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> |
1da177e4 | 134 | |
342709ef PE |
135 | #include "net-sysfs.h" |
136 | ||
d565b0a1 HX |
137 | /* Instead of increasing this, you should create a hash table. */ |
138 | #define MAX_GRO_SKBS 8 | |
139 | ||
5d38a079 HX |
140 | /* This should be increased if a protocol with a bigger head is added. */ |
141 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
142 | ||
1da177e4 | 143 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 144 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
145 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
146 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 147 | static struct list_head offload_base __read_mostly; |
1da177e4 | 148 | |
1da177e4 | 149 | /* |
7562f876 | 150 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
151 | * semaphore. |
152 | * | |
c6d14c84 | 153 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
154 | * |
155 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 156 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
157 | * actual updates. This allows pure readers to access the list even |
158 | * while a writer is preparing to update it. | |
159 | * | |
160 | * To put it another way, dev_base_lock is held for writing only to | |
161 | * protect against pure readers; the rtnl semaphore provides the | |
162 | * protection against other writers. | |
163 | * | |
164 | * See, for example usages, register_netdevice() and | |
165 | * unregister_netdevice(), which must be called with the rtnl | |
166 | * semaphore held. | |
167 | */ | |
1da177e4 | 168 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
169 | EXPORT_SYMBOL(dev_base_lock); |
170 | ||
af12fa6e ET |
171 | /* protects napi_hash addition/deletion and napi_gen_id */ |
172 | static DEFINE_SPINLOCK(napi_hash_lock); | |
173 | ||
174 | static unsigned int napi_gen_id; | |
175 | static DEFINE_HASHTABLE(napi_hash, 8); | |
176 | ||
18afa4b0 | 177 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 178 | |
4e985ada TG |
179 | static inline void dev_base_seq_inc(struct net *net) |
180 | { | |
181 | while (++net->dev_base_seq == 0); | |
182 | } | |
183 | ||
881d966b | 184 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 185 | { |
95c96174 ED |
186 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
187 | ||
08e9897d | 188 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
189 | } |
190 | ||
881d966b | 191 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 192 | { |
7c28bd0b | 193 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
194 | } |
195 | ||
e36fa2f7 | 196 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
197 | { |
198 | #ifdef CONFIG_RPS | |
e36fa2f7 | 199 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
200 | #endif |
201 | } | |
202 | ||
e36fa2f7 | 203 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
204 | { |
205 | #ifdef CONFIG_RPS | |
e36fa2f7 | 206 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
207 | #endif |
208 | } | |
209 | ||
ce286d32 | 210 | /* Device list insertion */ |
53759be9 | 211 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 212 | { |
c346dca1 | 213 | struct net *net = dev_net(dev); |
ce286d32 EB |
214 | |
215 | ASSERT_RTNL(); | |
216 | ||
217 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 218 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 219 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
220 | hlist_add_head_rcu(&dev->index_hlist, |
221 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 222 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
223 | |
224 | dev_base_seq_inc(net); | |
ce286d32 EB |
225 | } |
226 | ||
fb699dfd ED |
227 | /* Device list removal |
228 | * caller must respect a RCU grace period before freeing/reusing dev | |
229 | */ | |
ce286d32 EB |
230 | static void unlist_netdevice(struct net_device *dev) |
231 | { | |
232 | ASSERT_RTNL(); | |
233 | ||
234 | /* Unlink dev from the device chain */ | |
235 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 236 | list_del_rcu(&dev->dev_list); |
72c9528b | 237 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 238 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 239 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
240 | |
241 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
242 | } |
243 | ||
1da177e4 LT |
244 | /* |
245 | * Our notifier list | |
246 | */ | |
247 | ||
f07d5b94 | 248 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
249 | |
250 | /* | |
251 | * Device drivers call our routines to queue packets here. We empty the | |
252 | * queue in the local softnet handler. | |
253 | */ | |
bea3348e | 254 | |
9958da05 | 255 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 256 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 257 | |
cf508b12 | 258 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 259 | /* |
c773e847 | 260 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
261 | * according to dev->type |
262 | */ | |
263 | static const unsigned short netdev_lock_type[] = | |
264 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
265 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
266 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
267 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
268 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
269 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
270 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
271 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
272 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
273 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
274 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
275 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
276 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
277 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
278 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 279 | |
36cbd3dc | 280 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
281 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
282 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
283 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
284 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
285 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
286 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
287 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
288 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
289 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
290 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
291 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
292 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
293 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
294 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
295 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
296 | |
297 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 298 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
299 | |
300 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
301 | { | |
302 | int i; | |
303 | ||
304 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
305 | if (netdev_lock_type[i] == dev_type) | |
306 | return i; | |
307 | /* the last key is used by default */ | |
308 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
309 | } | |
310 | ||
cf508b12 DM |
311 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
312 | unsigned short dev_type) | |
723e98b7 JP |
313 | { |
314 | int i; | |
315 | ||
316 | i = netdev_lock_pos(dev_type); | |
317 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
318 | netdev_lock_name[i]); | |
319 | } | |
cf508b12 DM |
320 | |
321 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
322 | { | |
323 | int i; | |
324 | ||
325 | i = netdev_lock_pos(dev->type); | |
326 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
327 | &netdev_addr_lock_key[i], | |
328 | netdev_lock_name[i]); | |
329 | } | |
723e98b7 | 330 | #else |
cf508b12 DM |
331 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
332 | unsigned short dev_type) | |
333 | { | |
334 | } | |
335 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
336 | { |
337 | } | |
338 | #endif | |
1da177e4 LT |
339 | |
340 | /******************************************************************************* | |
341 | ||
342 | Protocol management and registration routines | |
343 | ||
344 | *******************************************************************************/ | |
345 | ||
1da177e4 LT |
346 | /* |
347 | * Add a protocol ID to the list. Now that the input handler is | |
348 | * smarter we can dispense with all the messy stuff that used to be | |
349 | * here. | |
350 | * | |
351 | * BEWARE!!! Protocol handlers, mangling input packets, | |
352 | * MUST BE last in hash buckets and checking protocol handlers | |
353 | * MUST start from promiscuous ptype_all chain in net_bh. | |
354 | * It is true now, do not change it. | |
355 | * Explanation follows: if protocol handler, mangling packet, will | |
356 | * be the first on list, it is not able to sense, that packet | |
357 | * is cloned and should be copied-on-write, so that it will | |
358 | * change it and subsequent readers will get broken packet. | |
359 | * --ANK (980803) | |
360 | */ | |
361 | ||
c07b68e8 ED |
362 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
363 | { | |
364 | if (pt->type == htons(ETH_P_ALL)) | |
365 | return &ptype_all; | |
366 | else | |
367 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
368 | } | |
369 | ||
1da177e4 LT |
370 | /** |
371 | * dev_add_pack - add packet handler | |
372 | * @pt: packet type declaration | |
373 | * | |
374 | * Add a protocol handler to the networking stack. The passed &packet_type | |
375 | * is linked into kernel lists and may not be freed until it has been | |
376 | * removed from the kernel lists. | |
377 | * | |
4ec93edb | 378 | * This call does not sleep therefore it can not |
1da177e4 LT |
379 | * guarantee all CPU's that are in middle of receiving packets |
380 | * will see the new packet type (until the next received packet). | |
381 | */ | |
382 | ||
383 | void dev_add_pack(struct packet_type *pt) | |
384 | { | |
c07b68e8 | 385 | struct list_head *head = ptype_head(pt); |
1da177e4 | 386 | |
c07b68e8 ED |
387 | spin_lock(&ptype_lock); |
388 | list_add_rcu(&pt->list, head); | |
389 | spin_unlock(&ptype_lock); | |
1da177e4 | 390 | } |
d1b19dff | 391 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 392 | |
1da177e4 LT |
393 | /** |
394 | * __dev_remove_pack - remove packet handler | |
395 | * @pt: packet type declaration | |
396 | * | |
397 | * Remove a protocol handler that was previously added to the kernel | |
398 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
399 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 400 | * returns. |
1da177e4 LT |
401 | * |
402 | * The packet type might still be in use by receivers | |
403 | * and must not be freed until after all the CPU's have gone | |
404 | * through a quiescent state. | |
405 | */ | |
406 | void __dev_remove_pack(struct packet_type *pt) | |
407 | { | |
c07b68e8 | 408 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
409 | struct packet_type *pt1; |
410 | ||
c07b68e8 | 411 | spin_lock(&ptype_lock); |
1da177e4 LT |
412 | |
413 | list_for_each_entry(pt1, head, list) { | |
414 | if (pt == pt1) { | |
415 | list_del_rcu(&pt->list); | |
416 | goto out; | |
417 | } | |
418 | } | |
419 | ||
7b6cd1ce | 420 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 421 | out: |
c07b68e8 | 422 | spin_unlock(&ptype_lock); |
1da177e4 | 423 | } |
d1b19dff ED |
424 | EXPORT_SYMBOL(__dev_remove_pack); |
425 | ||
1da177e4 LT |
426 | /** |
427 | * dev_remove_pack - remove packet handler | |
428 | * @pt: packet type declaration | |
429 | * | |
430 | * Remove a protocol handler that was previously added to the kernel | |
431 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
432 | * from the kernel lists and can be freed or reused once this function | |
433 | * returns. | |
434 | * | |
435 | * This call sleeps to guarantee that no CPU is looking at the packet | |
436 | * type after return. | |
437 | */ | |
438 | void dev_remove_pack(struct packet_type *pt) | |
439 | { | |
440 | __dev_remove_pack(pt); | |
4ec93edb | 441 | |
1da177e4 LT |
442 | synchronize_net(); |
443 | } | |
d1b19dff | 444 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 445 | |
62532da9 VY |
446 | |
447 | /** | |
448 | * dev_add_offload - register offload handlers | |
449 | * @po: protocol offload declaration | |
450 | * | |
451 | * Add protocol offload handlers to the networking stack. The passed | |
452 | * &proto_offload is linked into kernel lists and may not be freed until | |
453 | * it has been removed from the kernel lists. | |
454 | * | |
455 | * This call does not sleep therefore it can not | |
456 | * guarantee all CPU's that are in middle of receiving packets | |
457 | * will see the new offload handlers (until the next received packet). | |
458 | */ | |
459 | void dev_add_offload(struct packet_offload *po) | |
460 | { | |
461 | struct list_head *head = &offload_base; | |
462 | ||
463 | spin_lock(&offload_lock); | |
464 | list_add_rcu(&po->list, head); | |
465 | spin_unlock(&offload_lock); | |
466 | } | |
467 | EXPORT_SYMBOL(dev_add_offload); | |
468 | ||
469 | /** | |
470 | * __dev_remove_offload - remove offload handler | |
471 | * @po: packet offload declaration | |
472 | * | |
473 | * Remove a protocol offload handler that was previously added to the | |
474 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
475 | * is removed from the kernel lists and can be freed or reused once this | |
476 | * function returns. | |
477 | * | |
478 | * The packet type might still be in use by receivers | |
479 | * and must not be freed until after all the CPU's have gone | |
480 | * through a quiescent state. | |
481 | */ | |
482 | void __dev_remove_offload(struct packet_offload *po) | |
483 | { | |
484 | struct list_head *head = &offload_base; | |
485 | struct packet_offload *po1; | |
486 | ||
c53aa505 | 487 | spin_lock(&offload_lock); |
62532da9 VY |
488 | |
489 | list_for_each_entry(po1, head, list) { | |
490 | if (po == po1) { | |
491 | list_del_rcu(&po->list); | |
492 | goto out; | |
493 | } | |
494 | } | |
495 | ||
496 | pr_warn("dev_remove_offload: %p not found\n", po); | |
497 | out: | |
c53aa505 | 498 | spin_unlock(&offload_lock); |
62532da9 VY |
499 | } |
500 | EXPORT_SYMBOL(__dev_remove_offload); | |
501 | ||
502 | /** | |
503 | * dev_remove_offload - remove packet offload handler | |
504 | * @po: packet offload declaration | |
505 | * | |
506 | * Remove a packet offload handler that was previously added to the kernel | |
507 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
508 | * removed from the kernel lists and can be freed or reused once this | |
509 | * function returns. | |
510 | * | |
511 | * This call sleeps to guarantee that no CPU is looking at the packet | |
512 | * type after return. | |
513 | */ | |
514 | void dev_remove_offload(struct packet_offload *po) | |
515 | { | |
516 | __dev_remove_offload(po); | |
517 | ||
518 | synchronize_net(); | |
519 | } | |
520 | EXPORT_SYMBOL(dev_remove_offload); | |
521 | ||
1da177e4 LT |
522 | /****************************************************************************** |
523 | ||
524 | Device Boot-time Settings Routines | |
525 | ||
526 | *******************************************************************************/ | |
527 | ||
528 | /* Boot time configuration table */ | |
529 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
530 | ||
531 | /** | |
532 | * netdev_boot_setup_add - add new setup entry | |
533 | * @name: name of the device | |
534 | * @map: configured settings for the device | |
535 | * | |
536 | * Adds new setup entry to the dev_boot_setup list. The function | |
537 | * returns 0 on error and 1 on success. This is a generic routine to | |
538 | * all netdevices. | |
539 | */ | |
540 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
541 | { | |
542 | struct netdev_boot_setup *s; | |
543 | int i; | |
544 | ||
545 | s = dev_boot_setup; | |
546 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
547 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
548 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 549 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
550 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
551 | break; | |
552 | } | |
553 | } | |
554 | ||
555 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
556 | } | |
557 | ||
558 | /** | |
559 | * netdev_boot_setup_check - check boot time settings | |
560 | * @dev: the netdevice | |
561 | * | |
562 | * Check boot time settings for the device. | |
563 | * The found settings are set for the device to be used | |
564 | * later in the device probing. | |
565 | * Returns 0 if no settings found, 1 if they are. | |
566 | */ | |
567 | int netdev_boot_setup_check(struct net_device *dev) | |
568 | { | |
569 | struct netdev_boot_setup *s = dev_boot_setup; | |
570 | int i; | |
571 | ||
572 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
573 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 574 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
575 | dev->irq = s[i].map.irq; |
576 | dev->base_addr = s[i].map.base_addr; | |
577 | dev->mem_start = s[i].map.mem_start; | |
578 | dev->mem_end = s[i].map.mem_end; | |
579 | return 1; | |
580 | } | |
581 | } | |
582 | return 0; | |
583 | } | |
d1b19dff | 584 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
585 | |
586 | ||
587 | /** | |
588 | * netdev_boot_base - get address from boot time settings | |
589 | * @prefix: prefix for network device | |
590 | * @unit: id for network device | |
591 | * | |
592 | * Check boot time settings for the base address of device. | |
593 | * The found settings are set for the device to be used | |
594 | * later in the device probing. | |
595 | * Returns 0 if no settings found. | |
596 | */ | |
597 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
598 | { | |
599 | const struct netdev_boot_setup *s = dev_boot_setup; | |
600 | char name[IFNAMSIZ]; | |
601 | int i; | |
602 | ||
603 | sprintf(name, "%s%d", prefix, unit); | |
604 | ||
605 | /* | |
606 | * If device already registered then return base of 1 | |
607 | * to indicate not to probe for this interface | |
608 | */ | |
881d966b | 609 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
610 | return 1; |
611 | ||
612 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
613 | if (!strcmp(name, s[i].name)) | |
614 | return s[i].map.base_addr; | |
615 | return 0; | |
616 | } | |
617 | ||
618 | /* | |
619 | * Saves at boot time configured settings for any netdevice. | |
620 | */ | |
621 | int __init netdev_boot_setup(char *str) | |
622 | { | |
623 | int ints[5]; | |
624 | struct ifmap map; | |
625 | ||
626 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
627 | if (!str || !*str) | |
628 | return 0; | |
629 | ||
630 | /* Save settings */ | |
631 | memset(&map, 0, sizeof(map)); | |
632 | if (ints[0] > 0) | |
633 | map.irq = ints[1]; | |
634 | if (ints[0] > 1) | |
635 | map.base_addr = ints[2]; | |
636 | if (ints[0] > 2) | |
637 | map.mem_start = ints[3]; | |
638 | if (ints[0] > 3) | |
639 | map.mem_end = ints[4]; | |
640 | ||
641 | /* Add new entry to the list */ | |
642 | return netdev_boot_setup_add(str, &map); | |
643 | } | |
644 | ||
645 | __setup("netdev=", netdev_boot_setup); | |
646 | ||
647 | /******************************************************************************* | |
648 | ||
649 | Device Interface Subroutines | |
650 | ||
651 | *******************************************************************************/ | |
652 | ||
653 | /** | |
654 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 655 | * @net: the applicable net namespace |
1da177e4 LT |
656 | * @name: name to find |
657 | * | |
658 | * Find an interface by name. Must be called under RTNL semaphore | |
659 | * or @dev_base_lock. If the name is found a pointer to the device | |
660 | * is returned. If the name is not found then %NULL is returned. The | |
661 | * reference counters are not incremented so the caller must be | |
662 | * careful with locks. | |
663 | */ | |
664 | ||
881d966b | 665 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 666 | { |
0bd8d536 ED |
667 | struct net_device *dev; |
668 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 669 | |
b67bfe0d | 670 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
671 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
672 | return dev; | |
0bd8d536 | 673 | |
1da177e4 LT |
674 | return NULL; |
675 | } | |
d1b19dff | 676 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 677 | |
72c9528b ED |
678 | /** |
679 | * dev_get_by_name_rcu - find a device by its name | |
680 | * @net: the applicable net namespace | |
681 | * @name: name to find | |
682 | * | |
683 | * Find an interface by name. | |
684 | * If the name is found a pointer to the device is returned. | |
685 | * If the name is not found then %NULL is returned. | |
686 | * The reference counters are not incremented so the caller must be | |
687 | * careful with locks. The caller must hold RCU lock. | |
688 | */ | |
689 | ||
690 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
691 | { | |
72c9528b ED |
692 | struct net_device *dev; |
693 | struct hlist_head *head = dev_name_hash(net, name); | |
694 | ||
b67bfe0d | 695 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
696 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
697 | return dev; | |
698 | ||
699 | return NULL; | |
700 | } | |
701 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
702 | ||
1da177e4 LT |
703 | /** |
704 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 705 | * @net: the applicable net namespace |
1da177e4 LT |
706 | * @name: name to find |
707 | * | |
708 | * Find an interface by name. This can be called from any | |
709 | * context and does its own locking. The returned handle has | |
710 | * the usage count incremented and the caller must use dev_put() to | |
711 | * release it when it is no longer needed. %NULL is returned if no | |
712 | * matching device is found. | |
713 | */ | |
714 | ||
881d966b | 715 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
716 | { |
717 | struct net_device *dev; | |
718 | ||
72c9528b ED |
719 | rcu_read_lock(); |
720 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
721 | if (dev) |
722 | dev_hold(dev); | |
72c9528b | 723 | rcu_read_unlock(); |
1da177e4 LT |
724 | return dev; |
725 | } | |
d1b19dff | 726 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
727 | |
728 | /** | |
729 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 730 | * @net: the applicable net namespace |
1da177e4 LT |
731 | * @ifindex: index of device |
732 | * | |
733 | * Search for an interface by index. Returns %NULL if the device | |
734 | * is not found or a pointer to the device. The device has not | |
735 | * had its reference counter increased so the caller must be careful | |
736 | * about locking. The caller must hold either the RTNL semaphore | |
737 | * or @dev_base_lock. | |
738 | */ | |
739 | ||
881d966b | 740 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 741 | { |
0bd8d536 ED |
742 | struct net_device *dev; |
743 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 744 | |
b67bfe0d | 745 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
746 | if (dev->ifindex == ifindex) |
747 | return dev; | |
0bd8d536 | 748 | |
1da177e4 LT |
749 | return NULL; |
750 | } | |
d1b19dff | 751 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 752 | |
fb699dfd ED |
753 | /** |
754 | * dev_get_by_index_rcu - find a device by its ifindex | |
755 | * @net: the applicable net namespace | |
756 | * @ifindex: index of device | |
757 | * | |
758 | * Search for an interface by index. Returns %NULL if the device | |
759 | * is not found or a pointer to the device. The device has not | |
760 | * had its reference counter increased so the caller must be careful | |
761 | * about locking. The caller must hold RCU lock. | |
762 | */ | |
763 | ||
764 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
765 | { | |
fb699dfd ED |
766 | struct net_device *dev; |
767 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
768 | ||
b67bfe0d | 769 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
770 | if (dev->ifindex == ifindex) |
771 | return dev; | |
772 | ||
773 | return NULL; | |
774 | } | |
775 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
776 | ||
1da177e4 LT |
777 | |
778 | /** | |
779 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 780 | * @net: the applicable net namespace |
1da177e4 LT |
781 | * @ifindex: index of device |
782 | * | |
783 | * Search for an interface by index. Returns NULL if the device | |
784 | * is not found or a pointer to the device. The device returned has | |
785 | * had a reference added and the pointer is safe until the user calls | |
786 | * dev_put to indicate they have finished with it. | |
787 | */ | |
788 | ||
881d966b | 789 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
790 | { |
791 | struct net_device *dev; | |
792 | ||
fb699dfd ED |
793 | rcu_read_lock(); |
794 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
795 | if (dev) |
796 | dev_hold(dev); | |
fb699dfd | 797 | rcu_read_unlock(); |
1da177e4 LT |
798 | return dev; |
799 | } | |
d1b19dff | 800 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 801 | |
5dbe7c17 NS |
802 | /** |
803 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
804 | * @net: network namespace | |
805 | * @name: a pointer to the buffer where the name will be stored. | |
806 | * @ifindex: the ifindex of the interface to get the name from. | |
807 | * | |
808 | * The use of raw_seqcount_begin() and cond_resched() before | |
809 | * retrying is required as we want to give the writers a chance | |
810 | * to complete when CONFIG_PREEMPT is not set. | |
811 | */ | |
812 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
813 | { | |
814 | struct net_device *dev; | |
815 | unsigned int seq; | |
816 | ||
817 | retry: | |
818 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
819 | rcu_read_lock(); | |
820 | dev = dev_get_by_index_rcu(net, ifindex); | |
821 | if (!dev) { | |
822 | rcu_read_unlock(); | |
823 | return -ENODEV; | |
824 | } | |
825 | ||
826 | strcpy(name, dev->name); | |
827 | rcu_read_unlock(); | |
828 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
829 | cond_resched(); | |
830 | goto retry; | |
831 | } | |
832 | ||
833 | return 0; | |
834 | } | |
835 | ||
1da177e4 | 836 | /** |
941666c2 | 837 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 838 | * @net: the applicable net namespace |
1da177e4 LT |
839 | * @type: media type of device |
840 | * @ha: hardware address | |
841 | * | |
842 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
843 | * is not found or a pointer to the device. |
844 | * The caller must hold RCU or RTNL. | |
941666c2 | 845 | * The returned device has not had its ref count increased |
1da177e4 LT |
846 | * and the caller must therefore be careful about locking |
847 | * | |
1da177e4 LT |
848 | */ |
849 | ||
941666c2 ED |
850 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
851 | const char *ha) | |
1da177e4 LT |
852 | { |
853 | struct net_device *dev; | |
854 | ||
941666c2 | 855 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
856 | if (dev->type == type && |
857 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
858 | return dev; |
859 | ||
860 | return NULL; | |
1da177e4 | 861 | } |
941666c2 | 862 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 863 | |
881d966b | 864 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
865 | { |
866 | struct net_device *dev; | |
867 | ||
4e9cac2b | 868 | ASSERT_RTNL(); |
881d966b | 869 | for_each_netdev(net, dev) |
4e9cac2b | 870 | if (dev->type == type) |
7562f876 PE |
871 | return dev; |
872 | ||
873 | return NULL; | |
4e9cac2b | 874 | } |
4e9cac2b PM |
875 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
876 | ||
881d966b | 877 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 878 | { |
99fe3c39 | 879 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 880 | |
99fe3c39 ED |
881 | rcu_read_lock(); |
882 | for_each_netdev_rcu(net, dev) | |
883 | if (dev->type == type) { | |
884 | dev_hold(dev); | |
885 | ret = dev; | |
886 | break; | |
887 | } | |
888 | rcu_read_unlock(); | |
889 | return ret; | |
1da177e4 | 890 | } |
1da177e4 LT |
891 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
892 | ||
893 | /** | |
bb69ae04 | 894 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 895 | * @net: the applicable net namespace |
1da177e4 LT |
896 | * @if_flags: IFF_* values |
897 | * @mask: bitmask of bits in if_flags to check | |
898 | * | |
899 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
900 | * is not found or a pointer to the device. Must be called inside |
901 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
902 | */ |
903 | ||
bb69ae04 | 904 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 905 | unsigned short mask) |
1da177e4 | 906 | { |
7562f876 | 907 | struct net_device *dev, *ret; |
1da177e4 | 908 | |
7562f876 | 909 | ret = NULL; |
c6d14c84 | 910 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 911 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 912 | ret = dev; |
1da177e4 LT |
913 | break; |
914 | } | |
915 | } | |
7562f876 | 916 | return ret; |
1da177e4 | 917 | } |
bb69ae04 | 918 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
919 | |
920 | /** | |
921 | * dev_valid_name - check if name is okay for network device | |
922 | * @name: name string | |
923 | * | |
924 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
925 | * to allow sysfs to work. We also disallow any kind of |
926 | * whitespace. | |
1da177e4 | 927 | */ |
95f050bf | 928 | bool dev_valid_name(const char *name) |
1da177e4 | 929 | { |
c7fa9d18 | 930 | if (*name == '\0') |
95f050bf | 931 | return false; |
b6fe17d6 | 932 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 933 | return false; |
c7fa9d18 | 934 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 935 | return false; |
c7fa9d18 DM |
936 | |
937 | while (*name) { | |
938 | if (*name == '/' || isspace(*name)) | |
95f050bf | 939 | return false; |
c7fa9d18 DM |
940 | name++; |
941 | } | |
95f050bf | 942 | return true; |
1da177e4 | 943 | } |
d1b19dff | 944 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
945 | |
946 | /** | |
b267b179 EB |
947 | * __dev_alloc_name - allocate a name for a device |
948 | * @net: network namespace to allocate the device name in | |
1da177e4 | 949 | * @name: name format string |
b267b179 | 950 | * @buf: scratch buffer and result name string |
1da177e4 LT |
951 | * |
952 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
953 | * id. It scans list of devices to build up a free map, then chooses |
954 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
955 | * while allocating the name and adding the device in order to avoid | |
956 | * duplicates. | |
957 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
958 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
959 | */ |
960 | ||
b267b179 | 961 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
962 | { |
963 | int i = 0; | |
1da177e4 LT |
964 | const char *p; |
965 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 966 | unsigned long *inuse; |
1da177e4 LT |
967 | struct net_device *d; |
968 | ||
969 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
970 | if (p) { | |
971 | /* | |
972 | * Verify the string as this thing may have come from | |
973 | * the user. There must be either one "%d" and no other "%" | |
974 | * characters. | |
975 | */ | |
976 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
977 | return -EINVAL; | |
978 | ||
979 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 980 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
981 | if (!inuse) |
982 | return -ENOMEM; | |
983 | ||
881d966b | 984 | for_each_netdev(net, d) { |
1da177e4 LT |
985 | if (!sscanf(d->name, name, &i)) |
986 | continue; | |
987 | if (i < 0 || i >= max_netdevices) | |
988 | continue; | |
989 | ||
990 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 991 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
992 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
993 | set_bit(i, inuse); | |
994 | } | |
995 | ||
996 | i = find_first_zero_bit(inuse, max_netdevices); | |
997 | free_page((unsigned long) inuse); | |
998 | } | |
999 | ||
d9031024 OP |
1000 | if (buf != name) |
1001 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1002 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1003 | return i; |
1da177e4 LT |
1004 | |
1005 | /* It is possible to run out of possible slots | |
1006 | * when the name is long and there isn't enough space left | |
1007 | * for the digits, or if all bits are used. | |
1008 | */ | |
1009 | return -ENFILE; | |
1010 | } | |
1011 | ||
b267b179 EB |
1012 | /** |
1013 | * dev_alloc_name - allocate a name for a device | |
1014 | * @dev: device | |
1015 | * @name: name format string | |
1016 | * | |
1017 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1018 | * id. It scans list of devices to build up a free map, then chooses | |
1019 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1020 | * while allocating the name and adding the device in order to avoid | |
1021 | * duplicates. | |
1022 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1023 | * Returns the number of the unit assigned or a negative errno code. | |
1024 | */ | |
1025 | ||
1026 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1027 | { | |
1028 | char buf[IFNAMSIZ]; | |
1029 | struct net *net; | |
1030 | int ret; | |
1031 | ||
c346dca1 YH |
1032 | BUG_ON(!dev_net(dev)); |
1033 | net = dev_net(dev); | |
b267b179 EB |
1034 | ret = __dev_alloc_name(net, name, buf); |
1035 | if (ret >= 0) | |
1036 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1037 | return ret; | |
1038 | } | |
d1b19dff | 1039 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1040 | |
828de4f6 G |
1041 | static int dev_alloc_name_ns(struct net *net, |
1042 | struct net_device *dev, | |
1043 | const char *name) | |
d9031024 | 1044 | { |
828de4f6 G |
1045 | char buf[IFNAMSIZ]; |
1046 | int ret; | |
8ce6cebc | 1047 | |
828de4f6 G |
1048 | ret = __dev_alloc_name(net, name, buf); |
1049 | if (ret >= 0) | |
1050 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1051 | return ret; | |
1052 | } | |
1053 | ||
1054 | static int dev_get_valid_name(struct net *net, | |
1055 | struct net_device *dev, | |
1056 | const char *name) | |
1057 | { | |
1058 | BUG_ON(!net); | |
8ce6cebc | 1059 | |
d9031024 OP |
1060 | if (!dev_valid_name(name)) |
1061 | return -EINVAL; | |
1062 | ||
1c5cae81 | 1063 | if (strchr(name, '%')) |
828de4f6 | 1064 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1065 | else if (__dev_get_by_name(net, name)) |
1066 | return -EEXIST; | |
8ce6cebc DL |
1067 | else if (dev->name != name) |
1068 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1069 | |
1070 | return 0; | |
1071 | } | |
1da177e4 LT |
1072 | |
1073 | /** | |
1074 | * dev_change_name - change name of a device | |
1075 | * @dev: device | |
1076 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1077 | * | |
1078 | * Change name of a device, can pass format strings "eth%d". | |
1079 | * for wildcarding. | |
1080 | */ | |
cf04a4c7 | 1081 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1082 | { |
fcc5a03a | 1083 | char oldname[IFNAMSIZ]; |
1da177e4 | 1084 | int err = 0; |
fcc5a03a | 1085 | int ret; |
881d966b | 1086 | struct net *net; |
1da177e4 LT |
1087 | |
1088 | ASSERT_RTNL(); | |
c346dca1 | 1089 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1090 | |
c346dca1 | 1091 | net = dev_net(dev); |
1da177e4 LT |
1092 | if (dev->flags & IFF_UP) |
1093 | return -EBUSY; | |
1094 | ||
30e6c9fa | 1095 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1096 | |
1097 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1098 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1099 | return 0; |
c91f6df2 | 1100 | } |
c8d90dca | 1101 | |
fcc5a03a HX |
1102 | memcpy(oldname, dev->name, IFNAMSIZ); |
1103 | ||
828de4f6 | 1104 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1105 | if (err < 0) { |
30e6c9fa | 1106 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1107 | return err; |
c91f6df2 | 1108 | } |
1da177e4 | 1109 | |
fcc5a03a | 1110 | rollback: |
a1b3f594 EB |
1111 | ret = device_rename(&dev->dev, dev->name); |
1112 | if (ret) { | |
1113 | memcpy(dev->name, oldname, IFNAMSIZ); | |
30e6c9fa | 1114 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1115 | return ret; |
dcc99773 | 1116 | } |
7f988eab | 1117 | |
30e6c9fa | 1118 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1119 | |
7f988eab | 1120 | write_lock_bh(&dev_base_lock); |
372b2312 | 1121 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1122 | write_unlock_bh(&dev_base_lock); |
1123 | ||
1124 | synchronize_rcu(); | |
1125 | ||
1126 | write_lock_bh(&dev_base_lock); | |
1127 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1128 | write_unlock_bh(&dev_base_lock); |
1129 | ||
056925ab | 1130 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1131 | ret = notifier_to_errno(ret); |
1132 | ||
1133 | if (ret) { | |
91e9c07b ED |
1134 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1135 | if (err >= 0) { | |
fcc5a03a | 1136 | err = ret; |
30e6c9fa | 1137 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a HX |
1138 | memcpy(dev->name, oldname, IFNAMSIZ); |
1139 | goto rollback; | |
91e9c07b | 1140 | } else { |
7b6cd1ce | 1141 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1142 | dev->name, ret); |
fcc5a03a HX |
1143 | } |
1144 | } | |
1da177e4 LT |
1145 | |
1146 | return err; | |
1147 | } | |
1148 | ||
0b815a1a SH |
1149 | /** |
1150 | * dev_set_alias - change ifalias of a device | |
1151 | * @dev: device | |
1152 | * @alias: name up to IFALIASZ | |
f0db275a | 1153 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1154 | * |
1155 | * Set ifalias for a device, | |
1156 | */ | |
1157 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1158 | { | |
7364e445 AK |
1159 | char *new_ifalias; |
1160 | ||
0b815a1a SH |
1161 | ASSERT_RTNL(); |
1162 | ||
1163 | if (len >= IFALIASZ) | |
1164 | return -EINVAL; | |
1165 | ||
96ca4a2c | 1166 | if (!len) { |
388dfc2d SK |
1167 | kfree(dev->ifalias); |
1168 | dev->ifalias = NULL; | |
96ca4a2c OH |
1169 | return 0; |
1170 | } | |
1171 | ||
7364e445 AK |
1172 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1173 | if (!new_ifalias) | |
0b815a1a | 1174 | return -ENOMEM; |
7364e445 | 1175 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1176 | |
1177 | strlcpy(dev->ifalias, alias, len+1); | |
1178 | return len; | |
1179 | } | |
1180 | ||
1181 | ||
d8a33ac4 | 1182 | /** |
3041a069 | 1183 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1184 | * @dev: device to cause notification |
1185 | * | |
1186 | * Called to indicate a device has changed features. | |
1187 | */ | |
1188 | void netdev_features_change(struct net_device *dev) | |
1189 | { | |
056925ab | 1190 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1191 | } |
1192 | EXPORT_SYMBOL(netdev_features_change); | |
1193 | ||
1da177e4 LT |
1194 | /** |
1195 | * netdev_state_change - device changes state | |
1196 | * @dev: device to cause notification | |
1197 | * | |
1198 | * Called to indicate a device has changed state. This function calls | |
1199 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1200 | * to the routing socket. | |
1201 | */ | |
1202 | void netdev_state_change(struct net_device *dev) | |
1203 | { | |
1204 | if (dev->flags & IFF_UP) { | |
056925ab | 1205 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1206 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1207 | } | |
1208 | } | |
d1b19dff | 1209 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1210 | |
ee89bab1 AW |
1211 | /** |
1212 | * netdev_notify_peers - notify network peers about existence of @dev | |
1213 | * @dev: network device | |
1214 | * | |
1215 | * Generate traffic such that interested network peers are aware of | |
1216 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1217 | * a device wants to inform the rest of the network about some sort of | |
1218 | * reconfiguration such as a failover event or virtual machine | |
1219 | * migration. | |
1220 | */ | |
1221 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1222 | { |
ee89bab1 AW |
1223 | rtnl_lock(); |
1224 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1225 | rtnl_unlock(); | |
c1da4ac7 | 1226 | } |
ee89bab1 | 1227 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1228 | |
bd380811 | 1229 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1230 | { |
d314774c | 1231 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1232 | int ret; |
1da177e4 | 1233 | |
e46b66bc BH |
1234 | ASSERT_RTNL(); |
1235 | ||
1da177e4 LT |
1236 | if (!netif_device_present(dev)) |
1237 | return -ENODEV; | |
1238 | ||
ca99ca14 NH |
1239 | /* Block netpoll from trying to do any rx path servicing. |
1240 | * If we don't do this there is a chance ndo_poll_controller | |
1241 | * or ndo_poll may be running while we open the device | |
1242 | */ | |
da6e378b | 1243 | netpoll_rx_disable(dev); |
ca99ca14 | 1244 | |
3b8bcfd5 JB |
1245 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1246 | ret = notifier_to_errno(ret); | |
1247 | if (ret) | |
1248 | return ret; | |
1249 | ||
1da177e4 | 1250 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1251 | |
d314774c SH |
1252 | if (ops->ndo_validate_addr) |
1253 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1254 | |
d314774c SH |
1255 | if (!ret && ops->ndo_open) |
1256 | ret = ops->ndo_open(dev); | |
1da177e4 | 1257 | |
ca99ca14 NH |
1258 | netpoll_rx_enable(dev); |
1259 | ||
bada339b JG |
1260 | if (ret) |
1261 | clear_bit(__LINK_STATE_START, &dev->state); | |
1262 | else { | |
1da177e4 | 1263 | dev->flags |= IFF_UP; |
b4bd07c2 | 1264 | net_dmaengine_get(); |
4417da66 | 1265 | dev_set_rx_mode(dev); |
1da177e4 | 1266 | dev_activate(dev); |
7bf23575 | 1267 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1268 | } |
bada339b | 1269 | |
1da177e4 LT |
1270 | return ret; |
1271 | } | |
1272 | ||
1273 | /** | |
bd380811 PM |
1274 | * dev_open - prepare an interface for use. |
1275 | * @dev: device to open | |
1da177e4 | 1276 | * |
bd380811 PM |
1277 | * Takes a device from down to up state. The device's private open |
1278 | * function is invoked and then the multicast lists are loaded. Finally | |
1279 | * the device is moved into the up state and a %NETDEV_UP message is | |
1280 | * sent to the netdev notifier chain. | |
1281 | * | |
1282 | * Calling this function on an active interface is a nop. On a failure | |
1283 | * a negative errno code is returned. | |
1da177e4 | 1284 | */ |
bd380811 PM |
1285 | int dev_open(struct net_device *dev) |
1286 | { | |
1287 | int ret; | |
1288 | ||
bd380811 PM |
1289 | if (dev->flags & IFF_UP) |
1290 | return 0; | |
1291 | ||
bd380811 PM |
1292 | ret = __dev_open(dev); |
1293 | if (ret < 0) | |
1294 | return ret; | |
1295 | ||
bd380811 PM |
1296 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); |
1297 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1298 | ||
1299 | return ret; | |
1300 | } | |
1301 | EXPORT_SYMBOL(dev_open); | |
1302 | ||
44345724 | 1303 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1304 | { |
44345724 | 1305 | struct net_device *dev; |
e46b66bc | 1306 | |
bd380811 | 1307 | ASSERT_RTNL(); |
9d5010db DM |
1308 | might_sleep(); |
1309 | ||
44345724 | 1310 | list_for_each_entry(dev, head, unreg_list) { |
44345724 | 1311 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1312 | |
44345724 | 1313 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1314 | |
44345724 OP |
1315 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1316 | * can be even on different cpu. So just clear netif_running(). | |
1317 | * | |
1318 | * dev->stop() will invoke napi_disable() on all of it's | |
1319 | * napi_struct instances on this device. | |
1320 | */ | |
1321 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1322 | } | |
1da177e4 | 1323 | |
44345724 | 1324 | dev_deactivate_many(head); |
d8b2a4d2 | 1325 | |
44345724 OP |
1326 | list_for_each_entry(dev, head, unreg_list) { |
1327 | const struct net_device_ops *ops = dev->netdev_ops; | |
1da177e4 | 1328 | |
44345724 OP |
1329 | /* |
1330 | * Call the device specific close. This cannot fail. | |
1331 | * Only if device is UP | |
1332 | * | |
1333 | * We allow it to be called even after a DETACH hot-plug | |
1334 | * event. | |
1335 | */ | |
1336 | if (ops->ndo_stop) | |
1337 | ops->ndo_stop(dev); | |
1338 | ||
44345724 | 1339 | dev->flags &= ~IFF_UP; |
44345724 OP |
1340 | net_dmaengine_put(); |
1341 | } | |
1342 | ||
1343 | return 0; | |
1344 | } | |
1345 | ||
1346 | static int __dev_close(struct net_device *dev) | |
1347 | { | |
f87e6f47 | 1348 | int retval; |
44345724 OP |
1349 | LIST_HEAD(single); |
1350 | ||
ca99ca14 | 1351 | /* Temporarily disable netpoll until the interface is down */ |
da6e378b | 1352 | netpoll_rx_disable(dev); |
ca99ca14 | 1353 | |
44345724 | 1354 | list_add(&dev->unreg_list, &single); |
f87e6f47 LT |
1355 | retval = __dev_close_many(&single); |
1356 | list_del(&single); | |
ca99ca14 NH |
1357 | |
1358 | netpoll_rx_enable(dev); | |
f87e6f47 | 1359 | return retval; |
44345724 OP |
1360 | } |
1361 | ||
3fbd8758 | 1362 | static int dev_close_many(struct list_head *head) |
44345724 OP |
1363 | { |
1364 | struct net_device *dev, *tmp; | |
1365 | LIST_HEAD(tmp_list); | |
1da177e4 | 1366 | |
44345724 OP |
1367 | list_for_each_entry_safe(dev, tmp, head, unreg_list) |
1368 | if (!(dev->flags & IFF_UP)) | |
1369 | list_move(&dev->unreg_list, &tmp_list); | |
1370 | ||
1371 | __dev_close_many(head); | |
1da177e4 | 1372 | |
44345724 OP |
1373 | list_for_each_entry(dev, head, unreg_list) { |
1374 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1375 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1376 | } | |
bd380811 | 1377 | |
44345724 OP |
1378 | /* rollback_registered_many needs the complete original list */ |
1379 | list_splice(&tmp_list, head); | |
bd380811 PM |
1380 | return 0; |
1381 | } | |
1382 | ||
1383 | /** | |
1384 | * dev_close - shutdown an interface. | |
1385 | * @dev: device to shutdown | |
1386 | * | |
1387 | * This function moves an active device into down state. A | |
1388 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1389 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1390 | * chain. | |
1391 | */ | |
1392 | int dev_close(struct net_device *dev) | |
1393 | { | |
e14a5993 ED |
1394 | if (dev->flags & IFF_UP) { |
1395 | LIST_HEAD(single); | |
1da177e4 | 1396 | |
ca99ca14 | 1397 | /* Block netpoll rx while the interface is going down */ |
da6e378b | 1398 | netpoll_rx_disable(dev); |
ca99ca14 | 1399 | |
e14a5993 ED |
1400 | list_add(&dev->unreg_list, &single); |
1401 | dev_close_many(&single); | |
1402 | list_del(&single); | |
ca99ca14 NH |
1403 | |
1404 | netpoll_rx_enable(dev); | |
e14a5993 | 1405 | } |
da6e378b | 1406 | return 0; |
1da177e4 | 1407 | } |
d1b19dff | 1408 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1409 | |
1410 | ||
0187bdfb BH |
1411 | /** |
1412 | * dev_disable_lro - disable Large Receive Offload on a device | |
1413 | * @dev: device | |
1414 | * | |
1415 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1416 | * called under RTNL. This is needed if received packets may be | |
1417 | * forwarded to another interface. | |
1418 | */ | |
1419 | void dev_disable_lro(struct net_device *dev) | |
1420 | { | |
f11970e3 NH |
1421 | /* |
1422 | * If we're trying to disable lro on a vlan device | |
1423 | * use the underlying physical device instead | |
1424 | */ | |
1425 | if (is_vlan_dev(dev)) | |
1426 | dev = vlan_dev_real_dev(dev); | |
1427 | ||
bc5787c6 MM |
1428 | dev->wanted_features &= ~NETIF_F_LRO; |
1429 | netdev_update_features(dev); | |
27660515 | 1430 | |
22d5969f MM |
1431 | if (unlikely(dev->features & NETIF_F_LRO)) |
1432 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
0187bdfb BH |
1433 | } |
1434 | EXPORT_SYMBOL(dev_disable_lro); | |
1435 | ||
351638e7 JP |
1436 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1437 | struct net_device *dev) | |
1438 | { | |
1439 | struct netdev_notifier_info info; | |
1440 | ||
1441 | netdev_notifier_info_init(&info, dev); | |
1442 | return nb->notifier_call(nb, val, &info); | |
1443 | } | |
0187bdfb | 1444 | |
881d966b EB |
1445 | static int dev_boot_phase = 1; |
1446 | ||
1da177e4 LT |
1447 | /** |
1448 | * register_netdevice_notifier - register a network notifier block | |
1449 | * @nb: notifier | |
1450 | * | |
1451 | * Register a notifier to be called when network device events occur. | |
1452 | * The notifier passed is linked into the kernel structures and must | |
1453 | * not be reused until it has been unregistered. A negative errno code | |
1454 | * is returned on a failure. | |
1455 | * | |
1456 | * When registered all registration and up events are replayed | |
4ec93edb | 1457 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1458 | * view of the network device list. |
1459 | */ | |
1460 | ||
1461 | int register_netdevice_notifier(struct notifier_block *nb) | |
1462 | { | |
1463 | struct net_device *dev; | |
fcc5a03a | 1464 | struct net_device *last; |
881d966b | 1465 | struct net *net; |
1da177e4 LT |
1466 | int err; |
1467 | ||
1468 | rtnl_lock(); | |
f07d5b94 | 1469 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1470 | if (err) |
1471 | goto unlock; | |
881d966b EB |
1472 | if (dev_boot_phase) |
1473 | goto unlock; | |
1474 | for_each_net(net) { | |
1475 | for_each_netdev(net, dev) { | |
351638e7 | 1476 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1477 | err = notifier_to_errno(err); |
1478 | if (err) | |
1479 | goto rollback; | |
1480 | ||
1481 | if (!(dev->flags & IFF_UP)) | |
1482 | continue; | |
1da177e4 | 1483 | |
351638e7 | 1484 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1485 | } |
1da177e4 | 1486 | } |
fcc5a03a HX |
1487 | |
1488 | unlock: | |
1da177e4 LT |
1489 | rtnl_unlock(); |
1490 | return err; | |
fcc5a03a HX |
1491 | |
1492 | rollback: | |
1493 | last = dev; | |
881d966b EB |
1494 | for_each_net(net) { |
1495 | for_each_netdev(net, dev) { | |
1496 | if (dev == last) | |
8f891489 | 1497 | goto outroll; |
fcc5a03a | 1498 | |
881d966b | 1499 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1500 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1501 | dev); | |
1502 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1503 | } |
351638e7 | 1504 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1505 | } |
fcc5a03a | 1506 | } |
c67625a1 | 1507 | |
8f891489 | 1508 | outroll: |
c67625a1 | 1509 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1510 | goto unlock; |
1da177e4 | 1511 | } |
d1b19dff | 1512 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1513 | |
1514 | /** | |
1515 | * unregister_netdevice_notifier - unregister a network notifier block | |
1516 | * @nb: notifier | |
1517 | * | |
1518 | * Unregister a notifier previously registered by | |
1519 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1520 | * kernel structures and may then be reused. A negative errno code | |
1521 | * is returned on a failure. | |
7d3d43da EB |
1522 | * |
1523 | * After unregistering unregister and down device events are synthesized | |
1524 | * for all devices on the device list to the removed notifier to remove | |
1525 | * the need for special case cleanup code. | |
1da177e4 LT |
1526 | */ |
1527 | ||
1528 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1529 | { | |
7d3d43da EB |
1530 | struct net_device *dev; |
1531 | struct net *net; | |
9f514950 HX |
1532 | int err; |
1533 | ||
1534 | rtnl_lock(); | |
f07d5b94 | 1535 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1536 | if (err) |
1537 | goto unlock; | |
1538 | ||
1539 | for_each_net(net) { | |
1540 | for_each_netdev(net, dev) { | |
1541 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1542 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1543 | dev); | |
1544 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1545 | } |
351638e7 | 1546 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1547 | } |
1548 | } | |
1549 | unlock: | |
9f514950 HX |
1550 | rtnl_unlock(); |
1551 | return err; | |
1da177e4 | 1552 | } |
d1b19dff | 1553 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1554 | |
351638e7 JP |
1555 | /** |
1556 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1557 | * @val: value passed unmodified to notifier function | |
1558 | * @dev: net_device pointer passed unmodified to notifier function | |
1559 | * @info: notifier information data | |
1560 | * | |
1561 | * Call all network notifier blocks. Parameters and return value | |
1562 | * are as for raw_notifier_call_chain(). | |
1563 | */ | |
1564 | ||
1565 | int call_netdevice_notifiers_info(unsigned long val, struct net_device *dev, | |
1566 | struct netdev_notifier_info *info) | |
1567 | { | |
1568 | ASSERT_RTNL(); | |
1569 | netdev_notifier_info_init(info, dev); | |
1570 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1571 | } | |
1572 | EXPORT_SYMBOL(call_netdevice_notifiers_info); | |
1573 | ||
1da177e4 LT |
1574 | /** |
1575 | * call_netdevice_notifiers - call all network notifier blocks | |
1576 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1577 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1578 | * |
1579 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1580 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1581 | */ |
1582 | ||
ad7379d4 | 1583 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1584 | { |
351638e7 JP |
1585 | struct netdev_notifier_info info; |
1586 | ||
1587 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1588 | } |
edf947f1 | 1589 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1590 | |
c5905afb | 1591 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1592 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1593 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1594 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1595 | * static_key_slow_dec() calls. |
b90e5794 ED |
1596 | */ |
1597 | static atomic_t netstamp_needed_deferred; | |
1598 | #endif | |
1da177e4 LT |
1599 | |
1600 | void net_enable_timestamp(void) | |
1601 | { | |
b90e5794 ED |
1602 | #ifdef HAVE_JUMP_LABEL |
1603 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1604 | ||
1605 | if (deferred) { | |
1606 | while (--deferred) | |
c5905afb | 1607 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1608 | return; |
1609 | } | |
1610 | #endif | |
c5905afb | 1611 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1612 | } |
d1b19dff | 1613 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1614 | |
1615 | void net_disable_timestamp(void) | |
1616 | { | |
b90e5794 ED |
1617 | #ifdef HAVE_JUMP_LABEL |
1618 | if (in_interrupt()) { | |
1619 | atomic_inc(&netstamp_needed_deferred); | |
1620 | return; | |
1621 | } | |
1622 | #endif | |
c5905afb | 1623 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1624 | } |
d1b19dff | 1625 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1626 | |
3b098e2d | 1627 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1628 | { |
588f0330 | 1629 | skb->tstamp.tv64 = 0; |
c5905afb | 1630 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1631 | __net_timestamp(skb); |
1da177e4 LT |
1632 | } |
1633 | ||
588f0330 | 1634 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1635 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1636 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1637 | __net_timestamp(SKB); \ | |
1638 | } \ | |
3b098e2d | 1639 | |
79b569f0 DL |
1640 | static inline bool is_skb_forwardable(struct net_device *dev, |
1641 | struct sk_buff *skb) | |
1642 | { | |
1643 | unsigned int len; | |
1644 | ||
1645 | if (!(dev->flags & IFF_UP)) | |
1646 | return false; | |
1647 | ||
1648 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1649 | if (skb->len <= len) | |
1650 | return true; | |
1651 | ||
1652 | /* if TSO is enabled, we don't care about the length as the packet | |
1653 | * could be forwarded without being segmented before | |
1654 | */ | |
1655 | if (skb_is_gso(skb)) | |
1656 | return true; | |
1657 | ||
1658 | return false; | |
1659 | } | |
1660 | ||
44540960 AB |
1661 | /** |
1662 | * dev_forward_skb - loopback an skb to another netif | |
1663 | * | |
1664 | * @dev: destination network device | |
1665 | * @skb: buffer to forward | |
1666 | * | |
1667 | * return values: | |
1668 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1669 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1670 | * |
1671 | * dev_forward_skb can be used for injecting an skb from the | |
1672 | * start_xmit function of one device into the receive queue | |
1673 | * of another device. | |
1674 | * | |
1675 | * The receiving device may be in another namespace, so | |
1676 | * we have to clear all information in the skb that could | |
1677 | * impact namespace isolation. | |
1678 | */ | |
1679 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1680 | { | |
48c83012 MT |
1681 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { |
1682 | if (skb_copy_ubufs(skb, GFP_ATOMIC)) { | |
1683 | atomic_long_inc(&dev->rx_dropped); | |
1684 | kfree_skb(skb); | |
1685 | return NET_RX_DROP; | |
1686 | } | |
1687 | } | |
1688 | ||
79b569f0 | 1689 | if (unlikely(!is_skb_forwardable(dev, skb))) { |
caf586e5 | 1690 | atomic_long_inc(&dev->rx_dropped); |
6ec82562 | 1691 | kfree_skb(skb); |
44540960 | 1692 | return NET_RX_DROP; |
6ec82562 | 1693 | } |
44540960 | 1694 | skb->protocol = eth_type_trans(skb, dev); |
06a23fe3 IY |
1695 | |
1696 | /* eth_type_trans() can set pkt_type. | |
64261f23 ND |
1697 | * call skb_scrub_packet() after it to clear pkt_type _after_ calling |
1698 | * eth_type_trans(). | |
06a23fe3 | 1699 | */ |
8b27f277 | 1700 | skb_scrub_packet(skb, true); |
06a23fe3 | 1701 | |
44540960 AB |
1702 | return netif_rx(skb); |
1703 | } | |
1704 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1705 | ||
71d9dec2 CG |
1706 | static inline int deliver_skb(struct sk_buff *skb, |
1707 | struct packet_type *pt_prev, | |
1708 | struct net_device *orig_dev) | |
1709 | { | |
1080e512 MT |
1710 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1711 | return -ENOMEM; | |
71d9dec2 CG |
1712 | atomic_inc(&skb->users); |
1713 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1714 | } | |
1715 | ||
c0de08d0 EL |
1716 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1717 | { | |
a3d744e9 | 1718 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1719 | return false; |
1720 | ||
1721 | if (ptype->id_match) | |
1722 | return ptype->id_match(ptype, skb->sk); | |
1723 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1724 | return true; | |
1725 | ||
1726 | return false; | |
1727 | } | |
1728 | ||
1da177e4 LT |
1729 | /* |
1730 | * Support routine. Sends outgoing frames to any network | |
1731 | * taps currently in use. | |
1732 | */ | |
1733 | ||
f6a78bfc | 1734 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1735 | { |
1736 | struct packet_type *ptype; | |
71d9dec2 CG |
1737 | struct sk_buff *skb2 = NULL; |
1738 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1739 | |
1da177e4 LT |
1740 | rcu_read_lock(); |
1741 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1742 | /* Never send packets back to the socket | |
1743 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1744 | */ | |
1745 | if ((ptype->dev == dev || !ptype->dev) && | |
c0de08d0 | 1746 | (!skb_loop_sk(ptype, skb))) { |
71d9dec2 CG |
1747 | if (pt_prev) { |
1748 | deliver_skb(skb2, pt_prev, skb->dev); | |
1749 | pt_prev = ptype; | |
1750 | continue; | |
1751 | } | |
1752 | ||
1753 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1754 | if (!skb2) |
1755 | break; | |
1756 | ||
70978182 ED |
1757 | net_timestamp_set(skb2); |
1758 | ||
1da177e4 LT |
1759 | /* skb->nh should be correctly |
1760 | set by sender, so that the second statement is | |
1761 | just protection against buggy protocols. | |
1762 | */ | |
459a98ed | 1763 | skb_reset_mac_header(skb2); |
1da177e4 | 1764 | |
d56f90a7 | 1765 | if (skb_network_header(skb2) < skb2->data || |
ced14f68 | 1766 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { |
e87cc472 JP |
1767 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", |
1768 | ntohs(skb2->protocol), | |
1769 | dev->name); | |
c1d2bbe1 | 1770 | skb_reset_network_header(skb2); |
1da177e4 LT |
1771 | } |
1772 | ||
b0e380b1 | 1773 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1774 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1775 | pt_prev = ptype; |
1da177e4 LT |
1776 | } |
1777 | } | |
71d9dec2 CG |
1778 | if (pt_prev) |
1779 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1780 | rcu_read_unlock(); |
1781 | } | |
1782 | ||
2c53040f BH |
1783 | /** |
1784 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1785 | * @dev: Network device |
1786 | * @txq: number of queues available | |
1787 | * | |
1788 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1789 | * valid. To resolve this verify the tc mapping remains valid and if | |
1790 | * not NULL the mapping. With no priorities mapping to this | |
1791 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1792 | * is invalid nothing can be done so disable priority mappings. If is | |
1793 | * expected that drivers will fix this mapping if they can before | |
1794 | * calling netif_set_real_num_tx_queues. | |
1795 | */ | |
bb134d22 | 1796 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1797 | { |
1798 | int i; | |
1799 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1800 | ||
1801 | /* If TC0 is invalidated disable TC mapping */ | |
1802 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1803 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1804 | dev->num_tc = 0; |
1805 | return; | |
1806 | } | |
1807 | ||
1808 | /* Invalidated prio to tc mappings set to TC0 */ | |
1809 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1810 | int q = netdev_get_prio_tc_map(dev, i); | |
1811 | ||
1812 | tc = &dev->tc_to_txq[q]; | |
1813 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1814 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1815 | i, q); | |
4f57c087 JF |
1816 | netdev_set_prio_tc_map(dev, i, 0); |
1817 | } | |
1818 | } | |
1819 | } | |
1820 | ||
537c00de AD |
1821 | #ifdef CONFIG_XPS |
1822 | static DEFINE_MUTEX(xps_map_mutex); | |
1823 | #define xmap_dereference(P) \ | |
1824 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1825 | ||
10cdc3f3 AD |
1826 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1827 | int cpu, u16 index) | |
537c00de | 1828 | { |
10cdc3f3 AD |
1829 | struct xps_map *map = NULL; |
1830 | int pos; | |
537c00de | 1831 | |
10cdc3f3 AD |
1832 | if (dev_maps) |
1833 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1834 | |
10cdc3f3 AD |
1835 | for (pos = 0; map && pos < map->len; pos++) { |
1836 | if (map->queues[pos] == index) { | |
537c00de AD |
1837 | if (map->len > 1) { |
1838 | map->queues[pos] = map->queues[--map->len]; | |
1839 | } else { | |
10cdc3f3 | 1840 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1841 | kfree_rcu(map, rcu); |
1842 | map = NULL; | |
1843 | } | |
10cdc3f3 | 1844 | break; |
537c00de | 1845 | } |
537c00de AD |
1846 | } |
1847 | ||
10cdc3f3 AD |
1848 | return map; |
1849 | } | |
1850 | ||
024e9679 | 1851 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
10cdc3f3 AD |
1852 | { |
1853 | struct xps_dev_maps *dev_maps; | |
024e9679 | 1854 | int cpu, i; |
10cdc3f3 AD |
1855 | bool active = false; |
1856 | ||
1857 | mutex_lock(&xps_map_mutex); | |
1858 | dev_maps = xmap_dereference(dev->xps_maps); | |
1859 | ||
1860 | if (!dev_maps) | |
1861 | goto out_no_maps; | |
1862 | ||
1863 | for_each_possible_cpu(cpu) { | |
024e9679 AD |
1864 | for (i = index; i < dev->num_tx_queues; i++) { |
1865 | if (!remove_xps_queue(dev_maps, cpu, i)) | |
1866 | break; | |
1867 | } | |
1868 | if (i == dev->num_tx_queues) | |
10cdc3f3 AD |
1869 | active = true; |
1870 | } | |
1871 | ||
1872 | if (!active) { | |
537c00de AD |
1873 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
1874 | kfree_rcu(dev_maps, rcu); | |
1875 | } | |
1876 | ||
024e9679 AD |
1877 | for (i = index; i < dev->num_tx_queues; i++) |
1878 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), | |
1879 | NUMA_NO_NODE); | |
1880 | ||
537c00de AD |
1881 | out_no_maps: |
1882 | mutex_unlock(&xps_map_mutex); | |
1883 | } | |
1884 | ||
01c5f864 AD |
1885 | static struct xps_map *expand_xps_map(struct xps_map *map, |
1886 | int cpu, u16 index) | |
1887 | { | |
1888 | struct xps_map *new_map; | |
1889 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
1890 | int i, pos; | |
1891 | ||
1892 | for (pos = 0; map && pos < map->len; pos++) { | |
1893 | if (map->queues[pos] != index) | |
1894 | continue; | |
1895 | return map; | |
1896 | } | |
1897 | ||
1898 | /* Need to add queue to this CPU's existing map */ | |
1899 | if (map) { | |
1900 | if (pos < map->alloc_len) | |
1901 | return map; | |
1902 | ||
1903 | alloc_len = map->alloc_len * 2; | |
1904 | } | |
1905 | ||
1906 | /* Need to allocate new map to store queue on this CPU's map */ | |
1907 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
1908 | cpu_to_node(cpu)); | |
1909 | if (!new_map) | |
1910 | return NULL; | |
1911 | ||
1912 | for (i = 0; i < pos; i++) | |
1913 | new_map->queues[i] = map->queues[i]; | |
1914 | new_map->alloc_len = alloc_len; | |
1915 | new_map->len = pos; | |
1916 | ||
1917 | return new_map; | |
1918 | } | |
1919 | ||
537c00de AD |
1920 | int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index) |
1921 | { | |
01c5f864 | 1922 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 1923 | struct xps_map *map, *new_map; |
537c00de | 1924 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
1925 | int cpu, numa_node_id = -2; |
1926 | bool active = false; | |
537c00de AD |
1927 | |
1928 | mutex_lock(&xps_map_mutex); | |
1929 | ||
1930 | dev_maps = xmap_dereference(dev->xps_maps); | |
1931 | ||
01c5f864 AD |
1932 | /* allocate memory for queue storage */ |
1933 | for_each_online_cpu(cpu) { | |
1934 | if (!cpumask_test_cpu(cpu, mask)) | |
1935 | continue; | |
1936 | ||
1937 | if (!new_dev_maps) | |
1938 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
1939 | if (!new_dev_maps) { |
1940 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 1941 | return -ENOMEM; |
2bb60cb9 | 1942 | } |
01c5f864 AD |
1943 | |
1944 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
1945 | NULL; | |
1946 | ||
1947 | map = expand_xps_map(map, cpu, index); | |
1948 | if (!map) | |
1949 | goto error; | |
1950 | ||
1951 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
1952 | } | |
1953 | ||
1954 | if (!new_dev_maps) | |
1955 | goto out_no_new_maps; | |
1956 | ||
537c00de | 1957 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
1958 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
1959 | /* add queue to CPU maps */ | |
1960 | int pos = 0; | |
1961 | ||
1962 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1963 | while ((pos < map->len) && (map->queues[pos] != index)) | |
1964 | pos++; | |
1965 | ||
1966 | if (pos == map->len) | |
1967 | map->queues[map->len++] = index; | |
537c00de | 1968 | #ifdef CONFIG_NUMA |
537c00de AD |
1969 | if (numa_node_id == -2) |
1970 | numa_node_id = cpu_to_node(cpu); | |
1971 | else if (numa_node_id != cpu_to_node(cpu)) | |
1972 | numa_node_id = -1; | |
537c00de | 1973 | #endif |
01c5f864 AD |
1974 | } else if (dev_maps) { |
1975 | /* fill in the new device map from the old device map */ | |
1976 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
1977 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 1978 | } |
01c5f864 | 1979 | |
537c00de AD |
1980 | } |
1981 | ||
01c5f864 AD |
1982 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
1983 | ||
537c00de | 1984 | /* Cleanup old maps */ |
01c5f864 AD |
1985 | if (dev_maps) { |
1986 | for_each_possible_cpu(cpu) { | |
1987 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1988 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
1989 | if (map && map != new_map) | |
1990 | kfree_rcu(map, rcu); | |
1991 | } | |
537c00de | 1992 | |
01c5f864 | 1993 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
1994 | } |
1995 | ||
01c5f864 AD |
1996 | dev_maps = new_dev_maps; |
1997 | active = true; | |
537c00de | 1998 | |
01c5f864 AD |
1999 | out_no_new_maps: |
2000 | /* update Tx queue numa node */ | |
537c00de AD |
2001 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2002 | (numa_node_id >= 0) ? numa_node_id : | |
2003 | NUMA_NO_NODE); | |
2004 | ||
01c5f864 AD |
2005 | if (!dev_maps) |
2006 | goto out_no_maps; | |
2007 | ||
2008 | /* removes queue from unused CPUs */ | |
2009 | for_each_possible_cpu(cpu) { | |
2010 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
2011 | continue; | |
2012 | ||
2013 | if (remove_xps_queue(dev_maps, cpu, index)) | |
2014 | active = true; | |
2015 | } | |
2016 | ||
2017 | /* free map if not active */ | |
2018 | if (!active) { | |
2019 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2020 | kfree_rcu(dev_maps, rcu); | |
2021 | } | |
2022 | ||
2023 | out_no_maps: | |
537c00de AD |
2024 | mutex_unlock(&xps_map_mutex); |
2025 | ||
2026 | return 0; | |
2027 | error: | |
01c5f864 AD |
2028 | /* remove any maps that we added */ |
2029 | for_each_possible_cpu(cpu) { | |
2030 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2031 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2032 | NULL; | |
2033 | if (new_map && new_map != map) | |
2034 | kfree(new_map); | |
2035 | } | |
2036 | ||
537c00de AD |
2037 | mutex_unlock(&xps_map_mutex); |
2038 | ||
537c00de AD |
2039 | kfree(new_dev_maps); |
2040 | return -ENOMEM; | |
2041 | } | |
2042 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2043 | ||
2044 | #endif | |
f0796d5c JF |
2045 | /* |
2046 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2047 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2048 | */ | |
e6484930 | 2049 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2050 | { |
1d24eb48 TH |
2051 | int rc; |
2052 | ||
e6484930 TH |
2053 | if (txq < 1 || txq > dev->num_tx_queues) |
2054 | return -EINVAL; | |
f0796d5c | 2055 | |
5c56580b BH |
2056 | if (dev->reg_state == NETREG_REGISTERED || |
2057 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2058 | ASSERT_RTNL(); |
2059 | ||
1d24eb48 TH |
2060 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2061 | txq); | |
bf264145 TH |
2062 | if (rc) |
2063 | return rc; | |
2064 | ||
4f57c087 JF |
2065 | if (dev->num_tc) |
2066 | netif_setup_tc(dev, txq); | |
2067 | ||
024e9679 | 2068 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2069 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2070 | #ifdef CONFIG_XPS |
2071 | netif_reset_xps_queues_gt(dev, txq); | |
2072 | #endif | |
2073 | } | |
f0796d5c | 2074 | } |
e6484930 TH |
2075 | |
2076 | dev->real_num_tx_queues = txq; | |
2077 | return 0; | |
f0796d5c JF |
2078 | } |
2079 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2080 | |
62fe0b40 BH |
2081 | #ifdef CONFIG_RPS |
2082 | /** | |
2083 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2084 | * @dev: Network device | |
2085 | * @rxq: Actual number of RX queues | |
2086 | * | |
2087 | * This must be called either with the rtnl_lock held or before | |
2088 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2089 | * negative error code. If called before registration, it always |
2090 | * succeeds. | |
62fe0b40 BH |
2091 | */ |
2092 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2093 | { | |
2094 | int rc; | |
2095 | ||
bd25fa7b TH |
2096 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2097 | return -EINVAL; | |
2098 | ||
62fe0b40 BH |
2099 | if (dev->reg_state == NETREG_REGISTERED) { |
2100 | ASSERT_RTNL(); | |
2101 | ||
62fe0b40 BH |
2102 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2103 | rxq); | |
2104 | if (rc) | |
2105 | return rc; | |
62fe0b40 BH |
2106 | } |
2107 | ||
2108 | dev->real_num_rx_queues = rxq; | |
2109 | return 0; | |
2110 | } | |
2111 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2112 | #endif | |
2113 | ||
2c53040f BH |
2114 | /** |
2115 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2116 | * |
2117 | * This routine should set an upper limit on the number of RSS queues | |
2118 | * used by default by multiqueue devices. | |
2119 | */ | |
a55b138b | 2120 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
2121 | { |
2122 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
2123 | } | |
2124 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2125 | ||
def82a1d | 2126 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 2127 | { |
def82a1d JP |
2128 | struct softnet_data *sd; |
2129 | unsigned long flags; | |
56079431 | 2130 | |
def82a1d JP |
2131 | local_irq_save(flags); |
2132 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
2133 | q->next_sched = NULL; |
2134 | *sd->output_queue_tailp = q; | |
2135 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2136 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2137 | local_irq_restore(flags); | |
2138 | } | |
2139 | ||
2140 | void __netif_schedule(struct Qdisc *q) | |
2141 | { | |
2142 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2143 | __netif_reschedule(q); | |
56079431 DV |
2144 | } |
2145 | EXPORT_SYMBOL(__netif_schedule); | |
2146 | ||
bea3348e | 2147 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 2148 | { |
3578b0c8 | 2149 | if (atomic_dec_and_test(&skb->users)) { |
bea3348e SH |
2150 | struct softnet_data *sd; |
2151 | unsigned long flags; | |
56079431 | 2152 | |
bea3348e SH |
2153 | local_irq_save(flags); |
2154 | sd = &__get_cpu_var(softnet_data); | |
2155 | skb->next = sd->completion_queue; | |
2156 | sd->completion_queue = skb; | |
2157 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2158 | local_irq_restore(flags); | |
2159 | } | |
56079431 | 2160 | } |
bea3348e | 2161 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
2162 | |
2163 | void dev_kfree_skb_any(struct sk_buff *skb) | |
2164 | { | |
2165 | if (in_irq() || irqs_disabled()) | |
2166 | dev_kfree_skb_irq(skb); | |
2167 | else | |
2168 | dev_kfree_skb(skb); | |
2169 | } | |
2170 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
2171 | ||
2172 | ||
bea3348e SH |
2173 | /** |
2174 | * netif_device_detach - mark device as removed | |
2175 | * @dev: network device | |
2176 | * | |
2177 | * Mark device as removed from system and therefore no longer available. | |
2178 | */ | |
56079431 DV |
2179 | void netif_device_detach(struct net_device *dev) |
2180 | { | |
2181 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2182 | netif_running(dev)) { | |
d543103a | 2183 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2184 | } |
2185 | } | |
2186 | EXPORT_SYMBOL(netif_device_detach); | |
2187 | ||
bea3348e SH |
2188 | /** |
2189 | * netif_device_attach - mark device as attached | |
2190 | * @dev: network device | |
2191 | * | |
2192 | * Mark device as attached from system and restart if needed. | |
2193 | */ | |
56079431 DV |
2194 | void netif_device_attach(struct net_device *dev) |
2195 | { | |
2196 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2197 | netif_running(dev)) { | |
d543103a | 2198 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2199 | __netdev_watchdog_up(dev); |
56079431 DV |
2200 | } |
2201 | } | |
2202 | EXPORT_SYMBOL(netif_device_attach); | |
2203 | ||
36c92474 BH |
2204 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2205 | { | |
65e9d2fa | 2206 | static const netdev_features_t null_features = 0; |
36c92474 BH |
2207 | struct net_device *dev = skb->dev; |
2208 | const char *driver = ""; | |
2209 | ||
c846ad9b BG |
2210 | if (!net_ratelimit()) |
2211 | return; | |
2212 | ||
36c92474 BH |
2213 | if (dev && dev->dev.parent) |
2214 | driver = dev_driver_string(dev->dev.parent); | |
2215 | ||
2216 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " | |
2217 | "gso_type=%d ip_summed=%d\n", | |
65e9d2fa MM |
2218 | driver, dev ? &dev->features : &null_features, |
2219 | skb->sk ? &skb->sk->sk_route_caps : &null_features, | |
36c92474 BH |
2220 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2221 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2222 | } | |
2223 | ||
1da177e4 LT |
2224 | /* |
2225 | * Invalidate hardware checksum when packet is to be mangled, and | |
2226 | * complete checksum manually on outgoing path. | |
2227 | */ | |
84fa7933 | 2228 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2229 | { |
d3bc23e7 | 2230 | __wsum csum; |
663ead3b | 2231 | int ret = 0, offset; |
1da177e4 | 2232 | |
84fa7933 | 2233 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2234 | goto out_set_summed; |
2235 | ||
2236 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2237 | skb_warn_bad_offload(skb); |
2238 | return -EINVAL; | |
1da177e4 LT |
2239 | } |
2240 | ||
cef401de ED |
2241 | /* Before computing a checksum, we should make sure no frag could |
2242 | * be modified by an external entity : checksum could be wrong. | |
2243 | */ | |
2244 | if (skb_has_shared_frag(skb)) { | |
2245 | ret = __skb_linearize(skb); | |
2246 | if (ret) | |
2247 | goto out; | |
2248 | } | |
2249 | ||
55508d60 | 2250 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2251 | BUG_ON(offset >= skb_headlen(skb)); |
2252 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2253 | ||
2254 | offset += skb->csum_offset; | |
2255 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2256 | ||
2257 | if (skb_cloned(skb) && | |
2258 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2259 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2260 | if (ret) | |
2261 | goto out; | |
2262 | } | |
2263 | ||
a030847e | 2264 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2265 | out_set_summed: |
1da177e4 | 2266 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2267 | out: |
1da177e4 LT |
2268 | return ret; |
2269 | } | |
d1b19dff | 2270 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2271 | |
ec5f0615 | 2272 | __be16 skb_network_protocol(struct sk_buff *skb) |
f6a78bfc | 2273 | { |
252e3346 | 2274 | __be16 type = skb->protocol; |
c80a8512 | 2275 | int vlan_depth = ETH_HLEN; |
f6a78bfc | 2276 | |
19acc327 PS |
2277 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2278 | if (type == htons(ETH_P_TEB)) { | |
2279 | struct ethhdr *eth; | |
2280 | ||
2281 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2282 | return 0; | |
2283 | ||
2284 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2285 | type = eth->h_proto; | |
2286 | } | |
2287 | ||
8ad227ff | 2288 | while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) { |
c8d5bcd1 | 2289 | struct vlan_hdr *vh; |
7b9c6090 | 2290 | |
c8d5bcd1 | 2291 | if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN))) |
ec5f0615 | 2292 | return 0; |
7b9c6090 | 2293 | |
c8d5bcd1 JG |
2294 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); |
2295 | type = vh->h_vlan_encapsulated_proto; | |
2296 | vlan_depth += VLAN_HLEN; | |
7b9c6090 JG |
2297 | } |
2298 | ||
ec5f0615 PS |
2299 | return type; |
2300 | } | |
2301 | ||
2302 | /** | |
2303 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2304 | * @skb: buffer to segment | |
2305 | * @features: features for the output path (see dev->features) | |
2306 | */ | |
2307 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2308 | netdev_features_t features) | |
2309 | { | |
2310 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2311 | struct packet_offload *ptype; | |
2312 | __be16 type = skb_network_protocol(skb); | |
2313 | ||
2314 | if (unlikely(!type)) | |
2315 | return ERR_PTR(-EINVAL); | |
2316 | ||
f6a78bfc HX |
2317 | __skb_pull(skb, skb->mac_len); |
2318 | ||
2319 | rcu_read_lock(); | |
22061d80 | 2320 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2321 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
84fa7933 | 2322 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
05e8ef4a PS |
2323 | int err; |
2324 | ||
f191a1d1 | 2325 | err = ptype->callbacks.gso_send_check(skb); |
a430a43d HX |
2326 | segs = ERR_PTR(err); |
2327 | if (err || skb_gso_ok(skb, features)) | |
2328 | break; | |
d56f90a7 ACM |
2329 | __skb_push(skb, (skb->data - |
2330 | skb_network_header(skb))); | |
a430a43d | 2331 | } |
f191a1d1 | 2332 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2333 | break; |
2334 | } | |
2335 | } | |
2336 | rcu_read_unlock(); | |
2337 | ||
98e399f8 | 2338 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2339 | |
f6a78bfc HX |
2340 | return segs; |
2341 | } | |
05e8ef4a PS |
2342 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2343 | ||
2344 | ||
2345 | /* openvswitch calls this on rx path, so we need a different check. | |
2346 | */ | |
2347 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2348 | { | |
2349 | if (tx_path) | |
2350 | return skb->ip_summed != CHECKSUM_PARTIAL; | |
2351 | else | |
2352 | return skb->ip_summed == CHECKSUM_NONE; | |
2353 | } | |
2354 | ||
2355 | /** | |
2356 | * __skb_gso_segment - Perform segmentation on skb. | |
2357 | * @skb: buffer to segment | |
2358 | * @features: features for the output path (see dev->features) | |
2359 | * @tx_path: whether it is called in TX path | |
2360 | * | |
2361 | * This function segments the given skb and returns a list of segments. | |
2362 | * | |
2363 | * It may return NULL if the skb requires no segmentation. This is | |
2364 | * only possible when GSO is used for verifying header integrity. | |
2365 | */ | |
2366 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2367 | netdev_features_t features, bool tx_path) | |
2368 | { | |
2369 | if (unlikely(skb_needs_check(skb, tx_path))) { | |
2370 | int err; | |
2371 | ||
2372 | skb_warn_bad_offload(skb); | |
2373 | ||
2374 | if (skb_header_cloned(skb) && | |
2375 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
2376 | return ERR_PTR(err); | |
2377 | } | |
2378 | ||
68c33163 | 2379 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
05e8ef4a PS |
2380 | skb_reset_mac_header(skb); |
2381 | skb_reset_mac_len(skb); | |
2382 | ||
2383 | return skb_mac_gso_segment(skb, features); | |
2384 | } | |
12b0004d | 2385 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2386 | |
fb286bb2 HX |
2387 | /* Take action when hardware reception checksum errors are detected. */ |
2388 | #ifdef CONFIG_BUG | |
2389 | void netdev_rx_csum_fault(struct net_device *dev) | |
2390 | { | |
2391 | if (net_ratelimit()) { | |
7b6cd1ce | 2392 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2393 | dump_stack(); |
2394 | } | |
2395 | } | |
2396 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2397 | #endif | |
2398 | ||
1da177e4 LT |
2399 | /* Actually, we should eliminate this check as soon as we know, that: |
2400 | * 1. IOMMU is present and allows to map all the memory. | |
2401 | * 2. No high memory really exists on this machine. | |
2402 | */ | |
2403 | ||
9092c658 | 2404 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2405 | { |
3d3a8533 | 2406 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2407 | int i; |
5acbbd42 | 2408 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2409 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2410 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2411 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2412 | return 1; |
ea2ab693 | 2413 | } |
5acbbd42 | 2414 | } |
1da177e4 | 2415 | |
5acbbd42 FT |
2416 | if (PCI_DMA_BUS_IS_PHYS) { |
2417 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2418 | |
9092c658 ED |
2419 | if (!pdev) |
2420 | return 0; | |
5acbbd42 | 2421 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2422 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2423 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2424 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2425 | return 1; | |
2426 | } | |
2427 | } | |
3d3a8533 | 2428 | #endif |
1da177e4 LT |
2429 | return 0; |
2430 | } | |
1da177e4 | 2431 | |
f6a78bfc HX |
2432 | struct dev_gso_cb { |
2433 | void (*destructor)(struct sk_buff *skb); | |
2434 | }; | |
2435 | ||
2436 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
2437 | ||
2438 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
2439 | { | |
2440 | struct dev_gso_cb *cb; | |
2441 | ||
2442 | do { | |
2443 | struct sk_buff *nskb = skb->next; | |
2444 | ||
2445 | skb->next = nskb->next; | |
2446 | nskb->next = NULL; | |
2447 | kfree_skb(nskb); | |
2448 | } while (skb->next); | |
2449 | ||
2450 | cb = DEV_GSO_CB(skb); | |
2451 | if (cb->destructor) | |
2452 | cb->destructor(skb); | |
2453 | } | |
2454 | ||
2455 | /** | |
2456 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
2457 | * @skb: buffer to segment | |
91ecb63c | 2458 | * @features: device features as applicable to this skb |
f6a78bfc HX |
2459 | * |
2460 | * This function segments the given skb and stores the list of segments | |
2461 | * in skb->next. | |
2462 | */ | |
c8f44aff | 2463 | static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features) |
f6a78bfc | 2464 | { |
f6a78bfc | 2465 | struct sk_buff *segs; |
576a30eb HX |
2466 | |
2467 | segs = skb_gso_segment(skb, features); | |
2468 | ||
2469 | /* Verifying header integrity only. */ | |
2470 | if (!segs) | |
2471 | return 0; | |
f6a78bfc | 2472 | |
801678c5 | 2473 | if (IS_ERR(segs)) |
f6a78bfc HX |
2474 | return PTR_ERR(segs); |
2475 | ||
2476 | skb->next = segs; | |
2477 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
2478 | skb->destructor = dev_gso_skb_destructor; | |
2479 | ||
2480 | return 0; | |
2481 | } | |
2482 | ||
c8f44aff | 2483 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
cdbaa0bb | 2484 | netdev_features_t features) |
f01a5236 | 2485 | { |
c0d680e5 | 2486 | if (skb->ip_summed != CHECKSUM_NONE && |
cdbaa0bb | 2487 | !can_checksum_protocol(features, skb_network_protocol(skb))) { |
f01a5236 | 2488 | features &= ~NETIF_F_ALL_CSUM; |
f01a5236 JG |
2489 | } else if (illegal_highdma(skb->dev, skb)) { |
2490 | features &= ~NETIF_F_SG; | |
2491 | } | |
2492 | ||
2493 | return features; | |
2494 | } | |
2495 | ||
c8f44aff | 2496 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 JG |
2497 | { |
2498 | __be16 protocol = skb->protocol; | |
c8f44aff | 2499 | netdev_features_t features = skb->dev->features; |
58e998c6 | 2500 | |
30b678d8 BH |
2501 | if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs) |
2502 | features &= ~NETIF_F_GSO_MASK; | |
2503 | ||
8ad227ff | 2504 | if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) { |
58e998c6 JG |
2505 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; |
2506 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 | 2507 | } else if (!vlan_tx_tag_present(skb)) { |
cdbaa0bb | 2508 | return harmonize_features(skb, features); |
f01a5236 | 2509 | } |
58e998c6 | 2510 | |
8ad227ff PM |
2511 | features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX | |
2512 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2513 | |
cdbaa0bb | 2514 | if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) |
f01a5236 | 2515 | features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | |
8ad227ff PM |
2516 | NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX | |
2517 | NETIF_F_HW_VLAN_STAG_TX; | |
cdbaa0bb AD |
2518 | |
2519 | return harmonize_features(skb, features); | |
58e998c6 | 2520 | } |
f01a5236 | 2521 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2522 | |
6afff0ca JF |
2523 | /* |
2524 | * Returns true if either: | |
2525 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
d1a53dfd | 2526 | * 2. skb is fragmented and the device does not support SG. |
6afff0ca JF |
2527 | */ |
2528 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
6708c9e5 | 2529 | netdev_features_t features) |
6afff0ca | 2530 | { |
02932ce9 JG |
2531 | return skb_is_nonlinear(skb) && |
2532 | ((skb_has_frag_list(skb) && | |
2533 | !(features & NETIF_F_FRAGLIST)) || | |
e1e78db6 | 2534 | (skb_shinfo(skb)->nr_frags && |
02932ce9 | 2535 | !(features & NETIF_F_SG))); |
6afff0ca JF |
2536 | } |
2537 | ||
fd2ea0a7 DM |
2538 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
2539 | struct netdev_queue *txq) | |
f6a78bfc | 2540 | { |
00829823 | 2541 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 2542 | int rc = NETDEV_TX_OK; |
ec764bf0 | 2543 | unsigned int skb_len; |
00829823 | 2544 | |
f6a78bfc | 2545 | if (likely(!skb->next)) { |
c8f44aff | 2546 | netdev_features_t features; |
fc741216 | 2547 | |
93f154b5 | 2548 | /* |
25985edc | 2549 | * If device doesn't need skb->dst, release it right now while |
93f154b5 ED |
2550 | * its hot in this cpu cache |
2551 | */ | |
adf30907 ED |
2552 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
2553 | skb_dst_drop(skb); | |
2554 | ||
fc741216 JG |
2555 | features = netif_skb_features(skb); |
2556 | ||
7b9c6090 | 2557 | if (vlan_tx_tag_present(skb) && |
86a9bad3 PM |
2558 | !vlan_hw_offload_capable(features, skb->vlan_proto)) { |
2559 | skb = __vlan_put_tag(skb, skb->vlan_proto, | |
2560 | vlan_tx_tag_get(skb)); | |
7b9c6090 JG |
2561 | if (unlikely(!skb)) |
2562 | goto out; | |
2563 | ||
2564 | skb->vlan_tci = 0; | |
2565 | } | |
2566 | ||
fc70fb64 AD |
2567 | /* If encapsulation offload request, verify we are testing |
2568 | * hardware encapsulation features instead of standard | |
2569 | * features for the netdev | |
2570 | */ | |
2571 | if (skb->encapsulation) | |
2572 | features &= dev->hw_enc_features; | |
2573 | ||
fc741216 | 2574 | if (netif_needs_gso(skb, features)) { |
91ecb63c | 2575 | if (unlikely(dev_gso_segment(skb, features))) |
9ccb8975 DM |
2576 | goto out_kfree_skb; |
2577 | if (skb->next) | |
2578 | goto gso; | |
6afff0ca | 2579 | } else { |
02932ce9 | 2580 | if (skb_needs_linearize(skb, features) && |
6afff0ca JF |
2581 | __skb_linearize(skb)) |
2582 | goto out_kfree_skb; | |
2583 | ||
2584 | /* If packet is not checksummed and device does not | |
2585 | * support checksumming for this protocol, complete | |
2586 | * checksumming here. | |
2587 | */ | |
2588 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
fc70fb64 AD |
2589 | if (skb->encapsulation) |
2590 | skb_set_inner_transport_header(skb, | |
2591 | skb_checksum_start_offset(skb)); | |
2592 | else | |
2593 | skb_set_transport_header(skb, | |
2594 | skb_checksum_start_offset(skb)); | |
03634668 | 2595 | if (!(features & NETIF_F_ALL_CSUM) && |
6afff0ca JF |
2596 | skb_checksum_help(skb)) |
2597 | goto out_kfree_skb; | |
2598 | } | |
9ccb8975 DM |
2599 | } |
2600 | ||
b40863c6 ED |
2601 | if (!list_empty(&ptype_all)) |
2602 | dev_queue_xmit_nit(skb, dev); | |
2603 | ||
ec764bf0 | 2604 | skb_len = skb->len; |
ac45f602 | 2605 | rc = ops->ndo_start_xmit(skb, dev); |
ec764bf0 | 2606 | trace_net_dev_xmit(skb, rc, dev, skb_len); |
ec634fe3 | 2607 | if (rc == NETDEV_TX_OK) |
08baf561 | 2608 | txq_trans_update(txq); |
ac45f602 | 2609 | return rc; |
f6a78bfc HX |
2610 | } |
2611 | ||
576a30eb | 2612 | gso: |
f6a78bfc HX |
2613 | do { |
2614 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
2615 | |
2616 | skb->next = nskb->next; | |
2617 | nskb->next = NULL; | |
068a2de5 | 2618 | |
b40863c6 ED |
2619 | if (!list_empty(&ptype_all)) |
2620 | dev_queue_xmit_nit(nskb, dev); | |
2621 | ||
ec764bf0 | 2622 | skb_len = nskb->len; |
00829823 | 2623 | rc = ops->ndo_start_xmit(nskb, dev); |
ec764bf0 | 2624 | trace_net_dev_xmit(nskb, rc, dev, skb_len); |
ec634fe3 | 2625 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
2626 | if (rc & ~NETDEV_TX_MASK) |
2627 | goto out_kfree_gso_skb; | |
f54d9e8d | 2628 | nskb->next = skb->next; |
f6a78bfc HX |
2629 | skb->next = nskb; |
2630 | return rc; | |
2631 | } | |
08baf561 | 2632 | txq_trans_update(txq); |
73466498 | 2633 | if (unlikely(netif_xmit_stopped(txq) && skb->next)) |
f54d9e8d | 2634 | return NETDEV_TX_BUSY; |
f6a78bfc | 2635 | } while (skb->next); |
4ec93edb | 2636 | |
572a9d7b | 2637 | out_kfree_gso_skb: |
0c772159 | 2638 | if (likely(skb->next == NULL)) { |
572a9d7b | 2639 | skb->destructor = DEV_GSO_CB(skb)->destructor; |
0c772159 SS |
2640 | consume_skb(skb); |
2641 | return rc; | |
2642 | } | |
f6a78bfc HX |
2643 | out_kfree_skb: |
2644 | kfree_skb(skb); | |
7b9c6090 | 2645 | out: |
572a9d7b | 2646 | return rc; |
f6a78bfc HX |
2647 | } |
2648 | ||
1def9238 ED |
2649 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
2650 | { | |
2651 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2652 | ||
2653 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2654 | ||
2655 | /* To get more precise estimation of bytes sent on wire, | |
2656 | * we add to pkt_len the headers size of all segments | |
2657 | */ | |
2658 | if (shinfo->gso_size) { | |
757b8b1d | 2659 | unsigned int hdr_len; |
15e5a030 | 2660 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 2661 | |
757b8b1d ED |
2662 | /* mac layer + network layer */ |
2663 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
2664 | ||
2665 | /* + transport layer */ | |
1def9238 ED |
2666 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
2667 | hdr_len += tcp_hdrlen(skb); | |
2668 | else | |
2669 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
2670 | |
2671 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
2672 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
2673 | shinfo->gso_size); | |
2674 | ||
2675 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
2676 | } |
2677 | } | |
2678 | ||
bbd8a0d3 KK |
2679 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2680 | struct net_device *dev, | |
2681 | struct netdev_queue *txq) | |
2682 | { | |
2683 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2684 | bool contended; |
bbd8a0d3 KK |
2685 | int rc; |
2686 | ||
1def9238 | 2687 | qdisc_pkt_len_init(skb); |
a2da570d | 2688 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
2689 | /* |
2690 | * Heuristic to force contended enqueues to serialize on a | |
2691 | * separate lock before trying to get qdisc main lock. | |
2692 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2693 | * and dequeue packets faster. | |
2694 | */ | |
a2da570d | 2695 | contended = qdisc_is_running(q); |
79640a4c ED |
2696 | if (unlikely(contended)) |
2697 | spin_lock(&q->busylock); | |
2698 | ||
bbd8a0d3 KK |
2699 | spin_lock(root_lock); |
2700 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2701 | kfree_skb(skb); | |
2702 | rc = NET_XMIT_DROP; | |
2703 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2704 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2705 | /* |
2706 | * This is a work-conserving queue; there are no old skbs | |
2707 | * waiting to be sent out; and the qdisc is not running - | |
2708 | * xmit the skb directly. | |
2709 | */ | |
7fee226a ED |
2710 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2711 | skb_dst_force(skb); | |
bfe0d029 | 2712 | |
bfe0d029 ED |
2713 | qdisc_bstats_update(q, skb); |
2714 | ||
79640a4c ED |
2715 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2716 | if (unlikely(contended)) { | |
2717 | spin_unlock(&q->busylock); | |
2718 | contended = false; | |
2719 | } | |
bbd8a0d3 | 2720 | __qdisc_run(q); |
79640a4c | 2721 | } else |
bc135b23 | 2722 | qdisc_run_end(q); |
bbd8a0d3 KK |
2723 | |
2724 | rc = NET_XMIT_SUCCESS; | |
2725 | } else { | |
7fee226a | 2726 | skb_dst_force(skb); |
a2da570d | 2727 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2728 | if (qdisc_run_begin(q)) { |
2729 | if (unlikely(contended)) { | |
2730 | spin_unlock(&q->busylock); | |
2731 | contended = false; | |
2732 | } | |
2733 | __qdisc_run(q); | |
2734 | } | |
bbd8a0d3 KK |
2735 | } |
2736 | spin_unlock(root_lock); | |
79640a4c ED |
2737 | if (unlikely(contended)) |
2738 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2739 | return rc; |
2740 | } | |
2741 | ||
5bc1421e NH |
2742 | #if IS_ENABLED(CONFIG_NETPRIO_CGROUP) |
2743 | static void skb_update_prio(struct sk_buff *skb) | |
2744 | { | |
6977a79d | 2745 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 2746 | |
91c68ce2 ED |
2747 | if (!skb->priority && skb->sk && map) { |
2748 | unsigned int prioidx = skb->sk->sk_cgrp_prioidx; | |
2749 | ||
2750 | if (prioidx < map->priomap_len) | |
2751 | skb->priority = map->priomap[prioidx]; | |
2752 | } | |
5bc1421e NH |
2753 | } |
2754 | #else | |
2755 | #define skb_update_prio(skb) | |
2756 | #endif | |
2757 | ||
745e20f1 | 2758 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2759 | #define RECURSION_LIMIT 10 |
745e20f1 | 2760 | |
95603e22 MM |
2761 | /** |
2762 | * dev_loopback_xmit - loop back @skb | |
2763 | * @skb: buffer to transmit | |
2764 | */ | |
2765 | int dev_loopback_xmit(struct sk_buff *skb) | |
2766 | { | |
2767 | skb_reset_mac_header(skb); | |
2768 | __skb_pull(skb, skb_network_offset(skb)); | |
2769 | skb->pkt_type = PACKET_LOOPBACK; | |
2770 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2771 | WARN_ON(!skb_dst(skb)); | |
2772 | skb_dst_force(skb); | |
2773 | netif_rx_ni(skb); | |
2774 | return 0; | |
2775 | } | |
2776 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2777 | ||
d29f749e DJ |
2778 | /** |
2779 | * dev_queue_xmit - transmit a buffer | |
2780 | * @skb: buffer to transmit | |
2781 | * | |
2782 | * Queue a buffer for transmission to a network device. The caller must | |
2783 | * have set the device and priority and built the buffer before calling | |
2784 | * this function. The function can be called from an interrupt. | |
2785 | * | |
2786 | * A negative errno code is returned on a failure. A success does not | |
2787 | * guarantee the frame will be transmitted as it may be dropped due | |
2788 | * to congestion or traffic shaping. | |
2789 | * | |
2790 | * ----------------------------------------------------------------------------------- | |
2791 | * I notice this method can also return errors from the queue disciplines, | |
2792 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2793 | * be positive. | |
2794 | * | |
2795 | * Regardless of the return value, the skb is consumed, so it is currently | |
2796 | * difficult to retry a send to this method. (You can bump the ref count | |
2797 | * before sending to hold a reference for retry if you are careful.) | |
2798 | * | |
2799 | * When calling this method, interrupts MUST be enabled. This is because | |
2800 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2801 | * --BLG | |
2802 | */ | |
1da177e4 LT |
2803 | int dev_queue_xmit(struct sk_buff *skb) |
2804 | { | |
2805 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2806 | struct netdev_queue *txq; |
1da177e4 LT |
2807 | struct Qdisc *q; |
2808 | int rc = -ENOMEM; | |
2809 | ||
6d1ccff6 ED |
2810 | skb_reset_mac_header(skb); |
2811 | ||
4ec93edb YH |
2812 | /* Disable soft irqs for various locks below. Also |
2813 | * stops preemption for RCU. | |
1da177e4 | 2814 | */ |
4ec93edb | 2815 | rcu_read_lock_bh(); |
1da177e4 | 2816 | |
5bc1421e NH |
2817 | skb_update_prio(skb); |
2818 | ||
8c4c49df | 2819 | txq = netdev_pick_tx(dev, skb); |
a898def2 | 2820 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2821 | |
1da177e4 | 2822 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2823 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2824 | #endif |
cf66ba58 | 2825 | trace_net_dev_queue(skb); |
1da177e4 | 2826 | if (q->enqueue) { |
bbd8a0d3 | 2827 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2828 | goto out; |
1da177e4 LT |
2829 | } |
2830 | ||
2831 | /* The device has no queue. Common case for software devices: | |
2832 | loopback, all the sorts of tunnels... | |
2833 | ||
932ff279 HX |
2834 | Really, it is unlikely that netif_tx_lock protection is necessary |
2835 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2836 | counters.) |
2837 | However, it is possible, that they rely on protection | |
2838 | made by us here. | |
2839 | ||
2840 | Check this and shot the lock. It is not prone from deadlocks. | |
2841 | Either shot noqueue qdisc, it is even simpler 8) | |
2842 | */ | |
2843 | if (dev->flags & IFF_UP) { | |
2844 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2845 | ||
c773e847 | 2846 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2847 | |
745e20f1 ED |
2848 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2849 | goto recursion_alert; | |
2850 | ||
c773e847 | 2851 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2852 | |
73466498 | 2853 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 2854 | __this_cpu_inc(xmit_recursion); |
572a9d7b | 2855 | rc = dev_hard_start_xmit(skb, dev, txq); |
745e20f1 | 2856 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2857 | if (dev_xmit_complete(rc)) { |
c773e847 | 2858 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2859 | goto out; |
2860 | } | |
2861 | } | |
c773e847 | 2862 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
2863 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
2864 | dev->name); | |
1da177e4 LT |
2865 | } else { |
2866 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2867 | * unfortunately |
2868 | */ | |
2869 | recursion_alert: | |
e87cc472 JP |
2870 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
2871 | dev->name); | |
1da177e4 LT |
2872 | } |
2873 | } | |
2874 | ||
2875 | rc = -ENETDOWN; | |
d4828d85 | 2876 | rcu_read_unlock_bh(); |
1da177e4 | 2877 | |
1da177e4 LT |
2878 | kfree_skb(skb); |
2879 | return rc; | |
2880 | out: | |
d4828d85 | 2881 | rcu_read_unlock_bh(); |
1da177e4 LT |
2882 | return rc; |
2883 | } | |
d1b19dff | 2884 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2885 | |
2886 | ||
2887 | /*======================================================================= | |
2888 | Receiver routines | |
2889 | =======================================================================*/ | |
2890 | ||
6b2bedc3 | 2891 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
2892 | EXPORT_SYMBOL(netdev_max_backlog); |
2893 | ||
3b098e2d | 2894 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2895 | int netdev_budget __read_mostly = 300; |
2896 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2897 | |
eecfd7c4 ED |
2898 | /* Called with irq disabled */ |
2899 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2900 | struct napi_struct *napi) | |
2901 | { | |
2902 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2903 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2904 | } | |
2905 | ||
bfb564e7 KK |
2906 | #ifdef CONFIG_RPS |
2907 | ||
2908 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 2909 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
2910 | EXPORT_SYMBOL(rps_sock_flow_table); |
2911 | ||
c5905afb | 2912 | struct static_key rps_needed __read_mostly; |
adc9300e | 2913 | |
c445477d BH |
2914 | static struct rps_dev_flow * |
2915 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2916 | struct rps_dev_flow *rflow, u16 next_cpu) | |
2917 | { | |
09994d1b | 2918 | if (next_cpu != RPS_NO_CPU) { |
c445477d BH |
2919 | #ifdef CONFIG_RFS_ACCEL |
2920 | struct netdev_rx_queue *rxqueue; | |
2921 | struct rps_dev_flow_table *flow_table; | |
2922 | struct rps_dev_flow *old_rflow; | |
2923 | u32 flow_id; | |
2924 | u16 rxq_index; | |
2925 | int rc; | |
2926 | ||
2927 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
2928 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
2929 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
2930 | goto out; |
2931 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
2932 | if (rxq_index == skb_get_rx_queue(skb)) | |
2933 | goto out; | |
2934 | ||
2935 | rxqueue = dev->_rx + rxq_index; | |
2936 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2937 | if (!flow_table) | |
2938 | goto out; | |
2939 | flow_id = skb->rxhash & flow_table->mask; | |
2940 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, | |
2941 | rxq_index, flow_id); | |
2942 | if (rc < 0) | |
2943 | goto out; | |
2944 | old_rflow = rflow; | |
2945 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
2946 | rflow->filter = rc; |
2947 | if (old_rflow->filter == rflow->filter) | |
2948 | old_rflow->filter = RPS_NO_FILTER; | |
2949 | out: | |
2950 | #endif | |
2951 | rflow->last_qtail = | |
09994d1b | 2952 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
2953 | } |
2954 | ||
09994d1b | 2955 | rflow->cpu = next_cpu; |
c445477d BH |
2956 | return rflow; |
2957 | } | |
2958 | ||
bfb564e7 KK |
2959 | /* |
2960 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2961 | * CPU from the RPS map of the receiving queue for a given skb. | |
2962 | * rcu_read_lock must be held on entry. | |
2963 | */ | |
2964 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2965 | struct rps_dev_flow **rflowp) | |
2966 | { | |
2967 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 2968 | struct rps_map *map; |
bfb564e7 KK |
2969 | struct rps_dev_flow_table *flow_table; |
2970 | struct rps_sock_flow_table *sock_flow_table; | |
2971 | int cpu = -1; | |
2972 | u16 tcpu; | |
2973 | ||
2974 | if (skb_rx_queue_recorded(skb)) { | |
2975 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
2976 | if (unlikely(index >= dev->real_num_rx_queues)) { |
2977 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
2978 | "%s received packet on queue %u, but number " | |
2979 | "of RX queues is %u\n", | |
2980 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
2981 | goto done; |
2982 | } | |
2983 | rxqueue = dev->_rx + index; | |
2984 | } else | |
2985 | rxqueue = dev->_rx; | |
2986 | ||
6e3f7faf ED |
2987 | map = rcu_dereference(rxqueue->rps_map); |
2988 | if (map) { | |
85875236 | 2989 | if (map->len == 1 && |
33d480ce | 2990 | !rcu_access_pointer(rxqueue->rps_flow_table)) { |
6febfca9 CG |
2991 | tcpu = map->cpus[0]; |
2992 | if (cpu_online(tcpu)) | |
2993 | cpu = tcpu; | |
2994 | goto done; | |
2995 | } | |
33d480ce | 2996 | } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) { |
bfb564e7 | 2997 | goto done; |
6febfca9 | 2998 | } |
bfb564e7 | 2999 | |
2d47b459 | 3000 | skb_reset_network_header(skb); |
bfb564e7 KK |
3001 | if (!skb_get_rxhash(skb)) |
3002 | goto done; | |
3003 | ||
fec5e652 TH |
3004 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
3005 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
3006 | if (flow_table && sock_flow_table) { | |
3007 | u16 next_cpu; | |
3008 | struct rps_dev_flow *rflow; | |
3009 | ||
3010 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
3011 | tcpu = rflow->cpu; | |
3012 | ||
3013 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
3014 | sock_flow_table->mask]; | |
3015 | ||
3016 | /* | |
3017 | * If the desired CPU (where last recvmsg was done) is | |
3018 | * different from current CPU (one in the rx-queue flow | |
3019 | * table entry), switch if one of the following holds: | |
3020 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
3021 | * - Current CPU is offline. | |
3022 | * - The current CPU's queue tail has advanced beyond the | |
3023 | * last packet that was enqueued using this table entry. | |
3024 | * This guarantees that all previous packets for the flow | |
3025 | * have been dequeued, thus preserving in order delivery. | |
3026 | */ | |
3027 | if (unlikely(tcpu != next_cpu) && | |
3028 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
3029 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
baefa31d TH |
3030 | rflow->last_qtail)) >= 0)) { |
3031 | tcpu = next_cpu; | |
c445477d | 3032 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3033 | } |
c445477d | 3034 | |
fec5e652 TH |
3035 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
3036 | *rflowp = rflow; | |
3037 | cpu = tcpu; | |
3038 | goto done; | |
3039 | } | |
3040 | } | |
3041 | ||
0a9627f2 | 3042 | if (map) { |
fec5e652 | 3043 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; |
0a9627f2 TH |
3044 | |
3045 | if (cpu_online(tcpu)) { | |
3046 | cpu = tcpu; | |
3047 | goto done; | |
3048 | } | |
3049 | } | |
3050 | ||
3051 | done: | |
0a9627f2 TH |
3052 | return cpu; |
3053 | } | |
3054 | ||
c445477d BH |
3055 | #ifdef CONFIG_RFS_ACCEL |
3056 | ||
3057 | /** | |
3058 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3059 | * @dev: Device on which the filter was set | |
3060 | * @rxq_index: RX queue index | |
3061 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3062 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3063 | * | |
3064 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3065 | * this function for each installed filter and remove the filters for | |
3066 | * which it returns %true. | |
3067 | */ | |
3068 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3069 | u32 flow_id, u16 filter_id) | |
3070 | { | |
3071 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3072 | struct rps_dev_flow_table *flow_table; | |
3073 | struct rps_dev_flow *rflow; | |
3074 | bool expire = true; | |
3075 | int cpu; | |
3076 | ||
3077 | rcu_read_lock(); | |
3078 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3079 | if (flow_table && flow_id <= flow_table->mask) { | |
3080 | rflow = &flow_table->flows[flow_id]; | |
3081 | cpu = ACCESS_ONCE(rflow->cpu); | |
3082 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
3083 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
3084 | rflow->last_qtail) < | |
3085 | (int)(10 * flow_table->mask))) | |
3086 | expire = false; | |
3087 | } | |
3088 | rcu_read_unlock(); | |
3089 | return expire; | |
3090 | } | |
3091 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3092 | ||
3093 | #endif /* CONFIG_RFS_ACCEL */ | |
3094 | ||
0a9627f2 | 3095 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3096 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3097 | { |
e36fa2f7 ED |
3098 | struct softnet_data *sd = data; |
3099 | ||
eecfd7c4 | 3100 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3101 | sd->received_rps++; |
0a9627f2 | 3102 | } |
e36fa2f7 | 3103 | |
fec5e652 | 3104 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3105 | |
e36fa2f7 ED |
3106 | /* |
3107 | * Check if this softnet_data structure is another cpu one | |
3108 | * If yes, queue it to our IPI list and return 1 | |
3109 | * If no, return 0 | |
3110 | */ | |
3111 | static int rps_ipi_queued(struct softnet_data *sd) | |
3112 | { | |
3113 | #ifdef CONFIG_RPS | |
3114 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
3115 | ||
3116 | if (sd != mysd) { | |
3117 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3118 | mysd->rps_ipi_list = sd; | |
3119 | ||
3120 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3121 | return 1; | |
3122 | } | |
3123 | #endif /* CONFIG_RPS */ | |
3124 | return 0; | |
3125 | } | |
3126 | ||
99bbc707 WB |
3127 | #ifdef CONFIG_NET_FLOW_LIMIT |
3128 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3129 | #endif | |
3130 | ||
3131 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3132 | { | |
3133 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3134 | struct sd_flow_limit *fl; | |
3135 | struct softnet_data *sd; | |
3136 | unsigned int old_flow, new_flow; | |
3137 | ||
3138 | if (qlen < (netdev_max_backlog >> 1)) | |
3139 | return false; | |
3140 | ||
3141 | sd = &__get_cpu_var(softnet_data); | |
3142 | ||
3143 | rcu_read_lock(); | |
3144 | fl = rcu_dereference(sd->flow_limit); | |
3145 | if (fl) { | |
3146 | new_flow = skb_get_rxhash(skb) & (fl->num_buckets - 1); | |
3147 | old_flow = fl->history[fl->history_head]; | |
3148 | fl->history[fl->history_head] = new_flow; | |
3149 | ||
3150 | fl->history_head++; | |
3151 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3152 | ||
3153 | if (likely(fl->buckets[old_flow])) | |
3154 | fl->buckets[old_flow]--; | |
3155 | ||
3156 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3157 | fl->count++; | |
3158 | rcu_read_unlock(); | |
3159 | return true; | |
3160 | } | |
3161 | } | |
3162 | rcu_read_unlock(); | |
3163 | #endif | |
3164 | return false; | |
3165 | } | |
3166 | ||
0a9627f2 TH |
3167 | /* |
3168 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3169 | * queue (may be a remote CPU queue). | |
3170 | */ | |
fec5e652 TH |
3171 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3172 | unsigned int *qtail) | |
0a9627f2 | 3173 | { |
e36fa2f7 | 3174 | struct softnet_data *sd; |
0a9627f2 | 3175 | unsigned long flags; |
99bbc707 | 3176 | unsigned int qlen; |
0a9627f2 | 3177 | |
e36fa2f7 | 3178 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3179 | |
3180 | local_irq_save(flags); | |
0a9627f2 | 3181 | |
e36fa2f7 | 3182 | rps_lock(sd); |
99bbc707 WB |
3183 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3184 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
6e7676c1 | 3185 | if (skb_queue_len(&sd->input_pkt_queue)) { |
0a9627f2 | 3186 | enqueue: |
e36fa2f7 | 3187 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3188 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3189 | rps_unlock(sd); |
152102c7 | 3190 | local_irq_restore(flags); |
0a9627f2 TH |
3191 | return NET_RX_SUCCESS; |
3192 | } | |
3193 | ||
ebda37c2 ED |
3194 | /* Schedule NAPI for backlog device |
3195 | * We can use non atomic operation since we own the queue lock | |
3196 | */ | |
3197 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3198 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3199 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3200 | } |
3201 | goto enqueue; | |
3202 | } | |
3203 | ||
dee42870 | 3204 | sd->dropped++; |
e36fa2f7 | 3205 | rps_unlock(sd); |
0a9627f2 | 3206 | |
0a9627f2 TH |
3207 | local_irq_restore(flags); |
3208 | ||
caf586e5 | 3209 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3210 | kfree_skb(skb); |
3211 | return NET_RX_DROP; | |
3212 | } | |
1da177e4 | 3213 | |
1da177e4 LT |
3214 | /** |
3215 | * netif_rx - post buffer to the network code | |
3216 | * @skb: buffer to post | |
3217 | * | |
3218 | * This function receives a packet from a device driver and queues it for | |
3219 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3220 | * may be dropped during processing for congestion control or by the | |
3221 | * protocol layers. | |
3222 | * | |
3223 | * return values: | |
3224 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
3225 | * NET_RX_DROP (packet was dropped) |
3226 | * | |
3227 | */ | |
3228 | ||
3229 | int netif_rx(struct sk_buff *skb) | |
3230 | { | |
b0e28f1e | 3231 | int ret; |
1da177e4 LT |
3232 | |
3233 | /* if netpoll wants it, pretend we never saw it */ | |
3234 | if (netpoll_rx(skb)) | |
3235 | return NET_RX_DROP; | |
3236 | ||
588f0330 | 3237 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3238 | |
cf66ba58 | 3239 | trace_netif_rx(skb); |
df334545 | 3240 | #ifdef CONFIG_RPS |
c5905afb | 3241 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3242 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3243 | int cpu; |
3244 | ||
cece1945 | 3245 | preempt_disable(); |
b0e28f1e | 3246 | rcu_read_lock(); |
fec5e652 TH |
3247 | |
3248 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3249 | if (cpu < 0) |
3250 | cpu = smp_processor_id(); | |
fec5e652 TH |
3251 | |
3252 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3253 | ||
b0e28f1e | 3254 | rcu_read_unlock(); |
cece1945 | 3255 | preempt_enable(); |
adc9300e ED |
3256 | } else |
3257 | #endif | |
fec5e652 TH |
3258 | { |
3259 | unsigned int qtail; | |
3260 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3261 | put_cpu(); | |
3262 | } | |
b0e28f1e | 3263 | return ret; |
1da177e4 | 3264 | } |
d1b19dff | 3265 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3266 | |
3267 | int netif_rx_ni(struct sk_buff *skb) | |
3268 | { | |
3269 | int err; | |
3270 | ||
3271 | preempt_disable(); | |
3272 | err = netif_rx(skb); | |
3273 | if (local_softirq_pending()) | |
3274 | do_softirq(); | |
3275 | preempt_enable(); | |
3276 | ||
3277 | return err; | |
3278 | } | |
1da177e4 LT |
3279 | EXPORT_SYMBOL(netif_rx_ni); |
3280 | ||
1da177e4 LT |
3281 | static void net_tx_action(struct softirq_action *h) |
3282 | { | |
3283 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
3284 | ||
3285 | if (sd->completion_queue) { | |
3286 | struct sk_buff *clist; | |
3287 | ||
3288 | local_irq_disable(); | |
3289 | clist = sd->completion_queue; | |
3290 | sd->completion_queue = NULL; | |
3291 | local_irq_enable(); | |
3292 | ||
3293 | while (clist) { | |
3294 | struct sk_buff *skb = clist; | |
3295 | clist = clist->next; | |
3296 | ||
547b792c | 3297 | WARN_ON(atomic_read(&skb->users)); |
07dc22e7 | 3298 | trace_kfree_skb(skb, net_tx_action); |
1da177e4 LT |
3299 | __kfree_skb(skb); |
3300 | } | |
3301 | } | |
3302 | ||
3303 | if (sd->output_queue) { | |
37437bb2 | 3304 | struct Qdisc *head; |
1da177e4 LT |
3305 | |
3306 | local_irq_disable(); | |
3307 | head = sd->output_queue; | |
3308 | sd->output_queue = NULL; | |
a9cbd588 | 3309 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3310 | local_irq_enable(); |
3311 | ||
3312 | while (head) { | |
37437bb2 DM |
3313 | struct Qdisc *q = head; |
3314 | spinlock_t *root_lock; | |
3315 | ||
1da177e4 LT |
3316 | head = head->next_sched; |
3317 | ||
5fb66229 | 3318 | root_lock = qdisc_lock(q); |
37437bb2 | 3319 | if (spin_trylock(root_lock)) { |
def82a1d JP |
3320 | smp_mb__before_clear_bit(); |
3321 | clear_bit(__QDISC_STATE_SCHED, | |
3322 | &q->state); | |
37437bb2 DM |
3323 | qdisc_run(q); |
3324 | spin_unlock(root_lock); | |
1da177e4 | 3325 | } else { |
195648bb | 3326 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3327 | &q->state)) { |
195648bb | 3328 | __netif_reschedule(q); |
e8a83e10 JP |
3329 | } else { |
3330 | smp_mb__before_clear_bit(); | |
3331 | clear_bit(__QDISC_STATE_SCHED, | |
3332 | &q->state); | |
3333 | } | |
1da177e4 LT |
3334 | } |
3335 | } | |
3336 | } | |
3337 | } | |
3338 | ||
ab95bfe0 JP |
3339 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3340 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3341 | /* This hook is defined here for ATM LANE */ |
3342 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3343 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3344 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3345 | #endif |
1da177e4 | 3346 | |
1da177e4 LT |
3347 | #ifdef CONFIG_NET_CLS_ACT |
3348 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3349 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3350 | * a compare and 2 stores extra right now if we dont have it on | |
3351 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3352 | * NOTE: This doesn't stop any functionality; if you dont have |
3353 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3354 | * |
3355 | */ | |
24824a09 | 3356 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3357 | { |
1da177e4 | 3358 | struct net_device *dev = skb->dev; |
f697c3e8 | 3359 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3360 | int result = TC_ACT_OK; |
3361 | struct Qdisc *q; | |
4ec93edb | 3362 | |
de384830 | 3363 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
e87cc472 JP |
3364 | net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n", |
3365 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3366 | return TC_ACT_SHOT; |
3367 | } | |
1da177e4 | 3368 | |
f697c3e8 HX |
3369 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3370 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3371 | |
83874000 | 3372 | q = rxq->qdisc; |
8d50b53d | 3373 | if (q != &noop_qdisc) { |
83874000 | 3374 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3375 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3376 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3377 | spin_unlock(qdisc_lock(q)); |
3378 | } | |
f697c3e8 HX |
3379 | |
3380 | return result; | |
3381 | } | |
86e65da9 | 3382 | |
f697c3e8 HX |
3383 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3384 | struct packet_type **pt_prev, | |
3385 | int *ret, struct net_device *orig_dev) | |
3386 | { | |
24824a09 ED |
3387 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3388 | ||
3389 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
f697c3e8 | 3390 | goto out; |
1da177e4 | 3391 | |
f697c3e8 HX |
3392 | if (*pt_prev) { |
3393 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3394 | *pt_prev = NULL; | |
1da177e4 LT |
3395 | } |
3396 | ||
24824a09 | 3397 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3398 | case TC_ACT_SHOT: |
3399 | case TC_ACT_STOLEN: | |
3400 | kfree_skb(skb); | |
3401 | return NULL; | |
3402 | } | |
3403 | ||
3404 | out: | |
3405 | skb->tc_verd = 0; | |
3406 | return skb; | |
1da177e4 LT |
3407 | } |
3408 | #endif | |
3409 | ||
ab95bfe0 JP |
3410 | /** |
3411 | * netdev_rx_handler_register - register receive handler | |
3412 | * @dev: device to register a handler for | |
3413 | * @rx_handler: receive handler to register | |
93e2c32b | 3414 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 JP |
3415 | * |
3416 | * Register a receive hander for a device. This handler will then be | |
3417 | * called from __netif_receive_skb. A negative errno code is returned | |
3418 | * on a failure. | |
3419 | * | |
3420 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3421 | * |
3422 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3423 | */ |
3424 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3425 | rx_handler_func_t *rx_handler, |
3426 | void *rx_handler_data) | |
ab95bfe0 JP |
3427 | { |
3428 | ASSERT_RTNL(); | |
3429 | ||
3430 | if (dev->rx_handler) | |
3431 | return -EBUSY; | |
3432 | ||
00cfec37 | 3433 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 3434 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3435 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3436 | ||
3437 | return 0; | |
3438 | } | |
3439 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3440 | ||
3441 | /** | |
3442 | * netdev_rx_handler_unregister - unregister receive handler | |
3443 | * @dev: device to unregister a handler from | |
3444 | * | |
166ec369 | 3445 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
3446 | * |
3447 | * The caller must hold the rtnl_mutex. | |
3448 | */ | |
3449 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3450 | { | |
3451 | ||
3452 | ASSERT_RTNL(); | |
a9b3cd7f | 3453 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
3454 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
3455 | * section has a guarantee to see a non NULL rx_handler_data | |
3456 | * as well. | |
3457 | */ | |
3458 | synchronize_net(); | |
a9b3cd7f | 3459 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
3460 | } |
3461 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3462 | ||
b4b9e355 MG |
3463 | /* |
3464 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3465 | * the special handling of PFMEMALLOC skbs. | |
3466 | */ | |
3467 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3468 | { | |
3469 | switch (skb->protocol) { | |
3470 | case __constant_htons(ETH_P_ARP): | |
3471 | case __constant_htons(ETH_P_IP): | |
3472 | case __constant_htons(ETH_P_IPV6): | |
3473 | case __constant_htons(ETH_P_8021Q): | |
8ad227ff | 3474 | case __constant_htons(ETH_P_8021AD): |
b4b9e355 MG |
3475 | return true; |
3476 | default: | |
3477 | return false; | |
3478 | } | |
3479 | } | |
3480 | ||
9754e293 | 3481 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
3482 | { |
3483 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3484 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3485 | struct net_device *orig_dev; |
63d8ea7f | 3486 | struct net_device *null_or_dev; |
8a4eb573 | 3487 | bool deliver_exact = false; |
1da177e4 | 3488 | int ret = NET_RX_DROP; |
252e3346 | 3489 | __be16 type; |
1da177e4 | 3490 | |
588f0330 | 3491 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3492 | |
cf66ba58 | 3493 | trace_netif_receive_skb(skb); |
9b22ea56 | 3494 | |
1da177e4 | 3495 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 3496 | if (netpoll_receive_skb(skb)) |
b4b9e355 | 3497 | goto out; |
1da177e4 | 3498 | |
cc9bd5ce | 3499 | orig_dev = skb->dev; |
8f903c70 | 3500 | |
c1d2bbe1 | 3501 | skb_reset_network_header(skb); |
fda55eca ED |
3502 | if (!skb_transport_header_was_set(skb)) |
3503 | skb_reset_transport_header(skb); | |
0b5c9db1 | 3504 | skb_reset_mac_len(skb); |
1da177e4 LT |
3505 | |
3506 | pt_prev = NULL; | |
3507 | ||
3508 | rcu_read_lock(); | |
3509 | ||
63d8ea7f | 3510 | another_round: |
b6858177 | 3511 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
3512 | |
3513 | __this_cpu_inc(softnet_data.processed); | |
3514 | ||
8ad227ff PM |
3515 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
3516 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
bcc6d479 JP |
3517 | skb = vlan_untag(skb); |
3518 | if (unlikely(!skb)) | |
b4b9e355 | 3519 | goto unlock; |
bcc6d479 JP |
3520 | } |
3521 | ||
1da177e4 LT |
3522 | #ifdef CONFIG_NET_CLS_ACT |
3523 | if (skb->tc_verd & TC_NCLS) { | |
3524 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3525 | goto ncls; | |
3526 | } | |
3527 | #endif | |
3528 | ||
9754e293 | 3529 | if (pfmemalloc) |
b4b9e355 MG |
3530 | goto skip_taps; |
3531 | ||
1da177e4 | 3532 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
63d8ea7f | 3533 | if (!ptype->dev || ptype->dev == skb->dev) { |
4ec93edb | 3534 | if (pt_prev) |
f2ccd8fa | 3535 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3536 | pt_prev = ptype; |
3537 | } | |
3538 | } | |
3539 | ||
b4b9e355 | 3540 | skip_taps: |
1da177e4 | 3541 | #ifdef CONFIG_NET_CLS_ACT |
f697c3e8 HX |
3542 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3543 | if (!skb) | |
b4b9e355 | 3544 | goto unlock; |
1da177e4 LT |
3545 | ncls: |
3546 | #endif | |
3547 | ||
9754e293 | 3548 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
3549 | goto drop; |
3550 | ||
2425717b JF |
3551 | if (vlan_tx_tag_present(skb)) { |
3552 | if (pt_prev) { | |
3553 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3554 | pt_prev = NULL; | |
3555 | } | |
48cc32d3 | 3556 | if (vlan_do_receive(&skb)) |
2425717b JF |
3557 | goto another_round; |
3558 | else if (unlikely(!skb)) | |
b4b9e355 | 3559 | goto unlock; |
2425717b JF |
3560 | } |
3561 | ||
48cc32d3 | 3562 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
3563 | if (rx_handler) { |
3564 | if (pt_prev) { | |
3565 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3566 | pt_prev = NULL; | |
3567 | } | |
8a4eb573 JP |
3568 | switch (rx_handler(&skb)) { |
3569 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 3570 | ret = NET_RX_SUCCESS; |
b4b9e355 | 3571 | goto unlock; |
8a4eb573 | 3572 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3573 | goto another_round; |
8a4eb573 JP |
3574 | case RX_HANDLER_EXACT: |
3575 | deliver_exact = true; | |
3576 | case RX_HANDLER_PASS: | |
3577 | break; | |
3578 | default: | |
3579 | BUG(); | |
3580 | } | |
ab95bfe0 | 3581 | } |
1da177e4 | 3582 | |
d4b812de ED |
3583 | if (unlikely(vlan_tx_tag_present(skb))) { |
3584 | if (vlan_tx_tag_get_id(skb)) | |
3585 | skb->pkt_type = PACKET_OTHERHOST; | |
3586 | /* Note: we might in the future use prio bits | |
3587 | * and set skb->priority like in vlan_do_receive() | |
3588 | * For the time being, just ignore Priority Code Point | |
3589 | */ | |
3590 | skb->vlan_tci = 0; | |
3591 | } | |
48cc32d3 | 3592 | |
63d8ea7f | 3593 | /* deliver only exact match when indicated */ |
8a4eb573 | 3594 | null_or_dev = deliver_exact ? skb->dev : NULL; |
1f3c8804 | 3595 | |
1da177e4 | 3596 | type = skb->protocol; |
82d8a867 PE |
3597 | list_for_each_entry_rcu(ptype, |
3598 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
63d8ea7f | 3599 | if (ptype->type == type && |
e3f48d37 JP |
3600 | (ptype->dev == null_or_dev || ptype->dev == skb->dev || |
3601 | ptype->dev == orig_dev)) { | |
4ec93edb | 3602 | if (pt_prev) |
f2ccd8fa | 3603 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3604 | pt_prev = ptype; |
3605 | } | |
3606 | } | |
3607 | ||
3608 | if (pt_prev) { | |
1080e512 | 3609 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 3610 | goto drop; |
1080e512 MT |
3611 | else |
3612 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 3613 | } else { |
b4b9e355 | 3614 | drop: |
caf586e5 | 3615 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3616 | kfree_skb(skb); |
3617 | /* Jamal, now you will not able to escape explaining | |
3618 | * me how you were going to use this. :-) | |
3619 | */ | |
3620 | ret = NET_RX_DROP; | |
3621 | } | |
3622 | ||
b4b9e355 | 3623 | unlock: |
1da177e4 | 3624 | rcu_read_unlock(); |
b4b9e355 | 3625 | out: |
9754e293 DM |
3626 | return ret; |
3627 | } | |
3628 | ||
3629 | static int __netif_receive_skb(struct sk_buff *skb) | |
3630 | { | |
3631 | int ret; | |
3632 | ||
3633 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
3634 | unsigned long pflags = current->flags; | |
3635 | ||
3636 | /* | |
3637 | * PFMEMALLOC skbs are special, they should | |
3638 | * - be delivered to SOCK_MEMALLOC sockets only | |
3639 | * - stay away from userspace | |
3640 | * - have bounded memory usage | |
3641 | * | |
3642 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
3643 | * context down to all allocation sites. | |
3644 | */ | |
3645 | current->flags |= PF_MEMALLOC; | |
3646 | ret = __netif_receive_skb_core(skb, true); | |
3647 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
3648 | } else | |
3649 | ret = __netif_receive_skb_core(skb, false); | |
3650 | ||
1da177e4 LT |
3651 | return ret; |
3652 | } | |
0a9627f2 TH |
3653 | |
3654 | /** | |
3655 | * netif_receive_skb - process receive buffer from network | |
3656 | * @skb: buffer to process | |
3657 | * | |
3658 | * netif_receive_skb() is the main receive data processing function. | |
3659 | * It always succeeds. The buffer may be dropped during processing | |
3660 | * for congestion control or by the protocol layers. | |
3661 | * | |
3662 | * This function may only be called from softirq context and interrupts | |
3663 | * should be enabled. | |
3664 | * | |
3665 | * Return values (usually ignored): | |
3666 | * NET_RX_SUCCESS: no congestion | |
3667 | * NET_RX_DROP: packet was dropped | |
3668 | */ | |
3669 | int netif_receive_skb(struct sk_buff *skb) | |
3670 | { | |
588f0330 | 3671 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3672 | |
c1f19b51 RC |
3673 | if (skb_defer_rx_timestamp(skb)) |
3674 | return NET_RX_SUCCESS; | |
3675 | ||
df334545 | 3676 | #ifdef CONFIG_RPS |
c5905afb | 3677 | if (static_key_false(&rps_needed)) { |
3b098e2d ED |
3678 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
3679 | int cpu, ret; | |
fec5e652 | 3680 | |
3b098e2d ED |
3681 | rcu_read_lock(); |
3682 | ||
3683 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3684 | |
3b098e2d ED |
3685 | if (cpu >= 0) { |
3686 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3687 | rcu_read_unlock(); | |
adc9300e | 3688 | return ret; |
3b098e2d | 3689 | } |
adc9300e | 3690 | rcu_read_unlock(); |
fec5e652 | 3691 | } |
1e94d72f | 3692 | #endif |
adc9300e | 3693 | return __netif_receive_skb(skb); |
0a9627f2 | 3694 | } |
d1b19dff | 3695 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3696 | |
88751275 ED |
3697 | /* Network device is going away, flush any packets still pending |
3698 | * Called with irqs disabled. | |
3699 | */ | |
152102c7 | 3700 | static void flush_backlog(void *arg) |
6e583ce5 | 3701 | { |
152102c7 | 3702 | struct net_device *dev = arg; |
e36fa2f7 | 3703 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3704 | struct sk_buff *skb, *tmp; |
3705 | ||
e36fa2f7 | 3706 | rps_lock(sd); |
6e7676c1 | 3707 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3708 | if (skb->dev == dev) { |
e36fa2f7 | 3709 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3710 | kfree_skb(skb); |
76cc8b13 | 3711 | input_queue_head_incr(sd); |
6e583ce5 | 3712 | } |
6e7676c1 | 3713 | } |
e36fa2f7 | 3714 | rps_unlock(sd); |
6e7676c1 CG |
3715 | |
3716 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3717 | if (skb->dev == dev) { | |
3718 | __skb_unlink(skb, &sd->process_queue); | |
3719 | kfree_skb(skb); | |
76cc8b13 | 3720 | input_queue_head_incr(sd); |
6e7676c1 CG |
3721 | } |
3722 | } | |
6e583ce5 SH |
3723 | } |
3724 | ||
d565b0a1 HX |
3725 | static int napi_gro_complete(struct sk_buff *skb) |
3726 | { | |
22061d80 | 3727 | struct packet_offload *ptype; |
d565b0a1 | 3728 | __be16 type = skb->protocol; |
22061d80 | 3729 | struct list_head *head = &offload_base; |
d565b0a1 HX |
3730 | int err = -ENOENT; |
3731 | ||
c3c7c254 ED |
3732 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
3733 | ||
fc59f9a3 HX |
3734 | if (NAPI_GRO_CB(skb)->count == 1) { |
3735 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3736 | goto out; |
fc59f9a3 | 3737 | } |
d565b0a1 HX |
3738 | |
3739 | rcu_read_lock(); | |
3740 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3741 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
3742 | continue; |
3743 | ||
f191a1d1 | 3744 | err = ptype->callbacks.gro_complete(skb); |
d565b0a1 HX |
3745 | break; |
3746 | } | |
3747 | rcu_read_unlock(); | |
3748 | ||
3749 | if (err) { | |
3750 | WARN_ON(&ptype->list == head); | |
3751 | kfree_skb(skb); | |
3752 | return NET_RX_SUCCESS; | |
3753 | } | |
3754 | ||
3755 | out: | |
d565b0a1 HX |
3756 | return netif_receive_skb(skb); |
3757 | } | |
3758 | ||
2e71a6f8 ED |
3759 | /* napi->gro_list contains packets ordered by age. |
3760 | * youngest packets at the head of it. | |
3761 | * Complete skbs in reverse order to reduce latencies. | |
3762 | */ | |
3763 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 3764 | { |
2e71a6f8 | 3765 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 3766 | |
2e71a6f8 ED |
3767 | /* scan list and build reverse chain */ |
3768 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
3769 | skb->prev = prev; | |
3770 | prev = skb; | |
3771 | } | |
3772 | ||
3773 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 3774 | skb->next = NULL; |
2e71a6f8 ED |
3775 | |
3776 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
3777 | return; | |
3778 | ||
3779 | prev = skb->prev; | |
d565b0a1 | 3780 | napi_gro_complete(skb); |
2e71a6f8 | 3781 | napi->gro_count--; |
d565b0a1 HX |
3782 | } |
3783 | ||
3784 | napi->gro_list = NULL; | |
3785 | } | |
86cac58b | 3786 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3787 | |
89c5fa33 ED |
3788 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
3789 | { | |
3790 | struct sk_buff *p; | |
3791 | unsigned int maclen = skb->dev->hard_header_len; | |
3792 | ||
3793 | for (p = napi->gro_list; p; p = p->next) { | |
3794 | unsigned long diffs; | |
3795 | ||
3796 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; | |
3797 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
3798 | if (maclen == ETH_HLEN) | |
3799 | diffs |= compare_ether_header(skb_mac_header(p), | |
3800 | skb_gro_mac_header(skb)); | |
3801 | else if (!diffs) | |
3802 | diffs = memcmp(skb_mac_header(p), | |
3803 | skb_gro_mac_header(skb), | |
3804 | maclen); | |
3805 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
3806 | NAPI_GRO_CB(p)->flush = 0; | |
3807 | } | |
3808 | } | |
3809 | ||
bb728820 | 3810 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3811 | { |
3812 | struct sk_buff **pp = NULL; | |
22061d80 | 3813 | struct packet_offload *ptype; |
d565b0a1 | 3814 | __be16 type = skb->protocol; |
22061d80 | 3815 | struct list_head *head = &offload_base; |
0da2afd5 | 3816 | int same_flow; |
5b252f0c | 3817 | enum gro_result ret; |
d565b0a1 | 3818 | |
ce9e76c8 | 3819 | if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb)) |
d565b0a1 HX |
3820 | goto normal; |
3821 | ||
21dc3301 | 3822 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) |
f17f5c91 HX |
3823 | goto normal; |
3824 | ||
89c5fa33 ED |
3825 | gro_list_prepare(napi, skb); |
3826 | ||
d565b0a1 HX |
3827 | rcu_read_lock(); |
3828 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3829 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
3830 | continue; |
3831 | ||
86911732 | 3832 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 3833 | skb_reset_mac_len(skb); |
d565b0a1 HX |
3834 | NAPI_GRO_CB(skb)->same_flow = 0; |
3835 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3836 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 3837 | |
f191a1d1 | 3838 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
3839 | break; |
3840 | } | |
3841 | rcu_read_unlock(); | |
3842 | ||
3843 | if (&ptype->list == head) | |
3844 | goto normal; | |
3845 | ||
0da2afd5 | 3846 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3847 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3848 | |
d565b0a1 HX |
3849 | if (pp) { |
3850 | struct sk_buff *nskb = *pp; | |
3851 | ||
3852 | *pp = nskb->next; | |
3853 | nskb->next = NULL; | |
3854 | napi_gro_complete(nskb); | |
4ae5544f | 3855 | napi->gro_count--; |
d565b0a1 HX |
3856 | } |
3857 | ||
0da2afd5 | 3858 | if (same_flow) |
d565b0a1 HX |
3859 | goto ok; |
3860 | ||
4ae5544f | 3861 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 3862 | goto normal; |
d565b0a1 | 3863 | |
4ae5544f | 3864 | napi->gro_count++; |
d565b0a1 | 3865 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 3866 | NAPI_GRO_CB(skb)->age = jiffies; |
86911732 | 3867 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3868 | skb->next = napi->gro_list; |
3869 | napi->gro_list = skb; | |
5d0d9be8 | 3870 | ret = GRO_HELD; |
d565b0a1 | 3871 | |
ad0f9904 | 3872 | pull: |
cb18978c HX |
3873 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
3874 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3875 | ||
3876 | BUG_ON(skb->end - skb->tail < grow); | |
3877 | ||
3878 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3879 | ||
3880 | skb->tail += grow; | |
3881 | skb->data_len -= grow; | |
3882 | ||
3883 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
9e903e08 | 3884 | skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow); |
cb18978c | 3885 | |
9e903e08 | 3886 | if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) { |
ea2ab693 | 3887 | skb_frag_unref(skb, 0); |
cb18978c HX |
3888 | memmove(skb_shinfo(skb)->frags, |
3889 | skb_shinfo(skb)->frags + 1, | |
e5093aec | 3890 | --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); |
cb18978c | 3891 | } |
ad0f9904 HX |
3892 | } |
3893 | ||
d565b0a1 | 3894 | ok: |
5d0d9be8 | 3895 | return ret; |
d565b0a1 HX |
3896 | |
3897 | normal: | |
ad0f9904 HX |
3898 | ret = GRO_NORMAL; |
3899 | goto pull; | |
5d38a079 | 3900 | } |
96e93eab | 3901 | |
5d38a079 | 3902 | |
bb728820 | 3903 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3904 | { |
5d0d9be8 HX |
3905 | switch (ret) { |
3906 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3907 | if (netif_receive_skb(skb)) |
3908 | ret = GRO_DROP; | |
3909 | break; | |
5d38a079 | 3910 | |
5d0d9be8 | 3911 | case GRO_DROP: |
5d38a079 HX |
3912 | kfree_skb(skb); |
3913 | break; | |
5b252f0c | 3914 | |
daa86548 | 3915 | case GRO_MERGED_FREE: |
d7e8883c ED |
3916 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
3917 | kmem_cache_free(skbuff_head_cache, skb); | |
3918 | else | |
3919 | __kfree_skb(skb); | |
daa86548 ED |
3920 | break; |
3921 | ||
5b252f0c BH |
3922 | case GRO_HELD: |
3923 | case GRO_MERGED: | |
3924 | break; | |
5d38a079 HX |
3925 | } |
3926 | ||
c7c4b3b6 | 3927 | return ret; |
5d0d9be8 | 3928 | } |
5d0d9be8 | 3929 | |
ca07e43e | 3930 | static void skb_gro_reset_offset(struct sk_buff *skb) |
78a478d0 | 3931 | { |
ca07e43e ED |
3932 | const struct skb_shared_info *pinfo = skb_shinfo(skb); |
3933 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
3934 | ||
78a478d0 HX |
3935 | NAPI_GRO_CB(skb)->data_offset = 0; |
3936 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 3937 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 3938 | |
ced14f68 | 3939 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && |
ca07e43e ED |
3940 | pinfo->nr_frags && |
3941 | !PageHighMem(skb_frag_page(frag0))) { | |
3942 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
3943 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
7489594c | 3944 | } |
78a478d0 | 3945 | } |
78a478d0 | 3946 | |
c7c4b3b6 | 3947 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 3948 | { |
86911732 HX |
3949 | skb_gro_reset_offset(skb); |
3950 | ||
89c5fa33 | 3951 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
3952 | } |
3953 | EXPORT_SYMBOL(napi_gro_receive); | |
3954 | ||
d0c2b0d2 | 3955 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 3956 | { |
96e93eab | 3957 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
3958 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
3959 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 3960 | skb->vlan_tci = 0; |
66c46d74 | 3961 | skb->dev = napi->dev; |
6d152e23 | 3962 | skb->skb_iif = 0; |
96e93eab HX |
3963 | |
3964 | napi->skb = skb; | |
3965 | } | |
96e93eab | 3966 | |
76620aaf | 3967 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 3968 | { |
5d38a079 | 3969 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
3970 | |
3971 | if (!skb) { | |
89d71a66 ED |
3972 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3973 | if (skb) | |
3974 | napi->skb = skb; | |
80595d59 | 3975 | } |
96e93eab HX |
3976 | return skb; |
3977 | } | |
76620aaf | 3978 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3979 | |
bb728820 | 3980 | static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
c7c4b3b6 | 3981 | gro_result_t ret) |
96e93eab | 3982 | { |
5d0d9be8 HX |
3983 | switch (ret) { |
3984 | case GRO_NORMAL: | |
86911732 | 3985 | case GRO_HELD: |
e76b69cc | 3986 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3987 | |
c7c4b3b6 BH |
3988 | if (ret == GRO_HELD) |
3989 | skb_gro_pull(skb, -ETH_HLEN); | |
3990 | else if (netif_receive_skb(skb)) | |
3991 | ret = GRO_DROP; | |
86911732 | 3992 | break; |
5d38a079 | 3993 | |
5d0d9be8 | 3994 | case GRO_DROP: |
5d0d9be8 HX |
3995 | case GRO_MERGED_FREE: |
3996 | napi_reuse_skb(napi, skb); | |
3997 | break; | |
5b252f0c BH |
3998 | |
3999 | case GRO_MERGED: | |
4000 | break; | |
5d0d9be8 | 4001 | } |
5d38a079 | 4002 | |
c7c4b3b6 | 4003 | return ret; |
5d38a079 | 4004 | } |
5d0d9be8 | 4005 | |
4adb9c4a | 4006 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4007 | { |
4008 | struct sk_buff *skb = napi->skb; | |
4009 | struct ethhdr *eth; | |
a5b1cf28 HX |
4010 | unsigned int hlen; |
4011 | unsigned int off; | |
76620aaf HX |
4012 | |
4013 | napi->skb = NULL; | |
4014 | ||
4015 | skb_reset_mac_header(skb); | |
4016 | skb_gro_reset_offset(skb); | |
4017 | ||
a5b1cf28 HX |
4018 | off = skb_gro_offset(skb); |
4019 | hlen = off + sizeof(*eth); | |
4020 | eth = skb_gro_header_fast(skb, off); | |
4021 | if (skb_gro_header_hard(skb, hlen)) { | |
4022 | eth = skb_gro_header_slow(skb, hlen, off); | |
4023 | if (unlikely(!eth)) { | |
4024 | napi_reuse_skb(napi, skb); | |
4025 | skb = NULL; | |
4026 | goto out; | |
4027 | } | |
76620aaf HX |
4028 | } |
4029 | ||
4030 | skb_gro_pull(skb, sizeof(*eth)); | |
4031 | ||
4032 | /* | |
4033 | * This works because the only protocols we care about don't require | |
4034 | * special handling. We'll fix it up properly at the end. | |
4035 | */ | |
4036 | skb->protocol = eth->h_proto; | |
4037 | ||
4038 | out: | |
4039 | return skb; | |
4040 | } | |
76620aaf | 4041 | |
c7c4b3b6 | 4042 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4043 | { |
76620aaf | 4044 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4045 | |
4046 | if (!skb) | |
c7c4b3b6 | 4047 | return GRO_DROP; |
5d0d9be8 | 4048 | |
89c5fa33 | 4049 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4050 | } |
5d38a079 HX |
4051 | EXPORT_SYMBOL(napi_gro_frags); |
4052 | ||
e326bed2 ED |
4053 | /* |
4054 | * net_rps_action sends any pending IPI's for rps. | |
4055 | * Note: called with local irq disabled, but exits with local irq enabled. | |
4056 | */ | |
4057 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4058 | { | |
4059 | #ifdef CONFIG_RPS | |
4060 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4061 | ||
4062 | if (remsd) { | |
4063 | sd->rps_ipi_list = NULL; | |
4064 | ||
4065 | local_irq_enable(); | |
4066 | ||
4067 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
4068 | while (remsd) { | |
4069 | struct softnet_data *next = remsd->rps_ipi_next; | |
4070 | ||
4071 | if (cpu_online(remsd->cpu)) | |
4072 | __smp_call_function_single(remsd->cpu, | |
4073 | &remsd->csd, 0); | |
4074 | remsd = next; | |
4075 | } | |
4076 | } else | |
4077 | #endif | |
4078 | local_irq_enable(); | |
4079 | } | |
4080 | ||
bea3348e | 4081 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
4082 | { |
4083 | int work = 0; | |
eecfd7c4 | 4084 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 4085 | |
e326bed2 ED |
4086 | #ifdef CONFIG_RPS |
4087 | /* Check if we have pending ipi, its better to send them now, | |
4088 | * not waiting net_rx_action() end. | |
4089 | */ | |
4090 | if (sd->rps_ipi_list) { | |
4091 | local_irq_disable(); | |
4092 | net_rps_action_and_irq_enable(sd); | |
4093 | } | |
4094 | #endif | |
bea3348e | 4095 | napi->weight = weight_p; |
6e7676c1 CG |
4096 | local_irq_disable(); |
4097 | while (work < quota) { | |
1da177e4 | 4098 | struct sk_buff *skb; |
6e7676c1 CG |
4099 | unsigned int qlen; |
4100 | ||
4101 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
4102 | local_irq_enable(); | |
4103 | __netif_receive_skb(skb); | |
6e7676c1 | 4104 | local_irq_disable(); |
76cc8b13 TH |
4105 | input_queue_head_incr(sd); |
4106 | if (++work >= quota) { | |
4107 | local_irq_enable(); | |
4108 | return work; | |
4109 | } | |
6e7676c1 | 4110 | } |
1da177e4 | 4111 | |
e36fa2f7 | 4112 | rps_lock(sd); |
6e7676c1 | 4113 | qlen = skb_queue_len(&sd->input_pkt_queue); |
76cc8b13 | 4114 | if (qlen) |
6e7676c1 CG |
4115 | skb_queue_splice_tail_init(&sd->input_pkt_queue, |
4116 | &sd->process_queue); | |
76cc8b13 | 4117 | |
6e7676c1 | 4118 | if (qlen < quota - work) { |
eecfd7c4 ED |
4119 | /* |
4120 | * Inline a custom version of __napi_complete(). | |
4121 | * only current cpu owns and manipulates this napi, | |
4122 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
4123 | * we can use a plain write instead of clear_bit(), | |
4124 | * and we dont need an smp_mb() memory barrier. | |
4125 | */ | |
4126 | list_del(&napi->poll_list); | |
4127 | napi->state = 0; | |
4128 | ||
6e7676c1 | 4129 | quota = work + qlen; |
bea3348e | 4130 | } |
e36fa2f7 | 4131 | rps_unlock(sd); |
6e7676c1 CG |
4132 | } |
4133 | local_irq_enable(); | |
1da177e4 | 4134 | |
bea3348e SH |
4135 | return work; |
4136 | } | |
1da177e4 | 4137 | |
bea3348e SH |
4138 | /** |
4139 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4140 | * @n: entry to schedule |
bea3348e SH |
4141 | * |
4142 | * The entry's receive function will be scheduled to run | |
4143 | */ | |
b5606c2d | 4144 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4145 | { |
4146 | unsigned long flags; | |
1da177e4 | 4147 | |
bea3348e | 4148 | local_irq_save(flags); |
eecfd7c4 | 4149 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 4150 | local_irq_restore(flags); |
1da177e4 | 4151 | } |
bea3348e SH |
4152 | EXPORT_SYMBOL(__napi_schedule); |
4153 | ||
d565b0a1 HX |
4154 | void __napi_complete(struct napi_struct *n) |
4155 | { | |
4156 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
4157 | BUG_ON(n->gro_list); | |
4158 | ||
4159 | list_del(&n->poll_list); | |
4160 | smp_mb__before_clear_bit(); | |
4161 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
4162 | } | |
4163 | EXPORT_SYMBOL(__napi_complete); | |
4164 | ||
4165 | void napi_complete(struct napi_struct *n) | |
4166 | { | |
4167 | unsigned long flags; | |
4168 | ||
4169 | /* | |
4170 | * don't let napi dequeue from the cpu poll list | |
4171 | * just in case its running on a different cpu | |
4172 | */ | |
4173 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4174 | return; | |
4175 | ||
2e71a6f8 | 4176 | napi_gro_flush(n, false); |
d565b0a1 HX |
4177 | local_irq_save(flags); |
4178 | __napi_complete(n); | |
4179 | local_irq_restore(flags); | |
4180 | } | |
4181 | EXPORT_SYMBOL(napi_complete); | |
4182 | ||
af12fa6e ET |
4183 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
4184 | struct napi_struct *napi_by_id(unsigned int napi_id) | |
4185 | { | |
4186 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
4187 | struct napi_struct *napi; | |
4188 | ||
4189 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
4190 | if (napi->napi_id == napi_id) | |
4191 | return napi; | |
4192 | ||
4193 | return NULL; | |
4194 | } | |
4195 | EXPORT_SYMBOL_GPL(napi_by_id); | |
4196 | ||
4197 | void napi_hash_add(struct napi_struct *napi) | |
4198 | { | |
4199 | if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) { | |
4200 | ||
4201 | spin_lock(&napi_hash_lock); | |
4202 | ||
4203 | /* 0 is not a valid id, we also skip an id that is taken | |
4204 | * we expect both events to be extremely rare | |
4205 | */ | |
4206 | napi->napi_id = 0; | |
4207 | while (!napi->napi_id) { | |
4208 | napi->napi_id = ++napi_gen_id; | |
4209 | if (napi_by_id(napi->napi_id)) | |
4210 | napi->napi_id = 0; | |
4211 | } | |
4212 | ||
4213 | hlist_add_head_rcu(&napi->napi_hash_node, | |
4214 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
4215 | ||
4216 | spin_unlock(&napi_hash_lock); | |
4217 | } | |
4218 | } | |
4219 | EXPORT_SYMBOL_GPL(napi_hash_add); | |
4220 | ||
4221 | /* Warning : caller is responsible to make sure rcu grace period | |
4222 | * is respected before freeing memory containing @napi | |
4223 | */ | |
4224 | void napi_hash_del(struct napi_struct *napi) | |
4225 | { | |
4226 | spin_lock(&napi_hash_lock); | |
4227 | ||
4228 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) | |
4229 | hlist_del_rcu(&napi->napi_hash_node); | |
4230 | ||
4231 | spin_unlock(&napi_hash_lock); | |
4232 | } | |
4233 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
4234 | ||
d565b0a1 HX |
4235 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
4236 | int (*poll)(struct napi_struct *, int), int weight) | |
4237 | { | |
4238 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 4239 | napi->gro_count = 0; |
d565b0a1 | 4240 | napi->gro_list = NULL; |
5d38a079 | 4241 | napi->skb = NULL; |
d565b0a1 | 4242 | napi->poll = poll; |
82dc3c63 ED |
4243 | if (weight > NAPI_POLL_WEIGHT) |
4244 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
4245 | weight, dev->name); | |
d565b0a1 HX |
4246 | napi->weight = weight; |
4247 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 4248 | napi->dev = dev; |
5d38a079 | 4249 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
4250 | spin_lock_init(&napi->poll_lock); |
4251 | napi->poll_owner = -1; | |
4252 | #endif | |
4253 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
4254 | } | |
4255 | EXPORT_SYMBOL(netif_napi_add); | |
4256 | ||
4257 | void netif_napi_del(struct napi_struct *napi) | |
4258 | { | |
4259 | struct sk_buff *skb, *next; | |
4260 | ||
d7b06636 | 4261 | list_del_init(&napi->dev_list); |
76620aaf | 4262 | napi_free_frags(napi); |
d565b0a1 HX |
4263 | |
4264 | for (skb = napi->gro_list; skb; skb = next) { | |
4265 | next = skb->next; | |
4266 | skb->next = NULL; | |
4267 | kfree_skb(skb); | |
4268 | } | |
4269 | ||
4270 | napi->gro_list = NULL; | |
4ae5544f | 4271 | napi->gro_count = 0; |
d565b0a1 HX |
4272 | } |
4273 | EXPORT_SYMBOL(netif_napi_del); | |
4274 | ||
1da177e4 LT |
4275 | static void net_rx_action(struct softirq_action *h) |
4276 | { | |
e326bed2 | 4277 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 4278 | unsigned long time_limit = jiffies + 2; |
51b0bded | 4279 | int budget = netdev_budget; |
53fb95d3 MM |
4280 | void *have; |
4281 | ||
1da177e4 LT |
4282 | local_irq_disable(); |
4283 | ||
e326bed2 | 4284 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
4285 | struct napi_struct *n; |
4286 | int work, weight; | |
1da177e4 | 4287 | |
bea3348e | 4288 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
4289 | * Allow this to run for 2 jiffies since which will allow |
4290 | * an average latency of 1.5/HZ. | |
bea3348e | 4291 | */ |
d1f41b67 | 4292 | if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit))) |
1da177e4 LT |
4293 | goto softnet_break; |
4294 | ||
4295 | local_irq_enable(); | |
4296 | ||
bea3348e SH |
4297 | /* Even though interrupts have been re-enabled, this |
4298 | * access is safe because interrupts can only add new | |
4299 | * entries to the tail of this list, and only ->poll() | |
4300 | * calls can remove this head entry from the list. | |
4301 | */ | |
e326bed2 | 4302 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 4303 | |
bea3348e SH |
4304 | have = netpoll_poll_lock(n); |
4305 | ||
4306 | weight = n->weight; | |
4307 | ||
0a7606c1 DM |
4308 | /* This NAPI_STATE_SCHED test is for avoiding a race |
4309 | * with netpoll's poll_napi(). Only the entity which | |
4310 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
4311 | * actually make the ->poll() call. Therefore we avoid | |
25985edc | 4312 | * accidentally calling ->poll() when NAPI is not scheduled. |
0a7606c1 DM |
4313 | */ |
4314 | work = 0; | |
4ea7e386 | 4315 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 4316 | work = n->poll(n, weight); |
4ea7e386 NH |
4317 | trace_napi_poll(n); |
4318 | } | |
bea3348e SH |
4319 | |
4320 | WARN_ON_ONCE(work > weight); | |
4321 | ||
4322 | budget -= work; | |
4323 | ||
4324 | local_irq_disable(); | |
4325 | ||
4326 | /* Drivers must not modify the NAPI state if they | |
4327 | * consume the entire weight. In such cases this code | |
4328 | * still "owns" the NAPI instance and therefore can | |
4329 | * move the instance around on the list at-will. | |
4330 | */ | |
fed17f30 | 4331 | if (unlikely(work == weight)) { |
ff780cd8 HX |
4332 | if (unlikely(napi_disable_pending(n))) { |
4333 | local_irq_enable(); | |
4334 | napi_complete(n); | |
4335 | local_irq_disable(); | |
2e71a6f8 ED |
4336 | } else { |
4337 | if (n->gro_list) { | |
4338 | /* flush too old packets | |
4339 | * If HZ < 1000, flush all packets. | |
4340 | */ | |
4341 | local_irq_enable(); | |
4342 | napi_gro_flush(n, HZ >= 1000); | |
4343 | local_irq_disable(); | |
4344 | } | |
e326bed2 | 4345 | list_move_tail(&n->poll_list, &sd->poll_list); |
2e71a6f8 | 4346 | } |
fed17f30 | 4347 | } |
bea3348e SH |
4348 | |
4349 | netpoll_poll_unlock(have); | |
1da177e4 LT |
4350 | } |
4351 | out: | |
e326bed2 | 4352 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 4353 | |
db217334 CL |
4354 | #ifdef CONFIG_NET_DMA |
4355 | /* | |
4356 | * There may not be any more sk_buffs coming right now, so push | |
4357 | * any pending DMA copies to hardware | |
4358 | */ | |
2ba05622 | 4359 | dma_issue_pending_all(); |
db217334 | 4360 | #endif |
bea3348e | 4361 | |
1da177e4 LT |
4362 | return; |
4363 | ||
4364 | softnet_break: | |
dee42870 | 4365 | sd->time_squeeze++; |
1da177e4 LT |
4366 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
4367 | goto out; | |
4368 | } | |
4369 | ||
aa9d8560 | 4370 | struct netdev_adjacent { |
9ff162a8 | 4371 | struct net_device *dev; |
5d261913 VF |
4372 | |
4373 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 4374 | bool master; |
5d261913 | 4375 | |
5d261913 VF |
4376 | /* counter for the number of times this device was added to us */ |
4377 | u16 ref_nr; | |
4378 | ||
402dae96 VF |
4379 | /* private field for the users */ |
4380 | void *private; | |
4381 | ||
9ff162a8 JP |
4382 | struct list_head list; |
4383 | struct rcu_head rcu; | |
9ff162a8 JP |
4384 | }; |
4385 | ||
5249dec7 VF |
4386 | static struct netdev_adjacent *__netdev_find_adj_rcu(struct net_device *dev, |
4387 | struct net_device *adj_dev, | |
4388 | struct list_head *adj_list) | |
4389 | { | |
4390 | struct netdev_adjacent *adj; | |
4391 | ||
4392 | list_for_each_entry_rcu(adj, adj_list, list) { | |
4393 | if (adj->dev == adj_dev) | |
4394 | return adj; | |
4395 | } | |
4396 | return NULL; | |
4397 | } | |
4398 | ||
5d261913 VF |
4399 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev, |
4400 | struct net_device *adj_dev, | |
2f268f12 | 4401 | struct list_head *adj_list) |
9ff162a8 | 4402 | { |
5d261913 | 4403 | struct netdev_adjacent *adj; |
5d261913 | 4404 | |
2f268f12 | 4405 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
4406 | if (adj->dev == adj_dev) |
4407 | return adj; | |
9ff162a8 JP |
4408 | } |
4409 | return NULL; | |
4410 | } | |
4411 | ||
4412 | /** | |
4413 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
4414 | * @dev: device | |
4415 | * @upper_dev: upper device to check | |
4416 | * | |
4417 | * Find out if a device is linked to specified upper device and return true | |
4418 | * in case it is. Note that this checks only immediate upper device, | |
4419 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
4420 | */ | |
4421 | bool netdev_has_upper_dev(struct net_device *dev, | |
4422 | struct net_device *upper_dev) | |
4423 | { | |
4424 | ASSERT_RTNL(); | |
4425 | ||
2f268f12 | 4426 | return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper); |
9ff162a8 JP |
4427 | } |
4428 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
4429 | ||
4430 | /** | |
4431 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
4432 | * @dev: device | |
4433 | * | |
4434 | * Find out if a device is linked to an upper device and return true in case | |
4435 | * it is. The caller must hold the RTNL lock. | |
4436 | */ | |
4437 | bool netdev_has_any_upper_dev(struct net_device *dev) | |
4438 | { | |
4439 | ASSERT_RTNL(); | |
4440 | ||
2f268f12 | 4441 | return !list_empty(&dev->all_adj_list.upper); |
9ff162a8 JP |
4442 | } |
4443 | EXPORT_SYMBOL(netdev_has_any_upper_dev); | |
4444 | ||
4445 | /** | |
4446 | * netdev_master_upper_dev_get - Get master upper device | |
4447 | * @dev: device | |
4448 | * | |
4449 | * Find a master upper device and return pointer to it or NULL in case | |
4450 | * it's not there. The caller must hold the RTNL lock. | |
4451 | */ | |
4452 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
4453 | { | |
aa9d8560 | 4454 | struct netdev_adjacent *upper; |
9ff162a8 JP |
4455 | |
4456 | ASSERT_RTNL(); | |
4457 | ||
2f268f12 | 4458 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
4459 | return NULL; |
4460 | ||
2f268f12 | 4461 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 4462 | struct netdev_adjacent, list); |
9ff162a8 JP |
4463 | if (likely(upper->master)) |
4464 | return upper->dev; | |
4465 | return NULL; | |
4466 | } | |
4467 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
4468 | ||
31088a11 VF |
4469 | /** |
4470 | * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list | |
48311f46 VF |
4471 | * @dev: device |
4472 | * @iter: list_head ** of the current position | |
4473 | * | |
4474 | * Gets the next device from the dev's upper list, starting from iter | |
4475 | * position. The caller must hold RCU read lock. | |
4476 | */ | |
2f268f12 VF |
4477 | struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, |
4478 | struct list_head **iter) | |
48311f46 VF |
4479 | { |
4480 | struct netdev_adjacent *upper; | |
4481 | ||
4482 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4483 | ||
4484 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4485 | ||
2f268f12 | 4486 | if (&upper->list == &dev->all_adj_list.upper) |
48311f46 VF |
4487 | return NULL; |
4488 | ||
4489 | *iter = &upper->list; | |
4490 | ||
4491 | return upper->dev; | |
4492 | } | |
2f268f12 | 4493 | EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); |
48311f46 | 4494 | |
31088a11 VF |
4495 | /** |
4496 | * netdev_lower_get_next_private - Get the next ->private from the | |
4497 | * lower neighbour list | |
4498 | * @dev: device | |
4499 | * @iter: list_head ** of the current position | |
4500 | * | |
4501 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
4502 | * list, starting from iter position. The caller must hold either hold the | |
4503 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
4504 | * list will remain unchainged. | |
4505 | */ | |
4506 | void *netdev_lower_get_next_private(struct net_device *dev, | |
4507 | struct list_head **iter) | |
4508 | { | |
4509 | struct netdev_adjacent *lower; | |
4510 | ||
4511 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
4512 | ||
4513 | if (&lower->list == &dev->adj_list.lower) | |
4514 | return NULL; | |
4515 | ||
4516 | if (iter) | |
4517 | *iter = lower->list.next; | |
4518 | ||
4519 | return lower->private; | |
4520 | } | |
4521 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
4522 | ||
4523 | /** | |
4524 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
4525 | * lower neighbour list, RCU | |
4526 | * variant | |
4527 | * @dev: device | |
4528 | * @iter: list_head ** of the current position | |
4529 | * | |
4530 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
4531 | * list, starting from iter position. The caller must hold RCU read lock. | |
4532 | */ | |
4533 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
4534 | struct list_head **iter) | |
4535 | { | |
4536 | struct netdev_adjacent *lower; | |
4537 | ||
4538 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4539 | ||
4540 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4541 | ||
4542 | if (&lower->list == &dev->adj_list.lower) | |
4543 | return NULL; | |
4544 | ||
4545 | if (iter) | |
4546 | *iter = &lower->list; | |
4547 | ||
4548 | return lower->private; | |
4549 | } | |
4550 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
4551 | ||
9ff162a8 JP |
4552 | /** |
4553 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
4554 | * @dev: device | |
4555 | * | |
4556 | * Find a master upper device and return pointer to it or NULL in case | |
4557 | * it's not there. The caller must hold the RCU read lock. | |
4558 | */ | |
4559 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
4560 | { | |
aa9d8560 | 4561 | struct netdev_adjacent *upper; |
9ff162a8 | 4562 | |
2f268f12 | 4563 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 4564 | struct netdev_adjacent, list); |
9ff162a8 JP |
4565 | if (upper && likely(upper->master)) |
4566 | return upper->dev; | |
4567 | return NULL; | |
4568 | } | |
4569 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
4570 | ||
5d261913 VF |
4571 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
4572 | struct net_device *adj_dev, | |
7863c054 | 4573 | struct list_head *dev_list, |
402dae96 | 4574 | void *private, bool master) |
5d261913 VF |
4575 | { |
4576 | struct netdev_adjacent *adj; | |
4577 | ||
7863c054 | 4578 | adj = __netdev_find_adj(dev, adj_dev, dev_list); |
5d261913 VF |
4579 | |
4580 | if (adj) { | |
5d261913 VF |
4581 | adj->ref_nr++; |
4582 | return 0; | |
4583 | } | |
4584 | ||
4585 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
4586 | if (!adj) | |
4587 | return -ENOMEM; | |
4588 | ||
4589 | adj->dev = adj_dev; | |
4590 | adj->master = master; | |
5d261913 | 4591 | adj->ref_nr = 1; |
402dae96 | 4592 | adj->private = private; |
5d261913 | 4593 | dev_hold(adj_dev); |
2f268f12 VF |
4594 | |
4595 | pr_debug("dev_hold for %s, because of link added from %s to %s\n", | |
4596 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 | 4597 | |
7863c054 | 4598 | /* Ensure that master link is always the first item in list. */ |
5d261913 | 4599 | if (master) |
7863c054 | 4600 | list_add_rcu(&adj->list, dev_list); |
5d261913 | 4601 | else |
7863c054 | 4602 | list_add_tail_rcu(&adj->list, dev_list); |
5d261913 VF |
4603 | |
4604 | return 0; | |
4605 | } | |
4606 | ||
5d261913 | 4607 | void __netdev_adjacent_dev_remove(struct net_device *dev, |
7863c054 VF |
4608 | struct net_device *adj_dev, |
4609 | struct list_head *dev_list) | |
5d261913 VF |
4610 | { |
4611 | struct netdev_adjacent *adj; | |
4612 | ||
7863c054 | 4613 | adj = __netdev_find_adj(dev, adj_dev, dev_list); |
5d261913 | 4614 | |
2f268f12 VF |
4615 | if (!adj) { |
4616 | pr_err("tried to remove device %s from %s\n", | |
4617 | dev->name, adj_dev->name); | |
5d261913 | 4618 | BUG(); |
2f268f12 | 4619 | } |
5d261913 VF |
4620 | |
4621 | if (adj->ref_nr > 1) { | |
2f268f12 VF |
4622 | pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name, |
4623 | adj->ref_nr-1); | |
5d261913 VF |
4624 | adj->ref_nr--; |
4625 | return; | |
4626 | } | |
4627 | ||
4628 | list_del_rcu(&adj->list); | |
2f268f12 VF |
4629 | pr_debug("dev_put for %s, because link removed from %s to %s\n", |
4630 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 VF |
4631 | dev_put(adj_dev); |
4632 | kfree_rcu(adj, rcu); | |
4633 | } | |
4634 | ||
2f268f12 VF |
4635 | int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
4636 | struct net_device *upper_dev, | |
4637 | struct list_head *up_list, | |
4638 | struct list_head *down_list, | |
402dae96 | 4639 | void *private, bool master) |
5d261913 VF |
4640 | { |
4641 | int ret; | |
4642 | ||
402dae96 VF |
4643 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private, |
4644 | master); | |
5d261913 VF |
4645 | if (ret) |
4646 | return ret; | |
4647 | ||
402dae96 VF |
4648 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private, |
4649 | false); | |
5d261913 | 4650 | if (ret) { |
2f268f12 | 4651 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5d261913 VF |
4652 | return ret; |
4653 | } | |
4654 | ||
4655 | return 0; | |
4656 | } | |
4657 | ||
2f268f12 VF |
4658 | int __netdev_adjacent_dev_link(struct net_device *dev, |
4659 | struct net_device *upper_dev) | |
5d261913 | 4660 | { |
2f268f12 VF |
4661 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
4662 | &dev->all_adj_list.upper, | |
4663 | &upper_dev->all_adj_list.lower, | |
402dae96 | 4664 | NULL, false); |
5d261913 VF |
4665 | } |
4666 | ||
2f268f12 VF |
4667 | void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
4668 | struct net_device *upper_dev, | |
4669 | struct list_head *up_list, | |
4670 | struct list_head *down_list) | |
5d261913 | 4671 | { |
2f268f12 VF |
4672 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
4673 | __netdev_adjacent_dev_remove(upper_dev, dev, down_list); | |
5d261913 VF |
4674 | } |
4675 | ||
4676 | void __netdev_adjacent_dev_unlink(struct net_device *dev, | |
4677 | struct net_device *upper_dev) | |
4678 | { | |
2f268f12 VF |
4679 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, |
4680 | &dev->all_adj_list.upper, | |
4681 | &upper_dev->all_adj_list.lower); | |
4682 | } | |
4683 | ||
4684 | int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, | |
4685 | struct net_device *upper_dev, | |
402dae96 | 4686 | void *private, bool master) |
2f268f12 VF |
4687 | { |
4688 | int ret = __netdev_adjacent_dev_link(dev, upper_dev); | |
4689 | ||
4690 | if (ret) | |
4691 | return ret; | |
4692 | ||
4693 | ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, | |
4694 | &dev->adj_list.upper, | |
4695 | &upper_dev->adj_list.lower, | |
402dae96 | 4696 | private, master); |
2f268f12 VF |
4697 | if (ret) { |
4698 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
4699 | return ret; | |
4700 | } | |
4701 | ||
4702 | return 0; | |
5d261913 VF |
4703 | } |
4704 | ||
2f268f12 VF |
4705 | void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
4706 | struct net_device *upper_dev) | |
4707 | { | |
4708 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
4709 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, | |
4710 | &dev->adj_list.upper, | |
4711 | &upper_dev->adj_list.lower); | |
4712 | } | |
5d261913 | 4713 | |
9ff162a8 | 4714 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 VF |
4715 | struct net_device *upper_dev, bool master, |
4716 | void *private) | |
9ff162a8 | 4717 | { |
5d261913 VF |
4718 | struct netdev_adjacent *i, *j, *to_i, *to_j; |
4719 | int ret = 0; | |
9ff162a8 JP |
4720 | |
4721 | ASSERT_RTNL(); | |
4722 | ||
4723 | if (dev == upper_dev) | |
4724 | return -EBUSY; | |
4725 | ||
4726 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
2f268f12 | 4727 | if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper)) |
9ff162a8 JP |
4728 | return -EBUSY; |
4729 | ||
2f268f12 | 4730 | if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper)) |
9ff162a8 JP |
4731 | return -EEXIST; |
4732 | ||
4733 | if (master && netdev_master_upper_dev_get(dev)) | |
4734 | return -EBUSY; | |
4735 | ||
402dae96 VF |
4736 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private, |
4737 | master); | |
5d261913 VF |
4738 | if (ret) |
4739 | return ret; | |
9ff162a8 | 4740 | |
5d261913 | 4741 | /* Now that we linked these devs, make all the upper_dev's |
2f268f12 | 4742 | * all_adj_list.upper visible to every dev's all_adj_list.lower an |
5d261913 VF |
4743 | * versa, and don't forget the devices itself. All of these |
4744 | * links are non-neighbours. | |
4745 | */ | |
2f268f12 VF |
4746 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
4747 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
4748 | pr_debug("Interlinking %s with %s, non-neighbour\n", | |
4749 | i->dev->name, j->dev->name); | |
5d261913 VF |
4750 | ret = __netdev_adjacent_dev_link(i->dev, j->dev); |
4751 | if (ret) | |
4752 | goto rollback_mesh; | |
4753 | } | |
4754 | } | |
4755 | ||
4756 | /* add dev to every upper_dev's upper device */ | |
2f268f12 VF |
4757 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
4758 | pr_debug("linking %s's upper device %s with %s\n", | |
4759 | upper_dev->name, i->dev->name, dev->name); | |
5d261913 VF |
4760 | ret = __netdev_adjacent_dev_link(dev, i->dev); |
4761 | if (ret) | |
4762 | goto rollback_upper_mesh; | |
4763 | } | |
4764 | ||
4765 | /* add upper_dev to every dev's lower device */ | |
2f268f12 VF |
4766 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
4767 | pr_debug("linking %s's lower device %s with %s\n", dev->name, | |
4768 | i->dev->name, upper_dev->name); | |
5d261913 VF |
4769 | ret = __netdev_adjacent_dev_link(i->dev, upper_dev); |
4770 | if (ret) | |
4771 | goto rollback_lower_mesh; | |
4772 | } | |
9ff162a8 | 4773 | |
42e52bf9 | 4774 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 | 4775 | return 0; |
5d261913 VF |
4776 | |
4777 | rollback_lower_mesh: | |
4778 | to_i = i; | |
2f268f12 | 4779 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5d261913 VF |
4780 | if (i == to_i) |
4781 | break; | |
4782 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); | |
4783 | } | |
4784 | ||
4785 | i = NULL; | |
4786 | ||
4787 | rollback_upper_mesh: | |
4788 | to_i = i; | |
2f268f12 | 4789 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5d261913 VF |
4790 | if (i == to_i) |
4791 | break; | |
4792 | __netdev_adjacent_dev_unlink(dev, i->dev); | |
4793 | } | |
4794 | ||
4795 | i = j = NULL; | |
4796 | ||
4797 | rollback_mesh: | |
4798 | to_i = i; | |
4799 | to_j = j; | |
2f268f12 VF |
4800 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
4801 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5d261913 VF |
4802 | if (i == to_i && j == to_j) |
4803 | break; | |
4804 | __netdev_adjacent_dev_unlink(i->dev, j->dev); | |
4805 | } | |
4806 | if (i == to_i) | |
4807 | break; | |
4808 | } | |
4809 | ||
2f268f12 | 4810 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
4811 | |
4812 | return ret; | |
9ff162a8 JP |
4813 | } |
4814 | ||
4815 | /** | |
4816 | * netdev_upper_dev_link - Add a link to the upper device | |
4817 | * @dev: device | |
4818 | * @upper_dev: new upper device | |
4819 | * | |
4820 | * Adds a link to device which is upper to this one. The caller must hold | |
4821 | * the RTNL lock. On a failure a negative errno code is returned. | |
4822 | * On success the reference counts are adjusted and the function | |
4823 | * returns zero. | |
4824 | */ | |
4825 | int netdev_upper_dev_link(struct net_device *dev, | |
4826 | struct net_device *upper_dev) | |
4827 | { | |
402dae96 | 4828 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL); |
9ff162a8 JP |
4829 | } |
4830 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
4831 | ||
4832 | /** | |
4833 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
4834 | * @dev: device | |
4835 | * @upper_dev: new upper device | |
4836 | * | |
4837 | * Adds a link to device which is upper to this one. In this case, only | |
4838 | * one master upper device can be linked, although other non-master devices | |
4839 | * might be linked as well. The caller must hold the RTNL lock. | |
4840 | * On a failure a negative errno code is returned. On success the reference | |
4841 | * counts are adjusted and the function returns zero. | |
4842 | */ | |
4843 | int netdev_master_upper_dev_link(struct net_device *dev, | |
4844 | struct net_device *upper_dev) | |
4845 | { | |
402dae96 | 4846 | return __netdev_upper_dev_link(dev, upper_dev, true, NULL); |
9ff162a8 JP |
4847 | } |
4848 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
4849 | ||
402dae96 VF |
4850 | int netdev_master_upper_dev_link_private(struct net_device *dev, |
4851 | struct net_device *upper_dev, | |
4852 | void *private) | |
4853 | { | |
4854 | return __netdev_upper_dev_link(dev, upper_dev, true, private); | |
4855 | } | |
4856 | EXPORT_SYMBOL(netdev_master_upper_dev_link_private); | |
4857 | ||
9ff162a8 JP |
4858 | /** |
4859 | * netdev_upper_dev_unlink - Removes a link to upper device | |
4860 | * @dev: device | |
4861 | * @upper_dev: new upper device | |
4862 | * | |
4863 | * Removes a link to device which is upper to this one. The caller must hold | |
4864 | * the RTNL lock. | |
4865 | */ | |
4866 | void netdev_upper_dev_unlink(struct net_device *dev, | |
4867 | struct net_device *upper_dev) | |
4868 | { | |
5d261913 | 4869 | struct netdev_adjacent *i, *j; |
9ff162a8 JP |
4870 | ASSERT_RTNL(); |
4871 | ||
2f268f12 | 4872 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
4873 | |
4874 | /* Here is the tricky part. We must remove all dev's lower | |
4875 | * devices from all upper_dev's upper devices and vice | |
4876 | * versa, to maintain the graph relationship. | |
4877 | */ | |
2f268f12 VF |
4878 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
4879 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) | |
5d261913 VF |
4880 | __netdev_adjacent_dev_unlink(i->dev, j->dev); |
4881 | ||
4882 | /* remove also the devices itself from lower/upper device | |
4883 | * list | |
4884 | */ | |
2f268f12 | 4885 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5d261913 VF |
4886 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); |
4887 | ||
2f268f12 | 4888 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) |
5d261913 VF |
4889 | __netdev_adjacent_dev_unlink(dev, i->dev); |
4890 | ||
42e52bf9 | 4891 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 JP |
4892 | } |
4893 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
4894 | ||
402dae96 VF |
4895 | void *netdev_lower_dev_get_private_rcu(struct net_device *dev, |
4896 | struct net_device *lower_dev) | |
4897 | { | |
4898 | struct netdev_adjacent *lower; | |
4899 | ||
4900 | if (!lower_dev) | |
4901 | return NULL; | |
4902 | lower = __netdev_find_adj_rcu(dev, lower_dev, &dev->adj_list.lower); | |
4903 | if (!lower) | |
4904 | return NULL; | |
4905 | ||
4906 | return lower->private; | |
4907 | } | |
4908 | EXPORT_SYMBOL(netdev_lower_dev_get_private_rcu); | |
4909 | ||
4910 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
4911 | struct net_device *lower_dev) | |
4912 | { | |
4913 | struct netdev_adjacent *lower; | |
4914 | ||
4915 | if (!lower_dev) | |
4916 | return NULL; | |
4917 | lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower); | |
4918 | if (!lower) | |
4919 | return NULL; | |
4920 | ||
4921 | return lower->private; | |
4922 | } | |
4923 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
4924 | ||
b6c40d68 PM |
4925 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
4926 | { | |
d314774c SH |
4927 | const struct net_device_ops *ops = dev->netdev_ops; |
4928 | ||
4929 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
4930 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
4931 | } |
4932 | ||
dad9b335 | 4933 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 | 4934 | { |
b536db93 | 4935 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
4936 | kuid_t uid; |
4937 | kgid_t gid; | |
1da177e4 | 4938 | |
24023451 PM |
4939 | ASSERT_RTNL(); |
4940 | ||
dad9b335 WC |
4941 | dev->flags |= IFF_PROMISC; |
4942 | dev->promiscuity += inc; | |
4943 | if (dev->promiscuity == 0) { | |
4944 | /* | |
4945 | * Avoid overflow. | |
4946 | * If inc causes overflow, untouch promisc and return error. | |
4947 | */ | |
4948 | if (inc < 0) | |
4949 | dev->flags &= ~IFF_PROMISC; | |
4950 | else { | |
4951 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
4952 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
4953 | dev->name); | |
dad9b335 WC |
4954 | return -EOVERFLOW; |
4955 | } | |
4956 | } | |
52609c0b | 4957 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
4958 | pr_info("device %s %s promiscuous mode\n", |
4959 | dev->name, | |
4960 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
4961 | if (audit_enabled) { |
4962 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
4963 | audit_log(current->audit_context, GFP_ATOMIC, |
4964 | AUDIT_ANOM_PROMISCUOUS, | |
4965 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
4966 | dev->name, (dev->flags & IFF_PROMISC), | |
4967 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 4968 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
4969 | from_kuid(&init_user_ns, uid), |
4970 | from_kgid(&init_user_ns, gid), | |
7759db82 | 4971 | audit_get_sessionid(current)); |
8192b0c4 | 4972 | } |
24023451 | 4973 | |
b6c40d68 | 4974 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 4975 | } |
dad9b335 | 4976 | return 0; |
1da177e4 LT |
4977 | } |
4978 | ||
4417da66 PM |
4979 | /** |
4980 | * dev_set_promiscuity - update promiscuity count on a device | |
4981 | * @dev: device | |
4982 | * @inc: modifier | |
4983 | * | |
4984 | * Add or remove promiscuity from a device. While the count in the device | |
4985 | * remains above zero the interface remains promiscuous. Once it hits zero | |
4986 | * the device reverts back to normal filtering operation. A negative inc | |
4987 | * value is used to drop promiscuity on the device. | |
dad9b335 | 4988 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 4989 | */ |
dad9b335 | 4990 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 4991 | { |
b536db93 | 4992 | unsigned int old_flags = dev->flags; |
dad9b335 | 4993 | int err; |
4417da66 | 4994 | |
dad9b335 | 4995 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 4996 | if (err < 0) |
dad9b335 | 4997 | return err; |
4417da66 PM |
4998 | if (dev->flags != old_flags) |
4999 | dev_set_rx_mode(dev); | |
dad9b335 | 5000 | return err; |
4417da66 | 5001 | } |
d1b19dff | 5002 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 5003 | |
1da177e4 LT |
5004 | /** |
5005 | * dev_set_allmulti - update allmulti count on a device | |
5006 | * @dev: device | |
5007 | * @inc: modifier | |
5008 | * | |
5009 | * Add or remove reception of all multicast frames to a device. While the | |
5010 | * count in the device remains above zero the interface remains listening | |
5011 | * to all interfaces. Once it hits zero the device reverts back to normal | |
5012 | * filtering operation. A negative @inc value is used to drop the counter | |
5013 | * when releasing a resource needing all multicasts. | |
dad9b335 | 5014 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
5015 | */ |
5016 | ||
dad9b335 | 5017 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 | 5018 | { |
b536db93 | 5019 | unsigned int old_flags = dev->flags; |
1da177e4 | 5020 | |
24023451 PM |
5021 | ASSERT_RTNL(); |
5022 | ||
1da177e4 | 5023 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
5024 | dev->allmulti += inc; |
5025 | if (dev->allmulti == 0) { | |
5026 | /* | |
5027 | * Avoid overflow. | |
5028 | * If inc causes overflow, untouch allmulti and return error. | |
5029 | */ | |
5030 | if (inc < 0) | |
5031 | dev->flags &= ~IFF_ALLMULTI; | |
5032 | else { | |
5033 | dev->allmulti -= inc; | |
7b6cd1ce JP |
5034 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
5035 | dev->name); | |
dad9b335 WC |
5036 | return -EOVERFLOW; |
5037 | } | |
5038 | } | |
24023451 | 5039 | if (dev->flags ^ old_flags) { |
b6c40d68 | 5040 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 5041 | dev_set_rx_mode(dev); |
24023451 | 5042 | } |
dad9b335 | 5043 | return 0; |
4417da66 | 5044 | } |
d1b19dff | 5045 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
5046 | |
5047 | /* | |
5048 | * Upload unicast and multicast address lists to device and | |
5049 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 5050 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
5051 | * are present. |
5052 | */ | |
5053 | void __dev_set_rx_mode(struct net_device *dev) | |
5054 | { | |
d314774c SH |
5055 | const struct net_device_ops *ops = dev->netdev_ops; |
5056 | ||
4417da66 PM |
5057 | /* dev_open will call this function so the list will stay sane. */ |
5058 | if (!(dev->flags&IFF_UP)) | |
5059 | return; | |
5060 | ||
5061 | if (!netif_device_present(dev)) | |
40b77c94 | 5062 | return; |
4417da66 | 5063 | |
01789349 | 5064 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
5065 | /* Unicast addresses changes may only happen under the rtnl, |
5066 | * therefore calling __dev_set_promiscuity here is safe. | |
5067 | */ | |
32e7bfc4 | 5068 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 | 5069 | __dev_set_promiscuity(dev, 1); |
2d348d1f | 5070 | dev->uc_promisc = true; |
32e7bfc4 | 5071 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 | 5072 | __dev_set_promiscuity(dev, -1); |
2d348d1f | 5073 | dev->uc_promisc = false; |
4417da66 | 5074 | } |
4417da66 | 5075 | } |
01789349 JP |
5076 | |
5077 | if (ops->ndo_set_rx_mode) | |
5078 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
5079 | } |
5080 | ||
5081 | void dev_set_rx_mode(struct net_device *dev) | |
5082 | { | |
b9e40857 | 5083 | netif_addr_lock_bh(dev); |
4417da66 | 5084 | __dev_set_rx_mode(dev); |
b9e40857 | 5085 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
5086 | } |
5087 | ||
f0db275a SH |
5088 | /** |
5089 | * dev_get_flags - get flags reported to userspace | |
5090 | * @dev: device | |
5091 | * | |
5092 | * Get the combination of flag bits exported through APIs to userspace. | |
5093 | */ | |
95c96174 | 5094 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 5095 | { |
95c96174 | 5096 | unsigned int flags; |
1da177e4 LT |
5097 | |
5098 | flags = (dev->flags & ~(IFF_PROMISC | | |
5099 | IFF_ALLMULTI | | |
b00055aa SR |
5100 | IFF_RUNNING | |
5101 | IFF_LOWER_UP | | |
5102 | IFF_DORMANT)) | | |
1da177e4 LT |
5103 | (dev->gflags & (IFF_PROMISC | |
5104 | IFF_ALLMULTI)); | |
5105 | ||
b00055aa SR |
5106 | if (netif_running(dev)) { |
5107 | if (netif_oper_up(dev)) | |
5108 | flags |= IFF_RUNNING; | |
5109 | if (netif_carrier_ok(dev)) | |
5110 | flags |= IFF_LOWER_UP; | |
5111 | if (netif_dormant(dev)) | |
5112 | flags |= IFF_DORMANT; | |
5113 | } | |
1da177e4 LT |
5114 | |
5115 | return flags; | |
5116 | } | |
d1b19dff | 5117 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 5118 | |
bd380811 | 5119 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 5120 | { |
b536db93 | 5121 | unsigned int old_flags = dev->flags; |
bd380811 | 5122 | int ret; |
1da177e4 | 5123 | |
24023451 PM |
5124 | ASSERT_RTNL(); |
5125 | ||
1da177e4 LT |
5126 | /* |
5127 | * Set the flags on our device. | |
5128 | */ | |
5129 | ||
5130 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
5131 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
5132 | IFF_AUTOMEDIA)) | | |
5133 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
5134 | IFF_ALLMULTI)); | |
5135 | ||
5136 | /* | |
5137 | * Load in the correct multicast list now the flags have changed. | |
5138 | */ | |
5139 | ||
b6c40d68 PM |
5140 | if ((old_flags ^ flags) & IFF_MULTICAST) |
5141 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 5142 | |
4417da66 | 5143 | dev_set_rx_mode(dev); |
1da177e4 LT |
5144 | |
5145 | /* | |
5146 | * Have we downed the interface. We handle IFF_UP ourselves | |
5147 | * according to user attempts to set it, rather than blindly | |
5148 | * setting it. | |
5149 | */ | |
5150 | ||
5151 | ret = 0; | |
5152 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 5153 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
5154 | |
5155 | if (!ret) | |
4417da66 | 5156 | dev_set_rx_mode(dev); |
1da177e4 LT |
5157 | } |
5158 | ||
1da177e4 | 5159 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
5160 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
5161 | ||
1da177e4 LT |
5162 | dev->gflags ^= IFF_PROMISC; |
5163 | dev_set_promiscuity(dev, inc); | |
5164 | } | |
5165 | ||
5166 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
5167 | is important. Some (broken) drivers set IFF_PROMISC, when | |
5168 | IFF_ALLMULTI is requested not asking us and not reporting. | |
5169 | */ | |
5170 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
5171 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
5172 | ||
1da177e4 LT |
5173 | dev->gflags ^= IFF_ALLMULTI; |
5174 | dev_set_allmulti(dev, inc); | |
5175 | } | |
5176 | ||
bd380811 PM |
5177 | return ret; |
5178 | } | |
5179 | ||
5180 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
5181 | { | |
5182 | unsigned int changes = dev->flags ^ old_flags; | |
5183 | ||
5184 | if (changes & IFF_UP) { | |
5185 | if (dev->flags & IFF_UP) | |
5186 | call_netdevice_notifiers(NETDEV_UP, dev); | |
5187 | else | |
5188 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
5189 | } | |
5190 | ||
5191 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
5192 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
5193 | struct netdev_notifier_change_info change_info; | |
5194 | ||
5195 | change_info.flags_changed = changes; | |
5196 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
5197 | &change_info.info); | |
5198 | } | |
bd380811 PM |
5199 | } |
5200 | ||
5201 | /** | |
5202 | * dev_change_flags - change device settings | |
5203 | * @dev: device | |
5204 | * @flags: device state flags | |
5205 | * | |
5206 | * Change settings on device based state flags. The flags are | |
5207 | * in the userspace exported format. | |
5208 | */ | |
b536db93 | 5209 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 5210 | { |
b536db93 ED |
5211 | int ret; |
5212 | unsigned int changes, old_flags = dev->flags; | |
bd380811 PM |
5213 | |
5214 | ret = __dev_change_flags(dev, flags); | |
5215 | if (ret < 0) | |
5216 | return ret; | |
5217 | ||
5218 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
5219 | if (changes) |
5220 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 5221 | |
bd380811 | 5222 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
5223 | return ret; |
5224 | } | |
d1b19dff | 5225 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 5226 | |
f0db275a SH |
5227 | /** |
5228 | * dev_set_mtu - Change maximum transfer unit | |
5229 | * @dev: device | |
5230 | * @new_mtu: new transfer unit | |
5231 | * | |
5232 | * Change the maximum transfer size of the network device. | |
5233 | */ | |
1da177e4 LT |
5234 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
5235 | { | |
d314774c | 5236 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
5237 | int err; |
5238 | ||
5239 | if (new_mtu == dev->mtu) | |
5240 | return 0; | |
5241 | ||
5242 | /* MTU must be positive. */ | |
5243 | if (new_mtu < 0) | |
5244 | return -EINVAL; | |
5245 | ||
5246 | if (!netif_device_present(dev)) | |
5247 | return -ENODEV; | |
5248 | ||
5249 | err = 0; | |
d314774c SH |
5250 | if (ops->ndo_change_mtu) |
5251 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
5252 | else |
5253 | dev->mtu = new_mtu; | |
d314774c | 5254 | |
e3d8fabe | 5255 | if (!err) |
056925ab | 5256 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
5257 | return err; |
5258 | } | |
d1b19dff | 5259 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 5260 | |
cbda10fa VD |
5261 | /** |
5262 | * dev_set_group - Change group this device belongs to | |
5263 | * @dev: device | |
5264 | * @new_group: group this device should belong to | |
5265 | */ | |
5266 | void dev_set_group(struct net_device *dev, int new_group) | |
5267 | { | |
5268 | dev->group = new_group; | |
5269 | } | |
5270 | EXPORT_SYMBOL(dev_set_group); | |
5271 | ||
f0db275a SH |
5272 | /** |
5273 | * dev_set_mac_address - Change Media Access Control Address | |
5274 | * @dev: device | |
5275 | * @sa: new address | |
5276 | * | |
5277 | * Change the hardware (MAC) address of the device | |
5278 | */ | |
1da177e4 LT |
5279 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
5280 | { | |
d314774c | 5281 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
5282 | int err; |
5283 | ||
d314774c | 5284 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
5285 | return -EOPNOTSUPP; |
5286 | if (sa->sa_family != dev->type) | |
5287 | return -EINVAL; | |
5288 | if (!netif_device_present(dev)) | |
5289 | return -ENODEV; | |
d314774c | 5290 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
5291 | if (err) |
5292 | return err; | |
fbdeca2d | 5293 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 5294 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 5295 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 5296 | return 0; |
1da177e4 | 5297 | } |
d1b19dff | 5298 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 5299 | |
4bf84c35 JP |
5300 | /** |
5301 | * dev_change_carrier - Change device carrier | |
5302 | * @dev: device | |
691b3b7e | 5303 | * @new_carrier: new value |
4bf84c35 JP |
5304 | * |
5305 | * Change device carrier | |
5306 | */ | |
5307 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
5308 | { | |
5309 | const struct net_device_ops *ops = dev->netdev_ops; | |
5310 | ||
5311 | if (!ops->ndo_change_carrier) | |
5312 | return -EOPNOTSUPP; | |
5313 | if (!netif_device_present(dev)) | |
5314 | return -ENODEV; | |
5315 | return ops->ndo_change_carrier(dev, new_carrier); | |
5316 | } | |
5317 | EXPORT_SYMBOL(dev_change_carrier); | |
5318 | ||
66b52b0d JP |
5319 | /** |
5320 | * dev_get_phys_port_id - Get device physical port ID | |
5321 | * @dev: device | |
5322 | * @ppid: port ID | |
5323 | * | |
5324 | * Get device physical port ID | |
5325 | */ | |
5326 | int dev_get_phys_port_id(struct net_device *dev, | |
5327 | struct netdev_phys_port_id *ppid) | |
5328 | { | |
5329 | const struct net_device_ops *ops = dev->netdev_ops; | |
5330 | ||
5331 | if (!ops->ndo_get_phys_port_id) | |
5332 | return -EOPNOTSUPP; | |
5333 | return ops->ndo_get_phys_port_id(dev, ppid); | |
5334 | } | |
5335 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
5336 | ||
1da177e4 LT |
5337 | /** |
5338 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5339 | * @net: the applicable net namespace |
1da177e4 LT |
5340 | * |
5341 | * Returns a suitable unique value for a new device interface | |
5342 | * number. The caller must hold the rtnl semaphore or the | |
5343 | * dev_base_lock to be sure it remains unique. | |
5344 | */ | |
881d966b | 5345 | static int dev_new_index(struct net *net) |
1da177e4 | 5346 | { |
aa79e66e | 5347 | int ifindex = net->ifindex; |
1da177e4 LT |
5348 | for (;;) { |
5349 | if (++ifindex <= 0) | |
5350 | ifindex = 1; | |
881d966b | 5351 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 5352 | return net->ifindex = ifindex; |
1da177e4 LT |
5353 | } |
5354 | } | |
5355 | ||
1da177e4 | 5356 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5357 | static LIST_HEAD(net_todo_list); |
1da177e4 | 5358 | |
6f05f629 | 5359 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5360 | { |
1da177e4 | 5361 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
5362 | } |
5363 | ||
9b5e383c | 5364 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5365 | { |
e93737b0 | 5366 | struct net_device *dev, *tmp; |
9b5e383c | 5367 | |
93ee31f1 DL |
5368 | BUG_ON(dev_boot_phase); |
5369 | ASSERT_RTNL(); | |
5370 | ||
e93737b0 | 5371 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5372 | /* Some devices call without registering |
e93737b0 KK |
5373 | * for initialization unwind. Remove those |
5374 | * devices and proceed with the remaining. | |
9b5e383c ED |
5375 | */ |
5376 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
5377 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
5378 | dev->name, dev); | |
93ee31f1 | 5379 | |
9b5e383c | 5380 | WARN_ON(1); |
e93737b0 KK |
5381 | list_del(&dev->unreg_list); |
5382 | continue; | |
9b5e383c | 5383 | } |
449f4544 | 5384 | dev->dismantle = true; |
9b5e383c | 5385 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5386 | } |
93ee31f1 | 5387 | |
44345724 OP |
5388 | /* If device is running, close it first. */ |
5389 | dev_close_many(head); | |
93ee31f1 | 5390 | |
44345724 | 5391 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5392 | /* And unlink it from device chain. */ |
5393 | unlist_netdevice(dev); | |
93ee31f1 | 5394 | |
9b5e383c ED |
5395 | dev->reg_state = NETREG_UNREGISTERING; |
5396 | } | |
93ee31f1 DL |
5397 | |
5398 | synchronize_net(); | |
5399 | ||
9b5e383c ED |
5400 | list_for_each_entry(dev, head, unreg_list) { |
5401 | /* Shutdown queueing discipline. */ | |
5402 | dev_shutdown(dev); | |
93ee31f1 DL |
5403 | |
5404 | ||
9b5e383c ED |
5405 | /* Notify protocols, that we are about to destroy |
5406 | this device. They should clean all the things. | |
5407 | */ | |
5408 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5409 | |
a2835763 PM |
5410 | if (!dev->rtnl_link_ops || |
5411 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5412 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
5413 | ||
9b5e383c ED |
5414 | /* |
5415 | * Flush the unicast and multicast chains | |
5416 | */ | |
a748ee24 | 5417 | dev_uc_flush(dev); |
22bedad3 | 5418 | dev_mc_flush(dev); |
93ee31f1 | 5419 | |
9b5e383c ED |
5420 | if (dev->netdev_ops->ndo_uninit) |
5421 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5422 | |
9ff162a8 JP |
5423 | /* Notifier chain MUST detach us all upper devices. */ |
5424 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 5425 | |
9b5e383c ED |
5426 | /* Remove entries from kobject tree */ |
5427 | netdev_unregister_kobject(dev); | |
024e9679 AD |
5428 | #ifdef CONFIG_XPS |
5429 | /* Remove XPS queueing entries */ | |
5430 | netif_reset_xps_queues_gt(dev, 0); | |
5431 | #endif | |
9b5e383c | 5432 | } |
93ee31f1 | 5433 | |
850a545b | 5434 | synchronize_net(); |
395264d5 | 5435 | |
a5ee1551 | 5436 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5437 | dev_put(dev); |
5438 | } | |
5439 | ||
5440 | static void rollback_registered(struct net_device *dev) | |
5441 | { | |
5442 | LIST_HEAD(single); | |
5443 | ||
5444 | list_add(&dev->unreg_list, &single); | |
5445 | rollback_registered_many(&single); | |
ceaaec98 | 5446 | list_del(&single); |
93ee31f1 DL |
5447 | } |
5448 | ||
c8f44aff MM |
5449 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
5450 | netdev_features_t features) | |
b63365a2 | 5451 | { |
57422dc5 MM |
5452 | /* Fix illegal checksum combinations */ |
5453 | if ((features & NETIF_F_HW_CSUM) && | |
5454 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5455 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
5456 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
5457 | } | |
5458 | ||
b63365a2 | 5459 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 5460 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 5461 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 5462 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
5463 | } |
5464 | ||
ec5f0615 PS |
5465 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
5466 | !(features & NETIF_F_IP_CSUM)) { | |
5467 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
5468 | features &= ~NETIF_F_TSO; | |
5469 | features &= ~NETIF_F_TSO_ECN; | |
5470 | } | |
5471 | ||
5472 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
5473 | !(features & NETIF_F_IPV6_CSUM)) { | |
5474 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
5475 | features &= ~NETIF_F_TSO6; | |
5476 | } | |
5477 | ||
31d8b9e0 BH |
5478 | /* TSO ECN requires that TSO is present as well. */ |
5479 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
5480 | features &= ~NETIF_F_TSO_ECN; | |
5481 | ||
212b573f MM |
5482 | /* Software GSO depends on SG. */ |
5483 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 5484 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
5485 | features &= ~NETIF_F_GSO; |
5486 | } | |
5487 | ||
acd1130e | 5488 | /* UFO needs SG and checksumming */ |
b63365a2 | 5489 | if (features & NETIF_F_UFO) { |
79032644 MM |
5490 | /* maybe split UFO into V4 and V6? */ |
5491 | if (!((features & NETIF_F_GEN_CSUM) || | |
5492 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
5493 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5494 | netdev_dbg(dev, |
acd1130e | 5495 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
5496 | features &= ~NETIF_F_UFO; |
5497 | } | |
5498 | ||
5499 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 5500 | netdev_dbg(dev, |
acd1130e | 5501 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
5502 | features &= ~NETIF_F_UFO; |
5503 | } | |
5504 | } | |
5505 | ||
5506 | return features; | |
5507 | } | |
b63365a2 | 5508 | |
6cb6a27c | 5509 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 5510 | { |
c8f44aff | 5511 | netdev_features_t features; |
5455c699 MM |
5512 | int err = 0; |
5513 | ||
87267485 MM |
5514 | ASSERT_RTNL(); |
5515 | ||
5455c699 MM |
5516 | features = netdev_get_wanted_features(dev); |
5517 | ||
5518 | if (dev->netdev_ops->ndo_fix_features) | |
5519 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
5520 | ||
5521 | /* driver might be less strict about feature dependencies */ | |
5522 | features = netdev_fix_features(dev, features); | |
5523 | ||
5524 | if (dev->features == features) | |
6cb6a27c | 5525 | return 0; |
5455c699 | 5526 | |
c8f44aff MM |
5527 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
5528 | &dev->features, &features); | |
5455c699 MM |
5529 | |
5530 | if (dev->netdev_ops->ndo_set_features) | |
5531 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5532 | ||
6cb6a27c | 5533 | if (unlikely(err < 0)) { |
5455c699 | 5534 | netdev_err(dev, |
c8f44aff MM |
5535 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
5536 | err, &features, &dev->features); | |
6cb6a27c MM |
5537 | return -1; |
5538 | } | |
5539 | ||
5540 | if (!err) | |
5541 | dev->features = features; | |
5542 | ||
5543 | return 1; | |
5544 | } | |
5545 | ||
afe12cc8 MM |
5546 | /** |
5547 | * netdev_update_features - recalculate device features | |
5548 | * @dev: the device to check | |
5549 | * | |
5550 | * Recalculate dev->features set and send notifications if it | |
5551 | * has changed. Should be called after driver or hardware dependent | |
5552 | * conditions might have changed that influence the features. | |
5553 | */ | |
6cb6a27c MM |
5554 | void netdev_update_features(struct net_device *dev) |
5555 | { | |
5556 | if (__netdev_update_features(dev)) | |
5557 | netdev_features_change(dev); | |
5455c699 MM |
5558 | } |
5559 | EXPORT_SYMBOL(netdev_update_features); | |
5560 | ||
afe12cc8 MM |
5561 | /** |
5562 | * netdev_change_features - recalculate device features | |
5563 | * @dev: the device to check | |
5564 | * | |
5565 | * Recalculate dev->features set and send notifications even | |
5566 | * if they have not changed. Should be called instead of | |
5567 | * netdev_update_features() if also dev->vlan_features might | |
5568 | * have changed to allow the changes to be propagated to stacked | |
5569 | * VLAN devices. | |
5570 | */ | |
5571 | void netdev_change_features(struct net_device *dev) | |
5572 | { | |
5573 | __netdev_update_features(dev); | |
5574 | netdev_features_change(dev); | |
5575 | } | |
5576 | EXPORT_SYMBOL(netdev_change_features); | |
5577 | ||
fc4a7489 PM |
5578 | /** |
5579 | * netif_stacked_transfer_operstate - transfer operstate | |
5580 | * @rootdev: the root or lower level device to transfer state from | |
5581 | * @dev: the device to transfer operstate to | |
5582 | * | |
5583 | * Transfer operational state from root to device. This is normally | |
5584 | * called when a stacking relationship exists between the root | |
5585 | * device and the device(a leaf device). | |
5586 | */ | |
5587 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5588 | struct net_device *dev) | |
5589 | { | |
5590 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5591 | netif_dormant_on(dev); | |
5592 | else | |
5593 | netif_dormant_off(dev); | |
5594 | ||
5595 | if (netif_carrier_ok(rootdev)) { | |
5596 | if (!netif_carrier_ok(dev)) | |
5597 | netif_carrier_on(dev); | |
5598 | } else { | |
5599 | if (netif_carrier_ok(dev)) | |
5600 | netif_carrier_off(dev); | |
5601 | } | |
5602 | } | |
5603 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5604 | ||
bf264145 | 5605 | #ifdef CONFIG_RPS |
1b4bf461 ED |
5606 | static int netif_alloc_rx_queues(struct net_device *dev) |
5607 | { | |
1b4bf461 | 5608 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 5609 | struct netdev_rx_queue *rx; |
1b4bf461 | 5610 | |
bd25fa7b | 5611 | BUG_ON(count < 1); |
1b4bf461 | 5612 | |
bd25fa7b | 5613 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
62b5942a | 5614 | if (!rx) |
bd25fa7b | 5615 | return -ENOMEM; |
62b5942a | 5616 | |
bd25fa7b TH |
5617 | dev->_rx = rx; |
5618 | ||
bd25fa7b | 5619 | for (i = 0; i < count; i++) |
fe822240 | 5620 | rx[i].dev = dev; |
1b4bf461 ED |
5621 | return 0; |
5622 | } | |
bf264145 | 5623 | #endif |
1b4bf461 | 5624 | |
aa942104 CG |
5625 | static void netdev_init_one_queue(struct net_device *dev, |
5626 | struct netdev_queue *queue, void *_unused) | |
5627 | { | |
5628 | /* Initialize queue lock */ | |
5629 | spin_lock_init(&queue->_xmit_lock); | |
5630 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
5631 | queue->xmit_lock_owner = -1; | |
b236da69 | 5632 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 5633 | queue->dev = dev; |
114cf580 TH |
5634 | #ifdef CONFIG_BQL |
5635 | dql_init(&queue->dql, HZ); | |
5636 | #endif | |
aa942104 CG |
5637 | } |
5638 | ||
60877a32 ED |
5639 | static void netif_free_tx_queues(struct net_device *dev) |
5640 | { | |
5641 | if (is_vmalloc_addr(dev->_tx)) | |
5642 | vfree(dev->_tx); | |
5643 | else | |
5644 | kfree(dev->_tx); | |
5645 | } | |
5646 | ||
e6484930 TH |
5647 | static int netif_alloc_netdev_queues(struct net_device *dev) |
5648 | { | |
5649 | unsigned int count = dev->num_tx_queues; | |
5650 | struct netdev_queue *tx; | |
60877a32 | 5651 | size_t sz = count * sizeof(*tx); |
e6484930 | 5652 | |
60877a32 | 5653 | BUG_ON(count < 1 || count > 0xffff); |
62b5942a | 5654 | |
60877a32 ED |
5655 | tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
5656 | if (!tx) { | |
5657 | tx = vzalloc(sz); | |
5658 | if (!tx) | |
5659 | return -ENOMEM; | |
5660 | } | |
e6484930 | 5661 | dev->_tx = tx; |
1d24eb48 | 5662 | |
e6484930 TH |
5663 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
5664 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
5665 | |
5666 | return 0; | |
e6484930 TH |
5667 | } |
5668 | ||
1da177e4 LT |
5669 | /** |
5670 | * register_netdevice - register a network device | |
5671 | * @dev: device to register | |
5672 | * | |
5673 | * Take a completed network device structure and add it to the kernel | |
5674 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5675 | * chain. 0 is returned on success. A negative errno code is returned | |
5676 | * on a failure to set up the device, or if the name is a duplicate. | |
5677 | * | |
5678 | * Callers must hold the rtnl semaphore. You may want | |
5679 | * register_netdev() instead of this. | |
5680 | * | |
5681 | * BUGS: | |
5682 | * The locking appears insufficient to guarantee two parallel registers | |
5683 | * will not get the same name. | |
5684 | */ | |
5685 | ||
5686 | int register_netdevice(struct net_device *dev) | |
5687 | { | |
1da177e4 | 5688 | int ret; |
d314774c | 5689 | struct net *net = dev_net(dev); |
1da177e4 LT |
5690 | |
5691 | BUG_ON(dev_boot_phase); | |
5692 | ASSERT_RTNL(); | |
5693 | ||
b17a7c17 SH |
5694 | might_sleep(); |
5695 | ||
1da177e4 LT |
5696 | /* When net_device's are persistent, this will be fatal. */ |
5697 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 5698 | BUG_ON(!net); |
1da177e4 | 5699 | |
f1f28aa3 | 5700 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 5701 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 5702 | |
1da177e4 LT |
5703 | dev->iflink = -1; |
5704 | ||
828de4f6 | 5705 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
5706 | if (ret < 0) |
5707 | goto out; | |
5708 | ||
1da177e4 | 5709 | /* Init, if this function is available */ |
d314774c SH |
5710 | if (dev->netdev_ops->ndo_init) { |
5711 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
5712 | if (ret) { |
5713 | if (ret > 0) | |
5714 | ret = -EIO; | |
90833aa4 | 5715 | goto out; |
1da177e4 LT |
5716 | } |
5717 | } | |
4ec93edb | 5718 | |
f646968f PM |
5719 | if (((dev->hw_features | dev->features) & |
5720 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
5721 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
5722 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
5723 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
5724 | ret = -EINVAL; | |
5725 | goto err_uninit; | |
5726 | } | |
5727 | ||
9c7dafbf PE |
5728 | ret = -EBUSY; |
5729 | if (!dev->ifindex) | |
5730 | dev->ifindex = dev_new_index(net); | |
5731 | else if (__dev_get_by_index(net, dev->ifindex)) | |
5732 | goto err_uninit; | |
5733 | ||
1da177e4 LT |
5734 | if (dev->iflink == -1) |
5735 | dev->iflink = dev->ifindex; | |
5736 | ||
5455c699 MM |
5737 | /* Transfer changeable features to wanted_features and enable |
5738 | * software offloads (GSO and GRO). | |
5739 | */ | |
5740 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
5741 | dev->features |= NETIF_F_SOFT_FEATURES; |
5742 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 5743 | |
c6e1a0d1 | 5744 | /* Turn on no cache copy if HW is doing checksum */ |
34324dc2 MM |
5745 | if (!(dev->flags & IFF_LOOPBACK)) { |
5746 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
5747 | if (dev->features & NETIF_F_ALL_CSUM) { | |
5748 | dev->wanted_features |= NETIF_F_NOCACHE_COPY; | |
5749 | dev->features |= NETIF_F_NOCACHE_COPY; | |
5750 | } | |
c6e1a0d1 TH |
5751 | } |
5752 | ||
1180e7d6 | 5753 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 5754 | */ |
1180e7d6 | 5755 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 5756 | |
ee579677 PS |
5757 | /* Make NETIF_F_SG inheritable to tunnel devices. |
5758 | */ | |
5759 | dev->hw_enc_features |= NETIF_F_SG; | |
5760 | ||
0d89d203 SH |
5761 | /* Make NETIF_F_SG inheritable to MPLS. |
5762 | */ | |
5763 | dev->mpls_features |= NETIF_F_SG; | |
5764 | ||
7ffbe3fd JB |
5765 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
5766 | ret = notifier_to_errno(ret); | |
5767 | if (ret) | |
5768 | goto err_uninit; | |
5769 | ||
8b41d188 | 5770 | ret = netdev_register_kobject(dev); |
b17a7c17 | 5771 | if (ret) |
7ce1b0ed | 5772 | goto err_uninit; |
b17a7c17 SH |
5773 | dev->reg_state = NETREG_REGISTERED; |
5774 | ||
6cb6a27c | 5775 | __netdev_update_features(dev); |
8e9b59b2 | 5776 | |
1da177e4 LT |
5777 | /* |
5778 | * Default initial state at registry is that the | |
5779 | * device is present. | |
5780 | */ | |
5781 | ||
5782 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5783 | ||
8f4cccbb BH |
5784 | linkwatch_init_dev(dev); |
5785 | ||
1da177e4 | 5786 | dev_init_scheduler(dev); |
1da177e4 | 5787 | dev_hold(dev); |
ce286d32 | 5788 | list_netdevice(dev); |
7bf23575 | 5789 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 5790 | |
948b337e JP |
5791 | /* If the device has permanent device address, driver should |
5792 | * set dev_addr and also addr_assign_type should be set to | |
5793 | * NET_ADDR_PERM (default value). | |
5794 | */ | |
5795 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
5796 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
5797 | ||
1da177e4 | 5798 | /* Notify protocols, that a new device appeared. */ |
056925ab | 5799 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 5800 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
5801 | if (ret) { |
5802 | rollback_registered(dev); | |
5803 | dev->reg_state = NETREG_UNREGISTERED; | |
5804 | } | |
d90a909e EB |
5805 | /* |
5806 | * Prevent userspace races by waiting until the network | |
5807 | * device is fully setup before sending notifications. | |
5808 | */ | |
a2835763 PM |
5809 | if (!dev->rtnl_link_ops || |
5810 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5811 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
5812 | |
5813 | out: | |
5814 | return ret; | |
7ce1b0ed HX |
5815 | |
5816 | err_uninit: | |
d314774c SH |
5817 | if (dev->netdev_ops->ndo_uninit) |
5818 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 5819 | goto out; |
1da177e4 | 5820 | } |
d1b19dff | 5821 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 5822 | |
937f1ba5 BH |
5823 | /** |
5824 | * init_dummy_netdev - init a dummy network device for NAPI | |
5825 | * @dev: device to init | |
5826 | * | |
5827 | * This takes a network device structure and initialize the minimum | |
5828 | * amount of fields so it can be used to schedule NAPI polls without | |
5829 | * registering a full blown interface. This is to be used by drivers | |
5830 | * that need to tie several hardware interfaces to a single NAPI | |
5831 | * poll scheduler due to HW limitations. | |
5832 | */ | |
5833 | int init_dummy_netdev(struct net_device *dev) | |
5834 | { | |
5835 | /* Clear everything. Note we don't initialize spinlocks | |
5836 | * are they aren't supposed to be taken by any of the | |
5837 | * NAPI code and this dummy netdev is supposed to be | |
5838 | * only ever used for NAPI polls | |
5839 | */ | |
5840 | memset(dev, 0, sizeof(struct net_device)); | |
5841 | ||
5842 | /* make sure we BUG if trying to hit standard | |
5843 | * register/unregister code path | |
5844 | */ | |
5845 | dev->reg_state = NETREG_DUMMY; | |
5846 | ||
937f1ba5 BH |
5847 | /* NAPI wants this */ |
5848 | INIT_LIST_HEAD(&dev->napi_list); | |
5849 | ||
5850 | /* a dummy interface is started by default */ | |
5851 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5852 | set_bit(__LINK_STATE_START, &dev->state); | |
5853 | ||
29b4433d ED |
5854 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
5855 | * because users of this 'device' dont need to change | |
5856 | * its refcount. | |
5857 | */ | |
5858 | ||
937f1ba5 BH |
5859 | return 0; |
5860 | } | |
5861 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5862 | ||
5863 | ||
1da177e4 LT |
5864 | /** |
5865 | * register_netdev - register a network device | |
5866 | * @dev: device to register | |
5867 | * | |
5868 | * Take a completed network device structure and add it to the kernel | |
5869 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5870 | * chain. 0 is returned on success. A negative errno code is returned | |
5871 | * on a failure to set up the device, or if the name is a duplicate. | |
5872 | * | |
38b4da38 | 5873 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5874 | * and expands the device name if you passed a format string to |
5875 | * alloc_netdev. | |
5876 | */ | |
5877 | int register_netdev(struct net_device *dev) | |
5878 | { | |
5879 | int err; | |
5880 | ||
5881 | rtnl_lock(); | |
1da177e4 | 5882 | err = register_netdevice(dev); |
1da177e4 LT |
5883 | rtnl_unlock(); |
5884 | return err; | |
5885 | } | |
5886 | EXPORT_SYMBOL(register_netdev); | |
5887 | ||
29b4433d ED |
5888 | int netdev_refcnt_read(const struct net_device *dev) |
5889 | { | |
5890 | int i, refcnt = 0; | |
5891 | ||
5892 | for_each_possible_cpu(i) | |
5893 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
5894 | return refcnt; | |
5895 | } | |
5896 | EXPORT_SYMBOL(netdev_refcnt_read); | |
5897 | ||
2c53040f | 5898 | /** |
1da177e4 | 5899 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 5900 | * @dev: target net_device |
1da177e4 LT |
5901 | * |
5902 | * This is called when unregistering network devices. | |
5903 | * | |
5904 | * Any protocol or device that holds a reference should register | |
5905 | * for netdevice notification, and cleanup and put back the | |
5906 | * reference if they receive an UNREGISTER event. | |
5907 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5908 | * call dev_put. |
1da177e4 LT |
5909 | */ |
5910 | static void netdev_wait_allrefs(struct net_device *dev) | |
5911 | { | |
5912 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 5913 | int refcnt; |
1da177e4 | 5914 | |
e014debe ED |
5915 | linkwatch_forget_dev(dev); |
5916 | ||
1da177e4 | 5917 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
5918 | refcnt = netdev_refcnt_read(dev); |
5919 | ||
5920 | while (refcnt != 0) { | |
1da177e4 | 5921 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 5922 | rtnl_lock(); |
1da177e4 LT |
5923 | |
5924 | /* Rebroadcast unregister notification */ | |
056925ab | 5925 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 5926 | |
748e2d93 | 5927 | __rtnl_unlock(); |
0115e8e3 | 5928 | rcu_barrier(); |
748e2d93 ED |
5929 | rtnl_lock(); |
5930 | ||
0115e8e3 | 5931 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
5932 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
5933 | &dev->state)) { | |
5934 | /* We must not have linkwatch events | |
5935 | * pending on unregister. If this | |
5936 | * happens, we simply run the queue | |
5937 | * unscheduled, resulting in a noop | |
5938 | * for this device. | |
5939 | */ | |
5940 | linkwatch_run_queue(); | |
5941 | } | |
5942 | ||
6756ae4b | 5943 | __rtnl_unlock(); |
1da177e4 LT |
5944 | |
5945 | rebroadcast_time = jiffies; | |
5946 | } | |
5947 | ||
5948 | msleep(250); | |
5949 | ||
29b4433d ED |
5950 | refcnt = netdev_refcnt_read(dev); |
5951 | ||
1da177e4 | 5952 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
5953 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
5954 | dev->name, refcnt); | |
1da177e4 LT |
5955 | warning_time = jiffies; |
5956 | } | |
5957 | } | |
5958 | } | |
5959 | ||
5960 | /* The sequence is: | |
5961 | * | |
5962 | * rtnl_lock(); | |
5963 | * ... | |
5964 | * register_netdevice(x1); | |
5965 | * register_netdevice(x2); | |
5966 | * ... | |
5967 | * unregister_netdevice(y1); | |
5968 | * unregister_netdevice(y2); | |
5969 | * ... | |
5970 | * rtnl_unlock(); | |
5971 | * free_netdev(y1); | |
5972 | * free_netdev(y2); | |
5973 | * | |
58ec3b4d | 5974 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5975 | * This allows us to deal with problems: |
b17a7c17 | 5976 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5977 | * without deadlocking with linkwatch via keventd. |
5978 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5979 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5980 | * |
5981 | * We must not return until all unregister events added during | |
5982 | * the interval the lock was held have been completed. | |
1da177e4 | 5983 | */ |
1da177e4 LT |
5984 | void netdev_run_todo(void) |
5985 | { | |
626ab0e6 | 5986 | struct list_head list; |
1da177e4 | 5987 | |
1da177e4 | 5988 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5989 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5990 | |
5991 | __rtnl_unlock(); | |
626ab0e6 | 5992 | |
0115e8e3 ED |
5993 | |
5994 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
5995 | if (!list_empty(&list)) |
5996 | rcu_barrier(); | |
5997 | ||
1da177e4 LT |
5998 | while (!list_empty(&list)) { |
5999 | struct net_device *dev | |
e5e26d75 | 6000 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
6001 | list_del(&dev->todo_list); |
6002 | ||
748e2d93 | 6003 | rtnl_lock(); |
0115e8e3 | 6004 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 6005 | __rtnl_unlock(); |
0115e8e3 | 6006 | |
b17a7c17 | 6007 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 6008 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
6009 | dev->name, dev->reg_state); |
6010 | dump_stack(); | |
6011 | continue; | |
6012 | } | |
1da177e4 | 6013 | |
b17a7c17 | 6014 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 6015 | |
152102c7 | 6016 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 6017 | |
b17a7c17 | 6018 | netdev_wait_allrefs(dev); |
1da177e4 | 6019 | |
b17a7c17 | 6020 | /* paranoia */ |
29b4433d | 6021 | BUG_ON(netdev_refcnt_read(dev)); |
33d480ce ED |
6022 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
6023 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 6024 | WARN_ON(dev->dn_ptr); |
1da177e4 | 6025 | |
b17a7c17 SH |
6026 | if (dev->destructor) |
6027 | dev->destructor(dev); | |
9093bbb2 SH |
6028 | |
6029 | /* Free network device */ | |
6030 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 6031 | } |
1da177e4 LT |
6032 | } |
6033 | ||
3cfde79c BH |
6034 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
6035 | * fields in the same order, with only the type differing. | |
6036 | */ | |
77a1abf5 ED |
6037 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
6038 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
6039 | { |
6040 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
6041 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
6042 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
6043 | #else |
6044 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
6045 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
6046 | u64 *dst = (u64 *)stats64; | |
6047 | ||
6048 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
6049 | sizeof(*stats64) / sizeof(u64)); | |
6050 | for (i = 0; i < n; i++) | |
6051 | dst[i] = src[i]; | |
6052 | #endif | |
6053 | } | |
77a1abf5 | 6054 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 6055 | |
eeda3fd6 SH |
6056 | /** |
6057 | * dev_get_stats - get network device statistics | |
6058 | * @dev: device to get statistics from | |
28172739 | 6059 | * @storage: place to store stats |
eeda3fd6 | 6060 | * |
d7753516 BH |
6061 | * Get network statistics from device. Return @storage. |
6062 | * The device driver may provide its own method by setting | |
6063 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
6064 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 6065 | */ |
d7753516 BH |
6066 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
6067 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 6068 | { |
eeda3fd6 SH |
6069 | const struct net_device_ops *ops = dev->netdev_ops; |
6070 | ||
28172739 ED |
6071 | if (ops->ndo_get_stats64) { |
6072 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
6073 | ops->ndo_get_stats64(dev, storage); |
6074 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 6075 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
6076 | } else { |
6077 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 6078 | } |
caf586e5 | 6079 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
28172739 | 6080 | return storage; |
c45d286e | 6081 | } |
eeda3fd6 | 6082 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 6083 | |
24824a09 | 6084 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 6085 | { |
24824a09 | 6086 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 6087 | |
24824a09 ED |
6088 | #ifdef CONFIG_NET_CLS_ACT |
6089 | if (queue) | |
6090 | return queue; | |
6091 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
6092 | if (!queue) | |
6093 | return NULL; | |
6094 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
6095 | queue->qdisc = &noop_qdisc; |
6096 | queue->qdisc_sleeping = &noop_qdisc; | |
6097 | rcu_assign_pointer(dev->ingress_queue, queue); | |
6098 | #endif | |
6099 | return queue; | |
bb949fbd DM |
6100 | } |
6101 | ||
2c60db03 ED |
6102 | static const struct ethtool_ops default_ethtool_ops; |
6103 | ||
d07d7507 SG |
6104 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
6105 | const struct ethtool_ops *ops) | |
6106 | { | |
6107 | if (dev->ethtool_ops == &default_ethtool_ops) | |
6108 | dev->ethtool_ops = ops; | |
6109 | } | |
6110 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
6111 | ||
1da177e4 | 6112 | /** |
36909ea4 | 6113 | * alloc_netdev_mqs - allocate network device |
1da177e4 LT |
6114 | * @sizeof_priv: size of private data to allocate space for |
6115 | * @name: device name format string | |
6116 | * @setup: callback to initialize device | |
36909ea4 TH |
6117 | * @txqs: the number of TX subqueues to allocate |
6118 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
6119 | * |
6120 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 | 6121 | * and performs basic initialization. Also allocates subquue structs |
36909ea4 | 6122 | * for each queue on the device. |
1da177e4 | 6123 | */ |
36909ea4 TH |
6124 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
6125 | void (*setup)(struct net_device *), | |
6126 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 6127 | { |
1da177e4 | 6128 | struct net_device *dev; |
7943986c | 6129 | size_t alloc_size; |
1ce8e7b5 | 6130 | struct net_device *p; |
1da177e4 | 6131 | |
b6fe17d6 SH |
6132 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
6133 | ||
36909ea4 | 6134 | if (txqs < 1) { |
7b6cd1ce | 6135 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
6136 | return NULL; |
6137 | } | |
6138 | ||
36909ea4 TH |
6139 | #ifdef CONFIG_RPS |
6140 | if (rxqs < 1) { | |
7b6cd1ce | 6141 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
6142 | return NULL; |
6143 | } | |
6144 | #endif | |
6145 | ||
fd2ea0a7 | 6146 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
6147 | if (sizeof_priv) { |
6148 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 6149 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
6150 | alloc_size += sizeof_priv; |
6151 | } | |
6152 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 6153 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 6154 | |
31380de9 | 6155 | p = kzalloc(alloc_size, GFP_KERNEL); |
62b5942a | 6156 | if (!p) |
1da177e4 | 6157 | return NULL; |
1da177e4 | 6158 | |
1ce8e7b5 | 6159 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 6160 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 6161 | |
29b4433d ED |
6162 | dev->pcpu_refcnt = alloc_percpu(int); |
6163 | if (!dev->pcpu_refcnt) | |
e6484930 | 6164 | goto free_p; |
ab9c73cc | 6165 | |
ab9c73cc | 6166 | if (dev_addr_init(dev)) |
29b4433d | 6167 | goto free_pcpu; |
ab9c73cc | 6168 | |
22bedad3 | 6169 | dev_mc_init(dev); |
a748ee24 | 6170 | dev_uc_init(dev); |
ccffad25 | 6171 | |
c346dca1 | 6172 | dev_net_set(dev, &init_net); |
1da177e4 | 6173 | |
8d3bdbd5 | 6174 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 6175 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 6176 | |
8d3bdbd5 DM |
6177 | INIT_LIST_HEAD(&dev->napi_list); |
6178 | INIT_LIST_HEAD(&dev->unreg_list); | |
6179 | INIT_LIST_HEAD(&dev->link_watch_list); | |
2f268f12 VF |
6180 | INIT_LIST_HEAD(&dev->adj_list.upper); |
6181 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
6182 | INIT_LIST_HEAD(&dev->all_adj_list.upper); | |
6183 | INIT_LIST_HEAD(&dev->all_adj_list.lower); | |
8d3bdbd5 DM |
6184 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
6185 | setup(dev); | |
6186 | ||
36909ea4 TH |
6187 | dev->num_tx_queues = txqs; |
6188 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 6189 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 6190 | goto free_all; |
e8a0464c | 6191 | |
df334545 | 6192 | #ifdef CONFIG_RPS |
36909ea4 TH |
6193 | dev->num_rx_queues = rxqs; |
6194 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 6195 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 6196 | goto free_all; |
df334545 | 6197 | #endif |
0a9627f2 | 6198 | |
1da177e4 | 6199 | strcpy(dev->name, name); |
cbda10fa | 6200 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
6201 | if (!dev->ethtool_ops) |
6202 | dev->ethtool_ops = &default_ethtool_ops; | |
1da177e4 | 6203 | return dev; |
ab9c73cc | 6204 | |
8d3bdbd5 DM |
6205 | free_all: |
6206 | free_netdev(dev); | |
6207 | return NULL; | |
6208 | ||
29b4433d ED |
6209 | free_pcpu: |
6210 | free_percpu(dev->pcpu_refcnt); | |
60877a32 | 6211 | netif_free_tx_queues(dev); |
fe822240 TH |
6212 | #ifdef CONFIG_RPS |
6213 | kfree(dev->_rx); | |
6214 | #endif | |
6215 | ||
ab9c73cc JP |
6216 | free_p: |
6217 | kfree(p); | |
6218 | return NULL; | |
1da177e4 | 6219 | } |
36909ea4 | 6220 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
6221 | |
6222 | /** | |
6223 | * free_netdev - free network device | |
6224 | * @dev: device | |
6225 | * | |
4ec93edb YH |
6226 | * This function does the last stage of destroying an allocated device |
6227 | * interface. The reference to the device object is released. | |
1da177e4 LT |
6228 | * If this is the last reference then it will be freed. |
6229 | */ | |
6230 | void free_netdev(struct net_device *dev) | |
6231 | { | |
d565b0a1 HX |
6232 | struct napi_struct *p, *n; |
6233 | ||
f3005d7f DL |
6234 | release_net(dev_net(dev)); |
6235 | ||
60877a32 | 6236 | netif_free_tx_queues(dev); |
fe822240 TH |
6237 | #ifdef CONFIG_RPS |
6238 | kfree(dev->_rx); | |
6239 | #endif | |
e8a0464c | 6240 | |
33d480ce | 6241 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 6242 | |
f001fde5 JP |
6243 | /* Flush device addresses */ |
6244 | dev_addr_flush(dev); | |
6245 | ||
d565b0a1 HX |
6246 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
6247 | netif_napi_del(p); | |
6248 | ||
29b4433d ED |
6249 | free_percpu(dev->pcpu_refcnt); |
6250 | dev->pcpu_refcnt = NULL; | |
6251 | ||
3041a069 | 6252 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
6253 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
6254 | kfree((char *)dev - dev->padded); | |
6255 | return; | |
6256 | } | |
6257 | ||
6258 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
6259 | dev->reg_state = NETREG_RELEASED; | |
6260 | ||
43cb76d9 GKH |
6261 | /* will free via device release */ |
6262 | put_device(&dev->dev); | |
1da177e4 | 6263 | } |
d1b19dff | 6264 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 6265 | |
f0db275a SH |
6266 | /** |
6267 | * synchronize_net - Synchronize with packet receive processing | |
6268 | * | |
6269 | * Wait for packets currently being received to be done. | |
6270 | * Does not block later packets from starting. | |
6271 | */ | |
4ec93edb | 6272 | void synchronize_net(void) |
1da177e4 LT |
6273 | { |
6274 | might_sleep(); | |
be3fc413 ED |
6275 | if (rtnl_is_locked()) |
6276 | synchronize_rcu_expedited(); | |
6277 | else | |
6278 | synchronize_rcu(); | |
1da177e4 | 6279 | } |
d1b19dff | 6280 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
6281 | |
6282 | /** | |
44a0873d | 6283 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 6284 | * @dev: device |
44a0873d | 6285 | * @head: list |
6ebfbc06 | 6286 | * |
1da177e4 | 6287 | * This function shuts down a device interface and removes it |
d59b54b1 | 6288 | * from the kernel tables. |
44a0873d | 6289 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
6290 | * |
6291 | * Callers must hold the rtnl semaphore. You may want | |
6292 | * unregister_netdev() instead of this. | |
6293 | */ | |
6294 | ||
44a0873d | 6295 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 6296 | { |
a6620712 HX |
6297 | ASSERT_RTNL(); |
6298 | ||
44a0873d | 6299 | if (head) { |
9fdce099 | 6300 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
6301 | } else { |
6302 | rollback_registered(dev); | |
6303 | /* Finish processing unregister after unlock */ | |
6304 | net_set_todo(dev); | |
6305 | } | |
1da177e4 | 6306 | } |
44a0873d | 6307 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 6308 | |
9b5e383c ED |
6309 | /** |
6310 | * unregister_netdevice_many - unregister many devices | |
6311 | * @head: list of devices | |
9b5e383c ED |
6312 | */ |
6313 | void unregister_netdevice_many(struct list_head *head) | |
6314 | { | |
6315 | struct net_device *dev; | |
6316 | ||
6317 | if (!list_empty(head)) { | |
6318 | rollback_registered_many(head); | |
6319 | list_for_each_entry(dev, head, unreg_list) | |
6320 | net_set_todo(dev); | |
6321 | } | |
6322 | } | |
63c8099d | 6323 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 6324 | |
1da177e4 LT |
6325 | /** |
6326 | * unregister_netdev - remove device from the kernel | |
6327 | * @dev: device | |
6328 | * | |
6329 | * This function shuts down a device interface and removes it | |
d59b54b1 | 6330 | * from the kernel tables. |
1da177e4 LT |
6331 | * |
6332 | * This is just a wrapper for unregister_netdevice that takes | |
6333 | * the rtnl semaphore. In general you want to use this and not | |
6334 | * unregister_netdevice. | |
6335 | */ | |
6336 | void unregister_netdev(struct net_device *dev) | |
6337 | { | |
6338 | rtnl_lock(); | |
6339 | unregister_netdevice(dev); | |
6340 | rtnl_unlock(); | |
6341 | } | |
1da177e4 LT |
6342 | EXPORT_SYMBOL(unregister_netdev); |
6343 | ||
ce286d32 EB |
6344 | /** |
6345 | * dev_change_net_namespace - move device to different nethost namespace | |
6346 | * @dev: device | |
6347 | * @net: network namespace | |
6348 | * @pat: If not NULL name pattern to try if the current device name | |
6349 | * is already taken in the destination network namespace. | |
6350 | * | |
6351 | * This function shuts down a device interface and moves it | |
6352 | * to a new network namespace. On success 0 is returned, on | |
6353 | * a failure a netagive errno code is returned. | |
6354 | * | |
6355 | * Callers must hold the rtnl semaphore. | |
6356 | */ | |
6357 | ||
6358 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
6359 | { | |
ce286d32 EB |
6360 | int err; |
6361 | ||
6362 | ASSERT_RTNL(); | |
6363 | ||
6364 | /* Don't allow namespace local devices to be moved. */ | |
6365 | err = -EINVAL; | |
6366 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6367 | goto out; | |
6368 | ||
6369 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
6370 | if (dev->reg_state != NETREG_REGISTERED) |
6371 | goto out; | |
6372 | ||
6373 | /* Get out if there is nothing todo */ | |
6374 | err = 0; | |
878628fb | 6375 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
6376 | goto out; |
6377 | ||
6378 | /* Pick the destination device name, and ensure | |
6379 | * we can use it in the destination network namespace. | |
6380 | */ | |
6381 | err = -EEXIST; | |
d9031024 | 6382 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
6383 | /* We get here if we can't use the current device name */ |
6384 | if (!pat) | |
6385 | goto out; | |
828de4f6 | 6386 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
6387 | goto out; |
6388 | } | |
6389 | ||
6390 | /* | |
6391 | * And now a mini version of register_netdevice unregister_netdevice. | |
6392 | */ | |
6393 | ||
6394 | /* If device is running close it first. */ | |
9b772652 | 6395 | dev_close(dev); |
ce286d32 EB |
6396 | |
6397 | /* And unlink it from device chain */ | |
6398 | err = -ENODEV; | |
6399 | unlist_netdevice(dev); | |
6400 | ||
6401 | synchronize_net(); | |
6402 | ||
6403 | /* Shutdown queueing discipline. */ | |
6404 | dev_shutdown(dev); | |
6405 | ||
6406 | /* Notify protocols, that we are about to destroy | |
6407 | this device. They should clean all the things. | |
3b27e105 DL |
6408 | |
6409 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
6410 | This is wanted because this way 8021q and macvlan know | |
6411 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
6412 | */ |
6413 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
6414 | rcu_barrier(); |
6415 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
d2237d35 | 6416 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); |
ce286d32 EB |
6417 | |
6418 | /* | |
6419 | * Flush the unicast and multicast chains | |
6420 | */ | |
a748ee24 | 6421 | dev_uc_flush(dev); |
22bedad3 | 6422 | dev_mc_flush(dev); |
ce286d32 | 6423 | |
4e66ae2e SH |
6424 | /* Send a netdev-removed uevent to the old namespace */ |
6425 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
6426 | ||
ce286d32 | 6427 | /* Actually switch the network namespace */ |
c346dca1 | 6428 | dev_net_set(dev, net); |
ce286d32 | 6429 | |
ce286d32 EB |
6430 | /* If there is an ifindex conflict assign a new one */ |
6431 | if (__dev_get_by_index(net, dev->ifindex)) { | |
6432 | int iflink = (dev->iflink == dev->ifindex); | |
6433 | dev->ifindex = dev_new_index(net); | |
6434 | if (iflink) | |
6435 | dev->iflink = dev->ifindex; | |
6436 | } | |
6437 | ||
4e66ae2e SH |
6438 | /* Send a netdev-add uevent to the new namespace */ |
6439 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
6440 | ||
8b41d188 | 6441 | /* Fixup kobjects */ |
a1b3f594 | 6442 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 6443 | WARN_ON(err); |
ce286d32 EB |
6444 | |
6445 | /* Add the device back in the hashes */ | |
6446 | list_netdevice(dev); | |
6447 | ||
6448 | /* Notify protocols, that a new device appeared. */ | |
6449 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
6450 | ||
d90a909e EB |
6451 | /* |
6452 | * Prevent userspace races by waiting until the network | |
6453 | * device is fully setup before sending notifications. | |
6454 | */ | |
6455 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
6456 | ||
ce286d32 EB |
6457 | synchronize_net(); |
6458 | err = 0; | |
6459 | out: | |
6460 | return err; | |
6461 | } | |
463d0183 | 6462 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 6463 | |
1da177e4 LT |
6464 | static int dev_cpu_callback(struct notifier_block *nfb, |
6465 | unsigned long action, | |
6466 | void *ocpu) | |
6467 | { | |
6468 | struct sk_buff **list_skb; | |
1da177e4 LT |
6469 | struct sk_buff *skb; |
6470 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
6471 | struct softnet_data *sd, *oldsd; | |
6472 | ||
8bb78442 | 6473 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
6474 | return NOTIFY_OK; |
6475 | ||
6476 | local_irq_disable(); | |
6477 | cpu = smp_processor_id(); | |
6478 | sd = &per_cpu(softnet_data, cpu); | |
6479 | oldsd = &per_cpu(softnet_data, oldcpu); | |
6480 | ||
6481 | /* Find end of our completion_queue. */ | |
6482 | list_skb = &sd->completion_queue; | |
6483 | while (*list_skb) | |
6484 | list_skb = &(*list_skb)->next; | |
6485 | /* Append completion queue from offline CPU. */ | |
6486 | *list_skb = oldsd->completion_queue; | |
6487 | oldsd->completion_queue = NULL; | |
6488 | ||
1da177e4 | 6489 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
6490 | if (oldsd->output_queue) { |
6491 | *sd->output_queue_tailp = oldsd->output_queue; | |
6492 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
6493 | oldsd->output_queue = NULL; | |
6494 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
6495 | } | |
264524d5 HC |
6496 | /* Append NAPI poll list from offline CPU. */ |
6497 | if (!list_empty(&oldsd->poll_list)) { | |
6498 | list_splice_init(&oldsd->poll_list, &sd->poll_list); | |
6499 | raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6500 | } | |
1da177e4 LT |
6501 | |
6502 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
6503 | local_irq_enable(); | |
6504 | ||
6505 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 6506 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
1da177e4 | 6507 | netif_rx(skb); |
76cc8b13 | 6508 | input_queue_head_incr(oldsd); |
fec5e652 | 6509 | } |
76cc8b13 | 6510 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
6e7676c1 | 6511 | netif_rx(skb); |
76cc8b13 TH |
6512 | input_queue_head_incr(oldsd); |
6513 | } | |
1da177e4 LT |
6514 | |
6515 | return NOTIFY_OK; | |
6516 | } | |
1da177e4 LT |
6517 | |
6518 | ||
7f353bf2 | 6519 | /** |
b63365a2 HX |
6520 | * netdev_increment_features - increment feature set by one |
6521 | * @all: current feature set | |
6522 | * @one: new feature set | |
6523 | * @mask: mask feature set | |
7f353bf2 HX |
6524 | * |
6525 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
6526 | * @one to the master device with current feature set @all. Will not |
6527 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 6528 | */ |
c8f44aff MM |
6529 | netdev_features_t netdev_increment_features(netdev_features_t all, |
6530 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 6531 | { |
1742f183 MM |
6532 | if (mask & NETIF_F_GEN_CSUM) |
6533 | mask |= NETIF_F_ALL_CSUM; | |
6534 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 6535 | |
1742f183 MM |
6536 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
6537 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 6538 | |
1742f183 MM |
6539 | /* If one device supports hw checksumming, set for all. */ |
6540 | if (all & NETIF_F_GEN_CSUM) | |
6541 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
6542 | |
6543 | return all; | |
6544 | } | |
b63365a2 | 6545 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 6546 | |
430f03cd | 6547 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
6548 | { |
6549 | int i; | |
6550 | struct hlist_head *hash; | |
6551 | ||
6552 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
6553 | if (hash != NULL) | |
6554 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
6555 | INIT_HLIST_HEAD(&hash[i]); | |
6556 | ||
6557 | return hash; | |
6558 | } | |
6559 | ||
881d966b | 6560 | /* Initialize per network namespace state */ |
4665079c | 6561 | static int __net_init netdev_init(struct net *net) |
881d966b | 6562 | { |
734b6541 RM |
6563 | if (net != &init_net) |
6564 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 6565 | |
30d97d35 PE |
6566 | net->dev_name_head = netdev_create_hash(); |
6567 | if (net->dev_name_head == NULL) | |
6568 | goto err_name; | |
881d966b | 6569 | |
30d97d35 PE |
6570 | net->dev_index_head = netdev_create_hash(); |
6571 | if (net->dev_index_head == NULL) | |
6572 | goto err_idx; | |
881d966b EB |
6573 | |
6574 | return 0; | |
30d97d35 PE |
6575 | |
6576 | err_idx: | |
6577 | kfree(net->dev_name_head); | |
6578 | err_name: | |
6579 | return -ENOMEM; | |
881d966b EB |
6580 | } |
6581 | ||
f0db275a SH |
6582 | /** |
6583 | * netdev_drivername - network driver for the device | |
6584 | * @dev: network device | |
f0db275a SH |
6585 | * |
6586 | * Determine network driver for device. | |
6587 | */ | |
3019de12 | 6588 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 6589 | { |
cf04a4c7 SH |
6590 | const struct device_driver *driver; |
6591 | const struct device *parent; | |
3019de12 | 6592 | const char *empty = ""; |
6579e57b AV |
6593 | |
6594 | parent = dev->dev.parent; | |
6579e57b | 6595 | if (!parent) |
3019de12 | 6596 | return empty; |
6579e57b AV |
6597 | |
6598 | driver = parent->driver; | |
6599 | if (driver && driver->name) | |
3019de12 DM |
6600 | return driver->name; |
6601 | return empty; | |
6579e57b AV |
6602 | } |
6603 | ||
b004ff49 | 6604 | static int __netdev_printk(const char *level, const struct net_device *dev, |
256df2f3 JP |
6605 | struct va_format *vaf) |
6606 | { | |
6607 | int r; | |
6608 | ||
b004ff49 | 6609 | if (dev && dev->dev.parent) { |
666f355f JP |
6610 | r = dev_printk_emit(level[1] - '0', |
6611 | dev->dev.parent, | |
6612 | "%s %s %s: %pV", | |
6613 | dev_driver_string(dev->dev.parent), | |
6614 | dev_name(dev->dev.parent), | |
6615 | netdev_name(dev), vaf); | |
b004ff49 | 6616 | } else if (dev) { |
256df2f3 | 6617 | r = printk("%s%s: %pV", level, netdev_name(dev), vaf); |
b004ff49 | 6618 | } else { |
256df2f3 | 6619 | r = printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 6620 | } |
256df2f3 JP |
6621 | |
6622 | return r; | |
6623 | } | |
6624 | ||
6625 | int netdev_printk(const char *level, const struct net_device *dev, | |
6626 | const char *format, ...) | |
6627 | { | |
6628 | struct va_format vaf; | |
6629 | va_list args; | |
6630 | int r; | |
6631 | ||
6632 | va_start(args, format); | |
6633 | ||
6634 | vaf.fmt = format; | |
6635 | vaf.va = &args; | |
6636 | ||
6637 | r = __netdev_printk(level, dev, &vaf); | |
b004ff49 | 6638 | |
256df2f3 JP |
6639 | va_end(args); |
6640 | ||
6641 | return r; | |
6642 | } | |
6643 | EXPORT_SYMBOL(netdev_printk); | |
6644 | ||
6645 | #define define_netdev_printk_level(func, level) \ | |
6646 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
6647 | { \ | |
6648 | int r; \ | |
6649 | struct va_format vaf; \ | |
6650 | va_list args; \ | |
6651 | \ | |
6652 | va_start(args, fmt); \ | |
6653 | \ | |
6654 | vaf.fmt = fmt; \ | |
6655 | vaf.va = &args; \ | |
6656 | \ | |
6657 | r = __netdev_printk(level, dev, &vaf); \ | |
b004ff49 | 6658 | \ |
256df2f3 JP |
6659 | va_end(args); \ |
6660 | \ | |
6661 | return r; \ | |
6662 | } \ | |
6663 | EXPORT_SYMBOL(func); | |
6664 | ||
6665 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
6666 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
6667 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
6668 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
6669 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
6670 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
6671 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
6672 | ||
4665079c | 6673 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
6674 | { |
6675 | kfree(net->dev_name_head); | |
6676 | kfree(net->dev_index_head); | |
6677 | } | |
6678 | ||
022cbae6 | 6679 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
6680 | .init = netdev_init, |
6681 | .exit = netdev_exit, | |
6682 | }; | |
6683 | ||
4665079c | 6684 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 6685 | { |
e008b5fc | 6686 | struct net_device *dev, *aux; |
ce286d32 | 6687 | /* |
e008b5fc | 6688 | * Push all migratable network devices back to the |
ce286d32 EB |
6689 | * initial network namespace |
6690 | */ | |
6691 | rtnl_lock(); | |
e008b5fc | 6692 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 6693 | int err; |
aca51397 | 6694 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
6695 | |
6696 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6697 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6698 | continue; | |
6699 | ||
e008b5fc EB |
6700 | /* Leave virtual devices for the generic cleanup */ |
6701 | if (dev->rtnl_link_ops) | |
6702 | continue; | |
d0c082ce | 6703 | |
25985edc | 6704 | /* Push remaining network devices to init_net */ |
aca51397 PE |
6705 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
6706 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 6707 | if (err) { |
7b6cd1ce JP |
6708 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
6709 | __func__, dev->name, err); | |
aca51397 | 6710 | BUG(); |
ce286d32 EB |
6711 | } |
6712 | } | |
6713 | rtnl_unlock(); | |
6714 | } | |
6715 | ||
04dc7f6b EB |
6716 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
6717 | { | |
6718 | /* At exit all network devices most be removed from a network | |
b595076a | 6719 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
6720 | * Do this across as many network namespaces as possible to |
6721 | * improve batching efficiency. | |
6722 | */ | |
6723 | struct net_device *dev; | |
6724 | struct net *net; | |
6725 | LIST_HEAD(dev_kill_list); | |
6726 | ||
6727 | rtnl_lock(); | |
6728 | list_for_each_entry(net, net_list, exit_list) { | |
6729 | for_each_netdev_reverse(net, dev) { | |
6730 | if (dev->rtnl_link_ops) | |
6731 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
6732 | else | |
6733 | unregister_netdevice_queue(dev, &dev_kill_list); | |
6734 | } | |
6735 | } | |
6736 | unregister_netdevice_many(&dev_kill_list); | |
ceaaec98 | 6737 | list_del(&dev_kill_list); |
04dc7f6b EB |
6738 | rtnl_unlock(); |
6739 | } | |
6740 | ||
022cbae6 | 6741 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 6742 | .exit = default_device_exit, |
04dc7f6b | 6743 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
6744 | }; |
6745 | ||
1da177e4 LT |
6746 | /* |
6747 | * Initialize the DEV module. At boot time this walks the device list and | |
6748 | * unhooks any devices that fail to initialise (normally hardware not | |
6749 | * present) and leaves us with a valid list of present and active devices. | |
6750 | * | |
6751 | */ | |
6752 | ||
6753 | /* | |
6754 | * This is called single threaded during boot, so no need | |
6755 | * to take the rtnl semaphore. | |
6756 | */ | |
6757 | static int __init net_dev_init(void) | |
6758 | { | |
6759 | int i, rc = -ENOMEM; | |
6760 | ||
6761 | BUG_ON(!dev_boot_phase); | |
6762 | ||
1da177e4 LT |
6763 | if (dev_proc_init()) |
6764 | goto out; | |
6765 | ||
8b41d188 | 6766 | if (netdev_kobject_init()) |
1da177e4 LT |
6767 | goto out; |
6768 | ||
6769 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 6770 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
6771 | INIT_LIST_HEAD(&ptype_base[i]); |
6772 | ||
62532da9 VY |
6773 | INIT_LIST_HEAD(&offload_base); |
6774 | ||
881d966b EB |
6775 | if (register_pernet_subsys(&netdev_net_ops)) |
6776 | goto out; | |
1da177e4 LT |
6777 | |
6778 | /* | |
6779 | * Initialise the packet receive queues. | |
6780 | */ | |
6781 | ||
6f912042 | 6782 | for_each_possible_cpu(i) { |
e36fa2f7 | 6783 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 6784 | |
dee42870 | 6785 | memset(sd, 0, sizeof(*sd)); |
e36fa2f7 | 6786 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 6787 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 ED |
6788 | sd->completion_queue = NULL; |
6789 | INIT_LIST_HEAD(&sd->poll_list); | |
a9cbd588 CG |
6790 | sd->output_queue = NULL; |
6791 | sd->output_queue_tailp = &sd->output_queue; | |
df334545 | 6792 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
6793 | sd->csd.func = rps_trigger_softirq; |
6794 | sd->csd.info = sd; | |
6795 | sd->csd.flags = 0; | |
6796 | sd->cpu = i; | |
1e94d72f | 6797 | #endif |
0a9627f2 | 6798 | |
e36fa2f7 ED |
6799 | sd->backlog.poll = process_backlog; |
6800 | sd->backlog.weight = weight_p; | |
6801 | sd->backlog.gro_list = NULL; | |
6802 | sd->backlog.gro_count = 0; | |
99bbc707 WB |
6803 | |
6804 | #ifdef CONFIG_NET_FLOW_LIMIT | |
6805 | sd->flow_limit = NULL; | |
6806 | #endif | |
1da177e4 LT |
6807 | } |
6808 | ||
1da177e4 LT |
6809 | dev_boot_phase = 0; |
6810 | ||
505d4f73 EB |
6811 | /* The loopback device is special if any other network devices |
6812 | * is present in a network namespace the loopback device must | |
6813 | * be present. Since we now dynamically allocate and free the | |
6814 | * loopback device ensure this invariant is maintained by | |
6815 | * keeping the loopback device as the first device on the | |
6816 | * list of network devices. Ensuring the loopback devices | |
6817 | * is the first device that appears and the last network device | |
6818 | * that disappears. | |
6819 | */ | |
6820 | if (register_pernet_device(&loopback_net_ops)) | |
6821 | goto out; | |
6822 | ||
6823 | if (register_pernet_device(&default_device_ops)) | |
6824 | goto out; | |
6825 | ||
962cf36c CM |
6826 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
6827 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
6828 | |
6829 | hotcpu_notifier(dev_cpu_callback, 0); | |
6830 | dst_init(); | |
1da177e4 LT |
6831 | rc = 0; |
6832 | out: | |
6833 | return rc; | |
6834 | } | |
6835 | ||
6836 | subsys_initcall(net_dev_init); |