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> |
25cd9ba0 | 121 | #include <net/mpls.h> |
8f0f2223 DM |
122 | #include <linux/ipv6.h> |
123 | #include <linux/in.h> | |
b6b2fed1 DM |
124 | #include <linux/jhash.h> |
125 | #include <linux/random.h> | |
9cbc1cb8 | 126 | #include <trace/events/napi.h> |
cf66ba58 | 127 | #include <trace/events/net.h> |
07dc22e7 | 128 | #include <trace/events/skb.h> |
5acbbd42 | 129 | #include <linux/pci.h> |
caeda9b9 | 130 | #include <linux/inetdevice.h> |
c445477d | 131 | #include <linux/cpu_rmap.h> |
c5905afb | 132 | #include <linux/static_key.h> |
af12fa6e | 133 | #include <linux/hashtable.h> |
60877a32 | 134 | #include <linux/vmalloc.h> |
529d0489 | 135 | #include <linux/if_macvlan.h> |
e7fd2885 | 136 | #include <linux/errqueue.h> |
3b47d303 | 137 | #include <linux/hrtimer.h> |
1da177e4 | 138 | |
342709ef PE |
139 | #include "net-sysfs.h" |
140 | ||
d565b0a1 HX |
141 | /* Instead of increasing this, you should create a hash table. */ |
142 | #define MAX_GRO_SKBS 8 | |
143 | ||
5d38a079 HX |
144 | /* This should be increased if a protocol with a bigger head is added. */ |
145 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
146 | ||
1da177e4 | 147 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 148 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
149 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
150 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 151 | static struct list_head offload_base __read_mostly; |
1da177e4 | 152 | |
ae78dbfa | 153 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
154 | static int call_netdevice_notifiers_info(unsigned long val, |
155 | struct net_device *dev, | |
156 | struct netdev_notifier_info *info); | |
ae78dbfa | 157 | |
1da177e4 | 158 | /* |
7562f876 | 159 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
160 | * semaphore. |
161 | * | |
c6d14c84 | 162 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
163 | * |
164 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 165 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
166 | * actual updates. This allows pure readers to access the list even |
167 | * while a writer is preparing to update it. | |
168 | * | |
169 | * To put it another way, dev_base_lock is held for writing only to | |
170 | * protect against pure readers; the rtnl semaphore provides the | |
171 | * protection against other writers. | |
172 | * | |
173 | * See, for example usages, register_netdevice() and | |
174 | * unregister_netdevice(), which must be called with the rtnl | |
175 | * semaphore held. | |
176 | */ | |
1da177e4 | 177 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
178 | EXPORT_SYMBOL(dev_base_lock); |
179 | ||
af12fa6e ET |
180 | /* protects napi_hash addition/deletion and napi_gen_id */ |
181 | static DEFINE_SPINLOCK(napi_hash_lock); | |
182 | ||
183 | static unsigned int napi_gen_id; | |
184 | static DEFINE_HASHTABLE(napi_hash, 8); | |
185 | ||
18afa4b0 | 186 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 187 | |
4e985ada TG |
188 | static inline void dev_base_seq_inc(struct net *net) |
189 | { | |
190 | while (++net->dev_base_seq == 0); | |
191 | } | |
192 | ||
881d966b | 193 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 194 | { |
95c96174 ED |
195 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
196 | ||
08e9897d | 197 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
198 | } |
199 | ||
881d966b | 200 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 201 | { |
7c28bd0b | 202 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
203 | } |
204 | ||
e36fa2f7 | 205 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
206 | { |
207 | #ifdef CONFIG_RPS | |
e36fa2f7 | 208 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
209 | #endif |
210 | } | |
211 | ||
e36fa2f7 | 212 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
213 | { |
214 | #ifdef CONFIG_RPS | |
e36fa2f7 | 215 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
216 | #endif |
217 | } | |
218 | ||
ce286d32 | 219 | /* Device list insertion */ |
53759be9 | 220 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 221 | { |
c346dca1 | 222 | struct net *net = dev_net(dev); |
ce286d32 EB |
223 | |
224 | ASSERT_RTNL(); | |
225 | ||
226 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 227 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 228 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
229 | hlist_add_head_rcu(&dev->index_hlist, |
230 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 231 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
232 | |
233 | dev_base_seq_inc(net); | |
ce286d32 EB |
234 | } |
235 | ||
fb699dfd ED |
236 | /* Device list removal |
237 | * caller must respect a RCU grace period before freeing/reusing dev | |
238 | */ | |
ce286d32 EB |
239 | static void unlist_netdevice(struct net_device *dev) |
240 | { | |
241 | ASSERT_RTNL(); | |
242 | ||
243 | /* Unlink dev from the device chain */ | |
244 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 245 | list_del_rcu(&dev->dev_list); |
72c9528b | 246 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 247 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 248 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
249 | |
250 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
251 | } |
252 | ||
1da177e4 LT |
253 | /* |
254 | * Our notifier list | |
255 | */ | |
256 | ||
f07d5b94 | 257 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
258 | |
259 | /* | |
260 | * Device drivers call our routines to queue packets here. We empty the | |
261 | * queue in the local softnet handler. | |
262 | */ | |
bea3348e | 263 | |
9958da05 | 264 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 265 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 266 | |
cf508b12 | 267 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 268 | /* |
c773e847 | 269 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
270 | * according to dev->type |
271 | */ | |
272 | static const unsigned short netdev_lock_type[] = | |
273 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
274 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
275 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
276 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
277 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
278 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
279 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
280 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
281 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
282 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
283 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
284 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
285 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
286 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
287 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 288 | |
36cbd3dc | 289 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
290 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
291 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
292 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
293 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
294 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
295 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
296 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
297 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
298 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
299 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
300 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
301 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
302 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
303 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
304 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
305 | |
306 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 307 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
308 | |
309 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
310 | { | |
311 | int i; | |
312 | ||
313 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
314 | if (netdev_lock_type[i] == dev_type) | |
315 | return i; | |
316 | /* the last key is used by default */ | |
317 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
318 | } | |
319 | ||
cf508b12 DM |
320 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
321 | unsigned short dev_type) | |
723e98b7 JP |
322 | { |
323 | int i; | |
324 | ||
325 | i = netdev_lock_pos(dev_type); | |
326 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
327 | netdev_lock_name[i]); | |
328 | } | |
cf508b12 DM |
329 | |
330 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
331 | { | |
332 | int i; | |
333 | ||
334 | i = netdev_lock_pos(dev->type); | |
335 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
336 | &netdev_addr_lock_key[i], | |
337 | netdev_lock_name[i]); | |
338 | } | |
723e98b7 | 339 | #else |
cf508b12 DM |
340 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
341 | unsigned short dev_type) | |
342 | { | |
343 | } | |
344 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
345 | { |
346 | } | |
347 | #endif | |
1da177e4 LT |
348 | |
349 | /******************************************************************************* | |
350 | ||
351 | Protocol management and registration routines | |
352 | ||
353 | *******************************************************************************/ | |
354 | ||
1da177e4 LT |
355 | /* |
356 | * Add a protocol ID to the list. Now that the input handler is | |
357 | * smarter we can dispense with all the messy stuff that used to be | |
358 | * here. | |
359 | * | |
360 | * BEWARE!!! Protocol handlers, mangling input packets, | |
361 | * MUST BE last in hash buckets and checking protocol handlers | |
362 | * MUST start from promiscuous ptype_all chain in net_bh. | |
363 | * It is true now, do not change it. | |
364 | * Explanation follows: if protocol handler, mangling packet, will | |
365 | * be the first on list, it is not able to sense, that packet | |
366 | * is cloned and should be copied-on-write, so that it will | |
367 | * change it and subsequent readers will get broken packet. | |
368 | * --ANK (980803) | |
369 | */ | |
370 | ||
c07b68e8 ED |
371 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
372 | { | |
373 | if (pt->type == htons(ETH_P_ALL)) | |
374 | return &ptype_all; | |
375 | else | |
376 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
377 | } | |
378 | ||
1da177e4 LT |
379 | /** |
380 | * dev_add_pack - add packet handler | |
381 | * @pt: packet type declaration | |
382 | * | |
383 | * Add a protocol handler to the networking stack. The passed &packet_type | |
384 | * is linked into kernel lists and may not be freed until it has been | |
385 | * removed from the kernel lists. | |
386 | * | |
4ec93edb | 387 | * This call does not sleep therefore it can not |
1da177e4 LT |
388 | * guarantee all CPU's that are in middle of receiving packets |
389 | * will see the new packet type (until the next received packet). | |
390 | */ | |
391 | ||
392 | void dev_add_pack(struct packet_type *pt) | |
393 | { | |
c07b68e8 | 394 | struct list_head *head = ptype_head(pt); |
1da177e4 | 395 | |
c07b68e8 ED |
396 | spin_lock(&ptype_lock); |
397 | list_add_rcu(&pt->list, head); | |
398 | spin_unlock(&ptype_lock); | |
1da177e4 | 399 | } |
d1b19dff | 400 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 401 | |
1da177e4 LT |
402 | /** |
403 | * __dev_remove_pack - remove packet handler | |
404 | * @pt: packet type declaration | |
405 | * | |
406 | * Remove a protocol handler that was previously added to the kernel | |
407 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
408 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 409 | * returns. |
1da177e4 LT |
410 | * |
411 | * The packet type might still be in use by receivers | |
412 | * and must not be freed until after all the CPU's have gone | |
413 | * through a quiescent state. | |
414 | */ | |
415 | void __dev_remove_pack(struct packet_type *pt) | |
416 | { | |
c07b68e8 | 417 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
418 | struct packet_type *pt1; |
419 | ||
c07b68e8 | 420 | spin_lock(&ptype_lock); |
1da177e4 LT |
421 | |
422 | list_for_each_entry(pt1, head, list) { | |
423 | if (pt == pt1) { | |
424 | list_del_rcu(&pt->list); | |
425 | goto out; | |
426 | } | |
427 | } | |
428 | ||
7b6cd1ce | 429 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 430 | out: |
c07b68e8 | 431 | spin_unlock(&ptype_lock); |
1da177e4 | 432 | } |
d1b19dff ED |
433 | EXPORT_SYMBOL(__dev_remove_pack); |
434 | ||
1da177e4 LT |
435 | /** |
436 | * dev_remove_pack - remove packet handler | |
437 | * @pt: packet type declaration | |
438 | * | |
439 | * Remove a protocol handler that was previously added to the kernel | |
440 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
441 | * from the kernel lists and can be freed or reused once this function | |
442 | * returns. | |
443 | * | |
444 | * This call sleeps to guarantee that no CPU is looking at the packet | |
445 | * type after return. | |
446 | */ | |
447 | void dev_remove_pack(struct packet_type *pt) | |
448 | { | |
449 | __dev_remove_pack(pt); | |
4ec93edb | 450 | |
1da177e4 LT |
451 | synchronize_net(); |
452 | } | |
d1b19dff | 453 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 454 | |
62532da9 VY |
455 | |
456 | /** | |
457 | * dev_add_offload - register offload handlers | |
458 | * @po: protocol offload declaration | |
459 | * | |
460 | * Add protocol offload handlers to the networking stack. The passed | |
461 | * &proto_offload is linked into kernel lists and may not be freed until | |
462 | * it has been removed from the kernel lists. | |
463 | * | |
464 | * This call does not sleep therefore it can not | |
465 | * guarantee all CPU's that are in middle of receiving packets | |
466 | * will see the new offload handlers (until the next received packet). | |
467 | */ | |
468 | void dev_add_offload(struct packet_offload *po) | |
469 | { | |
470 | struct list_head *head = &offload_base; | |
471 | ||
472 | spin_lock(&offload_lock); | |
473 | list_add_rcu(&po->list, head); | |
474 | spin_unlock(&offload_lock); | |
475 | } | |
476 | EXPORT_SYMBOL(dev_add_offload); | |
477 | ||
478 | /** | |
479 | * __dev_remove_offload - remove offload handler | |
480 | * @po: packet offload declaration | |
481 | * | |
482 | * Remove a protocol offload handler that was previously added to the | |
483 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
484 | * is removed from the kernel lists and can be freed or reused once this | |
485 | * function returns. | |
486 | * | |
487 | * The packet type might still be in use by receivers | |
488 | * and must not be freed until after all the CPU's have gone | |
489 | * through a quiescent state. | |
490 | */ | |
1d143d9f | 491 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
492 | { |
493 | struct list_head *head = &offload_base; | |
494 | struct packet_offload *po1; | |
495 | ||
c53aa505 | 496 | spin_lock(&offload_lock); |
62532da9 VY |
497 | |
498 | list_for_each_entry(po1, head, list) { | |
499 | if (po == po1) { | |
500 | list_del_rcu(&po->list); | |
501 | goto out; | |
502 | } | |
503 | } | |
504 | ||
505 | pr_warn("dev_remove_offload: %p not found\n", po); | |
506 | out: | |
c53aa505 | 507 | spin_unlock(&offload_lock); |
62532da9 | 508 | } |
62532da9 VY |
509 | |
510 | /** | |
511 | * dev_remove_offload - remove packet offload handler | |
512 | * @po: packet offload declaration | |
513 | * | |
514 | * Remove a packet offload handler that was previously added to the kernel | |
515 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
516 | * removed from the kernel lists and can be freed or reused once this | |
517 | * function returns. | |
518 | * | |
519 | * This call sleeps to guarantee that no CPU is looking at the packet | |
520 | * type after return. | |
521 | */ | |
522 | void dev_remove_offload(struct packet_offload *po) | |
523 | { | |
524 | __dev_remove_offload(po); | |
525 | ||
526 | synchronize_net(); | |
527 | } | |
528 | EXPORT_SYMBOL(dev_remove_offload); | |
529 | ||
1da177e4 LT |
530 | /****************************************************************************** |
531 | ||
532 | Device Boot-time Settings Routines | |
533 | ||
534 | *******************************************************************************/ | |
535 | ||
536 | /* Boot time configuration table */ | |
537 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
538 | ||
539 | /** | |
540 | * netdev_boot_setup_add - add new setup entry | |
541 | * @name: name of the device | |
542 | * @map: configured settings for the device | |
543 | * | |
544 | * Adds new setup entry to the dev_boot_setup list. The function | |
545 | * returns 0 on error and 1 on success. This is a generic routine to | |
546 | * all netdevices. | |
547 | */ | |
548 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
549 | { | |
550 | struct netdev_boot_setup *s; | |
551 | int i; | |
552 | ||
553 | s = dev_boot_setup; | |
554 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
555 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
556 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 557 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
558 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
559 | break; | |
560 | } | |
561 | } | |
562 | ||
563 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
564 | } | |
565 | ||
566 | /** | |
567 | * netdev_boot_setup_check - check boot time settings | |
568 | * @dev: the netdevice | |
569 | * | |
570 | * Check boot time settings for the device. | |
571 | * The found settings are set for the device to be used | |
572 | * later in the device probing. | |
573 | * Returns 0 if no settings found, 1 if they are. | |
574 | */ | |
575 | int netdev_boot_setup_check(struct net_device *dev) | |
576 | { | |
577 | struct netdev_boot_setup *s = dev_boot_setup; | |
578 | int i; | |
579 | ||
580 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
581 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 582 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
583 | dev->irq = s[i].map.irq; |
584 | dev->base_addr = s[i].map.base_addr; | |
585 | dev->mem_start = s[i].map.mem_start; | |
586 | dev->mem_end = s[i].map.mem_end; | |
587 | return 1; | |
588 | } | |
589 | } | |
590 | return 0; | |
591 | } | |
d1b19dff | 592 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
593 | |
594 | ||
595 | /** | |
596 | * netdev_boot_base - get address from boot time settings | |
597 | * @prefix: prefix for network device | |
598 | * @unit: id for network device | |
599 | * | |
600 | * Check boot time settings for the base address of device. | |
601 | * The found settings are set for the device to be used | |
602 | * later in the device probing. | |
603 | * Returns 0 if no settings found. | |
604 | */ | |
605 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
606 | { | |
607 | const struct netdev_boot_setup *s = dev_boot_setup; | |
608 | char name[IFNAMSIZ]; | |
609 | int i; | |
610 | ||
611 | sprintf(name, "%s%d", prefix, unit); | |
612 | ||
613 | /* | |
614 | * If device already registered then return base of 1 | |
615 | * to indicate not to probe for this interface | |
616 | */ | |
881d966b | 617 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
618 | return 1; |
619 | ||
620 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
621 | if (!strcmp(name, s[i].name)) | |
622 | return s[i].map.base_addr; | |
623 | return 0; | |
624 | } | |
625 | ||
626 | /* | |
627 | * Saves at boot time configured settings for any netdevice. | |
628 | */ | |
629 | int __init netdev_boot_setup(char *str) | |
630 | { | |
631 | int ints[5]; | |
632 | struct ifmap map; | |
633 | ||
634 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
635 | if (!str || !*str) | |
636 | return 0; | |
637 | ||
638 | /* Save settings */ | |
639 | memset(&map, 0, sizeof(map)); | |
640 | if (ints[0] > 0) | |
641 | map.irq = ints[1]; | |
642 | if (ints[0] > 1) | |
643 | map.base_addr = ints[2]; | |
644 | if (ints[0] > 2) | |
645 | map.mem_start = ints[3]; | |
646 | if (ints[0] > 3) | |
647 | map.mem_end = ints[4]; | |
648 | ||
649 | /* Add new entry to the list */ | |
650 | return netdev_boot_setup_add(str, &map); | |
651 | } | |
652 | ||
653 | __setup("netdev=", netdev_boot_setup); | |
654 | ||
655 | /******************************************************************************* | |
656 | ||
657 | Device Interface Subroutines | |
658 | ||
659 | *******************************************************************************/ | |
660 | ||
661 | /** | |
662 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 663 | * @net: the applicable net namespace |
1da177e4 LT |
664 | * @name: name to find |
665 | * | |
666 | * Find an interface by name. Must be called under RTNL semaphore | |
667 | * or @dev_base_lock. If the name is found a pointer to the device | |
668 | * is returned. If the name is not found then %NULL is returned. The | |
669 | * reference counters are not incremented so the caller must be | |
670 | * careful with locks. | |
671 | */ | |
672 | ||
881d966b | 673 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 674 | { |
0bd8d536 ED |
675 | struct net_device *dev; |
676 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 677 | |
b67bfe0d | 678 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
679 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
680 | return dev; | |
0bd8d536 | 681 | |
1da177e4 LT |
682 | return NULL; |
683 | } | |
d1b19dff | 684 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 685 | |
72c9528b ED |
686 | /** |
687 | * dev_get_by_name_rcu - find a device by its name | |
688 | * @net: the applicable net namespace | |
689 | * @name: name to find | |
690 | * | |
691 | * Find an interface by name. | |
692 | * If the name is found a pointer to the device is returned. | |
693 | * If the name is not found then %NULL is returned. | |
694 | * The reference counters are not incremented so the caller must be | |
695 | * careful with locks. The caller must hold RCU lock. | |
696 | */ | |
697 | ||
698 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
699 | { | |
72c9528b ED |
700 | struct net_device *dev; |
701 | struct hlist_head *head = dev_name_hash(net, name); | |
702 | ||
b67bfe0d | 703 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
704 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
705 | return dev; | |
706 | ||
707 | return NULL; | |
708 | } | |
709 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
710 | ||
1da177e4 LT |
711 | /** |
712 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 713 | * @net: the applicable net namespace |
1da177e4 LT |
714 | * @name: name to find |
715 | * | |
716 | * Find an interface by name. This can be called from any | |
717 | * context and does its own locking. The returned handle has | |
718 | * the usage count incremented and the caller must use dev_put() to | |
719 | * release it when it is no longer needed. %NULL is returned if no | |
720 | * matching device is found. | |
721 | */ | |
722 | ||
881d966b | 723 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
724 | { |
725 | struct net_device *dev; | |
726 | ||
72c9528b ED |
727 | rcu_read_lock(); |
728 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
729 | if (dev) |
730 | dev_hold(dev); | |
72c9528b | 731 | rcu_read_unlock(); |
1da177e4 LT |
732 | return dev; |
733 | } | |
d1b19dff | 734 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
735 | |
736 | /** | |
737 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 738 | * @net: the applicable net namespace |
1da177e4 LT |
739 | * @ifindex: index of device |
740 | * | |
741 | * Search for an interface by index. Returns %NULL if the device | |
742 | * is not found or a pointer to the device. The device has not | |
743 | * had its reference counter increased so the caller must be careful | |
744 | * about locking. The caller must hold either the RTNL semaphore | |
745 | * or @dev_base_lock. | |
746 | */ | |
747 | ||
881d966b | 748 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 749 | { |
0bd8d536 ED |
750 | struct net_device *dev; |
751 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 752 | |
b67bfe0d | 753 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
754 | if (dev->ifindex == ifindex) |
755 | return dev; | |
0bd8d536 | 756 | |
1da177e4 LT |
757 | return NULL; |
758 | } | |
d1b19dff | 759 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 760 | |
fb699dfd ED |
761 | /** |
762 | * dev_get_by_index_rcu - find a device by its ifindex | |
763 | * @net: the applicable net namespace | |
764 | * @ifindex: index of device | |
765 | * | |
766 | * Search for an interface by index. Returns %NULL if the device | |
767 | * is not found or a pointer to the device. The device has not | |
768 | * had its reference counter increased so the caller must be careful | |
769 | * about locking. The caller must hold RCU lock. | |
770 | */ | |
771 | ||
772 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
773 | { | |
fb699dfd ED |
774 | struct net_device *dev; |
775 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
776 | ||
b67bfe0d | 777 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
778 | if (dev->ifindex == ifindex) |
779 | return dev; | |
780 | ||
781 | return NULL; | |
782 | } | |
783 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
784 | ||
1da177e4 LT |
785 | |
786 | /** | |
787 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 788 | * @net: the applicable net namespace |
1da177e4 LT |
789 | * @ifindex: index of device |
790 | * | |
791 | * Search for an interface by index. Returns NULL if the device | |
792 | * is not found or a pointer to the device. The device returned has | |
793 | * had a reference added and the pointer is safe until the user calls | |
794 | * dev_put to indicate they have finished with it. | |
795 | */ | |
796 | ||
881d966b | 797 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
798 | { |
799 | struct net_device *dev; | |
800 | ||
fb699dfd ED |
801 | rcu_read_lock(); |
802 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
803 | if (dev) |
804 | dev_hold(dev); | |
fb699dfd | 805 | rcu_read_unlock(); |
1da177e4 LT |
806 | return dev; |
807 | } | |
d1b19dff | 808 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 809 | |
5dbe7c17 NS |
810 | /** |
811 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
812 | * @net: network namespace | |
813 | * @name: a pointer to the buffer where the name will be stored. | |
814 | * @ifindex: the ifindex of the interface to get the name from. | |
815 | * | |
816 | * The use of raw_seqcount_begin() and cond_resched() before | |
817 | * retrying is required as we want to give the writers a chance | |
818 | * to complete when CONFIG_PREEMPT is not set. | |
819 | */ | |
820 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
821 | { | |
822 | struct net_device *dev; | |
823 | unsigned int seq; | |
824 | ||
825 | retry: | |
826 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
827 | rcu_read_lock(); | |
828 | dev = dev_get_by_index_rcu(net, ifindex); | |
829 | if (!dev) { | |
830 | rcu_read_unlock(); | |
831 | return -ENODEV; | |
832 | } | |
833 | ||
834 | strcpy(name, dev->name); | |
835 | rcu_read_unlock(); | |
836 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
837 | cond_resched(); | |
838 | goto retry; | |
839 | } | |
840 | ||
841 | return 0; | |
842 | } | |
843 | ||
1da177e4 | 844 | /** |
941666c2 | 845 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 846 | * @net: the applicable net namespace |
1da177e4 LT |
847 | * @type: media type of device |
848 | * @ha: hardware address | |
849 | * | |
850 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
851 | * is not found or a pointer to the device. |
852 | * The caller must hold RCU or RTNL. | |
941666c2 | 853 | * The returned device has not had its ref count increased |
1da177e4 LT |
854 | * and the caller must therefore be careful about locking |
855 | * | |
1da177e4 LT |
856 | */ |
857 | ||
941666c2 ED |
858 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
859 | const char *ha) | |
1da177e4 LT |
860 | { |
861 | struct net_device *dev; | |
862 | ||
941666c2 | 863 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
864 | if (dev->type == type && |
865 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
866 | return dev; |
867 | ||
868 | return NULL; | |
1da177e4 | 869 | } |
941666c2 | 870 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 871 | |
881d966b | 872 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
873 | { |
874 | struct net_device *dev; | |
875 | ||
4e9cac2b | 876 | ASSERT_RTNL(); |
881d966b | 877 | for_each_netdev(net, dev) |
4e9cac2b | 878 | if (dev->type == type) |
7562f876 PE |
879 | return dev; |
880 | ||
881 | return NULL; | |
4e9cac2b | 882 | } |
4e9cac2b PM |
883 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
884 | ||
881d966b | 885 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 886 | { |
99fe3c39 | 887 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 888 | |
99fe3c39 ED |
889 | rcu_read_lock(); |
890 | for_each_netdev_rcu(net, dev) | |
891 | if (dev->type == type) { | |
892 | dev_hold(dev); | |
893 | ret = dev; | |
894 | break; | |
895 | } | |
896 | rcu_read_unlock(); | |
897 | return ret; | |
1da177e4 | 898 | } |
1da177e4 LT |
899 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
900 | ||
901 | /** | |
6c555490 | 902 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 903 | * @net: the applicable net namespace |
1da177e4 LT |
904 | * @if_flags: IFF_* values |
905 | * @mask: bitmask of bits in if_flags to check | |
906 | * | |
907 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 908 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 909 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
910 | */ |
911 | ||
6c555490 WC |
912 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
913 | unsigned short mask) | |
1da177e4 | 914 | { |
7562f876 | 915 | struct net_device *dev, *ret; |
1da177e4 | 916 | |
6c555490 WC |
917 | ASSERT_RTNL(); |
918 | ||
7562f876 | 919 | ret = NULL; |
6c555490 | 920 | for_each_netdev(net, dev) { |
1da177e4 | 921 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 922 | ret = dev; |
1da177e4 LT |
923 | break; |
924 | } | |
925 | } | |
7562f876 | 926 | return ret; |
1da177e4 | 927 | } |
6c555490 | 928 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
929 | |
930 | /** | |
931 | * dev_valid_name - check if name is okay for network device | |
932 | * @name: name string | |
933 | * | |
934 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
935 | * to allow sysfs to work. We also disallow any kind of |
936 | * whitespace. | |
1da177e4 | 937 | */ |
95f050bf | 938 | bool dev_valid_name(const char *name) |
1da177e4 | 939 | { |
c7fa9d18 | 940 | if (*name == '\0') |
95f050bf | 941 | return false; |
b6fe17d6 | 942 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 943 | return false; |
c7fa9d18 | 944 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 945 | return false; |
c7fa9d18 DM |
946 | |
947 | while (*name) { | |
948 | if (*name == '/' || isspace(*name)) | |
95f050bf | 949 | return false; |
c7fa9d18 DM |
950 | name++; |
951 | } | |
95f050bf | 952 | return true; |
1da177e4 | 953 | } |
d1b19dff | 954 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
955 | |
956 | /** | |
b267b179 EB |
957 | * __dev_alloc_name - allocate a name for a device |
958 | * @net: network namespace to allocate the device name in | |
1da177e4 | 959 | * @name: name format string |
b267b179 | 960 | * @buf: scratch buffer and result name string |
1da177e4 LT |
961 | * |
962 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
963 | * id. It scans list of devices to build up a free map, then chooses |
964 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
965 | * while allocating the name and adding the device in order to avoid | |
966 | * duplicates. | |
967 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
968 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
969 | */ |
970 | ||
b267b179 | 971 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
972 | { |
973 | int i = 0; | |
1da177e4 LT |
974 | const char *p; |
975 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 976 | unsigned long *inuse; |
1da177e4 LT |
977 | struct net_device *d; |
978 | ||
979 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
980 | if (p) { | |
981 | /* | |
982 | * Verify the string as this thing may have come from | |
983 | * the user. There must be either one "%d" and no other "%" | |
984 | * characters. | |
985 | */ | |
986 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
987 | return -EINVAL; | |
988 | ||
989 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 990 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
991 | if (!inuse) |
992 | return -ENOMEM; | |
993 | ||
881d966b | 994 | for_each_netdev(net, d) { |
1da177e4 LT |
995 | if (!sscanf(d->name, name, &i)) |
996 | continue; | |
997 | if (i < 0 || i >= max_netdevices) | |
998 | continue; | |
999 | ||
1000 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1001 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1002 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1003 | set_bit(i, inuse); | |
1004 | } | |
1005 | ||
1006 | i = find_first_zero_bit(inuse, max_netdevices); | |
1007 | free_page((unsigned long) inuse); | |
1008 | } | |
1009 | ||
d9031024 OP |
1010 | if (buf != name) |
1011 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1012 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1013 | return i; |
1da177e4 LT |
1014 | |
1015 | /* It is possible to run out of possible slots | |
1016 | * when the name is long and there isn't enough space left | |
1017 | * for the digits, or if all bits are used. | |
1018 | */ | |
1019 | return -ENFILE; | |
1020 | } | |
1021 | ||
b267b179 EB |
1022 | /** |
1023 | * dev_alloc_name - allocate a name for a device | |
1024 | * @dev: device | |
1025 | * @name: name format string | |
1026 | * | |
1027 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1028 | * id. It scans list of devices to build up a free map, then chooses | |
1029 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1030 | * while allocating the name and adding the device in order to avoid | |
1031 | * duplicates. | |
1032 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1033 | * Returns the number of the unit assigned or a negative errno code. | |
1034 | */ | |
1035 | ||
1036 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1037 | { | |
1038 | char buf[IFNAMSIZ]; | |
1039 | struct net *net; | |
1040 | int ret; | |
1041 | ||
c346dca1 YH |
1042 | BUG_ON(!dev_net(dev)); |
1043 | net = dev_net(dev); | |
b267b179 EB |
1044 | ret = __dev_alloc_name(net, name, buf); |
1045 | if (ret >= 0) | |
1046 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1047 | return ret; | |
1048 | } | |
d1b19dff | 1049 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1050 | |
828de4f6 G |
1051 | static int dev_alloc_name_ns(struct net *net, |
1052 | struct net_device *dev, | |
1053 | const char *name) | |
d9031024 | 1054 | { |
828de4f6 G |
1055 | char buf[IFNAMSIZ]; |
1056 | int ret; | |
8ce6cebc | 1057 | |
828de4f6 G |
1058 | ret = __dev_alloc_name(net, name, buf); |
1059 | if (ret >= 0) | |
1060 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1061 | return ret; | |
1062 | } | |
1063 | ||
1064 | static int dev_get_valid_name(struct net *net, | |
1065 | struct net_device *dev, | |
1066 | const char *name) | |
1067 | { | |
1068 | BUG_ON(!net); | |
8ce6cebc | 1069 | |
d9031024 OP |
1070 | if (!dev_valid_name(name)) |
1071 | return -EINVAL; | |
1072 | ||
1c5cae81 | 1073 | if (strchr(name, '%')) |
828de4f6 | 1074 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1075 | else if (__dev_get_by_name(net, name)) |
1076 | return -EEXIST; | |
8ce6cebc DL |
1077 | else if (dev->name != name) |
1078 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1079 | |
1080 | return 0; | |
1081 | } | |
1da177e4 LT |
1082 | |
1083 | /** | |
1084 | * dev_change_name - change name of a device | |
1085 | * @dev: device | |
1086 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1087 | * | |
1088 | * Change name of a device, can pass format strings "eth%d". | |
1089 | * for wildcarding. | |
1090 | */ | |
cf04a4c7 | 1091 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1092 | { |
238fa362 | 1093 | unsigned char old_assign_type; |
fcc5a03a | 1094 | char oldname[IFNAMSIZ]; |
1da177e4 | 1095 | int err = 0; |
fcc5a03a | 1096 | int ret; |
881d966b | 1097 | struct net *net; |
1da177e4 LT |
1098 | |
1099 | ASSERT_RTNL(); | |
c346dca1 | 1100 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1101 | |
c346dca1 | 1102 | net = dev_net(dev); |
1da177e4 LT |
1103 | if (dev->flags & IFF_UP) |
1104 | return -EBUSY; | |
1105 | ||
30e6c9fa | 1106 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1107 | |
1108 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1109 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1110 | return 0; |
c91f6df2 | 1111 | } |
c8d90dca | 1112 | |
fcc5a03a HX |
1113 | memcpy(oldname, dev->name, IFNAMSIZ); |
1114 | ||
828de4f6 | 1115 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1116 | if (err < 0) { |
30e6c9fa | 1117 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1118 | return err; |
c91f6df2 | 1119 | } |
1da177e4 | 1120 | |
6fe82a39 VF |
1121 | if (oldname[0] && !strchr(oldname, '%')) |
1122 | netdev_info(dev, "renamed from %s\n", oldname); | |
1123 | ||
238fa362 TG |
1124 | old_assign_type = dev->name_assign_type; |
1125 | dev->name_assign_type = NET_NAME_RENAMED; | |
1126 | ||
fcc5a03a | 1127 | rollback: |
a1b3f594 EB |
1128 | ret = device_rename(&dev->dev, dev->name); |
1129 | if (ret) { | |
1130 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1131 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1132 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1133 | return ret; |
dcc99773 | 1134 | } |
7f988eab | 1135 | |
30e6c9fa | 1136 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1137 | |
5bb025fa VF |
1138 | netdev_adjacent_rename_links(dev, oldname); |
1139 | ||
7f988eab | 1140 | write_lock_bh(&dev_base_lock); |
372b2312 | 1141 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1142 | write_unlock_bh(&dev_base_lock); |
1143 | ||
1144 | synchronize_rcu(); | |
1145 | ||
1146 | write_lock_bh(&dev_base_lock); | |
1147 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1148 | write_unlock_bh(&dev_base_lock); |
1149 | ||
056925ab | 1150 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1151 | ret = notifier_to_errno(ret); |
1152 | ||
1153 | if (ret) { | |
91e9c07b ED |
1154 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1155 | if (err >= 0) { | |
fcc5a03a | 1156 | err = ret; |
30e6c9fa | 1157 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1158 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1159 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1160 | dev->name_assign_type = old_assign_type; |
1161 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1162 | goto rollback; |
91e9c07b | 1163 | } else { |
7b6cd1ce | 1164 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1165 | dev->name, ret); |
fcc5a03a HX |
1166 | } |
1167 | } | |
1da177e4 LT |
1168 | |
1169 | return err; | |
1170 | } | |
1171 | ||
0b815a1a SH |
1172 | /** |
1173 | * dev_set_alias - change ifalias of a device | |
1174 | * @dev: device | |
1175 | * @alias: name up to IFALIASZ | |
f0db275a | 1176 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1177 | * |
1178 | * Set ifalias for a device, | |
1179 | */ | |
1180 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1181 | { | |
7364e445 AK |
1182 | char *new_ifalias; |
1183 | ||
0b815a1a SH |
1184 | ASSERT_RTNL(); |
1185 | ||
1186 | if (len >= IFALIASZ) | |
1187 | return -EINVAL; | |
1188 | ||
96ca4a2c | 1189 | if (!len) { |
388dfc2d SK |
1190 | kfree(dev->ifalias); |
1191 | dev->ifalias = NULL; | |
96ca4a2c OH |
1192 | return 0; |
1193 | } | |
1194 | ||
7364e445 AK |
1195 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1196 | if (!new_ifalias) | |
0b815a1a | 1197 | return -ENOMEM; |
7364e445 | 1198 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1199 | |
1200 | strlcpy(dev->ifalias, alias, len+1); | |
1201 | return len; | |
1202 | } | |
1203 | ||
1204 | ||
d8a33ac4 | 1205 | /** |
3041a069 | 1206 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1207 | * @dev: device to cause notification |
1208 | * | |
1209 | * Called to indicate a device has changed features. | |
1210 | */ | |
1211 | void netdev_features_change(struct net_device *dev) | |
1212 | { | |
056925ab | 1213 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1214 | } |
1215 | EXPORT_SYMBOL(netdev_features_change); | |
1216 | ||
1da177e4 LT |
1217 | /** |
1218 | * netdev_state_change - device changes state | |
1219 | * @dev: device to cause notification | |
1220 | * | |
1221 | * Called to indicate a device has changed state. This function calls | |
1222 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1223 | * to the routing socket. | |
1224 | */ | |
1225 | void netdev_state_change(struct net_device *dev) | |
1226 | { | |
1227 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1228 | struct netdev_notifier_change_info change_info; |
1229 | ||
1230 | change_info.flags_changed = 0; | |
1231 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1232 | &change_info.info); | |
7f294054 | 1233 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1234 | } |
1235 | } | |
d1b19dff | 1236 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1237 | |
ee89bab1 AW |
1238 | /** |
1239 | * netdev_notify_peers - notify network peers about existence of @dev | |
1240 | * @dev: network device | |
1241 | * | |
1242 | * Generate traffic such that interested network peers are aware of | |
1243 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1244 | * a device wants to inform the rest of the network about some sort of | |
1245 | * reconfiguration such as a failover event or virtual machine | |
1246 | * migration. | |
1247 | */ | |
1248 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1249 | { |
ee89bab1 AW |
1250 | rtnl_lock(); |
1251 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1252 | rtnl_unlock(); | |
c1da4ac7 | 1253 | } |
ee89bab1 | 1254 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1255 | |
bd380811 | 1256 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1257 | { |
d314774c | 1258 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1259 | int ret; |
1da177e4 | 1260 | |
e46b66bc BH |
1261 | ASSERT_RTNL(); |
1262 | ||
1da177e4 LT |
1263 | if (!netif_device_present(dev)) |
1264 | return -ENODEV; | |
1265 | ||
ca99ca14 NH |
1266 | /* Block netpoll from trying to do any rx path servicing. |
1267 | * If we don't do this there is a chance ndo_poll_controller | |
1268 | * or ndo_poll may be running while we open the device | |
1269 | */ | |
66b5552f | 1270 | netpoll_poll_disable(dev); |
ca99ca14 | 1271 | |
3b8bcfd5 JB |
1272 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1273 | ret = notifier_to_errno(ret); | |
1274 | if (ret) | |
1275 | return ret; | |
1276 | ||
1da177e4 | 1277 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1278 | |
d314774c SH |
1279 | if (ops->ndo_validate_addr) |
1280 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1281 | |
d314774c SH |
1282 | if (!ret && ops->ndo_open) |
1283 | ret = ops->ndo_open(dev); | |
1da177e4 | 1284 | |
66b5552f | 1285 | netpoll_poll_enable(dev); |
ca99ca14 | 1286 | |
bada339b JG |
1287 | if (ret) |
1288 | clear_bit(__LINK_STATE_START, &dev->state); | |
1289 | else { | |
1da177e4 | 1290 | dev->flags |= IFF_UP; |
4417da66 | 1291 | dev_set_rx_mode(dev); |
1da177e4 | 1292 | dev_activate(dev); |
7bf23575 | 1293 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1294 | } |
bada339b | 1295 | |
1da177e4 LT |
1296 | return ret; |
1297 | } | |
1298 | ||
1299 | /** | |
bd380811 PM |
1300 | * dev_open - prepare an interface for use. |
1301 | * @dev: device to open | |
1da177e4 | 1302 | * |
bd380811 PM |
1303 | * Takes a device from down to up state. The device's private open |
1304 | * function is invoked and then the multicast lists are loaded. Finally | |
1305 | * the device is moved into the up state and a %NETDEV_UP message is | |
1306 | * sent to the netdev notifier chain. | |
1307 | * | |
1308 | * Calling this function on an active interface is a nop. On a failure | |
1309 | * a negative errno code is returned. | |
1da177e4 | 1310 | */ |
bd380811 PM |
1311 | int dev_open(struct net_device *dev) |
1312 | { | |
1313 | int ret; | |
1314 | ||
bd380811 PM |
1315 | if (dev->flags & IFF_UP) |
1316 | return 0; | |
1317 | ||
bd380811 PM |
1318 | ret = __dev_open(dev); |
1319 | if (ret < 0) | |
1320 | return ret; | |
1321 | ||
7f294054 | 1322 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1323 | call_netdevice_notifiers(NETDEV_UP, dev); |
1324 | ||
1325 | return ret; | |
1326 | } | |
1327 | EXPORT_SYMBOL(dev_open); | |
1328 | ||
44345724 | 1329 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1330 | { |
44345724 | 1331 | struct net_device *dev; |
e46b66bc | 1332 | |
bd380811 | 1333 | ASSERT_RTNL(); |
9d5010db DM |
1334 | might_sleep(); |
1335 | ||
5cde2829 | 1336 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1337 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1338 | netpoll_poll_disable(dev); |
3f4df206 | 1339 | |
44345724 | 1340 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1341 | |
44345724 | 1342 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1343 | |
44345724 OP |
1344 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1345 | * can be even on different cpu. So just clear netif_running(). | |
1346 | * | |
1347 | * dev->stop() will invoke napi_disable() on all of it's | |
1348 | * napi_struct instances on this device. | |
1349 | */ | |
4e857c58 | 1350 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1351 | } |
1da177e4 | 1352 | |
44345724 | 1353 | dev_deactivate_many(head); |
d8b2a4d2 | 1354 | |
5cde2829 | 1355 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1356 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1357 | |
44345724 OP |
1358 | /* |
1359 | * Call the device specific close. This cannot fail. | |
1360 | * Only if device is UP | |
1361 | * | |
1362 | * We allow it to be called even after a DETACH hot-plug | |
1363 | * event. | |
1364 | */ | |
1365 | if (ops->ndo_stop) | |
1366 | ops->ndo_stop(dev); | |
1367 | ||
44345724 | 1368 | dev->flags &= ~IFF_UP; |
66b5552f | 1369 | netpoll_poll_enable(dev); |
44345724 OP |
1370 | } |
1371 | ||
1372 | return 0; | |
1373 | } | |
1374 | ||
1375 | static int __dev_close(struct net_device *dev) | |
1376 | { | |
f87e6f47 | 1377 | int retval; |
44345724 OP |
1378 | LIST_HEAD(single); |
1379 | ||
5cde2829 | 1380 | list_add(&dev->close_list, &single); |
f87e6f47 LT |
1381 | retval = __dev_close_many(&single); |
1382 | list_del(&single); | |
ca99ca14 | 1383 | |
f87e6f47 | 1384 | return retval; |
44345724 OP |
1385 | } |
1386 | ||
3fbd8758 | 1387 | static int dev_close_many(struct list_head *head) |
44345724 OP |
1388 | { |
1389 | struct net_device *dev, *tmp; | |
1da177e4 | 1390 | |
5cde2829 EB |
1391 | /* Remove the devices that don't need to be closed */ |
1392 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1393 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1394 | list_del_init(&dev->close_list); |
44345724 OP |
1395 | |
1396 | __dev_close_many(head); | |
1da177e4 | 1397 | |
5cde2829 | 1398 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1399 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1400 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
5cde2829 | 1401 | list_del_init(&dev->close_list); |
44345724 | 1402 | } |
bd380811 PM |
1403 | |
1404 | return 0; | |
1405 | } | |
1406 | ||
1407 | /** | |
1408 | * dev_close - shutdown an interface. | |
1409 | * @dev: device to shutdown | |
1410 | * | |
1411 | * This function moves an active device into down state. A | |
1412 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1413 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1414 | * chain. | |
1415 | */ | |
1416 | int dev_close(struct net_device *dev) | |
1417 | { | |
e14a5993 ED |
1418 | if (dev->flags & IFF_UP) { |
1419 | LIST_HEAD(single); | |
1da177e4 | 1420 | |
5cde2829 | 1421 | list_add(&dev->close_list, &single); |
e14a5993 ED |
1422 | dev_close_many(&single); |
1423 | list_del(&single); | |
1424 | } | |
da6e378b | 1425 | return 0; |
1da177e4 | 1426 | } |
d1b19dff | 1427 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1428 | |
1429 | ||
0187bdfb BH |
1430 | /** |
1431 | * dev_disable_lro - disable Large Receive Offload on a device | |
1432 | * @dev: device | |
1433 | * | |
1434 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1435 | * called under RTNL. This is needed if received packets may be | |
1436 | * forwarded to another interface. | |
1437 | */ | |
1438 | void dev_disable_lro(struct net_device *dev) | |
1439 | { | |
fbe168ba MK |
1440 | struct net_device *lower_dev; |
1441 | struct list_head *iter; | |
529d0489 | 1442 | |
bc5787c6 MM |
1443 | dev->wanted_features &= ~NETIF_F_LRO; |
1444 | netdev_update_features(dev); | |
27660515 | 1445 | |
22d5969f MM |
1446 | if (unlikely(dev->features & NETIF_F_LRO)) |
1447 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1448 | |
1449 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1450 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1451 | } |
1452 | EXPORT_SYMBOL(dev_disable_lro); | |
1453 | ||
351638e7 JP |
1454 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1455 | struct net_device *dev) | |
1456 | { | |
1457 | struct netdev_notifier_info info; | |
1458 | ||
1459 | netdev_notifier_info_init(&info, dev); | |
1460 | return nb->notifier_call(nb, val, &info); | |
1461 | } | |
0187bdfb | 1462 | |
881d966b EB |
1463 | static int dev_boot_phase = 1; |
1464 | ||
1da177e4 LT |
1465 | /** |
1466 | * register_netdevice_notifier - register a network notifier block | |
1467 | * @nb: notifier | |
1468 | * | |
1469 | * Register a notifier to be called when network device events occur. | |
1470 | * The notifier passed is linked into the kernel structures and must | |
1471 | * not be reused until it has been unregistered. A negative errno code | |
1472 | * is returned on a failure. | |
1473 | * | |
1474 | * When registered all registration and up events are replayed | |
4ec93edb | 1475 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1476 | * view of the network device list. |
1477 | */ | |
1478 | ||
1479 | int register_netdevice_notifier(struct notifier_block *nb) | |
1480 | { | |
1481 | struct net_device *dev; | |
fcc5a03a | 1482 | struct net_device *last; |
881d966b | 1483 | struct net *net; |
1da177e4 LT |
1484 | int err; |
1485 | ||
1486 | rtnl_lock(); | |
f07d5b94 | 1487 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1488 | if (err) |
1489 | goto unlock; | |
881d966b EB |
1490 | if (dev_boot_phase) |
1491 | goto unlock; | |
1492 | for_each_net(net) { | |
1493 | for_each_netdev(net, dev) { | |
351638e7 | 1494 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1495 | err = notifier_to_errno(err); |
1496 | if (err) | |
1497 | goto rollback; | |
1498 | ||
1499 | if (!(dev->flags & IFF_UP)) | |
1500 | continue; | |
1da177e4 | 1501 | |
351638e7 | 1502 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1503 | } |
1da177e4 | 1504 | } |
fcc5a03a HX |
1505 | |
1506 | unlock: | |
1da177e4 LT |
1507 | rtnl_unlock(); |
1508 | return err; | |
fcc5a03a HX |
1509 | |
1510 | rollback: | |
1511 | last = dev; | |
881d966b EB |
1512 | for_each_net(net) { |
1513 | for_each_netdev(net, dev) { | |
1514 | if (dev == last) | |
8f891489 | 1515 | goto outroll; |
fcc5a03a | 1516 | |
881d966b | 1517 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1518 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1519 | dev); | |
1520 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1521 | } |
351638e7 | 1522 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1523 | } |
fcc5a03a | 1524 | } |
c67625a1 | 1525 | |
8f891489 | 1526 | outroll: |
c67625a1 | 1527 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1528 | goto unlock; |
1da177e4 | 1529 | } |
d1b19dff | 1530 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1531 | |
1532 | /** | |
1533 | * unregister_netdevice_notifier - unregister a network notifier block | |
1534 | * @nb: notifier | |
1535 | * | |
1536 | * Unregister a notifier previously registered by | |
1537 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1538 | * kernel structures and may then be reused. A negative errno code | |
1539 | * is returned on a failure. | |
7d3d43da EB |
1540 | * |
1541 | * After unregistering unregister and down device events are synthesized | |
1542 | * for all devices on the device list to the removed notifier to remove | |
1543 | * the need for special case cleanup code. | |
1da177e4 LT |
1544 | */ |
1545 | ||
1546 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1547 | { | |
7d3d43da EB |
1548 | struct net_device *dev; |
1549 | struct net *net; | |
9f514950 HX |
1550 | int err; |
1551 | ||
1552 | rtnl_lock(); | |
f07d5b94 | 1553 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1554 | if (err) |
1555 | goto unlock; | |
1556 | ||
1557 | for_each_net(net) { | |
1558 | for_each_netdev(net, dev) { | |
1559 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1560 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1561 | dev); | |
1562 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1563 | } |
351638e7 | 1564 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1565 | } |
1566 | } | |
1567 | unlock: | |
9f514950 HX |
1568 | rtnl_unlock(); |
1569 | return err; | |
1da177e4 | 1570 | } |
d1b19dff | 1571 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1572 | |
351638e7 JP |
1573 | /** |
1574 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1575 | * @val: value passed unmodified to notifier function | |
1576 | * @dev: net_device pointer passed unmodified to notifier function | |
1577 | * @info: notifier information data | |
1578 | * | |
1579 | * Call all network notifier blocks. Parameters and return value | |
1580 | * are as for raw_notifier_call_chain(). | |
1581 | */ | |
1582 | ||
1d143d9f | 1583 | static int call_netdevice_notifiers_info(unsigned long val, |
1584 | struct net_device *dev, | |
1585 | struct netdev_notifier_info *info) | |
351638e7 JP |
1586 | { |
1587 | ASSERT_RTNL(); | |
1588 | netdev_notifier_info_init(info, dev); | |
1589 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1590 | } | |
351638e7 | 1591 | |
1da177e4 LT |
1592 | /** |
1593 | * call_netdevice_notifiers - call all network notifier blocks | |
1594 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1595 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1596 | * |
1597 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1598 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1599 | */ |
1600 | ||
ad7379d4 | 1601 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1602 | { |
351638e7 JP |
1603 | struct netdev_notifier_info info; |
1604 | ||
1605 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1606 | } |
edf947f1 | 1607 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1608 | |
c5905afb | 1609 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1610 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1611 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1612 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1613 | * static_key_slow_dec() calls. |
b90e5794 ED |
1614 | */ |
1615 | static atomic_t netstamp_needed_deferred; | |
1616 | #endif | |
1da177e4 LT |
1617 | |
1618 | void net_enable_timestamp(void) | |
1619 | { | |
b90e5794 ED |
1620 | #ifdef HAVE_JUMP_LABEL |
1621 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1622 | ||
1623 | if (deferred) { | |
1624 | while (--deferred) | |
c5905afb | 1625 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1626 | return; |
1627 | } | |
1628 | #endif | |
c5905afb | 1629 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1630 | } |
d1b19dff | 1631 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1632 | |
1633 | void net_disable_timestamp(void) | |
1634 | { | |
b90e5794 ED |
1635 | #ifdef HAVE_JUMP_LABEL |
1636 | if (in_interrupt()) { | |
1637 | atomic_inc(&netstamp_needed_deferred); | |
1638 | return; | |
1639 | } | |
1640 | #endif | |
c5905afb | 1641 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1642 | } |
d1b19dff | 1643 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1644 | |
3b098e2d | 1645 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1646 | { |
588f0330 | 1647 | skb->tstamp.tv64 = 0; |
c5905afb | 1648 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1649 | __net_timestamp(skb); |
1da177e4 LT |
1650 | } |
1651 | ||
588f0330 | 1652 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1653 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1654 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1655 | __net_timestamp(SKB); \ | |
1656 | } \ | |
3b098e2d | 1657 | |
1ee481fb | 1658 | bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb) |
79b569f0 DL |
1659 | { |
1660 | unsigned int len; | |
1661 | ||
1662 | if (!(dev->flags & IFF_UP)) | |
1663 | return false; | |
1664 | ||
1665 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1666 | if (skb->len <= len) | |
1667 | return true; | |
1668 | ||
1669 | /* if TSO is enabled, we don't care about the length as the packet | |
1670 | * could be forwarded without being segmented before | |
1671 | */ | |
1672 | if (skb_is_gso(skb)) | |
1673 | return true; | |
1674 | ||
1675 | return false; | |
1676 | } | |
1ee481fb | 1677 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1678 | |
a0265d28 HX |
1679 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1680 | { | |
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 | ||
1689 | if (unlikely(!is_skb_forwardable(dev, skb))) { | |
1690 | atomic_long_inc(&dev->rx_dropped); | |
1691 | kfree_skb(skb); | |
1692 | return NET_RX_DROP; | |
1693 | } | |
1694 | ||
1695 | skb_scrub_packet(skb, true); | |
1696 | skb->protocol = eth_type_trans(skb, dev); | |
2c26d34b | 1697 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); |
a0265d28 HX |
1698 | |
1699 | return 0; | |
1700 | } | |
1701 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1702 | ||
44540960 AB |
1703 | /** |
1704 | * dev_forward_skb - loopback an skb to another netif | |
1705 | * | |
1706 | * @dev: destination network device | |
1707 | * @skb: buffer to forward | |
1708 | * | |
1709 | * return values: | |
1710 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1711 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1712 | * |
1713 | * dev_forward_skb can be used for injecting an skb from the | |
1714 | * start_xmit function of one device into the receive queue | |
1715 | * of another device. | |
1716 | * | |
1717 | * The receiving device may be in another namespace, so | |
1718 | * we have to clear all information in the skb that could | |
1719 | * impact namespace isolation. | |
1720 | */ | |
1721 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1722 | { | |
a0265d28 | 1723 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1724 | } |
1725 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1726 | ||
71d9dec2 CG |
1727 | static inline int deliver_skb(struct sk_buff *skb, |
1728 | struct packet_type *pt_prev, | |
1729 | struct net_device *orig_dev) | |
1730 | { | |
1080e512 MT |
1731 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1732 | return -ENOMEM; | |
71d9dec2 CG |
1733 | atomic_inc(&skb->users); |
1734 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1735 | } | |
1736 | ||
c0de08d0 EL |
1737 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1738 | { | |
a3d744e9 | 1739 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1740 | return false; |
1741 | ||
1742 | if (ptype->id_match) | |
1743 | return ptype->id_match(ptype, skb->sk); | |
1744 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1745 | return true; | |
1746 | ||
1747 | return false; | |
1748 | } | |
1749 | ||
1da177e4 LT |
1750 | /* |
1751 | * Support routine. Sends outgoing frames to any network | |
1752 | * taps currently in use. | |
1753 | */ | |
1754 | ||
f6a78bfc | 1755 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1756 | { |
1757 | struct packet_type *ptype; | |
71d9dec2 CG |
1758 | struct sk_buff *skb2 = NULL; |
1759 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1760 | |
1da177e4 LT |
1761 | rcu_read_lock(); |
1762 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1763 | /* Never send packets back to the socket | |
1764 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1765 | */ | |
1766 | if ((ptype->dev == dev || !ptype->dev) && | |
c0de08d0 | 1767 | (!skb_loop_sk(ptype, skb))) { |
71d9dec2 CG |
1768 | if (pt_prev) { |
1769 | deliver_skb(skb2, pt_prev, skb->dev); | |
1770 | pt_prev = ptype; | |
1771 | continue; | |
1772 | } | |
1773 | ||
1774 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1775 | if (!skb2) |
1776 | break; | |
1777 | ||
70978182 ED |
1778 | net_timestamp_set(skb2); |
1779 | ||
1da177e4 LT |
1780 | /* skb->nh should be correctly |
1781 | set by sender, so that the second statement is | |
1782 | just protection against buggy protocols. | |
1783 | */ | |
459a98ed | 1784 | skb_reset_mac_header(skb2); |
1da177e4 | 1785 | |
d56f90a7 | 1786 | if (skb_network_header(skb2) < skb2->data || |
ced14f68 | 1787 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { |
e87cc472 JP |
1788 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", |
1789 | ntohs(skb2->protocol), | |
1790 | dev->name); | |
c1d2bbe1 | 1791 | skb_reset_network_header(skb2); |
1da177e4 LT |
1792 | } |
1793 | ||
b0e380b1 | 1794 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1795 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1796 | pt_prev = ptype; |
1da177e4 LT |
1797 | } |
1798 | } | |
71d9dec2 CG |
1799 | if (pt_prev) |
1800 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1801 | rcu_read_unlock(); |
1802 | } | |
1803 | ||
2c53040f BH |
1804 | /** |
1805 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1806 | * @dev: Network device |
1807 | * @txq: number of queues available | |
1808 | * | |
1809 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1810 | * valid. To resolve this verify the tc mapping remains valid and if | |
1811 | * not NULL the mapping. With no priorities mapping to this | |
1812 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1813 | * is invalid nothing can be done so disable priority mappings. If is | |
1814 | * expected that drivers will fix this mapping if they can before | |
1815 | * calling netif_set_real_num_tx_queues. | |
1816 | */ | |
bb134d22 | 1817 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1818 | { |
1819 | int i; | |
1820 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1821 | ||
1822 | /* If TC0 is invalidated disable TC mapping */ | |
1823 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1824 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1825 | dev->num_tc = 0; |
1826 | return; | |
1827 | } | |
1828 | ||
1829 | /* Invalidated prio to tc mappings set to TC0 */ | |
1830 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1831 | int q = netdev_get_prio_tc_map(dev, i); | |
1832 | ||
1833 | tc = &dev->tc_to_txq[q]; | |
1834 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1835 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1836 | i, q); | |
4f57c087 JF |
1837 | netdev_set_prio_tc_map(dev, i, 0); |
1838 | } | |
1839 | } | |
1840 | } | |
1841 | ||
537c00de AD |
1842 | #ifdef CONFIG_XPS |
1843 | static DEFINE_MUTEX(xps_map_mutex); | |
1844 | #define xmap_dereference(P) \ | |
1845 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1846 | ||
10cdc3f3 AD |
1847 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1848 | int cpu, u16 index) | |
537c00de | 1849 | { |
10cdc3f3 AD |
1850 | struct xps_map *map = NULL; |
1851 | int pos; | |
537c00de | 1852 | |
10cdc3f3 AD |
1853 | if (dev_maps) |
1854 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1855 | |
10cdc3f3 AD |
1856 | for (pos = 0; map && pos < map->len; pos++) { |
1857 | if (map->queues[pos] == index) { | |
537c00de AD |
1858 | if (map->len > 1) { |
1859 | map->queues[pos] = map->queues[--map->len]; | |
1860 | } else { | |
10cdc3f3 | 1861 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1862 | kfree_rcu(map, rcu); |
1863 | map = NULL; | |
1864 | } | |
10cdc3f3 | 1865 | break; |
537c00de | 1866 | } |
537c00de AD |
1867 | } |
1868 | ||
10cdc3f3 AD |
1869 | return map; |
1870 | } | |
1871 | ||
024e9679 | 1872 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
10cdc3f3 AD |
1873 | { |
1874 | struct xps_dev_maps *dev_maps; | |
024e9679 | 1875 | int cpu, i; |
10cdc3f3 AD |
1876 | bool active = false; |
1877 | ||
1878 | mutex_lock(&xps_map_mutex); | |
1879 | dev_maps = xmap_dereference(dev->xps_maps); | |
1880 | ||
1881 | if (!dev_maps) | |
1882 | goto out_no_maps; | |
1883 | ||
1884 | for_each_possible_cpu(cpu) { | |
024e9679 AD |
1885 | for (i = index; i < dev->num_tx_queues; i++) { |
1886 | if (!remove_xps_queue(dev_maps, cpu, i)) | |
1887 | break; | |
1888 | } | |
1889 | if (i == dev->num_tx_queues) | |
10cdc3f3 AD |
1890 | active = true; |
1891 | } | |
1892 | ||
1893 | if (!active) { | |
537c00de AD |
1894 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
1895 | kfree_rcu(dev_maps, rcu); | |
1896 | } | |
1897 | ||
024e9679 AD |
1898 | for (i = index; i < dev->num_tx_queues; i++) |
1899 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), | |
1900 | NUMA_NO_NODE); | |
1901 | ||
537c00de AD |
1902 | out_no_maps: |
1903 | mutex_unlock(&xps_map_mutex); | |
1904 | } | |
1905 | ||
01c5f864 AD |
1906 | static struct xps_map *expand_xps_map(struct xps_map *map, |
1907 | int cpu, u16 index) | |
1908 | { | |
1909 | struct xps_map *new_map; | |
1910 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
1911 | int i, pos; | |
1912 | ||
1913 | for (pos = 0; map && pos < map->len; pos++) { | |
1914 | if (map->queues[pos] != index) | |
1915 | continue; | |
1916 | return map; | |
1917 | } | |
1918 | ||
1919 | /* Need to add queue to this CPU's existing map */ | |
1920 | if (map) { | |
1921 | if (pos < map->alloc_len) | |
1922 | return map; | |
1923 | ||
1924 | alloc_len = map->alloc_len * 2; | |
1925 | } | |
1926 | ||
1927 | /* Need to allocate new map to store queue on this CPU's map */ | |
1928 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
1929 | cpu_to_node(cpu)); | |
1930 | if (!new_map) | |
1931 | return NULL; | |
1932 | ||
1933 | for (i = 0; i < pos; i++) | |
1934 | new_map->queues[i] = map->queues[i]; | |
1935 | new_map->alloc_len = alloc_len; | |
1936 | new_map->len = pos; | |
1937 | ||
1938 | return new_map; | |
1939 | } | |
1940 | ||
3573540c MT |
1941 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
1942 | u16 index) | |
537c00de | 1943 | { |
01c5f864 | 1944 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 1945 | struct xps_map *map, *new_map; |
537c00de | 1946 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
1947 | int cpu, numa_node_id = -2; |
1948 | bool active = false; | |
537c00de AD |
1949 | |
1950 | mutex_lock(&xps_map_mutex); | |
1951 | ||
1952 | dev_maps = xmap_dereference(dev->xps_maps); | |
1953 | ||
01c5f864 AD |
1954 | /* allocate memory for queue storage */ |
1955 | for_each_online_cpu(cpu) { | |
1956 | if (!cpumask_test_cpu(cpu, mask)) | |
1957 | continue; | |
1958 | ||
1959 | if (!new_dev_maps) | |
1960 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
1961 | if (!new_dev_maps) { |
1962 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 1963 | return -ENOMEM; |
2bb60cb9 | 1964 | } |
01c5f864 AD |
1965 | |
1966 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
1967 | NULL; | |
1968 | ||
1969 | map = expand_xps_map(map, cpu, index); | |
1970 | if (!map) | |
1971 | goto error; | |
1972 | ||
1973 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
1974 | } | |
1975 | ||
1976 | if (!new_dev_maps) | |
1977 | goto out_no_new_maps; | |
1978 | ||
537c00de | 1979 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
1980 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
1981 | /* add queue to CPU maps */ | |
1982 | int pos = 0; | |
1983 | ||
1984 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1985 | while ((pos < map->len) && (map->queues[pos] != index)) | |
1986 | pos++; | |
1987 | ||
1988 | if (pos == map->len) | |
1989 | map->queues[map->len++] = index; | |
537c00de | 1990 | #ifdef CONFIG_NUMA |
537c00de AD |
1991 | if (numa_node_id == -2) |
1992 | numa_node_id = cpu_to_node(cpu); | |
1993 | else if (numa_node_id != cpu_to_node(cpu)) | |
1994 | numa_node_id = -1; | |
537c00de | 1995 | #endif |
01c5f864 AD |
1996 | } else if (dev_maps) { |
1997 | /* fill in the new device map from the old device map */ | |
1998 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
1999 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 2000 | } |
01c5f864 | 2001 | |
537c00de AD |
2002 | } |
2003 | ||
01c5f864 AD |
2004 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2005 | ||
537c00de | 2006 | /* Cleanup old maps */ |
01c5f864 AD |
2007 | if (dev_maps) { |
2008 | for_each_possible_cpu(cpu) { | |
2009 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2010 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2011 | if (map && map != new_map) | |
2012 | kfree_rcu(map, rcu); | |
2013 | } | |
537c00de | 2014 | |
01c5f864 | 2015 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
2016 | } |
2017 | ||
01c5f864 AD |
2018 | dev_maps = new_dev_maps; |
2019 | active = true; | |
537c00de | 2020 | |
01c5f864 AD |
2021 | out_no_new_maps: |
2022 | /* update Tx queue numa node */ | |
537c00de AD |
2023 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2024 | (numa_node_id >= 0) ? numa_node_id : | |
2025 | NUMA_NO_NODE); | |
2026 | ||
01c5f864 AD |
2027 | if (!dev_maps) |
2028 | goto out_no_maps; | |
2029 | ||
2030 | /* removes queue from unused CPUs */ | |
2031 | for_each_possible_cpu(cpu) { | |
2032 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
2033 | continue; | |
2034 | ||
2035 | if (remove_xps_queue(dev_maps, cpu, index)) | |
2036 | active = true; | |
2037 | } | |
2038 | ||
2039 | /* free map if not active */ | |
2040 | if (!active) { | |
2041 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2042 | kfree_rcu(dev_maps, rcu); | |
2043 | } | |
2044 | ||
2045 | out_no_maps: | |
537c00de AD |
2046 | mutex_unlock(&xps_map_mutex); |
2047 | ||
2048 | return 0; | |
2049 | error: | |
01c5f864 AD |
2050 | /* remove any maps that we added */ |
2051 | for_each_possible_cpu(cpu) { | |
2052 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2053 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2054 | NULL; | |
2055 | if (new_map && new_map != map) | |
2056 | kfree(new_map); | |
2057 | } | |
2058 | ||
537c00de AD |
2059 | mutex_unlock(&xps_map_mutex); |
2060 | ||
537c00de AD |
2061 | kfree(new_dev_maps); |
2062 | return -ENOMEM; | |
2063 | } | |
2064 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2065 | ||
2066 | #endif | |
f0796d5c JF |
2067 | /* |
2068 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2069 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2070 | */ | |
e6484930 | 2071 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2072 | { |
1d24eb48 TH |
2073 | int rc; |
2074 | ||
e6484930 TH |
2075 | if (txq < 1 || txq > dev->num_tx_queues) |
2076 | return -EINVAL; | |
f0796d5c | 2077 | |
5c56580b BH |
2078 | if (dev->reg_state == NETREG_REGISTERED || |
2079 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2080 | ASSERT_RTNL(); |
2081 | ||
1d24eb48 TH |
2082 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2083 | txq); | |
bf264145 TH |
2084 | if (rc) |
2085 | return rc; | |
2086 | ||
4f57c087 JF |
2087 | if (dev->num_tc) |
2088 | netif_setup_tc(dev, txq); | |
2089 | ||
024e9679 | 2090 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2091 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2092 | #ifdef CONFIG_XPS |
2093 | netif_reset_xps_queues_gt(dev, txq); | |
2094 | #endif | |
2095 | } | |
f0796d5c | 2096 | } |
e6484930 TH |
2097 | |
2098 | dev->real_num_tx_queues = txq; | |
2099 | return 0; | |
f0796d5c JF |
2100 | } |
2101 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2102 | |
a953be53 | 2103 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2104 | /** |
2105 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2106 | * @dev: Network device | |
2107 | * @rxq: Actual number of RX queues | |
2108 | * | |
2109 | * This must be called either with the rtnl_lock held or before | |
2110 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2111 | * negative error code. If called before registration, it always |
2112 | * succeeds. | |
62fe0b40 BH |
2113 | */ |
2114 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2115 | { | |
2116 | int rc; | |
2117 | ||
bd25fa7b TH |
2118 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2119 | return -EINVAL; | |
2120 | ||
62fe0b40 BH |
2121 | if (dev->reg_state == NETREG_REGISTERED) { |
2122 | ASSERT_RTNL(); | |
2123 | ||
62fe0b40 BH |
2124 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2125 | rxq); | |
2126 | if (rc) | |
2127 | return rc; | |
62fe0b40 BH |
2128 | } |
2129 | ||
2130 | dev->real_num_rx_queues = rxq; | |
2131 | return 0; | |
2132 | } | |
2133 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2134 | #endif | |
2135 | ||
2c53040f BH |
2136 | /** |
2137 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2138 | * |
2139 | * This routine should set an upper limit on the number of RSS queues | |
2140 | * used by default by multiqueue devices. | |
2141 | */ | |
a55b138b | 2142 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
2143 | { |
2144 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
2145 | } | |
2146 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2147 | ||
def82a1d | 2148 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 2149 | { |
def82a1d JP |
2150 | struct softnet_data *sd; |
2151 | unsigned long flags; | |
56079431 | 2152 | |
def82a1d | 2153 | local_irq_save(flags); |
903ceff7 | 2154 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2155 | q->next_sched = NULL; |
2156 | *sd->output_queue_tailp = q; | |
2157 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2158 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2159 | local_irq_restore(flags); | |
2160 | } | |
2161 | ||
2162 | void __netif_schedule(struct Qdisc *q) | |
2163 | { | |
2164 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2165 | __netif_reschedule(q); | |
56079431 DV |
2166 | } |
2167 | EXPORT_SYMBOL(__netif_schedule); | |
2168 | ||
e6247027 ED |
2169 | struct dev_kfree_skb_cb { |
2170 | enum skb_free_reason reason; | |
2171 | }; | |
2172 | ||
2173 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2174 | { |
e6247027 ED |
2175 | return (struct dev_kfree_skb_cb *)skb->cb; |
2176 | } | |
2177 | ||
46e5da40 JF |
2178 | void netif_schedule_queue(struct netdev_queue *txq) |
2179 | { | |
2180 | rcu_read_lock(); | |
2181 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2182 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2183 | ||
2184 | __netif_schedule(q); | |
2185 | } | |
2186 | rcu_read_unlock(); | |
2187 | } | |
2188 | EXPORT_SYMBOL(netif_schedule_queue); | |
2189 | ||
2190 | /** | |
2191 | * netif_wake_subqueue - allow sending packets on subqueue | |
2192 | * @dev: network device | |
2193 | * @queue_index: sub queue index | |
2194 | * | |
2195 | * Resume individual transmit queue of a device with multiple transmit queues. | |
2196 | */ | |
2197 | void netif_wake_subqueue(struct net_device *dev, u16 queue_index) | |
2198 | { | |
2199 | struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index); | |
2200 | ||
2201 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) { | |
2202 | struct Qdisc *q; | |
2203 | ||
2204 | rcu_read_lock(); | |
2205 | q = rcu_dereference(txq->qdisc); | |
2206 | __netif_schedule(q); | |
2207 | rcu_read_unlock(); | |
2208 | } | |
2209 | } | |
2210 | EXPORT_SYMBOL(netif_wake_subqueue); | |
2211 | ||
2212 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) | |
2213 | { | |
2214 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2215 | struct Qdisc *q; | |
2216 | ||
2217 | rcu_read_lock(); | |
2218 | q = rcu_dereference(dev_queue->qdisc); | |
2219 | __netif_schedule(q); | |
2220 | rcu_read_unlock(); | |
2221 | } | |
2222 | } | |
2223 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2224 | ||
e6247027 | 2225 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2226 | { |
e6247027 | 2227 | unsigned long flags; |
56079431 | 2228 | |
e6247027 ED |
2229 | if (likely(atomic_read(&skb->users) == 1)) { |
2230 | smp_rmb(); | |
2231 | atomic_set(&skb->users, 0); | |
2232 | } else if (likely(!atomic_dec_and_test(&skb->users))) { | |
2233 | return; | |
bea3348e | 2234 | } |
e6247027 ED |
2235 | get_kfree_skb_cb(skb)->reason = reason; |
2236 | local_irq_save(flags); | |
2237 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2238 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2239 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2240 | local_irq_restore(flags); | |
56079431 | 2241 | } |
e6247027 | 2242 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2243 | |
e6247027 | 2244 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2245 | { |
2246 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2247 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2248 | else |
2249 | dev_kfree_skb(skb); | |
2250 | } | |
e6247027 | 2251 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2252 | |
2253 | ||
bea3348e SH |
2254 | /** |
2255 | * netif_device_detach - mark device as removed | |
2256 | * @dev: network device | |
2257 | * | |
2258 | * Mark device as removed from system and therefore no longer available. | |
2259 | */ | |
56079431 DV |
2260 | void netif_device_detach(struct net_device *dev) |
2261 | { | |
2262 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2263 | netif_running(dev)) { | |
d543103a | 2264 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2265 | } |
2266 | } | |
2267 | EXPORT_SYMBOL(netif_device_detach); | |
2268 | ||
bea3348e SH |
2269 | /** |
2270 | * netif_device_attach - mark device as attached | |
2271 | * @dev: network device | |
2272 | * | |
2273 | * Mark device as attached from system and restart if needed. | |
2274 | */ | |
56079431 DV |
2275 | void netif_device_attach(struct net_device *dev) |
2276 | { | |
2277 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2278 | netif_running(dev)) { | |
d543103a | 2279 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2280 | __netdev_watchdog_up(dev); |
56079431 DV |
2281 | } |
2282 | } | |
2283 | EXPORT_SYMBOL(netif_device_attach); | |
2284 | ||
36c92474 BH |
2285 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2286 | { | |
65e9d2fa | 2287 | static const netdev_features_t null_features = 0; |
36c92474 BH |
2288 | struct net_device *dev = skb->dev; |
2289 | const char *driver = ""; | |
2290 | ||
c846ad9b BG |
2291 | if (!net_ratelimit()) |
2292 | return; | |
2293 | ||
36c92474 BH |
2294 | if (dev && dev->dev.parent) |
2295 | driver = dev_driver_string(dev->dev.parent); | |
2296 | ||
2297 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " | |
2298 | "gso_type=%d ip_summed=%d\n", | |
65e9d2fa MM |
2299 | driver, dev ? &dev->features : &null_features, |
2300 | skb->sk ? &skb->sk->sk_route_caps : &null_features, | |
36c92474 BH |
2301 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2302 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2303 | } | |
2304 | ||
1da177e4 LT |
2305 | /* |
2306 | * Invalidate hardware checksum when packet is to be mangled, and | |
2307 | * complete checksum manually on outgoing path. | |
2308 | */ | |
84fa7933 | 2309 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2310 | { |
d3bc23e7 | 2311 | __wsum csum; |
663ead3b | 2312 | int ret = 0, offset; |
1da177e4 | 2313 | |
84fa7933 | 2314 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2315 | goto out_set_summed; |
2316 | ||
2317 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2318 | skb_warn_bad_offload(skb); |
2319 | return -EINVAL; | |
1da177e4 LT |
2320 | } |
2321 | ||
cef401de ED |
2322 | /* Before computing a checksum, we should make sure no frag could |
2323 | * be modified by an external entity : checksum could be wrong. | |
2324 | */ | |
2325 | if (skb_has_shared_frag(skb)) { | |
2326 | ret = __skb_linearize(skb); | |
2327 | if (ret) | |
2328 | goto out; | |
2329 | } | |
2330 | ||
55508d60 | 2331 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2332 | BUG_ON(offset >= skb_headlen(skb)); |
2333 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2334 | ||
2335 | offset += skb->csum_offset; | |
2336 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2337 | ||
2338 | if (skb_cloned(skb) && | |
2339 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2340 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2341 | if (ret) | |
2342 | goto out; | |
2343 | } | |
2344 | ||
a030847e | 2345 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2346 | out_set_summed: |
1da177e4 | 2347 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2348 | out: |
1da177e4 LT |
2349 | return ret; |
2350 | } | |
d1b19dff | 2351 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2352 | |
53d6471c | 2353 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2354 | { |
4b9b1cdf | 2355 | unsigned int vlan_depth = skb->mac_len; |
252e3346 | 2356 | __be16 type = skb->protocol; |
f6a78bfc | 2357 | |
19acc327 PS |
2358 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2359 | if (type == htons(ETH_P_TEB)) { | |
2360 | struct ethhdr *eth; | |
2361 | ||
2362 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2363 | return 0; | |
2364 | ||
2365 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2366 | type = eth->h_proto; | |
2367 | } | |
2368 | ||
4b9b1cdf NA |
2369 | /* if skb->protocol is 802.1Q/AD then the header should already be |
2370 | * present at mac_len - VLAN_HLEN (if mac_len > 0), or at | |
2371 | * ETH_HLEN otherwise | |
2372 | */ | |
2373 | if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) { | |
2374 | if (vlan_depth) { | |
80019d31 | 2375 | if (WARN_ON(vlan_depth < VLAN_HLEN)) |
4b9b1cdf NA |
2376 | return 0; |
2377 | vlan_depth -= VLAN_HLEN; | |
2378 | } else { | |
2379 | vlan_depth = ETH_HLEN; | |
2380 | } | |
2381 | do { | |
2382 | struct vlan_hdr *vh; | |
2383 | ||
2384 | if (unlikely(!pskb_may_pull(skb, | |
2385 | vlan_depth + VLAN_HLEN))) | |
2386 | return 0; | |
2387 | ||
2388 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); | |
2389 | type = vh->h_vlan_encapsulated_proto; | |
2390 | vlan_depth += VLAN_HLEN; | |
2391 | } while (type == htons(ETH_P_8021Q) || | |
2392 | type == htons(ETH_P_8021AD)); | |
7b9c6090 JG |
2393 | } |
2394 | ||
53d6471c VY |
2395 | *depth = vlan_depth; |
2396 | ||
ec5f0615 PS |
2397 | return type; |
2398 | } | |
2399 | ||
2400 | /** | |
2401 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2402 | * @skb: buffer to segment | |
2403 | * @features: features for the output path (see dev->features) | |
2404 | */ | |
2405 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2406 | netdev_features_t features) | |
2407 | { | |
2408 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2409 | struct packet_offload *ptype; | |
53d6471c VY |
2410 | int vlan_depth = skb->mac_len; |
2411 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2412 | |
2413 | if (unlikely(!type)) | |
2414 | return ERR_PTR(-EINVAL); | |
2415 | ||
53d6471c | 2416 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2417 | |
2418 | rcu_read_lock(); | |
22061d80 | 2419 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2420 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2421 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2422 | break; |
2423 | } | |
2424 | } | |
2425 | rcu_read_unlock(); | |
2426 | ||
98e399f8 | 2427 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2428 | |
f6a78bfc HX |
2429 | return segs; |
2430 | } | |
05e8ef4a PS |
2431 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2432 | ||
2433 | ||
2434 | /* openvswitch calls this on rx path, so we need a different check. | |
2435 | */ | |
2436 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2437 | { | |
2438 | if (tx_path) | |
2439 | return skb->ip_summed != CHECKSUM_PARTIAL; | |
2440 | else | |
2441 | return skb->ip_summed == CHECKSUM_NONE; | |
2442 | } | |
2443 | ||
2444 | /** | |
2445 | * __skb_gso_segment - Perform segmentation on skb. | |
2446 | * @skb: buffer to segment | |
2447 | * @features: features for the output path (see dev->features) | |
2448 | * @tx_path: whether it is called in TX path | |
2449 | * | |
2450 | * This function segments the given skb and returns a list of segments. | |
2451 | * | |
2452 | * It may return NULL if the skb requires no segmentation. This is | |
2453 | * only possible when GSO is used for verifying header integrity. | |
2454 | */ | |
2455 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2456 | netdev_features_t features, bool tx_path) | |
2457 | { | |
2458 | if (unlikely(skb_needs_check(skb, tx_path))) { | |
2459 | int err; | |
2460 | ||
2461 | skb_warn_bad_offload(skb); | |
2462 | ||
a40e0a66 | 2463 | err = skb_cow_head(skb, 0); |
2464 | if (err < 0) | |
05e8ef4a PS |
2465 | return ERR_PTR(err); |
2466 | } | |
2467 | ||
68c33163 | 2468 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2469 | SKB_GSO_CB(skb)->encap_level = 0; |
2470 | ||
05e8ef4a PS |
2471 | skb_reset_mac_header(skb); |
2472 | skb_reset_mac_len(skb); | |
2473 | ||
2474 | return skb_mac_gso_segment(skb, features); | |
2475 | } | |
12b0004d | 2476 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2477 | |
fb286bb2 HX |
2478 | /* Take action when hardware reception checksum errors are detected. */ |
2479 | #ifdef CONFIG_BUG | |
2480 | void netdev_rx_csum_fault(struct net_device *dev) | |
2481 | { | |
2482 | if (net_ratelimit()) { | |
7b6cd1ce | 2483 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2484 | dump_stack(); |
2485 | } | |
2486 | } | |
2487 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2488 | #endif | |
2489 | ||
1da177e4 LT |
2490 | /* Actually, we should eliminate this check as soon as we know, that: |
2491 | * 1. IOMMU is present and allows to map all the memory. | |
2492 | * 2. No high memory really exists on this machine. | |
2493 | */ | |
2494 | ||
c1e756bf | 2495 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2496 | { |
3d3a8533 | 2497 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2498 | int i; |
5acbbd42 | 2499 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2500 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2501 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2502 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2503 | return 1; |
ea2ab693 | 2504 | } |
5acbbd42 | 2505 | } |
1da177e4 | 2506 | |
5acbbd42 FT |
2507 | if (PCI_DMA_BUS_IS_PHYS) { |
2508 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2509 | |
9092c658 ED |
2510 | if (!pdev) |
2511 | return 0; | |
5acbbd42 | 2512 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2513 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2514 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2515 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2516 | return 1; | |
2517 | } | |
2518 | } | |
3d3a8533 | 2519 | #endif |
1da177e4 LT |
2520 | return 0; |
2521 | } | |
1da177e4 | 2522 | |
3b392ddb SH |
2523 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2524 | * instead of standard features for the netdev. | |
2525 | */ | |
d0edc7bf | 2526 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
2527 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
2528 | netdev_features_t features, | |
2529 | __be16 type) | |
2530 | { | |
25cd9ba0 | 2531 | if (eth_p_mpls(type)) |
3b392ddb SH |
2532 | features &= skb->dev->mpls_features; |
2533 | ||
2534 | return features; | |
2535 | } | |
2536 | #else | |
2537 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2538 | netdev_features_t features, | |
2539 | __be16 type) | |
2540 | { | |
2541 | return features; | |
2542 | } | |
2543 | #endif | |
2544 | ||
c8f44aff | 2545 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2546 | netdev_features_t features) |
f01a5236 | 2547 | { |
53d6471c | 2548 | int tmp; |
3b392ddb SH |
2549 | __be16 type; |
2550 | ||
2551 | type = skb_network_protocol(skb, &tmp); | |
2552 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2553 | |
c0d680e5 | 2554 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2555 | !can_checksum_protocol(features, type)) { |
f01a5236 | 2556 | features &= ~NETIF_F_ALL_CSUM; |
c1e756bf | 2557 | } else if (illegal_highdma(skb->dev, skb)) { |
f01a5236 JG |
2558 | features &= ~NETIF_F_SG; |
2559 | } | |
2560 | ||
2561 | return features; | |
2562 | } | |
2563 | ||
c1e756bf | 2564 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 2565 | { |
5f35227e | 2566 | struct net_device *dev = skb->dev; |
fcbeb976 ED |
2567 | netdev_features_t features = dev->features; |
2568 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
58e998c6 JG |
2569 | __be16 protocol = skb->protocol; |
2570 | ||
fcbeb976 | 2571 | if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs) |
30b678d8 BH |
2572 | features &= ~NETIF_F_GSO_MASK; |
2573 | ||
5f35227e JG |
2574 | /* If encapsulation offload request, verify we are testing |
2575 | * hardware encapsulation features instead of standard | |
2576 | * features for the netdev | |
2577 | */ | |
2578 | if (skb->encapsulation) | |
2579 | features &= dev->hw_enc_features; | |
2580 | ||
df8a39de | 2581 | if (!skb_vlan_tag_present(skb)) { |
796f2da8 TM |
2582 | if (unlikely(protocol == htons(ETH_P_8021Q) || |
2583 | protocol == htons(ETH_P_8021AD))) { | |
2584 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
2585 | protocol = veh->h_vlan_encapsulated_proto; | |
2586 | } else { | |
5f35227e | 2587 | goto finalize; |
796f2da8 | 2588 | } |
f01a5236 | 2589 | } |
58e998c6 | 2590 | |
db115037 | 2591 | features = netdev_intersect_features(features, |
fcbeb976 | 2592 | dev->vlan_features | |
db115037 MK |
2593 | NETIF_F_HW_VLAN_CTAG_TX | |
2594 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2595 | |
cdbaa0bb | 2596 | if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) |
db115037 MK |
2597 | features = netdev_intersect_features(features, |
2598 | NETIF_F_SG | | |
2599 | NETIF_F_HIGHDMA | | |
2600 | NETIF_F_FRAGLIST | | |
2601 | NETIF_F_GEN_CSUM | | |
2602 | NETIF_F_HW_VLAN_CTAG_TX | | |
2603 | NETIF_F_HW_VLAN_STAG_TX); | |
cdbaa0bb | 2604 | |
5f35227e JG |
2605 | finalize: |
2606 | if (dev->netdev_ops->ndo_features_check) | |
2607 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
2608 | features); | |
2609 | ||
c1e756bf | 2610 | return harmonize_features(skb, features); |
58e998c6 | 2611 | } |
c1e756bf | 2612 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2613 | |
2ea25513 | 2614 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2615 | struct netdev_queue *txq, bool more) |
f6a78bfc | 2616 | { |
2ea25513 DM |
2617 | unsigned int len; |
2618 | int rc; | |
00829823 | 2619 | |
2ea25513 DM |
2620 | if (!list_empty(&ptype_all)) |
2621 | dev_queue_xmit_nit(skb, dev); | |
fc741216 | 2622 | |
2ea25513 DM |
2623 | len = skb->len; |
2624 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 2625 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 2626 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 2627 | |
2ea25513 DM |
2628 | return rc; |
2629 | } | |
7b9c6090 | 2630 | |
8dcda22a DM |
2631 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
2632 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
2633 | { |
2634 | struct sk_buff *skb = first; | |
2635 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 2636 | |
7f2e870f DM |
2637 | while (skb) { |
2638 | struct sk_buff *next = skb->next; | |
fc70fb64 | 2639 | |
7f2e870f | 2640 | skb->next = NULL; |
95f6b3dd | 2641 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
2642 | if (unlikely(!dev_xmit_complete(rc))) { |
2643 | skb->next = next; | |
2644 | goto out; | |
2645 | } | |
6afff0ca | 2646 | |
7f2e870f DM |
2647 | skb = next; |
2648 | if (netif_xmit_stopped(txq) && skb) { | |
2649 | rc = NETDEV_TX_BUSY; | |
2650 | break; | |
9ccb8975 | 2651 | } |
7f2e870f | 2652 | } |
9ccb8975 | 2653 | |
7f2e870f DM |
2654 | out: |
2655 | *ret = rc; | |
2656 | return skb; | |
2657 | } | |
b40863c6 | 2658 | |
1ff0dc94 ED |
2659 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
2660 | netdev_features_t features) | |
f6a78bfc | 2661 | { |
df8a39de | 2662 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
2663 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
2664 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
2665 | return skb; |
2666 | } | |
f6a78bfc | 2667 | |
55a93b3e | 2668 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
2669 | { |
2670 | netdev_features_t features; | |
f6a78bfc | 2671 | |
eae3f88e DM |
2672 | if (skb->next) |
2673 | return skb; | |
068a2de5 | 2674 | |
eae3f88e DM |
2675 | features = netif_skb_features(skb); |
2676 | skb = validate_xmit_vlan(skb, features); | |
2677 | if (unlikely(!skb)) | |
2678 | goto out_null; | |
7b9c6090 | 2679 | |
04ffcb25 | 2680 | if (netif_needs_gso(dev, skb, features)) { |
ce93718f DM |
2681 | struct sk_buff *segs; |
2682 | ||
2683 | segs = skb_gso_segment(skb, features); | |
cecda693 | 2684 | if (IS_ERR(segs)) { |
af6dabc9 | 2685 | goto out_kfree_skb; |
cecda693 JW |
2686 | } else if (segs) { |
2687 | consume_skb(skb); | |
2688 | skb = segs; | |
f6a78bfc | 2689 | } |
eae3f88e DM |
2690 | } else { |
2691 | if (skb_needs_linearize(skb, features) && | |
2692 | __skb_linearize(skb)) | |
2693 | goto out_kfree_skb; | |
4ec93edb | 2694 | |
eae3f88e DM |
2695 | /* If packet is not checksummed and device does not |
2696 | * support checksumming for this protocol, complete | |
2697 | * checksumming here. | |
2698 | */ | |
2699 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
2700 | if (skb->encapsulation) | |
2701 | skb_set_inner_transport_header(skb, | |
2702 | skb_checksum_start_offset(skb)); | |
2703 | else | |
2704 | skb_set_transport_header(skb, | |
2705 | skb_checksum_start_offset(skb)); | |
2706 | if (!(features & NETIF_F_ALL_CSUM) && | |
2707 | skb_checksum_help(skb)) | |
2708 | goto out_kfree_skb; | |
7b9c6090 | 2709 | } |
0c772159 | 2710 | } |
7b9c6090 | 2711 | |
eae3f88e | 2712 | return skb; |
fc70fb64 | 2713 | |
f6a78bfc HX |
2714 | out_kfree_skb: |
2715 | kfree_skb(skb); | |
eae3f88e DM |
2716 | out_null: |
2717 | return NULL; | |
2718 | } | |
6afff0ca | 2719 | |
55a93b3e ED |
2720 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
2721 | { | |
2722 | struct sk_buff *next, *head = NULL, *tail; | |
2723 | ||
bec3cfdc | 2724 | for (; skb != NULL; skb = next) { |
55a93b3e ED |
2725 | next = skb->next; |
2726 | skb->next = NULL; | |
bec3cfdc ED |
2727 | |
2728 | /* in case skb wont be segmented, point to itself */ | |
2729 | skb->prev = skb; | |
2730 | ||
55a93b3e | 2731 | skb = validate_xmit_skb(skb, dev); |
bec3cfdc ED |
2732 | if (!skb) |
2733 | continue; | |
55a93b3e | 2734 | |
bec3cfdc ED |
2735 | if (!head) |
2736 | head = skb; | |
2737 | else | |
2738 | tail->next = skb; | |
2739 | /* If skb was segmented, skb->prev points to | |
2740 | * the last segment. If not, it still contains skb. | |
2741 | */ | |
2742 | tail = skb->prev; | |
55a93b3e ED |
2743 | } |
2744 | return head; | |
f6a78bfc HX |
2745 | } |
2746 | ||
1def9238 ED |
2747 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
2748 | { | |
2749 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2750 | ||
2751 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2752 | ||
2753 | /* To get more precise estimation of bytes sent on wire, | |
2754 | * we add to pkt_len the headers size of all segments | |
2755 | */ | |
2756 | if (shinfo->gso_size) { | |
757b8b1d | 2757 | unsigned int hdr_len; |
15e5a030 | 2758 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 2759 | |
757b8b1d ED |
2760 | /* mac layer + network layer */ |
2761 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
2762 | ||
2763 | /* + transport layer */ | |
1def9238 ED |
2764 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
2765 | hdr_len += tcp_hdrlen(skb); | |
2766 | else | |
2767 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
2768 | |
2769 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
2770 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
2771 | shinfo->gso_size); | |
2772 | ||
2773 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
2774 | } |
2775 | } | |
2776 | ||
bbd8a0d3 KK |
2777 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2778 | struct net_device *dev, | |
2779 | struct netdev_queue *txq) | |
2780 | { | |
2781 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2782 | bool contended; |
bbd8a0d3 KK |
2783 | int rc; |
2784 | ||
1def9238 | 2785 | qdisc_pkt_len_init(skb); |
a2da570d | 2786 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
2787 | /* |
2788 | * Heuristic to force contended enqueues to serialize on a | |
2789 | * separate lock before trying to get qdisc main lock. | |
9bf2b8c2 YX |
2790 | * This permits __QDISC___STATE_RUNNING owner to get the lock more |
2791 | * often and dequeue packets faster. | |
79640a4c | 2792 | */ |
a2da570d | 2793 | contended = qdisc_is_running(q); |
79640a4c ED |
2794 | if (unlikely(contended)) |
2795 | spin_lock(&q->busylock); | |
2796 | ||
bbd8a0d3 KK |
2797 | spin_lock(root_lock); |
2798 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2799 | kfree_skb(skb); | |
2800 | rc = NET_XMIT_DROP; | |
2801 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2802 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2803 | /* |
2804 | * This is a work-conserving queue; there are no old skbs | |
2805 | * waiting to be sent out; and the qdisc is not running - | |
2806 | * xmit the skb directly. | |
2807 | */ | |
bfe0d029 | 2808 | |
bfe0d029 ED |
2809 | qdisc_bstats_update(q, skb); |
2810 | ||
55a93b3e | 2811 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
2812 | if (unlikely(contended)) { |
2813 | spin_unlock(&q->busylock); | |
2814 | contended = false; | |
2815 | } | |
bbd8a0d3 | 2816 | __qdisc_run(q); |
79640a4c | 2817 | } else |
bc135b23 | 2818 | qdisc_run_end(q); |
bbd8a0d3 KK |
2819 | |
2820 | rc = NET_XMIT_SUCCESS; | |
2821 | } else { | |
a2da570d | 2822 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2823 | if (qdisc_run_begin(q)) { |
2824 | if (unlikely(contended)) { | |
2825 | spin_unlock(&q->busylock); | |
2826 | contended = false; | |
2827 | } | |
2828 | __qdisc_run(q); | |
2829 | } | |
bbd8a0d3 KK |
2830 | } |
2831 | spin_unlock(root_lock); | |
79640a4c ED |
2832 | if (unlikely(contended)) |
2833 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2834 | return rc; |
2835 | } | |
2836 | ||
86f8515f | 2837 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
2838 | static void skb_update_prio(struct sk_buff *skb) |
2839 | { | |
6977a79d | 2840 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 2841 | |
91c68ce2 ED |
2842 | if (!skb->priority && skb->sk && map) { |
2843 | unsigned int prioidx = skb->sk->sk_cgrp_prioidx; | |
2844 | ||
2845 | if (prioidx < map->priomap_len) | |
2846 | skb->priority = map->priomap[prioidx]; | |
2847 | } | |
5bc1421e NH |
2848 | } |
2849 | #else | |
2850 | #define skb_update_prio(skb) | |
2851 | #endif | |
2852 | ||
745e20f1 | 2853 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2854 | #define RECURSION_LIMIT 10 |
745e20f1 | 2855 | |
95603e22 MM |
2856 | /** |
2857 | * dev_loopback_xmit - loop back @skb | |
2858 | * @skb: buffer to transmit | |
2859 | */ | |
2860 | int dev_loopback_xmit(struct sk_buff *skb) | |
2861 | { | |
2862 | skb_reset_mac_header(skb); | |
2863 | __skb_pull(skb, skb_network_offset(skb)); | |
2864 | skb->pkt_type = PACKET_LOOPBACK; | |
2865 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2866 | WARN_ON(!skb_dst(skb)); | |
2867 | skb_dst_force(skb); | |
2868 | netif_rx_ni(skb); | |
2869 | return 0; | |
2870 | } | |
2871 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2872 | ||
d29f749e | 2873 | /** |
9d08dd3d | 2874 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 2875 | * @skb: buffer to transmit |
9d08dd3d | 2876 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
2877 | * |
2878 | * Queue a buffer for transmission to a network device. The caller must | |
2879 | * have set the device and priority and built the buffer before calling | |
2880 | * this function. The function can be called from an interrupt. | |
2881 | * | |
2882 | * A negative errno code is returned on a failure. A success does not | |
2883 | * guarantee the frame will be transmitted as it may be dropped due | |
2884 | * to congestion or traffic shaping. | |
2885 | * | |
2886 | * ----------------------------------------------------------------------------------- | |
2887 | * I notice this method can also return errors from the queue disciplines, | |
2888 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2889 | * be positive. | |
2890 | * | |
2891 | * Regardless of the return value, the skb is consumed, so it is currently | |
2892 | * difficult to retry a send to this method. (You can bump the ref count | |
2893 | * before sending to hold a reference for retry if you are careful.) | |
2894 | * | |
2895 | * When calling this method, interrupts MUST be enabled. This is because | |
2896 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2897 | * --BLG | |
2898 | */ | |
0a59f3a9 | 2899 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
2900 | { |
2901 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2902 | struct netdev_queue *txq; |
1da177e4 LT |
2903 | struct Qdisc *q; |
2904 | int rc = -ENOMEM; | |
2905 | ||
6d1ccff6 ED |
2906 | skb_reset_mac_header(skb); |
2907 | ||
e7fd2885 WB |
2908 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
2909 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
2910 | ||
4ec93edb YH |
2911 | /* Disable soft irqs for various locks below. Also |
2912 | * stops preemption for RCU. | |
1da177e4 | 2913 | */ |
4ec93edb | 2914 | rcu_read_lock_bh(); |
1da177e4 | 2915 | |
5bc1421e NH |
2916 | skb_update_prio(skb); |
2917 | ||
02875878 ED |
2918 | /* If device/qdisc don't need skb->dst, release it right now while |
2919 | * its hot in this cpu cache. | |
2920 | */ | |
2921 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2922 | skb_dst_drop(skb); | |
2923 | else | |
2924 | skb_dst_force(skb); | |
2925 | ||
f663dd9a | 2926 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 2927 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2928 | |
1da177e4 | 2929 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2930 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2931 | #endif |
cf66ba58 | 2932 | trace_net_dev_queue(skb); |
1da177e4 | 2933 | if (q->enqueue) { |
bbd8a0d3 | 2934 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2935 | goto out; |
1da177e4 LT |
2936 | } |
2937 | ||
2938 | /* The device has no queue. Common case for software devices: | |
2939 | loopback, all the sorts of tunnels... | |
2940 | ||
932ff279 HX |
2941 | Really, it is unlikely that netif_tx_lock protection is necessary |
2942 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2943 | counters.) |
2944 | However, it is possible, that they rely on protection | |
2945 | made by us here. | |
2946 | ||
2947 | Check this and shot the lock. It is not prone from deadlocks. | |
2948 | Either shot noqueue qdisc, it is even simpler 8) | |
2949 | */ | |
2950 | if (dev->flags & IFF_UP) { | |
2951 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2952 | ||
c773e847 | 2953 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2954 | |
745e20f1 ED |
2955 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2956 | goto recursion_alert; | |
2957 | ||
1f59533f JDB |
2958 | skb = validate_xmit_skb(skb, dev); |
2959 | if (!skb) | |
2960 | goto drop; | |
2961 | ||
c773e847 | 2962 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2963 | |
73466498 | 2964 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 2965 | __this_cpu_inc(xmit_recursion); |
ce93718f | 2966 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 2967 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2968 | if (dev_xmit_complete(rc)) { |
c773e847 | 2969 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2970 | goto out; |
2971 | } | |
2972 | } | |
c773e847 | 2973 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
2974 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
2975 | dev->name); | |
1da177e4 LT |
2976 | } else { |
2977 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2978 | * unfortunately |
2979 | */ | |
2980 | recursion_alert: | |
e87cc472 JP |
2981 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
2982 | dev->name); | |
1da177e4 LT |
2983 | } |
2984 | } | |
2985 | ||
2986 | rc = -ENETDOWN; | |
1f59533f | 2987 | drop: |
d4828d85 | 2988 | rcu_read_unlock_bh(); |
1da177e4 | 2989 | |
015f0688 | 2990 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 2991 | kfree_skb_list(skb); |
1da177e4 LT |
2992 | return rc; |
2993 | out: | |
d4828d85 | 2994 | rcu_read_unlock_bh(); |
1da177e4 LT |
2995 | return rc; |
2996 | } | |
f663dd9a JW |
2997 | |
2998 | int dev_queue_xmit(struct sk_buff *skb) | |
2999 | { | |
3000 | return __dev_queue_xmit(skb, NULL); | |
3001 | } | |
d1b19dff | 3002 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3003 | |
f663dd9a JW |
3004 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3005 | { | |
3006 | return __dev_queue_xmit(skb, accel_priv); | |
3007 | } | |
3008 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3009 | ||
1da177e4 LT |
3010 | |
3011 | /*======================================================================= | |
3012 | Receiver routines | |
3013 | =======================================================================*/ | |
3014 | ||
6b2bedc3 | 3015 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3016 | EXPORT_SYMBOL(netdev_max_backlog); |
3017 | ||
3b098e2d | 3018 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
3019 | int netdev_budget __read_mostly = 300; |
3020 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 3021 | |
eecfd7c4 ED |
3022 | /* Called with irq disabled */ |
3023 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3024 | struct napi_struct *napi) | |
3025 | { | |
3026 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3027 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3028 | } | |
3029 | ||
bfb564e7 KK |
3030 | #ifdef CONFIG_RPS |
3031 | ||
3032 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3033 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
3034 | EXPORT_SYMBOL(rps_sock_flow_table); |
3035 | ||
c5905afb | 3036 | struct static_key rps_needed __read_mostly; |
adc9300e | 3037 | |
c445477d BH |
3038 | static struct rps_dev_flow * |
3039 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3040 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3041 | { | |
09994d1b | 3042 | if (next_cpu != RPS_NO_CPU) { |
c445477d BH |
3043 | #ifdef CONFIG_RFS_ACCEL |
3044 | struct netdev_rx_queue *rxqueue; | |
3045 | struct rps_dev_flow_table *flow_table; | |
3046 | struct rps_dev_flow *old_rflow; | |
3047 | u32 flow_id; | |
3048 | u16 rxq_index; | |
3049 | int rc; | |
3050 | ||
3051 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3052 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3053 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3054 | goto out; |
3055 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3056 | if (rxq_index == skb_get_rx_queue(skb)) | |
3057 | goto out; | |
3058 | ||
3059 | rxqueue = dev->_rx + rxq_index; | |
3060 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3061 | if (!flow_table) | |
3062 | goto out; | |
61b905da | 3063 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3064 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3065 | rxq_index, flow_id); | |
3066 | if (rc < 0) | |
3067 | goto out; | |
3068 | old_rflow = rflow; | |
3069 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3070 | rflow->filter = rc; |
3071 | if (old_rflow->filter == rflow->filter) | |
3072 | old_rflow->filter = RPS_NO_FILTER; | |
3073 | out: | |
3074 | #endif | |
3075 | rflow->last_qtail = | |
09994d1b | 3076 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3077 | } |
3078 | ||
09994d1b | 3079 | rflow->cpu = next_cpu; |
c445477d BH |
3080 | return rflow; |
3081 | } | |
3082 | ||
bfb564e7 KK |
3083 | /* |
3084 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3085 | * CPU from the RPS map of the receiving queue for a given skb. | |
3086 | * rcu_read_lock must be held on entry. | |
3087 | */ | |
3088 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3089 | struct rps_dev_flow **rflowp) | |
3090 | { | |
3091 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 3092 | struct rps_map *map; |
bfb564e7 KK |
3093 | struct rps_dev_flow_table *flow_table; |
3094 | struct rps_sock_flow_table *sock_flow_table; | |
3095 | int cpu = -1; | |
3096 | u16 tcpu; | |
61b905da | 3097 | u32 hash; |
bfb564e7 KK |
3098 | |
3099 | if (skb_rx_queue_recorded(skb)) { | |
3100 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
3101 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3102 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3103 | "%s received packet on queue %u, but number " | |
3104 | "of RX queues is %u\n", | |
3105 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3106 | goto done; |
3107 | } | |
3108 | rxqueue = dev->_rx + index; | |
3109 | } else | |
3110 | rxqueue = dev->_rx; | |
3111 | ||
6e3f7faf ED |
3112 | map = rcu_dereference(rxqueue->rps_map); |
3113 | if (map) { | |
85875236 | 3114 | if (map->len == 1 && |
33d480ce | 3115 | !rcu_access_pointer(rxqueue->rps_flow_table)) { |
6febfca9 CG |
3116 | tcpu = map->cpus[0]; |
3117 | if (cpu_online(tcpu)) | |
3118 | cpu = tcpu; | |
3119 | goto done; | |
3120 | } | |
33d480ce | 3121 | } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) { |
bfb564e7 | 3122 | goto done; |
6febfca9 | 3123 | } |
bfb564e7 | 3124 | |
2d47b459 | 3125 | skb_reset_network_header(skb); |
61b905da TH |
3126 | hash = skb_get_hash(skb); |
3127 | if (!hash) | |
bfb564e7 KK |
3128 | goto done; |
3129 | ||
fec5e652 TH |
3130 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
3131 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
3132 | if (flow_table && sock_flow_table) { | |
3133 | u16 next_cpu; | |
3134 | struct rps_dev_flow *rflow; | |
3135 | ||
61b905da | 3136 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3137 | tcpu = rflow->cpu; |
3138 | ||
61b905da | 3139 | next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask]; |
fec5e652 TH |
3140 | |
3141 | /* | |
3142 | * If the desired CPU (where last recvmsg was done) is | |
3143 | * different from current CPU (one in the rx-queue flow | |
3144 | * table entry), switch if one of the following holds: | |
3145 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
3146 | * - Current CPU is offline. | |
3147 | * - The current CPU's queue tail has advanced beyond the | |
3148 | * last packet that was enqueued using this table entry. | |
3149 | * This guarantees that all previous packets for the flow | |
3150 | * have been dequeued, thus preserving in order delivery. | |
3151 | */ | |
3152 | if (unlikely(tcpu != next_cpu) && | |
3153 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
3154 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
baefa31d TH |
3155 | rflow->last_qtail)) >= 0)) { |
3156 | tcpu = next_cpu; | |
c445477d | 3157 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3158 | } |
c445477d | 3159 | |
fec5e652 TH |
3160 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
3161 | *rflowp = rflow; | |
3162 | cpu = tcpu; | |
3163 | goto done; | |
3164 | } | |
3165 | } | |
3166 | ||
0a9627f2 | 3167 | if (map) { |
8fc54f68 | 3168 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3169 | if (cpu_online(tcpu)) { |
3170 | cpu = tcpu; | |
3171 | goto done; | |
3172 | } | |
3173 | } | |
3174 | ||
3175 | done: | |
0a9627f2 TH |
3176 | return cpu; |
3177 | } | |
3178 | ||
c445477d BH |
3179 | #ifdef CONFIG_RFS_ACCEL |
3180 | ||
3181 | /** | |
3182 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3183 | * @dev: Device on which the filter was set | |
3184 | * @rxq_index: RX queue index | |
3185 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3186 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3187 | * | |
3188 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3189 | * this function for each installed filter and remove the filters for | |
3190 | * which it returns %true. | |
3191 | */ | |
3192 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3193 | u32 flow_id, u16 filter_id) | |
3194 | { | |
3195 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3196 | struct rps_dev_flow_table *flow_table; | |
3197 | struct rps_dev_flow *rflow; | |
3198 | bool expire = true; | |
3199 | int cpu; | |
3200 | ||
3201 | rcu_read_lock(); | |
3202 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3203 | if (flow_table && flow_id <= flow_table->mask) { | |
3204 | rflow = &flow_table->flows[flow_id]; | |
3205 | cpu = ACCESS_ONCE(rflow->cpu); | |
3206 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
3207 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
3208 | rflow->last_qtail) < | |
3209 | (int)(10 * flow_table->mask))) | |
3210 | expire = false; | |
3211 | } | |
3212 | rcu_read_unlock(); | |
3213 | return expire; | |
3214 | } | |
3215 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3216 | ||
3217 | #endif /* CONFIG_RFS_ACCEL */ | |
3218 | ||
0a9627f2 | 3219 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3220 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3221 | { |
e36fa2f7 ED |
3222 | struct softnet_data *sd = data; |
3223 | ||
eecfd7c4 | 3224 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3225 | sd->received_rps++; |
0a9627f2 | 3226 | } |
e36fa2f7 | 3227 | |
fec5e652 | 3228 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3229 | |
e36fa2f7 ED |
3230 | /* |
3231 | * Check if this softnet_data structure is another cpu one | |
3232 | * If yes, queue it to our IPI list and return 1 | |
3233 | * If no, return 0 | |
3234 | */ | |
3235 | static int rps_ipi_queued(struct softnet_data *sd) | |
3236 | { | |
3237 | #ifdef CONFIG_RPS | |
903ceff7 | 3238 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
3239 | |
3240 | if (sd != mysd) { | |
3241 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3242 | mysd->rps_ipi_list = sd; | |
3243 | ||
3244 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3245 | return 1; | |
3246 | } | |
3247 | #endif /* CONFIG_RPS */ | |
3248 | return 0; | |
3249 | } | |
3250 | ||
99bbc707 WB |
3251 | #ifdef CONFIG_NET_FLOW_LIMIT |
3252 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3253 | #endif | |
3254 | ||
3255 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3256 | { | |
3257 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3258 | struct sd_flow_limit *fl; | |
3259 | struct softnet_data *sd; | |
3260 | unsigned int old_flow, new_flow; | |
3261 | ||
3262 | if (qlen < (netdev_max_backlog >> 1)) | |
3263 | return false; | |
3264 | ||
903ceff7 | 3265 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
3266 | |
3267 | rcu_read_lock(); | |
3268 | fl = rcu_dereference(sd->flow_limit); | |
3269 | if (fl) { | |
3958afa1 | 3270 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3271 | old_flow = fl->history[fl->history_head]; |
3272 | fl->history[fl->history_head] = new_flow; | |
3273 | ||
3274 | fl->history_head++; | |
3275 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3276 | ||
3277 | if (likely(fl->buckets[old_flow])) | |
3278 | fl->buckets[old_flow]--; | |
3279 | ||
3280 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3281 | fl->count++; | |
3282 | rcu_read_unlock(); | |
3283 | return true; | |
3284 | } | |
3285 | } | |
3286 | rcu_read_unlock(); | |
3287 | #endif | |
3288 | return false; | |
3289 | } | |
3290 | ||
0a9627f2 TH |
3291 | /* |
3292 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3293 | * queue (may be a remote CPU queue). | |
3294 | */ | |
fec5e652 TH |
3295 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3296 | unsigned int *qtail) | |
0a9627f2 | 3297 | { |
e36fa2f7 | 3298 | struct softnet_data *sd; |
0a9627f2 | 3299 | unsigned long flags; |
99bbc707 | 3300 | unsigned int qlen; |
0a9627f2 | 3301 | |
e36fa2f7 | 3302 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3303 | |
3304 | local_irq_save(flags); | |
0a9627f2 | 3305 | |
e36fa2f7 | 3306 | rps_lock(sd); |
99bbc707 WB |
3307 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3308 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 3309 | if (qlen) { |
0a9627f2 | 3310 | enqueue: |
e36fa2f7 | 3311 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3312 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3313 | rps_unlock(sd); |
152102c7 | 3314 | local_irq_restore(flags); |
0a9627f2 TH |
3315 | return NET_RX_SUCCESS; |
3316 | } | |
3317 | ||
ebda37c2 ED |
3318 | /* Schedule NAPI for backlog device |
3319 | * We can use non atomic operation since we own the queue lock | |
3320 | */ | |
3321 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3322 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3323 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3324 | } |
3325 | goto enqueue; | |
3326 | } | |
3327 | ||
dee42870 | 3328 | sd->dropped++; |
e36fa2f7 | 3329 | rps_unlock(sd); |
0a9627f2 | 3330 | |
0a9627f2 TH |
3331 | local_irq_restore(flags); |
3332 | ||
caf586e5 | 3333 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3334 | kfree_skb(skb); |
3335 | return NET_RX_DROP; | |
3336 | } | |
1da177e4 | 3337 | |
ae78dbfa | 3338 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3339 | { |
b0e28f1e | 3340 | int ret; |
1da177e4 | 3341 | |
588f0330 | 3342 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3343 | |
cf66ba58 | 3344 | trace_netif_rx(skb); |
df334545 | 3345 | #ifdef CONFIG_RPS |
c5905afb | 3346 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3347 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3348 | int cpu; |
3349 | ||
cece1945 | 3350 | preempt_disable(); |
b0e28f1e | 3351 | rcu_read_lock(); |
fec5e652 TH |
3352 | |
3353 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3354 | if (cpu < 0) |
3355 | cpu = smp_processor_id(); | |
fec5e652 TH |
3356 | |
3357 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3358 | ||
b0e28f1e | 3359 | rcu_read_unlock(); |
cece1945 | 3360 | preempt_enable(); |
adc9300e ED |
3361 | } else |
3362 | #endif | |
fec5e652 TH |
3363 | { |
3364 | unsigned int qtail; | |
3365 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3366 | put_cpu(); | |
3367 | } | |
b0e28f1e | 3368 | return ret; |
1da177e4 | 3369 | } |
ae78dbfa BH |
3370 | |
3371 | /** | |
3372 | * netif_rx - post buffer to the network code | |
3373 | * @skb: buffer to post | |
3374 | * | |
3375 | * This function receives a packet from a device driver and queues it for | |
3376 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3377 | * may be dropped during processing for congestion control or by the | |
3378 | * protocol layers. | |
3379 | * | |
3380 | * return values: | |
3381 | * NET_RX_SUCCESS (no congestion) | |
3382 | * NET_RX_DROP (packet was dropped) | |
3383 | * | |
3384 | */ | |
3385 | ||
3386 | int netif_rx(struct sk_buff *skb) | |
3387 | { | |
3388 | trace_netif_rx_entry(skb); | |
3389 | ||
3390 | return netif_rx_internal(skb); | |
3391 | } | |
d1b19dff | 3392 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3393 | |
3394 | int netif_rx_ni(struct sk_buff *skb) | |
3395 | { | |
3396 | int err; | |
3397 | ||
ae78dbfa BH |
3398 | trace_netif_rx_ni_entry(skb); |
3399 | ||
1da177e4 | 3400 | preempt_disable(); |
ae78dbfa | 3401 | err = netif_rx_internal(skb); |
1da177e4 LT |
3402 | if (local_softirq_pending()) |
3403 | do_softirq(); | |
3404 | preempt_enable(); | |
3405 | ||
3406 | return err; | |
3407 | } | |
1da177e4 LT |
3408 | EXPORT_SYMBOL(netif_rx_ni); |
3409 | ||
1da177e4 LT |
3410 | static void net_tx_action(struct softirq_action *h) |
3411 | { | |
903ceff7 | 3412 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
3413 | |
3414 | if (sd->completion_queue) { | |
3415 | struct sk_buff *clist; | |
3416 | ||
3417 | local_irq_disable(); | |
3418 | clist = sd->completion_queue; | |
3419 | sd->completion_queue = NULL; | |
3420 | local_irq_enable(); | |
3421 | ||
3422 | while (clist) { | |
3423 | struct sk_buff *skb = clist; | |
3424 | clist = clist->next; | |
3425 | ||
547b792c | 3426 | WARN_ON(atomic_read(&skb->users)); |
e6247027 ED |
3427 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
3428 | trace_consume_skb(skb); | |
3429 | else | |
3430 | trace_kfree_skb(skb, net_tx_action); | |
1da177e4 LT |
3431 | __kfree_skb(skb); |
3432 | } | |
3433 | } | |
3434 | ||
3435 | if (sd->output_queue) { | |
37437bb2 | 3436 | struct Qdisc *head; |
1da177e4 LT |
3437 | |
3438 | local_irq_disable(); | |
3439 | head = sd->output_queue; | |
3440 | sd->output_queue = NULL; | |
a9cbd588 | 3441 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3442 | local_irq_enable(); |
3443 | ||
3444 | while (head) { | |
37437bb2 DM |
3445 | struct Qdisc *q = head; |
3446 | spinlock_t *root_lock; | |
3447 | ||
1da177e4 LT |
3448 | head = head->next_sched; |
3449 | ||
5fb66229 | 3450 | root_lock = qdisc_lock(q); |
37437bb2 | 3451 | if (spin_trylock(root_lock)) { |
4e857c58 | 3452 | smp_mb__before_atomic(); |
def82a1d JP |
3453 | clear_bit(__QDISC_STATE_SCHED, |
3454 | &q->state); | |
37437bb2 DM |
3455 | qdisc_run(q); |
3456 | spin_unlock(root_lock); | |
1da177e4 | 3457 | } else { |
195648bb | 3458 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3459 | &q->state)) { |
195648bb | 3460 | __netif_reschedule(q); |
e8a83e10 | 3461 | } else { |
4e857c58 | 3462 | smp_mb__before_atomic(); |
e8a83e10 JP |
3463 | clear_bit(__QDISC_STATE_SCHED, |
3464 | &q->state); | |
3465 | } | |
1da177e4 LT |
3466 | } |
3467 | } | |
3468 | } | |
3469 | } | |
3470 | ||
ab95bfe0 JP |
3471 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3472 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3473 | /* This hook is defined here for ATM LANE */ |
3474 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3475 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3476 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3477 | #endif |
1da177e4 | 3478 | |
1da177e4 LT |
3479 | #ifdef CONFIG_NET_CLS_ACT |
3480 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3481 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3482 | * a compare and 2 stores extra right now if we dont have it on | |
3483 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3484 | * NOTE: This doesn't stop any functionality; if you dont have |
3485 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3486 | * |
3487 | */ | |
24824a09 | 3488 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3489 | { |
1da177e4 | 3490 | struct net_device *dev = skb->dev; |
f697c3e8 | 3491 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3492 | int result = TC_ACT_OK; |
3493 | struct Qdisc *q; | |
4ec93edb | 3494 | |
de384830 | 3495 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
e87cc472 JP |
3496 | net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n", |
3497 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3498 | return TC_ACT_SHOT; |
3499 | } | |
1da177e4 | 3500 | |
f697c3e8 HX |
3501 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3502 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3503 | |
46e5da40 | 3504 | q = rcu_dereference(rxq->qdisc); |
8d50b53d | 3505 | if (q != &noop_qdisc) { |
83874000 | 3506 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3507 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3508 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3509 | spin_unlock(qdisc_lock(q)); |
3510 | } | |
f697c3e8 HX |
3511 | |
3512 | return result; | |
3513 | } | |
86e65da9 | 3514 | |
f697c3e8 HX |
3515 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3516 | struct packet_type **pt_prev, | |
3517 | int *ret, struct net_device *orig_dev) | |
3518 | { | |
24824a09 ED |
3519 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3520 | ||
46e5da40 | 3521 | if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc) |
f697c3e8 | 3522 | goto out; |
1da177e4 | 3523 | |
f697c3e8 HX |
3524 | if (*pt_prev) { |
3525 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3526 | *pt_prev = NULL; | |
1da177e4 LT |
3527 | } |
3528 | ||
24824a09 | 3529 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3530 | case TC_ACT_SHOT: |
3531 | case TC_ACT_STOLEN: | |
3532 | kfree_skb(skb); | |
3533 | return NULL; | |
3534 | } | |
3535 | ||
3536 | out: | |
3537 | skb->tc_verd = 0; | |
3538 | return skb; | |
1da177e4 LT |
3539 | } |
3540 | #endif | |
3541 | ||
ab95bfe0 JP |
3542 | /** |
3543 | * netdev_rx_handler_register - register receive handler | |
3544 | * @dev: device to register a handler for | |
3545 | * @rx_handler: receive handler to register | |
93e2c32b | 3546 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 3547 | * |
e227867f | 3548 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
3549 | * called from __netif_receive_skb. A negative errno code is returned |
3550 | * on a failure. | |
3551 | * | |
3552 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3553 | * |
3554 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3555 | */ |
3556 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3557 | rx_handler_func_t *rx_handler, |
3558 | void *rx_handler_data) | |
ab95bfe0 JP |
3559 | { |
3560 | ASSERT_RTNL(); | |
3561 | ||
3562 | if (dev->rx_handler) | |
3563 | return -EBUSY; | |
3564 | ||
00cfec37 | 3565 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 3566 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3567 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3568 | ||
3569 | return 0; | |
3570 | } | |
3571 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3572 | ||
3573 | /** | |
3574 | * netdev_rx_handler_unregister - unregister receive handler | |
3575 | * @dev: device to unregister a handler from | |
3576 | * | |
166ec369 | 3577 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
3578 | * |
3579 | * The caller must hold the rtnl_mutex. | |
3580 | */ | |
3581 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3582 | { | |
3583 | ||
3584 | ASSERT_RTNL(); | |
a9b3cd7f | 3585 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
3586 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
3587 | * section has a guarantee to see a non NULL rx_handler_data | |
3588 | * as well. | |
3589 | */ | |
3590 | synchronize_net(); | |
a9b3cd7f | 3591 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
3592 | } |
3593 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3594 | ||
b4b9e355 MG |
3595 | /* |
3596 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3597 | * the special handling of PFMEMALLOC skbs. | |
3598 | */ | |
3599 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3600 | { | |
3601 | switch (skb->protocol) { | |
2b8837ae JP |
3602 | case htons(ETH_P_ARP): |
3603 | case htons(ETH_P_IP): | |
3604 | case htons(ETH_P_IPV6): | |
3605 | case htons(ETH_P_8021Q): | |
3606 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
3607 | return true; |
3608 | default: | |
3609 | return false; | |
3610 | } | |
3611 | } | |
3612 | ||
9754e293 | 3613 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
3614 | { |
3615 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3616 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3617 | struct net_device *orig_dev; |
63d8ea7f | 3618 | struct net_device *null_or_dev; |
8a4eb573 | 3619 | bool deliver_exact = false; |
1da177e4 | 3620 | int ret = NET_RX_DROP; |
252e3346 | 3621 | __be16 type; |
1da177e4 | 3622 | |
588f0330 | 3623 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3624 | |
cf66ba58 | 3625 | trace_netif_receive_skb(skb); |
9b22ea56 | 3626 | |
cc9bd5ce | 3627 | orig_dev = skb->dev; |
8f903c70 | 3628 | |
c1d2bbe1 | 3629 | skb_reset_network_header(skb); |
fda55eca ED |
3630 | if (!skb_transport_header_was_set(skb)) |
3631 | skb_reset_transport_header(skb); | |
0b5c9db1 | 3632 | skb_reset_mac_len(skb); |
1da177e4 LT |
3633 | |
3634 | pt_prev = NULL; | |
3635 | ||
3636 | rcu_read_lock(); | |
3637 | ||
63d8ea7f | 3638 | another_round: |
b6858177 | 3639 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
3640 | |
3641 | __this_cpu_inc(softnet_data.processed); | |
3642 | ||
8ad227ff PM |
3643 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
3644 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 3645 | skb = skb_vlan_untag(skb); |
bcc6d479 | 3646 | if (unlikely(!skb)) |
b4b9e355 | 3647 | goto unlock; |
bcc6d479 JP |
3648 | } |
3649 | ||
1da177e4 LT |
3650 | #ifdef CONFIG_NET_CLS_ACT |
3651 | if (skb->tc_verd & TC_NCLS) { | |
3652 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3653 | goto ncls; | |
3654 | } | |
3655 | #endif | |
3656 | ||
9754e293 | 3657 | if (pfmemalloc) |
b4b9e355 MG |
3658 | goto skip_taps; |
3659 | ||
1da177e4 | 3660 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
63d8ea7f | 3661 | if (!ptype->dev || ptype->dev == skb->dev) { |
4ec93edb | 3662 | if (pt_prev) |
f2ccd8fa | 3663 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3664 | pt_prev = ptype; |
3665 | } | |
3666 | } | |
3667 | ||
b4b9e355 | 3668 | skip_taps: |
1da177e4 | 3669 | #ifdef CONFIG_NET_CLS_ACT |
f697c3e8 HX |
3670 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3671 | if (!skb) | |
b4b9e355 | 3672 | goto unlock; |
1da177e4 LT |
3673 | ncls: |
3674 | #endif | |
3675 | ||
9754e293 | 3676 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
3677 | goto drop; |
3678 | ||
df8a39de | 3679 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
3680 | if (pt_prev) { |
3681 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3682 | pt_prev = NULL; | |
3683 | } | |
48cc32d3 | 3684 | if (vlan_do_receive(&skb)) |
2425717b JF |
3685 | goto another_round; |
3686 | else if (unlikely(!skb)) | |
b4b9e355 | 3687 | goto unlock; |
2425717b JF |
3688 | } |
3689 | ||
48cc32d3 | 3690 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
3691 | if (rx_handler) { |
3692 | if (pt_prev) { | |
3693 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3694 | pt_prev = NULL; | |
3695 | } | |
8a4eb573 JP |
3696 | switch (rx_handler(&skb)) { |
3697 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 3698 | ret = NET_RX_SUCCESS; |
b4b9e355 | 3699 | goto unlock; |
8a4eb573 | 3700 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3701 | goto another_round; |
8a4eb573 JP |
3702 | case RX_HANDLER_EXACT: |
3703 | deliver_exact = true; | |
3704 | case RX_HANDLER_PASS: | |
3705 | break; | |
3706 | default: | |
3707 | BUG(); | |
3708 | } | |
ab95bfe0 | 3709 | } |
1da177e4 | 3710 | |
df8a39de JP |
3711 | if (unlikely(skb_vlan_tag_present(skb))) { |
3712 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
3713 | skb->pkt_type = PACKET_OTHERHOST; |
3714 | /* Note: we might in the future use prio bits | |
3715 | * and set skb->priority like in vlan_do_receive() | |
3716 | * For the time being, just ignore Priority Code Point | |
3717 | */ | |
3718 | skb->vlan_tci = 0; | |
3719 | } | |
48cc32d3 | 3720 | |
63d8ea7f | 3721 | /* deliver only exact match when indicated */ |
8a4eb573 | 3722 | null_or_dev = deliver_exact ? skb->dev : NULL; |
1f3c8804 | 3723 | |
1da177e4 | 3724 | type = skb->protocol; |
82d8a867 PE |
3725 | list_for_each_entry_rcu(ptype, |
3726 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
63d8ea7f | 3727 | if (ptype->type == type && |
e3f48d37 JP |
3728 | (ptype->dev == null_or_dev || ptype->dev == skb->dev || |
3729 | ptype->dev == orig_dev)) { | |
4ec93edb | 3730 | if (pt_prev) |
f2ccd8fa | 3731 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3732 | pt_prev = ptype; |
3733 | } | |
3734 | } | |
3735 | ||
3736 | if (pt_prev) { | |
1080e512 | 3737 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 3738 | goto drop; |
1080e512 MT |
3739 | else |
3740 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 3741 | } else { |
b4b9e355 | 3742 | drop: |
caf586e5 | 3743 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3744 | kfree_skb(skb); |
3745 | /* Jamal, now you will not able to escape explaining | |
3746 | * me how you were going to use this. :-) | |
3747 | */ | |
3748 | ret = NET_RX_DROP; | |
3749 | } | |
3750 | ||
b4b9e355 | 3751 | unlock: |
1da177e4 | 3752 | rcu_read_unlock(); |
9754e293 DM |
3753 | return ret; |
3754 | } | |
3755 | ||
3756 | static int __netif_receive_skb(struct sk_buff *skb) | |
3757 | { | |
3758 | int ret; | |
3759 | ||
3760 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
3761 | unsigned long pflags = current->flags; | |
3762 | ||
3763 | /* | |
3764 | * PFMEMALLOC skbs are special, they should | |
3765 | * - be delivered to SOCK_MEMALLOC sockets only | |
3766 | * - stay away from userspace | |
3767 | * - have bounded memory usage | |
3768 | * | |
3769 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
3770 | * context down to all allocation sites. | |
3771 | */ | |
3772 | current->flags |= PF_MEMALLOC; | |
3773 | ret = __netif_receive_skb_core(skb, true); | |
3774 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
3775 | } else | |
3776 | ret = __netif_receive_skb_core(skb, false); | |
3777 | ||
1da177e4 LT |
3778 | return ret; |
3779 | } | |
0a9627f2 | 3780 | |
ae78dbfa | 3781 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 3782 | { |
588f0330 | 3783 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3784 | |
c1f19b51 RC |
3785 | if (skb_defer_rx_timestamp(skb)) |
3786 | return NET_RX_SUCCESS; | |
3787 | ||
df334545 | 3788 | #ifdef CONFIG_RPS |
c5905afb | 3789 | if (static_key_false(&rps_needed)) { |
3b098e2d ED |
3790 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
3791 | int cpu, ret; | |
fec5e652 | 3792 | |
3b098e2d ED |
3793 | rcu_read_lock(); |
3794 | ||
3795 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3796 | |
3b098e2d ED |
3797 | if (cpu >= 0) { |
3798 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3799 | rcu_read_unlock(); | |
adc9300e | 3800 | return ret; |
3b098e2d | 3801 | } |
adc9300e | 3802 | rcu_read_unlock(); |
fec5e652 | 3803 | } |
1e94d72f | 3804 | #endif |
adc9300e | 3805 | return __netif_receive_skb(skb); |
0a9627f2 | 3806 | } |
ae78dbfa BH |
3807 | |
3808 | /** | |
3809 | * netif_receive_skb - process receive buffer from network | |
3810 | * @skb: buffer to process | |
3811 | * | |
3812 | * netif_receive_skb() is the main receive data processing function. | |
3813 | * It always succeeds. The buffer may be dropped during processing | |
3814 | * for congestion control or by the protocol layers. | |
3815 | * | |
3816 | * This function may only be called from softirq context and interrupts | |
3817 | * should be enabled. | |
3818 | * | |
3819 | * Return values (usually ignored): | |
3820 | * NET_RX_SUCCESS: no congestion | |
3821 | * NET_RX_DROP: packet was dropped | |
3822 | */ | |
3823 | int netif_receive_skb(struct sk_buff *skb) | |
3824 | { | |
3825 | trace_netif_receive_skb_entry(skb); | |
3826 | ||
3827 | return netif_receive_skb_internal(skb); | |
3828 | } | |
d1b19dff | 3829 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3830 | |
88751275 ED |
3831 | /* Network device is going away, flush any packets still pending |
3832 | * Called with irqs disabled. | |
3833 | */ | |
152102c7 | 3834 | static void flush_backlog(void *arg) |
6e583ce5 | 3835 | { |
152102c7 | 3836 | struct net_device *dev = arg; |
903ceff7 | 3837 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
6e583ce5 SH |
3838 | struct sk_buff *skb, *tmp; |
3839 | ||
e36fa2f7 | 3840 | rps_lock(sd); |
6e7676c1 | 3841 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3842 | if (skb->dev == dev) { |
e36fa2f7 | 3843 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3844 | kfree_skb(skb); |
76cc8b13 | 3845 | input_queue_head_incr(sd); |
6e583ce5 | 3846 | } |
6e7676c1 | 3847 | } |
e36fa2f7 | 3848 | rps_unlock(sd); |
6e7676c1 CG |
3849 | |
3850 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3851 | if (skb->dev == dev) { | |
3852 | __skb_unlink(skb, &sd->process_queue); | |
3853 | kfree_skb(skb); | |
76cc8b13 | 3854 | input_queue_head_incr(sd); |
6e7676c1 CG |
3855 | } |
3856 | } | |
6e583ce5 SH |
3857 | } |
3858 | ||
d565b0a1 HX |
3859 | static int napi_gro_complete(struct sk_buff *skb) |
3860 | { | |
22061d80 | 3861 | struct packet_offload *ptype; |
d565b0a1 | 3862 | __be16 type = skb->protocol; |
22061d80 | 3863 | struct list_head *head = &offload_base; |
d565b0a1 HX |
3864 | int err = -ENOENT; |
3865 | ||
c3c7c254 ED |
3866 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
3867 | ||
fc59f9a3 HX |
3868 | if (NAPI_GRO_CB(skb)->count == 1) { |
3869 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3870 | goto out; |
fc59f9a3 | 3871 | } |
d565b0a1 HX |
3872 | |
3873 | rcu_read_lock(); | |
3874 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3875 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
3876 | continue; |
3877 | ||
299603e8 | 3878 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
3879 | break; |
3880 | } | |
3881 | rcu_read_unlock(); | |
3882 | ||
3883 | if (err) { | |
3884 | WARN_ON(&ptype->list == head); | |
3885 | kfree_skb(skb); | |
3886 | return NET_RX_SUCCESS; | |
3887 | } | |
3888 | ||
3889 | out: | |
ae78dbfa | 3890 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
3891 | } |
3892 | ||
2e71a6f8 ED |
3893 | /* napi->gro_list contains packets ordered by age. |
3894 | * youngest packets at the head of it. | |
3895 | * Complete skbs in reverse order to reduce latencies. | |
3896 | */ | |
3897 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 3898 | { |
2e71a6f8 | 3899 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 3900 | |
2e71a6f8 ED |
3901 | /* scan list and build reverse chain */ |
3902 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
3903 | skb->prev = prev; | |
3904 | prev = skb; | |
3905 | } | |
3906 | ||
3907 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 3908 | skb->next = NULL; |
2e71a6f8 ED |
3909 | |
3910 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
3911 | return; | |
3912 | ||
3913 | prev = skb->prev; | |
d565b0a1 | 3914 | napi_gro_complete(skb); |
2e71a6f8 | 3915 | napi->gro_count--; |
d565b0a1 HX |
3916 | } |
3917 | ||
3918 | napi->gro_list = NULL; | |
3919 | } | |
86cac58b | 3920 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3921 | |
89c5fa33 ED |
3922 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
3923 | { | |
3924 | struct sk_buff *p; | |
3925 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 3926 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
3927 | |
3928 | for (p = napi->gro_list; p; p = p->next) { | |
3929 | unsigned long diffs; | |
3930 | ||
0b4cec8c TH |
3931 | NAPI_GRO_CB(p)->flush = 0; |
3932 | ||
3933 | if (hash != skb_get_hash_raw(p)) { | |
3934 | NAPI_GRO_CB(p)->same_flow = 0; | |
3935 | continue; | |
3936 | } | |
3937 | ||
89c5fa33 ED |
3938 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
3939 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
3940 | if (maclen == ETH_HLEN) | |
3941 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 3942 | skb_mac_header(skb)); |
89c5fa33 ED |
3943 | else if (!diffs) |
3944 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 3945 | skb_mac_header(skb), |
89c5fa33 ED |
3946 | maclen); |
3947 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
3948 | } |
3949 | } | |
3950 | ||
299603e8 JC |
3951 | static void skb_gro_reset_offset(struct sk_buff *skb) |
3952 | { | |
3953 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
3954 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
3955 | ||
3956 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3957 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
3958 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
3959 | ||
3960 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
3961 | pinfo->nr_frags && | |
3962 | !PageHighMem(skb_frag_page(frag0))) { | |
3963 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
3964 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
89c5fa33 ED |
3965 | } |
3966 | } | |
3967 | ||
a50e233c ED |
3968 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
3969 | { | |
3970 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
3971 | ||
3972 | BUG_ON(skb->end - skb->tail < grow); | |
3973 | ||
3974 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3975 | ||
3976 | skb->data_len -= grow; | |
3977 | skb->tail += grow; | |
3978 | ||
3979 | pinfo->frags[0].page_offset += grow; | |
3980 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
3981 | ||
3982 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
3983 | skb_frag_unref(skb, 0); | |
3984 | memmove(pinfo->frags, pinfo->frags + 1, | |
3985 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
3986 | } | |
3987 | } | |
3988 | ||
bb728820 | 3989 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3990 | { |
3991 | struct sk_buff **pp = NULL; | |
22061d80 | 3992 | struct packet_offload *ptype; |
d565b0a1 | 3993 | __be16 type = skb->protocol; |
22061d80 | 3994 | struct list_head *head = &offload_base; |
0da2afd5 | 3995 | int same_flow; |
5b252f0c | 3996 | enum gro_result ret; |
a50e233c | 3997 | int grow; |
d565b0a1 | 3998 | |
9c62a68d | 3999 | if (!(skb->dev->features & NETIF_F_GRO)) |
d565b0a1 HX |
4000 | goto normal; |
4001 | ||
5a212329 | 4002 | if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad) |
f17f5c91 HX |
4003 | goto normal; |
4004 | ||
89c5fa33 ED |
4005 | gro_list_prepare(napi, skb); |
4006 | ||
d565b0a1 HX |
4007 | rcu_read_lock(); |
4008 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4009 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4010 | continue; |
4011 | ||
86911732 | 4012 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4013 | skb_reset_mac_len(skb); |
d565b0a1 HX |
4014 | NAPI_GRO_CB(skb)->same_flow = 0; |
4015 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 4016 | NAPI_GRO_CB(skb)->free = 0; |
b582ef09 | 4017 | NAPI_GRO_CB(skb)->udp_mark = 0; |
d565b0a1 | 4018 | |
662880f4 TH |
4019 | /* Setup for GRO checksum validation */ |
4020 | switch (skb->ip_summed) { | |
4021 | case CHECKSUM_COMPLETE: | |
4022 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4023 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4024 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4025 | break; | |
4026 | case CHECKSUM_UNNECESSARY: | |
4027 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4028 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4029 | break; | |
4030 | default: | |
4031 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4032 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4033 | } | |
d565b0a1 | 4034 | |
f191a1d1 | 4035 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4036 | break; |
4037 | } | |
4038 | rcu_read_unlock(); | |
4039 | ||
4040 | if (&ptype->list == head) | |
4041 | goto normal; | |
4042 | ||
0da2afd5 | 4043 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4044 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4045 | |
d565b0a1 HX |
4046 | if (pp) { |
4047 | struct sk_buff *nskb = *pp; | |
4048 | ||
4049 | *pp = nskb->next; | |
4050 | nskb->next = NULL; | |
4051 | napi_gro_complete(nskb); | |
4ae5544f | 4052 | napi->gro_count--; |
d565b0a1 HX |
4053 | } |
4054 | ||
0da2afd5 | 4055 | if (same_flow) |
d565b0a1 HX |
4056 | goto ok; |
4057 | ||
600adc18 | 4058 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4059 | goto normal; |
d565b0a1 | 4060 | |
600adc18 ED |
4061 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4062 | struct sk_buff *nskb = napi->gro_list; | |
4063 | ||
4064 | /* locate the end of the list to select the 'oldest' flow */ | |
4065 | while (nskb->next) { | |
4066 | pp = &nskb->next; | |
4067 | nskb = *pp; | |
4068 | } | |
4069 | *pp = NULL; | |
4070 | nskb->next = NULL; | |
4071 | napi_gro_complete(nskb); | |
4072 | } else { | |
4073 | napi->gro_count++; | |
4074 | } | |
d565b0a1 | 4075 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4076 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4077 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4078 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4079 | skb->next = napi->gro_list; |
4080 | napi->gro_list = skb; | |
5d0d9be8 | 4081 | ret = GRO_HELD; |
d565b0a1 | 4082 | |
ad0f9904 | 4083 | pull: |
a50e233c ED |
4084 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4085 | if (grow > 0) | |
4086 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4087 | ok: |
5d0d9be8 | 4088 | return ret; |
d565b0a1 HX |
4089 | |
4090 | normal: | |
ad0f9904 HX |
4091 | ret = GRO_NORMAL; |
4092 | goto pull; | |
5d38a079 | 4093 | } |
96e93eab | 4094 | |
bf5a755f JC |
4095 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4096 | { | |
4097 | struct list_head *offload_head = &offload_base; | |
4098 | struct packet_offload *ptype; | |
4099 | ||
4100 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4101 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4102 | continue; | |
4103 | return ptype; | |
4104 | } | |
4105 | return NULL; | |
4106 | } | |
e27a2f83 | 4107 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4108 | |
4109 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4110 | { | |
4111 | struct list_head *offload_head = &offload_base; | |
4112 | struct packet_offload *ptype; | |
4113 | ||
4114 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4115 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4116 | continue; | |
4117 | return ptype; | |
4118 | } | |
4119 | return NULL; | |
4120 | } | |
e27a2f83 | 4121 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4122 | |
bb728820 | 4123 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4124 | { |
5d0d9be8 HX |
4125 | switch (ret) { |
4126 | case GRO_NORMAL: | |
ae78dbfa | 4127 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4128 | ret = GRO_DROP; |
4129 | break; | |
5d38a079 | 4130 | |
5d0d9be8 | 4131 | case GRO_DROP: |
5d38a079 HX |
4132 | kfree_skb(skb); |
4133 | break; | |
5b252f0c | 4134 | |
daa86548 | 4135 | case GRO_MERGED_FREE: |
d7e8883c ED |
4136 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
4137 | kmem_cache_free(skbuff_head_cache, skb); | |
4138 | else | |
4139 | __kfree_skb(skb); | |
daa86548 ED |
4140 | break; |
4141 | ||
5b252f0c BH |
4142 | case GRO_HELD: |
4143 | case GRO_MERGED: | |
4144 | break; | |
5d38a079 HX |
4145 | } |
4146 | ||
c7c4b3b6 | 4147 | return ret; |
5d0d9be8 | 4148 | } |
5d0d9be8 | 4149 | |
c7c4b3b6 | 4150 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4151 | { |
ae78dbfa | 4152 | trace_napi_gro_receive_entry(skb); |
86911732 | 4153 | |
a50e233c ED |
4154 | skb_gro_reset_offset(skb); |
4155 | ||
89c5fa33 | 4156 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4157 | } |
4158 | EXPORT_SYMBOL(napi_gro_receive); | |
4159 | ||
d0c2b0d2 | 4160 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4161 | { |
93a35f59 ED |
4162 | if (unlikely(skb->pfmemalloc)) { |
4163 | consume_skb(skb); | |
4164 | return; | |
4165 | } | |
96e93eab | 4166 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4167 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4168 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4169 | skb->vlan_tci = 0; |
66c46d74 | 4170 | skb->dev = napi->dev; |
6d152e23 | 4171 | skb->skb_iif = 0; |
c3caf119 JC |
4172 | skb->encapsulation = 0; |
4173 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4174 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
96e93eab HX |
4175 | |
4176 | napi->skb = skb; | |
4177 | } | |
96e93eab | 4178 | |
76620aaf | 4179 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4180 | { |
5d38a079 | 4181 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4182 | |
4183 | if (!skb) { | |
fd11a83d | 4184 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
84b9cd63 | 4185 | napi->skb = skb; |
80595d59 | 4186 | } |
96e93eab HX |
4187 | return skb; |
4188 | } | |
76620aaf | 4189 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4190 | |
a50e233c ED |
4191 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4192 | struct sk_buff *skb, | |
4193 | gro_result_t ret) | |
96e93eab | 4194 | { |
5d0d9be8 HX |
4195 | switch (ret) { |
4196 | case GRO_NORMAL: | |
a50e233c ED |
4197 | case GRO_HELD: |
4198 | __skb_push(skb, ETH_HLEN); | |
4199 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4200 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4201 | ret = GRO_DROP; |
86911732 | 4202 | break; |
5d38a079 | 4203 | |
5d0d9be8 | 4204 | case GRO_DROP: |
5d0d9be8 HX |
4205 | case GRO_MERGED_FREE: |
4206 | napi_reuse_skb(napi, skb); | |
4207 | break; | |
5b252f0c BH |
4208 | |
4209 | case GRO_MERGED: | |
4210 | break; | |
5d0d9be8 | 4211 | } |
5d38a079 | 4212 | |
c7c4b3b6 | 4213 | return ret; |
5d38a079 | 4214 | } |
5d0d9be8 | 4215 | |
a50e233c ED |
4216 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
4217 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
4218 | * We copy ethernet header into skb->data to have a common layout. | |
4219 | */ | |
4adb9c4a | 4220 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4221 | { |
4222 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
4223 | const struct ethhdr *eth; |
4224 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
4225 | |
4226 | napi->skb = NULL; | |
4227 | ||
a50e233c ED |
4228 | skb_reset_mac_header(skb); |
4229 | skb_gro_reset_offset(skb); | |
4230 | ||
4231 | eth = skb_gro_header_fast(skb, 0); | |
4232 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
4233 | eth = skb_gro_header_slow(skb, hlen, 0); | |
4234 | if (unlikely(!eth)) { | |
4235 | napi_reuse_skb(napi, skb); | |
4236 | return NULL; | |
4237 | } | |
4238 | } else { | |
4239 | gro_pull_from_frag0(skb, hlen); | |
4240 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
4241 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 4242 | } |
a50e233c ED |
4243 | __skb_pull(skb, hlen); |
4244 | ||
4245 | /* | |
4246 | * This works because the only protocols we care about don't require | |
4247 | * special handling. | |
4248 | * We'll fix it up properly in napi_frags_finish() | |
4249 | */ | |
4250 | skb->protocol = eth->h_proto; | |
76620aaf | 4251 | |
76620aaf HX |
4252 | return skb; |
4253 | } | |
76620aaf | 4254 | |
c7c4b3b6 | 4255 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4256 | { |
76620aaf | 4257 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4258 | |
4259 | if (!skb) | |
c7c4b3b6 | 4260 | return GRO_DROP; |
5d0d9be8 | 4261 | |
ae78dbfa BH |
4262 | trace_napi_gro_frags_entry(skb); |
4263 | ||
89c5fa33 | 4264 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4265 | } |
5d38a079 HX |
4266 | EXPORT_SYMBOL(napi_gro_frags); |
4267 | ||
573e8fca TH |
4268 | /* Compute the checksum from gro_offset and return the folded value |
4269 | * after adding in any pseudo checksum. | |
4270 | */ | |
4271 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
4272 | { | |
4273 | __wsum wsum; | |
4274 | __sum16 sum; | |
4275 | ||
4276 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
4277 | ||
4278 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
4279 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
4280 | if (likely(!sum)) { | |
4281 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
4282 | !skb->csum_complete_sw) | |
4283 | netdev_rx_csum_fault(skb->dev); | |
4284 | } | |
4285 | ||
4286 | NAPI_GRO_CB(skb)->csum = wsum; | |
4287 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4288 | ||
4289 | return sum; | |
4290 | } | |
4291 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
4292 | ||
e326bed2 | 4293 | /* |
855abcf0 | 4294 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
4295 | * Note: called with local irq disabled, but exits with local irq enabled. |
4296 | */ | |
4297 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4298 | { | |
4299 | #ifdef CONFIG_RPS | |
4300 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4301 | ||
4302 | if (remsd) { | |
4303 | sd->rps_ipi_list = NULL; | |
4304 | ||
4305 | local_irq_enable(); | |
4306 | ||
4307 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
4308 | while (remsd) { | |
4309 | struct softnet_data *next = remsd->rps_ipi_next; | |
4310 | ||
4311 | if (cpu_online(remsd->cpu)) | |
c46fff2a | 4312 | smp_call_function_single_async(remsd->cpu, |
fce8ad15 | 4313 | &remsd->csd); |
e326bed2 ED |
4314 | remsd = next; |
4315 | } | |
4316 | } else | |
4317 | #endif | |
4318 | local_irq_enable(); | |
4319 | } | |
4320 | ||
d75b1ade ED |
4321 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
4322 | { | |
4323 | #ifdef CONFIG_RPS | |
4324 | return sd->rps_ipi_list != NULL; | |
4325 | #else | |
4326 | return false; | |
4327 | #endif | |
4328 | } | |
4329 | ||
bea3348e | 4330 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
4331 | { |
4332 | int work = 0; | |
eecfd7c4 | 4333 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 4334 | |
e326bed2 ED |
4335 | /* Check if we have pending ipi, its better to send them now, |
4336 | * not waiting net_rx_action() end. | |
4337 | */ | |
d75b1ade | 4338 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
4339 | local_irq_disable(); |
4340 | net_rps_action_and_irq_enable(sd); | |
4341 | } | |
d75b1ade | 4342 | |
bea3348e | 4343 | napi->weight = weight_p; |
6e7676c1 | 4344 | local_irq_disable(); |
11ef7a89 | 4345 | while (1) { |
1da177e4 | 4346 | struct sk_buff *skb; |
6e7676c1 CG |
4347 | |
4348 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
4349 | local_irq_enable(); | |
4350 | __netif_receive_skb(skb); | |
6e7676c1 | 4351 | local_irq_disable(); |
76cc8b13 TH |
4352 | input_queue_head_incr(sd); |
4353 | if (++work >= quota) { | |
4354 | local_irq_enable(); | |
4355 | return work; | |
4356 | } | |
6e7676c1 | 4357 | } |
1da177e4 | 4358 | |
e36fa2f7 | 4359 | rps_lock(sd); |
11ef7a89 | 4360 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
4361 | /* |
4362 | * Inline a custom version of __napi_complete(). | |
4363 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
4364 | * and NAPI_STATE_SCHED is the only possible flag set |
4365 | * on backlog. | |
4366 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
4367 | * and we dont need an smp_mb() memory barrier. |
4368 | */ | |
eecfd7c4 | 4369 | napi->state = 0; |
11ef7a89 | 4370 | rps_unlock(sd); |
eecfd7c4 | 4371 | |
11ef7a89 | 4372 | break; |
bea3348e | 4373 | } |
11ef7a89 TH |
4374 | |
4375 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
4376 | &sd->process_queue); | |
e36fa2f7 | 4377 | rps_unlock(sd); |
6e7676c1 CG |
4378 | } |
4379 | local_irq_enable(); | |
1da177e4 | 4380 | |
bea3348e SH |
4381 | return work; |
4382 | } | |
1da177e4 | 4383 | |
bea3348e SH |
4384 | /** |
4385 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4386 | * @n: entry to schedule |
bea3348e | 4387 | * |
bc9ad166 ED |
4388 | * The entry's receive function will be scheduled to run. |
4389 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 4390 | */ |
b5606c2d | 4391 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4392 | { |
4393 | unsigned long flags; | |
1da177e4 | 4394 | |
bea3348e | 4395 | local_irq_save(flags); |
903ceff7 | 4396 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 4397 | local_irq_restore(flags); |
1da177e4 | 4398 | } |
bea3348e SH |
4399 | EXPORT_SYMBOL(__napi_schedule); |
4400 | ||
bc9ad166 ED |
4401 | /** |
4402 | * __napi_schedule_irqoff - schedule for receive | |
4403 | * @n: entry to schedule | |
4404 | * | |
4405 | * Variant of __napi_schedule() assuming hard irqs are masked | |
4406 | */ | |
4407 | void __napi_schedule_irqoff(struct napi_struct *n) | |
4408 | { | |
4409 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
4410 | } | |
4411 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
4412 | ||
d565b0a1 HX |
4413 | void __napi_complete(struct napi_struct *n) |
4414 | { | |
4415 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
d565b0a1 | 4416 | |
d75b1ade | 4417 | list_del_init(&n->poll_list); |
4e857c58 | 4418 | smp_mb__before_atomic(); |
d565b0a1 HX |
4419 | clear_bit(NAPI_STATE_SCHED, &n->state); |
4420 | } | |
4421 | EXPORT_SYMBOL(__napi_complete); | |
4422 | ||
3b47d303 | 4423 | void napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 HX |
4424 | { |
4425 | unsigned long flags; | |
4426 | ||
4427 | /* | |
4428 | * don't let napi dequeue from the cpu poll list | |
4429 | * just in case its running on a different cpu | |
4430 | */ | |
4431 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4432 | return; | |
4433 | ||
3b47d303 ED |
4434 | if (n->gro_list) { |
4435 | unsigned long timeout = 0; | |
d75b1ade | 4436 | |
3b47d303 ED |
4437 | if (work_done) |
4438 | timeout = n->dev->gro_flush_timeout; | |
4439 | ||
4440 | if (timeout) | |
4441 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
4442 | HRTIMER_MODE_REL_PINNED); | |
4443 | else | |
4444 | napi_gro_flush(n, false); | |
4445 | } | |
d75b1ade ED |
4446 | if (likely(list_empty(&n->poll_list))) { |
4447 | WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state)); | |
4448 | } else { | |
4449 | /* If n->poll_list is not empty, we need to mask irqs */ | |
4450 | local_irq_save(flags); | |
4451 | __napi_complete(n); | |
4452 | local_irq_restore(flags); | |
4453 | } | |
d565b0a1 | 4454 | } |
3b47d303 | 4455 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 4456 | |
af12fa6e ET |
4457 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
4458 | struct napi_struct *napi_by_id(unsigned int napi_id) | |
4459 | { | |
4460 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
4461 | struct napi_struct *napi; | |
4462 | ||
4463 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
4464 | if (napi->napi_id == napi_id) | |
4465 | return napi; | |
4466 | ||
4467 | return NULL; | |
4468 | } | |
4469 | EXPORT_SYMBOL_GPL(napi_by_id); | |
4470 | ||
4471 | void napi_hash_add(struct napi_struct *napi) | |
4472 | { | |
4473 | if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) { | |
4474 | ||
4475 | spin_lock(&napi_hash_lock); | |
4476 | ||
4477 | /* 0 is not a valid id, we also skip an id that is taken | |
4478 | * we expect both events to be extremely rare | |
4479 | */ | |
4480 | napi->napi_id = 0; | |
4481 | while (!napi->napi_id) { | |
4482 | napi->napi_id = ++napi_gen_id; | |
4483 | if (napi_by_id(napi->napi_id)) | |
4484 | napi->napi_id = 0; | |
4485 | } | |
4486 | ||
4487 | hlist_add_head_rcu(&napi->napi_hash_node, | |
4488 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
4489 | ||
4490 | spin_unlock(&napi_hash_lock); | |
4491 | } | |
4492 | } | |
4493 | EXPORT_SYMBOL_GPL(napi_hash_add); | |
4494 | ||
4495 | /* Warning : caller is responsible to make sure rcu grace period | |
4496 | * is respected before freeing memory containing @napi | |
4497 | */ | |
4498 | void napi_hash_del(struct napi_struct *napi) | |
4499 | { | |
4500 | spin_lock(&napi_hash_lock); | |
4501 | ||
4502 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) | |
4503 | hlist_del_rcu(&napi->napi_hash_node); | |
4504 | ||
4505 | spin_unlock(&napi_hash_lock); | |
4506 | } | |
4507 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
4508 | ||
3b47d303 ED |
4509 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
4510 | { | |
4511 | struct napi_struct *napi; | |
4512 | ||
4513 | napi = container_of(timer, struct napi_struct, timer); | |
4514 | if (napi->gro_list) | |
4515 | napi_schedule(napi); | |
4516 | ||
4517 | return HRTIMER_NORESTART; | |
4518 | } | |
4519 | ||
d565b0a1 HX |
4520 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
4521 | int (*poll)(struct napi_struct *, int), int weight) | |
4522 | { | |
4523 | INIT_LIST_HEAD(&napi->poll_list); | |
3b47d303 ED |
4524 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
4525 | napi->timer.function = napi_watchdog; | |
4ae5544f | 4526 | napi->gro_count = 0; |
d565b0a1 | 4527 | napi->gro_list = NULL; |
5d38a079 | 4528 | napi->skb = NULL; |
d565b0a1 | 4529 | napi->poll = poll; |
82dc3c63 ED |
4530 | if (weight > NAPI_POLL_WEIGHT) |
4531 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
4532 | weight, dev->name); | |
d565b0a1 HX |
4533 | napi->weight = weight; |
4534 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 4535 | napi->dev = dev; |
5d38a079 | 4536 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
4537 | spin_lock_init(&napi->poll_lock); |
4538 | napi->poll_owner = -1; | |
4539 | #endif | |
4540 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
4541 | } | |
4542 | EXPORT_SYMBOL(netif_napi_add); | |
4543 | ||
3b47d303 ED |
4544 | void napi_disable(struct napi_struct *n) |
4545 | { | |
4546 | might_sleep(); | |
4547 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
4548 | ||
4549 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
4550 | msleep(1); | |
4551 | ||
4552 | hrtimer_cancel(&n->timer); | |
4553 | ||
4554 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
4555 | } | |
4556 | EXPORT_SYMBOL(napi_disable); | |
4557 | ||
d565b0a1 HX |
4558 | void netif_napi_del(struct napi_struct *napi) |
4559 | { | |
d7b06636 | 4560 | list_del_init(&napi->dev_list); |
76620aaf | 4561 | napi_free_frags(napi); |
d565b0a1 | 4562 | |
289dccbe | 4563 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 4564 | napi->gro_list = NULL; |
4ae5544f | 4565 | napi->gro_count = 0; |
d565b0a1 HX |
4566 | } |
4567 | EXPORT_SYMBOL(netif_napi_del); | |
4568 | ||
726ce70e HX |
4569 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
4570 | { | |
4571 | void *have; | |
4572 | int work, weight; | |
4573 | ||
4574 | list_del_init(&n->poll_list); | |
4575 | ||
4576 | have = netpoll_poll_lock(n); | |
4577 | ||
4578 | weight = n->weight; | |
4579 | ||
4580 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
4581 | * with netpoll's poll_napi(). Only the entity which | |
4582 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
4583 | * actually make the ->poll() call. Therefore we avoid | |
4584 | * accidentally calling ->poll() when NAPI is not scheduled. | |
4585 | */ | |
4586 | work = 0; | |
4587 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
4588 | work = n->poll(n, weight); | |
4589 | trace_napi_poll(n); | |
4590 | } | |
4591 | ||
4592 | WARN_ON_ONCE(work > weight); | |
4593 | ||
4594 | if (likely(work < weight)) | |
4595 | goto out_unlock; | |
4596 | ||
4597 | /* Drivers must not modify the NAPI state if they | |
4598 | * consume the entire weight. In such cases this code | |
4599 | * still "owns" the NAPI instance and therefore can | |
4600 | * move the instance around on the list at-will. | |
4601 | */ | |
4602 | if (unlikely(napi_disable_pending(n))) { | |
4603 | napi_complete(n); | |
4604 | goto out_unlock; | |
4605 | } | |
4606 | ||
4607 | if (n->gro_list) { | |
4608 | /* flush too old packets | |
4609 | * If HZ < 1000, flush all packets. | |
4610 | */ | |
4611 | napi_gro_flush(n, HZ >= 1000); | |
4612 | } | |
4613 | ||
001ce546 HX |
4614 | /* Some drivers may have called napi_schedule |
4615 | * prior to exhausting their budget. | |
4616 | */ | |
4617 | if (unlikely(!list_empty(&n->poll_list))) { | |
4618 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
4619 | n->dev ? n->dev->name : "backlog"); | |
4620 | goto out_unlock; | |
4621 | } | |
4622 | ||
726ce70e HX |
4623 | list_add_tail(&n->poll_list, repoll); |
4624 | ||
4625 | out_unlock: | |
4626 | netpoll_poll_unlock(have); | |
4627 | ||
4628 | return work; | |
4629 | } | |
4630 | ||
1da177e4 LT |
4631 | static void net_rx_action(struct softirq_action *h) |
4632 | { | |
903ceff7 | 4633 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
24f8b238 | 4634 | unsigned long time_limit = jiffies + 2; |
51b0bded | 4635 | int budget = netdev_budget; |
d75b1ade ED |
4636 | LIST_HEAD(list); |
4637 | LIST_HEAD(repoll); | |
53fb95d3 | 4638 | |
1da177e4 | 4639 | local_irq_disable(); |
d75b1ade ED |
4640 | list_splice_init(&sd->poll_list, &list); |
4641 | local_irq_enable(); | |
1da177e4 | 4642 | |
ceb8d5bf | 4643 | for (;;) { |
bea3348e | 4644 | struct napi_struct *n; |
1da177e4 | 4645 | |
ceb8d5bf HX |
4646 | if (list_empty(&list)) { |
4647 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
4648 | return; | |
4649 | break; | |
4650 | } | |
4651 | ||
6bd373eb HX |
4652 | n = list_first_entry(&list, struct napi_struct, poll_list); |
4653 | budget -= napi_poll(n, &repoll); | |
4654 | ||
d75b1ade | 4655 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
4656 | * Allow this to run for 2 jiffies since which will allow |
4657 | * an average latency of 1.5/HZ. | |
bea3348e | 4658 | */ |
ceb8d5bf HX |
4659 | if (unlikely(budget <= 0 || |
4660 | time_after_eq(jiffies, time_limit))) { | |
4661 | sd->time_squeeze++; | |
4662 | break; | |
4663 | } | |
1da177e4 | 4664 | } |
d75b1ade | 4665 | |
d75b1ade ED |
4666 | local_irq_disable(); |
4667 | ||
4668 | list_splice_tail_init(&sd->poll_list, &list); | |
4669 | list_splice_tail(&repoll, &list); | |
4670 | list_splice(&list, &sd->poll_list); | |
4671 | if (!list_empty(&sd->poll_list)) | |
4672 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4673 | ||
e326bed2 | 4674 | net_rps_action_and_irq_enable(sd); |
1da177e4 LT |
4675 | } |
4676 | ||
aa9d8560 | 4677 | struct netdev_adjacent { |
9ff162a8 | 4678 | struct net_device *dev; |
5d261913 VF |
4679 | |
4680 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 4681 | bool master; |
5d261913 | 4682 | |
5d261913 VF |
4683 | /* counter for the number of times this device was added to us */ |
4684 | u16 ref_nr; | |
4685 | ||
402dae96 VF |
4686 | /* private field for the users */ |
4687 | void *private; | |
4688 | ||
9ff162a8 JP |
4689 | struct list_head list; |
4690 | struct rcu_head rcu; | |
9ff162a8 JP |
4691 | }; |
4692 | ||
5d261913 VF |
4693 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev, |
4694 | struct net_device *adj_dev, | |
2f268f12 | 4695 | struct list_head *adj_list) |
9ff162a8 | 4696 | { |
5d261913 | 4697 | struct netdev_adjacent *adj; |
5d261913 | 4698 | |
2f268f12 | 4699 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
4700 | if (adj->dev == adj_dev) |
4701 | return adj; | |
9ff162a8 JP |
4702 | } |
4703 | return NULL; | |
4704 | } | |
4705 | ||
4706 | /** | |
4707 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
4708 | * @dev: device | |
4709 | * @upper_dev: upper device to check | |
4710 | * | |
4711 | * Find out if a device is linked to specified upper device and return true | |
4712 | * in case it is. Note that this checks only immediate upper device, | |
4713 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
4714 | */ | |
4715 | bool netdev_has_upper_dev(struct net_device *dev, | |
4716 | struct net_device *upper_dev) | |
4717 | { | |
4718 | ASSERT_RTNL(); | |
4719 | ||
2f268f12 | 4720 | return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper); |
9ff162a8 JP |
4721 | } |
4722 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
4723 | ||
4724 | /** | |
4725 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
4726 | * @dev: device | |
4727 | * | |
4728 | * Find out if a device is linked to an upper device and return true in case | |
4729 | * it is. The caller must hold the RTNL lock. | |
4730 | */ | |
1d143d9f | 4731 | static bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
4732 | { |
4733 | ASSERT_RTNL(); | |
4734 | ||
2f268f12 | 4735 | return !list_empty(&dev->all_adj_list.upper); |
9ff162a8 | 4736 | } |
9ff162a8 JP |
4737 | |
4738 | /** | |
4739 | * netdev_master_upper_dev_get - Get master upper device | |
4740 | * @dev: device | |
4741 | * | |
4742 | * Find a master upper device and return pointer to it or NULL in case | |
4743 | * it's not there. The caller must hold the RTNL lock. | |
4744 | */ | |
4745 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
4746 | { | |
aa9d8560 | 4747 | struct netdev_adjacent *upper; |
9ff162a8 JP |
4748 | |
4749 | ASSERT_RTNL(); | |
4750 | ||
2f268f12 | 4751 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
4752 | return NULL; |
4753 | ||
2f268f12 | 4754 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 4755 | struct netdev_adjacent, list); |
9ff162a8 JP |
4756 | if (likely(upper->master)) |
4757 | return upper->dev; | |
4758 | return NULL; | |
4759 | } | |
4760 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
4761 | ||
b6ccba4c VF |
4762 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
4763 | { | |
4764 | struct netdev_adjacent *adj; | |
4765 | ||
4766 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
4767 | ||
4768 | return adj->private; | |
4769 | } | |
4770 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
4771 | ||
44a40855 VY |
4772 | /** |
4773 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
4774 | * @dev: device | |
4775 | * @iter: list_head ** of the current position | |
4776 | * | |
4777 | * Gets the next device from the dev's upper list, starting from iter | |
4778 | * position. The caller must hold RCU read lock. | |
4779 | */ | |
4780 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
4781 | struct list_head **iter) | |
4782 | { | |
4783 | struct netdev_adjacent *upper; | |
4784 | ||
4785 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
4786 | ||
4787 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4788 | ||
4789 | if (&upper->list == &dev->adj_list.upper) | |
4790 | return NULL; | |
4791 | ||
4792 | *iter = &upper->list; | |
4793 | ||
4794 | return upper->dev; | |
4795 | } | |
4796 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
4797 | ||
31088a11 VF |
4798 | /** |
4799 | * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list | |
48311f46 VF |
4800 | * @dev: device |
4801 | * @iter: list_head ** of the current position | |
4802 | * | |
4803 | * Gets the next device from the dev's upper list, starting from iter | |
4804 | * position. The caller must hold RCU read lock. | |
4805 | */ | |
2f268f12 VF |
4806 | struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, |
4807 | struct list_head **iter) | |
48311f46 VF |
4808 | { |
4809 | struct netdev_adjacent *upper; | |
4810 | ||
85328240 | 4811 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); |
48311f46 VF |
4812 | |
4813 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4814 | ||
2f268f12 | 4815 | if (&upper->list == &dev->all_adj_list.upper) |
48311f46 VF |
4816 | return NULL; |
4817 | ||
4818 | *iter = &upper->list; | |
4819 | ||
4820 | return upper->dev; | |
4821 | } | |
2f268f12 | 4822 | EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); |
48311f46 | 4823 | |
31088a11 VF |
4824 | /** |
4825 | * netdev_lower_get_next_private - Get the next ->private from the | |
4826 | * lower neighbour list | |
4827 | * @dev: device | |
4828 | * @iter: list_head ** of the current position | |
4829 | * | |
4830 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
4831 | * list, starting from iter position. The caller must hold either hold the | |
4832 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
4833 | * list will remain unchainged. | |
4834 | */ | |
4835 | void *netdev_lower_get_next_private(struct net_device *dev, | |
4836 | struct list_head **iter) | |
4837 | { | |
4838 | struct netdev_adjacent *lower; | |
4839 | ||
4840 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
4841 | ||
4842 | if (&lower->list == &dev->adj_list.lower) | |
4843 | return NULL; | |
4844 | ||
6859e7df | 4845 | *iter = lower->list.next; |
31088a11 VF |
4846 | |
4847 | return lower->private; | |
4848 | } | |
4849 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
4850 | ||
4851 | /** | |
4852 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
4853 | * lower neighbour list, RCU | |
4854 | * variant | |
4855 | * @dev: device | |
4856 | * @iter: list_head ** of the current position | |
4857 | * | |
4858 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
4859 | * list, starting from iter position. The caller must hold RCU read lock. | |
4860 | */ | |
4861 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
4862 | struct list_head **iter) | |
4863 | { | |
4864 | struct netdev_adjacent *lower; | |
4865 | ||
4866 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4867 | ||
4868 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4869 | ||
4870 | if (&lower->list == &dev->adj_list.lower) | |
4871 | return NULL; | |
4872 | ||
6859e7df | 4873 | *iter = &lower->list; |
31088a11 VF |
4874 | |
4875 | return lower->private; | |
4876 | } | |
4877 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
4878 | ||
4085ebe8 VY |
4879 | /** |
4880 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
4881 | * list | |
4882 | * @dev: device | |
4883 | * @iter: list_head ** of the current position | |
4884 | * | |
4885 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
4886 | * list, starting from iter position. The caller must hold RTNL lock or | |
4887 | * its own locking that guarantees that the neighbour lower | |
4888 | * list will remain unchainged. | |
4889 | */ | |
4890 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
4891 | { | |
4892 | struct netdev_adjacent *lower; | |
4893 | ||
4894 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
4895 | ||
4896 | if (&lower->list == &dev->adj_list.lower) | |
4897 | return NULL; | |
4898 | ||
4899 | *iter = &lower->list; | |
4900 | ||
4901 | return lower->dev; | |
4902 | } | |
4903 | EXPORT_SYMBOL(netdev_lower_get_next); | |
4904 | ||
e001bfad | 4905 | /** |
4906 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
4907 | * lower neighbour list, RCU | |
4908 | * variant | |
4909 | * @dev: device | |
4910 | * | |
4911 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
4912 | * list. The caller must hold RCU read lock. | |
4913 | */ | |
4914 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
4915 | { | |
4916 | struct netdev_adjacent *lower; | |
4917 | ||
4918 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
4919 | struct netdev_adjacent, list); | |
4920 | if (lower) | |
4921 | return lower->private; | |
4922 | return NULL; | |
4923 | } | |
4924 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
4925 | ||
9ff162a8 JP |
4926 | /** |
4927 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
4928 | * @dev: device | |
4929 | * | |
4930 | * Find a master upper device and return pointer to it or NULL in case | |
4931 | * it's not there. The caller must hold the RCU read lock. | |
4932 | */ | |
4933 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
4934 | { | |
aa9d8560 | 4935 | struct netdev_adjacent *upper; |
9ff162a8 | 4936 | |
2f268f12 | 4937 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 4938 | struct netdev_adjacent, list); |
9ff162a8 JP |
4939 | if (upper && likely(upper->master)) |
4940 | return upper->dev; | |
4941 | return NULL; | |
4942 | } | |
4943 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
4944 | ||
0a59f3a9 | 4945 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
4946 | struct net_device *adj_dev, |
4947 | struct list_head *dev_list) | |
4948 | { | |
4949 | char linkname[IFNAMSIZ+7]; | |
4950 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
4951 | "upper_%s" : "lower_%s", adj_dev->name); | |
4952 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
4953 | linkname); | |
4954 | } | |
0a59f3a9 | 4955 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
4956 | char *name, |
4957 | struct list_head *dev_list) | |
4958 | { | |
4959 | char linkname[IFNAMSIZ+7]; | |
4960 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
4961 | "upper_%s" : "lower_%s", name); | |
4962 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
4963 | } | |
4964 | ||
7ce64c79 AF |
4965 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
4966 | struct net_device *adj_dev, | |
4967 | struct list_head *dev_list) | |
4968 | { | |
4969 | return (dev_list == &dev->adj_list.upper || | |
4970 | dev_list == &dev->adj_list.lower) && | |
4971 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
4972 | } | |
3ee32707 | 4973 | |
5d261913 VF |
4974 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
4975 | struct net_device *adj_dev, | |
7863c054 | 4976 | struct list_head *dev_list, |
402dae96 | 4977 | void *private, bool master) |
5d261913 VF |
4978 | { |
4979 | struct netdev_adjacent *adj; | |
842d67a7 | 4980 | int ret; |
5d261913 | 4981 | |
7863c054 | 4982 | adj = __netdev_find_adj(dev, adj_dev, dev_list); |
5d261913 VF |
4983 | |
4984 | if (adj) { | |
5d261913 VF |
4985 | adj->ref_nr++; |
4986 | return 0; | |
4987 | } | |
4988 | ||
4989 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
4990 | if (!adj) | |
4991 | return -ENOMEM; | |
4992 | ||
4993 | adj->dev = adj_dev; | |
4994 | adj->master = master; | |
5d261913 | 4995 | adj->ref_nr = 1; |
402dae96 | 4996 | adj->private = private; |
5d261913 | 4997 | dev_hold(adj_dev); |
2f268f12 VF |
4998 | |
4999 | pr_debug("dev_hold for %s, because of link added from %s to %s\n", | |
5000 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 | 5001 | |
7ce64c79 | 5002 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 5003 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
5004 | if (ret) |
5005 | goto free_adj; | |
5006 | } | |
5007 | ||
7863c054 | 5008 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
5009 | if (master) { |
5010 | ret = sysfs_create_link(&(dev->dev.kobj), | |
5011 | &(adj_dev->dev.kobj), "master"); | |
5012 | if (ret) | |
5831d66e | 5013 | goto remove_symlinks; |
842d67a7 | 5014 | |
7863c054 | 5015 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 5016 | } else { |
7863c054 | 5017 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 5018 | } |
5d261913 VF |
5019 | |
5020 | return 0; | |
842d67a7 | 5021 | |
5831d66e | 5022 | remove_symlinks: |
7ce64c79 | 5023 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5024 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
5025 | free_adj: |
5026 | kfree(adj); | |
974daef7 | 5027 | dev_put(adj_dev); |
842d67a7 VF |
5028 | |
5029 | return ret; | |
5d261913 VF |
5030 | } |
5031 | ||
1d143d9f | 5032 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
5033 | struct net_device *adj_dev, | |
5034 | struct list_head *dev_list) | |
5d261913 VF |
5035 | { |
5036 | struct netdev_adjacent *adj; | |
5037 | ||
7863c054 | 5038 | adj = __netdev_find_adj(dev, adj_dev, dev_list); |
5d261913 | 5039 | |
2f268f12 VF |
5040 | if (!adj) { |
5041 | pr_err("tried to remove device %s from %s\n", | |
5042 | dev->name, adj_dev->name); | |
5d261913 | 5043 | BUG(); |
2f268f12 | 5044 | } |
5d261913 VF |
5045 | |
5046 | if (adj->ref_nr > 1) { | |
2f268f12 VF |
5047 | pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name, |
5048 | adj->ref_nr-1); | |
5d261913 VF |
5049 | adj->ref_nr--; |
5050 | return; | |
5051 | } | |
5052 | ||
842d67a7 VF |
5053 | if (adj->master) |
5054 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
5055 | ||
7ce64c79 | 5056 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5057 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 5058 | |
5d261913 | 5059 | list_del_rcu(&adj->list); |
2f268f12 VF |
5060 | pr_debug("dev_put for %s, because link removed from %s to %s\n", |
5061 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 VF |
5062 | dev_put(adj_dev); |
5063 | kfree_rcu(adj, rcu); | |
5064 | } | |
5065 | ||
1d143d9f | 5066 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
5067 | struct net_device *upper_dev, | |
5068 | struct list_head *up_list, | |
5069 | struct list_head *down_list, | |
5070 | void *private, bool master) | |
5d261913 VF |
5071 | { |
5072 | int ret; | |
5073 | ||
402dae96 VF |
5074 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private, |
5075 | master); | |
5d261913 VF |
5076 | if (ret) |
5077 | return ret; | |
5078 | ||
402dae96 VF |
5079 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private, |
5080 | false); | |
5d261913 | 5081 | if (ret) { |
2f268f12 | 5082 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5d261913 VF |
5083 | return ret; |
5084 | } | |
5085 | ||
5086 | return 0; | |
5087 | } | |
5088 | ||
1d143d9f | 5089 | static int __netdev_adjacent_dev_link(struct net_device *dev, |
5090 | struct net_device *upper_dev) | |
5d261913 | 5091 | { |
2f268f12 VF |
5092 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
5093 | &dev->all_adj_list.upper, | |
5094 | &upper_dev->all_adj_list.lower, | |
402dae96 | 5095 | NULL, false); |
5d261913 VF |
5096 | } |
5097 | ||
1d143d9f | 5098 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
5099 | struct net_device *upper_dev, | |
5100 | struct list_head *up_list, | |
5101 | struct list_head *down_list) | |
5d261913 | 5102 | { |
2f268f12 VF |
5103 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5104 | __netdev_adjacent_dev_remove(upper_dev, dev, down_list); | |
5d261913 VF |
5105 | } |
5106 | ||
1d143d9f | 5107 | static void __netdev_adjacent_dev_unlink(struct net_device *dev, |
5108 | struct net_device *upper_dev) | |
5d261913 | 5109 | { |
2f268f12 VF |
5110 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, |
5111 | &dev->all_adj_list.upper, | |
5112 | &upper_dev->all_adj_list.lower); | |
5113 | } | |
5114 | ||
1d143d9f | 5115 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
5116 | struct net_device *upper_dev, | |
5117 | void *private, bool master) | |
2f268f12 VF |
5118 | { |
5119 | int ret = __netdev_adjacent_dev_link(dev, upper_dev); | |
5120 | ||
5121 | if (ret) | |
5122 | return ret; | |
5123 | ||
5124 | ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, | |
5125 | &dev->adj_list.upper, | |
5126 | &upper_dev->adj_list.lower, | |
402dae96 | 5127 | private, master); |
2f268f12 VF |
5128 | if (ret) { |
5129 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5130 | return ret; | |
5131 | } | |
5132 | ||
5133 | return 0; | |
5d261913 VF |
5134 | } |
5135 | ||
1d143d9f | 5136 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
5137 | struct net_device *upper_dev) | |
2f268f12 VF |
5138 | { |
5139 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5140 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, | |
5141 | &dev->adj_list.upper, | |
5142 | &upper_dev->adj_list.lower); | |
5143 | } | |
5d261913 | 5144 | |
9ff162a8 | 5145 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 VF |
5146 | struct net_device *upper_dev, bool master, |
5147 | void *private) | |
9ff162a8 | 5148 | { |
5d261913 VF |
5149 | struct netdev_adjacent *i, *j, *to_i, *to_j; |
5150 | int ret = 0; | |
9ff162a8 JP |
5151 | |
5152 | ASSERT_RTNL(); | |
5153 | ||
5154 | if (dev == upper_dev) | |
5155 | return -EBUSY; | |
5156 | ||
5157 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
2f268f12 | 5158 | if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper)) |
9ff162a8 JP |
5159 | return -EBUSY; |
5160 | ||
2f268f12 | 5161 | if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper)) |
9ff162a8 JP |
5162 | return -EEXIST; |
5163 | ||
5164 | if (master && netdev_master_upper_dev_get(dev)) | |
5165 | return -EBUSY; | |
5166 | ||
402dae96 VF |
5167 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private, |
5168 | master); | |
5d261913 VF |
5169 | if (ret) |
5170 | return ret; | |
9ff162a8 | 5171 | |
5d261913 | 5172 | /* Now that we linked these devs, make all the upper_dev's |
2f268f12 | 5173 | * all_adj_list.upper visible to every dev's all_adj_list.lower an |
5d261913 VF |
5174 | * versa, and don't forget the devices itself. All of these |
5175 | * links are non-neighbours. | |
5176 | */ | |
2f268f12 VF |
5177 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5178 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5179 | pr_debug("Interlinking %s with %s, non-neighbour\n", | |
5180 | i->dev->name, j->dev->name); | |
5d261913 VF |
5181 | ret = __netdev_adjacent_dev_link(i->dev, j->dev); |
5182 | if (ret) | |
5183 | goto rollback_mesh; | |
5184 | } | |
5185 | } | |
5186 | ||
5187 | /* add dev to every upper_dev's upper device */ | |
2f268f12 VF |
5188 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5189 | pr_debug("linking %s's upper device %s with %s\n", | |
5190 | upper_dev->name, i->dev->name, dev->name); | |
5d261913 VF |
5191 | ret = __netdev_adjacent_dev_link(dev, i->dev); |
5192 | if (ret) | |
5193 | goto rollback_upper_mesh; | |
5194 | } | |
5195 | ||
5196 | /* add upper_dev to every dev's lower device */ | |
2f268f12 VF |
5197 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5198 | pr_debug("linking %s's lower device %s with %s\n", dev->name, | |
5199 | i->dev->name, upper_dev->name); | |
5d261913 VF |
5200 | ret = __netdev_adjacent_dev_link(i->dev, upper_dev); |
5201 | if (ret) | |
5202 | goto rollback_lower_mesh; | |
5203 | } | |
9ff162a8 | 5204 | |
42e52bf9 | 5205 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 | 5206 | return 0; |
5d261913 VF |
5207 | |
5208 | rollback_lower_mesh: | |
5209 | to_i = i; | |
2f268f12 | 5210 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5d261913 VF |
5211 | if (i == to_i) |
5212 | break; | |
5213 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); | |
5214 | } | |
5215 | ||
5216 | i = NULL; | |
5217 | ||
5218 | rollback_upper_mesh: | |
5219 | to_i = i; | |
2f268f12 | 5220 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5d261913 VF |
5221 | if (i == to_i) |
5222 | break; | |
5223 | __netdev_adjacent_dev_unlink(dev, i->dev); | |
5224 | } | |
5225 | ||
5226 | i = j = NULL; | |
5227 | ||
5228 | rollback_mesh: | |
5229 | to_i = i; | |
5230 | to_j = j; | |
2f268f12 VF |
5231 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5232 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5d261913 VF |
5233 | if (i == to_i && j == to_j) |
5234 | break; | |
5235 | __netdev_adjacent_dev_unlink(i->dev, j->dev); | |
5236 | } | |
5237 | if (i == to_i) | |
5238 | break; | |
5239 | } | |
5240 | ||
2f268f12 | 5241 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5242 | |
5243 | return ret; | |
9ff162a8 JP |
5244 | } |
5245 | ||
5246 | /** | |
5247 | * netdev_upper_dev_link - Add a link to the upper device | |
5248 | * @dev: device | |
5249 | * @upper_dev: new upper device | |
5250 | * | |
5251 | * Adds a link to device which is upper to this one. The caller must hold | |
5252 | * the RTNL lock. On a failure a negative errno code is returned. | |
5253 | * On success the reference counts are adjusted and the function | |
5254 | * returns zero. | |
5255 | */ | |
5256 | int netdev_upper_dev_link(struct net_device *dev, | |
5257 | struct net_device *upper_dev) | |
5258 | { | |
402dae96 | 5259 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL); |
9ff162a8 JP |
5260 | } |
5261 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
5262 | ||
5263 | /** | |
5264 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
5265 | * @dev: device | |
5266 | * @upper_dev: new upper device | |
5267 | * | |
5268 | * Adds a link to device which is upper to this one. In this case, only | |
5269 | * one master upper device can be linked, although other non-master devices | |
5270 | * might be linked as well. The caller must hold the RTNL lock. | |
5271 | * On a failure a negative errno code is returned. On success the reference | |
5272 | * counts are adjusted and the function returns zero. | |
5273 | */ | |
5274 | int netdev_master_upper_dev_link(struct net_device *dev, | |
5275 | struct net_device *upper_dev) | |
5276 | { | |
402dae96 | 5277 | return __netdev_upper_dev_link(dev, upper_dev, true, NULL); |
9ff162a8 JP |
5278 | } |
5279 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
5280 | ||
402dae96 VF |
5281 | int netdev_master_upper_dev_link_private(struct net_device *dev, |
5282 | struct net_device *upper_dev, | |
5283 | void *private) | |
5284 | { | |
5285 | return __netdev_upper_dev_link(dev, upper_dev, true, private); | |
5286 | } | |
5287 | EXPORT_SYMBOL(netdev_master_upper_dev_link_private); | |
5288 | ||
9ff162a8 JP |
5289 | /** |
5290 | * netdev_upper_dev_unlink - Removes a link to upper device | |
5291 | * @dev: device | |
5292 | * @upper_dev: new upper device | |
5293 | * | |
5294 | * Removes a link to device which is upper to this one. The caller must hold | |
5295 | * the RTNL lock. | |
5296 | */ | |
5297 | void netdev_upper_dev_unlink(struct net_device *dev, | |
5298 | struct net_device *upper_dev) | |
5299 | { | |
5d261913 | 5300 | struct netdev_adjacent *i, *j; |
9ff162a8 JP |
5301 | ASSERT_RTNL(); |
5302 | ||
2f268f12 | 5303 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5304 | |
5305 | /* Here is the tricky part. We must remove all dev's lower | |
5306 | * devices from all upper_dev's upper devices and vice | |
5307 | * versa, to maintain the graph relationship. | |
5308 | */ | |
2f268f12 VF |
5309 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5310 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) | |
5d261913 VF |
5311 | __netdev_adjacent_dev_unlink(i->dev, j->dev); |
5312 | ||
5313 | /* remove also the devices itself from lower/upper device | |
5314 | * list | |
5315 | */ | |
2f268f12 | 5316 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5d261913 VF |
5317 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); |
5318 | ||
2f268f12 | 5319 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) |
5d261913 VF |
5320 | __netdev_adjacent_dev_unlink(dev, i->dev); |
5321 | ||
42e52bf9 | 5322 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 JP |
5323 | } |
5324 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
5325 | ||
4c75431a AF |
5326 | void netdev_adjacent_add_links(struct net_device *dev) |
5327 | { | |
5328 | struct netdev_adjacent *iter; | |
5329 | ||
5330 | struct net *net = dev_net(dev); | |
5331 | ||
5332 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
5333 | if (!net_eq(net,dev_net(iter->dev))) | |
5334 | continue; | |
5335 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5336 | &iter->dev->adj_list.lower); | |
5337 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5338 | &dev->adj_list.upper); | |
5339 | } | |
5340 | ||
5341 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
5342 | if (!net_eq(net,dev_net(iter->dev))) | |
5343 | continue; | |
5344 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5345 | &iter->dev->adj_list.upper); | |
5346 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5347 | &dev->adj_list.lower); | |
5348 | } | |
5349 | } | |
5350 | ||
5351 | void netdev_adjacent_del_links(struct net_device *dev) | |
5352 | { | |
5353 | struct netdev_adjacent *iter; | |
5354 | ||
5355 | struct net *net = dev_net(dev); | |
5356 | ||
5357 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
5358 | if (!net_eq(net,dev_net(iter->dev))) | |
5359 | continue; | |
5360 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5361 | &iter->dev->adj_list.lower); | |
5362 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5363 | &dev->adj_list.upper); | |
5364 | } | |
5365 | ||
5366 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
5367 | if (!net_eq(net,dev_net(iter->dev))) | |
5368 | continue; | |
5369 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5370 | &iter->dev->adj_list.upper); | |
5371 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5372 | &dev->adj_list.lower); | |
5373 | } | |
5374 | } | |
5375 | ||
5bb025fa | 5376 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 5377 | { |
5bb025fa | 5378 | struct netdev_adjacent *iter; |
402dae96 | 5379 | |
4c75431a AF |
5380 | struct net *net = dev_net(dev); |
5381 | ||
5bb025fa | 5382 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
4c75431a AF |
5383 | if (!net_eq(net,dev_net(iter->dev))) |
5384 | continue; | |
5bb025fa VF |
5385 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
5386 | &iter->dev->adj_list.lower); | |
5387 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5388 | &iter->dev->adj_list.lower); | |
5389 | } | |
402dae96 | 5390 | |
5bb025fa | 5391 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
4c75431a AF |
5392 | if (!net_eq(net,dev_net(iter->dev))) |
5393 | continue; | |
5bb025fa VF |
5394 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
5395 | &iter->dev->adj_list.upper); | |
5396 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5397 | &iter->dev->adj_list.upper); | |
5398 | } | |
402dae96 | 5399 | } |
402dae96 VF |
5400 | |
5401 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
5402 | struct net_device *lower_dev) | |
5403 | { | |
5404 | struct netdev_adjacent *lower; | |
5405 | ||
5406 | if (!lower_dev) | |
5407 | return NULL; | |
5408 | lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower); | |
5409 | if (!lower) | |
5410 | return NULL; | |
5411 | ||
5412 | return lower->private; | |
5413 | } | |
5414 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
5415 | ||
4085ebe8 VY |
5416 | |
5417 | int dev_get_nest_level(struct net_device *dev, | |
5418 | bool (*type_check)(struct net_device *dev)) | |
5419 | { | |
5420 | struct net_device *lower = NULL; | |
5421 | struct list_head *iter; | |
5422 | int max_nest = -1; | |
5423 | int nest; | |
5424 | ||
5425 | ASSERT_RTNL(); | |
5426 | ||
5427 | netdev_for_each_lower_dev(dev, lower, iter) { | |
5428 | nest = dev_get_nest_level(lower, type_check); | |
5429 | if (max_nest < nest) | |
5430 | max_nest = nest; | |
5431 | } | |
5432 | ||
5433 | if (type_check(dev)) | |
5434 | max_nest++; | |
5435 | ||
5436 | return max_nest; | |
5437 | } | |
5438 | EXPORT_SYMBOL(dev_get_nest_level); | |
5439 | ||
b6c40d68 PM |
5440 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
5441 | { | |
d314774c SH |
5442 | const struct net_device_ops *ops = dev->netdev_ops; |
5443 | ||
d2615bf4 | 5444 | if (ops->ndo_change_rx_flags) |
d314774c | 5445 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
5446 | } |
5447 | ||
991fb3f7 | 5448 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5449 | { |
b536db93 | 5450 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
5451 | kuid_t uid; |
5452 | kgid_t gid; | |
1da177e4 | 5453 | |
24023451 PM |
5454 | ASSERT_RTNL(); |
5455 | ||
dad9b335 WC |
5456 | dev->flags |= IFF_PROMISC; |
5457 | dev->promiscuity += inc; | |
5458 | if (dev->promiscuity == 0) { | |
5459 | /* | |
5460 | * Avoid overflow. | |
5461 | * If inc causes overflow, untouch promisc and return error. | |
5462 | */ | |
5463 | if (inc < 0) | |
5464 | dev->flags &= ~IFF_PROMISC; | |
5465 | else { | |
5466 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
5467 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
5468 | dev->name); | |
dad9b335 WC |
5469 | return -EOVERFLOW; |
5470 | } | |
5471 | } | |
52609c0b | 5472 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
5473 | pr_info("device %s %s promiscuous mode\n", |
5474 | dev->name, | |
5475 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
5476 | if (audit_enabled) { |
5477 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
5478 | audit_log(current->audit_context, GFP_ATOMIC, |
5479 | AUDIT_ANOM_PROMISCUOUS, | |
5480 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
5481 | dev->name, (dev->flags & IFF_PROMISC), | |
5482 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 5483 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
5484 | from_kuid(&init_user_ns, uid), |
5485 | from_kgid(&init_user_ns, gid), | |
7759db82 | 5486 | audit_get_sessionid(current)); |
8192b0c4 | 5487 | } |
24023451 | 5488 | |
b6c40d68 | 5489 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 5490 | } |
991fb3f7 ND |
5491 | if (notify) |
5492 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 5493 | return 0; |
1da177e4 LT |
5494 | } |
5495 | ||
4417da66 PM |
5496 | /** |
5497 | * dev_set_promiscuity - update promiscuity count on a device | |
5498 | * @dev: device | |
5499 | * @inc: modifier | |
5500 | * | |
5501 | * Add or remove promiscuity from a device. While the count in the device | |
5502 | * remains above zero the interface remains promiscuous. Once it hits zero | |
5503 | * the device reverts back to normal filtering operation. A negative inc | |
5504 | * value is used to drop promiscuity on the device. | |
dad9b335 | 5505 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 5506 | */ |
dad9b335 | 5507 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 5508 | { |
b536db93 | 5509 | unsigned int old_flags = dev->flags; |
dad9b335 | 5510 | int err; |
4417da66 | 5511 | |
991fb3f7 | 5512 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 5513 | if (err < 0) |
dad9b335 | 5514 | return err; |
4417da66 PM |
5515 | if (dev->flags != old_flags) |
5516 | dev_set_rx_mode(dev); | |
dad9b335 | 5517 | return err; |
4417da66 | 5518 | } |
d1b19dff | 5519 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 5520 | |
991fb3f7 | 5521 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5522 | { |
991fb3f7 | 5523 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 5524 | |
24023451 PM |
5525 | ASSERT_RTNL(); |
5526 | ||
1da177e4 | 5527 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
5528 | dev->allmulti += inc; |
5529 | if (dev->allmulti == 0) { | |
5530 | /* | |
5531 | * Avoid overflow. | |
5532 | * If inc causes overflow, untouch allmulti and return error. | |
5533 | */ | |
5534 | if (inc < 0) | |
5535 | dev->flags &= ~IFF_ALLMULTI; | |
5536 | else { | |
5537 | dev->allmulti -= inc; | |
7b6cd1ce JP |
5538 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
5539 | dev->name); | |
dad9b335 WC |
5540 | return -EOVERFLOW; |
5541 | } | |
5542 | } | |
24023451 | 5543 | if (dev->flags ^ old_flags) { |
b6c40d68 | 5544 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 5545 | dev_set_rx_mode(dev); |
991fb3f7 ND |
5546 | if (notify) |
5547 | __dev_notify_flags(dev, old_flags, | |
5548 | dev->gflags ^ old_gflags); | |
24023451 | 5549 | } |
dad9b335 | 5550 | return 0; |
4417da66 | 5551 | } |
991fb3f7 ND |
5552 | |
5553 | /** | |
5554 | * dev_set_allmulti - update allmulti count on a device | |
5555 | * @dev: device | |
5556 | * @inc: modifier | |
5557 | * | |
5558 | * Add or remove reception of all multicast frames to a device. While the | |
5559 | * count in the device remains above zero the interface remains listening | |
5560 | * to all interfaces. Once it hits zero the device reverts back to normal | |
5561 | * filtering operation. A negative @inc value is used to drop the counter | |
5562 | * when releasing a resource needing all multicasts. | |
5563 | * Return 0 if successful or a negative errno code on error. | |
5564 | */ | |
5565 | ||
5566 | int dev_set_allmulti(struct net_device *dev, int inc) | |
5567 | { | |
5568 | return __dev_set_allmulti(dev, inc, true); | |
5569 | } | |
d1b19dff | 5570 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
5571 | |
5572 | /* | |
5573 | * Upload unicast and multicast address lists to device and | |
5574 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 5575 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
5576 | * are present. |
5577 | */ | |
5578 | void __dev_set_rx_mode(struct net_device *dev) | |
5579 | { | |
d314774c SH |
5580 | const struct net_device_ops *ops = dev->netdev_ops; |
5581 | ||
4417da66 PM |
5582 | /* dev_open will call this function so the list will stay sane. */ |
5583 | if (!(dev->flags&IFF_UP)) | |
5584 | return; | |
5585 | ||
5586 | if (!netif_device_present(dev)) | |
40b77c94 | 5587 | return; |
4417da66 | 5588 | |
01789349 | 5589 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
5590 | /* Unicast addresses changes may only happen under the rtnl, |
5591 | * therefore calling __dev_set_promiscuity here is safe. | |
5592 | */ | |
32e7bfc4 | 5593 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 5594 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 5595 | dev->uc_promisc = true; |
32e7bfc4 | 5596 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 5597 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 5598 | dev->uc_promisc = false; |
4417da66 | 5599 | } |
4417da66 | 5600 | } |
01789349 JP |
5601 | |
5602 | if (ops->ndo_set_rx_mode) | |
5603 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
5604 | } |
5605 | ||
5606 | void dev_set_rx_mode(struct net_device *dev) | |
5607 | { | |
b9e40857 | 5608 | netif_addr_lock_bh(dev); |
4417da66 | 5609 | __dev_set_rx_mode(dev); |
b9e40857 | 5610 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
5611 | } |
5612 | ||
f0db275a SH |
5613 | /** |
5614 | * dev_get_flags - get flags reported to userspace | |
5615 | * @dev: device | |
5616 | * | |
5617 | * Get the combination of flag bits exported through APIs to userspace. | |
5618 | */ | |
95c96174 | 5619 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 5620 | { |
95c96174 | 5621 | unsigned int flags; |
1da177e4 LT |
5622 | |
5623 | flags = (dev->flags & ~(IFF_PROMISC | | |
5624 | IFF_ALLMULTI | | |
b00055aa SR |
5625 | IFF_RUNNING | |
5626 | IFF_LOWER_UP | | |
5627 | IFF_DORMANT)) | | |
1da177e4 LT |
5628 | (dev->gflags & (IFF_PROMISC | |
5629 | IFF_ALLMULTI)); | |
5630 | ||
b00055aa SR |
5631 | if (netif_running(dev)) { |
5632 | if (netif_oper_up(dev)) | |
5633 | flags |= IFF_RUNNING; | |
5634 | if (netif_carrier_ok(dev)) | |
5635 | flags |= IFF_LOWER_UP; | |
5636 | if (netif_dormant(dev)) | |
5637 | flags |= IFF_DORMANT; | |
5638 | } | |
1da177e4 LT |
5639 | |
5640 | return flags; | |
5641 | } | |
d1b19dff | 5642 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 5643 | |
bd380811 | 5644 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 5645 | { |
b536db93 | 5646 | unsigned int old_flags = dev->flags; |
bd380811 | 5647 | int ret; |
1da177e4 | 5648 | |
24023451 PM |
5649 | ASSERT_RTNL(); |
5650 | ||
1da177e4 LT |
5651 | /* |
5652 | * Set the flags on our device. | |
5653 | */ | |
5654 | ||
5655 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
5656 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
5657 | IFF_AUTOMEDIA)) | | |
5658 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
5659 | IFF_ALLMULTI)); | |
5660 | ||
5661 | /* | |
5662 | * Load in the correct multicast list now the flags have changed. | |
5663 | */ | |
5664 | ||
b6c40d68 PM |
5665 | if ((old_flags ^ flags) & IFF_MULTICAST) |
5666 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 5667 | |
4417da66 | 5668 | dev_set_rx_mode(dev); |
1da177e4 LT |
5669 | |
5670 | /* | |
5671 | * Have we downed the interface. We handle IFF_UP ourselves | |
5672 | * according to user attempts to set it, rather than blindly | |
5673 | * setting it. | |
5674 | */ | |
5675 | ||
5676 | ret = 0; | |
d215d10f | 5677 | if ((old_flags ^ flags) & IFF_UP) |
bd380811 | 5678 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 | 5679 | |
1da177e4 | 5680 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 5681 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 5682 | unsigned int old_flags = dev->flags; |
d1b19dff | 5683 | |
1da177e4 | 5684 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
5685 | |
5686 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
5687 | if (dev->flags != old_flags) | |
5688 | dev_set_rx_mode(dev); | |
1da177e4 LT |
5689 | } |
5690 | ||
5691 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
5692 | is important. Some (broken) drivers set IFF_PROMISC, when | |
5693 | IFF_ALLMULTI is requested not asking us and not reporting. | |
5694 | */ | |
5695 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
5696 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
5697 | ||
1da177e4 | 5698 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 5699 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
5700 | } |
5701 | ||
bd380811 PM |
5702 | return ret; |
5703 | } | |
5704 | ||
a528c219 ND |
5705 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
5706 | unsigned int gchanges) | |
bd380811 PM |
5707 | { |
5708 | unsigned int changes = dev->flags ^ old_flags; | |
5709 | ||
a528c219 | 5710 | if (gchanges) |
7f294054 | 5711 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 5712 | |
bd380811 PM |
5713 | if (changes & IFF_UP) { |
5714 | if (dev->flags & IFF_UP) | |
5715 | call_netdevice_notifiers(NETDEV_UP, dev); | |
5716 | else | |
5717 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
5718 | } | |
5719 | ||
5720 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
5721 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
5722 | struct netdev_notifier_change_info change_info; | |
5723 | ||
5724 | change_info.flags_changed = changes; | |
5725 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
5726 | &change_info.info); | |
5727 | } | |
bd380811 PM |
5728 | } |
5729 | ||
5730 | /** | |
5731 | * dev_change_flags - change device settings | |
5732 | * @dev: device | |
5733 | * @flags: device state flags | |
5734 | * | |
5735 | * Change settings on device based state flags. The flags are | |
5736 | * in the userspace exported format. | |
5737 | */ | |
b536db93 | 5738 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 5739 | { |
b536db93 | 5740 | int ret; |
991fb3f7 | 5741 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
5742 | |
5743 | ret = __dev_change_flags(dev, flags); | |
5744 | if (ret < 0) | |
5745 | return ret; | |
5746 | ||
991fb3f7 | 5747 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 5748 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
5749 | return ret; |
5750 | } | |
d1b19dff | 5751 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 5752 | |
2315dc91 VF |
5753 | static int __dev_set_mtu(struct net_device *dev, int new_mtu) |
5754 | { | |
5755 | const struct net_device_ops *ops = dev->netdev_ops; | |
5756 | ||
5757 | if (ops->ndo_change_mtu) | |
5758 | return ops->ndo_change_mtu(dev, new_mtu); | |
5759 | ||
5760 | dev->mtu = new_mtu; | |
5761 | return 0; | |
5762 | } | |
5763 | ||
f0db275a SH |
5764 | /** |
5765 | * dev_set_mtu - Change maximum transfer unit | |
5766 | * @dev: device | |
5767 | * @new_mtu: new transfer unit | |
5768 | * | |
5769 | * Change the maximum transfer size of the network device. | |
5770 | */ | |
1da177e4 LT |
5771 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
5772 | { | |
2315dc91 | 5773 | int err, orig_mtu; |
1da177e4 LT |
5774 | |
5775 | if (new_mtu == dev->mtu) | |
5776 | return 0; | |
5777 | ||
5778 | /* MTU must be positive. */ | |
5779 | if (new_mtu < 0) | |
5780 | return -EINVAL; | |
5781 | ||
5782 | if (!netif_device_present(dev)) | |
5783 | return -ENODEV; | |
5784 | ||
1d486bfb VF |
5785 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
5786 | err = notifier_to_errno(err); | |
5787 | if (err) | |
5788 | return err; | |
d314774c | 5789 | |
2315dc91 VF |
5790 | orig_mtu = dev->mtu; |
5791 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 5792 | |
2315dc91 VF |
5793 | if (!err) { |
5794 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
5795 | err = notifier_to_errno(err); | |
5796 | if (err) { | |
5797 | /* setting mtu back and notifying everyone again, | |
5798 | * so that they have a chance to revert changes. | |
5799 | */ | |
5800 | __dev_set_mtu(dev, orig_mtu); | |
5801 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
5802 | } | |
5803 | } | |
1da177e4 LT |
5804 | return err; |
5805 | } | |
d1b19dff | 5806 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 5807 | |
cbda10fa VD |
5808 | /** |
5809 | * dev_set_group - Change group this device belongs to | |
5810 | * @dev: device | |
5811 | * @new_group: group this device should belong to | |
5812 | */ | |
5813 | void dev_set_group(struct net_device *dev, int new_group) | |
5814 | { | |
5815 | dev->group = new_group; | |
5816 | } | |
5817 | EXPORT_SYMBOL(dev_set_group); | |
5818 | ||
f0db275a SH |
5819 | /** |
5820 | * dev_set_mac_address - Change Media Access Control Address | |
5821 | * @dev: device | |
5822 | * @sa: new address | |
5823 | * | |
5824 | * Change the hardware (MAC) address of the device | |
5825 | */ | |
1da177e4 LT |
5826 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
5827 | { | |
d314774c | 5828 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
5829 | int err; |
5830 | ||
d314774c | 5831 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
5832 | return -EOPNOTSUPP; |
5833 | if (sa->sa_family != dev->type) | |
5834 | return -EINVAL; | |
5835 | if (!netif_device_present(dev)) | |
5836 | return -ENODEV; | |
d314774c | 5837 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
5838 | if (err) |
5839 | return err; | |
fbdeca2d | 5840 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 5841 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 5842 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 5843 | return 0; |
1da177e4 | 5844 | } |
d1b19dff | 5845 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 5846 | |
4bf84c35 JP |
5847 | /** |
5848 | * dev_change_carrier - Change device carrier | |
5849 | * @dev: device | |
691b3b7e | 5850 | * @new_carrier: new value |
4bf84c35 JP |
5851 | * |
5852 | * Change device carrier | |
5853 | */ | |
5854 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
5855 | { | |
5856 | const struct net_device_ops *ops = dev->netdev_ops; | |
5857 | ||
5858 | if (!ops->ndo_change_carrier) | |
5859 | return -EOPNOTSUPP; | |
5860 | if (!netif_device_present(dev)) | |
5861 | return -ENODEV; | |
5862 | return ops->ndo_change_carrier(dev, new_carrier); | |
5863 | } | |
5864 | EXPORT_SYMBOL(dev_change_carrier); | |
5865 | ||
66b52b0d JP |
5866 | /** |
5867 | * dev_get_phys_port_id - Get device physical port ID | |
5868 | * @dev: device | |
5869 | * @ppid: port ID | |
5870 | * | |
5871 | * Get device physical port ID | |
5872 | */ | |
5873 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 5874 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
5875 | { |
5876 | const struct net_device_ops *ops = dev->netdev_ops; | |
5877 | ||
5878 | if (!ops->ndo_get_phys_port_id) | |
5879 | return -EOPNOTSUPP; | |
5880 | return ops->ndo_get_phys_port_id(dev, ppid); | |
5881 | } | |
5882 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
5883 | ||
1da177e4 LT |
5884 | /** |
5885 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5886 | * @net: the applicable net namespace |
1da177e4 LT |
5887 | * |
5888 | * Returns a suitable unique value for a new device interface | |
5889 | * number. The caller must hold the rtnl semaphore or the | |
5890 | * dev_base_lock to be sure it remains unique. | |
5891 | */ | |
881d966b | 5892 | static int dev_new_index(struct net *net) |
1da177e4 | 5893 | { |
aa79e66e | 5894 | int ifindex = net->ifindex; |
1da177e4 LT |
5895 | for (;;) { |
5896 | if (++ifindex <= 0) | |
5897 | ifindex = 1; | |
881d966b | 5898 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 5899 | return net->ifindex = ifindex; |
1da177e4 LT |
5900 | } |
5901 | } | |
5902 | ||
1da177e4 | 5903 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5904 | static LIST_HEAD(net_todo_list); |
200b916f | 5905 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 5906 | |
6f05f629 | 5907 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5908 | { |
1da177e4 | 5909 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 5910 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
5911 | } |
5912 | ||
9b5e383c | 5913 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5914 | { |
e93737b0 | 5915 | struct net_device *dev, *tmp; |
5cde2829 | 5916 | LIST_HEAD(close_head); |
9b5e383c | 5917 | |
93ee31f1 DL |
5918 | BUG_ON(dev_boot_phase); |
5919 | ASSERT_RTNL(); | |
5920 | ||
e93737b0 | 5921 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5922 | /* Some devices call without registering |
e93737b0 KK |
5923 | * for initialization unwind. Remove those |
5924 | * devices and proceed with the remaining. | |
9b5e383c ED |
5925 | */ |
5926 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
5927 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
5928 | dev->name, dev); | |
93ee31f1 | 5929 | |
9b5e383c | 5930 | WARN_ON(1); |
e93737b0 KK |
5931 | list_del(&dev->unreg_list); |
5932 | continue; | |
9b5e383c | 5933 | } |
449f4544 | 5934 | dev->dismantle = true; |
9b5e383c | 5935 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5936 | } |
93ee31f1 | 5937 | |
44345724 | 5938 | /* If device is running, close it first. */ |
5cde2829 EB |
5939 | list_for_each_entry(dev, head, unreg_list) |
5940 | list_add_tail(&dev->close_list, &close_head); | |
5941 | dev_close_many(&close_head); | |
93ee31f1 | 5942 | |
44345724 | 5943 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5944 | /* And unlink it from device chain. */ |
5945 | unlist_netdevice(dev); | |
93ee31f1 | 5946 | |
9b5e383c ED |
5947 | dev->reg_state = NETREG_UNREGISTERING; |
5948 | } | |
93ee31f1 DL |
5949 | |
5950 | synchronize_net(); | |
5951 | ||
9b5e383c | 5952 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
5953 | struct sk_buff *skb = NULL; |
5954 | ||
9b5e383c ED |
5955 | /* Shutdown queueing discipline. */ |
5956 | dev_shutdown(dev); | |
93ee31f1 DL |
5957 | |
5958 | ||
9b5e383c ED |
5959 | /* Notify protocols, that we are about to destroy |
5960 | this device. They should clean all the things. | |
5961 | */ | |
5962 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5963 | |
395eea6c MB |
5964 | if (!dev->rtnl_link_ops || |
5965 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5966 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, | |
5967 | GFP_KERNEL); | |
5968 | ||
9b5e383c ED |
5969 | /* |
5970 | * Flush the unicast and multicast chains | |
5971 | */ | |
a748ee24 | 5972 | dev_uc_flush(dev); |
22bedad3 | 5973 | dev_mc_flush(dev); |
93ee31f1 | 5974 | |
9b5e383c ED |
5975 | if (dev->netdev_ops->ndo_uninit) |
5976 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5977 | |
395eea6c MB |
5978 | if (skb) |
5979 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 5980 | |
9ff162a8 JP |
5981 | /* Notifier chain MUST detach us all upper devices. */ |
5982 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 5983 | |
9b5e383c ED |
5984 | /* Remove entries from kobject tree */ |
5985 | netdev_unregister_kobject(dev); | |
024e9679 AD |
5986 | #ifdef CONFIG_XPS |
5987 | /* Remove XPS queueing entries */ | |
5988 | netif_reset_xps_queues_gt(dev, 0); | |
5989 | #endif | |
9b5e383c | 5990 | } |
93ee31f1 | 5991 | |
850a545b | 5992 | synchronize_net(); |
395264d5 | 5993 | |
a5ee1551 | 5994 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5995 | dev_put(dev); |
5996 | } | |
5997 | ||
5998 | static void rollback_registered(struct net_device *dev) | |
5999 | { | |
6000 | LIST_HEAD(single); | |
6001 | ||
6002 | list_add(&dev->unreg_list, &single); | |
6003 | rollback_registered_many(&single); | |
ceaaec98 | 6004 | list_del(&single); |
93ee31f1 DL |
6005 | } |
6006 | ||
c8f44aff MM |
6007 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
6008 | netdev_features_t features) | |
b63365a2 | 6009 | { |
57422dc5 MM |
6010 | /* Fix illegal checksum combinations */ |
6011 | if ((features & NETIF_F_HW_CSUM) && | |
6012 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 6013 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
6014 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
6015 | } | |
6016 | ||
b63365a2 | 6017 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 6018 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 6019 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 6020 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
6021 | } |
6022 | ||
ec5f0615 PS |
6023 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
6024 | !(features & NETIF_F_IP_CSUM)) { | |
6025 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
6026 | features &= ~NETIF_F_TSO; | |
6027 | features &= ~NETIF_F_TSO_ECN; | |
6028 | } | |
6029 | ||
6030 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
6031 | !(features & NETIF_F_IPV6_CSUM)) { | |
6032 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
6033 | features &= ~NETIF_F_TSO6; | |
6034 | } | |
6035 | ||
31d8b9e0 BH |
6036 | /* TSO ECN requires that TSO is present as well. */ |
6037 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
6038 | features &= ~NETIF_F_TSO_ECN; | |
6039 | ||
212b573f MM |
6040 | /* Software GSO depends on SG. */ |
6041 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 6042 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
6043 | features &= ~NETIF_F_GSO; |
6044 | } | |
6045 | ||
acd1130e | 6046 | /* UFO needs SG and checksumming */ |
b63365a2 | 6047 | if (features & NETIF_F_UFO) { |
79032644 MM |
6048 | /* maybe split UFO into V4 and V6? */ |
6049 | if (!((features & NETIF_F_GEN_CSUM) || | |
6050 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
6051 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 6052 | netdev_dbg(dev, |
acd1130e | 6053 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
6054 | features &= ~NETIF_F_UFO; |
6055 | } | |
6056 | ||
6057 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 6058 | netdev_dbg(dev, |
acd1130e | 6059 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
6060 | features &= ~NETIF_F_UFO; |
6061 | } | |
6062 | } | |
6063 | ||
d0290214 JP |
6064 | #ifdef CONFIG_NET_RX_BUSY_POLL |
6065 | if (dev->netdev_ops->ndo_busy_poll) | |
6066 | features |= NETIF_F_BUSY_POLL; | |
6067 | else | |
6068 | #endif | |
6069 | features &= ~NETIF_F_BUSY_POLL; | |
6070 | ||
b63365a2 HX |
6071 | return features; |
6072 | } | |
b63365a2 | 6073 | |
6cb6a27c | 6074 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 6075 | { |
c8f44aff | 6076 | netdev_features_t features; |
5455c699 MM |
6077 | int err = 0; |
6078 | ||
87267485 MM |
6079 | ASSERT_RTNL(); |
6080 | ||
5455c699 MM |
6081 | features = netdev_get_wanted_features(dev); |
6082 | ||
6083 | if (dev->netdev_ops->ndo_fix_features) | |
6084 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
6085 | ||
6086 | /* driver might be less strict about feature dependencies */ | |
6087 | features = netdev_fix_features(dev, features); | |
6088 | ||
6089 | if (dev->features == features) | |
6cb6a27c | 6090 | return 0; |
5455c699 | 6091 | |
c8f44aff MM |
6092 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
6093 | &dev->features, &features); | |
5455c699 MM |
6094 | |
6095 | if (dev->netdev_ops->ndo_set_features) | |
6096 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
6097 | ||
6cb6a27c | 6098 | if (unlikely(err < 0)) { |
5455c699 | 6099 | netdev_err(dev, |
c8f44aff MM |
6100 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
6101 | err, &features, &dev->features); | |
6cb6a27c MM |
6102 | return -1; |
6103 | } | |
6104 | ||
6105 | if (!err) | |
6106 | dev->features = features; | |
6107 | ||
6108 | return 1; | |
6109 | } | |
6110 | ||
afe12cc8 MM |
6111 | /** |
6112 | * netdev_update_features - recalculate device features | |
6113 | * @dev: the device to check | |
6114 | * | |
6115 | * Recalculate dev->features set and send notifications if it | |
6116 | * has changed. Should be called after driver or hardware dependent | |
6117 | * conditions might have changed that influence the features. | |
6118 | */ | |
6cb6a27c MM |
6119 | void netdev_update_features(struct net_device *dev) |
6120 | { | |
6121 | if (__netdev_update_features(dev)) | |
6122 | netdev_features_change(dev); | |
5455c699 MM |
6123 | } |
6124 | EXPORT_SYMBOL(netdev_update_features); | |
6125 | ||
afe12cc8 MM |
6126 | /** |
6127 | * netdev_change_features - recalculate device features | |
6128 | * @dev: the device to check | |
6129 | * | |
6130 | * Recalculate dev->features set and send notifications even | |
6131 | * if they have not changed. Should be called instead of | |
6132 | * netdev_update_features() if also dev->vlan_features might | |
6133 | * have changed to allow the changes to be propagated to stacked | |
6134 | * VLAN devices. | |
6135 | */ | |
6136 | void netdev_change_features(struct net_device *dev) | |
6137 | { | |
6138 | __netdev_update_features(dev); | |
6139 | netdev_features_change(dev); | |
6140 | } | |
6141 | EXPORT_SYMBOL(netdev_change_features); | |
6142 | ||
fc4a7489 PM |
6143 | /** |
6144 | * netif_stacked_transfer_operstate - transfer operstate | |
6145 | * @rootdev: the root or lower level device to transfer state from | |
6146 | * @dev: the device to transfer operstate to | |
6147 | * | |
6148 | * Transfer operational state from root to device. This is normally | |
6149 | * called when a stacking relationship exists between the root | |
6150 | * device and the device(a leaf device). | |
6151 | */ | |
6152 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
6153 | struct net_device *dev) | |
6154 | { | |
6155 | if (rootdev->operstate == IF_OPER_DORMANT) | |
6156 | netif_dormant_on(dev); | |
6157 | else | |
6158 | netif_dormant_off(dev); | |
6159 | ||
6160 | if (netif_carrier_ok(rootdev)) { | |
6161 | if (!netif_carrier_ok(dev)) | |
6162 | netif_carrier_on(dev); | |
6163 | } else { | |
6164 | if (netif_carrier_ok(dev)) | |
6165 | netif_carrier_off(dev); | |
6166 | } | |
6167 | } | |
6168 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
6169 | ||
a953be53 | 6170 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
6171 | static int netif_alloc_rx_queues(struct net_device *dev) |
6172 | { | |
1b4bf461 | 6173 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 6174 | struct netdev_rx_queue *rx; |
10595902 | 6175 | size_t sz = count * sizeof(*rx); |
1b4bf461 | 6176 | |
bd25fa7b | 6177 | BUG_ON(count < 1); |
1b4bf461 | 6178 | |
10595902 PG |
6179 | rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6180 | if (!rx) { | |
6181 | rx = vzalloc(sz); | |
6182 | if (!rx) | |
6183 | return -ENOMEM; | |
6184 | } | |
bd25fa7b TH |
6185 | dev->_rx = rx; |
6186 | ||
bd25fa7b | 6187 | for (i = 0; i < count; i++) |
fe822240 | 6188 | rx[i].dev = dev; |
1b4bf461 ED |
6189 | return 0; |
6190 | } | |
bf264145 | 6191 | #endif |
1b4bf461 | 6192 | |
aa942104 CG |
6193 | static void netdev_init_one_queue(struct net_device *dev, |
6194 | struct netdev_queue *queue, void *_unused) | |
6195 | { | |
6196 | /* Initialize queue lock */ | |
6197 | spin_lock_init(&queue->_xmit_lock); | |
6198 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
6199 | queue->xmit_lock_owner = -1; | |
b236da69 | 6200 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 6201 | queue->dev = dev; |
114cf580 TH |
6202 | #ifdef CONFIG_BQL |
6203 | dql_init(&queue->dql, HZ); | |
6204 | #endif | |
aa942104 CG |
6205 | } |
6206 | ||
60877a32 ED |
6207 | static void netif_free_tx_queues(struct net_device *dev) |
6208 | { | |
4cb28970 | 6209 | kvfree(dev->_tx); |
60877a32 ED |
6210 | } |
6211 | ||
e6484930 TH |
6212 | static int netif_alloc_netdev_queues(struct net_device *dev) |
6213 | { | |
6214 | unsigned int count = dev->num_tx_queues; | |
6215 | struct netdev_queue *tx; | |
60877a32 | 6216 | size_t sz = count * sizeof(*tx); |
e6484930 | 6217 | |
60877a32 | 6218 | BUG_ON(count < 1 || count > 0xffff); |
62b5942a | 6219 | |
60877a32 ED |
6220 | tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6221 | if (!tx) { | |
6222 | tx = vzalloc(sz); | |
6223 | if (!tx) | |
6224 | return -ENOMEM; | |
6225 | } | |
e6484930 | 6226 | dev->_tx = tx; |
1d24eb48 | 6227 | |
e6484930 TH |
6228 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
6229 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
6230 | |
6231 | return 0; | |
e6484930 TH |
6232 | } |
6233 | ||
1da177e4 LT |
6234 | /** |
6235 | * register_netdevice - register a network device | |
6236 | * @dev: device to register | |
6237 | * | |
6238 | * Take a completed network device structure and add it to the kernel | |
6239 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6240 | * chain. 0 is returned on success. A negative errno code is returned | |
6241 | * on a failure to set up the device, or if the name is a duplicate. | |
6242 | * | |
6243 | * Callers must hold the rtnl semaphore. You may want | |
6244 | * register_netdev() instead of this. | |
6245 | * | |
6246 | * BUGS: | |
6247 | * The locking appears insufficient to guarantee two parallel registers | |
6248 | * will not get the same name. | |
6249 | */ | |
6250 | ||
6251 | int register_netdevice(struct net_device *dev) | |
6252 | { | |
1da177e4 | 6253 | int ret; |
d314774c | 6254 | struct net *net = dev_net(dev); |
1da177e4 LT |
6255 | |
6256 | BUG_ON(dev_boot_phase); | |
6257 | ASSERT_RTNL(); | |
6258 | ||
b17a7c17 SH |
6259 | might_sleep(); |
6260 | ||
1da177e4 LT |
6261 | /* When net_device's are persistent, this will be fatal. */ |
6262 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 6263 | BUG_ON(!net); |
1da177e4 | 6264 | |
f1f28aa3 | 6265 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 6266 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 6267 | |
1da177e4 LT |
6268 | dev->iflink = -1; |
6269 | ||
828de4f6 | 6270 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
6271 | if (ret < 0) |
6272 | goto out; | |
6273 | ||
1da177e4 | 6274 | /* Init, if this function is available */ |
d314774c SH |
6275 | if (dev->netdev_ops->ndo_init) { |
6276 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
6277 | if (ret) { |
6278 | if (ret > 0) | |
6279 | ret = -EIO; | |
90833aa4 | 6280 | goto out; |
1da177e4 LT |
6281 | } |
6282 | } | |
4ec93edb | 6283 | |
f646968f PM |
6284 | if (((dev->hw_features | dev->features) & |
6285 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
6286 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
6287 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
6288 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
6289 | ret = -EINVAL; | |
6290 | goto err_uninit; | |
6291 | } | |
6292 | ||
9c7dafbf PE |
6293 | ret = -EBUSY; |
6294 | if (!dev->ifindex) | |
6295 | dev->ifindex = dev_new_index(net); | |
6296 | else if (__dev_get_by_index(net, dev->ifindex)) | |
6297 | goto err_uninit; | |
6298 | ||
1da177e4 LT |
6299 | if (dev->iflink == -1) |
6300 | dev->iflink = dev->ifindex; | |
6301 | ||
5455c699 MM |
6302 | /* Transfer changeable features to wanted_features and enable |
6303 | * software offloads (GSO and GRO). | |
6304 | */ | |
6305 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
6306 | dev->features |= NETIF_F_SOFT_FEATURES; |
6307 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 6308 | |
34324dc2 MM |
6309 | if (!(dev->flags & IFF_LOOPBACK)) { |
6310 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
c6e1a0d1 TH |
6311 | } |
6312 | ||
1180e7d6 | 6313 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 6314 | */ |
1180e7d6 | 6315 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 6316 | |
ee579677 PS |
6317 | /* Make NETIF_F_SG inheritable to tunnel devices. |
6318 | */ | |
6319 | dev->hw_enc_features |= NETIF_F_SG; | |
6320 | ||
0d89d203 SH |
6321 | /* Make NETIF_F_SG inheritable to MPLS. |
6322 | */ | |
6323 | dev->mpls_features |= NETIF_F_SG; | |
6324 | ||
7ffbe3fd JB |
6325 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
6326 | ret = notifier_to_errno(ret); | |
6327 | if (ret) | |
6328 | goto err_uninit; | |
6329 | ||
8b41d188 | 6330 | ret = netdev_register_kobject(dev); |
b17a7c17 | 6331 | if (ret) |
7ce1b0ed | 6332 | goto err_uninit; |
b17a7c17 SH |
6333 | dev->reg_state = NETREG_REGISTERED; |
6334 | ||
6cb6a27c | 6335 | __netdev_update_features(dev); |
8e9b59b2 | 6336 | |
1da177e4 LT |
6337 | /* |
6338 | * Default initial state at registry is that the | |
6339 | * device is present. | |
6340 | */ | |
6341 | ||
6342 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6343 | ||
8f4cccbb BH |
6344 | linkwatch_init_dev(dev); |
6345 | ||
1da177e4 | 6346 | dev_init_scheduler(dev); |
1da177e4 | 6347 | dev_hold(dev); |
ce286d32 | 6348 | list_netdevice(dev); |
7bf23575 | 6349 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 6350 | |
948b337e JP |
6351 | /* If the device has permanent device address, driver should |
6352 | * set dev_addr and also addr_assign_type should be set to | |
6353 | * NET_ADDR_PERM (default value). | |
6354 | */ | |
6355 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
6356 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
6357 | ||
1da177e4 | 6358 | /* Notify protocols, that a new device appeared. */ |
056925ab | 6359 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 6360 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
6361 | if (ret) { |
6362 | rollback_registered(dev); | |
6363 | dev->reg_state = NETREG_UNREGISTERED; | |
6364 | } | |
d90a909e EB |
6365 | /* |
6366 | * Prevent userspace races by waiting until the network | |
6367 | * device is fully setup before sending notifications. | |
6368 | */ | |
a2835763 PM |
6369 | if (!dev->rtnl_link_ops || |
6370 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 6371 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
6372 | |
6373 | out: | |
6374 | return ret; | |
7ce1b0ed HX |
6375 | |
6376 | err_uninit: | |
d314774c SH |
6377 | if (dev->netdev_ops->ndo_uninit) |
6378 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 6379 | goto out; |
1da177e4 | 6380 | } |
d1b19dff | 6381 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 6382 | |
937f1ba5 BH |
6383 | /** |
6384 | * init_dummy_netdev - init a dummy network device for NAPI | |
6385 | * @dev: device to init | |
6386 | * | |
6387 | * This takes a network device structure and initialize the minimum | |
6388 | * amount of fields so it can be used to schedule NAPI polls without | |
6389 | * registering a full blown interface. This is to be used by drivers | |
6390 | * that need to tie several hardware interfaces to a single NAPI | |
6391 | * poll scheduler due to HW limitations. | |
6392 | */ | |
6393 | int init_dummy_netdev(struct net_device *dev) | |
6394 | { | |
6395 | /* Clear everything. Note we don't initialize spinlocks | |
6396 | * are they aren't supposed to be taken by any of the | |
6397 | * NAPI code and this dummy netdev is supposed to be | |
6398 | * only ever used for NAPI polls | |
6399 | */ | |
6400 | memset(dev, 0, sizeof(struct net_device)); | |
6401 | ||
6402 | /* make sure we BUG if trying to hit standard | |
6403 | * register/unregister code path | |
6404 | */ | |
6405 | dev->reg_state = NETREG_DUMMY; | |
6406 | ||
937f1ba5 BH |
6407 | /* NAPI wants this */ |
6408 | INIT_LIST_HEAD(&dev->napi_list); | |
6409 | ||
6410 | /* a dummy interface is started by default */ | |
6411 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6412 | set_bit(__LINK_STATE_START, &dev->state); | |
6413 | ||
29b4433d ED |
6414 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
6415 | * because users of this 'device' dont need to change | |
6416 | * its refcount. | |
6417 | */ | |
6418 | ||
937f1ba5 BH |
6419 | return 0; |
6420 | } | |
6421 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
6422 | ||
6423 | ||
1da177e4 LT |
6424 | /** |
6425 | * register_netdev - register a network device | |
6426 | * @dev: device to register | |
6427 | * | |
6428 | * Take a completed network device structure and add it to the kernel | |
6429 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6430 | * chain. 0 is returned on success. A negative errno code is returned | |
6431 | * on a failure to set up the device, or if the name is a duplicate. | |
6432 | * | |
38b4da38 | 6433 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
6434 | * and expands the device name if you passed a format string to |
6435 | * alloc_netdev. | |
6436 | */ | |
6437 | int register_netdev(struct net_device *dev) | |
6438 | { | |
6439 | int err; | |
6440 | ||
6441 | rtnl_lock(); | |
1da177e4 | 6442 | err = register_netdevice(dev); |
1da177e4 LT |
6443 | rtnl_unlock(); |
6444 | return err; | |
6445 | } | |
6446 | EXPORT_SYMBOL(register_netdev); | |
6447 | ||
29b4433d ED |
6448 | int netdev_refcnt_read(const struct net_device *dev) |
6449 | { | |
6450 | int i, refcnt = 0; | |
6451 | ||
6452 | for_each_possible_cpu(i) | |
6453 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
6454 | return refcnt; | |
6455 | } | |
6456 | EXPORT_SYMBOL(netdev_refcnt_read); | |
6457 | ||
2c53040f | 6458 | /** |
1da177e4 | 6459 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 6460 | * @dev: target net_device |
1da177e4 LT |
6461 | * |
6462 | * This is called when unregistering network devices. | |
6463 | * | |
6464 | * Any protocol or device that holds a reference should register | |
6465 | * for netdevice notification, and cleanup and put back the | |
6466 | * reference if they receive an UNREGISTER event. | |
6467 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 6468 | * call dev_put. |
1da177e4 LT |
6469 | */ |
6470 | static void netdev_wait_allrefs(struct net_device *dev) | |
6471 | { | |
6472 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 6473 | int refcnt; |
1da177e4 | 6474 | |
e014debe ED |
6475 | linkwatch_forget_dev(dev); |
6476 | ||
1da177e4 | 6477 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
6478 | refcnt = netdev_refcnt_read(dev); |
6479 | ||
6480 | while (refcnt != 0) { | |
1da177e4 | 6481 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 6482 | rtnl_lock(); |
1da177e4 LT |
6483 | |
6484 | /* Rebroadcast unregister notification */ | |
056925ab | 6485 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 6486 | |
748e2d93 | 6487 | __rtnl_unlock(); |
0115e8e3 | 6488 | rcu_barrier(); |
748e2d93 ED |
6489 | rtnl_lock(); |
6490 | ||
0115e8e3 | 6491 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
6492 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
6493 | &dev->state)) { | |
6494 | /* We must not have linkwatch events | |
6495 | * pending on unregister. If this | |
6496 | * happens, we simply run the queue | |
6497 | * unscheduled, resulting in a noop | |
6498 | * for this device. | |
6499 | */ | |
6500 | linkwatch_run_queue(); | |
6501 | } | |
6502 | ||
6756ae4b | 6503 | __rtnl_unlock(); |
1da177e4 LT |
6504 | |
6505 | rebroadcast_time = jiffies; | |
6506 | } | |
6507 | ||
6508 | msleep(250); | |
6509 | ||
29b4433d ED |
6510 | refcnt = netdev_refcnt_read(dev); |
6511 | ||
1da177e4 | 6512 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
6513 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
6514 | dev->name, refcnt); | |
1da177e4 LT |
6515 | warning_time = jiffies; |
6516 | } | |
6517 | } | |
6518 | } | |
6519 | ||
6520 | /* The sequence is: | |
6521 | * | |
6522 | * rtnl_lock(); | |
6523 | * ... | |
6524 | * register_netdevice(x1); | |
6525 | * register_netdevice(x2); | |
6526 | * ... | |
6527 | * unregister_netdevice(y1); | |
6528 | * unregister_netdevice(y2); | |
6529 | * ... | |
6530 | * rtnl_unlock(); | |
6531 | * free_netdev(y1); | |
6532 | * free_netdev(y2); | |
6533 | * | |
58ec3b4d | 6534 | * We are invoked by rtnl_unlock(). |
1da177e4 | 6535 | * This allows us to deal with problems: |
b17a7c17 | 6536 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
6537 | * without deadlocking with linkwatch via keventd. |
6538 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
6539 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
6540 | * |
6541 | * We must not return until all unregister events added during | |
6542 | * the interval the lock was held have been completed. | |
1da177e4 | 6543 | */ |
1da177e4 LT |
6544 | void netdev_run_todo(void) |
6545 | { | |
626ab0e6 | 6546 | struct list_head list; |
1da177e4 | 6547 | |
1da177e4 | 6548 | /* Snapshot list, allow later requests */ |
626ab0e6 | 6549 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
6550 | |
6551 | __rtnl_unlock(); | |
626ab0e6 | 6552 | |
0115e8e3 ED |
6553 | |
6554 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
6555 | if (!list_empty(&list)) |
6556 | rcu_barrier(); | |
6557 | ||
1da177e4 LT |
6558 | while (!list_empty(&list)) { |
6559 | struct net_device *dev | |
e5e26d75 | 6560 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
6561 | list_del(&dev->todo_list); |
6562 | ||
748e2d93 | 6563 | rtnl_lock(); |
0115e8e3 | 6564 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 6565 | __rtnl_unlock(); |
0115e8e3 | 6566 | |
b17a7c17 | 6567 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 6568 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
6569 | dev->name, dev->reg_state); |
6570 | dump_stack(); | |
6571 | continue; | |
6572 | } | |
1da177e4 | 6573 | |
b17a7c17 | 6574 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 6575 | |
152102c7 | 6576 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 6577 | |
b17a7c17 | 6578 | netdev_wait_allrefs(dev); |
1da177e4 | 6579 | |
b17a7c17 | 6580 | /* paranoia */ |
29b4433d | 6581 | BUG_ON(netdev_refcnt_read(dev)); |
33d480ce ED |
6582 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
6583 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 6584 | WARN_ON(dev->dn_ptr); |
1da177e4 | 6585 | |
b17a7c17 SH |
6586 | if (dev->destructor) |
6587 | dev->destructor(dev); | |
9093bbb2 | 6588 | |
50624c93 EB |
6589 | /* Report a network device has been unregistered */ |
6590 | rtnl_lock(); | |
6591 | dev_net(dev)->dev_unreg_count--; | |
6592 | __rtnl_unlock(); | |
6593 | wake_up(&netdev_unregistering_wq); | |
6594 | ||
9093bbb2 SH |
6595 | /* Free network device */ |
6596 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 6597 | } |
1da177e4 LT |
6598 | } |
6599 | ||
3cfde79c BH |
6600 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
6601 | * fields in the same order, with only the type differing. | |
6602 | */ | |
77a1abf5 ED |
6603 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
6604 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
6605 | { |
6606 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
6607 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
6608 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
6609 | #else |
6610 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
6611 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
6612 | u64 *dst = (u64 *)stats64; | |
6613 | ||
6614 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
6615 | sizeof(*stats64) / sizeof(u64)); | |
6616 | for (i = 0; i < n; i++) | |
6617 | dst[i] = src[i]; | |
6618 | #endif | |
6619 | } | |
77a1abf5 | 6620 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 6621 | |
eeda3fd6 SH |
6622 | /** |
6623 | * dev_get_stats - get network device statistics | |
6624 | * @dev: device to get statistics from | |
28172739 | 6625 | * @storage: place to store stats |
eeda3fd6 | 6626 | * |
d7753516 BH |
6627 | * Get network statistics from device. Return @storage. |
6628 | * The device driver may provide its own method by setting | |
6629 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
6630 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 6631 | */ |
d7753516 BH |
6632 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
6633 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 6634 | { |
eeda3fd6 SH |
6635 | const struct net_device_ops *ops = dev->netdev_ops; |
6636 | ||
28172739 ED |
6637 | if (ops->ndo_get_stats64) { |
6638 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
6639 | ops->ndo_get_stats64(dev, storage); |
6640 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 6641 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
6642 | } else { |
6643 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 6644 | } |
caf586e5 | 6645 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
015f0688 | 6646 | storage->tx_dropped += atomic_long_read(&dev->tx_dropped); |
28172739 | 6647 | return storage; |
c45d286e | 6648 | } |
eeda3fd6 | 6649 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 6650 | |
24824a09 | 6651 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 6652 | { |
24824a09 | 6653 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 6654 | |
24824a09 ED |
6655 | #ifdef CONFIG_NET_CLS_ACT |
6656 | if (queue) | |
6657 | return queue; | |
6658 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
6659 | if (!queue) | |
6660 | return NULL; | |
6661 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
6662 | queue->qdisc = &noop_qdisc; |
6663 | queue->qdisc_sleeping = &noop_qdisc; | |
6664 | rcu_assign_pointer(dev->ingress_queue, queue); | |
6665 | #endif | |
6666 | return queue; | |
bb949fbd DM |
6667 | } |
6668 | ||
2c60db03 ED |
6669 | static const struct ethtool_ops default_ethtool_ops; |
6670 | ||
d07d7507 SG |
6671 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
6672 | const struct ethtool_ops *ops) | |
6673 | { | |
6674 | if (dev->ethtool_ops == &default_ethtool_ops) | |
6675 | dev->ethtool_ops = ops; | |
6676 | } | |
6677 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
6678 | ||
74d332c1 ED |
6679 | void netdev_freemem(struct net_device *dev) |
6680 | { | |
6681 | char *addr = (char *)dev - dev->padded; | |
6682 | ||
4cb28970 | 6683 | kvfree(addr); |
74d332c1 ED |
6684 | } |
6685 | ||
1da177e4 | 6686 | /** |
36909ea4 | 6687 | * alloc_netdev_mqs - allocate network device |
c835a677 TG |
6688 | * @sizeof_priv: size of private data to allocate space for |
6689 | * @name: device name format string | |
6690 | * @name_assign_type: origin of device name | |
6691 | * @setup: callback to initialize device | |
6692 | * @txqs: the number of TX subqueues to allocate | |
6693 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
6694 | * |
6695 | * Allocates a struct net_device with private data area for driver use | |
90e51adf | 6696 | * and performs basic initialization. Also allocates subqueue structs |
36909ea4 | 6697 | * for each queue on the device. |
1da177e4 | 6698 | */ |
36909ea4 | 6699 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 6700 | unsigned char name_assign_type, |
36909ea4 TH |
6701 | void (*setup)(struct net_device *), |
6702 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 6703 | { |
1da177e4 | 6704 | struct net_device *dev; |
7943986c | 6705 | size_t alloc_size; |
1ce8e7b5 | 6706 | struct net_device *p; |
1da177e4 | 6707 | |
b6fe17d6 SH |
6708 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
6709 | ||
36909ea4 | 6710 | if (txqs < 1) { |
7b6cd1ce | 6711 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
6712 | return NULL; |
6713 | } | |
6714 | ||
a953be53 | 6715 | #ifdef CONFIG_SYSFS |
36909ea4 | 6716 | if (rxqs < 1) { |
7b6cd1ce | 6717 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
6718 | return NULL; |
6719 | } | |
6720 | #endif | |
6721 | ||
fd2ea0a7 | 6722 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
6723 | if (sizeof_priv) { |
6724 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 6725 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
6726 | alloc_size += sizeof_priv; |
6727 | } | |
6728 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 6729 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 6730 | |
74d332c1 ED |
6731 | p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6732 | if (!p) | |
6733 | p = vzalloc(alloc_size); | |
62b5942a | 6734 | if (!p) |
1da177e4 | 6735 | return NULL; |
1da177e4 | 6736 | |
1ce8e7b5 | 6737 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 6738 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 6739 | |
29b4433d ED |
6740 | dev->pcpu_refcnt = alloc_percpu(int); |
6741 | if (!dev->pcpu_refcnt) | |
74d332c1 | 6742 | goto free_dev; |
ab9c73cc | 6743 | |
ab9c73cc | 6744 | if (dev_addr_init(dev)) |
29b4433d | 6745 | goto free_pcpu; |
ab9c73cc | 6746 | |
22bedad3 | 6747 | dev_mc_init(dev); |
a748ee24 | 6748 | dev_uc_init(dev); |
ccffad25 | 6749 | |
c346dca1 | 6750 | dev_net_set(dev, &init_net); |
1da177e4 | 6751 | |
8d3bdbd5 | 6752 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 6753 | dev->gso_max_segs = GSO_MAX_SEGS; |
fcbeb976 | 6754 | dev->gso_min_segs = 0; |
8d3bdbd5 | 6755 | |
8d3bdbd5 DM |
6756 | INIT_LIST_HEAD(&dev->napi_list); |
6757 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 6758 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 6759 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
6760 | INIT_LIST_HEAD(&dev->adj_list.upper); |
6761 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
6762 | INIT_LIST_HEAD(&dev->all_adj_list.upper); | |
6763 | INIT_LIST_HEAD(&dev->all_adj_list.lower); | |
02875878 | 6764 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
6765 | setup(dev); |
6766 | ||
36909ea4 TH |
6767 | dev->num_tx_queues = txqs; |
6768 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 6769 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 6770 | goto free_all; |
e8a0464c | 6771 | |
a953be53 | 6772 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
6773 | dev->num_rx_queues = rxqs; |
6774 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 6775 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 6776 | goto free_all; |
df334545 | 6777 | #endif |
0a9627f2 | 6778 | |
1da177e4 | 6779 | strcpy(dev->name, name); |
c835a677 | 6780 | dev->name_assign_type = name_assign_type; |
cbda10fa | 6781 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
6782 | if (!dev->ethtool_ops) |
6783 | dev->ethtool_ops = &default_ethtool_ops; | |
1da177e4 | 6784 | return dev; |
ab9c73cc | 6785 | |
8d3bdbd5 DM |
6786 | free_all: |
6787 | free_netdev(dev); | |
6788 | return NULL; | |
6789 | ||
29b4433d ED |
6790 | free_pcpu: |
6791 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
6792 | free_dev: |
6793 | netdev_freemem(dev); | |
ab9c73cc | 6794 | return NULL; |
1da177e4 | 6795 | } |
36909ea4 | 6796 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
6797 | |
6798 | /** | |
6799 | * free_netdev - free network device | |
6800 | * @dev: device | |
6801 | * | |
4ec93edb YH |
6802 | * This function does the last stage of destroying an allocated device |
6803 | * interface. The reference to the device object is released. | |
1da177e4 LT |
6804 | * If this is the last reference then it will be freed. |
6805 | */ | |
6806 | void free_netdev(struct net_device *dev) | |
6807 | { | |
d565b0a1 HX |
6808 | struct napi_struct *p, *n; |
6809 | ||
f3005d7f DL |
6810 | release_net(dev_net(dev)); |
6811 | ||
60877a32 | 6812 | netif_free_tx_queues(dev); |
a953be53 | 6813 | #ifdef CONFIG_SYSFS |
10595902 | 6814 | kvfree(dev->_rx); |
fe822240 | 6815 | #endif |
e8a0464c | 6816 | |
33d480ce | 6817 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 6818 | |
f001fde5 JP |
6819 | /* Flush device addresses */ |
6820 | dev_addr_flush(dev); | |
6821 | ||
d565b0a1 HX |
6822 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
6823 | netif_napi_del(p); | |
6824 | ||
29b4433d ED |
6825 | free_percpu(dev->pcpu_refcnt); |
6826 | dev->pcpu_refcnt = NULL; | |
6827 | ||
3041a069 | 6828 | /* Compatibility with error handling in drivers */ |
1da177e4 | 6829 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 6830 | netdev_freemem(dev); |
1da177e4 LT |
6831 | return; |
6832 | } | |
6833 | ||
6834 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
6835 | dev->reg_state = NETREG_RELEASED; | |
6836 | ||
43cb76d9 GKH |
6837 | /* will free via device release */ |
6838 | put_device(&dev->dev); | |
1da177e4 | 6839 | } |
d1b19dff | 6840 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 6841 | |
f0db275a SH |
6842 | /** |
6843 | * synchronize_net - Synchronize with packet receive processing | |
6844 | * | |
6845 | * Wait for packets currently being received to be done. | |
6846 | * Does not block later packets from starting. | |
6847 | */ | |
4ec93edb | 6848 | void synchronize_net(void) |
1da177e4 LT |
6849 | { |
6850 | might_sleep(); | |
be3fc413 ED |
6851 | if (rtnl_is_locked()) |
6852 | synchronize_rcu_expedited(); | |
6853 | else | |
6854 | synchronize_rcu(); | |
1da177e4 | 6855 | } |
d1b19dff | 6856 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
6857 | |
6858 | /** | |
44a0873d | 6859 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 6860 | * @dev: device |
44a0873d | 6861 | * @head: list |
6ebfbc06 | 6862 | * |
1da177e4 | 6863 | * This function shuts down a device interface and removes it |
d59b54b1 | 6864 | * from the kernel tables. |
44a0873d | 6865 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
6866 | * |
6867 | * Callers must hold the rtnl semaphore. You may want | |
6868 | * unregister_netdev() instead of this. | |
6869 | */ | |
6870 | ||
44a0873d | 6871 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 6872 | { |
a6620712 HX |
6873 | ASSERT_RTNL(); |
6874 | ||
44a0873d | 6875 | if (head) { |
9fdce099 | 6876 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
6877 | } else { |
6878 | rollback_registered(dev); | |
6879 | /* Finish processing unregister after unlock */ | |
6880 | net_set_todo(dev); | |
6881 | } | |
1da177e4 | 6882 | } |
44a0873d | 6883 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 6884 | |
9b5e383c ED |
6885 | /** |
6886 | * unregister_netdevice_many - unregister many devices | |
6887 | * @head: list of devices | |
87757a91 ED |
6888 | * |
6889 | * Note: As most callers use a stack allocated list_head, | |
6890 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
6891 | */ |
6892 | void unregister_netdevice_many(struct list_head *head) | |
6893 | { | |
6894 | struct net_device *dev; | |
6895 | ||
6896 | if (!list_empty(head)) { | |
6897 | rollback_registered_many(head); | |
6898 | list_for_each_entry(dev, head, unreg_list) | |
6899 | net_set_todo(dev); | |
87757a91 | 6900 | list_del(head); |
9b5e383c ED |
6901 | } |
6902 | } | |
63c8099d | 6903 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 6904 | |
1da177e4 LT |
6905 | /** |
6906 | * unregister_netdev - remove device from the kernel | |
6907 | * @dev: device | |
6908 | * | |
6909 | * This function shuts down a device interface and removes it | |
d59b54b1 | 6910 | * from the kernel tables. |
1da177e4 LT |
6911 | * |
6912 | * This is just a wrapper for unregister_netdevice that takes | |
6913 | * the rtnl semaphore. In general you want to use this and not | |
6914 | * unregister_netdevice. | |
6915 | */ | |
6916 | void unregister_netdev(struct net_device *dev) | |
6917 | { | |
6918 | rtnl_lock(); | |
6919 | unregister_netdevice(dev); | |
6920 | rtnl_unlock(); | |
6921 | } | |
1da177e4 LT |
6922 | EXPORT_SYMBOL(unregister_netdev); |
6923 | ||
ce286d32 EB |
6924 | /** |
6925 | * dev_change_net_namespace - move device to different nethost namespace | |
6926 | * @dev: device | |
6927 | * @net: network namespace | |
6928 | * @pat: If not NULL name pattern to try if the current device name | |
6929 | * is already taken in the destination network namespace. | |
6930 | * | |
6931 | * This function shuts down a device interface and moves it | |
6932 | * to a new network namespace. On success 0 is returned, on | |
6933 | * a failure a netagive errno code is returned. | |
6934 | * | |
6935 | * Callers must hold the rtnl semaphore. | |
6936 | */ | |
6937 | ||
6938 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
6939 | { | |
ce286d32 EB |
6940 | int err; |
6941 | ||
6942 | ASSERT_RTNL(); | |
6943 | ||
6944 | /* Don't allow namespace local devices to be moved. */ | |
6945 | err = -EINVAL; | |
6946 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6947 | goto out; | |
6948 | ||
6949 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
6950 | if (dev->reg_state != NETREG_REGISTERED) |
6951 | goto out; | |
6952 | ||
6953 | /* Get out if there is nothing todo */ | |
6954 | err = 0; | |
878628fb | 6955 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
6956 | goto out; |
6957 | ||
6958 | /* Pick the destination device name, and ensure | |
6959 | * we can use it in the destination network namespace. | |
6960 | */ | |
6961 | err = -EEXIST; | |
d9031024 | 6962 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
6963 | /* We get here if we can't use the current device name */ |
6964 | if (!pat) | |
6965 | goto out; | |
828de4f6 | 6966 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
6967 | goto out; |
6968 | } | |
6969 | ||
6970 | /* | |
6971 | * And now a mini version of register_netdevice unregister_netdevice. | |
6972 | */ | |
6973 | ||
6974 | /* If device is running close it first. */ | |
9b772652 | 6975 | dev_close(dev); |
ce286d32 EB |
6976 | |
6977 | /* And unlink it from device chain */ | |
6978 | err = -ENODEV; | |
6979 | unlist_netdevice(dev); | |
6980 | ||
6981 | synchronize_net(); | |
6982 | ||
6983 | /* Shutdown queueing discipline. */ | |
6984 | dev_shutdown(dev); | |
6985 | ||
6986 | /* Notify protocols, that we are about to destroy | |
6987 | this device. They should clean all the things. | |
3b27e105 DL |
6988 | |
6989 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
6990 | This is wanted because this way 8021q and macvlan know | |
6991 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
6992 | */ |
6993 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
6994 | rcu_barrier(); |
6995 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 6996 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
6997 | |
6998 | /* | |
6999 | * Flush the unicast and multicast chains | |
7000 | */ | |
a748ee24 | 7001 | dev_uc_flush(dev); |
22bedad3 | 7002 | dev_mc_flush(dev); |
ce286d32 | 7003 | |
4e66ae2e SH |
7004 | /* Send a netdev-removed uevent to the old namespace */ |
7005 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 7006 | netdev_adjacent_del_links(dev); |
4e66ae2e | 7007 | |
ce286d32 | 7008 | /* Actually switch the network namespace */ |
c346dca1 | 7009 | dev_net_set(dev, net); |
ce286d32 | 7010 | |
ce286d32 EB |
7011 | /* If there is an ifindex conflict assign a new one */ |
7012 | if (__dev_get_by_index(net, dev->ifindex)) { | |
7013 | int iflink = (dev->iflink == dev->ifindex); | |
7014 | dev->ifindex = dev_new_index(net); | |
7015 | if (iflink) | |
7016 | dev->iflink = dev->ifindex; | |
7017 | } | |
7018 | ||
4e66ae2e SH |
7019 | /* Send a netdev-add uevent to the new namespace */ |
7020 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 7021 | netdev_adjacent_add_links(dev); |
4e66ae2e | 7022 | |
8b41d188 | 7023 | /* Fixup kobjects */ |
a1b3f594 | 7024 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 7025 | WARN_ON(err); |
ce286d32 EB |
7026 | |
7027 | /* Add the device back in the hashes */ | |
7028 | list_netdevice(dev); | |
7029 | ||
7030 | /* Notify protocols, that a new device appeared. */ | |
7031 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
7032 | ||
d90a909e EB |
7033 | /* |
7034 | * Prevent userspace races by waiting until the network | |
7035 | * device is fully setup before sending notifications. | |
7036 | */ | |
7f294054 | 7037 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 7038 | |
ce286d32 EB |
7039 | synchronize_net(); |
7040 | err = 0; | |
7041 | out: | |
7042 | return err; | |
7043 | } | |
463d0183 | 7044 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 7045 | |
1da177e4 LT |
7046 | static int dev_cpu_callback(struct notifier_block *nfb, |
7047 | unsigned long action, | |
7048 | void *ocpu) | |
7049 | { | |
7050 | struct sk_buff **list_skb; | |
1da177e4 LT |
7051 | struct sk_buff *skb; |
7052 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
7053 | struct softnet_data *sd, *oldsd; | |
7054 | ||
8bb78442 | 7055 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
7056 | return NOTIFY_OK; |
7057 | ||
7058 | local_irq_disable(); | |
7059 | cpu = smp_processor_id(); | |
7060 | sd = &per_cpu(softnet_data, cpu); | |
7061 | oldsd = &per_cpu(softnet_data, oldcpu); | |
7062 | ||
7063 | /* Find end of our completion_queue. */ | |
7064 | list_skb = &sd->completion_queue; | |
7065 | while (*list_skb) | |
7066 | list_skb = &(*list_skb)->next; | |
7067 | /* Append completion queue from offline CPU. */ | |
7068 | *list_skb = oldsd->completion_queue; | |
7069 | oldsd->completion_queue = NULL; | |
7070 | ||
1da177e4 | 7071 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
7072 | if (oldsd->output_queue) { |
7073 | *sd->output_queue_tailp = oldsd->output_queue; | |
7074 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
7075 | oldsd->output_queue = NULL; | |
7076 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
7077 | } | |
ac64da0b ED |
7078 | /* Append NAPI poll list from offline CPU, with one exception : |
7079 | * process_backlog() must be called by cpu owning percpu backlog. | |
7080 | * We properly handle process_queue & input_pkt_queue later. | |
7081 | */ | |
7082 | while (!list_empty(&oldsd->poll_list)) { | |
7083 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
7084 | struct napi_struct, | |
7085 | poll_list); | |
7086 | ||
7087 | list_del_init(&napi->poll_list); | |
7088 | if (napi->poll == process_backlog) | |
7089 | napi->state = 0; | |
7090 | else | |
7091 | ____napi_schedule(sd, napi); | |
264524d5 | 7092 | } |
1da177e4 LT |
7093 | |
7094 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
7095 | local_irq_enable(); | |
7096 | ||
7097 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 7098 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
ae78dbfa | 7099 | netif_rx_internal(skb); |
76cc8b13 | 7100 | input_queue_head_incr(oldsd); |
fec5e652 | 7101 | } |
ac64da0b | 7102 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
ae78dbfa | 7103 | netif_rx_internal(skb); |
76cc8b13 TH |
7104 | input_queue_head_incr(oldsd); |
7105 | } | |
1da177e4 LT |
7106 | |
7107 | return NOTIFY_OK; | |
7108 | } | |
1da177e4 LT |
7109 | |
7110 | ||
7f353bf2 | 7111 | /** |
b63365a2 HX |
7112 | * netdev_increment_features - increment feature set by one |
7113 | * @all: current feature set | |
7114 | * @one: new feature set | |
7115 | * @mask: mask feature set | |
7f353bf2 HX |
7116 | * |
7117 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
7118 | * @one to the master device with current feature set @all. Will not |
7119 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 7120 | */ |
c8f44aff MM |
7121 | netdev_features_t netdev_increment_features(netdev_features_t all, |
7122 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 7123 | { |
1742f183 MM |
7124 | if (mask & NETIF_F_GEN_CSUM) |
7125 | mask |= NETIF_F_ALL_CSUM; | |
7126 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 7127 | |
1742f183 MM |
7128 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
7129 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 7130 | |
1742f183 MM |
7131 | /* If one device supports hw checksumming, set for all. */ |
7132 | if (all & NETIF_F_GEN_CSUM) | |
7133 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
7134 | |
7135 | return all; | |
7136 | } | |
b63365a2 | 7137 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 7138 | |
430f03cd | 7139 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
7140 | { |
7141 | int i; | |
7142 | struct hlist_head *hash; | |
7143 | ||
7144 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
7145 | if (hash != NULL) | |
7146 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
7147 | INIT_HLIST_HEAD(&hash[i]); | |
7148 | ||
7149 | return hash; | |
7150 | } | |
7151 | ||
881d966b | 7152 | /* Initialize per network namespace state */ |
4665079c | 7153 | static int __net_init netdev_init(struct net *net) |
881d966b | 7154 | { |
734b6541 RM |
7155 | if (net != &init_net) |
7156 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 7157 | |
30d97d35 PE |
7158 | net->dev_name_head = netdev_create_hash(); |
7159 | if (net->dev_name_head == NULL) | |
7160 | goto err_name; | |
881d966b | 7161 | |
30d97d35 PE |
7162 | net->dev_index_head = netdev_create_hash(); |
7163 | if (net->dev_index_head == NULL) | |
7164 | goto err_idx; | |
881d966b EB |
7165 | |
7166 | return 0; | |
30d97d35 PE |
7167 | |
7168 | err_idx: | |
7169 | kfree(net->dev_name_head); | |
7170 | err_name: | |
7171 | return -ENOMEM; | |
881d966b EB |
7172 | } |
7173 | ||
f0db275a SH |
7174 | /** |
7175 | * netdev_drivername - network driver for the device | |
7176 | * @dev: network device | |
f0db275a SH |
7177 | * |
7178 | * Determine network driver for device. | |
7179 | */ | |
3019de12 | 7180 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 7181 | { |
cf04a4c7 SH |
7182 | const struct device_driver *driver; |
7183 | const struct device *parent; | |
3019de12 | 7184 | const char *empty = ""; |
6579e57b AV |
7185 | |
7186 | parent = dev->dev.parent; | |
6579e57b | 7187 | if (!parent) |
3019de12 | 7188 | return empty; |
6579e57b AV |
7189 | |
7190 | driver = parent->driver; | |
7191 | if (driver && driver->name) | |
3019de12 DM |
7192 | return driver->name; |
7193 | return empty; | |
6579e57b AV |
7194 | } |
7195 | ||
6ea754eb JP |
7196 | static void __netdev_printk(const char *level, const struct net_device *dev, |
7197 | struct va_format *vaf) | |
256df2f3 | 7198 | { |
b004ff49 | 7199 | if (dev && dev->dev.parent) { |
6ea754eb JP |
7200 | dev_printk_emit(level[1] - '0', |
7201 | dev->dev.parent, | |
7202 | "%s %s %s%s: %pV", | |
7203 | dev_driver_string(dev->dev.parent), | |
7204 | dev_name(dev->dev.parent), | |
7205 | netdev_name(dev), netdev_reg_state(dev), | |
7206 | vaf); | |
b004ff49 | 7207 | } else if (dev) { |
6ea754eb JP |
7208 | printk("%s%s%s: %pV", |
7209 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 7210 | } else { |
6ea754eb | 7211 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 7212 | } |
256df2f3 JP |
7213 | } |
7214 | ||
6ea754eb JP |
7215 | void netdev_printk(const char *level, const struct net_device *dev, |
7216 | const char *format, ...) | |
256df2f3 JP |
7217 | { |
7218 | struct va_format vaf; | |
7219 | va_list args; | |
256df2f3 JP |
7220 | |
7221 | va_start(args, format); | |
7222 | ||
7223 | vaf.fmt = format; | |
7224 | vaf.va = &args; | |
7225 | ||
6ea754eb | 7226 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 7227 | |
256df2f3 | 7228 | va_end(args); |
256df2f3 JP |
7229 | } |
7230 | EXPORT_SYMBOL(netdev_printk); | |
7231 | ||
7232 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 7233 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 7234 | { \ |
256df2f3 JP |
7235 | struct va_format vaf; \ |
7236 | va_list args; \ | |
7237 | \ | |
7238 | va_start(args, fmt); \ | |
7239 | \ | |
7240 | vaf.fmt = fmt; \ | |
7241 | vaf.va = &args; \ | |
7242 | \ | |
6ea754eb | 7243 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 7244 | \ |
256df2f3 | 7245 | va_end(args); \ |
256df2f3 JP |
7246 | } \ |
7247 | EXPORT_SYMBOL(func); | |
7248 | ||
7249 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
7250 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
7251 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
7252 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
7253 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
7254 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
7255 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
7256 | ||
4665079c | 7257 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
7258 | { |
7259 | kfree(net->dev_name_head); | |
7260 | kfree(net->dev_index_head); | |
7261 | } | |
7262 | ||
022cbae6 | 7263 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
7264 | .init = netdev_init, |
7265 | .exit = netdev_exit, | |
7266 | }; | |
7267 | ||
4665079c | 7268 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 7269 | { |
e008b5fc | 7270 | struct net_device *dev, *aux; |
ce286d32 | 7271 | /* |
e008b5fc | 7272 | * Push all migratable network devices back to the |
ce286d32 EB |
7273 | * initial network namespace |
7274 | */ | |
7275 | rtnl_lock(); | |
e008b5fc | 7276 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 7277 | int err; |
aca51397 | 7278 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
7279 | |
7280 | /* Ignore unmoveable devices (i.e. loopback) */ | |
7281 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7282 | continue; | |
7283 | ||
e008b5fc EB |
7284 | /* Leave virtual devices for the generic cleanup */ |
7285 | if (dev->rtnl_link_ops) | |
7286 | continue; | |
d0c082ce | 7287 | |
25985edc | 7288 | /* Push remaining network devices to init_net */ |
aca51397 PE |
7289 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
7290 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 7291 | if (err) { |
7b6cd1ce JP |
7292 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
7293 | __func__, dev->name, err); | |
aca51397 | 7294 | BUG(); |
ce286d32 EB |
7295 | } |
7296 | } | |
7297 | rtnl_unlock(); | |
7298 | } | |
7299 | ||
50624c93 EB |
7300 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
7301 | { | |
7302 | /* Return with the rtnl_lock held when there are no network | |
7303 | * devices unregistering in any network namespace in net_list. | |
7304 | */ | |
7305 | struct net *net; | |
7306 | bool unregistering; | |
ff960a73 | 7307 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 7308 | |
ff960a73 | 7309 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 7310 | for (;;) { |
50624c93 EB |
7311 | unregistering = false; |
7312 | rtnl_lock(); | |
7313 | list_for_each_entry(net, net_list, exit_list) { | |
7314 | if (net->dev_unreg_count > 0) { | |
7315 | unregistering = true; | |
7316 | break; | |
7317 | } | |
7318 | } | |
7319 | if (!unregistering) | |
7320 | break; | |
7321 | __rtnl_unlock(); | |
ff960a73 PZ |
7322 | |
7323 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 7324 | } |
ff960a73 | 7325 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
7326 | } |
7327 | ||
04dc7f6b EB |
7328 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
7329 | { | |
7330 | /* At exit all network devices most be removed from a network | |
b595076a | 7331 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
7332 | * Do this across as many network namespaces as possible to |
7333 | * improve batching efficiency. | |
7334 | */ | |
7335 | struct net_device *dev; | |
7336 | struct net *net; | |
7337 | LIST_HEAD(dev_kill_list); | |
7338 | ||
50624c93 EB |
7339 | /* To prevent network device cleanup code from dereferencing |
7340 | * loopback devices or network devices that have been freed | |
7341 | * wait here for all pending unregistrations to complete, | |
7342 | * before unregistring the loopback device and allowing the | |
7343 | * network namespace be freed. | |
7344 | * | |
7345 | * The netdev todo list containing all network devices | |
7346 | * unregistrations that happen in default_device_exit_batch | |
7347 | * will run in the rtnl_unlock() at the end of | |
7348 | * default_device_exit_batch. | |
7349 | */ | |
7350 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
7351 | list_for_each_entry(net, net_list, exit_list) { |
7352 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 7353 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
7354 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
7355 | else | |
7356 | unregister_netdevice_queue(dev, &dev_kill_list); | |
7357 | } | |
7358 | } | |
7359 | unregister_netdevice_many(&dev_kill_list); | |
7360 | rtnl_unlock(); | |
7361 | } | |
7362 | ||
022cbae6 | 7363 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 7364 | .exit = default_device_exit, |
04dc7f6b | 7365 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
7366 | }; |
7367 | ||
1da177e4 LT |
7368 | /* |
7369 | * Initialize the DEV module. At boot time this walks the device list and | |
7370 | * unhooks any devices that fail to initialise (normally hardware not | |
7371 | * present) and leaves us with a valid list of present and active devices. | |
7372 | * | |
7373 | */ | |
7374 | ||
7375 | /* | |
7376 | * This is called single threaded during boot, so no need | |
7377 | * to take the rtnl semaphore. | |
7378 | */ | |
7379 | static int __init net_dev_init(void) | |
7380 | { | |
7381 | int i, rc = -ENOMEM; | |
7382 | ||
7383 | BUG_ON(!dev_boot_phase); | |
7384 | ||
1da177e4 LT |
7385 | if (dev_proc_init()) |
7386 | goto out; | |
7387 | ||
8b41d188 | 7388 | if (netdev_kobject_init()) |
1da177e4 LT |
7389 | goto out; |
7390 | ||
7391 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 7392 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
7393 | INIT_LIST_HEAD(&ptype_base[i]); |
7394 | ||
62532da9 VY |
7395 | INIT_LIST_HEAD(&offload_base); |
7396 | ||
881d966b EB |
7397 | if (register_pernet_subsys(&netdev_net_ops)) |
7398 | goto out; | |
1da177e4 LT |
7399 | |
7400 | /* | |
7401 | * Initialise the packet receive queues. | |
7402 | */ | |
7403 | ||
6f912042 | 7404 | for_each_possible_cpu(i) { |
e36fa2f7 | 7405 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 7406 | |
e36fa2f7 | 7407 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 7408 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 7409 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 7410 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 7411 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
7412 | sd->csd.func = rps_trigger_softirq; |
7413 | sd->csd.info = sd; | |
e36fa2f7 | 7414 | sd->cpu = i; |
1e94d72f | 7415 | #endif |
0a9627f2 | 7416 | |
e36fa2f7 ED |
7417 | sd->backlog.poll = process_backlog; |
7418 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
7419 | } |
7420 | ||
1da177e4 LT |
7421 | dev_boot_phase = 0; |
7422 | ||
505d4f73 EB |
7423 | /* The loopback device is special if any other network devices |
7424 | * is present in a network namespace the loopback device must | |
7425 | * be present. Since we now dynamically allocate and free the | |
7426 | * loopback device ensure this invariant is maintained by | |
7427 | * keeping the loopback device as the first device on the | |
7428 | * list of network devices. Ensuring the loopback devices | |
7429 | * is the first device that appears and the last network device | |
7430 | * that disappears. | |
7431 | */ | |
7432 | if (register_pernet_device(&loopback_net_ops)) | |
7433 | goto out; | |
7434 | ||
7435 | if (register_pernet_device(&default_device_ops)) | |
7436 | goto out; | |
7437 | ||
962cf36c CM |
7438 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
7439 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
7440 | |
7441 | hotcpu_notifier(dev_cpu_callback, 0); | |
7442 | dst_init(); | |
1da177e4 LT |
7443 | rc = 0; |
7444 | out: | |
7445 | return rc; | |
7446 | } | |
7447 | ||
7448 | subsys_initcall(net_dev_init); |