects napi_hash addition/deletion and napi_gen_id */ |
175 | static DEFINE_SPINLOCK(napi_hash_lock); | |
176 | ||
177 | static unsigned int napi_gen_id; | |
178 | static DEFINE_HASHTABLE(napi_hash, 8); | |
179 | ||
18afa4b0 | 180 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 181 | |
4e985ada TG |
182 | static inline void dev_base_seq_inc(struct net *net) |
183 | { | |
184 | while (++net->dev_base_seq == 0); | |
185 | } | |
186 | ||
881d966b | 187 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 188 | { |
95c96174 ED |
189 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
190 | ||
08e9897d | 191 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
192 | } |
193 | ||
881d966b | 194 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 195 | { |
7c28bd0b | 196 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
197 | } |
198 | ||
e36fa2f7 | 199 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
200 | { |
201 | #ifdef CONFIG_RPS | |
e36fa2f7 | 202 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
203 | #endif |
204 | } | |
205 | ||
e36fa2f7 | 206 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
207 | { |
208 | #ifdef CONFIG_RPS | |
e36fa2f7 | 209 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
210 | #endif |
211 | } | |
212 | ||
ce286d32 | 213 | /* Device list insertion */ |
53759be9 | 214 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 215 | { |
c346dca1 | 216 | struct net *net = dev_net(dev); |
ce286d32 EB |
217 | |
218 | ASSERT_RTNL(); | |
219 | ||
220 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 221 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 222 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
223 | hlist_add_head_rcu(&dev->index_hlist, |
224 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 225 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
226 | |
227 | dev_base_seq_inc(net); | |
ce286d32 EB |
228 | } |
229 | ||
fb699dfd ED |
230 | /* Device list removal |
231 | * caller must respect a RCU grace period before freeing/reusing dev | |
232 | */ | |
ce286d32 EB |
233 | static void unlist_netdevice(struct net_device *dev) |
234 | { | |
235 | ASSERT_RTNL(); | |
236 | ||
237 | /* Unlink dev from the device chain */ | |
238 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 239 | list_del_rcu(&dev->dev_list); |
72c9528b | 240 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 241 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 242 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
243 | |
244 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
245 | } |
246 | ||
1da177e4 LT |
247 | /* |
248 | * Our notifier list | |
249 | */ | |
250 | ||
f07d5b94 | 251 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
252 | |
253 | /* | |
254 | * Device drivers call our routines to queue packets here. We empty the | |
255 | * queue in the local softnet handler. | |
256 | */ | |
bea3348e | 257 | |
9958da05 | 258 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 259 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 260 | |
cf508b12 | 261 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 262 | /* |
c773e847 | 263 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
264 | * according to dev->type |
265 | */ | |
266 | static const unsigned short netdev_lock_type[] = | |
267 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
268 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
269 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
270 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
271 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
272 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
273 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
274 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
275 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
276 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
277 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
278 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
279 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
280 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
281 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 282 | |
36cbd3dc | 283 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
284 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
285 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
286 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
287 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
288 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
289 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
290 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
291 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
292 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
293 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
294 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
295 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
296 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
297 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
298 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
299 | |
300 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 301 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
302 | |
303 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
304 | { | |
305 | int i; | |
306 | ||
307 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
308 | if (netdev_lock_type[i] == dev_type) | |
309 | return i; | |
310 | /* the last key is used by default */ | |
311 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
312 | } | |
313 | ||
cf508b12 DM |
314 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
315 | unsigned short dev_type) | |
723e98b7 JP |
316 | { |
317 | int i; | |
318 | ||
319 | i = netdev_lock_pos(dev_type); | |
320 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
321 | netdev_lock_name[i]); | |
322 | } | |
cf508b12 DM |
323 | |
324 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
325 | { | |
326 | int i; | |
327 | ||
328 | i = netdev_lock_pos(dev->type); | |
329 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
330 | &netdev_addr_lock_key[i], | |
331 | netdev_lock_name[i]); | |
332 | } | |
723e98b7 | 333 | #else |
cf508b12 DM |
334 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
335 | unsigned short dev_type) | |
336 | { | |
337 | } | |
338 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
339 | { |
340 | } | |
341 | #endif | |
1da177e4 LT |
342 | |
343 | /******************************************************************************* | |
344 | ||
345 | Protocol management and registration routines | |
346 | ||
347 | *******************************************************************************/ | |
348 | ||
1da177e4 LT |
349 | /* |
350 | * Add a protocol ID to the list. Now that the input handler is | |
351 | * smarter we can dispense with all the messy stuff that used to be | |
352 | * here. | |
353 | * | |
354 | * BEWARE!!! Protocol handlers, mangling input packets, | |
355 | * MUST BE last in hash buckets and checking protocol handlers | |
356 | * MUST start from promiscuous ptype_all chain in net_bh. | |
357 | * It is true now, do not change it. | |
358 | * Explanation follows: if protocol handler, mangling packet, will | |
359 | * be the first on list, it is not able to sense, that packet | |
360 | * is cloned and should be copied-on-write, so that it will | |
361 | * change it and subsequent readers will get broken packet. | |
362 | * --ANK (980803) | |
363 | */ | |
364 | ||
c07b68e8 ED |
365 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
366 | { | |
367 | if (pt->type == htons(ETH_P_ALL)) | |
368 | return &ptype_all; | |
369 | else | |
370 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
371 | } | |
372 | ||
1da177e4 LT |
373 | /** |
374 | * dev_add_pack - add packet handler | |
375 | * @pt: packet type declaration | |
376 | * | |
377 | * Add a protocol handler to the networking stack. The passed &packet_type | |
378 | * is linked into kernel lists and may not be freed until it has been | |
379 | * removed from the kernel lists. | |
380 | * | |
4ec93edb | 381 | * This call does not sleep therefore it can not |
1da177e4 LT |
382 | * guarantee all CPU's that are in middle of receiving packets |
383 | * will see the new packet type (until the next received packet). | |
384 | */ | |
385 | ||
386 | void dev_add_pack(struct packet_type *pt) | |
387 | { | |
c07b68e8 | 388 | struct list_head *head = ptype_head(pt); |
1da177e4 | 389 | |
c07b68e8 ED |
390 | spin_lock(&ptype_lock); |
391 | list_add_rcu(&pt->list, head); | |
392 | spin_unlock(&ptype_lock); | |
1da177e4 | 393 | } |
d1b19dff | 394 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 395 | |
1da177e4 LT |
396 | /** |
397 | * __dev_remove_pack - remove packet handler | |
398 | * @pt: packet type declaration | |
399 | * | |
400 | * Remove a protocol handler that was previously added to the kernel | |
401 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
402 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 403 | * returns. |
1da177e4 LT |
404 | * |
405 | * The packet type might still be in use by receivers | |
406 | * and must not be freed until after all the CPU's have gone | |
407 | * through a quiescent state. | |
408 | */ | |
409 | void __dev_remove_pack(struct packet_type *pt) | |
410 | { | |
c07b68e8 | 411 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
412 | struct packet_type *pt1; |
413 | ||
c07b68e8 | 414 | spin_lock(&ptype_lock); |
1da177e4 LT |
415 | |
416 | list_for_each_entry(pt1, head, list) { | |
417 | if (pt == pt1) { | |
418 | list_del_rcu(&pt->list); | |
419 | goto out; | |
420 | } | |
421 | } | |
422 | ||
7b6cd1ce | 423 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 424 | out: |
c07b68e8 | 425 | spin_unlock(&ptype_lock); |
1da177e4 | 426 | } |
d1b19dff ED |
427 | EXPORT_SYMBOL(__dev_remove_pack); |
428 | ||
1da177e4 LT |
429 | /** |
430 | * dev_remove_pack - remove packet handler | |
431 | * @pt: packet type declaration | |
432 | * | |
433 | * Remove a protocol handler that was previously added to the kernel | |
434 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
435 | * from the kernel lists and can be freed or reused once this function | |
436 | * returns. | |
437 | * | |
438 | * This call sleeps to guarantee that no CPU is looking at the packet | |
439 | * type after return. | |
440 | */ | |
441 | void dev_remove_pack(struct packet_type *pt) | |
442 | { | |
443 | __dev_remove_pack(pt); | |
4ec93edb | 444 | |
1da177e4 LT |
445 | synchronize_net(); |
446 | } | |
d1b19dff | 447 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 448 | |
62532da9 VY |
449 | |
450 | /** | |
451 | * dev_add_offload - register offload handlers | |
452 | * @po: protocol offload declaration | |
453 | * | |
454 | * Add protocol offload handlers to the networking stack. The passed | |
455 | * &proto_offload is linked into kernel lists and may not be freed until | |
456 | * it has been removed from the kernel lists. | |
457 | * | |
458 | * This call does not sleep therefore it can not | |
459 | * guarantee all CPU's that are in middle of receiving packets | |
460 | * will see the new offload handlers (until the next received packet). | |
461 | */ | |
462 | void dev_add_offload(struct packet_offload *po) | |
463 | { | |
464 | struct list_head *head = &offload_base; | |
465 | ||
466 | spin_lock(&offload_lock); | |
467 | list_add_rcu(&po->list, head); | |
468 | spin_unlock(&offload_lock); | |
469 | } | |
470 | EXPORT_SYMBOL(dev_add_offload); | |
471 | ||
472 | /** | |
473 | * __dev_remove_offload - remove offload handler | |
474 | * @po: packet offload declaration | |
475 | * | |
476 | * Remove a protocol offload handler that was previously added to the | |
477 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
478 | * is removed from the kernel lists and can be freed or reused once this | |
479 | * function returns. | |
480 | * | |
481 | * The packet type might still be in use by receivers | |
482 | * and must not be freed until after all the CPU's have gone | |
483 | * through a quiescent state. | |
484 | */ | |
1d143d9f | 485 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
486 | { |
487 | struct list_head *head = &offload_base; | |
488 | struct packet_offload *po1; | |
489 | ||
c53aa505 | 490 | spin_lock(&offload_lock); |
62532da9 VY |
491 | |
492 | list_for_each_entry(po1, head, list) { | |
493 | if (po == po1) { | |
494 | list_del_rcu(&po->list); | |
495 | goto out; | |
496 | } | |
497 | } | |
498 | ||
499 | pr_warn("dev_remove_offload: %p not found\n", po); | |
500 | out: | |
c53aa505 | 501 | spin_unlock(&offload_lock); |
62532da9 | 502 | } |
62532da9 VY |
503 | |
504 | /** | |
505 | * dev_remove_offload - remove packet offload handler | |
506 | * @po: packet offload declaration | |
507 | * | |
508 | * Remove a packet offload handler that was previously added to the kernel | |
509 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
510 | * removed from the kernel lists and can be freed or reused once this | |
511 | * function returns. | |
512 | * | |
513 | * This call sleeps to guarantee that no CPU is looking at the packet | |
514 | * type after return. | |
515 | */ | |
516 | void dev_remove_offload(struct packet_offload *po) | |
517 | { | |
518 | __dev_remove_offload(po); | |
519 | ||
520 | synchronize_net(); | |
521 | } | |
522 | EXPORT_SYMBOL(dev_remove_offload); | |
523 | ||
1da177e4 LT |
524 | /****************************************************************************** |
525 | ||
526 | Device Boot-time Settings Routines | |
527 | ||
528 | *******************************************************************************/ | |
529 | ||
530 | /* Boot time configuration table */ | |
531 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
532 | ||
533 | /** | |
534 | * netdev_boot_setup_add - add new setup entry | |
535 | * @name: name of the device | |
536 | * @map: configured settings for the device | |
537 | * | |
538 | * Adds new setup entry to the dev_boot_setup list. The function | |
539 | * returns 0 on error and 1 on success. This is a generic routine to | |
540 | * all netdevices. | |
541 | */ | |
542 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
543 | { | |
544 | struct netdev_boot_setup *s; | |
545 | int i; | |
546 | ||
547 | s = dev_boot_setup; | |
548 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
549 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
550 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 551 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
552 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
553 | break; | |
554 | } | |
555 | } | |
556 | ||
557 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
558 | } | |
559 | ||
560 | /** | |
561 | * netdev_boot_setup_check - check boot time settings | |
562 | * @dev: the netdevice | |
563 | * | |
564 | * Check boot time settings for the device. | |
565 | * The found settings are set for the device to be used | |
566 | * later in the device probing. | |
567 | * Returns 0 if no settings found, 1 if they are. | |
568 | */ | |
569 | int netdev_boot_setup_check(struct net_device *dev) | |
570 | { | |
571 | struct netdev_boot_setup *s = dev_boot_setup; | |
572 | int i; | |
573 | ||
574 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
575 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 576 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
577 | dev->irq = s[i].map.irq; |
578 | dev->base_addr = s[i].map.base_addr; | |
579 | dev->mem_start = s[i].map.mem_start; | |
580 | dev->mem_end = s[i].map.mem_end; | |
581 | return 1; | |
582 | } | |
583 | } | |
584 | return 0; | |
585 | } | |
d1b19dff | 586 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
587 | |
588 | ||
589 | /** | |
590 | * netdev_boot_base - get address from boot time settings | |
591 | * @prefix: prefix for network device | |
592 | * @unit: id for network device | |
593 | * | |
594 | * Check boot time settings for the base address of device. | |
595 | * The found settings are set for the device to be used | |
596 | * later in the device probing. | |
597 | * Returns 0 if no settings found. | |
598 | */ | |
599 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
600 | { | |
601 | const struct netdev_boot_setup *s = dev_boot_setup; | |
602 | char name[IFNAMSIZ]; | |
603 | int i; | |
604 | ||
605 | sprintf(name, "%s%d", prefix, unit); | |
606 | ||
607 | /* | |
608 | * If device already registered then return base of 1 | |
609 | * to indicate not to probe for this interface | |
610 | */ | |
881d966b | 611 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
612 | return 1; |
613 | ||
614 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
615 | if (!strcmp(name, s[i].name)) | |
616 | return s[i].map.base_addr; | |
617 | return 0; | |
618 | } | |
619 | ||
620 | /* | |
621 | * Saves at boot time configured settings for any netdevice. | |
622 | */ | |
623 | int __init netdev_boot_setup(char *str) | |
624 | { | |
625 | int ints[5]; | |
626 | struct ifmap map; | |
627 | ||
628 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
629 | if (!str || !*str) | |
630 | return 0; | |
631 | ||
632 | /* Save settings */ | |
633 | memset(&map, 0, sizeof(map)); | |
634 | if (ints[0] > 0) | |
635 | map.irq = ints[1]; | |
636 | if (ints[0] > 1) | |
637 | map.base_addr = ints[2]; | |
638 | if (ints[0] > 2) | |
639 | map.mem_start = ints[3]; | |
640 | if (ints[0] > 3) | |
641 | map.mem_end = ints[4]; | |
642 | ||
643 | /* Add new entry to the list */ | |
644 | return netdev_boot_setup_add(str, &map); | |
645 | } | |
646 | ||
647 | __setup("netdev=", netdev_boot_setup); | |
648 | ||
649 | /******************************************************************************* | |
650 | ||
651 | Device Interface Subroutines | |
652 | ||
653 | *******************************************************************************/ | |
654 | ||
655 | /** | |
656 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 657 | * @net: the applicable net namespace |
1da177e4 LT |
658 | * @name: name to find |
659 | * | |
660 | * Find an interface by name. Must be called under RTNL semaphore | |
661 | * or @dev_base_lock. If the name is found a pointer to the device | |
662 | * is returned. If the name is not found then %NULL is returned. The | |
663 | * reference counters are not incremented so the caller must be | |
664 | * careful with locks. | |
665 | */ | |
666 | ||
881d966b | 667 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 668 | { |
0bd8d536 ED |
669 | struct net_device *dev; |
670 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 671 | |
b67bfe0d | 672 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
673 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
674 | return dev; | |
0bd8d536 | 675 | |
1da177e4 LT |
676 | return NULL; |
677 | } | |
d1b19dff | 678 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 679 | |
72c9528b ED |
680 | /** |
681 | * dev_get_by_name_rcu - find a device by its name | |
682 | * @net: the applicable net namespace | |
683 | * @name: name to find | |
684 | * | |
685 | * Find an interface by name. | |
686 | * If the name is found a pointer to the device is returned. | |
687 | * If the name is not found then %NULL is returned. | |
688 | * The reference counters are not incremented so the caller must be | |
689 | * careful with locks. The caller must hold RCU lock. | |
690 | */ | |
691 | ||
692 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
693 | { | |
72c9528b ED |
694 | struct net_device *dev; |
695 | struct hlist_head *head = dev_name_hash(net, name); | |
696 | ||
b67bfe0d | 697 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
698 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
699 | return dev; | |
700 | ||
701 | return NULL; | |
702 | } | |
703 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
704 | ||
1da177e4 LT |
705 | /** |
706 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 707 | * @net: the applicable net namespace |
1da177e4 LT |
708 | * @name: name to find |
709 | * | |
710 | * Find an interface by name. This can be called from any | |
711 | * context and does its own locking. The returned handle has | |
712 | * the usage count incremented and the caller must use dev_put() to | |
713 | * release it when it is no longer needed. %NULL is returned if no | |
714 | * matching device is found. | |
715 | */ | |
716 | ||
881d966b | 717 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
718 | { |
719 | struct net_device *dev; | |
720 | ||
72c9528b ED |
721 | rcu_read_lock(); |
722 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
723 | if (dev) |
724 | dev_hold(dev); | |
72c9528b | 725 | rcu_read_unlock(); |
1da177e4 LT |
726 | return dev; |
727 | } | |
d1b19dff | 728 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
729 | |
730 | /** | |
731 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 732 | * @net: the applicable net namespace |
1da177e4 LT |
733 | * @ifindex: index of device |
734 | * | |
735 | * Search for an interface by index. Returns %NULL if the device | |
736 | * is not found or a pointer to the device. The device has not | |
737 | * had its reference counter increased so the caller must be careful | |
738 | * about locking. The caller must hold either the RTNL semaphore | |
739 | * or @dev_base_lock. | |
740 | */ | |
741 | ||
881d966b | 742 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 743 | { |
0bd8d536 ED |
744 | struct net_device *dev; |
745 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 746 | |
b67bfe0d | 747 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
748 | if (dev->ifindex == ifindex) |
749 | return dev; | |
0bd8d536 | 750 | |
1da177e4 LT |
751 | return NULL; |
752 | } | |
d1b19dff | 753 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 754 | |
fb699dfd ED |
755 | /** |
756 | * dev_get_by_index_rcu - find a device by its ifindex | |
757 | * @net: the applicable net namespace | |
758 | * @ifindex: index of device | |
759 | * | |
760 | * Search for an interface by index. Returns %NULL if the device | |
761 | * is not found or a pointer to the device. The device has not | |
762 | * had its reference counter increased so the caller must be careful | |
763 | * about locking. The caller must hold RCU lock. | |
764 | */ | |
765 | ||
766 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
767 | { | |
fb699dfd ED |
768 | struct net_device *dev; |
769 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
770 | ||
b67bfe0d | 771 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
772 | if (dev->ifindex == ifindex) |
773 | return dev; | |
774 | ||
775 | return NULL; | |
776 | } | |
777 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
778 | ||
1da177e4 LT |
779 | |
780 | /** | |
781 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 782 | * @net: the applicable net namespace |
1da177e4 LT |
783 | * @ifindex: index of device |
784 | * | |
785 | * Search for an interface by index. Returns NULL if the device | |
786 | * is not found or a pointer to the device. The device returned has | |
787 | * had a reference added and the pointer is safe until the user calls | |
788 | * dev_put to indicate they have finished with it. | |
789 | */ | |
790 | ||
881d966b | 791 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
792 | { |
793 | struct net_device *dev; | |
794 | ||
fb699dfd ED |
795 | rcu_read_lock(); |
796 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
797 | if (dev) |
798 | dev_hold(dev); | |
fb699dfd | 799 | rcu_read_unlock(); |
1da177e4 LT |
800 | return dev; |
801 | } | |
d1b19dff | 802 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 803 | |
5dbe7c17 NS |
804 | /** |
805 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
806 | * @net: network namespace | |
807 | * @name: a pointer to the buffer where the name will be stored. | |
808 | * @ifindex: the ifindex of the interface to get the name from. | |
809 | * | |
810 | * The use of raw_seqcount_begin() and cond_resched() before | |
811 | * retrying is required as we want to give the writers a chance | |
812 | * to complete when CONFIG_PREEMPT is not set. | |
813 | */ | |
814 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
815 | { | |
816 | struct net_device *dev; | |
817 | unsigned int seq; | |
818 | ||
819 | retry: | |
820 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
821 | rcu_read_lock(); | |
822 | dev = dev_get_by_index_rcu(net, ifindex); | |
823 | if (!dev) { | |
824 | rcu_read_unlock(); | |
825 | return -ENODEV; | |
826 | } | |
827 | ||
828 | strcpy(name, dev->name); | |
829 | rcu_read_unlock(); | |
830 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
831 | cond_resched(); | |
832 | goto retry; | |
833 | } | |
834 | ||
835 | return 0; | |
836 | } | |
837 | ||
1da177e4 | 838 | /** |
941666c2 | 839 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 840 | * @net: the applicable net namespace |
1da177e4 LT |
841 | * @type: media type of device |
842 | * @ha: hardware address | |
843 | * | |
844 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
845 | * is not found or a pointer to the device. |
846 | * The caller must hold RCU or RTNL. | |
941666c2 | 847 | * The returned device has not had its ref count increased |
1da177e4 LT |
848 | * and the caller must therefore be careful about locking |
849 | * | |
1da177e4 LT |
850 | */ |
851 | ||
941666c2 ED |
852 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
853 | const char *ha) | |
1da177e4 LT |
854 | { |
855 | struct net_device *dev; | |
856 | ||
941666c2 | 857 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
858 | if (dev->type == type && |
859 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
860 | return dev; |
861 | ||
862 | return NULL; | |
1da177e4 | 863 | } |
941666c2 | 864 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 865 | |
881d966b | 866 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
867 | { |
868 | struct net_device *dev; | |
869 | ||
4e9cac2b | 870 | ASSERT_RTNL(); |
881d966b | 871 | for_each_netdev(net, dev) |
4e9cac2b | 872 | if (dev->type == type) |
7562f876 PE |
873 | return dev; |
874 | ||
875 | return NULL; | |
4e9cac2b | 876 | } |
4e9cac2b PM |
877 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
878 | ||
881d966b | 879 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 880 | { |
99fe3c39 | 881 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 882 | |
99fe3c39 ED |
883 | rcu_read_lock(); |
884 | for_each_netdev_rcu(net, dev) | |
885 | if (dev->type == type) { | |
886 | dev_hold(dev); | |
887 | ret = dev; | |
888 | break; | |
889 | } | |
890 | rcu_read_unlock(); | |
891 | return ret; | |
1da177e4 | 892 | } |
1da177e4 LT |
893 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
894 | ||
895 | /** | |
bb69ae04 | 896 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 897 | * @net: the applicable net namespace |
1da177e4 LT |
898 | * @if_flags: IFF_* values |
899 | * @mask: bitmask of bits in if_flags to check | |
900 | * | |
901 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
902 | * is not found or a pointer to the device. Must be called inside |
903 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
904 | */ |
905 | ||
bb69ae04 | 906 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 907 | unsigned short mask) |
1da177e4 | 908 | { |
7562f876 | 909 | struct net_device *dev, *ret; |
1da177e4 | 910 | |
7562f876 | 911 | ret = NULL; |
c6d14c84 | 912 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 913 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 914 | ret = dev; |
1da177e4 LT |
915 | break; |
916 | } | |
917 | } | |
7562f876 | 918 | return ret; |
1da177e4 | 919 | } |
bb69ae04 | 920 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
921 | |
922 | /** | |
923 | * dev_valid_name - check if name is okay for network device | |
924 | * @name: name string | |
925 | * | |
926 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
927 | * to allow sysfs to work. We also disallow any kind of |
928 | * whitespace. | |
1da177e4 | 929 | */ |
95f050bf | 930 | bool dev_valid_name(const char *name) |
1da177e4 | 931 | { |
c7fa9d18 | 932 | if (*name == '\0') |
95f050bf | 933 | return false; |
b6fe17d6 | 934 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 935 | return false; |
c7fa9d18 | 936 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 937 | return false; |
c7fa9d18 DM |
938 | |
939 | while (*name) { | |
940 | if (*name == '/' || isspace(*name)) | |
95f050bf | 941 | return false; |
c7fa9d18 DM |
942 | name++; |
943 | } | |
95f050bf | 944 | return true; |
1da177e4 | 945 | } |
d1b19dff | 946 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
947 | |
948 | /** | |
b267b179 EB |
949 | * __dev_alloc_name - allocate a name for a device |
950 | * @net: network namespace to allocate the device name in | |
1da177e4 | 951 | * @name: name format string |
b267b179 | 952 | * @buf: scratch buffer and result name string |
1da177e4 LT |
953 | * |
954 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
955 | * id. It scans list of devices to build up a free map, then chooses |
956 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
957 | * while allocating the name and adding the device in order to avoid | |
958 | * duplicates. | |
959 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
960 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
961 | */ |
962 | ||
b267b179 | 963 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
964 | { |
965 | int i = 0; | |
1da177e4 LT |
966 | const char *p; |
967 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 968 | unsigned long *inuse; |
1da177e4 LT |
969 | struct net_device *d; |
970 | ||
971 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
972 | if (p) { | |
973 | /* | |
974 | * Verify the string as this thing may have come from | |
975 | * the user. There must be either one "%d" and no other "%" | |
976 | * characters. | |
977 | */ | |
978 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
979 | return -EINVAL; | |
980 | ||
981 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 982 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
983 | if (!inuse) |
984 | return -ENOMEM; | |
985 | ||
881d966b | 986 | for_each_netdev(net, d) { |
1da177e4 LT |
987 | if (!sscanf(d->name, name, &i)) |
988 | continue; | |
989 | if (i < 0 || i >= max_netdevices) | |
990 | continue; | |
991 | ||
992 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 993 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
994 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
995 | set_bit(i, inuse); | |
996 | } | |
997 | ||
998 | i = find_first_zero_bit(inuse, max_netdevices); | |
999 | free_page((unsigned long) inuse); | |
1000 | } | |
1001 | ||
d9031024 OP |
1002 | if (buf != name) |
1003 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1004 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1005 | return i; |
1da177e4 LT |
1006 | |
1007 | /* It is possible to run out of possible slots | |
1008 | * when the name is long and there isn't enough space left | |
1009 | * for the digits, or if all bits are used. | |
1010 | */ | |
1011 | return -ENFILE; | |
1012 | } | |
1013 | ||
b267b179 EB |
1014 | /** |
1015 | * dev_alloc_name - allocate a name for a device | |
1016 | * @dev: device | |
1017 | * @name: name format string | |
1018 | * | |
1019 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1020 | * id. It scans list of devices to build up a free map, then chooses | |
1021 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1022 | * while allocating the name and adding the device in order to avoid | |
1023 | * duplicates. | |
1024 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1025 | * Returns the number of the unit assigned or a negative errno code. | |
1026 | */ | |
1027 | ||
1028 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1029 | { | |
1030 | char buf[IFNAMSIZ]; | |
1031 | struct net *net; | |
1032 | int ret; | |
1033 | ||
c346dca1 YH |
1034 | BUG_ON(!dev_net(dev)); |
1035 | net = dev_net(dev); | |
b267b179 EB |
1036 | ret = __dev_alloc_name(net, name, buf); |
1037 | if (ret >= 0) | |
1038 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1039 | return ret; | |
1040 | } | |
d1b19dff | 1041 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1042 | |
828de4f6 G |
1043 | static int dev_alloc_name_ns(struct net *net, |
1044 | struct net_device *dev, | |
1045 | const char *name) | |
d9031024 | 1046 | { |
828de4f6 G |
1047 | char buf[IFNAMSIZ]; |
1048 | int ret; | |
8ce6cebc | 1049 | |
828de4f6 G |
1050 | ret = __dev_alloc_name(net, name, buf); |
1051 | if (ret >= 0) | |
1052 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1053 | return ret; | |
1054 | } | |
1055 | ||
1056 | static int dev_get_valid_name(struct net *net, | |
1057 | struct net_device *dev, | |
1058 | const char *name) | |
1059 | { | |
1060 | BUG_ON(!net); | |
8ce6cebc | 1061 | |
d9031024 OP |
1062 | if (!dev_valid_name(name)) |
1063 | return -EINVAL; | |
1064 | ||
1c5cae81 | 1065 | if (strchr(name, '%')) |
828de4f6 | 1066 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1067 | else if (__dev_get_by_name(net, name)) |
1068 | return -EEXIST; | |
8ce6cebc DL |
1069 | else if (dev->name != name) |
1070 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1071 | |
1072 | return 0; | |
1073 | } | |
1da177e4 LT |
1074 | |
1075 | /** | |
1076 | * dev_change_name - change name of a device | |
1077 | * @dev: device | |
1078 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1079 | * | |
1080 | * Change name of a device, can pass format strings "eth%d". | |
1081 | * for wildcarding. | |
1082 | */ | |
cf04a4c7 | 1083 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1084 | { |
fcc5a03a | 1085 | char oldname[IFNAMSIZ]; |
1da177e4 | 1086 | int err = 0; |
fcc5a03a | 1087 | int ret; |
881d966b | 1088 | struct net *net; |
1da177e4 LT |
1089 | |
1090 | ASSERT_RTNL(); | |
c346dca1 | 1091 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1092 | |
c346dca1 | 1093 | net = dev_net(dev); |
1da177e4 LT |
1094 | if (dev->flags & IFF_UP) |
1095 | return -EBUSY; | |
1096 | ||
30e6c9fa | 1097 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1098 | |
1099 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1100 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1101 | return 0; |
c91f6df2 | 1102 | } |
c8d90dca | 1103 | |
fcc5a03a HX |
1104 | memcpy(oldname, dev->name, IFNAMSIZ); |
1105 | ||
828de4f6 | 1106 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1107 | if (err < 0) { |
30e6c9fa | 1108 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1109 | return err; |
c91f6df2 | 1110 | } |
1da177e4 | 1111 | |
fcc5a03a | 1112 | rollback: |
a1b3f594 EB |
1113 | ret = device_rename(&dev->dev, dev->name); |
1114 | if (ret) { | |
1115 | memcpy(dev->name, oldname, IFNAMSIZ); | |
30e6c9fa | 1116 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1117 | return ret; |
dcc99773 | 1118 | } |
7f988eab | 1119 | |
30e6c9fa | 1120 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1121 | |
5bb025fa VF |
1122 | netdev_adjacent_rename_links(dev, oldname); |
1123 | ||
7f988eab | 1124 | write_lock_bh(&dev_base_lock); |
372b2312 | 1125 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1126 | write_unlock_bh(&dev_base_lock); |
1127 | ||
1128 | synchronize_rcu(); | |
1129 | ||
1130 | write_lock_bh(&dev_base_lock); | |
1131 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1132 | write_unlock_bh(&dev_base_lock); |
1133 | ||
056925ab | 1134 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1135 | ret = notifier_to_errno(ret); |
1136 | ||
1137 | if (ret) { | |
91e9c07b ED |
1138 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1139 | if (err >= 0) { | |
fcc5a03a | 1140 | err = ret; |
30e6c9fa | 1141 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1142 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1143 | memcpy(oldname, newname, IFNAMSIZ); |
fcc5a03a | 1144 | goto rollback; |
91e9c07b | 1145 | } else { |
7b6cd1ce | 1146 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1147 | dev->name, ret); |
fcc5a03a HX |
1148 | } |
1149 | } | |
1da177e4 LT |
1150 | |
1151 | return err; | |
1152 | } | |
1153 | ||
0b815a1a SH |
1154 | /** |
1155 | * dev_set_alias - change ifalias of a device | |
1156 | * @dev: device | |
1157 | * @alias: name up to IFALIASZ | |
f0db275a | 1158 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1159 | * |
1160 | * Set ifalias for a device, | |
1161 | */ | |
1162 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1163 | { | |
7364e445 AK |
1164 | char *new_ifalias; |
1165 | ||
0b815a1a SH |
1166 | ASSERT_RTNL(); |
1167 | ||
1168 | if (len >= IFALIASZ) | |
1169 | return -EINVAL; | |
1170 | ||
96ca4a2c | 1171 | if (!len) { |
388dfc2d SK |
1172 | kfree(dev->ifalias); |
1173 | dev->ifalias = NULL; | |
96ca4a2c OH |
1174 | return 0; |
1175 | } | |
1176 | ||
7364e445 AK |
1177 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1178 | if (!new_ifalias) | |
0b815a1a | 1179 | return -ENOMEM; |
7364e445 | 1180 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1181 | |
1182 | strlcpy(dev->ifalias, alias, len+1); | |
1183 | return len; | |
1184 | } | |
1185 | ||
1186 | ||
d8a33ac4 | 1187 | /** |
3041a069 | 1188 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1189 | * @dev: device to cause notification |
1190 | * | |
1191 | * Called to indicate a device has changed features. | |
1192 | */ | |
1193 | void netdev_features_change(struct net_device *dev) | |
1194 | { | |
056925ab | 1195 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1196 | } |
1197 | EXPORT_SYMBOL(netdev_features_change); | |
1198 | ||
1da177e4 LT |
1199 | /** |
1200 | * netdev_state_change - device changes state | |
1201 | * @dev: device to cause notification | |
1202 | * | |
1203 | * Called to indicate a device has changed state. This function calls | |
1204 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1205 | * to the routing socket. | |
1206 | */ | |
1207 | void netdev_state_change(struct net_device *dev) | |
1208 | { | |
1209 | if (dev->flags & IFF_UP) { | |
056925ab | 1210 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
7f294054 | 1211 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1212 | } |
1213 | } | |
d1b19dff | 1214 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1215 | |
ee89bab1 AW |
1216 | /** |
1217 | * netdev_notify_peers - notify network peers about existence of @dev | |
1218 | * @dev: network device | |
1219 | * | |
1220 | * Generate traffic such that interested network peers are aware of | |
1221 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1222 | * a device wants to inform the rest of the network about some sort of | |
1223 | * reconfiguration such as a failover event or virtual machine | |
1224 | * migration. | |
1225 | */ | |
1226 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1227 | { |
ee89bab1 AW |
1228 | rtnl_lock(); |
1229 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1230 | rtnl_unlock(); | |
c1da4ac7 | 1231 | } |
ee89bab1 | 1232 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1233 | |
bd380811 | 1234 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1235 | { |
d314774c | 1236 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1237 | int ret; |
1da177e4 | 1238 | |
e46b66bc BH |
1239 | ASSERT_RTNL(); |
1240 | ||
1da177e4 LT |
1241 | if (!netif_device_present(dev)) |
1242 | return -ENODEV; | |
1243 | ||
ca99ca14 NH |
1244 | /* Block netpoll from trying to do any rx path servicing. |
1245 | * If we don't do this there is a chance ndo_poll_controller | |
1246 | * or ndo_poll may be running while we open the device | |
1247 | */ | |
66b5552f | 1248 | netpoll_poll_disable(dev); |
ca99ca14 | 1249 | |
3b8bcfd5 JB |
1250 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1251 | ret = notifier_to_errno(ret); | |
1252 | if (ret) | |
1253 | return ret; | |
1254 | ||
1da177e4 | 1255 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1256 | |
d314774c SH |
1257 | if (ops->ndo_validate_addr) |
1258 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1259 | |
d314774c SH |
1260 | if (!ret && ops->ndo_open) |
1261 | ret = ops->ndo_open(dev); | |
1da177e4 | 1262 | |
66b5552f | 1263 | netpoll_poll_enable(dev); |
ca99ca14 | 1264 | |
bada339b JG |
1265 | if (ret) |
1266 | clear_bit(__LINK_STATE_START, &dev->state); | |
1267 | else { | |
1da177e4 | 1268 | dev->flags |= IFF_UP; |
b4bd07c2 | 1269 | net_dmaengine_get(); |
4417da66 | 1270 | dev_set_rx_mode(dev); |
1da177e4 | 1271 | dev_activate(dev); |
7bf23575 | 1272 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1273 | } |
bada339b | 1274 | |
1da177e4 LT |
1275 | return ret; |
1276 | } | |
1277 | ||
1278 | /** | |
bd380811 PM |
1279 | * dev_open - prepare an interface for use. |
1280 | * @dev: device to open | |
1da177e4 | 1281 | * |
bd380811 PM |
1282 | * Takes a device from down to up state. The device's private open |
1283 | * function is invoked and then the multicast lists are loaded. Finally | |
1284 | * the device is moved into the up state and a %NETDEV_UP message is | |
1285 | * sent to the netdev notifier chain. | |
1286 | * | |
1287 | * Calling this function on an active interface is a nop. On a failure | |
1288 | * a negative errno code is returned. | |
1da177e4 | 1289 | */ |
bd380811 PM |
1290 | int dev_open(struct net_device *dev) |
1291 | { | |
1292 | int ret; | |
1293 | ||
bd380811 PM |
1294 | if (dev->flags & IFF_UP) |
1295 | return 0; | |
1296 | ||
bd380811 PM |
1297 | ret = __dev_open(dev); |
1298 | if (ret < 0) | |
1299 | return ret; | |
1300 | ||
7f294054 | 1301 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1302 | call_netdevice_notifiers(NETDEV_UP, dev); |
1303 | ||
1304 | return ret; | |
1305 | } | |
1306 | EXPORT_SYMBOL(dev_open); | |
1307 | ||
44345724 | 1308 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1309 | { |
44345724 | 1310 | struct net_device *dev; |
e46b66bc | 1311 | |
bd380811 | 1312 | ASSERT_RTNL(); |
9d5010db DM |
1313 | might_sleep(); |
1314 | ||
5cde2829 | 1315 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1316 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1317 | netpoll_poll_disable(dev); |
3f4df206 | 1318 | |
44345724 | 1319 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1320 | |
44345724 | 1321 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1322 | |
44345724 OP |
1323 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1324 | * can be even on different cpu. So just clear netif_running(). | |
1325 | * | |
1326 | * dev->stop() will invoke napi_disable() on all of it's | |
1327 | * napi_struct instances on this device. | |
1328 | */ | |
1329 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1330 | } | |
1da177e4 | 1331 | |
44345724 | 1332 | dev_deactivate_many(head); |
d8b2a4d2 | 1333 | |
5cde2829 | 1334 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1335 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1336 | |
44345724 OP |
1337 | /* |
1338 | * Call the device specific close. This cannot fail. | |
1339 | * Only if device is UP | |
1340 | * | |
1341 | * We allow it to be called even after a DETACH hot-plug | |
1342 | * event. | |
1343 | */ | |
1344 | if (ops->ndo_stop) | |
1345 | ops->ndo_stop(dev); | |
1346 | ||
44345724 | 1347 | dev->flags &= ~IFF_UP; |
44345724 | 1348 | net_dmaengine_put(); |
66b5552f | 1349 | netpoll_poll_enable(dev); |
44345724 OP |
1350 | } |
1351 | ||
1352 | return 0; | |
1353 | } | |
1354 | ||
1355 | static int __dev_close(struct net_device *dev) | |
1356 | { | |
f87e6f47 | 1357 | int retval; |
44345724 OP |
1358 | LIST_HEAD(single); |
1359 | ||
5cde2829 | 1360 | list_add(&dev->close_list, &single); |
f87e6f47 LT |
1361 | retval = __dev_close_many(&single); |
1362 | list_del(&single); | |
ca99ca14 | 1363 | |
f87e6f47 | 1364 | return retval; |
44345724 OP |
1365 | } |
1366 | ||
3fbd8758 | 1367 | static int dev_close_many(struct list_head *head) |
44345724 OP |
1368 | { |
1369 | struct net_device *dev, *tmp; | |
1da177e4 | 1370 | |
5cde2829 EB |
1371 | /* Remove the devices that don't need to be closed */ |
1372 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1373 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1374 | list_del_init(&dev->close_list); |
44345724 OP |
1375 | |
1376 | __dev_close_many(head); | |
1da177e4 | 1377 | |
5cde2829 | 1378 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1379 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1380 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
5cde2829 | 1381 | list_del_init(&dev->close_list); |
44345724 | 1382 | } |
bd380811 PM |
1383 | |
1384 | return 0; | |
1385 | } | |
1386 | ||
1387 | /** | |
1388 | * dev_close - shutdown an interface. | |
1389 | * @dev: device to shutdown | |
1390 | * | |
1391 | * This function moves an active device into down state. A | |
1392 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1393 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1394 | * chain. | |
1395 | */ | |
1396 | int dev_close(struct net_device *dev) | |
1397 | { | |
e14a5993 ED |
1398 | if (dev->flags & IFF_UP) { |
1399 | LIST_HEAD(single); | |
1da177e4 | 1400 | |
5cde2829 | 1401 | list_add(&dev->close_list, &single); |
e14a5993 ED |
1402 | dev_close_many(&single); |
1403 | list_del(&single); | |
1404 | } | |
da6e378b | 1405 | return 0; |
1da177e4 | 1406 | } |
d1b19dff | 1407 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1408 | |
1409 | ||
0187bdfb BH |
1410 | /** |
1411 | * dev_disable_lro - disable Large Receive Offload on a device | |
1412 | * @dev: device | |
1413 | * | |
1414 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1415 | * called under RTNL. This is needed if received packets may be | |
1416 | * forwarded to another interface. | |
1417 | */ | |
1418 | void dev_disable_lro(struct net_device *dev) | |
1419 | { | |
f11970e3 NH |
1420 | /* |
1421 | * If we're trying to disable lro on a vlan device | |
1422 | * use the underlying physical device instead | |
1423 | */ | |
1424 | if (is_vlan_dev(dev)) | |
1425 | dev = vlan_dev_real_dev(dev); | |
1426 | ||
529d0489 MK |
1427 | /* the same for macvlan devices */ |
1428 | if (netif_is_macvlan(dev)) | |
1429 | dev = macvlan_dev_real_dev(dev); | |
1430 | ||
bc5787c6 MM |
1431 | dev->wanted_features &= ~NETIF_F_LRO; |
1432 | netdev_update_features(dev); | |
27660515 | 1433 | |
22d5969f MM |
1434 | if (unlikely(dev->features & NETIF_F_LRO)) |
1435 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
0187bdfb BH |
1436 | } |
1437 | EXPORT_SYMBOL(dev_disable_lro); | |
1438 | ||
351638e7 JP |
1439 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1440 | struct net_device *dev) | |
1441 | { | |
1442 | struct netdev_notifier_info info; | |
1443 | ||
1444 | netdev_notifier_info_init(&info, dev); | |
1445 | return nb->notifier_call(nb, val, &info); | |
1446 | } | |
0187bdfb | 1447 | |
881d966b EB |
1448 | static int dev_boot_phase = 1; |
1449 | ||
1da177e4 LT |
1450 | /** |
1451 | * register_netdevice_notifier - register a network notifier block | |
1452 | * @nb: notifier | |
1453 | * | |
1454 | * Register a notifier to be called when network device events occur. | |
1455 | * The notifier passed is linked into the kernel structures and must | |
1456 | * not be reused until it has been unregistered. A negative errno code | |
1457 | * is returned on a failure. | |
1458 | * | |
1459 | * When registered all registration and up events are replayed | |
4ec93edb | 1460 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1461 | * view of the network device list. |
1462 | */ | |
1463 | ||
1464 | int register_netdevice_notifier(struct notifier_block *nb) | |
1465 | { | |
1466 | struct net_device *dev; | |
fcc5a03a | 1467 | struct net_device *last; |
881d966b | 1468 | struct net *net; |
1da177e4 LT |
1469 | int err; |
1470 | ||
1471 | rtnl_lock(); | |
f07d5b94 | 1472 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1473 | if (err) |
1474 | goto unlock; | |
881d966b EB |
1475 | if (dev_boot_phase) |
1476 | goto unlock; | |
1477 | for_each_net(net) { | |
1478 | for_each_netdev(net, dev) { | |
351638e7 | 1479 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1480 | err = notifier_to_errno(err); |
1481 | if (err) | |
1482 | goto rollback; | |
1483 | ||
1484 | if (!(dev->flags & IFF_UP)) | |
1485 | continue; | |
1da177e4 | 1486 | |
351638e7 | 1487 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1488 | } |
1da177e4 | 1489 | } |
fcc5a03a HX |
1490 | |
1491 | unlock: | |
1da177e4 LT |
1492 | rtnl_unlock(); |
1493 | return err; | |
fcc5a03a HX |
1494 | |
1495 | rollback: | |
1496 | last = dev; | |
881d966b EB |
1497 | for_each_net(net) { |
1498 | for_each_netdev(net, dev) { | |
1499 | if (dev == last) | |
8f891489 | 1500 | goto outroll; |
fcc5a03a | 1501 | |
881d966b | 1502 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1503 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1504 | dev); | |
1505 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1506 | } |
351638e7 | 1507 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1508 | } |
fcc5a03a | 1509 | } |
c67625a1 | 1510 | |
8f891489 | 1511 | outroll: |
c67625a1 | 1512 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1513 | goto unlock; |
1da177e4 | 1514 | } |
d1b19dff | 1515 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1516 | |
1517 | /** | |
1518 | * unregister_netdevice_notifier - unregister a network notifier block | |
1519 | * @nb: notifier | |
1520 | * | |
1521 | * Unregister a notifier previously registered by | |
1522 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1523 | * kernel structures and may then be reused. A negative errno code | |
1524 | * is returned on a failure. | |
7d3d43da EB |
1525 | * |
1526 | * After unregistering unregister and down device events are synthesized | |
1527 | * for all devices on the device list to the removed notifier to remove | |
1528 | * the need for special case cleanup code. | |
1da177e4 LT |
1529 | */ |
1530 | ||
1531 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1532 | { | |
7d3d43da EB |
1533 | struct net_device *dev; |
1534 | struct net *net; | |
9f514950 HX |
1535 | int err; |
1536 | ||
1537 | rtnl_lock(); | |
f07d5b94 | 1538 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1539 | if (err) |
1540 | goto unlock; | |
1541 | ||
1542 | for_each_net(net) { | |
1543 | for_each_netdev(net, dev) { | |
1544 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1545 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1546 | dev); | |
1547 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1548 | } |
351638e7 | 1549 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1550 | } |
1551 | } | |
1552 | unlock: | |
9f514950 HX |
1553 | rtnl_unlock(); |
1554 | return err; | |
1da177e4 | 1555 | } |
d1b19dff | 1556 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1557 | |
351638e7 JP |
1558 | /** |
1559 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1560 | * @val: value passed unmodified to notifier function | |
1561 | * @dev: net_device pointer passed unmodified to notifier function | |
1562 | * @info: notifier information data | |
1563 | * | |
1564 | * Call all network notifier blocks. Parameters and return value | |
1565 | * are as for raw_notifier_call_chain(). | |
1566 | */ | |
1567 | ||
1d143d9f | 1568 | static int call_netdevice_notifiers_info(unsigned long val, |
1569 | struct net_device *dev, | |
1570 | struct netdev_notifier_info *info) | |
351638e7 JP |
1571 | { |
1572 | ASSERT_RTNL(); | |
1573 | netdev_notifier_info_init(info, dev); | |
1574 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1575 | } | |
351638e7 | 1576 | |
1da177e4 LT |
1577 | /** |
1578 | * call_netdevice_notifiers - call all network notifier blocks | |
1579 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1580 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1581 | * |
1582 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1583 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1584 | */ |
1585 | ||
ad7379d4 | 1586 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1587 | { |
351638e7 JP |
1588 | struct netdev_notifier_info info; |
1589 | ||
1590 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1591 | } |
edf947f1 | 1592 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1593 | |
c5905afb | 1594 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1595 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1596 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1597 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1598 | * static_key_slow_dec() calls. |
b90e5794 ED |
1599 | */ |
1600 | static atomic_t netstamp_needed_deferred; | |
1601 | #endif | |
1da177e4 LT |
1602 | |
1603 | void net_enable_timestamp(void) | |
1604 | { | |
b90e5794 ED |
1605 | #ifdef HAVE_JUMP_LABEL |
1606 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1607 | ||
1608 | if (deferred) { | |
1609 | while (--deferred) | |
c5905afb | 1610 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1611 | return; |
1612 | } | |
1613 | #endif | |
c5905afb | 1614 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1615 | } |
d1b19dff | 1616 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1617 | |
1618 | void net_disable_timestamp(void) | |
1619 | { | |
b90e5794 ED |
1620 | #ifdef HAVE_JUMP_LABEL |
1621 | if (in_interrupt()) { | |
1622 | atomic_inc(&netstamp_needed_deferred); | |
1623 | return; | |
1624 | } | |
1625 | #endif | |
c5905afb | 1626 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1627 | } |
d1b19dff | 1628 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1629 | |
3b098e2d | 1630 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1631 | { |
588f0330 | 1632 | skb->tstamp.tv64 = 0; |
c5905afb | 1633 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1634 | __net_timestamp(skb); |
1da177e4 LT |
1635 | } |
1636 | ||
588f0330 | 1637 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1638 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1639 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1640 | __net_timestamp(SKB); \ | |
1641 | } \ | |
3b098e2d | 1642 | |
1ee481fb | 1643 | bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb) |
79b569f0 DL |
1644 | { |
1645 | unsigned int len; | |
1646 | ||
1647 | if (!(dev->flags & IFF_UP)) | |
1648 | return false; | |
1649 | ||
1650 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1651 | if (skb->len <= len) | |
1652 | return true; | |
1653 | ||
1654 | /* if TSO is enabled, we don't care about the length as the packet | |
1655 | * could be forwarded without being segmented before | |
1656 | */ | |
1657 | if (skb_is_gso(skb)) | |
1658 | return true; | |
1659 | ||
1660 | return false; | |
1661 | } | |
1ee481fb | 1662 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1663 | |
a0265d28 HX |
1664 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1665 | { | |
1666 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { | |
1667 | if (skb_copy_ubufs(skb, GFP_ATOMIC)) { | |
1668 | atomic_long_inc(&dev->rx_dropped); | |
1669 | kfree_skb(skb); | |
1670 | return NET_RX_DROP; | |
1671 | } | |
1672 | } | |
1673 | ||
1674 | if (unlikely(!is_skb_forwardable(dev, skb))) { | |
1675 | atomic_long_inc(&dev->rx_dropped); | |
1676 | kfree_skb(skb); | |
1677 | return NET_RX_DROP; | |
1678 | } | |
1679 | ||
1680 | skb_scrub_packet(skb, true); | |
1681 | skb->protocol = eth_type_trans(skb, dev); | |
1682 | ||
1683 | return 0; | |
1684 | } | |
1685 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1686 | ||
44540960 AB |
1687 | /** |
1688 | * dev_forward_skb - loopback an skb to another netif | |
1689 | * | |
1690 | * @dev: destination network device | |
1691 | * @skb: buffer to forward | |
1692 | * | |
1693 | * return values: | |
1694 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1695 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1696 | * |
1697 | * dev_forward_skb can be used for injecting an skb from the | |
1698 | * start_xmit function of one device into the receive queue | |
1699 | * of another device. | |
1700 | * | |
1701 | * The receiving device may be in another namespace, so | |
1702 | * we have to clear all information in the skb that could | |
1703 | * impact namespace isolation. | |
1704 | */ | |
1705 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1706 | { | |
a0265d28 | 1707 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1708 | } |
1709 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1710 | ||
71d9dec2 CG |
1711 | static inline int deliver_skb(struct sk_buff *skb, |
1712 | struct packet_type *pt_prev, | |
1713 | struct net_device *orig_dev) | |
1714 | { | |
1080e512 MT |
1715 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1716 | return -ENOMEM; | |
71d9dec2 CG |
1717 | atomic_inc(&skb->users); |
1718 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1719 | } | |
1720 | ||
c0de08d0 EL |
1721 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1722 | { | |
a3d744e9 | 1723 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1724 | return false; |
1725 | ||
1726 | if (ptype->id_match) | |
1727 | return ptype->id_match(ptype, skb->sk); | |
1728 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1729 | return true; | |
1730 | ||
1731 | return false; | |
1732 | } | |
1733 | ||
1da177e4 LT |
1734 | /* |
1735 | * Support routine. Sends outgoing frames to any network | |
1736 | * taps currently in use. | |
1737 | */ | |
1738 | ||
f6a78bfc | 1739 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1740 | { |
1741 | struct packet_type *ptype; | |
71d9dec2 CG |
1742 | struct sk_buff *skb2 = NULL; |
1743 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1744 | |
1da177e4 LT |
1745 | rcu_read_lock(); |
1746 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1747 | /* Never send packets back to the socket | |
1748 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1749 | */ | |
1750 | if ((ptype->dev == dev || !ptype->dev) && | |
c0de08d0 | 1751 | (!skb_loop_sk(ptype, skb))) { |
71d9dec2 CG |
1752 | if (pt_prev) { |
1753 | deliver_skb(skb2, pt_prev, skb->dev); | |
1754 | pt_prev = ptype; | |
1755 | continue; | |
1756 | } | |
1757 | ||
1758 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1759 | if (!skb2) |
1760 | break; | |
1761 | ||
70978182 ED |
1762 | net_timestamp_set(skb2); |
1763 | ||
1da177e4 LT |
1764 | /* skb->nh should be correctly |
1765 | set by sender, so that the second statement is | |
1766 | just protection against buggy protocols. | |
1767 | */ | |
459a98ed | 1768 | skb_reset_mac_header(skb2); |
1da177e4 | 1769 | |
d56f90a7 | 1770 | if (skb_network_header(skb2) < skb2->data || |
ced14f68 | 1771 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { |
e87cc472 JP |
1772 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", |
1773 | ntohs(skb2->protocol), | |
1774 | dev->name); | |
c1d2bbe1 | 1775 | skb_reset_network_header(skb2); |
1da177e4 LT |
1776 | } |
1777 | ||
b0e380b1 | 1778 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1779 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1780 | pt_prev = ptype; |
1da177e4 LT |
1781 | } |
1782 | } | |
71d9dec2 CG |
1783 | if (pt_prev) |
1784 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1785 | rcu_read_unlock(); |
1786 | } | |
1787 | ||
2c53040f BH |
1788 | /** |
1789 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1790 | * @dev: Network device |
1791 | * @txq: number of queues available | |
1792 | * | |
1793 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1794 | * valid. To resolve this verify the tc mapping remains valid and if | |
1795 | * not NULL the mapping. With no priorities mapping to this | |
1796 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1797 | * is invalid nothing can be done so disable priority mappings. If is | |
1798 | * expected that drivers will fix this mapping if they can before | |
1799 | * calling netif_set_real_num_tx_queues. | |
1800 | */ | |
bb134d22 | 1801 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1802 | { |
1803 | int i; | |
1804 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1805 | ||
1806 | /* If TC0 is invalidated disable TC mapping */ | |
1807 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1808 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1809 | dev->num_tc = 0; |
1810 | return; | |
1811 | } | |
1812 | ||
1813 | /* Invalidated prio to tc mappings set to TC0 */ | |
1814 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1815 | int q = netdev_get_prio_tc_map(dev, i); | |
1816 | ||
1817 | tc = &dev->tc_to_txq[q]; | |
1818 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1819 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1820 | i, q); | |
4f57c087 JF |
1821 | netdev_set_prio_tc_map(dev, i, 0); |
1822 | } | |
1823 | } | |
1824 | } | |
1825 | ||
537c00de AD |
1826 | #ifdef CONFIG_XPS |
1827 | static DEFINE_MUTEX(xps_map_mutex); | |
1828 | #define xmap_dereference(P) \ | |
1829 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1830 | ||
10cdc3f3 AD |
1831 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1832 | int cpu, u16 index) | |
537c00de | 1833 | { |
10cdc3f3 AD |
1834 | struct xps_map *map = NULL; |
1835 | int pos; | |
537c00de | 1836 | |
10cdc3f3 AD |
1837 | if (dev_maps) |
1838 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1839 | |
10cdc3f3 AD |
1840 | for (pos = 0; map && pos < map->len; pos++) { |
1841 | if (map->queues[pos] == index) { | |
537c00de AD |
1842 | if (map->len > 1) { |
1843 | map->queues[pos] = map->queues[--map->len]; | |
1844 | } else { | |
10cdc3f3 | 1845 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1846 | kfree_rcu(map, rcu); |
1847 | map = NULL; | |
1848 | } | |
10cdc3f3 | 1849 | break; |
537c00de | 1850 | } |
537c00de AD |
1851 | } |
1852 | ||
10cdc3f3 AD |
1853 | return map; |
1854 | } | |
1855 | ||
024e9679 | 1856 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
10cdc3f3 AD |
1857 | { |
1858 | struct xps_dev_maps *dev_maps; | |
024e9679 | 1859 | int cpu, i; |
10cdc3f3 AD |
1860 | bool active = false; |
1861 | ||
1862 | mutex_lock(&xps_map_mutex); | |
1863 | dev_maps = xmap_dereference(dev->xps_maps); | |
1864 | ||
1865 | if (!dev_maps) | |
1866 | goto out_no_maps; | |
1867 | ||
1868 | for_each_possible_cpu(cpu) { | |
024e9679 AD |
1869 | for (i = index; i < dev->num_tx_queues; i++) { |
1870 | if (!remove_xps_queue(dev_maps, cpu, i)) | |
1871 | break; | |
1872 | } | |
1873 | if (i == dev->num_tx_queues) | |
10cdc3f3 AD |
1874 | active = true; |
1875 | } | |
1876 | ||
1877 | if (!active) { | |
537c00de AD |
1878 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
1879 | kfree_rcu(dev_maps, rcu); | |
1880 | } | |
1881 | ||
024e9679 AD |
1882 | for (i = index; i < dev->num_tx_queues; i++) |
1883 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), | |
1884 | NUMA_NO_NODE); | |
1885 | ||
537c00de AD |
1886 | out_no_maps: |
1887 | mutex_unlock(&xps_map_mutex); | |
1888 | } | |
1889 | ||
01c5f864 AD |
1890 | static struct xps_map *expand_xps_map(struct xps_map *map, |
1891 | int cpu, u16 index) | |
1892 | { | |
1893 | struct xps_map *new_map; | |
1894 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
1895 | int i, pos; | |
1896 | ||
1897 | for (pos = 0; map && pos < map->len; pos++) { | |
1898 | if (map->queues[pos] != index) | |
1899 | continue; | |
1900 | return map; | |
1901 | } | |
1902 | ||
1903 | /* Need to add queue to this CPU's existing map */ | |
1904 | if (map) { | |
1905 | if (pos < map->alloc_len) | |
1906 | return map; | |
1907 | ||
1908 | alloc_len = map->alloc_len * 2; | |
1909 | } | |
1910 | ||
1911 | /* Need to allocate new map to store queue on this CPU's map */ | |
1912 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
1913 | cpu_to_node(cpu)); | |
1914 | if (!new_map) | |
1915 | return NULL; | |
1916 | ||
1917 | for (i = 0; i < pos; i++) | |
1918 | new_map->queues[i] = map->queues[i]; | |
1919 | new_map->alloc_len = alloc_len; | |
1920 | new_map->len = pos; | |
1921 | ||
1922 | return new_map; | |
1923 | } | |
1924 | ||
3573540c MT |
1925 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
1926 | u16 index) | |
537c00de | 1927 | { |
01c5f864 | 1928 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 1929 | struct xps_map *map, *new_map; |
537c00de | 1930 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
1931 | int cpu, numa_node_id = -2; |
1932 | bool active = false; | |
537c00de AD |
1933 | |
1934 | mutex_lock(&xps_map_mutex); | |
1935 | ||
1936 | dev_maps = xmap_dereference(dev->xps_maps); | |
1937 | ||
01c5f864 AD |
1938 | /* allocate memory for queue storage */ |
1939 | for_each_online_cpu(cpu) { | |
1940 | if (!cpumask_test_cpu(cpu, mask)) | |
1941 | continue; | |
1942 | ||
1943 | if (!new_dev_maps) | |
1944 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
1945 | if (!new_dev_maps) { |
1946 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 1947 | return -ENOMEM; |
2bb60cb9 | 1948 | } |
01c5f864 AD |
1949 | |
1950 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
1951 | NULL; | |
1952 | ||
1953 | map = expand_xps_map(map, cpu, index); | |
1954 | if (!map) | |
1955 | goto error; | |
1956 | ||
1957 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
1958 | } | |
1959 | ||
1960 | if (!new_dev_maps) | |
1961 | goto out_no_new_maps; | |
1962 | ||
537c00de | 1963 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
1964 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
1965 | /* add queue to CPU maps */ | |
1966 | int pos = 0; | |
1967 | ||
1968 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1969 | while ((pos < map->len) && (map->queues[pos] != index)) | |
1970 | pos++; | |
1971 | ||
1972 | if (pos == map->len) | |
1973 | map->queues[map->len++] = index; | |
537c00de | 1974 | #ifdef CONFIG_NUMA |
537c00de AD |
1975 | if (numa_node_id == -2) |
1976 | numa_node_id = cpu_to_node(cpu); | |
1977 | else if (numa_node_id != cpu_to_node(cpu)) | |
1978 | numa_node_id = -1; | |
537c00de | 1979 | #endif |
01c5f864 AD |
1980 | } else if (dev_maps) { |
1981 | /* fill in the new device map from the old device map */ | |
1982 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
1983 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 1984 | } |
01c5f864 | 1985 | |
537c00de AD |
1986 | } |
1987 | ||
01c5f864 AD |
1988 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
1989 | ||
537c00de | 1990 | /* Cleanup old maps */ |
01c5f864 AD |
1991 | if (dev_maps) { |
1992 | for_each_possible_cpu(cpu) { | |
1993 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1994 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
1995 | if (map && map != new_map) | |
1996 | kfree_rcu(map, rcu); | |
1997 | } | |
537c00de | 1998 | |
01c5f864 | 1999 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
2000 | } |
2001 | ||
01c5f864 AD |
2002 | dev_maps = new_dev_maps; |
2003 | active = true; | |
537c00de | 2004 | |
01c5f864 AD |
2005 | out_no_new_maps: |
2006 | /* update Tx queue numa node */ | |
537c00de AD |
2007 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2008 | (numa_node_id >= 0) ? numa_node_id : | |
2009 | NUMA_NO_NODE); | |
2010 | ||
01c5f864 AD |
2011 | if (!dev_maps) |
2012 | goto out_no_maps; | |
2013 | ||
2014 | /* removes queue from unused CPUs */ | |
2015 | for_each_possible_cpu(cpu) { | |
2016 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
2017 | continue; | |
2018 | ||
2019 | if (remove_xps_queue(dev_maps, cpu, index)) | |
2020 | active = true; | |
2021 | } | |
2022 | ||
2023 | /* free map if not active */ | |
2024 | if (!active) { | |
2025 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2026 | kfree_rcu(dev_maps, rcu); | |
2027 | } | |
2028 | ||
2029 | out_no_maps: | |
537c00de AD |
2030 | mutex_unlock(&xps_map_mutex); |
2031 | ||
2032 | return 0; | |
2033 | error: | |
01c5f864 AD |
2034 | /* remove any maps that we added */ |
2035 | for_each_possible_cpu(cpu) { | |
2036 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2037 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2038 | NULL; | |
2039 | if (new_map && new_map != map) | |
2040 | kfree(new_map); | |
2041 | } | |
2042 | ||
537c00de AD |
2043 | mutex_unlock(&xps_map_mutex); |
2044 | ||
537c00de AD |
2045 | kfree(new_dev_maps); |
2046 | return -ENOMEM; | |
2047 | } | |
2048 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2049 | ||
2050 | #endif | |
f0796d5c JF |
2051 | /* |
2052 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2053 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2054 | */ | |
e6484930 | 2055 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2056 | { |
1d24eb48 TH |
2057 | int rc; |
2058 | ||
e6484930 TH |
2059 | if (txq < 1 || txq > dev->num_tx_queues) |
2060 | return -EINVAL; | |
f0796d5c | 2061 | |
5c56580b BH |
2062 | if (dev->reg_state == NETREG_REGISTERED || |
2063 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2064 | ASSERT_RTNL(); |
2065 | ||
1d24eb48 TH |
2066 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2067 | txq); | |
bf264145 TH |
2068 | if (rc) |
2069 | return rc; | |
2070 | ||
4f57c087 JF |
2071 | if (dev->num_tc) |
2072 | netif_setup_tc(dev, txq); | |
2073 | ||
024e9679 | 2074 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2075 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2076 | #ifdef CONFIG_XPS |
2077 | netif_reset_xps_queues_gt(dev, txq); | |
2078 | #endif | |
2079 | } | |
f0796d5c | 2080 | } |
e6484930 TH |
2081 | |
2082 | dev->real_num_tx_queues = txq; | |
2083 | return 0; | |
f0796d5c JF |
2084 | } |
2085 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2086 | |
a953be53 | 2087 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2088 | /** |
2089 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2090 | * @dev: Network device | |
2091 | * @rxq: Actual number of RX queues | |
2092 | * | |
2093 | * This must be called either with the rtnl_lock held or before | |
2094 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2095 | * negative error code. If called before registration, it always |
2096 | * succeeds. | |
62fe0b40 BH |
2097 | */ |
2098 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2099 | { | |
2100 | int rc; | |
2101 | ||
bd25fa7b TH |
2102 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2103 | return -EINVAL; | |
2104 | ||
62fe0b40 BH |
2105 | if (dev->reg_state == NETREG_REGISTERED) { |
2106 | ASSERT_RTNL(); | |
2107 | ||
62fe0b40 BH |
2108 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2109 | rxq); | |
2110 | if (rc) | |
2111 | return rc; | |
62fe0b40 BH |
2112 | } |
2113 | ||
2114 | dev->real_num_rx_queues = rxq; | |
2115 | return 0; | |
2116 | } | |
2117 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2118 | #endif | |
2119 | ||
2c53040f BH |
2120 | /** |
2121 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2122 | * |
2123 | * This routine should set an upper limit on the number of RSS queues | |
2124 | * used by default by multiqueue devices. | |
2125 | */ | |
a55b138b | 2126 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
2127 | { |
2128 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
2129 | } | |
2130 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2131 | ||
def82a1d | 2132 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 2133 | { |
def82a1d JP |
2134 | struct softnet_data *sd; |
2135 | unsigned long flags; | |
56079431 | 2136 | |
def82a1d JP |
2137 | local_irq_save(flags); |
2138 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
2139 | q->next_sched = NULL; |
2140 | *sd->output_queue_tailp = q; | |
2141 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2142 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2143 | local_irq_restore(flags); | |
2144 | } | |
2145 | ||
2146 | void __netif_schedule(struct Qdisc *q) | |
2147 | { | |
2148 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2149 | __netif_reschedule(q); | |
56079431 DV |
2150 | } |
2151 | EXPORT_SYMBOL(__netif_schedule); | |
2152 | ||
e6247027 ED |
2153 | struct dev_kfree_skb_cb { |
2154 | enum skb_free_reason reason; | |
2155 | }; | |
2156 | ||
2157 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2158 | { |
e6247027 ED |
2159 | return (struct dev_kfree_skb_cb *)skb->cb; |
2160 | } | |
2161 | ||
2162 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) | |
56079431 | 2163 | { |
e6247027 | 2164 | unsigned long flags; |
56079431 | 2165 | |
e6247027 ED |
2166 | if (likely(atomic_read(&skb->users) == 1)) { |
2167 | smp_rmb(); | |
2168 | atomic_set(&skb->users, 0); | |
2169 | } else if (likely(!atomic_dec_and_test(&skb->users))) { | |
2170 | return; | |
bea3348e | 2171 | } |
e6247027 ED |
2172 | get_kfree_skb_cb(skb)->reason = reason; |
2173 | local_irq_save(flags); | |
2174 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2175 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2176 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2177 | local_irq_restore(flags); | |
56079431 | 2178 | } |
e6247027 | 2179 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2180 | |
e6247027 | 2181 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2182 | { |
2183 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2184 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2185 | else |
2186 | dev_kfree_skb(skb); | |
2187 | } | |
e6247027 | 2188 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2189 | |
2190 | ||
bea3348e SH |
2191 | /** |
2192 | * netif_device_detach - mark device as removed | |
2193 | * @dev: network device | |
2194 | * | |
2195 | * Mark device as removed from system and therefore no longer available. | |
2196 | */ | |
56079431 DV |
2197 | void netif_device_detach(struct net_device *dev) |
2198 | { | |
2199 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2200 | netif_running(dev)) { | |
d543103a | 2201 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2202 | } |
2203 | } | |
2204 | EXPORT_SYMBOL(netif_device_detach); | |
2205 | ||
bea3348e SH |
2206 | /** |
2207 | * netif_device_attach - mark device as attached | |
2208 | * @dev: network device | |
2209 | * | |
2210 | * Mark device as attached from system and restart if needed. | |
2211 | */ | |
56079431 DV |
2212 | void netif_device_attach(struct net_device *dev) |
2213 | { | |
2214 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2215 | netif_running(dev)) { | |
d543103a | 2216 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2217 | __netdev_watchdog_up(dev); |
56079431 DV |
2218 | } |
2219 | } | |
2220 | EXPORT_SYMBOL(netif_device_attach); | |
2221 | ||
36c92474 BH |
2222 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2223 | { | |
65e9d2fa | 2224 | static const netdev_features_t null_features = 0; |
36c92474 BH |
2225 | struct net_device *dev = skb->dev; |
2226 | const char *driver = ""; | |
2227 | ||
c846ad9b BG |
2228 | if (!net_ratelimit()) |
2229 | return; | |
2230 | ||
36c92474 BH |
2231 | if (dev && dev->dev.parent) |
2232 | driver = dev_driver_string(dev->dev.parent); | |
2233 | ||
2234 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " | |
2235 | "gso_type=%d ip_summed=%d\n", | |
65e9d2fa MM |
2236 | driver, dev ? &dev->features : &null_features, |
2237 | skb->sk ? &skb->sk->sk_route_caps : &null_features, | |
36c92474 BH |
2238 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2239 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2240 | } | |
2241 | ||
1da177e4 LT |
2242 | /* |
2243 | * Invalidate hardware checksum when packet is to be mangled, and | |
2244 | * complete checksum manually on outgoing path. | |
2245 | */ | |
84fa7933 | 2246 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2247 | { |
d3bc23e7 | 2248 | __wsum csum; |
663ead3b | 2249 | int ret = 0, offset; |
1da177e4 | 2250 | |
84fa7933 | 2251 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2252 | goto out_set_summed; |
2253 | ||
2254 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2255 | skb_warn_bad_offload(skb); |
2256 | return -EINVAL; | |
1da177e4 LT |
2257 | } |
2258 | ||
cef401de ED |
2259 | /* Before computing a checksum, we should make sure no frag could |
2260 | * be modified by an external entity : checksum could be wrong. | |
2261 | */ | |
2262 | if (skb_has_shared_frag(skb)) { | |
2263 | ret = __skb_linearize(skb); | |
2264 | if (ret) | |
2265 | goto out; | |
2266 | } | |
2267 | ||
55508d60 | 2268 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2269 | BUG_ON(offset >= skb_headlen(skb)); |
2270 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2271 | ||
2272 | offset += skb->csum_offset; | |
2273 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2274 | ||
2275 | if (skb_cloned(skb) && | |
2276 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2277 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2278 | if (ret) | |
2279 | goto out; | |
2280 | } | |
2281 | ||
a030847e | 2282 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2283 | out_set_summed: |
1da177e4 | 2284 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2285 | out: |
1da177e4 LT |
2286 | return ret; |
2287 | } | |
d1b19dff | 2288 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2289 | |
53d6471c | 2290 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2291 | { |
252e3346 | 2292 | __be16 type = skb->protocol; |
1e785f48 | 2293 | int vlan_depth = skb->mac_len; |
f6a78bfc | 2294 | |
19acc327 PS |
2295 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2296 | if (type == htons(ETH_P_TEB)) { | |
2297 | struct ethhdr *eth; | |
2298 | ||
2299 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2300 | return 0; | |
2301 | ||
2302 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2303 | type = eth->h_proto; | |
2304 | } | |
2305 | ||
8ad227ff | 2306 | while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) { |
c8d5bcd1 | 2307 | struct vlan_hdr *vh; |
7b9c6090 | 2308 | |
c8d5bcd1 | 2309 | if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN))) |
ec5f0615 | 2310 | return 0; |
7b9c6090 | 2311 | |
c8d5bcd1 JG |
2312 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); |
2313 | type = vh->h_vlan_encapsulated_proto; | |
2314 | vlan_depth += VLAN_HLEN; | |
7b9c6090 JG |
2315 | } |
2316 | ||
53d6471c VY |
2317 | *depth = vlan_depth; |
2318 | ||
ec5f0615 PS |
2319 | return type; |
2320 | } | |
2321 | ||
2322 | /** | |
2323 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2324 | * @skb: buffer to segment | |
2325 | * @features: features for the output path (see dev->features) | |
2326 | */ | |
2327 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2328 | netdev_features_t features) | |
2329 | { | |
2330 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2331 | struct packet_offload *ptype; | |
53d6471c VY |
2332 | int vlan_depth = skb->mac_len; |
2333 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2334 | |
2335 | if (unlikely(!type)) | |
2336 | return ERR_PTR(-EINVAL); | |
2337 | ||
53d6471c | 2338 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2339 | |
2340 | rcu_read_lock(); | |
22061d80 | 2341 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2342 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
84fa7933 | 2343 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
05e8ef4a PS |
2344 | int err; |
2345 | ||
f191a1d1 | 2346 | err = ptype->callbacks.gso_send_check(skb); |
a430a43d HX |
2347 | segs = ERR_PTR(err); |
2348 | if (err || skb_gso_ok(skb, features)) | |
2349 | break; | |
d56f90a7 ACM |
2350 | __skb_push(skb, (skb->data - |
2351 | skb_network_header(skb))); | |
a430a43d | 2352 | } |
f191a1d1 | 2353 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2354 | break; |
2355 | } | |
2356 | } | |
2357 | rcu_read_unlock(); | |
2358 | ||
98e399f8 | 2359 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2360 | |
f6a78bfc HX |
2361 | return segs; |
2362 | } | |
05e8ef4a PS |
2363 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2364 | ||
2365 | ||
2366 | /* openvswitch calls this on rx path, so we need a different check. | |
2367 | */ | |
2368 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2369 | { | |
2370 | if (tx_path) | |
2371 | return skb->ip_summed != CHECKSUM_PARTIAL; | |
2372 | else | |
2373 | return skb->ip_summed == CHECKSUM_NONE; | |
2374 | } | |
2375 | ||
2376 | /** | |
2377 | * __skb_gso_segment - Perform segmentation on skb. | |
2378 | * @skb: buffer to segment | |
2379 | * @features: features for the output path (see dev->features) | |
2380 | * @tx_path: whether it is called in TX path | |
2381 | * | |
2382 | * This function segments the given skb and returns a list of segments. | |
2383 | * | |
2384 | * It may return NULL if the skb requires no segmentation. This is | |
2385 | * only possible when GSO is used for verifying header integrity. | |
2386 | */ | |
2387 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2388 | netdev_features_t features, bool tx_path) | |
2389 | { | |
2390 | if (unlikely(skb_needs_check(skb, tx_path))) { | |
2391 | int err; | |
2392 | ||
2393 | skb_warn_bad_offload(skb); | |
2394 | ||
2395 | if (skb_header_cloned(skb) && | |
2396 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
2397 | return ERR_PTR(err); | |
2398 | } | |
2399 | ||
68c33163 | 2400 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2401 | SKB_GSO_CB(skb)->encap_level = 0; |
2402 | ||
05e8ef4a PS |
2403 | skb_reset_mac_header(skb); |
2404 | skb_reset_mac_len(skb); | |
2405 | ||
2406 | return skb_mac_gso_segment(skb, features); | |
2407 | } | |
12b0004d | 2408 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2409 | |
fb286bb2 HX |
2410 | /* Take action when hardware reception checksum errors are detected. */ |
2411 | #ifdef CONFIG_BUG | |
2412 | void netdev_rx_csum_fault(struct net_device *dev) | |
2413 | { | |
2414 | if (net_ratelimit()) { | |
7b6cd1ce | 2415 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2416 | dump_stack(); |
2417 | } | |
2418 | } | |
2419 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2420 | #endif | |
2421 | ||
1da177e4 LT |
2422 | /* Actually, we should eliminate this check as soon as we know, that: |
2423 | * 1. IOMMU is present and allows to map all the memory. | |
2424 | * 2. No high memory really exists on this machine. | |
2425 | */ | |
2426 | ||
c1e756bf | 2427 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2428 | { |
3d3a8533 | 2429 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2430 | int i; |
5acbbd42 | 2431 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2432 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2433 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2434 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2435 | return 1; |
ea2ab693 | 2436 | } |
5acbbd42 | 2437 | } |
1da177e4 | 2438 | |
5acbbd42 FT |
2439 | if (PCI_DMA_BUS_IS_PHYS) { |
2440 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2441 | |
9092c658 ED |
2442 | if (!pdev) |
2443 | return 0; | |
5acbbd42 | 2444 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2445 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2446 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2447 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2448 | return 1; | |
2449 | } | |
2450 | } | |
3d3a8533 | 2451 | #endif |
1da177e4 LT |
2452 | return 0; |
2453 | } | |
1da177e4 | 2454 | |
f6a78bfc HX |
2455 | struct dev_gso_cb { |
2456 | void (*destructor)(struct sk_buff *skb); | |
2457 | }; | |
2458 | ||
2459 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
2460 | ||
2461 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
2462 | { | |
2463 | struct dev_gso_cb *cb; | |
2464 | ||
289dccbe ED |
2465 | kfree_skb_list(skb->next); |
2466 | skb->next = NULL; | |
f6a78bfc HX |
2467 | |
2468 | cb = DEV_GSO_CB(skb); | |
2469 | if (cb->destructor) | |
2470 | cb->destructor(skb); | |
2471 | } | |
2472 | ||
2473 | /** | |
2474 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
2475 | * @skb: buffer to segment | |
91ecb63c | 2476 | * @features: device features as applicable to this skb |
f6a78bfc HX |
2477 | * |
2478 | * This function segments the given skb and stores the list of segments | |
2479 | * in skb->next. | |
2480 | */ | |
c8f44aff | 2481 | static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features) |
f6a78bfc | 2482 | { |
f6a78bfc | 2483 | struct sk_buff *segs; |
576a30eb HX |
2484 | |
2485 | segs = skb_gso_segment(skb, features); | |
2486 | ||
2487 | /* Verifying header integrity only. */ | |
2488 | if (!segs) | |
2489 | return 0; | |
f6a78bfc | 2490 | |
801678c5 | 2491 | if (IS_ERR(segs)) |
f6a78bfc HX |
2492 | return PTR_ERR(segs); |
2493 | ||
2494 | skb->next = segs; | |
2495 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
2496 | skb->destructor = dev_gso_skb_destructor; | |
2497 | ||
2498 | return 0; | |
2499 | } | |
2500 | ||
c8f44aff | 2501 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2502 | netdev_features_t features) |
f01a5236 | 2503 | { |
53d6471c VY |
2504 | int tmp; |
2505 | ||
c0d680e5 | 2506 | if (skb->ip_summed != CHECKSUM_NONE && |
53d6471c | 2507 | !can_checksum_protocol(features, skb_network_protocol(skb, &tmp))) { |
f01a5236 | 2508 | features &= ~NETIF_F_ALL_CSUM; |
c1e756bf | 2509 | } else if (illegal_highdma(skb->dev, skb)) { |
f01a5236 JG |
2510 | features &= ~NETIF_F_SG; |
2511 | } | |
2512 | ||
2513 | return features; | |
2514 | } | |
2515 | ||
c1e756bf | 2516 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 JG |
2517 | { |
2518 | __be16 protocol = skb->protocol; | |
c1e756bf | 2519 | netdev_features_t features = skb->dev->features; |
58e998c6 | 2520 | |
c1e756bf | 2521 | if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs) |
30b678d8 BH |
2522 | features &= ~NETIF_F_GSO_MASK; |
2523 | ||
8ad227ff | 2524 | if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) { |
58e998c6 JG |
2525 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; |
2526 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 | 2527 | } else if (!vlan_tx_tag_present(skb)) { |
c1e756bf | 2528 | return harmonize_features(skb, features); |
f01a5236 | 2529 | } |
58e998c6 | 2530 | |
c1e756bf | 2531 | features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX | |
8ad227ff | 2532 | NETIF_F_HW_VLAN_STAG_TX); |
f01a5236 | 2533 | |
cdbaa0bb | 2534 | if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) |
f01a5236 | 2535 | features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | |
8ad227ff PM |
2536 | NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX | |
2537 | NETIF_F_HW_VLAN_STAG_TX; | |
cdbaa0bb | 2538 | |
c1e756bf | 2539 | return harmonize_features(skb, features); |
58e998c6 | 2540 | } |
c1e756bf | 2541 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2542 | |
fd2ea0a7 | 2543 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
f663dd9a | 2544 | struct netdev_queue *txq) |
f6a78bfc | 2545 | { |
00829823 | 2546 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 2547 | int rc = NETDEV_TX_OK; |
ec764bf0 | 2548 | unsigned int skb_len; |
00829823 | 2549 | |
f6a78bfc | 2550 | if (likely(!skb->next)) { |
c8f44aff | 2551 | netdev_features_t features; |
fc741216 | 2552 | |
93f154b5 | 2553 | /* |
25985edc | 2554 | * If device doesn't need skb->dst, release it right now while |
93f154b5 ED |
2555 | * its hot in this cpu cache |
2556 | */ | |
adf30907 ED |
2557 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
2558 | skb_dst_drop(skb); | |
2559 | ||
fc741216 JG |
2560 | features = netif_skb_features(skb); |
2561 | ||
7b9c6090 | 2562 | if (vlan_tx_tag_present(skb) && |
86a9bad3 PM |
2563 | !vlan_hw_offload_capable(features, skb->vlan_proto)) { |
2564 | skb = __vlan_put_tag(skb, skb->vlan_proto, | |
2565 | vlan_tx_tag_get(skb)); | |
7b9c6090 JG |
2566 | if (unlikely(!skb)) |
2567 | goto out; | |
2568 | ||
2569 | skb->vlan_tci = 0; | |
2570 | } | |
2571 | ||
fc70fb64 AD |
2572 | /* If encapsulation offload request, verify we are testing |
2573 | * hardware encapsulation features instead of standard | |
2574 | * features for the netdev | |
2575 | */ | |
2576 | if (skb->encapsulation) | |
2577 | features &= dev->hw_enc_features; | |
2578 | ||
fc741216 | 2579 | if (netif_needs_gso(skb, features)) { |
91ecb63c | 2580 | if (unlikely(dev_gso_segment(skb, features))) |
9ccb8975 DM |
2581 | goto out_kfree_skb; |
2582 | if (skb->next) | |
2583 | goto gso; | |
6afff0ca | 2584 | } else { |
02932ce9 | 2585 | if (skb_needs_linearize(skb, features) && |
6afff0ca JF |
2586 | __skb_linearize(skb)) |
2587 | goto out_kfree_skb; | |
2588 | ||
2589 | /* If packet is not checksummed and device does not | |
2590 | * support checksumming for this protocol, complete | |
2591 | * checksumming here. | |
2592 | */ | |
2593 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
fc70fb64 AD |
2594 | if (skb->encapsulation) |
2595 | skb_set_inner_transport_header(skb, | |
2596 | skb_checksum_start_offset(skb)); | |
2597 | else | |
2598 | skb_set_transport_header(skb, | |
2599 | skb_checksum_start_offset(skb)); | |
03634668 | 2600 | if (!(features & NETIF_F_ALL_CSUM) && |
6afff0ca JF |
2601 | skb_checksum_help(skb)) |
2602 | goto out_kfree_skb; | |
2603 | } | |
9ccb8975 DM |
2604 | } |
2605 | ||
b40863c6 ED |
2606 | if (!list_empty(&ptype_all)) |
2607 | dev_queue_xmit_nit(skb, dev); | |
2608 | ||
ec764bf0 | 2609 | skb_len = skb->len; |
d87d04a7 | 2610 | trace_net_dev_start_xmit(skb, dev); |
20567661 | 2611 | rc = ops->ndo_start_xmit(skb, dev); |
ec764bf0 | 2612 | trace_net_dev_xmit(skb, rc, dev, skb_len); |
f663dd9a | 2613 | if (rc == NETDEV_TX_OK) |
08baf561 | 2614 | txq_trans_update(txq); |
ac45f602 | 2615 | return rc; |
f6a78bfc HX |
2616 | } |
2617 | ||
576a30eb | 2618 | gso: |
f6a78bfc HX |
2619 | do { |
2620 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
2621 | |
2622 | skb->next = nskb->next; | |
2623 | nskb->next = NULL; | |
068a2de5 | 2624 | |
b40863c6 ED |
2625 | if (!list_empty(&ptype_all)) |
2626 | dev_queue_xmit_nit(nskb, dev); | |
2627 | ||
ec764bf0 | 2628 | skb_len = nskb->len; |
d87d04a7 | 2629 | trace_net_dev_start_xmit(nskb, dev); |
f663dd9a | 2630 | rc = ops->ndo_start_xmit(nskb, dev); |
ec764bf0 | 2631 | trace_net_dev_xmit(nskb, rc, dev, skb_len); |
ec634fe3 | 2632 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
2633 | if (rc & ~NETDEV_TX_MASK) |
2634 | goto out_kfree_gso_skb; | |
f54d9e8d | 2635 | nskb->next = skb->next; |
f6a78bfc HX |
2636 | skb->next = nskb; |
2637 | return rc; | |
2638 | } | |
08baf561 | 2639 | txq_trans_update(txq); |
73466498 | 2640 | if (unlikely(netif_xmit_stopped(txq) && skb->next)) |
f54d9e8d | 2641 | return NETDEV_TX_BUSY; |
f6a78bfc | 2642 | } while (skb->next); |
4ec93edb | 2643 | |
572a9d7b | 2644 | out_kfree_gso_skb: |
0c772159 | 2645 | if (likely(skb->next == NULL)) { |
572a9d7b | 2646 | skb->destructor = DEV_GSO_CB(skb)->destructor; |
0c772159 SS |
2647 | consume_skb(skb); |
2648 | return rc; | |
2649 | } | |
f6a78bfc HX |
2650 | out_kfree_skb: |
2651 | kfree_skb(skb); | |
7b9c6090 | 2652 | out: |
572a9d7b | 2653 | return rc; |
f6a78bfc | 2654 | } |
a6cc0cfa | 2655 | EXPORT_SYMBOL_GPL(dev_hard_start_xmit); |
f6a78bfc | 2656 | |
1def9238 ED |
2657 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
2658 | { | |
2659 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2660 | ||
2661 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2662 | ||
2663 | /* To get more precise estimation of bytes sent on wire, | |
2664 | * we add to pkt_len the headers size of all segments | |
2665 | */ | |
2666 | if (shinfo->gso_size) { | |
757b8b1d | 2667 | unsigned int hdr_len; |
15e5a030 | 2668 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 2669 | |
757b8b1d ED |
2670 | /* mac layer + network layer */ |
2671 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
2672 | ||
2673 | /* + transport layer */ | |
1def9238 ED |
2674 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
2675 | hdr_len += tcp_hdrlen(skb); | |
2676 | else | |
2677 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
2678 | |
2679 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
2680 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
2681 | shinfo->gso_size); | |
2682 | ||
2683 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
2684 | } |
2685 | } | |
2686 | ||
bbd8a0d3 KK |
2687 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2688 | struct net_device *dev, | |
2689 | struct netdev_queue *txq) | |
2690 | { | |
2691 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2692 | bool contended; |
bbd8a0d3 KK |
2693 | int rc; |
2694 | ||
1def9238 | 2695 | qdisc_pkt_len_init(skb); |
a2da570d | 2696 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
2697 | /* |
2698 | * Heuristic to force contended enqueues to serialize on a | |
2699 | * separate lock before trying to get qdisc main lock. | |
2700 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2701 | * and dequeue packets faster. | |
2702 | */ | |
a2da570d | 2703 | contended = qdisc_is_running(q); |
79640a4c ED |
2704 | if (unlikely(contended)) |
2705 | spin_lock(&q->busylock); | |
2706 | ||
bbd8a0d3 KK |
2707 | spin_lock(root_lock); |
2708 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2709 | kfree_skb(skb); | |
2710 | rc = NET_XMIT_DROP; | |
2711 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2712 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2713 | /* |
2714 | * This is a work-conserving queue; there are no old skbs | |
2715 | * waiting to be sent out; and the qdisc is not running - | |
2716 | * xmit the skb directly. | |
2717 | */ | |
7fee226a ED |
2718 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2719 | skb_dst_force(skb); | |
bfe0d029 | 2720 | |
bfe0d029 ED |
2721 | qdisc_bstats_update(q, skb); |
2722 | ||
79640a4c ED |
2723 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2724 | if (unlikely(contended)) { | |
2725 | spin_unlock(&q->busylock); | |
2726 | contended = false; | |
2727 | } | |
bbd8a0d3 | 2728 | __qdisc_run(q); |
79640a4c | 2729 | } else |
bc135b23 | 2730 | qdisc_run_end(q); |
bbd8a0d3 KK |
2731 | |
2732 | rc = NET_XMIT_SUCCESS; | |
2733 | } else { | |
7fee226a | 2734 | skb_dst_force(skb); |
a2da570d | 2735 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2736 | if (qdisc_run_begin(q)) { |
2737 | if (unlikely(contended)) { | |
2738 | spin_unlock(&q->busylock); | |
2739 | contended = false; | |
2740 | } | |
2741 | __qdisc_run(q); | |
2742 | } | |
bbd8a0d3 KK |
2743 | } |
2744 | spin_unlock(root_lock); | |
79640a4c ED |
2745 | if (unlikely(contended)) |
2746 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2747 | return rc; |
2748 | } | |
2749 | ||
86f8515f | 2750 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
2751 | static void skb_update_prio(struct sk_buff *skb) |
2752 | { | |
6977a79d | 2753 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 2754 | |
91c68ce2 ED |
2755 | if (!skb->priority && skb->sk && map) { |
2756 | unsigned int prioidx = skb->sk->sk_cgrp_prioidx; | |
2757 | ||
2758 | if (prioidx < map->priomap_len) | |
2759 | skb->priority = map->priomap[prioidx]; | |
2760 | } | |
5bc1421e NH |
2761 | } |
2762 | #else | |
2763 | #define skb_update_prio(skb) | |
2764 | #endif | |
2765 | ||
745e20f1 | 2766 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2767 | #define RECURSION_LIMIT 10 |
745e20f1 | 2768 | |
95603e22 MM |
2769 | /** |
2770 | * dev_loopback_xmit - loop back @skb | |
2771 | * @skb: buffer to transmit | |
2772 | */ | |
2773 | int dev_loopback_xmit(struct sk_buff *skb) | |
2774 | { | |
2775 | skb_reset_mac_header(skb); | |
2776 | __skb_pull(skb, skb_network_offset(skb)); | |
2777 | skb->pkt_type = PACKET_LOOPBACK; | |
2778 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2779 | WARN_ON(!skb_dst(skb)); | |
2780 | skb_dst_force(skb); | |
2781 | netif_rx_ni(skb); | |
2782 | return 0; | |
2783 | } | |
2784 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2785 | ||
d29f749e | 2786 | /** |
9d08dd3d | 2787 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 2788 | * @skb: buffer to transmit |
9d08dd3d | 2789 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
2790 | * |
2791 | * Queue a buffer for transmission to a network device. The caller must | |
2792 | * have set the device and priority and built the buffer before calling | |
2793 | * this function. The function can be called from an interrupt. | |
2794 | * | |
2795 | * A negative errno code is returned on a failure. A success does not | |
2796 | * guarantee the frame will be transmitted as it may be dropped due | |
2797 | * to congestion or traffic shaping. | |
2798 | * | |
2799 | * ----------------------------------------------------------------------------------- | |
2800 | * I notice this method can also return errors from the queue disciplines, | |
2801 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2802 | * be positive. | |
2803 | * | |
2804 | * Regardless of the return value, the skb is consumed, so it is currently | |
2805 | * difficult to retry a send to this method. (You can bump the ref count | |
2806 | * before sending to hold a reference for retry if you are careful.) | |
2807 | * | |
2808 | * When calling this method, interrupts MUST be enabled. This is because | |
2809 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2810 | * --BLG | |
2811 | */ | |
0a59f3a9 | 2812 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
2813 | { |
2814 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2815 | struct netdev_queue *txq; |
1da177e4 LT |
2816 | struct Qdisc *q; |
2817 | int rc = -ENOMEM; | |
2818 | ||
6d1ccff6 ED |
2819 | skb_reset_mac_header(skb); |
2820 | ||
4ec93edb YH |
2821 | /* Disable soft irqs for various locks below. Also |
2822 | * stops preemption for RCU. | |
1da177e4 | 2823 | */ |
4ec93edb | 2824 | rcu_read_lock_bh(); |
1da177e4 | 2825 | |
5bc1421e NH |
2826 | skb_update_prio(skb); |
2827 | ||
f663dd9a | 2828 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 2829 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2830 | |
1da177e4 | 2831 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2832 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2833 | #endif |
cf66ba58 | 2834 | trace_net_dev_queue(skb); |
1da177e4 | 2835 | if (q->enqueue) { |
bbd8a0d3 | 2836 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2837 | goto out; |
1da177e4 LT |
2838 | } |
2839 | ||
2840 | /* The device has no queue. Common case for software devices: | |
2841 | loopback, all the sorts of tunnels... | |
2842 | ||
932ff279 HX |
2843 | Really, it is unlikely that netif_tx_lock protection is necessary |
2844 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2845 | counters.) |
2846 | However, it is possible, that they rely on protection | |
2847 | made by us here. | |
2848 | ||
2849 | Check this and shot the lock. It is not prone from deadlocks. | |
2850 | Either shot noqueue qdisc, it is even simpler 8) | |
2851 | */ | |
2852 | if (dev->flags & IFF_UP) { | |
2853 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2854 | ||
c773e847 | 2855 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2856 | |
745e20f1 ED |
2857 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2858 | goto recursion_alert; | |
2859 | ||
c773e847 | 2860 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2861 | |
73466498 | 2862 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 2863 | __this_cpu_inc(xmit_recursion); |
f663dd9a | 2864 | rc = dev_hard_start_xmit(skb, dev, txq); |
745e20f1 | 2865 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2866 | if (dev_xmit_complete(rc)) { |
c773e847 | 2867 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2868 | goto out; |
2869 | } | |
2870 | } | |
c773e847 | 2871 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
2872 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
2873 | dev->name); | |
1da177e4 LT |
2874 | } else { |
2875 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2876 | * unfortunately |
2877 | */ | |
2878 | recursion_alert: | |
e87cc472 JP |
2879 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
2880 | dev->name); | |
1da177e4 LT |
2881 | } |
2882 | } | |
2883 | ||
2884 | rc = -ENETDOWN; | |
d4828d85 | 2885 | rcu_read_unlock_bh(); |
1da177e4 | 2886 | |
015f0688 | 2887 | atomic_long_inc(&dev->tx_dropped); |
1da177e4 LT |
2888 | kfree_skb(skb); |
2889 | return rc; | |
2890 | out: | |
d4828d85 | 2891 | rcu_read_unlock_bh(); |
1da177e4 LT |
2892 | return rc; |
2893 | } | |
f663dd9a JW |
2894 | |
2895 | int dev_queue_xmit(struct sk_buff *skb) | |
2896 | { | |
2897 | return __dev_queue_xmit(skb, NULL); | |
2898 | } | |
d1b19dff | 2899 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 2900 | |
f663dd9a JW |
2901 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
2902 | { | |
2903 | return __dev_queue_xmit(skb, accel_priv); | |
2904 | } | |
2905 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
2906 | ||
1da177e4 LT |
2907 | |
2908 | /*======================================================================= | |
2909 | Receiver routines | |
2910 | =======================================================================*/ | |
2911 | ||
6b2bedc3 | 2912 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
2913 | EXPORT_SYMBOL(netdev_max_backlog); |
2914 | ||
3b098e2d | 2915 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2916 | int netdev_budget __read_mostly = 300; |
2917 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2918 | |
eecfd7c4 ED |
2919 | /* Called with irq disabled */ |
2920 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2921 | struct napi_struct *napi) | |
2922 | { | |
2923 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2924 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2925 | } | |
2926 | ||
bfb564e7 KK |
2927 | #ifdef CONFIG_RPS |
2928 | ||
2929 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 2930 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
2931 | EXPORT_SYMBOL(rps_sock_flow_table); |
2932 | ||
c5905afb | 2933 | struct static_key rps_needed __read_mostly; |
adc9300e | 2934 | |
c445477d BH |
2935 | static struct rps_dev_flow * |
2936 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2937 | struct rps_dev_flow *rflow, u16 next_cpu) | |
2938 | { | |
09994d1b | 2939 | if (next_cpu != RPS_NO_CPU) { |
c445477d BH |
2940 | #ifdef CONFIG_RFS_ACCEL |
2941 | struct netdev_rx_queue *rxqueue; | |
2942 | struct rps_dev_flow_table *flow_table; | |
2943 | struct rps_dev_flow *old_rflow; | |
2944 | u32 flow_id; | |
2945 | u16 rxq_index; | |
2946 | int rc; | |
2947 | ||
2948 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
2949 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
2950 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
2951 | goto out; |
2952 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
2953 | if (rxq_index == skb_get_rx_queue(skb)) | |
2954 | goto out; | |
2955 | ||
2956 | rxqueue = dev->_rx + rxq_index; | |
2957 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2958 | if (!flow_table) | |
2959 | goto out; | |
61b905da | 2960 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
2961 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
2962 | rxq_index, flow_id); | |
2963 | if (rc < 0) | |
2964 | goto out; | |
2965 | old_rflow = rflow; | |
2966 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
2967 | rflow->filter = rc; |
2968 | if (old_rflow->filter == rflow->filter) | |
2969 | old_rflow->filter = RPS_NO_FILTER; | |
2970 | out: | |
2971 | #endif | |
2972 | rflow->last_qtail = | |
09994d1b | 2973 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
2974 | } |
2975 | ||
09994d1b | 2976 | rflow->cpu = next_cpu; |
c445477d BH |
2977 | return rflow; |
2978 | } | |
2979 | ||
bfb564e7 KK |
2980 | /* |
2981 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2982 | * CPU from the RPS map of the receiving queue for a given skb. | |
2983 | * rcu_read_lock must be held on entry. | |
2984 | */ | |
2985 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2986 | struct rps_dev_flow **rflowp) | |
2987 | { | |
2988 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 2989 | struct rps_map *map; |
bfb564e7 KK |
2990 | struct rps_dev_flow_table *flow_table; |
2991 | struct rps_sock_flow_table *sock_flow_table; | |
2992 | int cpu = -1; | |
2993 | u16 tcpu; | |
61b905da | 2994 | u32 hash; |
bfb564e7 KK |
2995 | |
2996 | if (skb_rx_queue_recorded(skb)) { | |
2997 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
2998 | if (unlikely(index >= dev->real_num_rx_queues)) { |
2999 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3000 | "%s received packet on queue %u, but number " | |
3001 | "of RX queues is %u\n", | |
3002 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3003 | goto done; |
3004 | } | |
3005 | rxqueue = dev->_rx + index; | |
3006 | } else | |
3007 | rxqueue = dev->_rx; | |
3008 | ||
6e3f7faf ED |
3009 | map = rcu_dereference(rxqueue->rps_map); |
3010 | if (map) { | |
85875236 | 3011 | if (map->len == 1 && |
33d480ce | 3012 | !rcu_access_pointer(rxqueue->rps_flow_table)) { |
6febfca9 CG |
3013 | tcpu = map->cpus[0]; |
3014 | if (cpu_online(tcpu)) | |
3015 | cpu = tcpu; | |
3016 | goto done; | |
3017 | } | |
33d480ce | 3018 | } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) { |
bfb564e7 | 3019 | goto done; |
6febfca9 | 3020 | } |
bfb564e7 | 3021 | |
2d47b459 | 3022 | skb_reset_network_header(skb); |
61b905da TH |
3023 | hash = skb_get_hash(skb); |
3024 | if (!hash) | |
bfb564e7 KK |
3025 | goto done; |
3026 | ||
fec5e652 TH |
3027 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
3028 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
3029 | if (flow_table && sock_flow_table) { | |
3030 | u16 next_cpu; | |
3031 | struct rps_dev_flow *rflow; | |
3032 | ||
61b905da | 3033 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3034 | tcpu = rflow->cpu; |
3035 | ||
61b905da | 3036 | next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask]; |
fec5e652 TH |
3037 | |
3038 | /* | |
3039 | * If the desired CPU (where last recvmsg was done) is | |
3040 | * different from current CPU (one in the rx-queue flow | |
3041 | * table entry), switch if one of the following holds: | |
3042 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
3043 | * - Current CPU is offline. | |
3044 | * - The current CPU's queue tail has advanced beyond the | |
3045 | * last packet that was enqueued using this table entry. | |
3046 | * This guarantees that all previous packets for the flow | |
3047 | * have been dequeued, thus preserving in order delivery. | |
3048 | */ | |
3049 | if (unlikely(tcpu != next_cpu) && | |
3050 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
3051 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
baefa31d TH |
3052 | rflow->last_qtail)) >= 0)) { |
3053 | tcpu = next_cpu; | |
c445477d | 3054 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3055 | } |
c445477d | 3056 | |
fec5e652 TH |
3057 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
3058 | *rflowp = rflow; | |
3059 | cpu = tcpu; | |
3060 | goto done; | |
3061 | } | |
3062 | } | |
3063 | ||
0a9627f2 | 3064 | if (map) { |
61b905da | 3065 | tcpu = map->cpus[((u64) hash * map->len) >> 32]; |
0a9627f2 TH |
3066 | |
3067 | if (cpu_online(tcpu)) { | |
3068 | cpu = tcpu; | |
3069 | goto done; | |
3070 | } | |
3071 | } | |
3072 | ||
3073 | done: | |
0a9627f2 TH |
3074 | return cpu; |
3075 | } | |
3076 | ||
c445477d BH |
3077 | #ifdef CONFIG_RFS_ACCEL |
3078 | ||
3079 | /** | |
3080 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3081 | * @dev: Device on which the filter was set | |
3082 | * @rxq_index: RX queue index | |
3083 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3084 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3085 | * | |
3086 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3087 | * this function for each installed filter and remove the filters for | |
3088 | * which it returns %true. | |
3089 | */ | |
3090 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3091 | u32 flow_id, u16 filter_id) | |
3092 | { | |
3093 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3094 | struct rps_dev_flow_table *flow_table; | |
3095 | struct rps_dev_flow *rflow; | |
3096 | bool expire = true; | |
3097 | int cpu; | |
3098 | ||
3099 | rcu_read_lock(); | |
3100 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3101 | if (flow_table && flow_id <= flow_table->mask) { | |
3102 | rflow = &flow_table->flows[flow_id]; | |
3103 | cpu = ACCESS_ONCE(rflow->cpu); | |
3104 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
3105 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
3106 | rflow->last_qtail) < | |
3107 | (int)(10 * flow_table->mask))) | |
3108 | expire = false; | |
3109 | } | |
3110 | rcu_read_unlock(); | |
3111 | return expire; | |
3112 | } | |
3113 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3114 | ||
3115 | #endif /* CONFIG_RFS_ACCEL */ | |
3116 | ||
0a9627f2 | 3117 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3118 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3119 | { |
e36fa2f7 ED |
3120 | struct softnet_data *sd = data; |
3121 | ||
eecfd7c4 | 3122 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3123 | sd->received_rps++; |
0a9627f2 | 3124 | } |
e36fa2f7 | 3125 | |
fec5e652 | 3126 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3127 | |
e36fa2f7 ED |
3128 | /* |
3129 | * Check if this softnet_data structure is another cpu one | |
3130 | * If yes, queue it to our IPI list and return 1 | |
3131 | * If no, return 0 | |
3132 | */ | |
3133 | static int rps_ipi_queued(struct softnet_data *sd) | |
3134 | { | |
3135 | #ifdef CONFIG_RPS | |
3136 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
3137 | ||
3138 | if (sd != mysd) { | |
3139 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3140 | mysd->rps_ipi_list = sd; | |
3141 | ||
3142 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3143 | return 1; | |
3144 | } | |
3145 | #endif /* CONFIG_RPS */ | |
3146 | return 0; | |
3147 | } | |
3148 | ||
99bbc707 WB |
3149 | #ifdef CONFIG_NET_FLOW_LIMIT |
3150 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3151 | #endif | |
3152 | ||
3153 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3154 | { | |
3155 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3156 | struct sd_flow_limit *fl; | |
3157 | struct softnet_data *sd; | |
3158 | unsigned int old_flow, new_flow; | |
3159 | ||
3160 | if (qlen < (netdev_max_backlog >> 1)) | |
3161 | return false; | |
3162 | ||
3163 | sd = &__get_cpu_var(softnet_data); | |
3164 | ||
3165 | rcu_read_lock(); | |
3166 | fl = rcu_dereference(sd->flow_limit); | |
3167 | if (fl) { | |
3958afa1 | 3168 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3169 | old_flow = fl->history[fl->history_head]; |
3170 | fl->history[fl->history_head] = new_flow; | |
3171 | ||
3172 | fl->history_head++; | |
3173 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3174 | ||
3175 | if (likely(fl->buckets[old_flow])) | |
3176 | fl->buckets[old_flow]--; | |
3177 | ||
3178 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3179 | fl->count++; | |
3180 | rcu_read_unlock(); | |
3181 | return true; | |
3182 | } | |
3183 | } | |
3184 | rcu_read_unlock(); | |
3185 | #endif | |
3186 | return false; | |
3187 | } | |
3188 | ||
0a9627f2 TH |
3189 | /* |
3190 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3191 | * queue (may be a remote CPU queue). | |
3192 | */ | |
fec5e652 TH |
3193 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3194 | unsigned int *qtail) | |
0a9627f2 | 3195 | { |
e36fa2f7 | 3196 | struct softnet_data *sd; |
0a9627f2 | 3197 | unsigned long flags; |
99bbc707 | 3198 | unsigned int qlen; |
0a9627f2 | 3199 | |
e36fa2f7 | 3200 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3201 | |
3202 | local_irq_save(flags); | |
0a9627f2 | 3203 | |
e36fa2f7 | 3204 | rps_lock(sd); |
99bbc707 WB |
3205 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3206 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
6e7676c1 | 3207 | if (skb_queue_len(&sd->input_pkt_queue)) { |
0a9627f2 | 3208 | enqueue: |
e36fa2f7 | 3209 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3210 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3211 | rps_unlock(sd); |
152102c7 | 3212 | local_irq_restore(flags); |
0a9627f2 TH |
3213 | return NET_RX_SUCCESS; |
3214 | } | |
3215 | ||
ebda37c2 ED |
3216 | /* Schedule NAPI for backlog device |
3217 | * We can use non atomic operation since we own the queue lock | |
3218 | */ | |
3219 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3220 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3221 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3222 | } |
3223 | goto enqueue; | |
3224 | } | |
3225 | ||
dee42870 | 3226 | sd->dropped++; |
e36fa2f7 | 3227 | rps_unlock(sd); |
0a9627f2 | 3228 | |
0a9627f2 TH |
3229 | local_irq_restore(flags); |
3230 | ||
caf586e5 | 3231 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3232 | kfree_skb(skb); |
3233 | return NET_RX_DROP; | |
3234 | } | |
1da177e4 | 3235 | |
ae78dbfa | 3236 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3237 | { |
b0e28f1e | 3238 | int ret; |
1da177e4 | 3239 | |
588f0330 | 3240 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3241 | |
cf66ba58 | 3242 | trace_netif_rx(skb); |
df334545 | 3243 | #ifdef CONFIG_RPS |
c5905afb | 3244 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3245 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3246 | int cpu; |
3247 | ||
cece1945 | 3248 | preempt_disable(); |
b0e28f1e | 3249 | rcu_read_lock(); |
fec5e652 TH |
3250 | |
3251 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3252 | if (cpu < 0) |
3253 | cpu = smp_processor_id(); | |
fec5e652 TH |
3254 | |
3255 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3256 | ||
b0e28f1e | 3257 | rcu_read_unlock(); |
cece1945 | 3258 | preempt_enable(); |
adc9300e ED |
3259 | } else |
3260 | #endif | |
fec5e652 TH |
3261 | { |
3262 | unsigned int qtail; | |
3263 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3264 | put_cpu(); | |
3265 | } | |
b0e28f1e | 3266 | return ret; |
1da177e4 | 3267 | } |
ae78dbfa BH |
3268 | |
3269 | /** | |
3270 | * netif_rx - post buffer to the network code | |
3271 | * @skb: buffer to post | |
3272 | * | |
3273 | * This function receives a packet from a device driver and queues it for | |
3274 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3275 | * may be dropped during processing for congestion control or by the | |
3276 | * protocol layers. | |
3277 | * | |
3278 | * return values: | |
3279 | * NET_RX_SUCCESS (no congestion) | |
3280 | * NET_RX_DROP (packet was dropped) | |
3281 | * | |
3282 | */ | |
3283 | ||
3284 | int netif_rx(struct sk_buff *skb) | |
3285 | { | |
3286 | trace_netif_rx_entry(skb); | |
3287 | ||
3288 | return netif_rx_internal(skb); | |
3289 | } | |
d1b19dff | 3290 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3291 | |
3292 | int netif_rx_ni(struct sk_buff *skb) | |
3293 | { | |
3294 | int err; | |
3295 | ||
ae78dbfa BH |
3296 | trace_netif_rx_ni_entry(skb); |
3297 | ||
1da177e4 | 3298 | preempt_disable(); |
ae78dbfa | 3299 | err = netif_rx_internal(skb); |
1da177e4 LT |
3300 | if (local_softirq_pending()) |
3301 | do_softirq(); | |
3302 | preempt_enable(); | |
3303 | ||
3304 | return err; | |
3305 | } | |
1da177e4 LT |
3306 | EXPORT_SYMBOL(netif_rx_ni); |
3307 | ||
1da177e4 LT |
3308 | static void net_tx_action(struct softirq_action *h) |
3309 | { | |
3310 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
3311 | ||
3312 | if (sd->completion_queue) { | |
3313 | struct sk_buff *clist; | |
3314 | ||
3315 | local_irq_disable(); | |
3316 | clist = sd->completion_queue; | |
3317 | sd->completion_queue = NULL; | |
3318 | local_irq_enable(); | |
3319 | ||
3320 | while (clist) { | |
3321 | struct sk_buff *skb = clist; | |
3322 | clist = clist->next; | |
3323 | ||
547b792c | 3324 | WARN_ON(atomic_read(&skb->users)); |
e6247027 ED |
3325 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
3326 | trace_consume_skb(skb); | |
3327 | else | |
3328 | trace_kfree_skb(skb, net_tx_action); | |
1da177e4 LT |
3329 | __kfree_skb(skb); |
3330 | } | |
3331 | } | |
3332 | ||
3333 | if (sd->output_queue) { | |
37437bb2 | 3334 | struct Qdisc *head; |
1da177e4 LT |
3335 | |
3336 | local_irq_disable(); | |
3337 | head = sd->output_queue; | |
3338 | sd->output_queue = NULL; | |
a9cbd588 | 3339 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3340 | local_irq_enable(); |
3341 | ||
3342 | while (head) { | |
37437bb2 DM |
3343 | struct Qdisc *q = head; |
3344 | spinlock_t *root_lock; | |
3345 | ||
1da177e4 LT |
3346 | head = head->next_sched; |
3347 | ||
5fb66229 | 3348 | root_lock = qdisc_lock(q); |
37437bb2 | 3349 | if (spin_trylock(root_lock)) { |
def82a1d JP |
3350 | smp_mb__before_clear_bit(); |
3351 | clear_bit(__QDISC_STATE_SCHED, | |
3352 | &q->state); | |
37437bb2 DM |
3353 | qdisc_run(q); |
3354 | spin_unlock(root_lock); | |
1da177e4 | 3355 | } else { |
195648bb | 3356 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3357 | &q->state)) { |
195648bb | 3358 | __netif_reschedule(q); |
e8a83e10 JP |
3359 | } else { |
3360 | smp_mb__before_clear_bit(); | |
3361 | clear_bit(__QDISC_STATE_SCHED, | |
3362 | &q->state); | |
3363 | } | |
1da177e4 LT |
3364 | } |
3365 | } | |
3366 | } | |
3367 | } | |
3368 | ||
ab95bfe0 JP |
3369 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3370 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3371 | /* This hook is defined here for ATM LANE */ |
3372 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3373 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3374 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3375 | #endif |
1da177e4 | 3376 | |
1da177e4 LT |
3377 | #ifdef CONFIG_NET_CLS_ACT |
3378 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3379 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3380 | * a compare and 2 stores extra right now if we dont have it on | |
3381 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3382 | * NOTE: This doesn't stop any functionality; if you dont have |
3383 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3384 | * |
3385 | */ | |
24824a09 | 3386 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3387 | { |
1da177e4 | 3388 | struct net_device *dev = skb->dev; |
f697c3e8 | 3389 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3390 | int result = TC_ACT_OK; |
3391 | struct Qdisc *q; | |
4ec93edb | 3392 | |
de384830 | 3393 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
e87cc472 JP |
3394 | net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n", |
3395 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3396 | return TC_ACT_SHOT; |
3397 | } | |
1da177e4 | 3398 | |
f697c3e8 HX |
3399 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3400 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3401 | |
83874000 | 3402 | q = rxq->qdisc; |
8d50b53d | 3403 | if (q != &noop_qdisc) { |
83874000 | 3404 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3405 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3406 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3407 | spin_unlock(qdisc_lock(q)); |
3408 | } | |
f697c3e8 HX |
3409 | |
3410 | return result; | |
3411 | } | |
86e65da9 | 3412 | |
f697c3e8 HX |
3413 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3414 | struct packet_type **pt_prev, | |
3415 | int *ret, struct net_device *orig_dev) | |
3416 | { | |
24824a09 ED |
3417 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3418 | ||
3419 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
f697c3e8 | 3420 | goto out; |
1da177e4 | 3421 | |
f697c3e8 HX |
3422 | if (*pt_prev) { |
3423 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3424 | *pt_prev = NULL; | |
1da177e4 LT |
3425 | } |
3426 | ||
24824a09 | 3427 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3428 | case TC_ACT_SHOT: |
3429 | case TC_ACT_STOLEN: | |
3430 | kfree_skb(skb); | |
3431 | return NULL; | |
3432 | } | |
3433 | ||
3434 | out: | |
3435 | skb->tc_verd = 0; | |
3436 | return skb; | |
1da177e4 LT |
3437 | } |
3438 | #endif | |
3439 | ||
ab95bfe0 JP |
3440 | /** |
3441 | * netdev_rx_handler_register - register receive handler | |
3442 | * @dev: device to register a handler for | |
3443 | * @rx_handler: receive handler to register | |
93e2c32b | 3444 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 3445 | * |
e227867f | 3446 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
3447 | * called from __netif_receive_skb. A negative errno code is returned |
3448 | * on a failure. | |
3449 | * | |
3450 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3451 | * |
3452 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3453 | */ |
3454 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3455 | rx_handler_func_t *rx_handler, |
3456 | void *rx_handler_data) | |
ab95bfe0 JP |
3457 | { |
3458 | ASSERT_RTNL(); | |
3459 | ||
3460 | if (dev->rx_handler) | |
3461 | return -EBUSY; | |
3462 | ||
00cfec37 | 3463 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 3464 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3465 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3466 | ||
3467 | return 0; | |
3468 | } | |
3469 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3470 | ||
3471 | /** | |
3472 | * netdev_rx_handler_unregister - unregister receive handler | |
3473 | * @dev: device to unregister a handler from | |
3474 | * | |
166ec369 | 3475 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
3476 | * |
3477 | * The caller must hold the rtnl_mutex. | |
3478 | */ | |
3479 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3480 | { | |
3481 | ||
3482 | ASSERT_RTNL(); | |
a9b3cd7f | 3483 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
3484 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
3485 | * section has a guarantee to see a non NULL rx_handler_data | |
3486 | * as well. | |
3487 | */ | |
3488 | synchronize_net(); | |
a9b3cd7f | 3489 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
3490 | } |
3491 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3492 | ||
b4b9e355 MG |
3493 | /* |
3494 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3495 | * the special handling of PFMEMALLOC skbs. | |
3496 | */ | |
3497 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3498 | { | |
3499 | switch (skb->protocol) { | |
2b8837ae JP |
3500 | case htons(ETH_P_ARP): |
3501 | case htons(ETH_P_IP): | |
3502 | case htons(ETH_P_IPV6): | |
3503 | case htons(ETH_P_8021Q): | |
3504 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
3505 | return true; |
3506 | default: | |
3507 | return false; | |
3508 | } | |
3509 | } | |
3510 | ||
9754e293 | 3511 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
3512 | { |
3513 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3514 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3515 | struct net_device *orig_dev; |
63d8ea7f | 3516 | struct net_device *null_or_dev; |
8a4eb573 | 3517 | bool deliver_exact = false; |
1da177e4 | 3518 | int ret = NET_RX_DROP; |
252e3346 | 3519 | __be16 type; |
1da177e4 | 3520 | |
588f0330 | 3521 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3522 | |
cf66ba58 | 3523 | trace_netif_receive_skb(skb); |
9b22ea56 | 3524 | |
cc9bd5ce | 3525 | orig_dev = skb->dev; |
8f903c70 | 3526 | |
c1d2bbe1 | 3527 | skb_reset_network_header(skb); |
fda55eca ED |
3528 | if (!skb_transport_header_was_set(skb)) |
3529 | skb_reset_transport_header(skb); | |
0b5c9db1 | 3530 | skb_reset_mac_len(skb); |
1da177e4 LT |
3531 | |
3532 | pt_prev = NULL; | |
3533 | ||
3534 | rcu_read_lock(); | |
3535 | ||
63d8ea7f | 3536 | another_round: |
b6858177 | 3537 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
3538 | |
3539 | __this_cpu_inc(softnet_data.processed); | |
3540 | ||
8ad227ff PM |
3541 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
3542 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
bcc6d479 JP |
3543 | skb = vlan_untag(skb); |
3544 | if (unlikely(!skb)) | |
b4b9e355 | 3545 | goto unlock; |
bcc6d479 JP |
3546 | } |
3547 | ||
1da177e4 LT |
3548 | #ifdef CONFIG_NET_CLS_ACT |
3549 | if (skb->tc_verd & TC_NCLS) { | |
3550 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3551 | goto ncls; | |
3552 | } | |
3553 | #endif | |
3554 | ||
9754e293 | 3555 | if (pfmemalloc) |
b4b9e355 MG |
3556 | goto skip_taps; |
3557 | ||
1da177e4 | 3558 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
63d8ea7f | 3559 | if (!ptype->dev || ptype->dev == skb->dev) { |
4ec93edb | 3560 | if (pt_prev) |
f2ccd8fa | 3561 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3562 | pt_prev = ptype; |
3563 | } | |
3564 | } | |
3565 | ||
b4b9e355 | 3566 | skip_taps: |
1da177e4 | 3567 | #ifdef CONFIG_NET_CLS_ACT |
f697c3e8 HX |
3568 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3569 | if (!skb) | |
b4b9e355 | 3570 | goto unlock; |
1da177e4 LT |
3571 | ncls: |
3572 | #endif | |
3573 | ||
9754e293 | 3574 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
3575 | goto drop; |
3576 | ||
2425717b JF |
3577 | if (vlan_tx_tag_present(skb)) { |
3578 | if (pt_prev) { | |
3579 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3580 | pt_prev = NULL; | |
3581 | } | |
48cc32d3 | 3582 | if (vlan_do_receive(&skb)) |
2425717b JF |
3583 | goto another_round; |
3584 | else if (unlikely(!skb)) | |
b4b9e355 | 3585 | goto unlock; |
2425717b JF |
3586 | } |
3587 | ||
48cc32d3 | 3588 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
3589 | if (rx_handler) { |
3590 | if (pt_prev) { | |
3591 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3592 | pt_prev = NULL; | |
3593 | } | |
8a4eb573 JP |
3594 | switch (rx_handler(&skb)) { |
3595 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 3596 | ret = NET_RX_SUCCESS; |
b4b9e355 | 3597 | goto unlock; |
8a4eb573 | 3598 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3599 | goto another_round; |
8a4eb573 JP |
3600 | case RX_HANDLER_EXACT: |
3601 | deliver_exact = true; | |
3602 | case RX_HANDLER_PASS: | |
3603 | break; | |
3604 | default: | |
3605 | BUG(); | |
3606 | } | |
ab95bfe0 | 3607 | } |
1da177e4 | 3608 | |
d4b812de ED |
3609 | if (unlikely(vlan_tx_tag_present(skb))) { |
3610 | if (vlan_tx_tag_get_id(skb)) | |
3611 | skb->pkt_type = PACKET_OTHERHOST; | |
3612 | /* Note: we might in the future use prio bits | |
3613 | * and set skb->priority like in vlan_do_receive() | |
3614 | * For the time being, just ignore Priority Code Point | |
3615 | */ | |
3616 | skb->vlan_tci = 0; | |
3617 | } | |
48cc32d3 | 3618 | |
63d8ea7f | 3619 | /* deliver only exact match when indicated */ |
8a4eb573 | 3620 | null_or_dev = deliver_exact ? skb->dev : NULL; |
1f3c8804 | 3621 | |
1da177e4 | 3622 | type = skb->protocol; |
82d8a867 PE |
3623 | list_for_each_entry_rcu(ptype, |
3624 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
63d8ea7f | 3625 | if (ptype->type == type && |
e3f48d37 JP |
3626 | (ptype->dev == null_or_dev || ptype->dev == skb->dev || |
3627 | ptype->dev == orig_dev)) { | |
4ec93edb | 3628 | if (pt_prev) |
f2ccd8fa | 3629 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3630 | pt_prev = ptype; |
3631 | } | |
3632 | } | |
3633 | ||
3634 | if (pt_prev) { | |
1080e512 | 3635 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 3636 | goto drop; |
1080e512 MT |
3637 | else |
3638 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 3639 | } else { |
b4b9e355 | 3640 | drop: |
caf586e5 | 3641 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3642 | kfree_skb(skb); |
3643 | /* Jamal, now you will not able to escape explaining | |
3644 | * me how you were going to use this. :-) | |
3645 | */ | |
3646 | ret = NET_RX_DROP; | |
3647 | } | |
3648 | ||
b4b9e355 | 3649 | unlock: |
1da177e4 | 3650 | rcu_read_unlock(); |
9754e293 DM |
3651 | return ret; |
3652 | } | |
3653 | ||
3654 | static int __netif_receive_skb(struct sk_buff *skb) | |
3655 | { | |
3656 | int ret; | |
3657 | ||
3658 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
3659 | unsigned long pflags = current->flags; | |
3660 | ||
3661 | /* | |
3662 | * PFMEMALLOC skbs are special, they should | |
3663 | * - be delivered to SOCK_MEMALLOC sockets only | |
3664 | * - stay away from userspace | |
3665 | * - have bounded memory usage | |
3666 | * | |
3667 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
3668 | * context down to all allocation sites. | |
3669 | */ | |
3670 | current->flags |= PF_MEMALLOC; | |
3671 | ret = __netif_receive_skb_core(skb, true); | |
3672 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
3673 | } else | |
3674 | ret = __netif_receive_skb_core(skb, false); | |
3675 | ||
1da177e4 LT |
3676 | return ret; |
3677 | } | |
0a9627f2 | 3678 | |
ae78dbfa | 3679 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 3680 | { |
588f0330 | 3681 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3682 | |
c1f19b51 RC |
3683 | if (skb_defer_rx_timestamp(skb)) |
3684 | return NET_RX_SUCCESS; | |
3685 | ||
df334545 | 3686 | #ifdef CONFIG_RPS |
c5905afb | 3687 | if (static_key_false(&rps_needed)) { |
3b098e2d ED |
3688 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
3689 | int cpu, ret; | |
fec5e652 | 3690 | |
3b098e2d ED |
3691 | rcu_read_lock(); |
3692 | ||
3693 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3694 | |
3b098e2d ED |
3695 | if (cpu >= 0) { |
3696 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3697 | rcu_read_unlock(); | |
adc9300e | 3698 | return ret; |
3b098e2d | 3699 | } |
adc9300e | 3700 | rcu_read_unlock(); |
fec5e652 | 3701 | } |
1e94d72f | 3702 | #endif |
adc9300e | 3703 | return __netif_receive_skb(skb); |
0a9627f2 | 3704 | } |
ae78dbfa BH |
3705 | |
3706 | /** | |
3707 | * netif_receive_skb - process receive buffer from network | |
3708 | * @skb: buffer to process | |
3709 | * | |
3710 | * netif_receive_skb() is the main receive data processing function. | |
3711 | * It always succeeds. The buffer may be dropped during processing | |
3712 | * for congestion control or by the protocol layers. | |
3713 | * | |
3714 | * This function may only be called from softirq context and interrupts | |
3715 | * should be enabled. | |
3716 | * | |
3717 | * Return values (usually ignored): | |
3718 | * NET_RX_SUCCESS: no congestion | |
3719 | * NET_RX_DROP: packet was dropped | |
3720 | */ | |
3721 | int netif_receive_skb(struct sk_buff *skb) | |
3722 | { | |
3723 | trace_netif_receive_skb_entry(skb); | |
3724 | ||
3725 | return netif_receive_skb_internal(skb); | |
3726 | } | |
d1b19dff | 3727 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3728 | |
88751275 ED |
3729 | /* Network device is going away, flush any packets still pending |
3730 | * Called with irqs disabled. | |
3731 | */ | |
152102c7 | 3732 | static void flush_backlog(void *arg) |
6e583ce5 | 3733 | { |
152102c7 | 3734 | struct net_device *dev = arg; |
e36fa2f7 | 3735 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3736 | struct sk_buff *skb, *tmp; |
3737 | ||
e36fa2f7 | 3738 | rps_lock(sd); |
6e7676c1 | 3739 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3740 | if (skb->dev == dev) { |
e36fa2f7 | 3741 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3742 | kfree_skb(skb); |
76cc8b13 | 3743 | input_queue_head_incr(sd); |
6e583ce5 | 3744 | } |
6e7676c1 | 3745 | } |
e36fa2f7 | 3746 | rps_unlock(sd); |
6e7676c1 CG |
3747 | |
3748 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3749 | if (skb->dev == dev) { | |
3750 | __skb_unlink(skb, &sd->process_queue); | |
3751 | kfree_skb(skb); | |
76cc8b13 | 3752 | input_queue_head_incr(sd); |
6e7676c1 CG |
3753 | } |
3754 | } | |
6e583ce5 SH |
3755 | } |
3756 | ||
d565b0a1 HX |
3757 | static int napi_gro_complete(struct sk_buff *skb) |
3758 | { | |
22061d80 | 3759 | struct packet_offload *ptype; |
d565b0a1 | 3760 | __be16 type = skb->protocol; |
22061d80 | 3761 | struct list_head *head = &offload_base; |
d565b0a1 HX |
3762 | int err = -ENOENT; |
3763 | ||
c3c7c254 ED |
3764 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
3765 | ||
fc59f9a3 HX |
3766 | if (NAPI_GRO_CB(skb)->count == 1) { |
3767 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3768 | goto out; |
fc59f9a3 | 3769 | } |
d565b0a1 HX |
3770 | |
3771 | rcu_read_lock(); | |
3772 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3773 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
3774 | continue; |
3775 | ||
299603e8 | 3776 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
3777 | break; |
3778 | } | |
3779 | rcu_read_unlock(); | |
3780 | ||
3781 | if (err) { | |
3782 | WARN_ON(&ptype->list == head); | |
3783 | kfree_skb(skb); | |
3784 | return NET_RX_SUCCESS; | |
3785 | } | |
3786 | ||
3787 | out: | |
ae78dbfa | 3788 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
3789 | } |
3790 | ||
2e71a6f8 ED |
3791 | /* napi->gro_list contains packets ordered by age. |
3792 | * youngest packets at the head of it. | |
3793 | * Complete skbs in reverse order to reduce latencies. | |
3794 | */ | |
3795 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 3796 | { |
2e71a6f8 | 3797 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 3798 | |
2e71a6f8 ED |
3799 | /* scan list and build reverse chain */ |
3800 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
3801 | skb->prev = prev; | |
3802 | prev = skb; | |
3803 | } | |
3804 | ||
3805 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 3806 | skb->next = NULL; |
2e71a6f8 ED |
3807 | |
3808 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
3809 | return; | |
3810 | ||
3811 | prev = skb->prev; | |
d565b0a1 | 3812 | napi_gro_complete(skb); |
2e71a6f8 | 3813 | napi->gro_count--; |
d565b0a1 HX |
3814 | } |
3815 | ||
3816 | napi->gro_list = NULL; | |
3817 | } | |
86cac58b | 3818 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3819 | |
89c5fa33 ED |
3820 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
3821 | { | |
3822 | struct sk_buff *p; | |
3823 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 3824 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
3825 | |
3826 | for (p = napi->gro_list; p; p = p->next) { | |
3827 | unsigned long diffs; | |
3828 | ||
0b4cec8c TH |
3829 | NAPI_GRO_CB(p)->flush = 0; |
3830 | ||
3831 | if (hash != skb_get_hash_raw(p)) { | |
3832 | NAPI_GRO_CB(p)->same_flow = 0; | |
3833 | continue; | |
3834 | } | |
3835 | ||
89c5fa33 ED |
3836 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
3837 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
3838 | if (maclen == ETH_HLEN) | |
3839 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 3840 | skb_mac_header(skb)); |
89c5fa33 ED |
3841 | else if (!diffs) |
3842 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 3843 | skb_mac_header(skb), |
89c5fa33 ED |
3844 | maclen); |
3845 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
3846 | } |
3847 | } | |
3848 | ||
299603e8 JC |
3849 | static void skb_gro_reset_offset(struct sk_buff *skb) |
3850 | { | |
3851 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
3852 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
3853 | ||
3854 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3855 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
3856 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
3857 | ||
3858 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
3859 | pinfo->nr_frags && | |
3860 | !PageHighMem(skb_frag_page(frag0))) { | |
3861 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
3862 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
89c5fa33 ED |
3863 | } |
3864 | } | |
3865 | ||
a50e233c ED |
3866 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
3867 | { | |
3868 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
3869 | ||
3870 | BUG_ON(skb->end - skb->tail < grow); | |
3871 | ||
3872 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3873 | ||
3874 | skb->data_len -= grow; | |
3875 | skb->tail += grow; | |
3876 | ||
3877 | pinfo->frags[0].page_offset += grow; | |
3878 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
3879 | ||
3880 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
3881 | skb_frag_unref(skb, 0); | |
3882 | memmove(pinfo->frags, pinfo->frags + 1, | |
3883 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
3884 | } | |
3885 | } | |
3886 | ||
bb728820 | 3887 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3888 | { |
3889 | struct sk_buff **pp = NULL; | |
22061d80 | 3890 | struct packet_offload *ptype; |
d565b0a1 | 3891 | __be16 type = skb->protocol; |
22061d80 | 3892 | struct list_head *head = &offload_base; |
0da2afd5 | 3893 | int same_flow; |
5b252f0c | 3894 | enum gro_result ret; |
a50e233c | 3895 | int grow; |
d565b0a1 | 3896 | |
9c62a68d | 3897 | if (!(skb->dev->features & NETIF_F_GRO)) |
d565b0a1 HX |
3898 | goto normal; |
3899 | ||
21dc3301 | 3900 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) |
f17f5c91 HX |
3901 | goto normal; |
3902 | ||
89c5fa33 | 3903 | gro_list_prepare(napi, skb); |
bf5a755f | 3904 | NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */ |
89c5fa33 | 3905 | |
d565b0a1 HX |
3906 | rcu_read_lock(); |
3907 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3908 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
3909 | continue; |
3910 | ||
86911732 | 3911 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 3912 | skb_reset_mac_len(skb); |
d565b0a1 HX |
3913 | NAPI_GRO_CB(skb)->same_flow = 0; |
3914 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3915 | NAPI_GRO_CB(skb)->free = 0; |
b582ef09 | 3916 | NAPI_GRO_CB(skb)->udp_mark = 0; |
d565b0a1 | 3917 | |
f191a1d1 | 3918 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
3919 | break; |
3920 | } | |
3921 | rcu_read_unlock(); | |
3922 | ||
3923 | if (&ptype->list == head) | |
3924 | goto normal; | |
3925 | ||
0da2afd5 | 3926 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3927 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3928 | |
d565b0a1 HX |
3929 | if (pp) { |
3930 | struct sk_buff *nskb = *pp; | |
3931 | ||
3932 | *pp = nskb->next; | |
3933 | nskb->next = NULL; | |
3934 | napi_gro_complete(nskb); | |
4ae5544f | 3935 | napi->gro_count--; |
d565b0a1 HX |
3936 | } |
3937 | ||
0da2afd5 | 3938 | if (same_flow) |
d565b0a1 HX |
3939 | goto ok; |
3940 | ||
600adc18 | 3941 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 3942 | goto normal; |
d565b0a1 | 3943 | |
600adc18 ED |
3944 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
3945 | struct sk_buff *nskb = napi->gro_list; | |
3946 | ||
3947 | /* locate the end of the list to select the 'oldest' flow */ | |
3948 | while (nskb->next) { | |
3949 | pp = &nskb->next; | |
3950 | nskb = *pp; | |
3951 | } | |
3952 | *pp = NULL; | |
3953 | nskb->next = NULL; | |
3954 | napi_gro_complete(nskb); | |
3955 | } else { | |
3956 | napi->gro_count++; | |
3957 | } | |
d565b0a1 | 3958 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 3959 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 3960 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 3961 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3962 | skb->next = napi->gro_list; |
3963 | napi->gro_list = skb; | |
5d0d9be8 | 3964 | ret = GRO_HELD; |
d565b0a1 | 3965 | |
ad0f9904 | 3966 | pull: |
a50e233c ED |
3967 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
3968 | if (grow > 0) | |
3969 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 3970 | ok: |
5d0d9be8 | 3971 | return ret; |
d565b0a1 HX |
3972 | |
3973 | normal: | |
ad0f9904 HX |
3974 | ret = GRO_NORMAL; |
3975 | goto pull; | |
5d38a079 | 3976 | } |
96e93eab | 3977 | |
bf5a755f JC |
3978 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
3979 | { | |
3980 | struct list_head *offload_head = &offload_base; | |
3981 | struct packet_offload *ptype; | |
3982 | ||
3983 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
3984 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
3985 | continue; | |
3986 | return ptype; | |
3987 | } | |
3988 | return NULL; | |
3989 | } | |
e27a2f83 | 3990 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
3991 | |
3992 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
3993 | { | |
3994 | struct list_head *offload_head = &offload_base; | |
3995 | struct packet_offload *ptype; | |
3996 | ||
3997 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
3998 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
3999 | continue; | |
4000 | return ptype; | |
4001 | } | |
4002 | return NULL; | |
4003 | } | |
e27a2f83 | 4004 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4005 | |
bb728820 | 4006 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4007 | { |
5d0d9be8 HX |
4008 | switch (ret) { |
4009 | case GRO_NORMAL: | |
ae78dbfa | 4010 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4011 | ret = GRO_DROP; |
4012 | break; | |
5d38a079 | 4013 | |
5d0d9be8 | 4014 | case GRO_DROP: |
5d38a079 HX |
4015 | kfree_skb(skb); |
4016 | break; | |
5b252f0c | 4017 | |
daa86548 | 4018 | case GRO_MERGED_FREE: |
d7e8883c ED |
4019 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
4020 | kmem_cache_free(skbuff_head_cache, skb); | |
4021 | else | |
4022 | __kfree_skb(skb); | |
daa86548 ED |
4023 | break; |
4024 | ||
5b252f0c BH |
4025 | case GRO_HELD: |
4026 | case GRO_MERGED: | |
4027 | break; | |
5d38a079 HX |
4028 | } |
4029 | ||
c7c4b3b6 | 4030 | return ret; |
5d0d9be8 | 4031 | } |
5d0d9be8 | 4032 | |
c7c4b3b6 | 4033 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4034 | { |
ae78dbfa | 4035 | trace_napi_gro_receive_entry(skb); |
86911732 | 4036 | |
a50e233c ED |
4037 | skb_gro_reset_offset(skb); |
4038 | ||
89c5fa33 | 4039 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4040 | } |
4041 | EXPORT_SYMBOL(napi_gro_receive); | |
4042 | ||
d0c2b0d2 | 4043 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4044 | { |
96e93eab | 4045 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4046 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4047 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4048 | skb->vlan_tci = 0; |
66c46d74 | 4049 | skb->dev = napi->dev; |
6d152e23 | 4050 | skb->skb_iif = 0; |
e33d0ba8 | 4051 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
96e93eab HX |
4052 | |
4053 | napi->skb = skb; | |
4054 | } | |
96e93eab | 4055 | |
76620aaf | 4056 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4057 | { |
5d38a079 | 4058 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4059 | |
4060 | if (!skb) { | |
89d71a66 | 4061 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
84b9cd63 | 4062 | napi->skb = skb; |
80595d59 | 4063 | } |
96e93eab HX |
4064 | return skb; |
4065 | } | |
76620aaf | 4066 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4067 | |
a50e233c ED |
4068 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4069 | struct sk_buff *skb, | |
4070 | gro_result_t ret) | |
96e93eab | 4071 | { |
5d0d9be8 HX |
4072 | switch (ret) { |
4073 | case GRO_NORMAL: | |
a50e233c ED |
4074 | case GRO_HELD: |
4075 | __skb_push(skb, ETH_HLEN); | |
4076 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4077 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4078 | ret = GRO_DROP; |
86911732 | 4079 | break; |
5d38a079 | 4080 | |
5d0d9be8 | 4081 | case GRO_DROP: |
5d0d9be8 HX |
4082 | case GRO_MERGED_FREE: |
4083 | napi_reuse_skb(napi, skb); | |
4084 | break; | |
5b252f0c BH |
4085 | |
4086 | case GRO_MERGED: | |
4087 | break; | |
5d0d9be8 | 4088 | } |
5d38a079 | 4089 | |
c7c4b3b6 | 4090 | return ret; |
5d38a079 | 4091 | } |
5d0d9be8 | 4092 | |
a50e233c ED |
4093 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
4094 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
4095 | * We copy ethernet header into skb->data to have a common layout. | |
4096 | */ | |
4adb9c4a | 4097 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4098 | { |
4099 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
4100 | const struct ethhdr *eth; |
4101 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
4102 | |
4103 | napi->skb = NULL; | |
4104 | ||
a50e233c ED |
4105 | skb_reset_mac_header(skb); |
4106 | skb_gro_reset_offset(skb); | |
4107 | ||
4108 | eth = skb_gro_header_fast(skb, 0); | |
4109 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
4110 | eth = skb_gro_header_slow(skb, hlen, 0); | |
4111 | if (unlikely(!eth)) { | |
4112 | napi_reuse_skb(napi, skb); | |
4113 | return NULL; | |
4114 | } | |
4115 | } else { | |
4116 | gro_pull_from_frag0(skb, hlen); | |
4117 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
4118 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 4119 | } |
a50e233c ED |
4120 | __skb_pull(skb, hlen); |
4121 | ||
4122 | /* | |
4123 | * This works because the only protocols we care about don't require | |
4124 | * special handling. | |
4125 | * We'll fix it up properly in napi_frags_finish() | |
4126 | */ | |
4127 | skb->protocol = eth->h_proto; | |
76620aaf | 4128 | |
76620aaf HX |
4129 | return skb; |
4130 | } | |
76620aaf | 4131 | |
c7c4b3b6 | 4132 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4133 | { |
76620aaf | 4134 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4135 | |
4136 | if (!skb) | |
c7c4b3b6 | 4137 | return GRO_DROP; |
5d0d9be8 | 4138 | |
ae78dbfa BH |
4139 | trace_napi_gro_frags_entry(skb); |
4140 | ||
89c5fa33 | 4141 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4142 | } |
5d38a079 HX |
4143 | EXPORT_SYMBOL(napi_gro_frags); |
4144 | ||
e326bed2 | 4145 | /* |
855abcf0 | 4146 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
4147 | * Note: called with local irq disabled, but exits with local irq enabled. |
4148 | */ | |
4149 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4150 | { | |
4151 | #ifdef CONFIG_RPS | |
4152 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4153 | ||
4154 | if (remsd) { | |
4155 | sd->rps_ipi_list = NULL; | |
4156 | ||
4157 | local_irq_enable(); | |
4158 | ||
4159 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
4160 | while (remsd) { | |
4161 | struct softnet_data *next = remsd->rps_ipi_next; | |
4162 | ||
4163 | if (cpu_online(remsd->cpu)) | |
c46fff2a | 4164 | smp_call_function_single_async(remsd->cpu, |
fce8ad15 | 4165 | &remsd->csd); |
e326bed2 ED |
4166 | remsd = next; |
4167 | } | |
4168 | } else | |
4169 | #endif | |
4170 | local_irq_enable(); | |
4171 | } | |
4172 | ||
bea3348e | 4173 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
4174 | { |
4175 | int work = 0; | |
eecfd7c4 | 4176 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 4177 | |
e326bed2 ED |
4178 | #ifdef CONFIG_RPS |
4179 | /* Check if we have pending ipi, its better to send them now, | |
4180 | * not waiting net_rx_action() end. | |
4181 | */ | |
4182 | if (sd->rps_ipi_list) { | |
4183 | local_irq_disable(); | |
4184 | net_rps_action_and_irq_enable(sd); | |
4185 | } | |
4186 | #endif | |
bea3348e | 4187 | napi->weight = weight_p; |
6e7676c1 CG |
4188 | local_irq_disable(); |
4189 | while (work < quota) { | |
1da177e4 | 4190 | struct sk_buff *skb; |
6e7676c1 CG |
4191 | unsigned int qlen; |
4192 | ||
4193 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
4194 | local_irq_enable(); | |
4195 | __netif_receive_skb(skb); | |
6e7676c1 | 4196 | local_irq_disable(); |
76cc8b13 TH |
4197 | input_queue_head_incr(sd); |
4198 | if (++work >= quota) { | |
4199 | local_irq_enable(); | |
4200 | return work; | |
4201 | } | |
6e7676c1 | 4202 | } |
1da177e4 | 4203 | |
e36fa2f7 | 4204 | rps_lock(sd); |
6e7676c1 | 4205 | qlen = skb_queue_len(&sd->input_pkt_queue); |
76cc8b13 | 4206 | if (qlen) |
6e7676c1 CG |
4207 | skb_queue_splice_tail_init(&sd->input_pkt_queue, |
4208 | &sd->process_queue); | |
76cc8b13 | 4209 | |
6e7676c1 | 4210 | if (qlen < quota - work) { |
eecfd7c4 ED |
4211 | /* |
4212 | * Inline a custom version of __napi_complete(). | |
4213 | * only current cpu owns and manipulates this napi, | |
4214 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
4215 | * we can use a plain write instead of clear_bit(), | |
4216 | * and we dont need an smp_mb() memory barrier. | |
4217 | */ | |
4218 | list_del(&napi->poll_list); | |
4219 | napi->state = 0; | |
4220 | ||
6e7676c1 | 4221 | quota = work + qlen; |
bea3348e | 4222 | } |
e36fa2f7 | 4223 | rps_unlock(sd); |
6e7676c1 CG |
4224 | } |
4225 | local_irq_enable(); | |
1da177e4 | 4226 | |
bea3348e SH |
4227 | return work; |
4228 | } | |
1da177e4 | 4229 | |
bea3348e SH |
4230 | /** |
4231 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4232 | * @n: entry to schedule |
bea3348e SH |
4233 | * |
4234 | * The entry's receive function will be scheduled to run | |
4235 | */ | |
b5606c2d | 4236 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4237 | { |
4238 | unsigned long flags; | |
1da177e4 | 4239 | |
bea3348e | 4240 | local_irq_save(flags); |
eecfd7c4 | 4241 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 4242 | local_irq_restore(flags); |
1da177e4 | 4243 | } |
bea3348e SH |
4244 | EXPORT_SYMBOL(__napi_schedule); |
4245 | ||
d565b0a1 HX |
4246 | void __napi_complete(struct napi_struct *n) |
4247 | { | |
4248 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
4249 | BUG_ON(n->gro_list); | |
4250 | ||
4251 | list_del(&n->poll_list); | |
4252 | smp_mb__before_clear_bit(); | |
4253 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
4254 | } | |
4255 | EXPORT_SYMBOL(__napi_complete); | |
4256 | ||
4257 | void napi_complete(struct napi_struct *n) | |
4258 | { | |
4259 | unsigned long flags; | |
4260 | ||
4261 | /* | |
4262 | * don't let napi dequeue from the cpu poll list | |
4263 | * just in case its running on a different cpu | |
4264 | */ | |
4265 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4266 | return; | |
4267 | ||
2e71a6f8 | 4268 | napi_gro_flush(n, false); |
d565b0a1 HX |
4269 | local_irq_save(flags); |
4270 | __napi_complete(n); | |
4271 | local_irq_restore(flags); | |
4272 | } | |
4273 | EXPORT_SYMBOL(napi_complete); | |
4274 | ||
af12fa6e ET |
4275 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
4276 | struct napi_struct *napi_by_id(unsigned int napi_id) | |
4277 | { | |
4278 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
4279 | struct napi_struct *napi; | |
4280 | ||
4281 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
4282 | if (napi->napi_id == napi_id) | |
4283 | return napi; | |
4284 | ||
4285 | return NULL; | |
4286 | } | |
4287 | EXPORT_SYMBOL_GPL(napi_by_id); | |
4288 | ||
4289 | void napi_hash_add(struct napi_struct *napi) | |
4290 | { | |
4291 | if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) { | |
4292 | ||
4293 | spin_lock(&napi_hash_lock); | |
4294 | ||
4295 | /* 0 is not a valid id, we also skip an id that is taken | |
4296 | * we expect both events to be extremely rare | |
4297 | */ | |
4298 | napi->napi_id = 0; | |
4299 | while (!napi->napi_id) { | |
4300 | napi->napi_id = ++napi_gen_id; | |
4301 | if (napi_by_id(napi->napi_id)) | |
4302 | napi->napi_id = 0; | |
4303 | } | |
4304 | ||
4305 | hlist_add_head_rcu(&napi->napi_hash_node, | |
4306 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
4307 | ||
4308 | spin_unlock(&napi_hash_lock); | |
4309 | } | |
4310 | } | |
4311 | EXPORT_SYMBOL_GPL(napi_hash_add); | |
4312 | ||
4313 | /* Warning : caller is responsible to make sure rcu grace period | |
4314 | * is respected before freeing memory containing @napi | |
4315 | */ | |
4316 | void napi_hash_del(struct napi_struct *napi) | |
4317 | { | |
4318 | spin_lock(&napi_hash_lock); | |
4319 | ||
4320 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) | |
4321 | hlist_del_rcu(&napi->napi_hash_node); | |
4322 | ||
4323 | spin_unlock(&napi_hash_lock); | |
4324 | } | |
4325 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
4326 | ||
d565b0a1 HX |
4327 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
4328 | int (*poll)(struct napi_struct *, int), int weight) | |
4329 | { | |
4330 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 4331 | napi->gro_count = 0; |
d565b0a1 | 4332 | napi->gro_list = NULL; |
5d38a079 | 4333 | napi->skb = NULL; |
d565b0a1 | 4334 | napi->poll = poll; |
82dc3c63 ED |
4335 | if (weight > NAPI_POLL_WEIGHT) |
4336 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
4337 | weight, dev->name); | |
d565b0a1 HX |
4338 | napi->weight = weight; |
4339 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 4340 | napi->dev = dev; |
5d38a079 | 4341 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
4342 | spin_lock_init(&napi->poll_lock); |
4343 | napi->poll_owner = -1; | |
4344 | #endif | |
4345 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
4346 | } | |
4347 | EXPORT_SYMBOL(netif_napi_add); | |
4348 | ||
4349 | void netif_napi_del(struct napi_struct *napi) | |
4350 | { | |
d7b06636 | 4351 | list_del_init(&napi->dev_list); |
76620aaf | 4352 | napi_free_frags(napi); |
d565b0a1 | 4353 | |
289dccbe | 4354 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 4355 | napi->gro_list = NULL; |
4ae5544f | 4356 | napi->gro_count = 0; |
d565b0a1 HX |
4357 | } |
4358 | EXPORT_SYMBOL(netif_napi_del); | |
4359 | ||
1da177e4 LT |
4360 | static void net_rx_action(struct softirq_action *h) |
4361 | { | |
e326bed2 | 4362 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 4363 | unsigned long time_limit = jiffies + 2; |
51b0bded | 4364 | int budget = netdev_budget; |
53fb95d3 MM |
4365 | void *have; |
4366 | ||
1da177e4 LT |
4367 | local_irq_disable(); |
4368 | ||
e326bed2 | 4369 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
4370 | struct napi_struct *n; |
4371 | int work, weight; | |
1da177e4 | 4372 | |
bea3348e | 4373 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
4374 | * Allow this to run for 2 jiffies since which will allow |
4375 | * an average latency of 1.5/HZ. | |
bea3348e | 4376 | */ |
d1f41b67 | 4377 | if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit))) |
1da177e4 LT |
4378 | goto softnet_break; |
4379 | ||
4380 | local_irq_enable(); | |
4381 | ||
bea3348e SH |
4382 | /* Even though interrupts have been re-enabled, this |
4383 | * access is safe because interrupts can only add new | |
4384 | * entries to the tail of this list, and only ->poll() | |
4385 | * calls can remove this head entry from the list. | |
4386 | */ | |
e326bed2 | 4387 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 4388 | |
bea3348e SH |
4389 | have = netpoll_poll_lock(n); |
4390 | ||
4391 | weight = n->weight; | |
4392 | ||
0a7606c1 DM |
4393 | /* This NAPI_STATE_SCHED test is for avoiding a race |
4394 | * with netpoll's poll_napi(). Only the entity which | |
4395 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
4396 | * actually make the ->poll() call. Therefore we avoid | |
25985edc | 4397 | * accidentally calling ->poll() when NAPI is not scheduled. |
0a7606c1 DM |
4398 | */ |
4399 | work = 0; | |
4ea7e386 | 4400 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 4401 | work = n->poll(n, weight); |
4ea7e386 NH |
4402 | trace_napi_poll(n); |
4403 | } | |
bea3348e SH |
4404 | |
4405 | WARN_ON_ONCE(work > weight); | |
4406 | ||
4407 | budget -= work; | |
4408 | ||
4409 | local_irq_disable(); | |
4410 | ||
4411 | /* Drivers must not modify the NAPI state if they | |
4412 | * consume the entire weight. In such cases this code | |
4413 | * still "owns" the NAPI instance and therefore can | |
4414 | * move the instance around on the list at-will. | |
4415 | */ | |
fed17f30 | 4416 | if (unlikely(work == weight)) { |
ff780cd8 HX |
4417 | if (unlikely(napi_disable_pending(n))) { |
4418 | local_irq_enable(); | |
4419 | napi_complete(n); | |
4420 | local_irq_disable(); | |
2e71a6f8 ED |
4421 | } else { |
4422 | if (n->gro_list) { | |
4423 | /* flush too old packets | |
4424 | * If HZ < 1000, flush all packets. | |
4425 | */ | |
4426 | local_irq_enable(); | |
4427 | napi_gro_flush(n, HZ >= 1000); | |
4428 | local_irq_disable(); | |
4429 | } | |
e326bed2 | 4430 | list_move_tail(&n->poll_list, &sd->poll_list); |
2e71a6f8 | 4431 | } |
fed17f30 | 4432 | } |
bea3348e SH |
4433 | |
4434 | netpoll_poll_unlock(have); | |
1da177e4 LT |
4435 | } |
4436 | out: | |
e326bed2 | 4437 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 4438 | |
db217334 CL |
4439 | #ifdef CONFIG_NET_DMA |
4440 | /* | |
4441 | * There may not be any more sk_buffs coming right now, so push | |
4442 | * any pending DMA copies to hardware | |
4443 | */ | |
2ba05622 | 4444 | dma_issue_pending_all(); |
db217334 | 4445 | #endif |
bea3348e | 4446 | |
1da177e4 LT |
4447 | return; |
4448 | ||
4449 | softnet_break: | |
dee42870 | 4450 | sd->time_squeeze++; |
1da177e4 LT |
4451 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
4452 | goto out; | |
4453 | } | |
4454 | ||
aa9d8560 | 4455 | struct netdev_adjacent { |
9ff162a8 | 4456 | struct net_device *dev; |
5d261913 VF |
4457 | |
4458 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 4459 | bool master; |
5d261913 | 4460 | |
5d261913 VF |
4461 | /* counter for the number of times this device was added to us */ |
4462 | u16 ref_nr; | |
4463 | ||
402dae96 VF |
4464 | /* private field for the users */ |
4465 | void *private; | |
4466 | ||
9ff162a8 JP |
4467 | struct list_head list; |
4468 | struct rcu_head rcu; | |
9ff162a8 JP |
4469 | }; |
4470 | ||
5d261913 VF |
4471 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev, |
4472 | struct net_device *adj_dev, | |
2f268f12 | 4473 | struct list_head *adj_list) |
9ff162a8 | 4474 | { |
5d261913 | 4475 | struct netdev_adjacent *adj; |
5d261913 | 4476 | |
2f268f12 | 4477 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
4478 | if (adj->dev == adj_dev) |
4479 | return adj; | |
9ff162a8 JP |
4480 | } |
4481 | return NULL; | |
4482 | } | |
4483 | ||
4484 | /** | |
4485 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
4486 | * @dev: device | |
4487 | * @upper_dev: upper device to check | |
4488 | * | |
4489 | * Find out if a device is linked to specified upper device and return true | |
4490 | * in case it is. Note that this checks only immediate upper device, | |
4491 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
4492 | */ | |
4493 | bool netdev_has_upper_dev(struct net_device *dev, | |
4494 | struct net_device *upper_dev) | |
4495 | { | |
4496 | ASSERT_RTNL(); | |
4497 | ||
2f268f12 | 4498 | return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper); |
9ff162a8 JP |
4499 | } |
4500 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
4501 | ||
4502 | /** | |
4503 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
4504 | * @dev: device | |
4505 | * | |
4506 | * Find out if a device is linked to an upper device and return true in case | |
4507 | * it is. The caller must hold the RTNL lock. | |
4508 | */ | |
1d143d9f | 4509 | static bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
4510 | { |
4511 | ASSERT_RTNL(); | |
4512 | ||
2f268f12 | 4513 | return !list_empty(&dev->all_adj_list.upper); |
9ff162a8 | 4514 | } |
9ff162a8 JP |
4515 | |
4516 | /** | |
4517 | * netdev_master_upper_dev_get - Get master upper device | |
4518 | * @dev: device | |
4519 | * | |
4520 | * Find a master upper device and return pointer to it or NULL in case | |
4521 | * it's not there. The caller must hold the RTNL lock. | |
4522 | */ | |
4523 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
4524 | { | |
aa9d8560 | 4525 | struct netdev_adjacent *upper; |
9ff162a8 JP |
4526 | |
4527 | ASSERT_RTNL(); | |
4528 | ||
2f268f12 | 4529 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
4530 | return NULL; |
4531 | ||
2f268f12 | 4532 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 4533 | struct netdev_adjacent, list); |
9ff162a8 JP |
4534 | if (likely(upper->master)) |
4535 | return upper->dev; | |
4536 | return NULL; | |
4537 | } | |
4538 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
4539 | ||
b6ccba4c VF |
4540 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
4541 | { | |
4542 | struct netdev_adjacent *adj; | |
4543 | ||
4544 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
4545 | ||
4546 | return adj->private; | |
4547 | } | |
4548 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
4549 | ||
44a40855 VY |
4550 | /** |
4551 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
4552 | * @dev: device | |
4553 | * @iter: list_head ** of the current position | |
4554 | * | |
4555 | * Gets the next device from the dev's upper list, starting from iter | |
4556 | * position. The caller must hold RCU read lock. | |
4557 | */ | |
4558 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
4559 | struct list_head **iter) | |
4560 | { | |
4561 | struct netdev_adjacent *upper; | |
4562 | ||
4563 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
4564 | ||
4565 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4566 | ||
4567 | if (&upper->list == &dev->adj_list.upper) | |
4568 | return NULL; | |
4569 | ||
4570 | *iter = &upper->list; | |
4571 | ||
4572 | return upper->dev; | |
4573 | } | |
4574 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
4575 | ||
31088a11 VF |
4576 | /** |
4577 | * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list | |
48311f46 VF |
4578 | * @dev: device |
4579 | * @iter: list_head ** of the current position | |
4580 | * | |
4581 | * Gets the next device from the dev's upper list, starting from iter | |
4582 | * position. The caller must hold RCU read lock. | |
4583 | */ | |
2f268f12 VF |
4584 | struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, |
4585 | struct list_head **iter) | |
48311f46 VF |
4586 | { |
4587 | struct netdev_adjacent *upper; | |
4588 | ||
85328240 | 4589 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); |
48311f46 VF |
4590 | |
4591 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4592 | ||
2f268f12 | 4593 | if (&upper->list == &dev->all_adj_list.upper) |
48311f46 VF |
4594 | return NULL; |
4595 | ||
4596 | *iter = &upper->list; | |
4597 | ||
4598 | return upper->dev; | |
4599 | } | |
2f268f12 | 4600 | EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); |
48311f46 | 4601 | |
31088a11 VF |
4602 | /** |
4603 | * netdev_lower_get_next_private - Get the next ->private from the | |
4604 | * lower neighbour list | |
4605 | * @dev: device | |
4606 | * @iter: list_head ** of the current position | |
4607 | * | |
4608 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
4609 | * list, starting from iter position. The caller must hold either hold the | |
4610 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
4611 | * list will remain unchainged. | |
4612 | */ | |
4613 | void *netdev_lower_get_next_private(struct net_device *dev, | |
4614 | struct list_head **iter) | |
4615 | { | |
4616 | struct netdev_adjacent *lower; | |
4617 | ||
4618 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
4619 | ||
4620 | if (&lower->list == &dev->adj_list.lower) | |
4621 | return NULL; | |
4622 | ||
6859e7df | 4623 | *iter = lower->list.next; |
31088a11 VF |
4624 | |
4625 | return lower->private; | |
4626 | } | |
4627 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
4628 | ||
4629 | /** | |
4630 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
4631 | * lower neighbour list, RCU | |
4632 | * variant | |
4633 | * @dev: device | |
4634 | * @iter: list_head ** of the current position | |
4635 | * | |
4636 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
4637 | * list, starting from iter position. The caller must hold RCU read lock. | |
4638 | */ | |
4639 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
4640 | struct list_head **iter) | |
4641 | { | |
4642 | struct netdev_adjacent *lower; | |
4643 | ||
4644 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4645 | ||
4646 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4647 | ||
4648 | if (&lower->list == &dev->adj_list.lower) | |
4649 | return NULL; | |
4650 | ||
6859e7df | 4651 | *iter = &lower->list; |
31088a11 VF |
4652 | |
4653 | return lower->private; | |
4654 | } | |
4655 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
4656 | ||
4085ebe8 VY |
4657 | /** |
4658 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
4659 | * list | |
4660 | * @dev: device | |
4661 | * @iter: list_head ** of the current position | |
4662 | * | |
4663 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
4664 | * list, starting from iter position. The caller must hold RTNL lock or | |
4665 | * its own locking that guarantees that the neighbour lower | |
4666 | * list will remain unchainged. | |
4667 | */ | |
4668 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
4669 | { | |
4670 | struct netdev_adjacent *lower; | |
4671 | ||
4672 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
4673 | ||
4674 | if (&lower->list == &dev->adj_list.lower) | |
4675 | return NULL; | |
4676 | ||
4677 | *iter = &lower->list; | |
4678 | ||
4679 | return lower->dev; | |
4680 | } | |
4681 | EXPORT_SYMBOL(netdev_lower_get_next); | |
4682 | ||
e001bfad | 4683 | /** |
4684 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
4685 | * lower neighbour list, RCU | |
4686 | * variant | |
4687 | * @dev: device | |
4688 | * | |
4689 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
4690 | * list. The caller must hold RCU read lock. | |
4691 | */ | |
4692 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
4693 | { | |
4694 | struct netdev_adjacent *lower; | |
4695 | ||
4696 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
4697 | struct netdev_adjacent, list); | |
4698 | if (lower) | |
4699 | return lower->private; | |
4700 | return NULL; | |
4701 | } | |
4702 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
4703 | ||
9ff162a8 JP |
4704 | /** |
4705 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
4706 | * @dev: device | |
4707 | * | |
4708 | * Find a master upper device and return pointer to it or NULL in case | |
4709 | * it's not there. The caller must hold the RCU read lock. | |
4710 | */ | |
4711 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
4712 | { | |
aa9d8560 | 4713 | struct netdev_adjacent *upper; |
9ff162a8 | 4714 | |
2f268f12 | 4715 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 4716 | struct netdev_adjacent, list); |
9ff162a8 JP |
4717 | if (upper && likely(upper->master)) |
4718 | return upper->dev; | |
4719 | return NULL; | |
4720 | } | |
4721 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
4722 | ||
0a59f3a9 | 4723 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
4724 | struct net_device *adj_dev, |
4725 | struct list_head *dev_list) | |
4726 | { | |
4727 | char linkname[IFNAMSIZ+7]; | |
4728 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
4729 | "upper_%s" : "lower_%s", adj_dev->name); | |
4730 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
4731 | linkname); | |
4732 | } | |
0a59f3a9 | 4733 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
4734 | char *name, |
4735 | struct list_head *dev_list) | |
4736 | { | |
4737 | char linkname[IFNAMSIZ+7]; | |
4738 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
4739 | "upper_%s" : "lower_%s", name); | |
4740 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
4741 | } | |
4742 | ||
4743 | #define netdev_adjacent_is_neigh_list(dev, dev_list) \ | |
4744 | (dev_list == &dev->adj_list.upper || \ | |
4745 | dev_list == &dev->adj_list.lower) | |
4746 | ||
5d261913 VF |
4747 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
4748 | struct net_device *adj_dev, | |
7863c054 | 4749 | struct list_head *dev_list, |
402dae96 | 4750 | void *private, bool master) |
5d261913 VF |
4751 | { |
4752 | struct netdev_adjacent *adj; | |
842d67a7 | 4753 | int ret; |
5d261913 | 4754 | |
7863c054 | 4755 | adj = __netdev_find_adj(dev, adj_dev, dev_list); |
5d261913 VF |
4756 | |
4757 | if (adj) { | |
5d261913 VF |
4758 | adj->ref_nr++; |
4759 | return 0; | |
4760 | } | |
4761 | ||
4762 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
4763 | if (!adj) | |
4764 | return -ENOMEM; | |
4765 | ||
4766 | adj->dev = adj_dev; | |
4767 | adj->master = master; | |
5d261913 | 4768 | adj->ref_nr = 1; |
402dae96 | 4769 | adj->private = private; |
5d261913 | 4770 | dev_hold(adj_dev); |
2f268f12 VF |
4771 | |
4772 | pr_debug("dev_hold for %s, because of link added from %s to %s\n", | |
4773 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 | 4774 | |
3ee32707 VF |
4775 | if (netdev_adjacent_is_neigh_list(dev, dev_list)) { |
4776 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); | |
5831d66e VF |
4777 | if (ret) |
4778 | goto free_adj; | |
4779 | } | |
4780 | ||
7863c054 | 4781 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
4782 | if (master) { |
4783 | ret = sysfs_create_link(&(dev->dev.kobj), | |
4784 | &(adj_dev->dev.kobj), "master"); | |
4785 | if (ret) | |
5831d66e | 4786 | goto remove_symlinks; |
842d67a7 | 4787 | |
7863c054 | 4788 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 4789 | } else { |
7863c054 | 4790 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 4791 | } |
5d261913 VF |
4792 | |
4793 | return 0; | |
842d67a7 | 4794 | |
5831d66e | 4795 | remove_symlinks: |
3ee32707 VF |
4796 | if (netdev_adjacent_is_neigh_list(dev, dev_list)) |
4797 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); | |
842d67a7 VF |
4798 | free_adj: |
4799 | kfree(adj); | |
974daef7 | 4800 | dev_put(adj_dev); |
842d67a7 VF |
4801 | |
4802 | return ret; | |
5d261913 VF |
4803 | } |
4804 | ||
1d143d9f | 4805 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
4806 | struct net_device *adj_dev, | |
4807 | struct list_head *dev_list) | |
5d261913 VF |
4808 | { |
4809 | struct netdev_adjacent *adj; | |
4810 | ||
7863c054 | 4811 | adj = __netdev_find_adj(dev, adj_dev, dev_list); |
5d261913 | 4812 | |
2f268f12 VF |
4813 | if (!adj) { |
4814 | pr_err("tried to remove device %s from %s\n", | |
4815 | dev->name, adj_dev->name); | |
5d261913 | 4816 | BUG(); |
2f268f12 | 4817 | } |
5d261913 VF |
4818 | |
4819 | if (adj->ref_nr > 1) { | |
2f268f12 VF |
4820 | pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name, |
4821 | adj->ref_nr-1); | |
5d261913 VF |
4822 | adj->ref_nr--; |
4823 | return; | |
4824 | } | |
4825 | ||
842d67a7 VF |
4826 | if (adj->master) |
4827 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
4828 | ||
3ee32707 VF |
4829 | if (netdev_adjacent_is_neigh_list(dev, dev_list)) |
4830 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); | |
5831d66e | 4831 | |
5d261913 | 4832 | list_del_rcu(&adj->list); |
2f268f12 VF |
4833 | pr_debug("dev_put for %s, because link removed from %s to %s\n", |
4834 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 VF |
4835 | dev_put(adj_dev); |
4836 | kfree_rcu(adj, rcu); | |
4837 | } | |
4838 | ||
1d143d9f | 4839 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
4840 | struct net_device *upper_dev, | |
4841 | struct list_head *up_list, | |
4842 | struct list_head *down_list, | |
4843 | void *private, bool master) | |
5d261913 VF |
4844 | { |
4845 | int ret; | |
4846 | ||
402dae96 VF |
4847 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private, |
4848 | master); | |
5d261913 VF |
4849 | if (ret) |
4850 | return ret; | |
4851 | ||
402dae96 VF |
4852 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private, |
4853 | false); | |
5d261913 | 4854 | if (ret) { |
2f268f12 | 4855 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5d261913 VF |
4856 | return ret; |
4857 | } | |
4858 | ||
4859 | return 0; | |
4860 | } | |
4861 | ||
1d143d9f | 4862 | static int __netdev_adjacent_dev_link(struct net_device *dev, |
4863 | struct net_device *upper_dev) | |
5d261913 | 4864 | { |
2f268f12 VF |
4865 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
4866 | &dev->all_adj_list.upper, | |
4867 | &upper_dev->all_adj_list.lower, | |
402dae96 | 4868 | NULL, false); |
5d261913 VF |
4869 | } |
4870 | ||
1d143d9f | 4871 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
4872 | struct net_device *upper_dev, | |
4873 | struct list_head *up_list, | |
4874 | struct list_head *down_list) | |
5d261913 | 4875 | { |
2f268f12 VF |
4876 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
4877 | __netdev_adjacent_dev_remove(upper_dev, dev, down_list); | |
5d261913 VF |
4878 | } |
4879 | ||
1d143d9f | 4880 | static void __netdev_adjacent_dev_unlink(struct net_device *dev, |
4881 | struct net_device *upper_dev) | |
5d261913 | 4882 | { |
2f268f12 VF |
4883 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, |
4884 | &dev->all_adj_list.upper, | |
4885 | &upper_dev->all_adj_list.lower); | |
4886 | } | |
4887 | ||
1d143d9f | 4888 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
4889 | struct net_device *upper_dev, | |
4890 | void *private, bool master) | |
2f268f12 VF |
4891 | { |
4892 | int ret = __netdev_adjacent_dev_link(dev, upper_dev); | |
4893 | ||
4894 | if (ret) | |
4895 | return ret; | |
4896 | ||
4897 | ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, | |
4898 | &dev->adj_list.upper, | |
4899 | &upper_dev->adj_list.lower, | |
402dae96 | 4900 | private, master); |
2f268f12 VF |
4901 | if (ret) { |
4902 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
4903 | return ret; | |
4904 | } | |
4905 | ||
4906 | return 0; | |
5d261913 VF |
4907 | } |
4908 | ||
1d143d9f | 4909 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
4910 | struct net_device *upper_dev) | |
2f268f12 VF |
4911 | { |
4912 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
4913 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, | |
4914 | &dev->adj_list.upper, | |
4915 | &upper_dev->adj_list.lower); | |
4916 | } | |
5d261913 | 4917 | |
9ff162a8 | 4918 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 VF |
4919 | struct net_device *upper_dev, bool master, |
4920 | void *private) | |
9ff162a8 | 4921 | { |
5d261913 VF |
4922 | struct netdev_adjacent *i, *j, *to_i, *to_j; |
4923 | int ret = 0; | |
9ff162a8 JP |
4924 | |
4925 | ASSERT_RTNL(); | |
4926 | ||
4927 | if (dev == upper_dev) | |
4928 | return -EBUSY; | |
4929 | ||
4930 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
2f268f12 | 4931 | if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper)) |
9ff162a8 JP |
4932 | return -EBUSY; |
4933 | ||
2f268f12 | 4934 | if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper)) |
9ff162a8 JP |
4935 | return -EEXIST; |
4936 | ||
4937 | if (master && netdev_master_upper_dev_get(dev)) | |
4938 | return -EBUSY; | |
4939 | ||
402dae96 VF |
4940 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private, |
4941 | master); | |
5d261913 VF |
4942 | if (ret) |
4943 | return ret; | |
9ff162a8 | 4944 | |
5d261913 | 4945 | /* Now that we linked these devs, make all the upper_dev's |
2f268f12 | 4946 | * all_adj_list.upper visible to every dev's all_adj_list.lower an |
5d261913 VF |
4947 | * versa, and don't forget the devices itself. All of these |
4948 | * links are non-neighbours. | |
4949 | */ | |
2f268f12 VF |
4950 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
4951 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
4952 | pr_debug("Interlinking %s with %s, non-neighbour\n", | |
4953 | i->dev->name, j->dev->name); | |
5d261913 VF |
4954 | ret = __netdev_adjacent_dev_link(i->dev, j->dev); |
4955 | if (ret) | |
4956 | goto rollback_mesh; | |
4957 | } | |
4958 | } | |
4959 | ||
4960 | /* add dev to every upper_dev's upper device */ | |
2f268f12 VF |
4961 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
4962 | pr_debug("linking %s's upper device %s with %s\n", | |
4963 | upper_dev->name, i->dev->name, dev->name); | |
5d261913 VF |
4964 | ret = __netdev_adjacent_dev_link(dev, i->dev); |
4965 | if (ret) | |
4966 | goto rollback_upper_mesh; | |
4967 | } | |
4968 | ||
4969 | /* add upper_dev to every dev's lower device */ | |
2f268f12 VF |
4970 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
4971 | pr_debug("linking %s's lower device %s with %s\n", dev->name, | |
4972 | i->dev->name, upper_dev->name); | |
5d261913 VF |
4973 | ret = __netdev_adjacent_dev_link(i->dev, upper_dev); |
4974 | if (ret) | |
4975 | goto rollback_lower_mesh; | |
4976 | } | |
9ff162a8 | 4977 | |
42e52bf9 | 4978 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 | 4979 | return 0; |
5d261913 VF |
4980 | |
4981 | rollback_lower_mesh: | |
4982 | to_i = i; | |
2f268f12 | 4983 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5d261913 VF |
4984 | if (i == to_i) |
4985 | break; | |
4986 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); | |
4987 | } | |
4988 | ||
4989 | i = NULL; | |
4990 | ||
4991 | rollback_upper_mesh: | |
4992 | to_i = i; | |
2f268f12 | 4993 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5d261913 VF |
4994 | if (i == to_i) |
4995 | break; | |
4996 | __netdev_adjacent_dev_unlink(dev, i->dev); | |
4997 | } | |
4998 | ||
4999 | i = j = NULL; | |
5000 | ||
5001 | rollback_mesh: | |
5002 | to_i = i; | |
5003 | to_j = j; | |
2f268f12 VF |
5004 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5005 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5d261913 VF |
5006 | if (i == to_i && j == to_j) |
5007 | break; | |
5008 | __netdev_adjacent_dev_unlink(i->dev, j->dev); | |
5009 | } | |
5010 | if (i == to_i) | |
5011 | break; | |
5012 | } | |
5013 | ||
2f268f12 | 5014 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5015 | |
5016 | return ret; | |
9ff162a8 JP |
5017 | } |
5018 | ||
5019 | /** | |
5020 | * netdev_upper_dev_link - Add a link to the upper device | |
5021 | * @dev: device | |
5022 | * @upper_dev: new upper device | |
5023 | * | |
5024 | * Adds a link to device which is upper to this one. The caller must hold | |
5025 | * the RTNL lock. On a failure a negative errno code is returned. | |
5026 | * On success the reference counts are adjusted and the function | |
5027 | * returns zero. | |
5028 | */ | |
5029 | int netdev_upper_dev_link(struct net_device *dev, | |
5030 | struct net_device *upper_dev) | |
5031 | { | |
402dae96 | 5032 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL); |
9ff162a8 JP |
5033 | } |
5034 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
5035 | ||
5036 | /** | |
5037 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
5038 | * @dev: device | |
5039 | * @upper_dev: new upper device | |
5040 | * | |
5041 | * Adds a link to device which is upper to this one. In this case, only | |
5042 | * one master upper device can be linked, although other non-master devices | |
5043 | * might be linked as well. The caller must hold the RTNL lock. | |
5044 | * On a failure a negative errno code is returned. On success the reference | |
5045 | * counts are adjusted and the function returns zero. | |
5046 | */ | |
5047 | int netdev_master_upper_dev_link(struct net_device *dev, | |
5048 | struct net_device *upper_dev) | |
5049 | { | |
402dae96 | 5050 | return __netdev_upper_dev_link(dev, upper_dev, true, NULL); |
9ff162a8 JP |
5051 | } |
5052 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
5053 | ||
402dae96 VF |
5054 | int netdev_master_upper_dev_link_private(struct net_device *dev, |
5055 | struct net_device *upper_dev, | |
5056 | void *private) | |
5057 | { | |
5058 | return __netdev_upper_dev_link(dev, upper_dev, true, private); | |
5059 | } | |
5060 | EXPORT_SYMBOL(netdev_master_upper_dev_link_private); | |
5061 | ||
9ff162a8 JP |
5062 | /** |
5063 | * netdev_upper_dev_unlink - Removes a link to upper device | |
5064 | * @dev: device | |
5065 | * @upper_dev: new upper device | |
5066 | * | |
5067 | * Removes a link to device which is upper to this one. The caller must hold | |
5068 | * the RTNL lock. | |
5069 | */ | |
5070 | void netdev_upper_dev_unlink(struct net_device *dev, | |
5071 | struct net_device *upper_dev) | |
5072 | { | |
5d261913 | 5073 | struct netdev_adjacent *i, *j; |
9ff162a8 JP |
5074 | ASSERT_RTNL(); |
5075 | ||
2f268f12 | 5076 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5077 | |
5078 | /* Here is the tricky part. We must remove all dev's lower | |
5079 | * devices from all upper_dev's upper devices and vice | |
5080 | * versa, to maintain the graph relationship. | |
5081 | */ | |
2f268f12 VF |
5082 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5083 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) | |
5d261913 VF |
5084 | __netdev_adjacent_dev_unlink(i->dev, j->dev); |
5085 | ||
5086 | /* remove also the devices itself from lower/upper device | |
5087 | * list | |
5088 | */ | |
2f268f12 | 5089 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5d261913 VF |
5090 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); |
5091 | ||
2f268f12 | 5092 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) |
5d261913 VF |
5093 | __netdev_adjacent_dev_unlink(dev, i->dev); |
5094 | ||
42e52bf9 | 5095 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 JP |
5096 | } |
5097 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
5098 | ||
5bb025fa | 5099 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 5100 | { |
5bb025fa | 5101 | struct netdev_adjacent *iter; |
402dae96 | 5102 | |
5bb025fa VF |
5103 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
5104 | netdev_adjacent_sysfs_del(iter->dev, oldname, | |
5105 | &iter->dev->adj_list.lower); | |
5106 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5107 | &iter->dev->adj_list.lower); | |
5108 | } | |
402dae96 | 5109 | |
5bb025fa VF |
5110 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
5111 | netdev_adjacent_sysfs_del(iter->dev, oldname, | |
5112 | &iter->dev->adj_list.upper); | |
5113 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5114 | &iter->dev->adj_list.upper); | |
5115 | } | |
402dae96 | 5116 | } |
402dae96 VF |
5117 | |
5118 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
5119 | struct net_device *lower_dev) | |
5120 | { | |
5121 | struct netdev_adjacent *lower; | |
5122 | ||
5123 | if (!lower_dev) | |
5124 | return NULL; | |
5125 | lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower); | |
5126 | if (!lower) | |
5127 | return NULL; | |
5128 | ||
5129 | return lower->private; | |
5130 | } | |
5131 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
5132 | ||
4085ebe8 VY |
5133 | |
5134 | int dev_get_nest_level(struct net_device *dev, | |
5135 | bool (*type_check)(struct net_device *dev)) | |
5136 | { | |
5137 | struct net_device *lower = NULL; | |
5138 | struct list_head *iter; | |
5139 | int max_nest = -1; | |
5140 | int nest; | |
5141 | ||
5142 | ASSERT_RTNL(); | |
5143 | ||
5144 | netdev_for_each_lower_dev(dev, lower, iter) { | |
5145 | nest = dev_get_nest_level(lower, type_check); | |
5146 | if (max_nest < nest) | |
5147 | max_nest = nest; | |
5148 | } | |
5149 | ||
5150 | if (type_check(dev)) | |
5151 | max_nest++; | |
5152 | ||
5153 | return max_nest; | |
5154 | } | |
5155 | EXPORT_SYMBOL(dev_get_nest_level); | |
5156 | ||
b6c40d68 PM |
5157 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
5158 | { | |
d314774c SH |
5159 | const struct net_device_ops *ops = dev->netdev_ops; |
5160 | ||
d2615bf4 | 5161 | if (ops->ndo_change_rx_flags) |
d314774c | 5162 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
5163 | } |
5164 | ||
991fb3f7 | 5165 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5166 | { |
b536db93 | 5167 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
5168 | kuid_t uid; |
5169 | kgid_t gid; | |
1da177e4 | 5170 | |
24023451 PM |
5171 | ASSERT_RTNL(); |
5172 | ||
dad9b335 WC |
5173 | dev->flags |= IFF_PROMISC; |
5174 | dev->promiscuity += inc; | |
5175 | if (dev->promiscuity == 0) { | |
5176 | /* | |
5177 | * Avoid overflow. | |
5178 | * If inc causes overflow, untouch promisc and return error. | |
5179 | */ | |
5180 | if (inc < 0) | |
5181 | dev->flags &= ~IFF_PROMISC; | |
5182 | else { | |
5183 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
5184 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
5185 | dev->name); | |
dad9b335 WC |
5186 | return -EOVERFLOW; |
5187 | } | |
5188 | } | |
52609c0b | 5189 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
5190 | pr_info("device %s %s promiscuous mode\n", |
5191 | dev->name, | |
5192 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
5193 | if (audit_enabled) { |
5194 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
5195 | audit_log(current->audit_context, GFP_ATOMIC, |
5196 | AUDIT_ANOM_PROMISCUOUS, | |
5197 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
5198 | dev->name, (dev->flags & IFF_PROMISC), | |
5199 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 5200 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
5201 | from_kuid(&init_user_ns, uid), |
5202 | from_kgid(&init_user_ns, gid), | |
7759db82 | 5203 | audit_get_sessionid(current)); |
8192b0c4 | 5204 | } |
24023451 | 5205 | |
b6c40d68 | 5206 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 5207 | } |
991fb3f7 ND |
5208 | if (notify) |
5209 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 5210 | return 0; |
1da177e4 LT |
5211 | } |
5212 | ||
4417da66 PM |
5213 | /** |
5214 | * dev_set_promiscuity - update promiscuity count on a device | |
5215 | * @dev: device | |
5216 | * @inc: modifier | |
5217 | * | |
5218 | * Add or remove promiscuity from a device. While the count in the device | |
5219 | * remains above zero the interface remains promiscuous. Once it hits zero | |
5220 | * the device reverts back to normal filtering operation. A negative inc | |
5221 | * value is used to drop promiscuity on the device. | |
dad9b335 | 5222 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 5223 | */ |
dad9b335 | 5224 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 5225 | { |
b536db93 | 5226 | unsigned int old_flags = dev->flags; |
dad9b335 | 5227 | int err; |
4417da66 | 5228 | |
991fb3f7 | 5229 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 5230 | if (err < 0) |
dad9b335 | 5231 | return err; |
4417da66 PM |
5232 | if (dev->flags != old_flags) |
5233 | dev_set_rx_mode(dev); | |
dad9b335 | 5234 | return err; |
4417da66 | 5235 | } |
d1b19dff | 5236 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 5237 | |
991fb3f7 | 5238 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5239 | { |
991fb3f7 | 5240 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 5241 | |
24023451 PM |
5242 | ASSERT_RTNL(); |
5243 | ||
1da177e4 | 5244 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
5245 | dev->allmulti += inc; |
5246 | if (dev->allmulti == 0) { | |
5247 | /* | |
5248 | * Avoid overflow. | |
5249 | * If inc causes overflow, untouch allmulti and return error. | |
5250 | */ | |
5251 | if (inc < 0) | |
5252 | dev->flags &= ~IFF_ALLMULTI; | |
5253 | else { | |
5254 | dev->allmulti -= inc; | |
7b6cd1ce JP |
5255 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
5256 | dev->name); | |
dad9b335 WC |
5257 | return -EOVERFLOW; |
5258 | } | |
5259 | } | |
24023451 | 5260 | if (dev->flags ^ old_flags) { |
b6c40d68 | 5261 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 5262 | dev_set_rx_mode(dev); |
991fb3f7 ND |
5263 | if (notify) |
5264 | __dev_notify_flags(dev, old_flags, | |
5265 | dev->gflags ^ old_gflags); | |
24023451 | 5266 | } |
dad9b335 | 5267 | return 0; |
4417da66 | 5268 | } |
991fb3f7 ND |
5269 | |
5270 | /** | |
5271 | * dev_set_allmulti - update allmulti count on a device | |
5272 | * @dev: device | |
5273 | * @inc: modifier | |
5274 | * | |
5275 | * Add or remove reception of all multicast frames to a device. While the | |
5276 | * count in the device remains above zero the interface remains listening | |
5277 | * to all interfaces. Once it hits zero the device reverts back to normal | |
5278 | * filtering operation. A negative @inc value is used to drop the counter | |
5279 | * when releasing a resource needing all multicasts. | |
5280 | * Return 0 if successful or a negative errno code on error. | |
5281 | */ | |
5282 | ||
5283 | int dev_set_allmulti(struct net_device *dev, int inc) | |
5284 | { | |
5285 | return __dev_set_allmulti(dev, inc, true); | |
5286 | } | |
d1b19dff | 5287 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
5288 | |
5289 | /* | |
5290 | * Upload unicast and multicast address lists to device and | |
5291 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 5292 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
5293 | * are present. |
5294 | */ | |
5295 | void __dev_set_rx_mode(struct net_device *dev) | |
5296 | { | |
d314774c SH |
5297 | const struct net_device_ops *ops = dev->netdev_ops; |
5298 | ||
4417da66 PM |
5299 | /* dev_open will call this function so the list will stay sane. */ |
5300 | if (!(dev->flags&IFF_UP)) | |
5301 | return; | |
5302 | ||
5303 | if (!netif_device_present(dev)) | |
40b77c94 | 5304 | return; |
4417da66 | 5305 | |
01789349 | 5306 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
5307 | /* Unicast addresses changes may only happen under the rtnl, |
5308 | * therefore calling __dev_set_promiscuity here is safe. | |
5309 | */ | |
32e7bfc4 | 5310 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 5311 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 5312 | dev->uc_promisc = true; |
32e7bfc4 | 5313 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 5314 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 5315 | dev->uc_promisc = false; |
4417da66 | 5316 | } |
4417da66 | 5317 | } |
01789349 JP |
5318 | |
5319 | if (ops->ndo_set_rx_mode) | |
5320 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
5321 | } |
5322 | ||
5323 | void dev_set_rx_mode(struct net_device *dev) | |
5324 | { | |
b9e40857 | 5325 | netif_addr_lock_bh(dev); |
4417da66 | 5326 | __dev_set_rx_mode(dev); |
b9e40857 | 5327 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
5328 | } |
5329 | ||
f0db275a SH |
5330 | /** |
5331 | * dev_get_flags - get flags reported to userspace | |
5332 | * @dev: device | |
5333 | * | |
5334 | * Get the combination of flag bits exported through APIs to userspace. | |
5335 | */ | |
95c96174 | 5336 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 5337 | { |
95c96174 | 5338 | unsigned int flags; |
1da177e4 LT |
5339 | |
5340 | flags = (dev->flags & ~(IFF_PROMISC | | |
5341 | IFF_ALLMULTI | | |
b00055aa SR |
5342 | IFF_RUNNING | |
5343 | IFF_LOWER_UP | | |
5344 | IFF_DORMANT)) | | |
1da177e4 LT |
5345 | (dev->gflags & (IFF_PROMISC | |
5346 | IFF_ALLMULTI)); | |
5347 | ||
b00055aa SR |
5348 | if (netif_running(dev)) { |
5349 | if (netif_oper_up(dev)) | |
5350 | flags |= IFF_RUNNING; | |
5351 | if (netif_carrier_ok(dev)) | |
5352 | flags |= IFF_LOWER_UP; | |
5353 | if (netif_dormant(dev)) | |
5354 | flags |= IFF_DORMANT; | |
5355 | } | |
1da177e4 LT |
5356 | |
5357 | return flags; | |
5358 | } | |
d1b19dff | 5359 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 5360 | |
bd380811 | 5361 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 5362 | { |
b536db93 | 5363 | unsigned int old_flags = dev->flags; |
bd380811 | 5364 | int ret; |
1da177e4 | 5365 | |
24023451 PM |
5366 | ASSERT_RTNL(); |
5367 | ||
1da177e4 LT |
5368 | /* |
5369 | * Set the flags on our device. | |
5370 | */ | |
5371 | ||
5372 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
5373 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
5374 | IFF_AUTOMEDIA)) | | |
5375 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
5376 | IFF_ALLMULTI)); | |
5377 | ||
5378 | /* | |
5379 | * Load in the correct multicast list now the flags have changed. | |
5380 | */ | |
5381 | ||
b6c40d68 PM |
5382 | if ((old_flags ^ flags) & IFF_MULTICAST) |
5383 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 5384 | |
4417da66 | 5385 | dev_set_rx_mode(dev); |
1da177e4 LT |
5386 | |
5387 | /* | |
5388 | * Have we downed the interface. We handle IFF_UP ourselves | |
5389 | * according to user attempts to set it, rather than blindly | |
5390 | * setting it. | |
5391 | */ | |
5392 | ||
5393 | ret = 0; | |
5394 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 5395 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
5396 | |
5397 | if (!ret) | |
4417da66 | 5398 | dev_set_rx_mode(dev); |
1da177e4 LT |
5399 | } |
5400 | ||
1da177e4 | 5401 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 5402 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 5403 | unsigned int old_flags = dev->flags; |
d1b19dff | 5404 | |
1da177e4 | 5405 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
5406 | |
5407 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
5408 | if (dev->flags != old_flags) | |
5409 | dev_set_rx_mode(dev); | |
1da177e4 LT |
5410 | } |
5411 | ||
5412 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
5413 | is important. Some (broken) drivers set IFF_PROMISC, when | |
5414 | IFF_ALLMULTI is requested not asking us and not reporting. | |
5415 | */ | |
5416 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
5417 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
5418 | ||
1da177e4 | 5419 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 5420 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
5421 | } |
5422 | ||
bd380811 PM |
5423 | return ret; |
5424 | } | |
5425 | ||
a528c219 ND |
5426 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
5427 | unsigned int gchanges) | |
bd380811 PM |
5428 | { |
5429 | unsigned int changes = dev->flags ^ old_flags; | |
5430 | ||
a528c219 | 5431 | if (gchanges) |
7f294054 | 5432 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 5433 | |
bd380811 PM |
5434 | if (changes & IFF_UP) { |
5435 | if (dev->flags & IFF_UP) | |
5436 | call_netdevice_notifiers(NETDEV_UP, dev); | |
5437 | else | |
5438 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
5439 | } | |
5440 | ||
5441 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
5442 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
5443 | struct netdev_notifier_change_info change_info; | |
5444 | ||
5445 | change_info.flags_changed = changes; | |
5446 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
5447 | &change_info.info); | |
5448 | } | |
bd380811 PM |
5449 | } |
5450 | ||
5451 | /** | |
5452 | * dev_change_flags - change device settings | |
5453 | * @dev: device | |
5454 | * @flags: device state flags | |
5455 | * | |
5456 | * Change settings on device based state flags. The flags are | |
5457 | * in the userspace exported format. | |
5458 | */ | |
b536db93 | 5459 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 5460 | { |
b536db93 | 5461 | int ret; |
991fb3f7 | 5462 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
5463 | |
5464 | ret = __dev_change_flags(dev, flags); | |
5465 | if (ret < 0) | |
5466 | return ret; | |
5467 | ||
991fb3f7 | 5468 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 5469 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
5470 | return ret; |
5471 | } | |
d1b19dff | 5472 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 5473 | |
2315dc91 VF |
5474 | static int __dev_set_mtu(struct net_device *dev, int new_mtu) |
5475 | { | |
5476 | const struct net_device_ops *ops = dev->netdev_ops; | |
5477 | ||
5478 | if (ops->ndo_change_mtu) | |
5479 | return ops->ndo_change_mtu(dev, new_mtu); | |
5480 | ||
5481 | dev->mtu = new_mtu; | |
5482 | return 0; | |
5483 | } | |
5484 | ||
f0db275a SH |
5485 | /** |
5486 | * dev_set_mtu - Change maximum transfer unit | |
5487 | * @dev: device | |
5488 | * @new_mtu: new transfer unit | |
5489 | * | |
5490 | * Change the maximum transfer size of the network device. | |
5491 | */ | |
1da177e4 LT |
5492 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
5493 | { | |
2315dc91 | 5494 | int err, orig_mtu; |
1da177e4 LT |
5495 | |
5496 | if (new_mtu == dev->mtu) | |
5497 | return 0; | |
5498 | ||
5499 | /* MTU must be positive. */ | |
5500 | if (new_mtu < 0) | |
5501 | return -EINVAL; | |
5502 | ||
5503 | if (!netif_device_present(dev)) | |
5504 | return -ENODEV; | |
5505 | ||
1d486bfb VF |
5506 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
5507 | err = notifier_to_errno(err); | |
5508 | if (err) | |
5509 | return err; | |
d314774c | 5510 | |
2315dc91 VF |
5511 | orig_mtu = dev->mtu; |
5512 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 5513 | |
2315dc91 VF |
5514 | if (!err) { |
5515 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
5516 | err = notifier_to_errno(err); | |
5517 | if (err) { | |
5518 | /* setting mtu back and notifying everyone again, | |
5519 | * so that they have a chance to revert changes. | |
5520 | */ | |
5521 | __dev_set_mtu(dev, orig_mtu); | |
5522 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
5523 | } | |
5524 | } | |
1da177e4 LT |
5525 | return err; |
5526 | } | |
d1b19dff | 5527 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 5528 | |
cbda10fa VD |
5529 | /** |
5530 | * dev_set_group - Change group this device belongs to | |
5531 | * @dev: device | |
5532 | * @new_group: group this device should belong to | |
5533 | */ | |
5534 | void dev_set_group(struct net_device *dev, int new_group) | |
5535 | { | |
5536 | dev->group = new_group; | |
5537 | } | |
5538 | EXPORT_SYMBOL(dev_set_group); | |
5539 | ||
f0db275a SH |
5540 | /** |
5541 | * dev_set_mac_address - Change Media Access Control Address | |
5542 | * @dev: device | |
5543 | * @sa: new address | |
5544 | * | |
5545 | * Change the hardware (MAC) address of the device | |
5546 | */ | |
1da177e4 LT |
5547 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
5548 | { | |
d314774c | 5549 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
5550 | int err; |
5551 | ||
d314774c | 5552 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
5553 | return -EOPNOTSUPP; |
5554 | if (sa->sa_family != dev->type) | |
5555 | return -EINVAL; | |
5556 | if (!netif_device_present(dev)) | |
5557 | return -ENODEV; | |
d314774c | 5558 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
5559 | if (err) |
5560 | return err; | |
fbdeca2d | 5561 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 5562 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 5563 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 5564 | return 0; |
1da177e4 | 5565 | } |
d1b19dff | 5566 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 5567 | |
4bf84c35 JP |
5568 | /** |
5569 | * dev_change_carrier - Change device carrier | |
5570 | * @dev: device | |
691b3b7e | 5571 | * @new_carrier: new value |
4bf84c35 JP |
5572 | * |
5573 | * Change device carrier | |
5574 | */ | |
5575 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
5576 | { | |
5577 | const struct net_device_ops *ops = dev->netdev_ops; | |
5578 | ||
5579 | if (!ops->ndo_change_carrier) | |
5580 | return -EOPNOTSUPP; | |
5581 | if (!netif_device_present(dev)) | |
5582 | return -ENODEV; | |
5583 | return ops->ndo_change_carrier(dev, new_carrier); | |
5584 | } | |
5585 | EXPORT_SYMBOL(dev_change_carrier); | |
5586 | ||
66b52b0d JP |
5587 | /** |
5588 | * dev_get_phys_port_id - Get device physical port ID | |
5589 | * @dev: device | |
5590 | * @ppid: port ID | |
5591 | * | |
5592 | * Get device physical port ID | |
5593 | */ | |
5594 | int dev_get_phys_port_id(struct net_device *dev, | |
5595 | struct netdev_phys_port_id *ppid) | |
5596 | { | |
5597 | const struct net_device_ops *ops = dev->netdev_ops; | |
5598 | ||
5599 | if (!ops->ndo_get_phys_port_id) | |
5600 | return -EOPNOTSUPP; | |
5601 | return ops->ndo_get_phys_port_id(dev, ppid); | |
5602 | } | |
5603 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
5604 | ||
1da177e4 LT |
5605 | /** |
5606 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5607 | * @net: the applicable net namespace |
1da177e4 LT |
5608 | * |
5609 | * Returns a suitable unique value for a new device interface | |
5610 | * number. The caller must hold the rtnl semaphore or the | |
5611 | * dev_base_lock to be sure it remains unique. | |
5612 | */ | |
881d966b | 5613 | static int dev_new_index(struct net *net) |
1da177e4 | 5614 | { |
aa79e66e | 5615 | int ifindex = net->ifindex; |
1da177e4 LT |
5616 | for (;;) { |
5617 | if (++ifindex <= 0) | |
5618 | ifindex = 1; | |
881d966b | 5619 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 5620 | return net->ifindex = ifindex; |
1da177e4 LT |
5621 | } |
5622 | } | |
5623 | ||
1da177e4 | 5624 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5625 | static LIST_HEAD(net_todo_list); |
200b916f | 5626 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 5627 | |
6f05f629 | 5628 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5629 | { |
1da177e4 | 5630 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 5631 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
5632 | } |
5633 | ||
9b5e383c | 5634 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5635 | { |
e93737b0 | 5636 | struct net_device *dev, *tmp; |
5cde2829 | 5637 | LIST_HEAD(close_head); |
9b5e383c | 5638 | |
93ee31f1 DL |
5639 | BUG_ON(dev_boot_phase); |
5640 | ASSERT_RTNL(); | |
5641 | ||
e93737b0 | 5642 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5643 | /* Some devices call without registering |
e93737b0 KK |
5644 | * for initialization unwind. Remove those |
5645 | * devices and proceed with the remaining. | |
9b5e383c ED |
5646 | */ |
5647 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
5648 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
5649 | dev->name, dev); | |
93ee31f1 | 5650 | |
9b5e383c | 5651 | WARN_ON(1); |
e93737b0 KK |
5652 | list_del(&dev->unreg_list); |
5653 | continue; | |
9b5e383c | 5654 | } |
449f4544 | 5655 | dev->dismantle = true; |
9b5e383c | 5656 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5657 | } |
93ee31f1 | 5658 | |
44345724 | 5659 | /* If device is running, close it first. */ |
5cde2829 EB |
5660 | list_for_each_entry(dev, head, unreg_list) |
5661 | list_add_tail(&dev->close_list, &close_head); | |
5662 | dev_close_many(&close_head); | |
93ee31f1 | 5663 | |
44345724 | 5664 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5665 | /* And unlink it from device chain. */ |
5666 | unlist_netdevice(dev); | |
93ee31f1 | 5667 | |
9b5e383c ED |
5668 | dev->reg_state = NETREG_UNREGISTERING; |
5669 | } | |
93ee31f1 DL |
5670 | |
5671 | synchronize_net(); | |
5672 | ||
9b5e383c ED |
5673 | list_for_each_entry(dev, head, unreg_list) { |
5674 | /* Shutdown queueing discipline. */ | |
5675 | dev_shutdown(dev); | |
93ee31f1 DL |
5676 | |
5677 | ||
9b5e383c ED |
5678 | /* Notify protocols, that we are about to destroy |
5679 | this device. They should clean all the things. | |
5680 | */ | |
5681 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5682 | |
9b5e383c ED |
5683 | /* |
5684 | * Flush the unicast and multicast chains | |
5685 | */ | |
a748ee24 | 5686 | dev_uc_flush(dev); |
22bedad3 | 5687 | dev_mc_flush(dev); |
93ee31f1 | 5688 | |
9b5e383c ED |
5689 | if (dev->netdev_ops->ndo_uninit) |
5690 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5691 | |
56bfa7ee RP |
5692 | if (!dev->rtnl_link_ops || |
5693 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5694 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); | |
5695 | ||
9ff162a8 JP |
5696 | /* Notifier chain MUST detach us all upper devices. */ |
5697 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 5698 | |
9b5e383c ED |
5699 | /* Remove entries from kobject tree */ |
5700 | netdev_unregister_kobject(dev); | |
024e9679 AD |
5701 | #ifdef CONFIG_XPS |
5702 | /* Remove XPS queueing entries */ | |
5703 | netif_reset_xps_queues_gt(dev, 0); | |
5704 | #endif | |
9b5e383c | 5705 | } |
93ee31f1 | 5706 | |
850a545b | 5707 | synchronize_net(); |
395264d5 | 5708 | |
a5ee1551 | 5709 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5710 | dev_put(dev); |
5711 | } | |
5712 | ||
5713 | static void rollback_registered(struct net_device *dev) | |
5714 | { | |
5715 | LIST_HEAD(single); | |
5716 | ||
5717 | list_add(&dev->unreg_list, &single); | |
5718 | rollback_registered_many(&single); | |
ceaaec98 | 5719 | list_del(&single); |
93ee31f1 DL |
5720 | } |
5721 | ||
c8f44aff MM |
5722 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
5723 | netdev_features_t features) | |
b63365a2 | 5724 | { |
57422dc5 MM |
5725 | /* Fix illegal checksum combinations */ |
5726 | if ((features & NETIF_F_HW_CSUM) && | |
5727 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5728 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
5729 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
5730 | } | |
5731 | ||
b63365a2 | 5732 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 5733 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 5734 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 5735 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
5736 | } |
5737 | ||
ec5f0615 PS |
5738 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
5739 | !(features & NETIF_F_IP_CSUM)) { | |
5740 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
5741 | features &= ~NETIF_F_TSO; | |
5742 | features &= ~NETIF_F_TSO_ECN; | |
5743 | } | |
5744 | ||
5745 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
5746 | !(features & NETIF_F_IPV6_CSUM)) { | |
5747 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
5748 | features &= ~NETIF_F_TSO6; | |
5749 | } | |
5750 | ||
31d8b9e0 BH |
5751 | /* TSO ECN requires that TSO is present as well. */ |
5752 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
5753 | features &= ~NETIF_F_TSO_ECN; | |
5754 | ||
212b573f MM |
5755 | /* Software GSO depends on SG. */ |
5756 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 5757 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
5758 | features &= ~NETIF_F_GSO; |
5759 | } | |
5760 | ||
acd1130e | 5761 | /* UFO needs SG and checksumming */ |
b63365a2 | 5762 | if (features & NETIF_F_UFO) { |
79032644 MM |
5763 | /* maybe split UFO into V4 and V6? */ |
5764 | if (!((features & NETIF_F_GEN_CSUM) || | |
5765 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
5766 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5767 | netdev_dbg(dev, |
acd1130e | 5768 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
5769 | features &= ~NETIF_F_UFO; |
5770 | } | |
5771 | ||
5772 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 5773 | netdev_dbg(dev, |
acd1130e | 5774 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
5775 | features &= ~NETIF_F_UFO; |
5776 | } | |
5777 | } | |
5778 | ||
d0290214 JP |
5779 | #ifdef CONFIG_NET_RX_BUSY_POLL |
5780 | if (dev->netdev_ops->ndo_busy_poll) | |
5781 | features |= NETIF_F_BUSY_POLL; | |
5782 | else | |
5783 | #endif | |
5784 | features &= ~NETIF_F_BUSY_POLL; | |
5785 | ||
b63365a2 HX |
5786 | return features; |
5787 | } | |
b63365a2 | 5788 | |
6cb6a27c | 5789 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 5790 | { |
c8f44aff | 5791 | netdev_features_t features; |
5455c699 MM |
5792 | int err = 0; |
5793 | ||
87267485 MM |
5794 | ASSERT_RTNL(); |
5795 | ||
5455c699 MM |
5796 | features = netdev_get_wanted_features(dev); |
5797 | ||
5798 | if (dev->netdev_ops->ndo_fix_features) | |
5799 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
5800 | ||
5801 | /* driver might be less strict about feature dependencies */ | |
5802 | features = netdev_fix_features(dev, features); | |
5803 | ||
5804 | if (dev->features == features) | |
6cb6a27c | 5805 | return 0; |
5455c699 | 5806 | |
c8f44aff MM |
5807 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
5808 | &dev->features, &features); | |
5455c699 MM |
5809 | |
5810 | if (dev->netdev_ops->ndo_set_features) | |
5811 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5812 | ||
6cb6a27c | 5813 | if (unlikely(err < 0)) { |
5455c699 | 5814 | netdev_err(dev, |
c8f44aff MM |
5815 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
5816 | err, &features, &dev->features); | |
6cb6a27c MM |
5817 | return -1; |
5818 | } | |
5819 | ||
5820 | if (!err) | |
5821 | dev->features = features; | |
5822 | ||
5823 | return 1; | |
5824 | } | |
5825 | ||
afe12cc8 MM |
5826 | /** |
5827 | * netdev_update_features - recalculate device features | |
5828 | * @dev: the device to check | |
5829 | * | |
5830 | * Recalculate dev->features set and send notifications if it | |
5831 | * has changed. Should be called after driver or hardware dependent | |
5832 | * conditions might have changed that influence the features. | |
5833 | */ | |
6cb6a27c MM |
5834 | void netdev_update_features(struct net_device *dev) |
5835 | { | |
5836 | if (__netdev_update_features(dev)) | |
5837 | netdev_features_change(dev); | |
5455c699 MM |
5838 | } |
5839 | EXPORT_SYMBOL(netdev_update_features); | |
5840 | ||
afe12cc8 MM |
5841 | /** |
5842 | * netdev_change_features - recalculate device features | |
5843 | * @dev: the device to check | |
5844 | * | |
5845 | * Recalculate dev->features set and send notifications even | |
5846 | * if they have not changed. Should be called instead of | |
5847 | * netdev_update_features() if also dev->vlan_features might | |
5848 | * have changed to allow the changes to be propagated to stacked | |
5849 | * VLAN devices. | |
5850 | */ | |
5851 | void netdev_change_features(struct net_device *dev) | |
5852 | { | |
5853 | __netdev_update_features(dev); | |
5854 | netdev_features_change(dev); | |
5855 | } | |
5856 | EXPORT_SYMBOL(netdev_change_features); | |
5857 | ||
fc4a7489 PM |
5858 | /** |
5859 | * netif_stacked_transfer_operstate - transfer operstate | |
5860 | * @rootdev: the root or lower level device to transfer state from | |
5861 | * @dev: the device to transfer operstate to | |
5862 | * | |
5863 | * Transfer operational state from root to device. This is normally | |
5864 | * called when a stacking relationship exists between the root | |
5865 | * device and the device(a leaf device). | |
5866 | */ | |
5867 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5868 | struct net_device *dev) | |
5869 | { | |
5870 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5871 | netif_dormant_on(dev); | |
5872 | else | |
5873 | netif_dormant_off(dev); | |
5874 | ||
5875 | if (netif_carrier_ok(rootdev)) { | |
5876 | if (!netif_carrier_ok(dev)) | |
5877 | netif_carrier_on(dev); | |
5878 | } else { | |
5879 | if (netif_carrier_ok(dev)) | |
5880 | netif_carrier_off(dev); | |
5881 | } | |
5882 | } | |
5883 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5884 | ||
a953be53 | 5885 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
5886 | static int netif_alloc_rx_queues(struct net_device *dev) |
5887 | { | |
1b4bf461 | 5888 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 5889 | struct netdev_rx_queue *rx; |
1b4bf461 | 5890 | |
bd25fa7b | 5891 | BUG_ON(count < 1); |
1b4bf461 | 5892 | |
bd25fa7b | 5893 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
62b5942a | 5894 | if (!rx) |
bd25fa7b | 5895 | return -ENOMEM; |
62b5942a | 5896 | |
bd25fa7b TH |
5897 | dev->_rx = rx; |
5898 | ||
bd25fa7b | 5899 | for (i = 0; i < count; i++) |
fe822240 | 5900 | rx[i].dev = dev; |
1b4bf461 ED |
5901 | return 0; |
5902 | } | |
bf264145 | 5903 | #endif |
1b4bf461 | 5904 | |
aa942104 CG |
5905 | static void netdev_init_one_queue(struct net_device *dev, |
5906 | struct netdev_queue *queue, void *_unused) | |
5907 | { | |
5908 | /* Initialize queue lock */ | |
5909 | spin_lock_init(&queue->_xmit_lock); | |
5910 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
5911 | queue->xmit_lock_owner = -1; | |
b236da69 | 5912 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 5913 | queue->dev = dev; |
114cf580 TH |
5914 | #ifdef CONFIG_BQL |
5915 | dql_init(&queue->dql, HZ); | |
5916 | #endif | |
aa942104 CG |
5917 | } |
5918 | ||
60877a32 ED |
5919 | static void netif_free_tx_queues(struct net_device *dev) |
5920 | { | |
5921 | if (is_vmalloc_addr(dev->_tx)) | |
5922 | vfree(dev->_tx); | |
5923 | else | |
5924 | kfree(dev->_tx); | |
5925 | } | |
5926 | ||
e6484930 TH |
5927 | static int netif_alloc_netdev_queues(struct net_device *dev) |
5928 | { | |
5929 | unsigned int count = dev->num_tx_queues; | |
5930 | struct netdev_queue *tx; | |
60877a32 | 5931 | size_t sz = count * sizeof(*tx); |
e6484930 | 5932 | |
60877a32 | 5933 | BUG_ON(count < 1 || count > 0xffff); |
62b5942a | 5934 | |
60877a32 ED |
5935 | tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
5936 | if (!tx) { | |
5937 | tx = vzalloc(sz); | |
5938 | if (!tx) | |
5939 | return -ENOMEM; | |
5940 | } | |
e6484930 | 5941 | dev->_tx = tx; |
1d24eb48 | 5942 | |
e6484930 TH |
5943 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
5944 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
5945 | |
5946 | return 0; | |
e6484930 TH |
5947 | } |
5948 | ||
1da177e4 LT |
5949 | /** |
5950 | * register_netdevice - register a network device | |
5951 | * @dev: device to register | |
5952 | * | |
5953 | * Take a completed network device structure and add it to the kernel | |
5954 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5955 | * chain. 0 is returned on success. A negative errno code is returned | |
5956 | * on a failure to set up the device, or if the name is a duplicate. | |
5957 | * | |
5958 | * Callers must hold the rtnl semaphore. You may want | |
5959 | * register_netdev() instead of this. | |
5960 | * | |
5961 | * BUGS: | |
5962 | * The locking appears insufficient to guarantee two parallel registers | |
5963 | * will not get the same name. | |
5964 | */ | |
5965 | ||
5966 | int register_netdevice(struct net_device *dev) | |
5967 | { | |
1da177e4 | 5968 | int ret; |
d314774c | 5969 | struct net *net = dev_net(dev); |
1da177e4 LT |
5970 | |
5971 | BUG_ON(dev_boot_phase); | |
5972 | ASSERT_RTNL(); | |
5973 | ||
b17a7c17 SH |
5974 | might_sleep(); |
5975 | ||
1da177e4 LT |
5976 | /* When net_device's are persistent, this will be fatal. */ |
5977 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 5978 | BUG_ON(!net); |
1da177e4 | 5979 | |
f1f28aa3 | 5980 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 5981 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 5982 | |
1da177e4 LT |
5983 | dev->iflink = -1; |
5984 | ||
828de4f6 | 5985 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
5986 | if (ret < 0) |
5987 | goto out; | |
5988 | ||
1da177e4 | 5989 | /* Init, if this function is available */ |
d314774c SH |
5990 | if (dev->netdev_ops->ndo_init) { |
5991 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
5992 | if (ret) { |
5993 | if (ret > 0) | |
5994 | ret = -EIO; | |
90833aa4 | 5995 | goto out; |
1da177e4 LT |
5996 | } |
5997 | } | |
4ec93edb | 5998 | |
f646968f PM |
5999 | if (((dev->hw_features | dev->features) & |
6000 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
6001 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
6002 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
6003 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
6004 | ret = -EINVAL; | |
6005 | goto err_uninit; | |
6006 | } | |
6007 | ||
9c7dafbf PE |
6008 | ret = -EBUSY; |
6009 | if (!dev->ifindex) | |
6010 | dev->ifindex = dev_new_index(net); | |
6011 | else if (__dev_get_by_index(net, dev->ifindex)) | |
6012 | goto err_uninit; | |
6013 | ||
1da177e4 LT |
6014 | if (dev->iflink == -1) |
6015 | dev->iflink = dev->ifindex; | |
6016 | ||
5455c699 MM |
6017 | /* Transfer changeable features to wanted_features and enable |
6018 | * software offloads (GSO and GRO). | |
6019 | */ | |
6020 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
6021 | dev->features |= NETIF_F_SOFT_FEATURES; |
6022 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 6023 | |
34324dc2 MM |
6024 | if (!(dev->flags & IFF_LOOPBACK)) { |
6025 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
c6e1a0d1 TH |
6026 | } |
6027 | ||
1180e7d6 | 6028 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 6029 | */ |
1180e7d6 | 6030 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 6031 | |
ee579677 PS |
6032 | /* Make NETIF_F_SG inheritable to tunnel devices. |
6033 | */ | |
6034 | dev->hw_enc_features |= NETIF_F_SG; | |
6035 | ||
0d89d203 SH |
6036 | /* Make NETIF_F_SG inheritable to MPLS. |
6037 | */ | |
6038 | dev->mpls_features |= NETIF_F_SG; | |
6039 | ||
7ffbe3fd JB |
6040 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
6041 | ret = notifier_to_errno(ret); | |
6042 | if (ret) | |
6043 | goto err_uninit; | |
6044 | ||
8b41d188 | 6045 | ret = netdev_register_kobject(dev); |
b17a7c17 | 6046 | if (ret) |
7ce1b0ed | 6047 | goto err_uninit; |
b17a7c17 SH |
6048 | dev->reg_state = NETREG_REGISTERED; |
6049 | ||
6cb6a27c | 6050 | __netdev_update_features(dev); |
8e9b59b2 | 6051 | |
1da177e4 LT |
6052 | /* |
6053 | * Default initial state at registry is that the | |
6054 | * device is present. | |
6055 | */ | |
6056 | ||
6057 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6058 | ||
8f4cccbb BH |
6059 | linkwatch_init_dev(dev); |
6060 | ||
1da177e4 | 6061 | dev_init_scheduler(dev); |
1da177e4 | 6062 | dev_hold(dev); |
ce286d32 | 6063 | list_netdevice(dev); |
7bf23575 | 6064 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 6065 | |
948b337e JP |
6066 | /* If the device has permanent device address, driver should |
6067 | * set dev_addr and also addr_assign_type should be set to | |
6068 | * NET_ADDR_PERM (default value). | |
6069 | */ | |
6070 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
6071 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
6072 | ||
1da177e4 | 6073 | /* Notify protocols, that a new device appeared. */ |
056925ab | 6074 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 6075 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
6076 | if (ret) { |
6077 | rollback_registered(dev); | |
6078 | dev->reg_state = NETREG_UNREGISTERED; | |
6079 | } | |
d90a909e EB |
6080 | /* |
6081 | * Prevent userspace races by waiting until the network | |
6082 | * device is fully setup before sending notifications. | |
6083 | */ | |
a2835763 PM |
6084 | if (!dev->rtnl_link_ops || |
6085 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 6086 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
6087 | |
6088 | out: | |
6089 | return ret; | |
7ce1b0ed HX |
6090 | |
6091 | err_uninit: | |
d314774c SH |
6092 | if (dev->netdev_ops->ndo_uninit) |
6093 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 6094 | goto out; |
1da177e4 | 6095 | } |
d1b19dff | 6096 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 6097 | |
937f1ba5 BH |
6098 | /** |
6099 | * init_dummy_netdev - init a dummy network device for NAPI | |
6100 | * @dev: device to init | |
6101 | * | |
6102 | * This takes a network device structure and initialize the minimum | |
6103 | * amount of fields so it can be used to schedule NAPI polls without | |
6104 | * registering a full blown interface. This is to be used by drivers | |
6105 | * that need to tie several hardware interfaces to a single NAPI | |
6106 | * poll scheduler due to HW limitations. | |
6107 | */ | |
6108 | int init_dummy_netdev(struct net_device *dev) | |
6109 | { | |
6110 | /* Clear everything. Note we don't initialize spinlocks | |
6111 | * are they aren't supposed to be taken by any of the | |
6112 | * NAPI code and this dummy netdev is supposed to be | |
6113 | * only ever used for NAPI polls | |
6114 | */ | |
6115 | memset(dev, 0, sizeof(struct net_device)); | |
6116 | ||
6117 | /* make sure we BUG if trying to hit standard | |
6118 | * register/unregister code path | |
6119 | */ | |
6120 | dev->reg_state = NETREG_DUMMY; | |
6121 | ||
937f1ba5 BH |
6122 | /* NAPI wants this */ |
6123 | INIT_LIST_HEAD(&dev->napi_list); | |
6124 | ||
6125 | /* a dummy interface is started by default */ | |
6126 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6127 | set_bit(__LINK_STATE_START, &dev->state); | |
6128 | ||
29b4433d ED |
6129 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
6130 | * because users of this 'device' dont need to change | |
6131 | * its refcount. | |
6132 | */ | |
6133 | ||
937f1ba5 BH |
6134 | return 0; |
6135 | } | |
6136 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
6137 | ||
6138 | ||
1da177e4 LT |
6139 | /** |
6140 | * register_netdev - register a network device | |
6141 | * @dev: device to register | |
6142 | * | |
6143 | * Take a completed network device structure and add it to the kernel | |
6144 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6145 | * chain. 0 is returned on success. A negative errno code is returned | |
6146 | * on a failure to set up the device, or if the name is a duplicate. | |
6147 | * | |
38b4da38 | 6148 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
6149 | * and expands the device name if you passed a format string to |
6150 | * alloc_netdev. | |
6151 | */ | |
6152 | int register_netdev(struct net_device *dev) | |
6153 | { | |
6154 | int err; | |
6155 | ||
6156 | rtnl_lock(); | |
1da177e4 | 6157 | err = register_netdevice(dev); |
1da177e4 LT |
6158 | rtnl_unlock(); |
6159 | return err; | |
6160 | } | |
6161 | EXPORT_SYMBOL(register_netdev); | |
6162 | ||
29b4433d ED |
6163 | int netdev_refcnt_read(const struct net_device *dev) |
6164 | { | |
6165 | int i, refcnt = 0; | |
6166 | ||
6167 | for_each_possible_cpu(i) | |
6168 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
6169 | return refcnt; | |
6170 | } | |
6171 | EXPORT_SYMBOL(netdev_refcnt_read); | |
6172 | ||
2c53040f | 6173 | /** |
1da177e4 | 6174 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 6175 | * @dev: target net_device |
1da177e4 LT |
6176 | * |
6177 | * This is called when unregistering network devices. | |
6178 | * | |
6179 | * Any protocol or device that holds a reference should register | |
6180 | * for netdevice notification, and cleanup and put back the | |
6181 | * reference if they receive an UNREGISTER event. | |
6182 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 6183 | * call dev_put. |
1da177e4 LT |
6184 | */ |
6185 | static void netdev_wait_allrefs(struct net_device *dev) | |
6186 | { | |
6187 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 6188 | int refcnt; |
1da177e4 | 6189 | |
e014debe ED |
6190 | linkwatch_forget_dev(dev); |
6191 | ||
1da177e4 | 6192 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
6193 | refcnt = netdev_refcnt_read(dev); |
6194 | ||
6195 | while (refcnt != 0) { | |
1da177e4 | 6196 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 6197 | rtnl_lock(); |
1da177e4 LT |
6198 | |
6199 | /* Rebroadcast unregister notification */ | |
056925ab | 6200 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 6201 | |
748e2d93 | 6202 | __rtnl_unlock(); |
0115e8e3 | 6203 | rcu_barrier(); |
748e2d93 ED |
6204 | rtnl_lock(); |
6205 | ||
0115e8e3 | 6206 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
6207 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
6208 | &dev->state)) { | |
6209 | /* We must not have linkwatch events | |
6210 | * pending on unregister. If this | |
6211 | * happens, we simply run the queue | |
6212 | * unscheduled, resulting in a noop | |
6213 | * for this device. | |
6214 | */ | |
6215 | linkwatch_run_queue(); | |
6216 | } | |
6217 | ||
6756ae4b | 6218 | __rtnl_unlock(); |
1da177e4 LT |
6219 | |
6220 | rebroadcast_time = jiffies; | |
6221 | } | |
6222 | ||
6223 | msleep(250); | |
6224 | ||
29b4433d ED |
6225 | refcnt = netdev_refcnt_read(dev); |
6226 | ||
1da177e4 | 6227 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
6228 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
6229 | dev->name, refcnt); | |
1da177e4 LT |
6230 | warning_time = jiffies; |
6231 | } | |
6232 | } | |
6233 | } | |
6234 | ||
6235 | /* The sequence is: | |
6236 | * | |
6237 | * rtnl_lock(); | |
6238 | * ... | |
6239 | * register_netdevice(x1); | |
6240 | * register_netdevice(x2); | |
6241 | * ... | |
6242 | * unregister_netdevice(y1); | |
6243 | * unregister_netdevice(y2); | |
6244 | * ... | |
6245 | * rtnl_unlock(); | |
6246 | * free_netdev(y1); | |
6247 | * free_netdev(y2); | |
6248 | * | |
58ec3b4d | 6249 | * We are invoked by rtnl_unlock(). |
1da177e4 | 6250 | * This allows us to deal with problems: |
b17a7c17 | 6251 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
6252 | * without deadlocking with linkwatch via keventd. |
6253 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
6254 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
6255 | * |
6256 | * We must not return until all unregister events added during | |
6257 | * the interval the lock was held have been completed. | |
1da177e4 | 6258 | */ |
1da177e4 LT |
6259 | void netdev_run_todo(void) |
6260 | { | |
626ab0e6 | 6261 | struct list_head list; |
1da177e4 | 6262 | |
1da177e4 | 6263 | /* Snapshot list, allow later requests */ |
626ab0e6 | 6264 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
6265 | |
6266 | __rtnl_unlock(); | |
626ab0e6 | 6267 | |
0115e8e3 ED |
6268 | |
6269 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
6270 | if (!list_empty(&list)) |
6271 | rcu_barrier(); | |
6272 | ||
1da177e4 LT |
6273 | while (!list_empty(&list)) { |
6274 | struct net_device *dev | |
e5e26d75 | 6275 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
6276 | list_del(&dev->todo_list); |
6277 | ||
748e2d93 | 6278 | rtnl_lock(); |
0115e8e3 | 6279 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 6280 | __rtnl_unlock(); |
0115e8e3 | 6281 | |
b17a7c17 | 6282 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 6283 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
6284 | dev->name, dev->reg_state); |
6285 | dump_stack(); | |
6286 | continue; | |
6287 | } | |
1da177e4 | 6288 | |
b17a7c17 | 6289 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 6290 | |
152102c7 | 6291 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 6292 | |
b17a7c17 | 6293 | netdev_wait_allrefs(dev); |
1da177e4 | 6294 | |
b17a7c17 | 6295 | /* paranoia */ |
29b4433d | 6296 | BUG_ON(netdev_refcnt_read(dev)); |
33d480ce ED |
6297 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
6298 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 6299 | WARN_ON(dev->dn_ptr); |
1da177e4 | 6300 | |
b17a7c17 SH |
6301 | if (dev->destructor) |
6302 | dev->destructor(dev); | |
9093bbb2 | 6303 | |
50624c93 EB |
6304 | /* Report a network device has been unregistered */ |
6305 | rtnl_lock(); | |
6306 | dev_net(dev)->dev_unreg_count--; | |
6307 | __rtnl_unlock(); | |
6308 | wake_up(&netdev_unregistering_wq); | |
6309 | ||
9093bbb2 SH |
6310 | /* Free network device */ |
6311 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 6312 | } |
1da177e4 LT |
6313 | } |
6314 | ||
3cfde79c BH |
6315 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
6316 | * fields in the same order, with only the type differing. | |
6317 | */ | |
77a1abf5 ED |
6318 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
6319 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
6320 | { |
6321 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
6322 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
6323 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
6324 | #else |
6325 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
6326 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
6327 | u64 *dst = (u64 *)stats64; | |
6328 | ||
6329 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
6330 | sizeof(*stats64) / sizeof(u64)); | |
6331 | for (i = 0; i < n; i++) | |
6332 | dst[i] = src[i]; | |
6333 | #endif | |
6334 | } | |
77a1abf5 | 6335 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 6336 | |
eeda3fd6 SH |
6337 | /** |
6338 | * dev_get_stats - get network device statistics | |
6339 | * @dev: device to get statistics from | |
28172739 | 6340 | * @storage: place to store stats |
eeda3fd6 | 6341 | * |
d7753516 BH |
6342 | * Get network statistics from device. Return @storage. |
6343 | * The device driver may provide its own method by setting | |
6344 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
6345 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 6346 | */ |
d7753516 BH |
6347 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
6348 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 6349 | { |
eeda3fd6 SH |
6350 | const struct net_device_ops *ops = dev->netdev_ops; |
6351 | ||
28172739 ED |
6352 | if (ops->ndo_get_stats64) { |
6353 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
6354 | ops->ndo_get_stats64(dev, storage); |
6355 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 6356 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
6357 | } else { |
6358 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 6359 | } |
caf586e5 | 6360 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
015f0688 | 6361 | storage->tx_dropped += atomic_long_read(&dev->tx_dropped); |
28172739 | 6362 | return storage; |
c45d286e | 6363 | } |
eeda3fd6 | 6364 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 6365 | |
24824a09 | 6366 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 6367 | { |
24824a09 | 6368 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 6369 | |
24824a09 ED |
6370 | #ifdef CONFIG_NET_CLS_ACT |
6371 | if (queue) | |
6372 | return queue; | |
6373 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
6374 | if (!queue) | |
6375 | return NULL; | |
6376 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
6377 | queue->qdisc = &noop_qdisc; |
6378 | queue->qdisc_sleeping = &noop_qdisc; | |
6379 | rcu_assign_pointer(dev->ingress_queue, queue); | |
6380 | #endif | |
6381 | return queue; | |
bb949fbd DM |
6382 | } |
6383 | ||
2c60db03 ED |
6384 | static const struct ethtool_ops default_ethtool_ops; |
6385 | ||
d07d7507 SG |
6386 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
6387 | const struct ethtool_ops *ops) | |
6388 | { | |
6389 | if (dev->ethtool_ops == &default_ethtool_ops) | |
6390 | dev->ethtool_ops = ops; | |
6391 | } | |
6392 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
6393 | ||
74d332c1 ED |
6394 | void netdev_freemem(struct net_device *dev) |
6395 | { | |
6396 | char *addr = (char *)dev - dev->padded; | |
6397 | ||
6398 | if (is_vmalloc_addr(addr)) | |
6399 | vfree(addr); | |
6400 | else | |
6401 | kfree(addr); | |
6402 | } | |
6403 | ||
1da177e4 | 6404 | /** |
36909ea4 | 6405 | * alloc_netdev_mqs - allocate network device |
1da177e4 LT |
6406 | * @sizeof_priv: size of private data to allocate space for |
6407 | * @name: device name format string | |
6408 | * @setup: callback to initialize device | |
36909ea4 TH |
6409 | * @txqs: the number of TX subqueues to allocate |
6410 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
6411 | * |
6412 | * Allocates a struct net_device with private data area for driver use | |
90e51adf | 6413 | * and performs basic initialization. Also allocates subqueue structs |
36909ea4 | 6414 | * for each queue on the device. |
1da177e4 | 6415 | */ |
36909ea4 TH |
6416 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
6417 | void (*setup)(struct net_device *), | |
6418 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 6419 | { |
1da177e4 | 6420 | struct net_device *dev; |
7943986c | 6421 | size_t alloc_size; |
1ce8e7b5 | 6422 | struct net_device *p; |
1da177e4 | 6423 | |
b6fe17d6 SH |
6424 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
6425 | ||
36909ea4 | 6426 | if (txqs < 1) { |
7b6cd1ce | 6427 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
6428 | return NULL; |
6429 | } | |
6430 | ||
a953be53 | 6431 | #ifdef CONFIG_SYSFS |
36909ea4 | 6432 | if (rxqs < 1) { |
7b6cd1ce | 6433 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
6434 | return NULL; |
6435 | } | |
6436 | #endif | |
6437 | ||
fd2ea0a7 | 6438 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
6439 | if (sizeof_priv) { |
6440 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 6441 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
6442 | alloc_size += sizeof_priv; |
6443 | } | |
6444 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 6445 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 6446 | |
74d332c1 ED |
6447 | p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6448 | if (!p) | |
6449 | p = vzalloc(alloc_size); | |
62b5942a | 6450 | if (!p) |
1da177e4 | 6451 | return NULL; |
1da177e4 | 6452 | |
1ce8e7b5 | 6453 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 6454 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 6455 | |
29b4433d ED |
6456 | dev->pcpu_refcnt = alloc_percpu(int); |
6457 | if (!dev->pcpu_refcnt) | |
74d332c1 | 6458 | goto free_dev; |
ab9c73cc | 6459 | |
ab9c73cc | 6460 | if (dev_addr_init(dev)) |
29b4433d | 6461 | goto free_pcpu; |
ab9c73cc | 6462 | |
22bedad3 | 6463 | dev_mc_init(dev); |
a748ee24 | 6464 | dev_uc_init(dev); |
ccffad25 | 6465 | |
c346dca1 | 6466 | dev_net_set(dev, &init_net); |
1da177e4 | 6467 | |
8d3bdbd5 | 6468 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 6469 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 6470 | |
8d3bdbd5 DM |
6471 | INIT_LIST_HEAD(&dev->napi_list); |
6472 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 6473 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 6474 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
6475 | INIT_LIST_HEAD(&dev->adj_list.upper); |
6476 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
6477 | INIT_LIST_HEAD(&dev->all_adj_list.upper); | |
6478 | INIT_LIST_HEAD(&dev->all_adj_list.lower); | |
8d3bdbd5 DM |
6479 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
6480 | setup(dev); | |
6481 | ||
36909ea4 TH |
6482 | dev->num_tx_queues = txqs; |
6483 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 6484 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 6485 | goto free_all; |
e8a0464c | 6486 | |
a953be53 | 6487 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
6488 | dev->num_rx_queues = rxqs; |
6489 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 6490 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 6491 | goto free_all; |
df334545 | 6492 | #endif |
0a9627f2 | 6493 | |
1da177e4 | 6494 | strcpy(dev->name, name); |
cbda10fa | 6495 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
6496 | if (!dev->ethtool_ops) |
6497 | dev->ethtool_ops = &default_ethtool_ops; | |
1da177e4 | 6498 | return dev; |
ab9c73cc | 6499 | |
8d3bdbd5 DM |
6500 | free_all: |
6501 | free_netdev(dev); | |
6502 | return NULL; | |
6503 | ||
29b4433d ED |
6504 | free_pcpu: |
6505 | free_percpu(dev->pcpu_refcnt); | |
60877a32 | 6506 | netif_free_tx_queues(dev); |
a953be53 | 6507 | #ifdef CONFIG_SYSFS |
fe822240 TH |
6508 | kfree(dev->_rx); |
6509 | #endif | |
6510 | ||
74d332c1 ED |
6511 | free_dev: |
6512 | netdev_freemem(dev); | |
ab9c73cc | 6513 | return NULL; |
1da177e4 | 6514 | } |
36909ea4 | 6515 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
6516 | |
6517 | /** | |
6518 | * free_netdev - free network device | |
6519 | * @dev: device | |
6520 | * | |
4ec93edb YH |
6521 | * This function does the last stage of destroying an allocated device |
6522 | * interface. The reference to the device object is released. | |
1da177e4 LT |
6523 | * If this is the last reference then it will be freed. |
6524 | */ | |
6525 | void free_netdev(struct net_device *dev) | |
6526 | { | |
d565b0a1 HX |
6527 | struct napi_struct *p, *n; |
6528 | ||
f3005d7f DL |
6529 | release_net(dev_net(dev)); |
6530 | ||
60877a32 | 6531 | netif_free_tx_queues(dev); |
a953be53 | 6532 | #ifdef CONFIG_SYSFS |
fe822240 TH |
6533 | kfree(dev->_rx); |
6534 | #endif | |
e8a0464c | 6535 | |
33d480ce | 6536 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 6537 | |
f001fde5 JP |
6538 | /* Flush device addresses */ |
6539 | dev_addr_flush(dev); | |
6540 | ||
d565b0a1 HX |
6541 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
6542 | netif_napi_del(p); | |
6543 | ||
29b4433d ED |
6544 | free_percpu(dev->pcpu_refcnt); |
6545 | dev->pcpu_refcnt = NULL; | |
6546 | ||
3041a069 | 6547 | /* Compatibility with error handling in drivers */ |
1da177e4 | 6548 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 6549 | netdev_freemem(dev); |
1da177e4 LT |
6550 | return; |
6551 | } | |
6552 | ||
6553 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
6554 | dev->reg_state = NETREG_RELEASED; | |
6555 | ||
43cb76d9 GKH |
6556 | /* will free via device release */ |
6557 | put_device(&dev->dev); | |
1da177e4 | 6558 | } |
d1b19dff | 6559 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 6560 | |
f0db275a SH |
6561 | /** |
6562 | * synchronize_net - Synchronize with packet receive processing | |
6563 | * | |
6564 | * Wait for packets currently being received to be done. | |
6565 | * Does not block later packets from starting. | |
6566 | */ | |
4ec93edb | 6567 | void synchronize_net(void) |
1da177e4 LT |
6568 | { |
6569 | might_sleep(); | |
be3fc413 ED |
6570 | if (rtnl_is_locked()) |
6571 | synchronize_rcu_expedited(); | |
6572 | else | |
6573 | synchronize_rcu(); | |
1da177e4 | 6574 | } |
d1b19dff | 6575 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
6576 | |
6577 | /** | |
44a0873d | 6578 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 6579 | * @dev: device |
44a0873d | 6580 | * @head: list |
6ebfbc06 | 6581 | * |
1da177e4 | 6582 | * This function shuts down a device interface and removes it |
d59b54b1 | 6583 | * from the kernel tables. |
44a0873d | 6584 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
6585 | * |
6586 | * Callers must hold the rtnl semaphore. You may want | |
6587 | * unregister_netdev() instead of this. | |
6588 | */ | |
6589 | ||
44a0873d | 6590 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 6591 | { |
a6620712 HX |
6592 | ASSERT_RTNL(); |
6593 | ||
44a0873d | 6594 | if (head) { |
9fdce099 | 6595 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
6596 | } else { |
6597 | rollback_registered(dev); | |
6598 | /* Finish processing unregister after unlock */ | |
6599 | net_set_todo(dev); | |
6600 | } | |
1da177e4 | 6601 | } |
44a0873d | 6602 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 6603 | |
9b5e383c ED |
6604 | /** |
6605 | * unregister_netdevice_many - unregister many devices | |
6606 | * @head: list of devices | |
9b5e383c ED |
6607 | */ |
6608 | void unregister_netdevice_many(struct list_head *head) | |
6609 | { | |
6610 | struct net_device *dev; | |
6611 | ||
6612 | if (!list_empty(head)) { | |
6613 | rollback_registered_many(head); | |
6614 | list_for_each_entry(dev, head, unreg_list) | |
6615 | net_set_todo(dev); | |
6616 | } | |
6617 | } | |
63c8099d | 6618 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 6619 | |
1da177e4 LT |
6620 | /** |
6621 | * unregister_netdev - remove device from the kernel | |
6622 | * @dev: device | |
6623 | * | |
6624 | * This function shuts down a device interface and removes it | |
d59b54b1 | 6625 | * from the kernel tables. |
1da177e4 LT |
6626 | * |
6627 | * This is just a wrapper for unregister_netdevice that takes | |
6628 | * the rtnl semaphore. In general you want to use this and not | |
6629 | * unregister_netdevice. | |
6630 | */ | |
6631 | void unregister_netdev(struct net_device *dev) | |
6632 | { | |
6633 | rtnl_lock(); | |
6634 | unregister_netdevice(dev); | |
6635 | rtnl_unlock(); | |
6636 | } | |
1da177e4 LT |
6637 | EXPORT_SYMBOL(unregister_netdev); |
6638 | ||
ce286d32 EB |
6639 | /** |
6640 | * dev_change_net_namespace - move device to different nethost namespace | |
6641 | * @dev: device | |
6642 | * @net: network namespace | |
6643 | * @pat: If not NULL name pattern to try if the current device name | |
6644 | * is already taken in the destination network namespace. | |
6645 | * | |
6646 | * This function shuts down a device interface and moves it | |
6647 | * to a new network namespace. On success 0 is returned, on | |
6648 | * a failure a netagive errno code is returned. | |
6649 | * | |
6650 | * Callers must hold the rtnl semaphore. | |
6651 | */ | |
6652 | ||
6653 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
6654 | { | |
ce286d32 EB |
6655 | int err; |
6656 | ||
6657 | ASSERT_RTNL(); | |
6658 | ||
6659 | /* Don't allow namespace local devices to be moved. */ | |
6660 | err = -EINVAL; | |
6661 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6662 | goto out; | |
6663 | ||
6664 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
6665 | if (dev->reg_state != NETREG_REGISTERED) |
6666 | goto out; | |
6667 | ||
6668 | /* Get out if there is nothing todo */ | |
6669 | err = 0; | |
878628fb | 6670 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
6671 | goto out; |
6672 | ||
6673 | /* Pick the destination device name, and ensure | |
6674 | * we can use it in the destination network namespace. | |
6675 | */ | |
6676 | err = -EEXIST; | |
d9031024 | 6677 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
6678 | /* We get here if we can't use the current device name */ |
6679 | if (!pat) | |
6680 | goto out; | |
828de4f6 | 6681 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
6682 | goto out; |
6683 | } | |
6684 | ||
6685 | /* | |
6686 | * And now a mini version of register_netdevice unregister_netdevice. | |
6687 | */ | |
6688 | ||
6689 | /* If device is running close it first. */ | |
9b772652 | 6690 | dev_close(dev); |
ce286d32 EB |
6691 | |
6692 | /* And unlink it from device chain */ | |
6693 | err = -ENODEV; | |
6694 | unlist_netdevice(dev); | |
6695 | ||
6696 | synchronize_net(); | |
6697 | ||
6698 | /* Shutdown queueing discipline. */ | |
6699 | dev_shutdown(dev); | |
6700 | ||
6701 | /* Notify protocols, that we are about to destroy | |
6702 | this device. They should clean all the things. | |
3b27e105 DL |
6703 | |
6704 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
6705 | This is wanted because this way 8021q and macvlan know | |
6706 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
6707 | */ |
6708 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
6709 | rcu_barrier(); |
6710 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 6711 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
6712 | |
6713 | /* | |
6714 | * Flush the unicast and multicast chains | |
6715 | */ | |
a748ee24 | 6716 | dev_uc_flush(dev); |
22bedad3 | 6717 | dev_mc_flush(dev); |
ce286d32 | 6718 | |
4e66ae2e SH |
6719 | /* Send a netdev-removed uevent to the old namespace */ |
6720 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
6721 | ||
ce286d32 | 6722 | /* Actually switch the network namespace */ |
c346dca1 | 6723 | dev_net_set(dev, net); |
ce286d32 | 6724 | |
ce286d32 EB |
6725 | /* If there is an ifindex conflict assign a new one */ |
6726 | if (__dev_get_by_index(net, dev->ifindex)) { | |
6727 | int iflink = (dev->iflink == dev->ifindex); | |
6728 | dev->ifindex = dev_new_index(net); | |
6729 | if (iflink) | |
6730 | dev->iflink = dev->ifindex; | |
6731 | } | |
6732 | ||
4e66ae2e SH |
6733 | /* Send a netdev-add uevent to the new namespace */ |
6734 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
6735 | ||
8b41d188 | 6736 | /* Fixup kobjects */ |
a1b3f594 | 6737 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 6738 | WARN_ON(err); |
ce286d32 EB |
6739 | |
6740 | /* Add the device back in the hashes */ | |
6741 | list_netdevice(dev); | |
6742 | ||
6743 | /* Notify protocols, that a new device appeared. */ | |
6744 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
6745 | ||
d90a909e EB |
6746 | /* |
6747 | * Prevent userspace races by waiting until the network | |
6748 | * device is fully setup before sending notifications. | |
6749 | */ | |
7f294054 | 6750 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 6751 | |
ce286d32 EB |
6752 | synchronize_net(); |
6753 | err = 0; | |
6754 | out: | |
6755 | return err; | |
6756 | } | |
463d0183 | 6757 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 6758 | |
1da177e4 LT |
6759 | static int dev_cpu_callback(struct notifier_block *nfb, |
6760 | unsigned long action, | |
6761 | void *ocpu) | |
6762 | { | |
6763 | struct sk_buff **list_skb; | |
1da177e4 LT |
6764 | struct sk_buff *skb; |
6765 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
6766 | struct softnet_data *sd, *oldsd; | |
6767 | ||
8bb78442 | 6768 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
6769 | return NOTIFY_OK; |
6770 | ||
6771 | local_irq_disable(); | |
6772 | cpu = smp_processor_id(); | |
6773 | sd = &per_cpu(softnet_data, cpu); | |
6774 | oldsd = &per_cpu(softnet_data, oldcpu); | |
6775 | ||
6776 | /* Find end of our completion_queue. */ | |
6777 | list_skb = &sd->completion_queue; | |
6778 | while (*list_skb) | |
6779 | list_skb = &(*list_skb)->next; | |
6780 | /* Append completion queue from offline CPU. */ | |
6781 | *list_skb = oldsd->completion_queue; | |
6782 | oldsd->completion_queue = NULL; | |
6783 | ||
1da177e4 | 6784 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
6785 | if (oldsd->output_queue) { |
6786 | *sd->output_queue_tailp = oldsd->output_queue; | |
6787 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
6788 | oldsd->output_queue = NULL; | |
6789 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
6790 | } | |
264524d5 HC |
6791 | /* Append NAPI poll list from offline CPU. */ |
6792 | if (!list_empty(&oldsd->poll_list)) { | |
6793 | list_splice_init(&oldsd->poll_list, &sd->poll_list); | |
6794 | raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6795 | } | |
1da177e4 LT |
6796 | |
6797 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
6798 | local_irq_enable(); | |
6799 | ||
6800 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 6801 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
ae78dbfa | 6802 | netif_rx_internal(skb); |
76cc8b13 | 6803 | input_queue_head_incr(oldsd); |
fec5e652 | 6804 | } |
76cc8b13 | 6805 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
ae78dbfa | 6806 | netif_rx_internal(skb); |
76cc8b13 TH |
6807 | input_queue_head_incr(oldsd); |
6808 | } | |
1da177e4 LT |
6809 | |
6810 | return NOTIFY_OK; | |
6811 | } | |
1da177e4 LT |
6812 | |
6813 | ||
7f353bf2 | 6814 | /** |
b63365a2 HX |
6815 | * netdev_increment_features - increment feature set by one |
6816 | * @all: current feature set | |
6817 | * @one: new feature set | |
6818 | * @mask: mask feature set | |
7f353bf2 HX |
6819 | * |
6820 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
6821 | * @one to the master device with current feature set @all. Will not |
6822 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 6823 | */ |
c8f44aff MM |
6824 | netdev_features_t netdev_increment_features(netdev_features_t all, |
6825 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 6826 | { |
1742f183 MM |
6827 | if (mask & NETIF_F_GEN_CSUM) |
6828 | mask |= NETIF_F_ALL_CSUM; | |
6829 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 6830 | |
1742f183 MM |
6831 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
6832 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 6833 | |
1742f183 MM |
6834 | /* If one device supports hw checksumming, set for all. */ |
6835 | if (all & NETIF_F_GEN_CSUM) | |
6836 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
6837 | |
6838 | return all; | |
6839 | } | |
b63365a2 | 6840 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 6841 | |
430f03cd | 6842 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
6843 | { |
6844 | int i; | |
6845 | struct hlist_head *hash; | |
6846 | ||
6847 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
6848 | if (hash != NULL) | |
6849 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
6850 | INIT_HLIST_HEAD(&hash[i]); | |
6851 | ||
6852 | return hash; | |
6853 | } | |
6854 | ||
881d966b | 6855 | /* Initialize per network namespace state */ |
4665079c | 6856 | static int __net_init netdev_init(struct net *net) |
881d966b | 6857 | { |
734b6541 RM |
6858 | if (net != &init_net) |
6859 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 6860 | |
30d97d35 PE |
6861 | net->dev_name_head = netdev_create_hash(); |
6862 | if (net->dev_name_head == NULL) | |
6863 | goto err_name; | |
881d966b | 6864 | |
30d97d35 PE |
6865 | net->dev_index_head = netdev_create_hash(); |
6866 | if (net->dev_index_head == NULL) | |
6867 | goto err_idx; | |
881d966b EB |
6868 | |
6869 | return 0; | |
30d97d35 PE |
6870 | |
6871 | err_idx: | |
6872 | kfree(net->dev_name_head); | |
6873 | err_name: | |
6874 | return -ENOMEM; | |
881d966b EB |
6875 | } |
6876 | ||
f0db275a SH |
6877 | /** |
6878 | * netdev_drivername - network driver for the device | |
6879 | * @dev: network device | |
f0db275a SH |
6880 | * |
6881 | * Determine network driver for device. | |
6882 | */ | |
3019de12 | 6883 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 6884 | { |
cf04a4c7 SH |
6885 | const struct device_driver *driver; |
6886 | const struct device *parent; | |
3019de12 | 6887 | const char *empty = ""; |
6579e57b AV |
6888 | |
6889 | parent = dev->dev.parent; | |
6579e57b | 6890 | if (!parent) |
3019de12 | 6891 | return empty; |
6579e57b AV |
6892 | |
6893 | driver = parent->driver; | |
6894 | if (driver && driver->name) | |
3019de12 DM |
6895 | return driver->name; |
6896 | return empty; | |
6579e57b AV |
6897 | } |
6898 | ||
b004ff49 | 6899 | static int __netdev_printk(const char *level, const struct net_device *dev, |
256df2f3 JP |
6900 | struct va_format *vaf) |
6901 | { | |
6902 | int r; | |
6903 | ||
b004ff49 | 6904 | if (dev && dev->dev.parent) { |
666f355f JP |
6905 | r = dev_printk_emit(level[1] - '0', |
6906 | dev->dev.parent, | |
6907 | "%s %s %s: %pV", | |
6908 | dev_driver_string(dev->dev.parent), | |
6909 | dev_name(dev->dev.parent), | |
6910 | netdev_name(dev), vaf); | |
b004ff49 | 6911 | } else if (dev) { |
256df2f3 | 6912 | r = printk("%s%s: %pV", level, netdev_name(dev), vaf); |
b004ff49 | 6913 | } else { |
256df2f3 | 6914 | r = printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 6915 | } |
256df2f3 JP |
6916 | |
6917 | return r; | |
6918 | } | |
6919 | ||
6920 | int netdev_printk(const char *level, const struct net_device *dev, | |
6921 | const char *format, ...) | |
6922 | { | |
6923 | struct va_format vaf; | |
6924 | va_list args; | |
6925 | int r; | |
6926 | ||
6927 | va_start(args, format); | |
6928 | ||
6929 | vaf.fmt = format; | |
6930 | vaf.va = &args; | |
6931 | ||
6932 | r = __netdev_printk(level, dev, &vaf); | |
b004ff49 | 6933 | |
256df2f3 JP |
6934 | va_end(args); |
6935 | ||
6936 | return r; | |
6937 | } | |
6938 | EXPORT_SYMBOL(netdev_printk); | |
6939 | ||
6940 | #define define_netdev_printk_level(func, level) \ | |
6941 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
6942 | { \ | |
6943 | int r; \ | |
6944 | struct va_format vaf; \ | |
6945 | va_list args; \ | |
6946 | \ | |
6947 | va_start(args, fmt); \ | |
6948 | \ | |
6949 | vaf.fmt = fmt; \ | |
6950 | vaf.va = &args; \ | |
6951 | \ | |
6952 | r = __netdev_printk(level, dev, &vaf); \ | |
b004ff49 | 6953 | \ |
256df2f3 JP |
6954 | va_end(args); \ |
6955 | \ | |
6956 | return r; \ | |
6957 | } \ | |
6958 | EXPORT_SYMBOL(func); | |
6959 | ||
6960 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
6961 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
6962 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
6963 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
6964 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
6965 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
6966 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
6967 | ||
4665079c | 6968 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
6969 | { |
6970 | kfree(net->dev_name_head); | |
6971 | kfree(net->dev_index_head); | |
6972 | } | |
6973 | ||
022cbae6 | 6974 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
6975 | .init = netdev_init, |
6976 | .exit = netdev_exit, | |
6977 | }; | |
6978 | ||
4665079c | 6979 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 6980 | { |
e008b5fc | 6981 | struct net_device *dev, *aux; |
ce286d32 | 6982 | /* |
e008b5fc | 6983 | * Push all migratable network devices back to the |
ce286d32 EB |
6984 | * initial network namespace |
6985 | */ | |
6986 | rtnl_lock(); | |
e008b5fc | 6987 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 6988 | int err; |
aca51397 | 6989 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
6990 | |
6991 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6992 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6993 | continue; | |
6994 | ||
e008b5fc EB |
6995 | /* Leave virtual devices for the generic cleanup */ |
6996 | if (dev->rtnl_link_ops) | |
6997 | continue; | |
d0c082ce | 6998 | |
25985edc | 6999 | /* Push remaining network devices to init_net */ |
aca51397 PE |
7000 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
7001 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 7002 | if (err) { |
7b6cd1ce JP |
7003 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
7004 | __func__, dev->name, err); | |
aca51397 | 7005 | BUG(); |
ce286d32 EB |
7006 | } |
7007 | } | |
7008 | rtnl_unlock(); | |
7009 | } | |
7010 | ||
50624c93 EB |
7011 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
7012 | { | |
7013 | /* Return with the rtnl_lock held when there are no network | |
7014 | * devices unregistering in any network namespace in net_list. | |
7015 | */ | |
7016 | struct net *net; | |
7017 | bool unregistering; | |
7018 | DEFINE_WAIT(wait); | |
7019 | ||
7020 | for (;;) { | |
7021 | prepare_to_wait(&netdev_unregistering_wq, &wait, | |
7022 | TASK_UNINTERRUPTIBLE); | |
7023 | unregistering = false; | |
7024 | rtnl_lock(); | |
7025 | list_for_each_entry(net, net_list, exit_list) { | |
7026 | if (net->dev_unreg_count > 0) { | |
7027 | unregistering = true; | |
7028 | break; | |
7029 | } | |
7030 | } | |
7031 | if (!unregistering) | |
7032 | break; | |
7033 | __rtnl_unlock(); | |
7034 | schedule(); | |
7035 | } | |
7036 | finish_wait(&netdev_unregistering_wq, &wait); | |
7037 | } | |
7038 | ||
04dc7f6b EB |
7039 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
7040 | { | |
7041 | /* At exit all network devices most be removed from a network | |
b595076a | 7042 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
7043 | * Do this across as many network namespaces as possible to |
7044 | * improve batching efficiency. | |
7045 | */ | |
7046 | struct net_device *dev; | |
7047 | struct net *net; | |
7048 | LIST_HEAD(dev_kill_list); | |
7049 | ||
50624c93 EB |
7050 | /* To prevent network device cleanup code from dereferencing |
7051 | * loopback devices or network devices that have been freed | |
7052 | * wait here for all pending unregistrations to complete, | |
7053 | * before unregistring the loopback device and allowing the | |
7054 | * network namespace be freed. | |
7055 | * | |
7056 | * The netdev todo list containing all network devices | |
7057 | * unregistrations that happen in default_device_exit_batch | |
7058 | * will run in the rtnl_unlock() at the end of | |
7059 | * default_device_exit_batch. | |
7060 | */ | |
7061 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
7062 | list_for_each_entry(net, net_list, exit_list) { |
7063 | for_each_netdev_reverse(net, dev) { | |
7064 | if (dev->rtnl_link_ops) | |
7065 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
7066 | else | |
7067 | unregister_netdevice_queue(dev, &dev_kill_list); | |
7068 | } | |
7069 | } | |
7070 | unregister_netdevice_many(&dev_kill_list); | |
ceaaec98 | 7071 | list_del(&dev_kill_list); |
04dc7f6b EB |
7072 | rtnl_unlock(); |
7073 | } | |
7074 | ||
022cbae6 | 7075 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 7076 | .exit = default_device_exit, |
04dc7f6b | 7077 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
7078 | }; |
7079 | ||
1da177e4 LT |
7080 | /* |
7081 | * Initialize the DEV module. At boot time this walks the device list and | |
7082 | * unhooks any devices that fail to initialise (normally hardware not | |
7083 | * present) and leaves us with a valid list of present and active devices. | |
7084 | * | |
7085 | */ | |
7086 | ||
7087 | /* | |
7088 | * This is called single threaded during boot, so no need | |
7089 | * to take the rtnl semaphore. | |
7090 | */ | |
7091 | static int __init net_dev_init(void) | |
7092 | { | |
7093 | int i, rc = -ENOMEM; | |
7094 | ||
7095 | BUG_ON(!dev_boot_phase); | |
7096 | ||
1da177e4 LT |
7097 | if (dev_proc_init()) |
7098 | goto out; | |
7099 | ||
8b41d188 | 7100 | if (netdev_kobject_init()) |
1da177e4 LT |
7101 | goto out; |
7102 | ||
7103 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 7104 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
7105 | INIT_LIST_HEAD(&ptype_base[i]); |
7106 | ||
62532da9 VY |
7107 | INIT_LIST_HEAD(&offload_base); |
7108 | ||
881d966b EB |
7109 | if (register_pernet_subsys(&netdev_net_ops)) |
7110 | goto out; | |
1da177e4 LT |
7111 | |
7112 | /* | |
7113 | * Initialise the packet receive queues. | |
7114 | */ | |
7115 | ||
6f912042 | 7116 | for_each_possible_cpu(i) { |
e36fa2f7 | 7117 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 7118 | |
e36fa2f7 | 7119 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 7120 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 7121 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 7122 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 7123 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
7124 | sd->csd.func = rps_trigger_softirq; |
7125 | sd->csd.info = sd; | |
e36fa2f7 | 7126 | sd->cpu = i; |
1e94d72f | 7127 | #endif |
0a9627f2 | 7128 | |
e36fa2f7 ED |
7129 | sd->backlog.poll = process_backlog; |
7130 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
7131 | } |
7132 | ||
1da177e4 LT |
7133 | dev_boot_phase = 0; |
7134 | ||
505d4f73 EB |
7135 | /* The loopback device is special if any other network devices |
7136 | * is present in a network namespace the loopback device must | |
7137 | * be present. Since we now dynamically allocate and free the | |
7138 | * loopback device ensure this invariant is maintained by | |
7139 | * keeping the loopback device as the first device on the | |
7140 | * list of network devices. Ensuring the loopback devices | |
7141 | * is the first device that appears and the last network device | |
7142 | * that disappears. | |
7143 | */ | |
7144 | if (register_pernet_device(&loopback_net_ops)) | |
7145 | goto out; | |
7146 | ||
7147 | if (register_pernet_device(&default_device_ops)) | |
7148 | goto out; | |
7149 | ||
962cf36c CM |
7150 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
7151 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
7152 | |
7153 | hotcpu_notifier(dev_cpu_callback, 0); | |
7154 | dst_init(); | |
1da177e4 LT |
7155 | rc = 0; |
7156 | out: | |
7157 | return rc; | |
7158 | } | |
7159 | ||
7160 | subsys_initcall(net_dev_init); |