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 | 98 | #include <net/sock.h> |
02d62e86 | 99 | #include <net/busy_poll.h> |
1da177e4 | 100 | #include <linux/rtnetlink.h> |
1da177e4 | 101 | #include <linux/stat.h> |
1da177e4 | 102 | #include <net/dst.h> |
fc4099f1 | 103 | #include <net/dst_metadata.h> |
1da177e4 LT |
104 | #include <net/pkt_sched.h> |
105 | #include <net/checksum.h> | |
44540960 | 106 | #include <net/xfrm.h> |
1da177e4 LT |
107 | #include <linux/highmem.h> |
108 | #include <linux/init.h> | |
1da177e4 | 109 | #include <linux/module.h> |
1da177e4 LT |
110 | #include <linux/netpoll.h> |
111 | #include <linux/rcupdate.h> | |
112 | #include <linux/delay.h> | |
1da177e4 | 113 | #include <net/iw_handler.h> |
1da177e4 | 114 | #include <asm/current.h> |
5bdb9886 | 115 | #include <linux/audit.h> |
db217334 | 116 | #include <linux/dmaengine.h> |
f6a78bfc | 117 | #include <linux/err.h> |
c7fa9d18 | 118 | #include <linux/ctype.h> |
723e98b7 | 119 | #include <linux/if_arp.h> |
6de329e2 | 120 | #include <linux/if_vlan.h> |
8f0f2223 | 121 | #include <linux/ip.h> |
ad55dcaf | 122 | #include <net/ip.h> |
25cd9ba0 | 123 | #include <net/mpls.h> |
8f0f2223 DM |
124 | #include <linux/ipv6.h> |
125 | #include <linux/in.h> | |
b6b2fed1 DM |
126 | #include <linux/jhash.h> |
127 | #include <linux/random.h> | |
9cbc1cb8 | 128 | #include <trace/events/napi.h> |
cf66ba58 | 129 | #include <trace/events/net.h> |
07dc22e7 | 130 | #include <trace/events/skb.h> |
5acbbd42 | 131 | #include <linux/pci.h> |
caeda9b9 | 132 | #include <linux/inetdevice.h> |
c445477d | 133 | #include <linux/cpu_rmap.h> |
c5905afb | 134 | #include <linux/static_key.h> |
af12fa6e | 135 | #include <linux/hashtable.h> |
60877a32 | 136 | #include <linux/vmalloc.h> |
529d0489 | 137 | #include <linux/if_macvlan.h> |
e7fd2885 | 138 | #include <linux/errqueue.h> |
3b47d303 | 139 | #include <linux/hrtimer.h> |
e687ad60 | 140 | #include <linux/netfilter_ingress.h> |
1da177e4 | 141 | |
342709ef PE |
142 | #include "net-sysfs.h" |
143 | ||
d565b0a1 HX |
144 | /* Instead of increasing this, you should create a hash table. */ |
145 | #define MAX_GRO_SKBS 8 | |
146 | ||
5d38a079 HX |
147 | /* This should be increased if a protocol with a bigger head is added. */ |
148 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
149 | ||
1da177e4 | 150 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 151 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
152 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
153 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 154 | static struct list_head offload_base __read_mostly; |
1da177e4 | 155 | |
ae78dbfa | 156 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
157 | static int call_netdevice_notifiers_info(unsigned long val, |
158 | struct net_device *dev, | |
159 | struct netdev_notifier_info *info); | |
ae78dbfa | 160 | |
1da177e4 | 161 | /* |
7562f876 | 162 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
163 | * semaphore. |
164 | * | |
c6d14c84 | 165 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
166 | * |
167 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 168 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
169 | * actual updates. This allows pure readers to access the list even |
170 | * while a writer is preparing to update it. | |
171 | * | |
172 | * To put it another way, dev_base_lock is held for writing only to | |
173 | * protect against pure readers; the rtnl semaphore provides the | |
174 | * protection against other writers. | |
175 | * | |
176 | * See, for example usages, register_netdevice() and | |
177 | * unregister_netdevice(), which must be called with the rtnl | |
178 | * semaphore held. | |
179 | */ | |
1da177e4 | 180 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
181 | EXPORT_SYMBOL(dev_base_lock); |
182 | ||
af12fa6e ET |
183 | /* protects napi_hash addition/deletion and napi_gen_id */ |
184 | static DEFINE_SPINLOCK(napi_hash_lock); | |
185 | ||
52bd2d62 | 186 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 187 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 188 | |
18afa4b0 | 189 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 190 | |
4e985ada TG |
191 | static inline void dev_base_seq_inc(struct net *net) |
192 | { | |
193 | while (++net->dev_base_seq == 0); | |
194 | } | |
195 | ||
881d966b | 196 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 197 | { |
95c96174 ED |
198 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
199 | ||
08e9897d | 200 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
201 | } |
202 | ||
881d966b | 203 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 204 | { |
7c28bd0b | 205 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
206 | } |
207 | ||
e36fa2f7 | 208 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
209 | { |
210 | #ifdef CONFIG_RPS | |
e36fa2f7 | 211 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
212 | #endif |
213 | } | |
214 | ||
e36fa2f7 | 215 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
216 | { |
217 | #ifdef CONFIG_RPS | |
e36fa2f7 | 218 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
219 | #endif |
220 | } | |
221 | ||
ce286d32 | 222 | /* Device list insertion */ |
53759be9 | 223 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 224 | { |
c346dca1 | 225 | struct net *net = dev_net(dev); |
ce286d32 EB |
226 | |
227 | ASSERT_RTNL(); | |
228 | ||
229 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 230 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 231 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
232 | hlist_add_head_rcu(&dev->index_hlist, |
233 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 234 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
235 | |
236 | dev_base_seq_inc(net); | |
ce286d32 EB |
237 | } |
238 | ||
fb699dfd ED |
239 | /* Device list removal |
240 | * caller must respect a RCU grace period before freeing/reusing dev | |
241 | */ | |
ce286d32 EB |
242 | static void unlist_netdevice(struct net_device *dev) |
243 | { | |
244 | ASSERT_RTNL(); | |
245 | ||
246 | /* Unlink dev from the device chain */ | |
247 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 248 | list_del_rcu(&dev->dev_list); |
72c9528b | 249 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 250 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 251 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
252 | |
253 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
254 | } |
255 | ||
1da177e4 LT |
256 | /* |
257 | * Our notifier list | |
258 | */ | |
259 | ||
f07d5b94 | 260 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
261 | |
262 | /* | |
263 | * Device drivers call our routines to queue packets here. We empty the | |
264 | * queue in the local softnet handler. | |
265 | */ | |
bea3348e | 266 | |
9958da05 | 267 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 268 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 269 | |
cf508b12 | 270 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 271 | /* |
c773e847 | 272 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
273 | * according to dev->type |
274 | */ | |
275 | static const unsigned short netdev_lock_type[] = | |
276 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
277 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
278 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
279 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
280 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
281 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
282 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
283 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
284 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
285 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
286 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
287 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
288 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
289 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
290 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 291 | |
36cbd3dc | 292 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
293 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
294 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
295 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
296 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
297 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
298 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
299 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
300 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
301 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
302 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
303 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
304 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
305 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
306 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
307 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
308 | |
309 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 310 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
311 | |
312 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
313 | { | |
314 | int i; | |
315 | ||
316 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
317 | if (netdev_lock_type[i] == dev_type) | |
318 | return i; | |
319 | /* the last key is used by default */ | |
320 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
321 | } | |
322 | ||
cf508b12 DM |
323 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
324 | unsigned short dev_type) | |
723e98b7 JP |
325 | { |
326 | int i; | |
327 | ||
328 | i = netdev_lock_pos(dev_type); | |
329 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
330 | netdev_lock_name[i]); | |
331 | } | |
cf508b12 DM |
332 | |
333 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
334 | { | |
335 | int i; | |
336 | ||
337 | i = netdev_lock_pos(dev->type); | |
338 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
339 | &netdev_addr_lock_key[i], | |
340 | netdev_lock_name[i]); | |
341 | } | |
723e98b7 | 342 | #else |
cf508b12 DM |
343 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
344 | unsigned short dev_type) | |
345 | { | |
346 | } | |
347 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
348 | { |
349 | } | |
350 | #endif | |
1da177e4 LT |
351 | |
352 | /******************************************************************************* | |
353 | ||
354 | Protocol management and registration routines | |
355 | ||
356 | *******************************************************************************/ | |
357 | ||
1da177e4 LT |
358 | /* |
359 | * Add a protocol ID to the list. Now that the input handler is | |
360 | * smarter we can dispense with all the messy stuff that used to be | |
361 | * here. | |
362 | * | |
363 | * BEWARE!!! Protocol handlers, mangling input packets, | |
364 | * MUST BE last in hash buckets and checking protocol handlers | |
365 | * MUST start from promiscuous ptype_all chain in net_bh. | |
366 | * It is true now, do not change it. | |
367 | * Explanation follows: if protocol handler, mangling packet, will | |
368 | * be the first on list, it is not able to sense, that packet | |
369 | * is cloned and should be copied-on-write, so that it will | |
370 | * change it and subsequent readers will get broken packet. | |
371 | * --ANK (980803) | |
372 | */ | |
373 | ||
c07b68e8 ED |
374 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
375 | { | |
376 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 377 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 378 | else |
7866a621 SN |
379 | return pt->dev ? &pt->dev->ptype_specific : |
380 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
381 | } |
382 | ||
1da177e4 LT |
383 | /** |
384 | * dev_add_pack - add packet handler | |
385 | * @pt: packet type declaration | |
386 | * | |
387 | * Add a protocol handler to the networking stack. The passed &packet_type | |
388 | * is linked into kernel lists and may not be freed until it has been | |
389 | * removed from the kernel lists. | |
390 | * | |
4ec93edb | 391 | * This call does not sleep therefore it can not |
1da177e4 LT |
392 | * guarantee all CPU's that are in middle of receiving packets |
393 | * will see the new packet type (until the next received packet). | |
394 | */ | |
395 | ||
396 | void dev_add_pack(struct packet_type *pt) | |
397 | { | |
c07b68e8 | 398 | struct list_head *head = ptype_head(pt); |
1da177e4 | 399 | |
c07b68e8 ED |
400 | spin_lock(&ptype_lock); |
401 | list_add_rcu(&pt->list, head); | |
402 | spin_unlock(&ptype_lock); | |
1da177e4 | 403 | } |
d1b19dff | 404 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 405 | |
1da177e4 LT |
406 | /** |
407 | * __dev_remove_pack - remove packet handler | |
408 | * @pt: packet type declaration | |
409 | * | |
410 | * Remove a protocol handler that was previously added to the kernel | |
411 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
412 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 413 | * returns. |
1da177e4 LT |
414 | * |
415 | * The packet type might still be in use by receivers | |
416 | * and must not be freed until after all the CPU's have gone | |
417 | * through a quiescent state. | |
418 | */ | |
419 | void __dev_remove_pack(struct packet_type *pt) | |
420 | { | |
c07b68e8 | 421 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
422 | struct packet_type *pt1; |
423 | ||
c07b68e8 | 424 | spin_lock(&ptype_lock); |
1da177e4 LT |
425 | |
426 | list_for_each_entry(pt1, head, list) { | |
427 | if (pt == pt1) { | |
428 | list_del_rcu(&pt->list); | |
429 | goto out; | |
430 | } | |
431 | } | |
432 | ||
7b6cd1ce | 433 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 434 | out: |
c07b68e8 | 435 | spin_unlock(&ptype_lock); |
1da177e4 | 436 | } |
d1b19dff ED |
437 | EXPORT_SYMBOL(__dev_remove_pack); |
438 | ||
1da177e4 LT |
439 | /** |
440 | * dev_remove_pack - remove packet handler | |
441 | * @pt: packet type declaration | |
442 | * | |
443 | * Remove a protocol handler that was previously added to the kernel | |
444 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
445 | * from the kernel lists and can be freed or reused once this function | |
446 | * returns. | |
447 | * | |
448 | * This call sleeps to guarantee that no CPU is looking at the packet | |
449 | * type after return. | |
450 | */ | |
451 | void dev_remove_pack(struct packet_type *pt) | |
452 | { | |
453 | __dev_remove_pack(pt); | |
4ec93edb | 454 | |
1da177e4 LT |
455 | synchronize_net(); |
456 | } | |
d1b19dff | 457 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 458 | |
62532da9 VY |
459 | |
460 | /** | |
461 | * dev_add_offload - register offload handlers | |
462 | * @po: protocol offload declaration | |
463 | * | |
464 | * Add protocol offload handlers to the networking stack. The passed | |
465 | * &proto_offload is linked into kernel lists and may not be freed until | |
466 | * it has been removed from the kernel lists. | |
467 | * | |
468 | * This call does not sleep therefore it can not | |
469 | * guarantee all CPU's that are in middle of receiving packets | |
470 | * will see the new offload handlers (until the next received packet). | |
471 | */ | |
472 | void dev_add_offload(struct packet_offload *po) | |
473 | { | |
bdef7de4 | 474 | struct packet_offload *elem; |
62532da9 VY |
475 | |
476 | spin_lock(&offload_lock); | |
bdef7de4 DM |
477 | list_for_each_entry(elem, &offload_base, list) { |
478 | if (po->priority < elem->priority) | |
479 | break; | |
480 | } | |
481 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
482 | spin_unlock(&offload_lock); |
483 | } | |
484 | EXPORT_SYMBOL(dev_add_offload); | |
485 | ||
486 | /** | |
487 | * __dev_remove_offload - remove offload handler | |
488 | * @po: packet offload declaration | |
489 | * | |
490 | * Remove a protocol offload handler that was previously added to the | |
491 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
492 | * is removed from the kernel lists and can be freed or reused once this | |
493 | * function returns. | |
494 | * | |
495 | * The packet type might still be in use by receivers | |
496 | * and must not be freed until after all the CPU's have gone | |
497 | * through a quiescent state. | |
498 | */ | |
1d143d9f | 499 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
500 | { |
501 | struct list_head *head = &offload_base; | |
502 | struct packet_offload *po1; | |
503 | ||
c53aa505 | 504 | spin_lock(&offload_lock); |
62532da9 VY |
505 | |
506 | list_for_each_entry(po1, head, list) { | |
507 | if (po == po1) { | |
508 | list_del_rcu(&po->list); | |
509 | goto out; | |
510 | } | |
511 | } | |
512 | ||
513 | pr_warn("dev_remove_offload: %p not found\n", po); | |
514 | out: | |
c53aa505 | 515 | spin_unlock(&offload_lock); |
62532da9 | 516 | } |
62532da9 VY |
517 | |
518 | /** | |
519 | * dev_remove_offload - remove packet offload handler | |
520 | * @po: packet offload declaration | |
521 | * | |
522 | * Remove a packet offload handler that was previously added to the kernel | |
523 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
524 | * removed from the kernel lists and can be freed or reused once this | |
525 | * function returns. | |
526 | * | |
527 | * This call sleeps to guarantee that no CPU is looking at the packet | |
528 | * type after return. | |
529 | */ | |
530 | void dev_remove_offload(struct packet_offload *po) | |
531 | { | |
532 | __dev_remove_offload(po); | |
533 | ||
534 | synchronize_net(); | |
535 | } | |
536 | EXPORT_SYMBOL(dev_remove_offload); | |
537 | ||
1da177e4 LT |
538 | /****************************************************************************** |
539 | ||
540 | Device Boot-time Settings Routines | |
541 | ||
542 | *******************************************************************************/ | |
543 | ||
544 | /* Boot time configuration table */ | |
545 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
546 | ||
547 | /** | |
548 | * netdev_boot_setup_add - add new setup entry | |
549 | * @name: name of the device | |
550 | * @map: configured settings for the device | |
551 | * | |
552 | * Adds new setup entry to the dev_boot_setup list. The function | |
553 | * returns 0 on error and 1 on success. This is a generic routine to | |
554 | * all netdevices. | |
555 | */ | |
556 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
557 | { | |
558 | struct netdev_boot_setup *s; | |
559 | int i; | |
560 | ||
561 | s = dev_boot_setup; | |
562 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
563 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
564 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 565 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
566 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
567 | break; | |
568 | } | |
569 | } | |
570 | ||
571 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
572 | } | |
573 | ||
574 | /** | |
575 | * netdev_boot_setup_check - check boot time settings | |
576 | * @dev: the netdevice | |
577 | * | |
578 | * Check boot time settings for the device. | |
579 | * The found settings are set for the device to be used | |
580 | * later in the device probing. | |
581 | * Returns 0 if no settings found, 1 if they are. | |
582 | */ | |
583 | int netdev_boot_setup_check(struct net_device *dev) | |
584 | { | |
585 | struct netdev_boot_setup *s = dev_boot_setup; | |
586 | int i; | |
587 | ||
588 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
589 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 590 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
591 | dev->irq = s[i].map.irq; |
592 | dev->base_addr = s[i].map.base_addr; | |
593 | dev->mem_start = s[i].map.mem_start; | |
594 | dev->mem_end = s[i].map.mem_end; | |
595 | return 1; | |
596 | } | |
597 | } | |
598 | return 0; | |
599 | } | |
d1b19dff | 600 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
601 | |
602 | ||
603 | /** | |
604 | * netdev_boot_base - get address from boot time settings | |
605 | * @prefix: prefix for network device | |
606 | * @unit: id for network device | |
607 | * | |
608 | * Check boot time settings for the base address of device. | |
609 | * The found settings are set for the device to be used | |
610 | * later in the device probing. | |
611 | * Returns 0 if no settings found. | |
612 | */ | |
613 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
614 | { | |
615 | const struct netdev_boot_setup *s = dev_boot_setup; | |
616 | char name[IFNAMSIZ]; | |
617 | int i; | |
618 | ||
619 | sprintf(name, "%s%d", prefix, unit); | |
620 | ||
621 | /* | |
622 | * If device already registered then return base of 1 | |
623 | * to indicate not to probe for this interface | |
624 | */ | |
881d966b | 625 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
626 | return 1; |
627 | ||
628 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
629 | if (!strcmp(name, s[i].name)) | |
630 | return s[i].map.base_addr; | |
631 | return 0; | |
632 | } | |
633 | ||
634 | /* | |
635 | * Saves at boot time configured settings for any netdevice. | |
636 | */ | |
637 | int __init netdev_boot_setup(char *str) | |
638 | { | |
639 | int ints[5]; | |
640 | struct ifmap map; | |
641 | ||
642 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
643 | if (!str || !*str) | |
644 | return 0; | |
645 | ||
646 | /* Save settings */ | |
647 | memset(&map, 0, sizeof(map)); | |
648 | if (ints[0] > 0) | |
649 | map.irq = ints[1]; | |
650 | if (ints[0] > 1) | |
651 | map.base_addr = ints[2]; | |
652 | if (ints[0] > 2) | |
653 | map.mem_start = ints[3]; | |
654 | if (ints[0] > 3) | |
655 | map.mem_end = ints[4]; | |
656 | ||
657 | /* Add new entry to the list */ | |
658 | return netdev_boot_setup_add(str, &map); | |
659 | } | |
660 | ||
661 | __setup("netdev=", netdev_boot_setup); | |
662 | ||
663 | /******************************************************************************* | |
664 | ||
665 | Device Interface Subroutines | |
666 | ||
667 | *******************************************************************************/ | |
668 | ||
a54acb3a ND |
669 | /** |
670 | * dev_get_iflink - get 'iflink' value of a interface | |
671 | * @dev: targeted interface | |
672 | * | |
673 | * Indicates the ifindex the interface is linked to. | |
674 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
675 | */ | |
676 | ||
677 | int dev_get_iflink(const struct net_device *dev) | |
678 | { | |
679 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
680 | return dev->netdev_ops->ndo_get_iflink(dev); | |
681 | ||
7a66bbc9 | 682 | return dev->ifindex; |
a54acb3a ND |
683 | } |
684 | EXPORT_SYMBOL(dev_get_iflink); | |
685 | ||
fc4099f1 PS |
686 | /** |
687 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
688 | * @dev: targeted interface | |
689 | * @skb: The packet. | |
690 | * | |
691 | * For better visibility of tunnel traffic OVS needs to retrieve | |
692 | * egress tunnel information for a packet. Following API allows | |
693 | * user to get this info. | |
694 | */ | |
695 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
696 | { | |
697 | struct ip_tunnel_info *info; | |
698 | ||
699 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
700 | return -EINVAL; | |
701 | ||
702 | info = skb_tunnel_info_unclone(skb); | |
703 | if (!info) | |
704 | return -ENOMEM; | |
705 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
706 | return -EINVAL; | |
707 | ||
708 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
709 | } | |
710 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
711 | ||
1da177e4 LT |
712 | /** |
713 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 714 | * @net: the applicable net namespace |
1da177e4 LT |
715 | * @name: name to find |
716 | * | |
717 | * Find an interface by name. Must be called under RTNL semaphore | |
718 | * or @dev_base_lock. If the name is found a pointer to the device | |
719 | * is returned. If the name is not found then %NULL is returned. The | |
720 | * reference counters are not incremented so the caller must be | |
721 | * careful with locks. | |
722 | */ | |
723 | ||
881d966b | 724 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 725 | { |
0bd8d536 ED |
726 | struct net_device *dev; |
727 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 728 | |
b67bfe0d | 729 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
730 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
731 | return dev; | |
0bd8d536 | 732 | |
1da177e4 LT |
733 | return NULL; |
734 | } | |
d1b19dff | 735 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 736 | |
72c9528b ED |
737 | /** |
738 | * dev_get_by_name_rcu - find a device by its name | |
739 | * @net: the applicable net namespace | |
740 | * @name: name to find | |
741 | * | |
742 | * Find an interface by name. | |
743 | * If the name is found a pointer to the device is returned. | |
744 | * If the name is not found then %NULL is returned. | |
745 | * The reference counters are not incremented so the caller must be | |
746 | * careful with locks. The caller must hold RCU lock. | |
747 | */ | |
748 | ||
749 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
750 | { | |
72c9528b ED |
751 | struct net_device *dev; |
752 | struct hlist_head *head = dev_name_hash(net, name); | |
753 | ||
b67bfe0d | 754 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
755 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
756 | return dev; | |
757 | ||
758 | return NULL; | |
759 | } | |
760 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
761 | ||
1da177e4 LT |
762 | /** |
763 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 764 | * @net: the applicable net namespace |
1da177e4 LT |
765 | * @name: name to find |
766 | * | |
767 | * Find an interface by name. This can be called from any | |
768 | * context and does its own locking. The returned handle has | |
769 | * the usage count incremented and the caller must use dev_put() to | |
770 | * release it when it is no longer needed. %NULL is returned if no | |
771 | * matching device is found. | |
772 | */ | |
773 | ||
881d966b | 774 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
775 | { |
776 | struct net_device *dev; | |
777 | ||
72c9528b ED |
778 | rcu_read_lock(); |
779 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
780 | if (dev) |
781 | dev_hold(dev); | |
72c9528b | 782 | rcu_read_unlock(); |
1da177e4 LT |
783 | return dev; |
784 | } | |
d1b19dff | 785 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
786 | |
787 | /** | |
788 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 789 | * @net: the applicable net namespace |
1da177e4 LT |
790 | * @ifindex: index of device |
791 | * | |
792 | * Search for an interface by index. Returns %NULL if the device | |
793 | * is not found or a pointer to the device. The device has not | |
794 | * had its reference counter increased so the caller must be careful | |
795 | * about locking. The caller must hold either the RTNL semaphore | |
796 | * or @dev_base_lock. | |
797 | */ | |
798 | ||
881d966b | 799 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 800 | { |
0bd8d536 ED |
801 | struct net_device *dev; |
802 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 803 | |
b67bfe0d | 804 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
805 | if (dev->ifindex == ifindex) |
806 | return dev; | |
0bd8d536 | 807 | |
1da177e4 LT |
808 | return NULL; |
809 | } | |
d1b19dff | 810 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 811 | |
fb699dfd ED |
812 | /** |
813 | * dev_get_by_index_rcu - find a device by its ifindex | |
814 | * @net: the applicable net namespace | |
815 | * @ifindex: index of device | |
816 | * | |
817 | * Search for an interface by index. Returns %NULL if the device | |
818 | * is not found or a pointer to the device. The device has not | |
819 | * had its reference counter increased so the caller must be careful | |
820 | * about locking. The caller must hold RCU lock. | |
821 | */ | |
822 | ||
823 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
824 | { | |
fb699dfd ED |
825 | struct net_device *dev; |
826 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
827 | ||
b67bfe0d | 828 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
829 | if (dev->ifindex == ifindex) |
830 | return dev; | |
831 | ||
832 | return NULL; | |
833 | } | |
834 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
835 | ||
1da177e4 LT |
836 | |
837 | /** | |
838 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 839 | * @net: the applicable net namespace |
1da177e4 LT |
840 | * @ifindex: index of device |
841 | * | |
842 | * Search for an interface by index. Returns NULL if the device | |
843 | * is not found or a pointer to the device. The device returned has | |
844 | * had a reference added and the pointer is safe until the user calls | |
845 | * dev_put to indicate they have finished with it. | |
846 | */ | |
847 | ||
881d966b | 848 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
849 | { |
850 | struct net_device *dev; | |
851 | ||
fb699dfd ED |
852 | rcu_read_lock(); |
853 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
854 | if (dev) |
855 | dev_hold(dev); | |
fb699dfd | 856 | rcu_read_unlock(); |
1da177e4 LT |
857 | return dev; |
858 | } | |
d1b19dff | 859 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 860 | |
5dbe7c17 NS |
861 | /** |
862 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
863 | * @net: network namespace | |
864 | * @name: a pointer to the buffer where the name will be stored. | |
865 | * @ifindex: the ifindex of the interface to get the name from. | |
866 | * | |
867 | * The use of raw_seqcount_begin() and cond_resched() before | |
868 | * retrying is required as we want to give the writers a chance | |
869 | * to complete when CONFIG_PREEMPT is not set. | |
870 | */ | |
871 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
872 | { | |
873 | struct net_device *dev; | |
874 | unsigned int seq; | |
875 | ||
876 | retry: | |
877 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
878 | rcu_read_lock(); | |
879 | dev = dev_get_by_index_rcu(net, ifindex); | |
880 | if (!dev) { | |
881 | rcu_read_unlock(); | |
882 | return -ENODEV; | |
883 | } | |
884 | ||
885 | strcpy(name, dev->name); | |
886 | rcu_read_unlock(); | |
887 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
888 | cond_resched(); | |
889 | goto retry; | |
890 | } | |
891 | ||
892 | return 0; | |
893 | } | |
894 | ||
1da177e4 | 895 | /** |
941666c2 | 896 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 897 | * @net: the applicable net namespace |
1da177e4 LT |
898 | * @type: media type of device |
899 | * @ha: hardware address | |
900 | * | |
901 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
902 | * is not found or a pointer to the device. |
903 | * The caller must hold RCU or RTNL. | |
941666c2 | 904 | * The returned device has not had its ref count increased |
1da177e4 LT |
905 | * and the caller must therefore be careful about locking |
906 | * | |
1da177e4 LT |
907 | */ |
908 | ||
941666c2 ED |
909 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
910 | const char *ha) | |
1da177e4 LT |
911 | { |
912 | struct net_device *dev; | |
913 | ||
941666c2 | 914 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
915 | if (dev->type == type && |
916 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
917 | return dev; |
918 | ||
919 | return NULL; | |
1da177e4 | 920 | } |
941666c2 | 921 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 922 | |
881d966b | 923 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
924 | { |
925 | struct net_device *dev; | |
926 | ||
4e9cac2b | 927 | ASSERT_RTNL(); |
881d966b | 928 | for_each_netdev(net, dev) |
4e9cac2b | 929 | if (dev->type == type) |
7562f876 PE |
930 | return dev; |
931 | ||
932 | return NULL; | |
4e9cac2b | 933 | } |
4e9cac2b PM |
934 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
935 | ||
881d966b | 936 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 937 | { |
99fe3c39 | 938 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 939 | |
99fe3c39 ED |
940 | rcu_read_lock(); |
941 | for_each_netdev_rcu(net, dev) | |
942 | if (dev->type == type) { | |
943 | dev_hold(dev); | |
944 | ret = dev; | |
945 | break; | |
946 | } | |
947 | rcu_read_unlock(); | |
948 | return ret; | |
1da177e4 | 949 | } |
1da177e4 LT |
950 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
951 | ||
952 | /** | |
6c555490 | 953 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 954 | * @net: the applicable net namespace |
1da177e4 LT |
955 | * @if_flags: IFF_* values |
956 | * @mask: bitmask of bits in if_flags to check | |
957 | * | |
958 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 959 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 960 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
961 | */ |
962 | ||
6c555490 WC |
963 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
964 | unsigned short mask) | |
1da177e4 | 965 | { |
7562f876 | 966 | struct net_device *dev, *ret; |
1da177e4 | 967 | |
6c555490 WC |
968 | ASSERT_RTNL(); |
969 | ||
7562f876 | 970 | ret = NULL; |
6c555490 | 971 | for_each_netdev(net, dev) { |
1da177e4 | 972 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 973 | ret = dev; |
1da177e4 LT |
974 | break; |
975 | } | |
976 | } | |
7562f876 | 977 | return ret; |
1da177e4 | 978 | } |
6c555490 | 979 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
980 | |
981 | /** | |
982 | * dev_valid_name - check if name is okay for network device | |
983 | * @name: name string | |
984 | * | |
985 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
986 | * to allow sysfs to work. We also disallow any kind of |
987 | * whitespace. | |
1da177e4 | 988 | */ |
95f050bf | 989 | bool dev_valid_name(const char *name) |
1da177e4 | 990 | { |
c7fa9d18 | 991 | if (*name == '\0') |
95f050bf | 992 | return false; |
b6fe17d6 | 993 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 994 | return false; |
c7fa9d18 | 995 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 996 | return false; |
c7fa9d18 DM |
997 | |
998 | while (*name) { | |
a4176a93 | 999 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1000 | return false; |
c7fa9d18 DM |
1001 | name++; |
1002 | } | |
95f050bf | 1003 | return true; |
1da177e4 | 1004 | } |
d1b19dff | 1005 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1006 | |
1007 | /** | |
b267b179 EB |
1008 | * __dev_alloc_name - allocate a name for a device |
1009 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1010 | * @name: name format string |
b267b179 | 1011 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1012 | * |
1013 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1014 | * id. It scans list of devices to build up a free map, then chooses |
1015 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1016 | * while allocating the name and adding the device in order to avoid | |
1017 | * duplicates. | |
1018 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1019 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1020 | */ |
1021 | ||
b267b179 | 1022 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1023 | { |
1024 | int i = 0; | |
1da177e4 LT |
1025 | const char *p; |
1026 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1027 | unsigned long *inuse; |
1da177e4 LT |
1028 | struct net_device *d; |
1029 | ||
1030 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
1031 | if (p) { | |
1032 | /* | |
1033 | * Verify the string as this thing may have come from | |
1034 | * the user. There must be either one "%d" and no other "%" | |
1035 | * characters. | |
1036 | */ | |
1037 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1038 | return -EINVAL; | |
1039 | ||
1040 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1041 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1042 | if (!inuse) |
1043 | return -ENOMEM; | |
1044 | ||
881d966b | 1045 | for_each_netdev(net, d) { |
1da177e4 LT |
1046 | if (!sscanf(d->name, name, &i)) |
1047 | continue; | |
1048 | if (i < 0 || i >= max_netdevices) | |
1049 | continue; | |
1050 | ||
1051 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1052 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1053 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1054 | set_bit(i, inuse); | |
1055 | } | |
1056 | ||
1057 | i = find_first_zero_bit(inuse, max_netdevices); | |
1058 | free_page((unsigned long) inuse); | |
1059 | } | |
1060 | ||
d9031024 OP |
1061 | if (buf != name) |
1062 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1063 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1064 | return i; |
1da177e4 LT |
1065 | |
1066 | /* It is possible to run out of possible slots | |
1067 | * when the name is long and there isn't enough space left | |
1068 | * for the digits, or if all bits are used. | |
1069 | */ | |
1070 | return -ENFILE; | |
1071 | } | |
1072 | ||
b267b179 EB |
1073 | /** |
1074 | * dev_alloc_name - allocate a name for a device | |
1075 | * @dev: device | |
1076 | * @name: name format string | |
1077 | * | |
1078 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1079 | * id. It scans list of devices to build up a free map, then chooses | |
1080 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1081 | * while allocating the name and adding the device in order to avoid | |
1082 | * duplicates. | |
1083 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1084 | * Returns the number of the unit assigned or a negative errno code. | |
1085 | */ | |
1086 | ||
1087 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1088 | { | |
1089 | char buf[IFNAMSIZ]; | |
1090 | struct net *net; | |
1091 | int ret; | |
1092 | ||
c346dca1 YH |
1093 | BUG_ON(!dev_net(dev)); |
1094 | net = dev_net(dev); | |
b267b179 EB |
1095 | ret = __dev_alloc_name(net, name, buf); |
1096 | if (ret >= 0) | |
1097 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1098 | return ret; | |
1099 | } | |
d1b19dff | 1100 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1101 | |
828de4f6 G |
1102 | static int dev_alloc_name_ns(struct net *net, |
1103 | struct net_device *dev, | |
1104 | const char *name) | |
d9031024 | 1105 | { |
828de4f6 G |
1106 | char buf[IFNAMSIZ]; |
1107 | int ret; | |
8ce6cebc | 1108 | |
828de4f6 G |
1109 | ret = __dev_alloc_name(net, name, buf); |
1110 | if (ret >= 0) | |
1111 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1112 | return ret; | |
1113 | } | |
1114 | ||
1115 | static int dev_get_valid_name(struct net *net, | |
1116 | struct net_device *dev, | |
1117 | const char *name) | |
1118 | { | |
1119 | BUG_ON(!net); | |
8ce6cebc | 1120 | |
d9031024 OP |
1121 | if (!dev_valid_name(name)) |
1122 | return -EINVAL; | |
1123 | ||
1c5cae81 | 1124 | if (strchr(name, '%')) |
828de4f6 | 1125 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1126 | else if (__dev_get_by_name(net, name)) |
1127 | return -EEXIST; | |
8ce6cebc DL |
1128 | else if (dev->name != name) |
1129 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1130 | |
1131 | return 0; | |
1132 | } | |
1da177e4 LT |
1133 | |
1134 | /** | |
1135 | * dev_change_name - change name of a device | |
1136 | * @dev: device | |
1137 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1138 | * | |
1139 | * Change name of a device, can pass format strings "eth%d". | |
1140 | * for wildcarding. | |
1141 | */ | |
cf04a4c7 | 1142 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1143 | { |
238fa362 | 1144 | unsigned char old_assign_type; |
fcc5a03a | 1145 | char oldname[IFNAMSIZ]; |
1da177e4 | 1146 | int err = 0; |
fcc5a03a | 1147 | int ret; |
881d966b | 1148 | struct net *net; |
1da177e4 LT |
1149 | |
1150 | ASSERT_RTNL(); | |
c346dca1 | 1151 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1152 | |
c346dca1 | 1153 | net = dev_net(dev); |
1da177e4 LT |
1154 | if (dev->flags & IFF_UP) |
1155 | return -EBUSY; | |
1156 | ||
30e6c9fa | 1157 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1158 | |
1159 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1160 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1161 | return 0; |
c91f6df2 | 1162 | } |
c8d90dca | 1163 | |
fcc5a03a HX |
1164 | memcpy(oldname, dev->name, IFNAMSIZ); |
1165 | ||
828de4f6 | 1166 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1167 | if (err < 0) { |
30e6c9fa | 1168 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1169 | return err; |
c91f6df2 | 1170 | } |
1da177e4 | 1171 | |
6fe82a39 VF |
1172 | if (oldname[0] && !strchr(oldname, '%')) |
1173 | netdev_info(dev, "renamed from %s\n", oldname); | |
1174 | ||
238fa362 TG |
1175 | old_assign_type = dev->name_assign_type; |
1176 | dev->name_assign_type = NET_NAME_RENAMED; | |
1177 | ||
fcc5a03a | 1178 | rollback: |
a1b3f594 EB |
1179 | ret = device_rename(&dev->dev, dev->name); |
1180 | if (ret) { | |
1181 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1182 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1183 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1184 | return ret; |
dcc99773 | 1185 | } |
7f988eab | 1186 | |
30e6c9fa | 1187 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1188 | |
5bb025fa VF |
1189 | netdev_adjacent_rename_links(dev, oldname); |
1190 | ||
7f988eab | 1191 | write_lock_bh(&dev_base_lock); |
372b2312 | 1192 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1193 | write_unlock_bh(&dev_base_lock); |
1194 | ||
1195 | synchronize_rcu(); | |
1196 | ||
1197 | write_lock_bh(&dev_base_lock); | |
1198 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1199 | write_unlock_bh(&dev_base_lock); |
1200 | ||
056925ab | 1201 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1202 | ret = notifier_to_errno(ret); |
1203 | ||
1204 | if (ret) { | |
91e9c07b ED |
1205 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1206 | if (err >= 0) { | |
fcc5a03a | 1207 | err = ret; |
30e6c9fa | 1208 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1209 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1210 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1211 | dev->name_assign_type = old_assign_type; |
1212 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1213 | goto rollback; |
91e9c07b | 1214 | } else { |
7b6cd1ce | 1215 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1216 | dev->name, ret); |
fcc5a03a HX |
1217 | } |
1218 | } | |
1da177e4 LT |
1219 | |
1220 | return err; | |
1221 | } | |
1222 | ||
0b815a1a SH |
1223 | /** |
1224 | * dev_set_alias - change ifalias of a device | |
1225 | * @dev: device | |
1226 | * @alias: name up to IFALIASZ | |
f0db275a | 1227 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1228 | * |
1229 | * Set ifalias for a device, | |
1230 | */ | |
1231 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1232 | { | |
7364e445 AK |
1233 | char *new_ifalias; |
1234 | ||
0b815a1a SH |
1235 | ASSERT_RTNL(); |
1236 | ||
1237 | if (len >= IFALIASZ) | |
1238 | return -EINVAL; | |
1239 | ||
96ca4a2c | 1240 | if (!len) { |
388dfc2d SK |
1241 | kfree(dev->ifalias); |
1242 | dev->ifalias = NULL; | |
96ca4a2c OH |
1243 | return 0; |
1244 | } | |
1245 | ||
7364e445 AK |
1246 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1247 | if (!new_ifalias) | |
0b815a1a | 1248 | return -ENOMEM; |
7364e445 | 1249 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1250 | |
1251 | strlcpy(dev->ifalias, alias, len+1); | |
1252 | return len; | |
1253 | } | |
1254 | ||
1255 | ||
d8a33ac4 | 1256 | /** |
3041a069 | 1257 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1258 | * @dev: device to cause notification |
1259 | * | |
1260 | * Called to indicate a device has changed features. | |
1261 | */ | |
1262 | void netdev_features_change(struct net_device *dev) | |
1263 | { | |
056925ab | 1264 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1265 | } |
1266 | EXPORT_SYMBOL(netdev_features_change); | |
1267 | ||
1da177e4 LT |
1268 | /** |
1269 | * netdev_state_change - device changes state | |
1270 | * @dev: device to cause notification | |
1271 | * | |
1272 | * Called to indicate a device has changed state. This function calls | |
1273 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1274 | * to the routing socket. | |
1275 | */ | |
1276 | void netdev_state_change(struct net_device *dev) | |
1277 | { | |
1278 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1279 | struct netdev_notifier_change_info change_info; |
1280 | ||
1281 | change_info.flags_changed = 0; | |
1282 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1283 | &change_info.info); | |
7f294054 | 1284 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1285 | } |
1286 | } | |
d1b19dff | 1287 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1288 | |
ee89bab1 AW |
1289 | /** |
1290 | * netdev_notify_peers - notify network peers about existence of @dev | |
1291 | * @dev: network device | |
1292 | * | |
1293 | * Generate traffic such that interested network peers are aware of | |
1294 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1295 | * a device wants to inform the rest of the network about some sort of | |
1296 | * reconfiguration such as a failover event or virtual machine | |
1297 | * migration. | |
1298 | */ | |
1299 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1300 | { |
ee89bab1 AW |
1301 | rtnl_lock(); |
1302 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1303 | rtnl_unlock(); | |
c1da4ac7 | 1304 | } |
ee89bab1 | 1305 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1306 | |
bd380811 | 1307 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1308 | { |
d314774c | 1309 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1310 | int ret; |
1da177e4 | 1311 | |
e46b66bc BH |
1312 | ASSERT_RTNL(); |
1313 | ||
1da177e4 LT |
1314 | if (!netif_device_present(dev)) |
1315 | return -ENODEV; | |
1316 | ||
ca99ca14 NH |
1317 | /* Block netpoll from trying to do any rx path servicing. |
1318 | * If we don't do this there is a chance ndo_poll_controller | |
1319 | * or ndo_poll may be running while we open the device | |
1320 | */ | |
66b5552f | 1321 | netpoll_poll_disable(dev); |
ca99ca14 | 1322 | |
3b8bcfd5 JB |
1323 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1324 | ret = notifier_to_errno(ret); | |
1325 | if (ret) | |
1326 | return ret; | |
1327 | ||
1da177e4 | 1328 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1329 | |
d314774c SH |
1330 | if (ops->ndo_validate_addr) |
1331 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1332 | |
d314774c SH |
1333 | if (!ret && ops->ndo_open) |
1334 | ret = ops->ndo_open(dev); | |
1da177e4 | 1335 | |
66b5552f | 1336 | netpoll_poll_enable(dev); |
ca99ca14 | 1337 | |
bada339b JG |
1338 | if (ret) |
1339 | clear_bit(__LINK_STATE_START, &dev->state); | |
1340 | else { | |
1da177e4 | 1341 | dev->flags |= IFF_UP; |
4417da66 | 1342 | dev_set_rx_mode(dev); |
1da177e4 | 1343 | dev_activate(dev); |
7bf23575 | 1344 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1345 | } |
bada339b | 1346 | |
1da177e4 LT |
1347 | return ret; |
1348 | } | |
1349 | ||
1350 | /** | |
bd380811 PM |
1351 | * dev_open - prepare an interface for use. |
1352 | * @dev: device to open | |
1da177e4 | 1353 | * |
bd380811 PM |
1354 | * Takes a device from down to up state. The device's private open |
1355 | * function is invoked and then the multicast lists are loaded. Finally | |
1356 | * the device is moved into the up state and a %NETDEV_UP message is | |
1357 | * sent to the netdev notifier chain. | |
1358 | * | |
1359 | * Calling this function on an active interface is a nop. On a failure | |
1360 | * a negative errno code is returned. | |
1da177e4 | 1361 | */ |
bd380811 PM |
1362 | int dev_open(struct net_device *dev) |
1363 | { | |
1364 | int ret; | |
1365 | ||
bd380811 PM |
1366 | if (dev->flags & IFF_UP) |
1367 | return 0; | |
1368 | ||
bd380811 PM |
1369 | ret = __dev_open(dev); |
1370 | if (ret < 0) | |
1371 | return ret; | |
1372 | ||
7f294054 | 1373 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1374 | call_netdevice_notifiers(NETDEV_UP, dev); |
1375 | ||
1376 | return ret; | |
1377 | } | |
1378 | EXPORT_SYMBOL(dev_open); | |
1379 | ||
44345724 | 1380 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1381 | { |
44345724 | 1382 | struct net_device *dev; |
e46b66bc | 1383 | |
bd380811 | 1384 | ASSERT_RTNL(); |
9d5010db DM |
1385 | might_sleep(); |
1386 | ||
5cde2829 | 1387 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1388 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1389 | netpoll_poll_disable(dev); |
3f4df206 | 1390 | |
44345724 | 1391 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1392 | |
44345724 | 1393 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1394 | |
44345724 OP |
1395 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1396 | * can be even on different cpu. So just clear netif_running(). | |
1397 | * | |
1398 | * dev->stop() will invoke napi_disable() on all of it's | |
1399 | * napi_struct instances on this device. | |
1400 | */ | |
4e857c58 | 1401 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1402 | } |
1da177e4 | 1403 | |
44345724 | 1404 | dev_deactivate_many(head); |
d8b2a4d2 | 1405 | |
5cde2829 | 1406 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1407 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1408 | |
44345724 OP |
1409 | /* |
1410 | * Call the device specific close. This cannot fail. | |
1411 | * Only if device is UP | |
1412 | * | |
1413 | * We allow it to be called even after a DETACH hot-plug | |
1414 | * event. | |
1415 | */ | |
1416 | if (ops->ndo_stop) | |
1417 | ops->ndo_stop(dev); | |
1418 | ||
44345724 | 1419 | dev->flags &= ~IFF_UP; |
66b5552f | 1420 | netpoll_poll_enable(dev); |
44345724 OP |
1421 | } |
1422 | ||
1423 | return 0; | |
1424 | } | |
1425 | ||
1426 | static int __dev_close(struct net_device *dev) | |
1427 | { | |
f87e6f47 | 1428 | int retval; |
44345724 OP |
1429 | LIST_HEAD(single); |
1430 | ||
5cde2829 | 1431 | list_add(&dev->close_list, &single); |
f87e6f47 LT |
1432 | retval = __dev_close_many(&single); |
1433 | list_del(&single); | |
ca99ca14 | 1434 | |
f87e6f47 | 1435 | return retval; |
44345724 OP |
1436 | } |
1437 | ||
99c4a26a | 1438 | int dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1439 | { |
1440 | struct net_device *dev, *tmp; | |
1da177e4 | 1441 | |
5cde2829 EB |
1442 | /* Remove the devices that don't need to be closed */ |
1443 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1444 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1445 | list_del_init(&dev->close_list); |
44345724 OP |
1446 | |
1447 | __dev_close_many(head); | |
1da177e4 | 1448 | |
5cde2829 | 1449 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1450 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1451 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1452 | if (unlink) |
1453 | list_del_init(&dev->close_list); | |
44345724 | 1454 | } |
bd380811 PM |
1455 | |
1456 | return 0; | |
1457 | } | |
99c4a26a | 1458 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1459 | |
1460 | /** | |
1461 | * dev_close - shutdown an interface. | |
1462 | * @dev: device to shutdown | |
1463 | * | |
1464 | * This function moves an active device into down state. A | |
1465 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1466 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1467 | * chain. | |
1468 | */ | |
1469 | int dev_close(struct net_device *dev) | |
1470 | { | |
e14a5993 ED |
1471 | if (dev->flags & IFF_UP) { |
1472 | LIST_HEAD(single); | |
1da177e4 | 1473 | |
5cde2829 | 1474 | list_add(&dev->close_list, &single); |
99c4a26a | 1475 | dev_close_many(&single, true); |
e14a5993 ED |
1476 | list_del(&single); |
1477 | } | |
da6e378b | 1478 | return 0; |
1da177e4 | 1479 | } |
d1b19dff | 1480 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1481 | |
1482 | ||
0187bdfb BH |
1483 | /** |
1484 | * dev_disable_lro - disable Large Receive Offload on a device | |
1485 | * @dev: device | |
1486 | * | |
1487 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1488 | * called under RTNL. This is needed if received packets may be | |
1489 | * forwarded to another interface. | |
1490 | */ | |
1491 | void dev_disable_lro(struct net_device *dev) | |
1492 | { | |
fbe168ba MK |
1493 | struct net_device *lower_dev; |
1494 | struct list_head *iter; | |
529d0489 | 1495 | |
bc5787c6 MM |
1496 | dev->wanted_features &= ~NETIF_F_LRO; |
1497 | netdev_update_features(dev); | |
27660515 | 1498 | |
22d5969f MM |
1499 | if (unlikely(dev->features & NETIF_F_LRO)) |
1500 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1501 | |
1502 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1503 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1504 | } |
1505 | EXPORT_SYMBOL(dev_disable_lro); | |
1506 | ||
351638e7 JP |
1507 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1508 | struct net_device *dev) | |
1509 | { | |
1510 | struct netdev_notifier_info info; | |
1511 | ||
1512 | netdev_notifier_info_init(&info, dev); | |
1513 | return nb->notifier_call(nb, val, &info); | |
1514 | } | |
0187bdfb | 1515 | |
881d966b EB |
1516 | static int dev_boot_phase = 1; |
1517 | ||
1da177e4 LT |
1518 | /** |
1519 | * register_netdevice_notifier - register a network notifier block | |
1520 | * @nb: notifier | |
1521 | * | |
1522 | * Register a notifier to be called when network device events occur. | |
1523 | * The notifier passed is linked into the kernel structures and must | |
1524 | * not be reused until it has been unregistered. A negative errno code | |
1525 | * is returned on a failure. | |
1526 | * | |
1527 | * When registered all registration and up events are replayed | |
4ec93edb | 1528 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1529 | * view of the network device list. |
1530 | */ | |
1531 | ||
1532 | int register_netdevice_notifier(struct notifier_block *nb) | |
1533 | { | |
1534 | struct net_device *dev; | |
fcc5a03a | 1535 | struct net_device *last; |
881d966b | 1536 | struct net *net; |
1da177e4 LT |
1537 | int err; |
1538 | ||
1539 | rtnl_lock(); | |
f07d5b94 | 1540 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1541 | if (err) |
1542 | goto unlock; | |
881d966b EB |
1543 | if (dev_boot_phase) |
1544 | goto unlock; | |
1545 | for_each_net(net) { | |
1546 | for_each_netdev(net, dev) { | |
351638e7 | 1547 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1548 | err = notifier_to_errno(err); |
1549 | if (err) | |
1550 | goto rollback; | |
1551 | ||
1552 | if (!(dev->flags & IFF_UP)) | |
1553 | continue; | |
1da177e4 | 1554 | |
351638e7 | 1555 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1556 | } |
1da177e4 | 1557 | } |
fcc5a03a HX |
1558 | |
1559 | unlock: | |
1da177e4 LT |
1560 | rtnl_unlock(); |
1561 | return err; | |
fcc5a03a HX |
1562 | |
1563 | rollback: | |
1564 | last = dev; | |
881d966b EB |
1565 | for_each_net(net) { |
1566 | for_each_netdev(net, dev) { | |
1567 | if (dev == last) | |
8f891489 | 1568 | goto outroll; |
fcc5a03a | 1569 | |
881d966b | 1570 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1571 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1572 | dev); | |
1573 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1574 | } |
351638e7 | 1575 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1576 | } |
fcc5a03a | 1577 | } |
c67625a1 | 1578 | |
8f891489 | 1579 | outroll: |
c67625a1 | 1580 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1581 | goto unlock; |
1da177e4 | 1582 | } |
d1b19dff | 1583 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1584 | |
1585 | /** | |
1586 | * unregister_netdevice_notifier - unregister a network notifier block | |
1587 | * @nb: notifier | |
1588 | * | |
1589 | * Unregister a notifier previously registered by | |
1590 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1591 | * kernel structures and may then be reused. A negative errno code | |
1592 | * is returned on a failure. | |
7d3d43da EB |
1593 | * |
1594 | * After unregistering unregister and down device events are synthesized | |
1595 | * for all devices on the device list to the removed notifier to remove | |
1596 | * the need for special case cleanup code. | |
1da177e4 LT |
1597 | */ |
1598 | ||
1599 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1600 | { | |
7d3d43da EB |
1601 | struct net_device *dev; |
1602 | struct net *net; | |
9f514950 HX |
1603 | int err; |
1604 | ||
1605 | rtnl_lock(); | |
f07d5b94 | 1606 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1607 | if (err) |
1608 | goto unlock; | |
1609 | ||
1610 | for_each_net(net) { | |
1611 | for_each_netdev(net, dev) { | |
1612 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1613 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1614 | dev); | |
1615 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1616 | } |
351638e7 | 1617 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1618 | } |
1619 | } | |
1620 | unlock: | |
9f514950 HX |
1621 | rtnl_unlock(); |
1622 | return err; | |
1da177e4 | 1623 | } |
d1b19dff | 1624 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1625 | |
351638e7 JP |
1626 | /** |
1627 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1628 | * @val: value passed unmodified to notifier function | |
1629 | * @dev: net_device pointer passed unmodified to notifier function | |
1630 | * @info: notifier information data | |
1631 | * | |
1632 | * Call all network notifier blocks. Parameters and return value | |
1633 | * are as for raw_notifier_call_chain(). | |
1634 | */ | |
1635 | ||
1d143d9f | 1636 | static int call_netdevice_notifiers_info(unsigned long val, |
1637 | struct net_device *dev, | |
1638 | struct netdev_notifier_info *info) | |
351638e7 JP |
1639 | { |
1640 | ASSERT_RTNL(); | |
1641 | netdev_notifier_info_init(info, dev); | |
1642 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1643 | } | |
351638e7 | 1644 | |
1da177e4 LT |
1645 | /** |
1646 | * call_netdevice_notifiers - call all network notifier blocks | |
1647 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1648 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1649 | * |
1650 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1651 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1652 | */ |
1653 | ||
ad7379d4 | 1654 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1655 | { |
351638e7 JP |
1656 | struct netdev_notifier_info info; |
1657 | ||
1658 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1659 | } |
edf947f1 | 1660 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1661 | |
1cf51900 | 1662 | #ifdef CONFIG_NET_INGRESS |
4577139b DB |
1663 | static struct static_key ingress_needed __read_mostly; |
1664 | ||
1665 | void net_inc_ingress_queue(void) | |
1666 | { | |
1667 | static_key_slow_inc(&ingress_needed); | |
1668 | } | |
1669 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
1670 | ||
1671 | void net_dec_ingress_queue(void) | |
1672 | { | |
1673 | static_key_slow_dec(&ingress_needed); | |
1674 | } | |
1675 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
1676 | #endif | |
1677 | ||
c5905afb | 1678 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1679 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1680 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1681 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1682 | * static_key_slow_dec() calls. |
b90e5794 ED |
1683 | */ |
1684 | static atomic_t netstamp_needed_deferred; | |
1685 | #endif | |
1da177e4 LT |
1686 | |
1687 | void net_enable_timestamp(void) | |
1688 | { | |
b90e5794 ED |
1689 | #ifdef HAVE_JUMP_LABEL |
1690 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1691 | ||
1692 | if (deferred) { | |
1693 | while (--deferred) | |
c5905afb | 1694 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1695 | return; |
1696 | } | |
1697 | #endif | |
c5905afb | 1698 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1699 | } |
d1b19dff | 1700 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1701 | |
1702 | void net_disable_timestamp(void) | |
1703 | { | |
b90e5794 ED |
1704 | #ifdef HAVE_JUMP_LABEL |
1705 | if (in_interrupt()) { | |
1706 | atomic_inc(&netstamp_needed_deferred); | |
1707 | return; | |
1708 | } | |
1709 | #endif | |
c5905afb | 1710 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1711 | } |
d1b19dff | 1712 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1713 | |
3b098e2d | 1714 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1715 | { |
588f0330 | 1716 | skb->tstamp.tv64 = 0; |
c5905afb | 1717 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1718 | __net_timestamp(skb); |
1da177e4 LT |
1719 | } |
1720 | ||
588f0330 | 1721 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1722 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1723 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1724 | __net_timestamp(SKB); \ | |
1725 | } \ | |
3b098e2d | 1726 | |
1ee481fb | 1727 | bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb) |
79b569f0 DL |
1728 | { |
1729 | unsigned int len; | |
1730 | ||
1731 | if (!(dev->flags & IFF_UP)) | |
1732 | return false; | |
1733 | ||
1734 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1735 | if (skb->len <= len) | |
1736 | return true; | |
1737 | ||
1738 | /* if TSO is enabled, we don't care about the length as the packet | |
1739 | * could be forwarded without being segmented before | |
1740 | */ | |
1741 | if (skb_is_gso(skb)) | |
1742 | return true; | |
1743 | ||
1744 | return false; | |
1745 | } | |
1ee481fb | 1746 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1747 | |
a0265d28 HX |
1748 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1749 | { | |
bbbf2df0 WB |
1750 | if (skb_orphan_frags(skb, GFP_ATOMIC) || |
1751 | unlikely(!is_skb_forwardable(dev, skb))) { | |
a0265d28 HX |
1752 | atomic_long_inc(&dev->rx_dropped); |
1753 | kfree_skb(skb); | |
1754 | return NET_RX_DROP; | |
1755 | } | |
1756 | ||
1757 | skb_scrub_packet(skb, true); | |
08b4b8ea | 1758 | skb->priority = 0; |
a0265d28 | 1759 | skb->protocol = eth_type_trans(skb, dev); |
2c26d34b | 1760 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); |
a0265d28 HX |
1761 | |
1762 | return 0; | |
1763 | } | |
1764 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1765 | ||
44540960 AB |
1766 | /** |
1767 | * dev_forward_skb - loopback an skb to another netif | |
1768 | * | |
1769 | * @dev: destination network device | |
1770 | * @skb: buffer to forward | |
1771 | * | |
1772 | * return values: | |
1773 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1774 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1775 | * |
1776 | * dev_forward_skb can be used for injecting an skb from the | |
1777 | * start_xmit function of one device into the receive queue | |
1778 | * of another device. | |
1779 | * | |
1780 | * The receiving device may be in another namespace, so | |
1781 | * we have to clear all information in the skb that could | |
1782 | * impact namespace isolation. | |
1783 | */ | |
1784 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1785 | { | |
a0265d28 | 1786 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1787 | } |
1788 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1789 | ||
71d9dec2 CG |
1790 | static inline int deliver_skb(struct sk_buff *skb, |
1791 | struct packet_type *pt_prev, | |
1792 | struct net_device *orig_dev) | |
1793 | { | |
1080e512 MT |
1794 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1795 | return -ENOMEM; | |
71d9dec2 CG |
1796 | atomic_inc(&skb->users); |
1797 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1798 | } | |
1799 | ||
7866a621 SN |
1800 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
1801 | struct packet_type **pt, | |
fbcb2170 JP |
1802 | struct net_device *orig_dev, |
1803 | __be16 type, | |
7866a621 SN |
1804 | struct list_head *ptype_list) |
1805 | { | |
1806 | struct packet_type *ptype, *pt_prev = *pt; | |
1807 | ||
1808 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1809 | if (ptype->type != type) | |
1810 | continue; | |
1811 | if (pt_prev) | |
fbcb2170 | 1812 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
1813 | pt_prev = ptype; |
1814 | } | |
1815 | *pt = pt_prev; | |
1816 | } | |
1817 | ||
c0de08d0 EL |
1818 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1819 | { | |
a3d744e9 | 1820 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1821 | return false; |
1822 | ||
1823 | if (ptype->id_match) | |
1824 | return ptype->id_match(ptype, skb->sk); | |
1825 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1826 | return true; | |
1827 | ||
1828 | return false; | |
1829 | } | |
1830 | ||
1da177e4 LT |
1831 | /* |
1832 | * Support routine. Sends outgoing frames to any network | |
1833 | * taps currently in use. | |
1834 | */ | |
1835 | ||
f6a78bfc | 1836 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1837 | { |
1838 | struct packet_type *ptype; | |
71d9dec2 CG |
1839 | struct sk_buff *skb2 = NULL; |
1840 | struct packet_type *pt_prev = NULL; | |
7866a621 | 1841 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 1842 | |
1da177e4 | 1843 | rcu_read_lock(); |
7866a621 SN |
1844 | again: |
1845 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1da177e4 LT |
1846 | /* Never send packets back to the socket |
1847 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1848 | */ | |
7866a621 SN |
1849 | if (skb_loop_sk(ptype, skb)) |
1850 | continue; | |
71d9dec2 | 1851 | |
7866a621 SN |
1852 | if (pt_prev) { |
1853 | deliver_skb(skb2, pt_prev, skb->dev); | |
1854 | pt_prev = ptype; | |
1855 | continue; | |
1856 | } | |
1da177e4 | 1857 | |
7866a621 SN |
1858 | /* need to clone skb, done only once */ |
1859 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1860 | if (!skb2) | |
1861 | goto out_unlock; | |
70978182 | 1862 | |
7866a621 | 1863 | net_timestamp_set(skb2); |
1da177e4 | 1864 | |
7866a621 SN |
1865 | /* skb->nh should be correctly |
1866 | * set by sender, so that the second statement is | |
1867 | * just protection against buggy protocols. | |
1868 | */ | |
1869 | skb_reset_mac_header(skb2); | |
1870 | ||
1871 | if (skb_network_header(skb2) < skb2->data || | |
1872 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
1873 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
1874 | ntohs(skb2->protocol), | |
1875 | dev->name); | |
1876 | skb_reset_network_header(skb2); | |
1da177e4 | 1877 | } |
7866a621 SN |
1878 | |
1879 | skb2->transport_header = skb2->network_header; | |
1880 | skb2->pkt_type = PACKET_OUTGOING; | |
1881 | pt_prev = ptype; | |
1882 | } | |
1883 | ||
1884 | if (ptype_list == &ptype_all) { | |
1885 | ptype_list = &dev->ptype_all; | |
1886 | goto again; | |
1da177e4 | 1887 | } |
7866a621 | 1888 | out_unlock: |
71d9dec2 CG |
1889 | if (pt_prev) |
1890 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1891 | rcu_read_unlock(); |
1892 | } | |
1893 | ||
2c53040f BH |
1894 | /** |
1895 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1896 | * @dev: Network device |
1897 | * @txq: number of queues available | |
1898 | * | |
1899 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1900 | * valid. To resolve this verify the tc mapping remains valid and if | |
1901 | * not NULL the mapping. With no priorities mapping to this | |
1902 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1903 | * is invalid nothing can be done so disable priority mappings. If is | |
1904 | * expected that drivers will fix this mapping if they can before | |
1905 | * calling netif_set_real_num_tx_queues. | |
1906 | */ | |
bb134d22 | 1907 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1908 | { |
1909 | int i; | |
1910 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1911 | ||
1912 | /* If TC0 is invalidated disable TC mapping */ | |
1913 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1914 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1915 | dev->num_tc = 0; |
1916 | return; | |
1917 | } | |
1918 | ||
1919 | /* Invalidated prio to tc mappings set to TC0 */ | |
1920 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1921 | int q = netdev_get_prio_tc_map(dev, i); | |
1922 | ||
1923 | tc = &dev->tc_to_txq[q]; | |
1924 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1925 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1926 | i, q); | |
4f57c087 JF |
1927 | netdev_set_prio_tc_map(dev, i, 0); |
1928 | } | |
1929 | } | |
1930 | } | |
1931 | ||
537c00de AD |
1932 | #ifdef CONFIG_XPS |
1933 | static DEFINE_MUTEX(xps_map_mutex); | |
1934 | #define xmap_dereference(P) \ | |
1935 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1936 | ||
10cdc3f3 AD |
1937 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1938 | int cpu, u16 index) | |
537c00de | 1939 | { |
10cdc3f3 AD |
1940 | struct xps_map *map = NULL; |
1941 | int pos; | |
537c00de | 1942 | |
10cdc3f3 AD |
1943 | if (dev_maps) |
1944 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1945 | |
10cdc3f3 AD |
1946 | for (pos = 0; map && pos < map->len; pos++) { |
1947 | if (map->queues[pos] == index) { | |
537c00de AD |
1948 | if (map->len > 1) { |
1949 | map->queues[pos] = map->queues[--map->len]; | |
1950 | } else { | |
10cdc3f3 | 1951 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1952 | kfree_rcu(map, rcu); |
1953 | map = NULL; | |
1954 | } | |
10cdc3f3 | 1955 | break; |
537c00de | 1956 | } |
537c00de AD |
1957 | } |
1958 | ||
10cdc3f3 AD |
1959 | return map; |
1960 | } | |
1961 | ||
024e9679 | 1962 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
10cdc3f3 AD |
1963 | { |
1964 | struct xps_dev_maps *dev_maps; | |
024e9679 | 1965 | int cpu, i; |
10cdc3f3 AD |
1966 | bool active = false; |
1967 | ||
1968 | mutex_lock(&xps_map_mutex); | |
1969 | dev_maps = xmap_dereference(dev->xps_maps); | |
1970 | ||
1971 | if (!dev_maps) | |
1972 | goto out_no_maps; | |
1973 | ||
1974 | for_each_possible_cpu(cpu) { | |
024e9679 AD |
1975 | for (i = index; i < dev->num_tx_queues; i++) { |
1976 | if (!remove_xps_queue(dev_maps, cpu, i)) | |
1977 | break; | |
1978 | } | |
1979 | if (i == dev->num_tx_queues) | |
10cdc3f3 AD |
1980 | active = true; |
1981 | } | |
1982 | ||
1983 | if (!active) { | |
537c00de AD |
1984 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
1985 | kfree_rcu(dev_maps, rcu); | |
1986 | } | |
1987 | ||
024e9679 AD |
1988 | for (i = index; i < dev->num_tx_queues; i++) |
1989 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), | |
1990 | NUMA_NO_NODE); | |
1991 | ||
537c00de AD |
1992 | out_no_maps: |
1993 | mutex_unlock(&xps_map_mutex); | |
1994 | } | |
1995 | ||
01c5f864 AD |
1996 | static struct xps_map *expand_xps_map(struct xps_map *map, |
1997 | int cpu, u16 index) | |
1998 | { | |
1999 | struct xps_map *new_map; | |
2000 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2001 | int i, pos; | |
2002 | ||
2003 | for (pos = 0; map && pos < map->len; pos++) { | |
2004 | if (map->queues[pos] != index) | |
2005 | continue; | |
2006 | return map; | |
2007 | } | |
2008 | ||
2009 | /* Need to add queue to this CPU's existing map */ | |
2010 | if (map) { | |
2011 | if (pos < map->alloc_len) | |
2012 | return map; | |
2013 | ||
2014 | alloc_len = map->alloc_len * 2; | |
2015 | } | |
2016 | ||
2017 | /* Need to allocate new map to store queue on this CPU's map */ | |
2018 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2019 | cpu_to_node(cpu)); | |
2020 | if (!new_map) | |
2021 | return NULL; | |
2022 | ||
2023 | for (i = 0; i < pos; i++) | |
2024 | new_map->queues[i] = map->queues[i]; | |
2025 | new_map->alloc_len = alloc_len; | |
2026 | new_map->len = pos; | |
2027 | ||
2028 | return new_map; | |
2029 | } | |
2030 | ||
3573540c MT |
2031 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
2032 | u16 index) | |
537c00de | 2033 | { |
01c5f864 | 2034 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 2035 | struct xps_map *map, *new_map; |
537c00de | 2036 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
2037 | int cpu, numa_node_id = -2; |
2038 | bool active = false; | |
537c00de AD |
2039 | |
2040 | mutex_lock(&xps_map_mutex); | |
2041 | ||
2042 | dev_maps = xmap_dereference(dev->xps_maps); | |
2043 | ||
01c5f864 AD |
2044 | /* allocate memory for queue storage */ |
2045 | for_each_online_cpu(cpu) { | |
2046 | if (!cpumask_test_cpu(cpu, mask)) | |
2047 | continue; | |
2048 | ||
2049 | if (!new_dev_maps) | |
2050 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2051 | if (!new_dev_maps) { |
2052 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2053 | return -ENOMEM; |
2bb60cb9 | 2054 | } |
01c5f864 AD |
2055 | |
2056 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2057 | NULL; | |
2058 | ||
2059 | map = expand_xps_map(map, cpu, index); | |
2060 | if (!map) | |
2061 | goto error; | |
2062 | ||
2063 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
2064 | } | |
2065 | ||
2066 | if (!new_dev_maps) | |
2067 | goto out_no_new_maps; | |
2068 | ||
537c00de | 2069 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
2070 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
2071 | /* add queue to CPU maps */ | |
2072 | int pos = 0; | |
2073 | ||
2074 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2075 | while ((pos < map->len) && (map->queues[pos] != index)) | |
2076 | pos++; | |
2077 | ||
2078 | if (pos == map->len) | |
2079 | map->queues[map->len++] = index; | |
537c00de | 2080 | #ifdef CONFIG_NUMA |
537c00de AD |
2081 | if (numa_node_id == -2) |
2082 | numa_node_id = cpu_to_node(cpu); | |
2083 | else if (numa_node_id != cpu_to_node(cpu)) | |
2084 | numa_node_id = -1; | |
537c00de | 2085 | #endif |
01c5f864 AD |
2086 | } else if (dev_maps) { |
2087 | /* fill in the new device map from the old device map */ | |
2088 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2089 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 2090 | } |
01c5f864 | 2091 | |
537c00de AD |
2092 | } |
2093 | ||
01c5f864 AD |
2094 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2095 | ||
537c00de | 2096 | /* Cleanup old maps */ |
01c5f864 AD |
2097 | if (dev_maps) { |
2098 | for_each_possible_cpu(cpu) { | |
2099 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2100 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2101 | if (map && map != new_map) | |
2102 | kfree_rcu(map, rcu); | |
2103 | } | |
537c00de | 2104 | |
01c5f864 | 2105 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
2106 | } |
2107 | ||
01c5f864 AD |
2108 | dev_maps = new_dev_maps; |
2109 | active = true; | |
537c00de | 2110 | |
01c5f864 AD |
2111 | out_no_new_maps: |
2112 | /* update Tx queue numa node */ | |
537c00de AD |
2113 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2114 | (numa_node_id >= 0) ? numa_node_id : | |
2115 | NUMA_NO_NODE); | |
2116 | ||
01c5f864 AD |
2117 | if (!dev_maps) |
2118 | goto out_no_maps; | |
2119 | ||
2120 | /* removes queue from unused CPUs */ | |
2121 | for_each_possible_cpu(cpu) { | |
2122 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
2123 | continue; | |
2124 | ||
2125 | if (remove_xps_queue(dev_maps, cpu, index)) | |
2126 | active = true; | |
2127 | } | |
2128 | ||
2129 | /* free map if not active */ | |
2130 | if (!active) { | |
2131 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2132 | kfree_rcu(dev_maps, rcu); | |
2133 | } | |
2134 | ||
2135 | out_no_maps: | |
537c00de AD |
2136 | mutex_unlock(&xps_map_mutex); |
2137 | ||
2138 | return 0; | |
2139 | error: | |
01c5f864 AD |
2140 | /* remove any maps that we added */ |
2141 | for_each_possible_cpu(cpu) { | |
2142 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2143 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2144 | NULL; | |
2145 | if (new_map && new_map != map) | |
2146 | kfree(new_map); | |
2147 | } | |
2148 | ||
537c00de AD |
2149 | mutex_unlock(&xps_map_mutex); |
2150 | ||
537c00de AD |
2151 | kfree(new_dev_maps); |
2152 | return -ENOMEM; | |
2153 | } | |
2154 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2155 | ||
2156 | #endif | |
f0796d5c JF |
2157 | /* |
2158 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2159 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2160 | */ | |
e6484930 | 2161 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2162 | { |
1d24eb48 TH |
2163 | int rc; |
2164 | ||
e6484930 TH |
2165 | if (txq < 1 || txq > dev->num_tx_queues) |
2166 | return -EINVAL; | |
f0796d5c | 2167 | |
5c56580b BH |
2168 | if (dev->reg_state == NETREG_REGISTERED || |
2169 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2170 | ASSERT_RTNL(); |
2171 | ||
1d24eb48 TH |
2172 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2173 | txq); | |
bf264145 TH |
2174 | if (rc) |
2175 | return rc; | |
2176 | ||
4f57c087 JF |
2177 | if (dev->num_tc) |
2178 | netif_setup_tc(dev, txq); | |
2179 | ||
024e9679 | 2180 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2181 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2182 | #ifdef CONFIG_XPS |
2183 | netif_reset_xps_queues_gt(dev, txq); | |
2184 | #endif | |
2185 | } | |
f0796d5c | 2186 | } |
e6484930 TH |
2187 | |
2188 | dev->real_num_tx_queues = txq; | |
2189 | return 0; | |
f0796d5c JF |
2190 | } |
2191 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2192 | |
a953be53 | 2193 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2194 | /** |
2195 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2196 | * @dev: Network device | |
2197 | * @rxq: Actual number of RX queues | |
2198 | * | |
2199 | * This must be called either with the rtnl_lock held or before | |
2200 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2201 | * negative error code. If called before registration, it always |
2202 | * succeeds. | |
62fe0b40 BH |
2203 | */ |
2204 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2205 | { | |
2206 | int rc; | |
2207 | ||
bd25fa7b TH |
2208 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2209 | return -EINVAL; | |
2210 | ||
62fe0b40 BH |
2211 | if (dev->reg_state == NETREG_REGISTERED) { |
2212 | ASSERT_RTNL(); | |
2213 | ||
62fe0b40 BH |
2214 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2215 | rxq); | |
2216 | if (rc) | |
2217 | return rc; | |
62fe0b40 BH |
2218 | } |
2219 | ||
2220 | dev->real_num_rx_queues = rxq; | |
2221 | return 0; | |
2222 | } | |
2223 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2224 | #endif | |
2225 | ||
2c53040f BH |
2226 | /** |
2227 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2228 | * |
2229 | * This routine should set an upper limit on the number of RSS queues | |
2230 | * used by default by multiqueue devices. | |
2231 | */ | |
a55b138b | 2232 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
2233 | { |
2234 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
2235 | } | |
2236 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2237 | ||
def82a1d | 2238 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 2239 | { |
def82a1d JP |
2240 | struct softnet_data *sd; |
2241 | unsigned long flags; | |
56079431 | 2242 | |
def82a1d | 2243 | local_irq_save(flags); |
903ceff7 | 2244 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2245 | q->next_sched = NULL; |
2246 | *sd->output_queue_tailp = q; | |
2247 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2248 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2249 | local_irq_restore(flags); | |
2250 | } | |
2251 | ||
2252 | void __netif_schedule(struct Qdisc *q) | |
2253 | { | |
2254 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2255 | __netif_reschedule(q); | |
56079431 DV |
2256 | } |
2257 | EXPORT_SYMBOL(__netif_schedule); | |
2258 | ||
e6247027 ED |
2259 | struct dev_kfree_skb_cb { |
2260 | enum skb_free_reason reason; | |
2261 | }; | |
2262 | ||
2263 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2264 | { |
e6247027 ED |
2265 | return (struct dev_kfree_skb_cb *)skb->cb; |
2266 | } | |
2267 | ||
46e5da40 JF |
2268 | void netif_schedule_queue(struct netdev_queue *txq) |
2269 | { | |
2270 | rcu_read_lock(); | |
2271 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2272 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2273 | ||
2274 | __netif_schedule(q); | |
2275 | } | |
2276 | rcu_read_unlock(); | |
2277 | } | |
2278 | EXPORT_SYMBOL(netif_schedule_queue); | |
2279 | ||
2280 | /** | |
2281 | * netif_wake_subqueue - allow sending packets on subqueue | |
2282 | * @dev: network device | |
2283 | * @queue_index: sub queue index | |
2284 | * | |
2285 | * Resume individual transmit queue of a device with multiple transmit queues. | |
2286 | */ | |
2287 | void netif_wake_subqueue(struct net_device *dev, u16 queue_index) | |
2288 | { | |
2289 | struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index); | |
2290 | ||
2291 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) { | |
2292 | struct Qdisc *q; | |
2293 | ||
2294 | rcu_read_lock(); | |
2295 | q = rcu_dereference(txq->qdisc); | |
2296 | __netif_schedule(q); | |
2297 | rcu_read_unlock(); | |
2298 | } | |
2299 | } | |
2300 | EXPORT_SYMBOL(netif_wake_subqueue); | |
2301 | ||
2302 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) | |
2303 | { | |
2304 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2305 | struct Qdisc *q; | |
2306 | ||
2307 | rcu_read_lock(); | |
2308 | q = rcu_dereference(dev_queue->qdisc); | |
2309 | __netif_schedule(q); | |
2310 | rcu_read_unlock(); | |
2311 | } | |
2312 | } | |
2313 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2314 | ||
e6247027 | 2315 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2316 | { |
e6247027 | 2317 | unsigned long flags; |
56079431 | 2318 | |
e6247027 ED |
2319 | if (likely(atomic_read(&skb->users) == 1)) { |
2320 | smp_rmb(); | |
2321 | atomic_set(&skb->users, 0); | |
2322 | } else if (likely(!atomic_dec_and_test(&skb->users))) { | |
2323 | return; | |
bea3348e | 2324 | } |
e6247027 ED |
2325 | get_kfree_skb_cb(skb)->reason = reason; |
2326 | local_irq_save(flags); | |
2327 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2328 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2329 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2330 | local_irq_restore(flags); | |
56079431 | 2331 | } |
e6247027 | 2332 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2333 | |
e6247027 | 2334 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2335 | { |
2336 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2337 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2338 | else |
2339 | dev_kfree_skb(skb); | |
2340 | } | |
e6247027 | 2341 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2342 | |
2343 | ||
bea3348e SH |
2344 | /** |
2345 | * netif_device_detach - mark device as removed | |
2346 | * @dev: network device | |
2347 | * | |
2348 | * Mark device as removed from system and therefore no longer available. | |
2349 | */ | |
56079431 DV |
2350 | void netif_device_detach(struct net_device *dev) |
2351 | { | |
2352 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2353 | netif_running(dev)) { | |
d543103a | 2354 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2355 | } |
2356 | } | |
2357 | EXPORT_SYMBOL(netif_device_detach); | |
2358 | ||
bea3348e SH |
2359 | /** |
2360 | * netif_device_attach - mark device as attached | |
2361 | * @dev: network device | |
2362 | * | |
2363 | * Mark device as attached from system and restart if needed. | |
2364 | */ | |
56079431 DV |
2365 | void netif_device_attach(struct net_device *dev) |
2366 | { | |
2367 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2368 | netif_running(dev)) { | |
d543103a | 2369 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2370 | __netdev_watchdog_up(dev); |
56079431 DV |
2371 | } |
2372 | } | |
2373 | EXPORT_SYMBOL(netif_device_attach); | |
2374 | ||
5605c762 JP |
2375 | /* |
2376 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2377 | * to be used as a distribution range. | |
2378 | */ | |
2379 | u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, | |
2380 | unsigned int num_tx_queues) | |
2381 | { | |
2382 | u32 hash; | |
2383 | u16 qoffset = 0; | |
2384 | u16 qcount = num_tx_queues; | |
2385 | ||
2386 | if (skb_rx_queue_recorded(skb)) { | |
2387 | hash = skb_get_rx_queue(skb); | |
2388 | while (unlikely(hash >= num_tx_queues)) | |
2389 | hash -= num_tx_queues; | |
2390 | return hash; | |
2391 | } | |
2392 | ||
2393 | if (dev->num_tc) { | |
2394 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
2395 | qoffset = dev->tc_to_txq[tc].offset; | |
2396 | qcount = dev->tc_to_txq[tc].count; | |
2397 | } | |
2398 | ||
2399 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
2400 | } | |
2401 | EXPORT_SYMBOL(__skb_tx_hash); | |
2402 | ||
36c92474 BH |
2403 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2404 | { | |
65e9d2fa | 2405 | static const netdev_features_t null_features = 0; |
36c92474 | 2406 | struct net_device *dev = skb->dev; |
88ad4175 | 2407 | const char *name = ""; |
36c92474 | 2408 | |
c846ad9b BG |
2409 | if (!net_ratelimit()) |
2410 | return; | |
2411 | ||
88ad4175 BM |
2412 | if (dev) { |
2413 | if (dev->dev.parent) | |
2414 | name = dev_driver_string(dev->dev.parent); | |
2415 | else | |
2416 | name = netdev_name(dev); | |
2417 | } | |
36c92474 BH |
2418 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " |
2419 | "gso_type=%d ip_summed=%d\n", | |
88ad4175 | 2420 | name, dev ? &dev->features : &null_features, |
65e9d2fa | 2421 | skb->sk ? &skb->sk->sk_route_caps : &null_features, |
36c92474 BH |
2422 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2423 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2424 | } | |
2425 | ||
1da177e4 LT |
2426 | /* |
2427 | * Invalidate hardware checksum when packet is to be mangled, and | |
2428 | * complete checksum manually on outgoing path. | |
2429 | */ | |
84fa7933 | 2430 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2431 | { |
d3bc23e7 | 2432 | __wsum csum; |
663ead3b | 2433 | int ret = 0, offset; |
1da177e4 | 2434 | |
84fa7933 | 2435 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2436 | goto out_set_summed; |
2437 | ||
2438 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2439 | skb_warn_bad_offload(skb); |
2440 | return -EINVAL; | |
1da177e4 LT |
2441 | } |
2442 | ||
cef401de ED |
2443 | /* Before computing a checksum, we should make sure no frag could |
2444 | * be modified by an external entity : checksum could be wrong. | |
2445 | */ | |
2446 | if (skb_has_shared_frag(skb)) { | |
2447 | ret = __skb_linearize(skb); | |
2448 | if (ret) | |
2449 | goto out; | |
2450 | } | |
2451 | ||
55508d60 | 2452 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2453 | BUG_ON(offset >= skb_headlen(skb)); |
2454 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2455 | ||
2456 | offset += skb->csum_offset; | |
2457 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2458 | ||
2459 | if (skb_cloned(skb) && | |
2460 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2461 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2462 | if (ret) | |
2463 | goto out; | |
2464 | } | |
2465 | ||
a030847e | 2466 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2467 | out_set_summed: |
1da177e4 | 2468 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2469 | out: |
1da177e4 LT |
2470 | return ret; |
2471 | } | |
d1b19dff | 2472 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2473 | |
53d6471c | 2474 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2475 | { |
252e3346 | 2476 | __be16 type = skb->protocol; |
f6a78bfc | 2477 | |
19acc327 PS |
2478 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2479 | if (type == htons(ETH_P_TEB)) { | |
2480 | struct ethhdr *eth; | |
2481 | ||
2482 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2483 | return 0; | |
2484 | ||
2485 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2486 | type = eth->h_proto; | |
2487 | } | |
2488 | ||
d4bcef3f | 2489 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
2490 | } |
2491 | ||
2492 | /** | |
2493 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2494 | * @skb: buffer to segment | |
2495 | * @features: features for the output path (see dev->features) | |
2496 | */ | |
2497 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2498 | netdev_features_t features) | |
2499 | { | |
2500 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2501 | struct packet_offload *ptype; | |
53d6471c VY |
2502 | int vlan_depth = skb->mac_len; |
2503 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2504 | |
2505 | if (unlikely(!type)) | |
2506 | return ERR_PTR(-EINVAL); | |
2507 | ||
53d6471c | 2508 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2509 | |
2510 | rcu_read_lock(); | |
22061d80 | 2511 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2512 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2513 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2514 | break; |
2515 | } | |
2516 | } | |
2517 | rcu_read_unlock(); | |
2518 | ||
98e399f8 | 2519 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2520 | |
f6a78bfc HX |
2521 | return segs; |
2522 | } | |
05e8ef4a PS |
2523 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2524 | ||
2525 | ||
2526 | /* openvswitch calls this on rx path, so we need a different check. | |
2527 | */ | |
2528 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2529 | { | |
2530 | if (tx_path) | |
2531 | return skb->ip_summed != CHECKSUM_PARTIAL; | |
2532 | else | |
2533 | return skb->ip_summed == CHECKSUM_NONE; | |
2534 | } | |
2535 | ||
2536 | /** | |
2537 | * __skb_gso_segment - Perform segmentation on skb. | |
2538 | * @skb: buffer to segment | |
2539 | * @features: features for the output path (see dev->features) | |
2540 | * @tx_path: whether it is called in TX path | |
2541 | * | |
2542 | * This function segments the given skb and returns a list of segments. | |
2543 | * | |
2544 | * It may return NULL if the skb requires no segmentation. This is | |
2545 | * only possible when GSO is used for verifying header integrity. | |
2546 | */ | |
2547 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2548 | netdev_features_t features, bool tx_path) | |
2549 | { | |
2550 | if (unlikely(skb_needs_check(skb, tx_path))) { | |
2551 | int err; | |
2552 | ||
2553 | skb_warn_bad_offload(skb); | |
2554 | ||
a40e0a66 | 2555 | err = skb_cow_head(skb, 0); |
2556 | if (err < 0) | |
05e8ef4a PS |
2557 | return ERR_PTR(err); |
2558 | } | |
2559 | ||
68c33163 | 2560 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2561 | SKB_GSO_CB(skb)->encap_level = 0; |
2562 | ||
05e8ef4a PS |
2563 | skb_reset_mac_header(skb); |
2564 | skb_reset_mac_len(skb); | |
2565 | ||
2566 | return skb_mac_gso_segment(skb, features); | |
2567 | } | |
12b0004d | 2568 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2569 | |
fb286bb2 HX |
2570 | /* Take action when hardware reception checksum errors are detected. */ |
2571 | #ifdef CONFIG_BUG | |
2572 | void netdev_rx_csum_fault(struct net_device *dev) | |
2573 | { | |
2574 | if (net_ratelimit()) { | |
7b6cd1ce | 2575 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2576 | dump_stack(); |
2577 | } | |
2578 | } | |
2579 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2580 | #endif | |
2581 | ||
1da177e4 LT |
2582 | /* Actually, we should eliminate this check as soon as we know, that: |
2583 | * 1. IOMMU is present and allows to map all the memory. | |
2584 | * 2. No high memory really exists on this machine. | |
2585 | */ | |
2586 | ||
c1e756bf | 2587 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2588 | { |
3d3a8533 | 2589 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2590 | int i; |
5acbbd42 | 2591 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2592 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2593 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2594 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2595 | return 1; |
ea2ab693 | 2596 | } |
5acbbd42 | 2597 | } |
1da177e4 | 2598 | |
5acbbd42 FT |
2599 | if (PCI_DMA_BUS_IS_PHYS) { |
2600 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2601 | |
9092c658 ED |
2602 | if (!pdev) |
2603 | return 0; | |
5acbbd42 | 2604 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2605 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2606 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2607 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2608 | return 1; | |
2609 | } | |
2610 | } | |
3d3a8533 | 2611 | #endif |
1da177e4 LT |
2612 | return 0; |
2613 | } | |
1da177e4 | 2614 | |
3b392ddb SH |
2615 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2616 | * instead of standard features for the netdev. | |
2617 | */ | |
d0edc7bf | 2618 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
2619 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
2620 | netdev_features_t features, | |
2621 | __be16 type) | |
2622 | { | |
25cd9ba0 | 2623 | if (eth_p_mpls(type)) |
3b392ddb SH |
2624 | features &= skb->dev->mpls_features; |
2625 | ||
2626 | return features; | |
2627 | } | |
2628 | #else | |
2629 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2630 | netdev_features_t features, | |
2631 | __be16 type) | |
2632 | { | |
2633 | return features; | |
2634 | } | |
2635 | #endif | |
2636 | ||
c8f44aff | 2637 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2638 | netdev_features_t features) |
f01a5236 | 2639 | { |
53d6471c | 2640 | int tmp; |
3b392ddb SH |
2641 | __be16 type; |
2642 | ||
2643 | type = skb_network_protocol(skb, &tmp); | |
2644 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2645 | |
c0d680e5 | 2646 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2647 | !can_checksum_protocol(features, type)) { |
f01a5236 | 2648 | features &= ~NETIF_F_ALL_CSUM; |
c1e756bf | 2649 | } else if (illegal_highdma(skb->dev, skb)) { |
f01a5236 JG |
2650 | features &= ~NETIF_F_SG; |
2651 | } | |
2652 | ||
2653 | return features; | |
2654 | } | |
2655 | ||
e38f3025 TM |
2656 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
2657 | struct net_device *dev, | |
2658 | netdev_features_t features) | |
2659 | { | |
2660 | return features; | |
2661 | } | |
2662 | EXPORT_SYMBOL(passthru_features_check); | |
2663 | ||
8cb65d00 TM |
2664 | static netdev_features_t dflt_features_check(const struct sk_buff *skb, |
2665 | struct net_device *dev, | |
2666 | netdev_features_t features) | |
2667 | { | |
2668 | return vlan_features_check(skb, features); | |
2669 | } | |
2670 | ||
c1e756bf | 2671 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 2672 | { |
5f35227e | 2673 | struct net_device *dev = skb->dev; |
fcbeb976 ED |
2674 | netdev_features_t features = dev->features; |
2675 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
58e998c6 | 2676 | |
fcbeb976 | 2677 | if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs) |
30b678d8 BH |
2678 | features &= ~NETIF_F_GSO_MASK; |
2679 | ||
5f35227e JG |
2680 | /* If encapsulation offload request, verify we are testing |
2681 | * hardware encapsulation features instead of standard | |
2682 | * features for the netdev | |
2683 | */ | |
2684 | if (skb->encapsulation) | |
2685 | features &= dev->hw_enc_features; | |
2686 | ||
f5a7fb88 TM |
2687 | if (skb_vlan_tagged(skb)) |
2688 | features = netdev_intersect_features(features, | |
2689 | dev->vlan_features | | |
2690 | NETIF_F_HW_VLAN_CTAG_TX | | |
2691 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2692 | |
5f35227e JG |
2693 | if (dev->netdev_ops->ndo_features_check) |
2694 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
2695 | features); | |
8cb65d00 TM |
2696 | else |
2697 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 2698 | |
c1e756bf | 2699 | return harmonize_features(skb, features); |
58e998c6 | 2700 | } |
c1e756bf | 2701 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2702 | |
2ea25513 | 2703 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2704 | struct netdev_queue *txq, bool more) |
f6a78bfc | 2705 | { |
2ea25513 DM |
2706 | unsigned int len; |
2707 | int rc; | |
00829823 | 2708 | |
7866a621 | 2709 | if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) |
2ea25513 | 2710 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 2711 | |
2ea25513 DM |
2712 | len = skb->len; |
2713 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 2714 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 2715 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 2716 | |
2ea25513 DM |
2717 | return rc; |
2718 | } | |
7b9c6090 | 2719 | |
8dcda22a DM |
2720 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
2721 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
2722 | { |
2723 | struct sk_buff *skb = first; | |
2724 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 2725 | |
7f2e870f DM |
2726 | while (skb) { |
2727 | struct sk_buff *next = skb->next; | |
fc70fb64 | 2728 | |
7f2e870f | 2729 | skb->next = NULL; |
95f6b3dd | 2730 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
2731 | if (unlikely(!dev_xmit_complete(rc))) { |
2732 | skb->next = next; | |
2733 | goto out; | |
2734 | } | |
6afff0ca | 2735 | |
7f2e870f DM |
2736 | skb = next; |
2737 | if (netif_xmit_stopped(txq) && skb) { | |
2738 | rc = NETDEV_TX_BUSY; | |
2739 | break; | |
9ccb8975 | 2740 | } |
7f2e870f | 2741 | } |
9ccb8975 | 2742 | |
7f2e870f DM |
2743 | out: |
2744 | *ret = rc; | |
2745 | return skb; | |
2746 | } | |
b40863c6 | 2747 | |
1ff0dc94 ED |
2748 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
2749 | netdev_features_t features) | |
f6a78bfc | 2750 | { |
df8a39de | 2751 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
2752 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
2753 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
2754 | return skb; |
2755 | } | |
f6a78bfc | 2756 | |
55a93b3e | 2757 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
2758 | { |
2759 | netdev_features_t features; | |
f6a78bfc | 2760 | |
eae3f88e DM |
2761 | if (skb->next) |
2762 | return skb; | |
068a2de5 | 2763 | |
eae3f88e DM |
2764 | features = netif_skb_features(skb); |
2765 | skb = validate_xmit_vlan(skb, features); | |
2766 | if (unlikely(!skb)) | |
2767 | goto out_null; | |
7b9c6090 | 2768 | |
8b86a61d | 2769 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
2770 | struct sk_buff *segs; |
2771 | ||
2772 | segs = skb_gso_segment(skb, features); | |
cecda693 | 2773 | if (IS_ERR(segs)) { |
af6dabc9 | 2774 | goto out_kfree_skb; |
cecda693 JW |
2775 | } else if (segs) { |
2776 | consume_skb(skb); | |
2777 | skb = segs; | |
f6a78bfc | 2778 | } |
eae3f88e DM |
2779 | } else { |
2780 | if (skb_needs_linearize(skb, features) && | |
2781 | __skb_linearize(skb)) | |
2782 | goto out_kfree_skb; | |
4ec93edb | 2783 | |
eae3f88e DM |
2784 | /* If packet is not checksummed and device does not |
2785 | * support checksumming for this protocol, complete | |
2786 | * checksumming here. | |
2787 | */ | |
2788 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
2789 | if (skb->encapsulation) | |
2790 | skb_set_inner_transport_header(skb, | |
2791 | skb_checksum_start_offset(skb)); | |
2792 | else | |
2793 | skb_set_transport_header(skb, | |
2794 | skb_checksum_start_offset(skb)); | |
2795 | if (!(features & NETIF_F_ALL_CSUM) && | |
2796 | skb_checksum_help(skb)) | |
2797 | goto out_kfree_skb; | |
7b9c6090 | 2798 | } |
0c772159 | 2799 | } |
7b9c6090 | 2800 | |
eae3f88e | 2801 | return skb; |
fc70fb64 | 2802 | |
f6a78bfc HX |
2803 | out_kfree_skb: |
2804 | kfree_skb(skb); | |
eae3f88e DM |
2805 | out_null: |
2806 | return NULL; | |
2807 | } | |
6afff0ca | 2808 | |
55a93b3e ED |
2809 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
2810 | { | |
2811 | struct sk_buff *next, *head = NULL, *tail; | |
2812 | ||
bec3cfdc | 2813 | for (; skb != NULL; skb = next) { |
55a93b3e ED |
2814 | next = skb->next; |
2815 | skb->next = NULL; | |
bec3cfdc ED |
2816 | |
2817 | /* in case skb wont be segmented, point to itself */ | |
2818 | skb->prev = skb; | |
2819 | ||
55a93b3e | 2820 | skb = validate_xmit_skb(skb, dev); |
bec3cfdc ED |
2821 | if (!skb) |
2822 | continue; | |
55a93b3e | 2823 | |
bec3cfdc ED |
2824 | if (!head) |
2825 | head = skb; | |
2826 | else | |
2827 | tail->next = skb; | |
2828 | /* If skb was segmented, skb->prev points to | |
2829 | * the last segment. If not, it still contains skb. | |
2830 | */ | |
2831 | tail = skb->prev; | |
55a93b3e ED |
2832 | } |
2833 | return head; | |
f6a78bfc HX |
2834 | } |
2835 | ||
1def9238 ED |
2836 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
2837 | { | |
2838 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2839 | ||
2840 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2841 | ||
2842 | /* To get more precise estimation of bytes sent on wire, | |
2843 | * we add to pkt_len the headers size of all segments | |
2844 | */ | |
2845 | if (shinfo->gso_size) { | |
757b8b1d | 2846 | unsigned int hdr_len; |
15e5a030 | 2847 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 2848 | |
757b8b1d ED |
2849 | /* mac layer + network layer */ |
2850 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
2851 | ||
2852 | /* + transport layer */ | |
1def9238 ED |
2853 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
2854 | hdr_len += tcp_hdrlen(skb); | |
2855 | else | |
2856 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
2857 | |
2858 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
2859 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
2860 | shinfo->gso_size); | |
2861 | ||
2862 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
2863 | } |
2864 | } | |
2865 | ||
bbd8a0d3 KK |
2866 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2867 | struct net_device *dev, | |
2868 | struct netdev_queue *txq) | |
2869 | { | |
2870 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2871 | bool contended; |
bbd8a0d3 KK |
2872 | int rc; |
2873 | ||
1def9238 | 2874 | qdisc_pkt_len_init(skb); |
a2da570d | 2875 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
2876 | /* |
2877 | * Heuristic to force contended enqueues to serialize on a | |
2878 | * separate lock before trying to get qdisc main lock. | |
9bf2b8c2 YX |
2879 | * This permits __QDISC___STATE_RUNNING owner to get the lock more |
2880 | * often and dequeue packets faster. | |
79640a4c | 2881 | */ |
a2da570d | 2882 | contended = qdisc_is_running(q); |
79640a4c ED |
2883 | if (unlikely(contended)) |
2884 | spin_lock(&q->busylock); | |
2885 | ||
bbd8a0d3 KK |
2886 | spin_lock(root_lock); |
2887 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2888 | kfree_skb(skb); | |
2889 | rc = NET_XMIT_DROP; | |
2890 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2891 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2892 | /* |
2893 | * This is a work-conserving queue; there are no old skbs | |
2894 | * waiting to be sent out; and the qdisc is not running - | |
2895 | * xmit the skb directly. | |
2896 | */ | |
bfe0d029 | 2897 | |
bfe0d029 ED |
2898 | qdisc_bstats_update(q, skb); |
2899 | ||
55a93b3e | 2900 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
2901 | if (unlikely(contended)) { |
2902 | spin_unlock(&q->busylock); | |
2903 | contended = false; | |
2904 | } | |
bbd8a0d3 | 2905 | __qdisc_run(q); |
79640a4c | 2906 | } else |
bc135b23 | 2907 | qdisc_run_end(q); |
bbd8a0d3 KK |
2908 | |
2909 | rc = NET_XMIT_SUCCESS; | |
2910 | } else { | |
a2da570d | 2911 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2912 | if (qdisc_run_begin(q)) { |
2913 | if (unlikely(contended)) { | |
2914 | spin_unlock(&q->busylock); | |
2915 | contended = false; | |
2916 | } | |
2917 | __qdisc_run(q); | |
2918 | } | |
bbd8a0d3 KK |
2919 | } |
2920 | spin_unlock(root_lock); | |
79640a4c ED |
2921 | if (unlikely(contended)) |
2922 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2923 | return rc; |
2924 | } | |
2925 | ||
86f8515f | 2926 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
2927 | static void skb_update_prio(struct sk_buff *skb) |
2928 | { | |
6977a79d | 2929 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 2930 | |
91c68ce2 ED |
2931 | if (!skb->priority && skb->sk && map) { |
2932 | unsigned int prioidx = skb->sk->sk_cgrp_prioidx; | |
2933 | ||
2934 | if (prioidx < map->priomap_len) | |
2935 | skb->priority = map->priomap[prioidx]; | |
2936 | } | |
5bc1421e NH |
2937 | } |
2938 | #else | |
2939 | #define skb_update_prio(skb) | |
2940 | #endif | |
2941 | ||
f60e5990 | 2942 | DEFINE_PER_CPU(int, xmit_recursion); |
2943 | EXPORT_SYMBOL(xmit_recursion); | |
2944 | ||
11a766ce | 2945 | #define RECURSION_LIMIT 10 |
745e20f1 | 2946 | |
95603e22 MM |
2947 | /** |
2948 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
2949 | * @net: network namespace this loopback is happening in |
2950 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
2951 | * @skb: buffer to transmit |
2952 | */ | |
0c4b51f0 | 2953 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
2954 | { |
2955 | skb_reset_mac_header(skb); | |
2956 | __skb_pull(skb, skb_network_offset(skb)); | |
2957 | skb->pkt_type = PACKET_LOOPBACK; | |
2958 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2959 | WARN_ON(!skb_dst(skb)); | |
2960 | skb_dst_force(skb); | |
2961 | netif_rx_ni(skb); | |
2962 | return 0; | |
2963 | } | |
2964 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2965 | ||
638b2a69 JP |
2966 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
2967 | { | |
2968 | #ifdef CONFIG_XPS | |
2969 | struct xps_dev_maps *dev_maps; | |
2970 | struct xps_map *map; | |
2971 | int queue_index = -1; | |
2972 | ||
2973 | rcu_read_lock(); | |
2974 | dev_maps = rcu_dereference(dev->xps_maps); | |
2975 | if (dev_maps) { | |
2976 | map = rcu_dereference( | |
2977 | dev_maps->cpu_map[skb->sender_cpu - 1]); | |
2978 | if (map) { | |
2979 | if (map->len == 1) | |
2980 | queue_index = map->queues[0]; | |
2981 | else | |
2982 | queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), | |
2983 | map->len)]; | |
2984 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
2985 | queue_index = -1; | |
2986 | } | |
2987 | } | |
2988 | rcu_read_unlock(); | |
2989 | ||
2990 | return queue_index; | |
2991 | #else | |
2992 | return -1; | |
2993 | #endif | |
2994 | } | |
2995 | ||
2996 | static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) | |
2997 | { | |
2998 | struct sock *sk = skb->sk; | |
2999 | int queue_index = sk_tx_queue_get(sk); | |
3000 | ||
3001 | if (queue_index < 0 || skb->ooo_okay || | |
3002 | queue_index >= dev->real_num_tx_queues) { | |
3003 | int new_index = get_xps_queue(dev, skb); | |
3004 | if (new_index < 0) | |
3005 | new_index = skb_tx_hash(dev, skb); | |
3006 | ||
3007 | if (queue_index != new_index && sk && | |
004a5d01 | 3008 | sk_fullsock(sk) && |
638b2a69 JP |
3009 | rcu_access_pointer(sk->sk_dst_cache)) |
3010 | sk_tx_queue_set(sk, new_index); | |
3011 | ||
3012 | queue_index = new_index; | |
3013 | } | |
3014 | ||
3015 | return queue_index; | |
3016 | } | |
3017 | ||
3018 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, | |
3019 | struct sk_buff *skb, | |
3020 | void *accel_priv) | |
3021 | { | |
3022 | int queue_index = 0; | |
3023 | ||
3024 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
3025 | u32 sender_cpu = skb->sender_cpu - 1; |
3026 | ||
3027 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
3028 | skb->sender_cpu = raw_smp_processor_id() + 1; |
3029 | #endif | |
3030 | ||
3031 | if (dev->real_num_tx_queues != 1) { | |
3032 | const struct net_device_ops *ops = dev->netdev_ops; | |
3033 | if (ops->ndo_select_queue) | |
3034 | queue_index = ops->ndo_select_queue(dev, skb, accel_priv, | |
3035 | __netdev_pick_tx); | |
3036 | else | |
3037 | queue_index = __netdev_pick_tx(dev, skb); | |
3038 | ||
3039 | if (!accel_priv) | |
3040 | queue_index = netdev_cap_txqueue(dev, queue_index); | |
3041 | } | |
3042 | ||
3043 | skb_set_queue_mapping(skb, queue_index); | |
3044 | return netdev_get_tx_queue(dev, queue_index); | |
3045 | } | |
3046 | ||
d29f749e | 3047 | /** |
9d08dd3d | 3048 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3049 | * @skb: buffer to transmit |
9d08dd3d | 3050 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
3051 | * |
3052 | * Queue a buffer for transmission to a network device. The caller must | |
3053 | * have set the device and priority and built the buffer before calling | |
3054 | * this function. The function can be called from an interrupt. | |
3055 | * | |
3056 | * A negative errno code is returned on a failure. A success does not | |
3057 | * guarantee the frame will be transmitted as it may be dropped due | |
3058 | * to congestion or traffic shaping. | |
3059 | * | |
3060 | * ----------------------------------------------------------------------------------- | |
3061 | * I notice this method can also return errors from the queue disciplines, | |
3062 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3063 | * be positive. | |
3064 | * | |
3065 | * Regardless of the return value, the skb is consumed, so it is currently | |
3066 | * difficult to retry a send to this method. (You can bump the ref count | |
3067 | * before sending to hold a reference for retry if you are careful.) | |
3068 | * | |
3069 | * When calling this method, interrupts MUST be enabled. This is because | |
3070 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3071 | * --BLG | |
3072 | */ | |
0a59f3a9 | 3073 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
3074 | { |
3075 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3076 | struct netdev_queue *txq; |
1da177e4 LT |
3077 | struct Qdisc *q; |
3078 | int rc = -ENOMEM; | |
3079 | ||
6d1ccff6 ED |
3080 | skb_reset_mac_header(skb); |
3081 | ||
e7fd2885 WB |
3082 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3083 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3084 | ||
4ec93edb YH |
3085 | /* Disable soft irqs for various locks below. Also |
3086 | * stops preemption for RCU. | |
1da177e4 | 3087 | */ |
4ec93edb | 3088 | rcu_read_lock_bh(); |
1da177e4 | 3089 | |
5bc1421e NH |
3090 | skb_update_prio(skb); |
3091 | ||
02875878 ED |
3092 | /* If device/qdisc don't need skb->dst, release it right now while |
3093 | * its hot in this cpu cache. | |
3094 | */ | |
3095 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
3096 | skb_dst_drop(skb); | |
3097 | else | |
3098 | skb_dst_force(skb); | |
3099 | ||
0c4f691f SF |
3100 | #ifdef CONFIG_NET_SWITCHDEV |
3101 | /* Don't forward if offload device already forwarded */ | |
3102 | if (skb->offload_fwd_mark && | |
3103 | skb->offload_fwd_mark == dev->offload_fwd_mark) { | |
3104 | consume_skb(skb); | |
3105 | rc = NET_XMIT_SUCCESS; | |
3106 | goto out; | |
3107 | } | |
3108 | #endif | |
3109 | ||
f663dd9a | 3110 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 3111 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 3112 | |
1da177e4 | 3113 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 3114 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 3115 | #endif |
cf66ba58 | 3116 | trace_net_dev_queue(skb); |
1da177e4 | 3117 | if (q->enqueue) { |
bbd8a0d3 | 3118 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 3119 | goto out; |
1da177e4 LT |
3120 | } |
3121 | ||
3122 | /* The device has no queue. Common case for software devices: | |
3123 | loopback, all the sorts of tunnels... | |
3124 | ||
932ff279 HX |
3125 | Really, it is unlikely that netif_tx_lock protection is necessary |
3126 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
3127 | counters.) |
3128 | However, it is possible, that they rely on protection | |
3129 | made by us here. | |
3130 | ||
3131 | Check this and shot the lock. It is not prone from deadlocks. | |
3132 | Either shot noqueue qdisc, it is even simpler 8) | |
3133 | */ | |
3134 | if (dev->flags & IFF_UP) { | |
3135 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
3136 | ||
c773e847 | 3137 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 3138 | |
745e20f1 ED |
3139 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
3140 | goto recursion_alert; | |
3141 | ||
1f59533f JDB |
3142 | skb = validate_xmit_skb(skb, dev); |
3143 | if (!skb) | |
3144 | goto drop; | |
3145 | ||
c773e847 | 3146 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 3147 | |
73466498 | 3148 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 3149 | __this_cpu_inc(xmit_recursion); |
ce93718f | 3150 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 3151 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 3152 | if (dev_xmit_complete(rc)) { |
c773e847 | 3153 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
3154 | goto out; |
3155 | } | |
3156 | } | |
c773e847 | 3157 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
3158 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
3159 | dev->name); | |
1da177e4 LT |
3160 | } else { |
3161 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
3162 | * unfortunately |
3163 | */ | |
3164 | recursion_alert: | |
e87cc472 JP |
3165 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
3166 | dev->name); | |
1da177e4 LT |
3167 | } |
3168 | } | |
3169 | ||
3170 | rc = -ENETDOWN; | |
1f59533f | 3171 | drop: |
d4828d85 | 3172 | rcu_read_unlock_bh(); |
1da177e4 | 3173 | |
015f0688 | 3174 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 3175 | kfree_skb_list(skb); |
1da177e4 LT |
3176 | return rc; |
3177 | out: | |
d4828d85 | 3178 | rcu_read_unlock_bh(); |
1da177e4 LT |
3179 | return rc; |
3180 | } | |
f663dd9a | 3181 | |
2b4aa3ce | 3182 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
3183 | { |
3184 | return __dev_queue_xmit(skb, NULL); | |
3185 | } | |
2b4aa3ce | 3186 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3187 | |
f663dd9a JW |
3188 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3189 | { | |
3190 | return __dev_queue_xmit(skb, accel_priv); | |
3191 | } | |
3192 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3193 | ||
1da177e4 LT |
3194 | |
3195 | /*======================================================================= | |
3196 | Receiver routines | |
3197 | =======================================================================*/ | |
3198 | ||
6b2bedc3 | 3199 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3200 | EXPORT_SYMBOL(netdev_max_backlog); |
3201 | ||
3b098e2d | 3202 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
3203 | int netdev_budget __read_mostly = 300; |
3204 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 3205 | |
eecfd7c4 ED |
3206 | /* Called with irq disabled */ |
3207 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3208 | struct napi_struct *napi) | |
3209 | { | |
3210 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3211 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3212 | } | |
3213 | ||
bfb564e7 KK |
3214 | #ifdef CONFIG_RPS |
3215 | ||
3216 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3217 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 3218 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
3219 | u32 rps_cpu_mask __read_mostly; |
3220 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 3221 | |
c5905afb | 3222 | struct static_key rps_needed __read_mostly; |
adc9300e | 3223 | |
c445477d BH |
3224 | static struct rps_dev_flow * |
3225 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3226 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3227 | { | |
a31196b0 | 3228 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
3229 | #ifdef CONFIG_RFS_ACCEL |
3230 | struct netdev_rx_queue *rxqueue; | |
3231 | struct rps_dev_flow_table *flow_table; | |
3232 | struct rps_dev_flow *old_rflow; | |
3233 | u32 flow_id; | |
3234 | u16 rxq_index; | |
3235 | int rc; | |
3236 | ||
3237 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3238 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3239 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3240 | goto out; |
3241 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3242 | if (rxq_index == skb_get_rx_queue(skb)) | |
3243 | goto out; | |
3244 | ||
3245 | rxqueue = dev->_rx + rxq_index; | |
3246 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3247 | if (!flow_table) | |
3248 | goto out; | |
61b905da | 3249 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3250 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3251 | rxq_index, flow_id); | |
3252 | if (rc < 0) | |
3253 | goto out; | |
3254 | old_rflow = rflow; | |
3255 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3256 | rflow->filter = rc; |
3257 | if (old_rflow->filter == rflow->filter) | |
3258 | old_rflow->filter = RPS_NO_FILTER; | |
3259 | out: | |
3260 | #endif | |
3261 | rflow->last_qtail = | |
09994d1b | 3262 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3263 | } |
3264 | ||
09994d1b | 3265 | rflow->cpu = next_cpu; |
c445477d BH |
3266 | return rflow; |
3267 | } | |
3268 | ||
bfb564e7 KK |
3269 | /* |
3270 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3271 | * CPU from the RPS map of the receiving queue for a given skb. | |
3272 | * rcu_read_lock must be held on entry. | |
3273 | */ | |
3274 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3275 | struct rps_dev_flow **rflowp) | |
3276 | { | |
567e4b79 ED |
3277 | const struct rps_sock_flow_table *sock_flow_table; |
3278 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 3279 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 3280 | struct rps_map *map; |
bfb564e7 | 3281 | int cpu = -1; |
567e4b79 | 3282 | u32 tcpu; |
61b905da | 3283 | u32 hash; |
bfb564e7 KK |
3284 | |
3285 | if (skb_rx_queue_recorded(skb)) { | |
3286 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 3287 | |
62fe0b40 BH |
3288 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3289 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3290 | "%s received packet on queue %u, but number " | |
3291 | "of RX queues is %u\n", | |
3292 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3293 | goto done; |
3294 | } | |
567e4b79 ED |
3295 | rxqueue += index; |
3296 | } | |
bfb564e7 | 3297 | |
567e4b79 ED |
3298 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
3299 | ||
3300 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 3301 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 3302 | if (!flow_table && !map) |
bfb564e7 KK |
3303 | goto done; |
3304 | ||
2d47b459 | 3305 | skb_reset_network_header(skb); |
61b905da TH |
3306 | hash = skb_get_hash(skb); |
3307 | if (!hash) | |
bfb564e7 KK |
3308 | goto done; |
3309 | ||
fec5e652 TH |
3310 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
3311 | if (flow_table && sock_flow_table) { | |
fec5e652 | 3312 | struct rps_dev_flow *rflow; |
567e4b79 ED |
3313 | u32 next_cpu; |
3314 | u32 ident; | |
3315 | ||
3316 | /* First check into global flow table if there is a match */ | |
3317 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
3318 | if ((ident ^ hash) & ~rps_cpu_mask) | |
3319 | goto try_rps; | |
fec5e652 | 3320 | |
567e4b79 ED |
3321 | next_cpu = ident & rps_cpu_mask; |
3322 | ||
3323 | /* OK, now we know there is a match, | |
3324 | * we can look at the local (per receive queue) flow table | |
3325 | */ | |
61b905da | 3326 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3327 | tcpu = rflow->cpu; |
3328 | ||
fec5e652 TH |
3329 | /* |
3330 | * If the desired CPU (where last recvmsg was done) is | |
3331 | * different from current CPU (one in the rx-queue flow | |
3332 | * table entry), switch if one of the following holds: | |
a31196b0 | 3333 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
3334 | * - Current CPU is offline. |
3335 | * - The current CPU's queue tail has advanced beyond the | |
3336 | * last packet that was enqueued using this table entry. | |
3337 | * This guarantees that all previous packets for the flow | |
3338 | * have been dequeued, thus preserving in order delivery. | |
3339 | */ | |
3340 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 3341 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 3342 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
3343 | rflow->last_qtail)) >= 0)) { |
3344 | tcpu = next_cpu; | |
c445477d | 3345 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3346 | } |
c445477d | 3347 | |
a31196b0 | 3348 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
3349 | *rflowp = rflow; |
3350 | cpu = tcpu; | |
3351 | goto done; | |
3352 | } | |
3353 | } | |
3354 | ||
567e4b79 ED |
3355 | try_rps: |
3356 | ||
0a9627f2 | 3357 | if (map) { |
8fc54f68 | 3358 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3359 | if (cpu_online(tcpu)) { |
3360 | cpu = tcpu; | |
3361 | goto done; | |
3362 | } | |
3363 | } | |
3364 | ||
3365 | done: | |
0a9627f2 TH |
3366 | return cpu; |
3367 | } | |
3368 | ||
c445477d BH |
3369 | #ifdef CONFIG_RFS_ACCEL |
3370 | ||
3371 | /** | |
3372 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3373 | * @dev: Device on which the filter was set | |
3374 | * @rxq_index: RX queue index | |
3375 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3376 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3377 | * | |
3378 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3379 | * this function for each installed filter and remove the filters for | |
3380 | * which it returns %true. | |
3381 | */ | |
3382 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3383 | u32 flow_id, u16 filter_id) | |
3384 | { | |
3385 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3386 | struct rps_dev_flow_table *flow_table; | |
3387 | struct rps_dev_flow *rflow; | |
3388 | bool expire = true; | |
a31196b0 | 3389 | unsigned int cpu; |
c445477d BH |
3390 | |
3391 | rcu_read_lock(); | |
3392 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3393 | if (flow_table && flow_id <= flow_table->mask) { | |
3394 | rflow = &flow_table->flows[flow_id]; | |
3395 | cpu = ACCESS_ONCE(rflow->cpu); | |
a31196b0 | 3396 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
3397 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
3398 | rflow->last_qtail) < | |
3399 | (int)(10 * flow_table->mask))) | |
3400 | expire = false; | |
3401 | } | |
3402 | rcu_read_unlock(); | |
3403 | return expire; | |
3404 | } | |
3405 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3406 | ||
3407 | #endif /* CONFIG_RFS_ACCEL */ | |
3408 | ||
0a9627f2 | 3409 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3410 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3411 | { |
e36fa2f7 ED |
3412 | struct softnet_data *sd = data; |
3413 | ||
eecfd7c4 | 3414 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3415 | sd->received_rps++; |
0a9627f2 | 3416 | } |
e36fa2f7 | 3417 | |
fec5e652 | 3418 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3419 | |
e36fa2f7 ED |
3420 | /* |
3421 | * Check if this softnet_data structure is another cpu one | |
3422 | * If yes, queue it to our IPI list and return 1 | |
3423 | * If no, return 0 | |
3424 | */ | |
3425 | static int rps_ipi_queued(struct softnet_data *sd) | |
3426 | { | |
3427 | #ifdef CONFIG_RPS | |
903ceff7 | 3428 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
3429 | |
3430 | if (sd != mysd) { | |
3431 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3432 | mysd->rps_ipi_list = sd; | |
3433 | ||
3434 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3435 | return 1; | |
3436 | } | |
3437 | #endif /* CONFIG_RPS */ | |
3438 | return 0; | |
3439 | } | |
3440 | ||
99bbc707 WB |
3441 | #ifdef CONFIG_NET_FLOW_LIMIT |
3442 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3443 | #endif | |
3444 | ||
3445 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3446 | { | |
3447 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3448 | struct sd_flow_limit *fl; | |
3449 | struct softnet_data *sd; | |
3450 | unsigned int old_flow, new_flow; | |
3451 | ||
3452 | if (qlen < (netdev_max_backlog >> 1)) | |
3453 | return false; | |
3454 | ||
903ceff7 | 3455 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
3456 | |
3457 | rcu_read_lock(); | |
3458 | fl = rcu_dereference(sd->flow_limit); | |
3459 | if (fl) { | |
3958afa1 | 3460 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3461 | old_flow = fl->history[fl->history_head]; |
3462 | fl->history[fl->history_head] = new_flow; | |
3463 | ||
3464 | fl->history_head++; | |
3465 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3466 | ||
3467 | if (likely(fl->buckets[old_flow])) | |
3468 | fl->buckets[old_flow]--; | |
3469 | ||
3470 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3471 | fl->count++; | |
3472 | rcu_read_unlock(); | |
3473 | return true; | |
3474 | } | |
3475 | } | |
3476 | rcu_read_unlock(); | |
3477 | #endif | |
3478 | return false; | |
3479 | } | |
3480 | ||
0a9627f2 TH |
3481 | /* |
3482 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3483 | * queue (may be a remote CPU queue). | |
3484 | */ | |
fec5e652 TH |
3485 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3486 | unsigned int *qtail) | |
0a9627f2 | 3487 | { |
e36fa2f7 | 3488 | struct softnet_data *sd; |
0a9627f2 | 3489 | unsigned long flags; |
99bbc707 | 3490 | unsigned int qlen; |
0a9627f2 | 3491 | |
e36fa2f7 | 3492 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3493 | |
3494 | local_irq_save(flags); | |
0a9627f2 | 3495 | |
e36fa2f7 | 3496 | rps_lock(sd); |
e9e4dd32 JA |
3497 | if (!netif_running(skb->dev)) |
3498 | goto drop; | |
99bbc707 WB |
3499 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3500 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 3501 | if (qlen) { |
0a9627f2 | 3502 | enqueue: |
e36fa2f7 | 3503 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3504 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3505 | rps_unlock(sd); |
152102c7 | 3506 | local_irq_restore(flags); |
0a9627f2 TH |
3507 | return NET_RX_SUCCESS; |
3508 | } | |
3509 | ||
ebda37c2 ED |
3510 | /* Schedule NAPI for backlog device |
3511 | * We can use non atomic operation since we own the queue lock | |
3512 | */ | |
3513 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3514 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3515 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3516 | } |
3517 | goto enqueue; | |
3518 | } | |
3519 | ||
e9e4dd32 | 3520 | drop: |
dee42870 | 3521 | sd->dropped++; |
e36fa2f7 | 3522 | rps_unlock(sd); |
0a9627f2 | 3523 | |
0a9627f2 TH |
3524 | local_irq_restore(flags); |
3525 | ||
caf586e5 | 3526 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3527 | kfree_skb(skb); |
3528 | return NET_RX_DROP; | |
3529 | } | |
1da177e4 | 3530 | |
ae78dbfa | 3531 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3532 | { |
b0e28f1e | 3533 | int ret; |
1da177e4 | 3534 | |
588f0330 | 3535 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3536 | |
cf66ba58 | 3537 | trace_netif_rx(skb); |
df334545 | 3538 | #ifdef CONFIG_RPS |
c5905afb | 3539 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3540 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3541 | int cpu; |
3542 | ||
cece1945 | 3543 | preempt_disable(); |
b0e28f1e | 3544 | rcu_read_lock(); |
fec5e652 TH |
3545 | |
3546 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3547 | if (cpu < 0) |
3548 | cpu = smp_processor_id(); | |
fec5e652 TH |
3549 | |
3550 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3551 | ||
b0e28f1e | 3552 | rcu_read_unlock(); |
cece1945 | 3553 | preempt_enable(); |
adc9300e ED |
3554 | } else |
3555 | #endif | |
fec5e652 TH |
3556 | { |
3557 | unsigned int qtail; | |
3558 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3559 | put_cpu(); | |
3560 | } | |
b0e28f1e | 3561 | return ret; |
1da177e4 | 3562 | } |
ae78dbfa BH |
3563 | |
3564 | /** | |
3565 | * netif_rx - post buffer to the network code | |
3566 | * @skb: buffer to post | |
3567 | * | |
3568 | * This function receives a packet from a device driver and queues it for | |
3569 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3570 | * may be dropped during processing for congestion control or by the | |
3571 | * protocol layers. | |
3572 | * | |
3573 | * return values: | |
3574 | * NET_RX_SUCCESS (no congestion) | |
3575 | * NET_RX_DROP (packet was dropped) | |
3576 | * | |
3577 | */ | |
3578 | ||
3579 | int netif_rx(struct sk_buff *skb) | |
3580 | { | |
3581 | trace_netif_rx_entry(skb); | |
3582 | ||
3583 | return netif_rx_internal(skb); | |
3584 | } | |
d1b19dff | 3585 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3586 | |
3587 | int netif_rx_ni(struct sk_buff *skb) | |
3588 | { | |
3589 | int err; | |
3590 | ||
ae78dbfa BH |
3591 | trace_netif_rx_ni_entry(skb); |
3592 | ||
1da177e4 | 3593 | preempt_disable(); |
ae78dbfa | 3594 | err = netif_rx_internal(skb); |
1da177e4 LT |
3595 | if (local_softirq_pending()) |
3596 | do_softirq(); | |
3597 | preempt_enable(); | |
3598 | ||
3599 | return err; | |
3600 | } | |
1da177e4 LT |
3601 | EXPORT_SYMBOL(netif_rx_ni); |
3602 | ||
1da177e4 LT |
3603 | static void net_tx_action(struct softirq_action *h) |
3604 | { | |
903ceff7 | 3605 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
3606 | |
3607 | if (sd->completion_queue) { | |
3608 | struct sk_buff *clist; | |
3609 | ||
3610 | local_irq_disable(); | |
3611 | clist = sd->completion_queue; | |
3612 | sd->completion_queue = NULL; | |
3613 | local_irq_enable(); | |
3614 | ||
3615 | while (clist) { | |
3616 | struct sk_buff *skb = clist; | |
3617 | clist = clist->next; | |
3618 | ||
547b792c | 3619 | WARN_ON(atomic_read(&skb->users)); |
e6247027 ED |
3620 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
3621 | trace_consume_skb(skb); | |
3622 | else | |
3623 | trace_kfree_skb(skb, net_tx_action); | |
1da177e4 LT |
3624 | __kfree_skb(skb); |
3625 | } | |
3626 | } | |
3627 | ||
3628 | if (sd->output_queue) { | |
37437bb2 | 3629 | struct Qdisc *head; |
1da177e4 LT |
3630 | |
3631 | local_irq_disable(); | |
3632 | head = sd->output_queue; | |
3633 | sd->output_queue = NULL; | |
a9cbd588 | 3634 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3635 | local_irq_enable(); |
3636 | ||
3637 | while (head) { | |
37437bb2 DM |
3638 | struct Qdisc *q = head; |
3639 | spinlock_t *root_lock; | |
3640 | ||
1da177e4 LT |
3641 | head = head->next_sched; |
3642 | ||
5fb66229 | 3643 | root_lock = qdisc_lock(q); |
37437bb2 | 3644 | if (spin_trylock(root_lock)) { |
4e857c58 | 3645 | smp_mb__before_atomic(); |
def82a1d JP |
3646 | clear_bit(__QDISC_STATE_SCHED, |
3647 | &q->state); | |
37437bb2 DM |
3648 | qdisc_run(q); |
3649 | spin_unlock(root_lock); | |
1da177e4 | 3650 | } else { |
195648bb | 3651 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3652 | &q->state)) { |
195648bb | 3653 | __netif_reschedule(q); |
e8a83e10 | 3654 | } else { |
4e857c58 | 3655 | smp_mb__before_atomic(); |
e8a83e10 JP |
3656 | clear_bit(__QDISC_STATE_SCHED, |
3657 | &q->state); | |
3658 | } | |
1da177e4 LT |
3659 | } |
3660 | } | |
3661 | } | |
3662 | } | |
3663 | ||
ab95bfe0 JP |
3664 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3665 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3666 | /* This hook is defined here for ATM LANE */ |
3667 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3668 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3669 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3670 | #endif |
1da177e4 | 3671 | |
f697c3e8 HX |
3672 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3673 | struct packet_type **pt_prev, | |
3674 | int *ret, struct net_device *orig_dev) | |
3675 | { | |
e7582bab | 3676 | #ifdef CONFIG_NET_CLS_ACT |
d2788d34 DB |
3677 | struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list); |
3678 | struct tcf_result cl_res; | |
24824a09 | 3679 | |
c9e99fd0 DB |
3680 | /* If there's at least one ingress present somewhere (so |
3681 | * we get here via enabled static key), remaining devices | |
3682 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 3683 | * out here. |
c9e99fd0 | 3684 | */ |
d2788d34 | 3685 | if (!cl) |
4577139b | 3686 | return skb; |
f697c3e8 HX |
3687 | if (*pt_prev) { |
3688 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3689 | *pt_prev = NULL; | |
1da177e4 LT |
3690 | } |
3691 | ||
3365495c | 3692 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
c9e99fd0 | 3693 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); |
24ea591d | 3694 | qdisc_bstats_cpu_update(cl->q, skb); |
c9e99fd0 | 3695 | |
3b3ae880 | 3696 | switch (tc_classify(skb, cl, &cl_res, false)) { |
d2788d34 DB |
3697 | case TC_ACT_OK: |
3698 | case TC_ACT_RECLASSIFY: | |
3699 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3700 | break; | |
3701 | case TC_ACT_SHOT: | |
24ea591d | 3702 | qdisc_qstats_cpu_drop(cl->q); |
d2788d34 DB |
3703 | case TC_ACT_STOLEN: |
3704 | case TC_ACT_QUEUED: | |
3705 | kfree_skb(skb); | |
3706 | return NULL; | |
27b29f63 AS |
3707 | case TC_ACT_REDIRECT: |
3708 | /* skb_mac_header check was done by cls/act_bpf, so | |
3709 | * we can safely push the L2 header back before | |
3710 | * redirecting to another netdev | |
3711 | */ | |
3712 | __skb_push(skb, skb->mac_len); | |
3713 | skb_do_redirect(skb); | |
3714 | return NULL; | |
d2788d34 DB |
3715 | default: |
3716 | break; | |
f697c3e8 | 3717 | } |
e7582bab | 3718 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
3719 | return skb; |
3720 | } | |
1da177e4 | 3721 | |
ab95bfe0 JP |
3722 | /** |
3723 | * netdev_rx_handler_register - register receive handler | |
3724 | * @dev: device to register a handler for | |
3725 | * @rx_handler: receive handler to register | |
93e2c32b | 3726 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 3727 | * |
e227867f | 3728 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
3729 | * called from __netif_receive_skb. A negative errno code is returned |
3730 | * on a failure. | |
3731 | * | |
3732 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3733 | * |
3734 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3735 | */ |
3736 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3737 | rx_handler_func_t *rx_handler, |
3738 | void *rx_handler_data) | |
ab95bfe0 JP |
3739 | { |
3740 | ASSERT_RTNL(); | |
3741 | ||
3742 | if (dev->rx_handler) | |
3743 | return -EBUSY; | |
3744 | ||
00cfec37 | 3745 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 3746 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3747 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3748 | ||
3749 | return 0; | |
3750 | } | |
3751 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3752 | ||
3753 | /** | |
3754 | * netdev_rx_handler_unregister - unregister receive handler | |
3755 | * @dev: device to unregister a handler from | |
3756 | * | |
166ec369 | 3757 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
3758 | * |
3759 | * The caller must hold the rtnl_mutex. | |
3760 | */ | |
3761 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3762 | { | |
3763 | ||
3764 | ASSERT_RTNL(); | |
a9b3cd7f | 3765 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
3766 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
3767 | * section has a guarantee to see a non NULL rx_handler_data | |
3768 | * as well. | |
3769 | */ | |
3770 | synchronize_net(); | |
a9b3cd7f | 3771 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
3772 | } |
3773 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3774 | ||
b4b9e355 MG |
3775 | /* |
3776 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3777 | * the special handling of PFMEMALLOC skbs. | |
3778 | */ | |
3779 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3780 | { | |
3781 | switch (skb->protocol) { | |
2b8837ae JP |
3782 | case htons(ETH_P_ARP): |
3783 | case htons(ETH_P_IP): | |
3784 | case htons(ETH_P_IPV6): | |
3785 | case htons(ETH_P_8021Q): | |
3786 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
3787 | return true; |
3788 | default: | |
3789 | return false; | |
3790 | } | |
3791 | } | |
3792 | ||
e687ad60 PN |
3793 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
3794 | int *ret, struct net_device *orig_dev) | |
3795 | { | |
e7582bab | 3796 | #ifdef CONFIG_NETFILTER_INGRESS |
e687ad60 PN |
3797 | if (nf_hook_ingress_active(skb)) { |
3798 | if (*pt_prev) { | |
3799 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3800 | *pt_prev = NULL; | |
3801 | } | |
3802 | ||
3803 | return nf_hook_ingress(skb); | |
3804 | } | |
e7582bab | 3805 | #endif /* CONFIG_NETFILTER_INGRESS */ |
e687ad60 PN |
3806 | return 0; |
3807 | } | |
e687ad60 | 3808 | |
9754e293 | 3809 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
3810 | { |
3811 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3812 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3813 | struct net_device *orig_dev; |
8a4eb573 | 3814 | bool deliver_exact = false; |
1da177e4 | 3815 | int ret = NET_RX_DROP; |
252e3346 | 3816 | __be16 type; |
1da177e4 | 3817 | |
588f0330 | 3818 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3819 | |
cf66ba58 | 3820 | trace_netif_receive_skb(skb); |
9b22ea56 | 3821 | |
cc9bd5ce | 3822 | orig_dev = skb->dev; |
8f903c70 | 3823 | |
c1d2bbe1 | 3824 | skb_reset_network_header(skb); |
fda55eca ED |
3825 | if (!skb_transport_header_was_set(skb)) |
3826 | skb_reset_transport_header(skb); | |
0b5c9db1 | 3827 | skb_reset_mac_len(skb); |
1da177e4 LT |
3828 | |
3829 | pt_prev = NULL; | |
3830 | ||
63d8ea7f | 3831 | another_round: |
b6858177 | 3832 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
3833 | |
3834 | __this_cpu_inc(softnet_data.processed); | |
3835 | ||
8ad227ff PM |
3836 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
3837 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 3838 | skb = skb_vlan_untag(skb); |
bcc6d479 | 3839 | if (unlikely(!skb)) |
2c17d27c | 3840 | goto out; |
bcc6d479 JP |
3841 | } |
3842 | ||
1da177e4 LT |
3843 | #ifdef CONFIG_NET_CLS_ACT |
3844 | if (skb->tc_verd & TC_NCLS) { | |
3845 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3846 | goto ncls; | |
3847 | } | |
3848 | #endif | |
3849 | ||
9754e293 | 3850 | if (pfmemalloc) |
b4b9e355 MG |
3851 | goto skip_taps; |
3852 | ||
1da177e4 | 3853 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
3854 | if (pt_prev) |
3855 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3856 | pt_prev = ptype; | |
3857 | } | |
3858 | ||
3859 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
3860 | if (pt_prev) | |
3861 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3862 | pt_prev = ptype; | |
1da177e4 LT |
3863 | } |
3864 | ||
b4b9e355 | 3865 | skip_taps: |
1cf51900 | 3866 | #ifdef CONFIG_NET_INGRESS |
4577139b DB |
3867 | if (static_key_false(&ingress_needed)) { |
3868 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); | |
3869 | if (!skb) | |
2c17d27c | 3870 | goto out; |
e687ad60 PN |
3871 | |
3872 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 3873 | goto out; |
4577139b | 3874 | } |
1cf51900 PN |
3875 | #endif |
3876 | #ifdef CONFIG_NET_CLS_ACT | |
4577139b | 3877 | skb->tc_verd = 0; |
1da177e4 LT |
3878 | ncls: |
3879 | #endif | |
9754e293 | 3880 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
3881 | goto drop; |
3882 | ||
df8a39de | 3883 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
3884 | if (pt_prev) { |
3885 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3886 | pt_prev = NULL; | |
3887 | } | |
48cc32d3 | 3888 | if (vlan_do_receive(&skb)) |
2425717b JF |
3889 | goto another_round; |
3890 | else if (unlikely(!skb)) | |
2c17d27c | 3891 | goto out; |
2425717b JF |
3892 | } |
3893 | ||
48cc32d3 | 3894 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
3895 | if (rx_handler) { |
3896 | if (pt_prev) { | |
3897 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3898 | pt_prev = NULL; | |
3899 | } | |
8a4eb573 JP |
3900 | switch (rx_handler(&skb)) { |
3901 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 3902 | ret = NET_RX_SUCCESS; |
2c17d27c | 3903 | goto out; |
8a4eb573 | 3904 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3905 | goto another_round; |
8a4eb573 JP |
3906 | case RX_HANDLER_EXACT: |
3907 | deliver_exact = true; | |
3908 | case RX_HANDLER_PASS: | |
3909 | break; | |
3910 | default: | |
3911 | BUG(); | |
3912 | } | |
ab95bfe0 | 3913 | } |
1da177e4 | 3914 | |
df8a39de JP |
3915 | if (unlikely(skb_vlan_tag_present(skb))) { |
3916 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
3917 | skb->pkt_type = PACKET_OTHERHOST; |
3918 | /* Note: we might in the future use prio bits | |
3919 | * and set skb->priority like in vlan_do_receive() | |
3920 | * For the time being, just ignore Priority Code Point | |
3921 | */ | |
3922 | skb->vlan_tci = 0; | |
3923 | } | |
48cc32d3 | 3924 | |
7866a621 SN |
3925 | type = skb->protocol; |
3926 | ||
63d8ea7f | 3927 | /* deliver only exact match when indicated */ |
7866a621 SN |
3928 | if (likely(!deliver_exact)) { |
3929 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
3930 | &ptype_base[ntohs(type) & | |
3931 | PTYPE_HASH_MASK]); | |
3932 | } | |
1f3c8804 | 3933 | |
7866a621 SN |
3934 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
3935 | &orig_dev->ptype_specific); | |
3936 | ||
3937 | if (unlikely(skb->dev != orig_dev)) { | |
3938 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
3939 | &skb->dev->ptype_specific); | |
1da177e4 LT |
3940 | } |
3941 | ||
3942 | if (pt_prev) { | |
1080e512 | 3943 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 3944 | goto drop; |
1080e512 MT |
3945 | else |
3946 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 3947 | } else { |
b4b9e355 | 3948 | drop: |
caf586e5 | 3949 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3950 | kfree_skb(skb); |
3951 | /* Jamal, now you will not able to escape explaining | |
3952 | * me how you were going to use this. :-) | |
3953 | */ | |
3954 | ret = NET_RX_DROP; | |
3955 | } | |
3956 | ||
2c17d27c | 3957 | out: |
9754e293 DM |
3958 | return ret; |
3959 | } | |
3960 | ||
3961 | static int __netif_receive_skb(struct sk_buff *skb) | |
3962 | { | |
3963 | int ret; | |
3964 | ||
3965 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
3966 | unsigned long pflags = current->flags; | |
3967 | ||
3968 | /* | |
3969 | * PFMEMALLOC skbs are special, they should | |
3970 | * - be delivered to SOCK_MEMALLOC sockets only | |
3971 | * - stay away from userspace | |
3972 | * - have bounded memory usage | |
3973 | * | |
3974 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
3975 | * context down to all allocation sites. | |
3976 | */ | |
3977 | current->flags |= PF_MEMALLOC; | |
3978 | ret = __netif_receive_skb_core(skb, true); | |
3979 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
3980 | } else | |
3981 | ret = __netif_receive_skb_core(skb, false); | |
3982 | ||
1da177e4 LT |
3983 | return ret; |
3984 | } | |
0a9627f2 | 3985 | |
ae78dbfa | 3986 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 3987 | { |
2c17d27c JA |
3988 | int ret; |
3989 | ||
588f0330 | 3990 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3991 | |
c1f19b51 RC |
3992 | if (skb_defer_rx_timestamp(skb)) |
3993 | return NET_RX_SUCCESS; | |
3994 | ||
2c17d27c JA |
3995 | rcu_read_lock(); |
3996 | ||
df334545 | 3997 | #ifdef CONFIG_RPS |
c5905afb | 3998 | if (static_key_false(&rps_needed)) { |
3b098e2d | 3999 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 4000 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 4001 | |
3b098e2d ED |
4002 | if (cpu >= 0) { |
4003 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4004 | rcu_read_unlock(); | |
adc9300e | 4005 | return ret; |
3b098e2d | 4006 | } |
fec5e652 | 4007 | } |
1e94d72f | 4008 | #endif |
2c17d27c JA |
4009 | ret = __netif_receive_skb(skb); |
4010 | rcu_read_unlock(); | |
4011 | return ret; | |
0a9627f2 | 4012 | } |
ae78dbfa BH |
4013 | |
4014 | /** | |
4015 | * netif_receive_skb - process receive buffer from network | |
4016 | * @skb: buffer to process | |
4017 | * | |
4018 | * netif_receive_skb() is the main receive data processing function. | |
4019 | * It always succeeds. The buffer may be dropped during processing | |
4020 | * for congestion control or by the protocol layers. | |
4021 | * | |
4022 | * This function may only be called from softirq context and interrupts | |
4023 | * should be enabled. | |
4024 | * | |
4025 | * Return values (usually ignored): | |
4026 | * NET_RX_SUCCESS: no congestion | |
4027 | * NET_RX_DROP: packet was dropped | |
4028 | */ | |
04eb4489 | 4029 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa BH |
4030 | { |
4031 | trace_netif_receive_skb_entry(skb); | |
4032 | ||
4033 | return netif_receive_skb_internal(skb); | |
4034 | } | |
04eb4489 | 4035 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 4036 | |
88751275 ED |
4037 | /* Network device is going away, flush any packets still pending |
4038 | * Called with irqs disabled. | |
4039 | */ | |
152102c7 | 4040 | static void flush_backlog(void *arg) |
6e583ce5 | 4041 | { |
152102c7 | 4042 | struct net_device *dev = arg; |
903ceff7 | 4043 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
6e583ce5 SH |
4044 | struct sk_buff *skb, *tmp; |
4045 | ||
e36fa2f7 | 4046 | rps_lock(sd); |
6e7676c1 | 4047 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 4048 | if (skb->dev == dev) { |
e36fa2f7 | 4049 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 4050 | kfree_skb(skb); |
76cc8b13 | 4051 | input_queue_head_incr(sd); |
6e583ce5 | 4052 | } |
6e7676c1 | 4053 | } |
e36fa2f7 | 4054 | rps_unlock(sd); |
6e7676c1 CG |
4055 | |
4056 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
4057 | if (skb->dev == dev) { | |
4058 | __skb_unlink(skb, &sd->process_queue); | |
4059 | kfree_skb(skb); | |
76cc8b13 | 4060 | input_queue_head_incr(sd); |
6e7676c1 CG |
4061 | } |
4062 | } | |
6e583ce5 SH |
4063 | } |
4064 | ||
d565b0a1 HX |
4065 | static int napi_gro_complete(struct sk_buff *skb) |
4066 | { | |
22061d80 | 4067 | struct packet_offload *ptype; |
d565b0a1 | 4068 | __be16 type = skb->protocol; |
22061d80 | 4069 | struct list_head *head = &offload_base; |
d565b0a1 HX |
4070 | int err = -ENOENT; |
4071 | ||
c3c7c254 ED |
4072 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
4073 | ||
fc59f9a3 HX |
4074 | if (NAPI_GRO_CB(skb)->count == 1) { |
4075 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 4076 | goto out; |
fc59f9a3 | 4077 | } |
d565b0a1 HX |
4078 | |
4079 | rcu_read_lock(); | |
4080 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4081 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
4082 | continue; |
4083 | ||
299603e8 | 4084 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
4085 | break; |
4086 | } | |
4087 | rcu_read_unlock(); | |
4088 | ||
4089 | if (err) { | |
4090 | WARN_ON(&ptype->list == head); | |
4091 | kfree_skb(skb); | |
4092 | return NET_RX_SUCCESS; | |
4093 | } | |
4094 | ||
4095 | out: | |
ae78dbfa | 4096 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
4097 | } |
4098 | ||
2e71a6f8 ED |
4099 | /* napi->gro_list contains packets ordered by age. |
4100 | * youngest packets at the head of it. | |
4101 | * Complete skbs in reverse order to reduce latencies. | |
4102 | */ | |
4103 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 4104 | { |
2e71a6f8 | 4105 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 4106 | |
2e71a6f8 ED |
4107 | /* scan list and build reverse chain */ |
4108 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
4109 | skb->prev = prev; | |
4110 | prev = skb; | |
4111 | } | |
4112 | ||
4113 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 4114 | skb->next = NULL; |
2e71a6f8 ED |
4115 | |
4116 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
4117 | return; | |
4118 | ||
4119 | prev = skb->prev; | |
d565b0a1 | 4120 | napi_gro_complete(skb); |
2e71a6f8 | 4121 | napi->gro_count--; |
d565b0a1 HX |
4122 | } |
4123 | ||
4124 | napi->gro_list = NULL; | |
4125 | } | |
86cac58b | 4126 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 4127 | |
89c5fa33 ED |
4128 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
4129 | { | |
4130 | struct sk_buff *p; | |
4131 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 4132 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
4133 | |
4134 | for (p = napi->gro_list; p; p = p->next) { | |
4135 | unsigned long diffs; | |
4136 | ||
0b4cec8c TH |
4137 | NAPI_GRO_CB(p)->flush = 0; |
4138 | ||
4139 | if (hash != skb_get_hash_raw(p)) { | |
4140 | NAPI_GRO_CB(p)->same_flow = 0; | |
4141 | continue; | |
4142 | } | |
4143 | ||
89c5fa33 ED |
4144 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
4145 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
4146 | if (maclen == ETH_HLEN) | |
4147 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 4148 | skb_mac_header(skb)); |
89c5fa33 ED |
4149 | else if (!diffs) |
4150 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 4151 | skb_mac_header(skb), |
89c5fa33 ED |
4152 | maclen); |
4153 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
4154 | } |
4155 | } | |
4156 | ||
299603e8 JC |
4157 | static void skb_gro_reset_offset(struct sk_buff *skb) |
4158 | { | |
4159 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4160 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
4161 | ||
4162 | NAPI_GRO_CB(skb)->data_offset = 0; | |
4163 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
4164 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
4165 | ||
4166 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
4167 | pinfo->nr_frags && | |
4168 | !PageHighMem(skb_frag_page(frag0))) { | |
4169 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
4170 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
89c5fa33 ED |
4171 | } |
4172 | } | |
4173 | ||
a50e233c ED |
4174 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
4175 | { | |
4176 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4177 | ||
4178 | BUG_ON(skb->end - skb->tail < grow); | |
4179 | ||
4180 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
4181 | ||
4182 | skb->data_len -= grow; | |
4183 | skb->tail += grow; | |
4184 | ||
4185 | pinfo->frags[0].page_offset += grow; | |
4186 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
4187 | ||
4188 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
4189 | skb_frag_unref(skb, 0); | |
4190 | memmove(pinfo->frags, pinfo->frags + 1, | |
4191 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
4192 | } | |
4193 | } | |
4194 | ||
bb728820 | 4195 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
4196 | { |
4197 | struct sk_buff **pp = NULL; | |
22061d80 | 4198 | struct packet_offload *ptype; |
d565b0a1 | 4199 | __be16 type = skb->protocol; |
22061d80 | 4200 | struct list_head *head = &offload_base; |
0da2afd5 | 4201 | int same_flow; |
5b252f0c | 4202 | enum gro_result ret; |
a50e233c | 4203 | int grow; |
d565b0a1 | 4204 | |
9c62a68d | 4205 | if (!(skb->dev->features & NETIF_F_GRO)) |
d565b0a1 HX |
4206 | goto normal; |
4207 | ||
5a212329 | 4208 | if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad) |
f17f5c91 HX |
4209 | goto normal; |
4210 | ||
89c5fa33 ED |
4211 | gro_list_prepare(napi, skb); |
4212 | ||
d565b0a1 HX |
4213 | rcu_read_lock(); |
4214 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4215 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4216 | continue; |
4217 | ||
86911732 | 4218 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4219 | skb_reset_mac_len(skb); |
d565b0a1 HX |
4220 | NAPI_GRO_CB(skb)->same_flow = 0; |
4221 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 4222 | NAPI_GRO_CB(skb)->free = 0; |
b582ef09 | 4223 | NAPI_GRO_CB(skb)->udp_mark = 0; |
15e2396d | 4224 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 4225 | |
662880f4 TH |
4226 | /* Setup for GRO checksum validation */ |
4227 | switch (skb->ip_summed) { | |
4228 | case CHECKSUM_COMPLETE: | |
4229 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4230 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4231 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4232 | break; | |
4233 | case CHECKSUM_UNNECESSARY: | |
4234 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4235 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4236 | break; | |
4237 | default: | |
4238 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4239 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4240 | } | |
d565b0a1 | 4241 | |
f191a1d1 | 4242 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4243 | break; |
4244 | } | |
4245 | rcu_read_unlock(); | |
4246 | ||
4247 | if (&ptype->list == head) | |
4248 | goto normal; | |
4249 | ||
0da2afd5 | 4250 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4251 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4252 | |
d565b0a1 HX |
4253 | if (pp) { |
4254 | struct sk_buff *nskb = *pp; | |
4255 | ||
4256 | *pp = nskb->next; | |
4257 | nskb->next = NULL; | |
4258 | napi_gro_complete(nskb); | |
4ae5544f | 4259 | napi->gro_count--; |
d565b0a1 HX |
4260 | } |
4261 | ||
0da2afd5 | 4262 | if (same_flow) |
d565b0a1 HX |
4263 | goto ok; |
4264 | ||
600adc18 | 4265 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4266 | goto normal; |
d565b0a1 | 4267 | |
600adc18 ED |
4268 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4269 | struct sk_buff *nskb = napi->gro_list; | |
4270 | ||
4271 | /* locate the end of the list to select the 'oldest' flow */ | |
4272 | while (nskb->next) { | |
4273 | pp = &nskb->next; | |
4274 | nskb = *pp; | |
4275 | } | |
4276 | *pp = NULL; | |
4277 | nskb->next = NULL; | |
4278 | napi_gro_complete(nskb); | |
4279 | } else { | |
4280 | napi->gro_count++; | |
4281 | } | |
d565b0a1 | 4282 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4283 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4284 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4285 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4286 | skb->next = napi->gro_list; |
4287 | napi->gro_list = skb; | |
5d0d9be8 | 4288 | ret = GRO_HELD; |
d565b0a1 | 4289 | |
ad0f9904 | 4290 | pull: |
a50e233c ED |
4291 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4292 | if (grow > 0) | |
4293 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4294 | ok: |
5d0d9be8 | 4295 | return ret; |
d565b0a1 HX |
4296 | |
4297 | normal: | |
ad0f9904 HX |
4298 | ret = GRO_NORMAL; |
4299 | goto pull; | |
5d38a079 | 4300 | } |
96e93eab | 4301 | |
bf5a755f JC |
4302 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4303 | { | |
4304 | struct list_head *offload_head = &offload_base; | |
4305 | struct packet_offload *ptype; | |
4306 | ||
4307 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4308 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4309 | continue; | |
4310 | return ptype; | |
4311 | } | |
4312 | return NULL; | |
4313 | } | |
e27a2f83 | 4314 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4315 | |
4316 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4317 | { | |
4318 | struct list_head *offload_head = &offload_base; | |
4319 | struct packet_offload *ptype; | |
4320 | ||
4321 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4322 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4323 | continue; | |
4324 | return ptype; | |
4325 | } | |
4326 | return NULL; | |
4327 | } | |
e27a2f83 | 4328 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4329 | |
bb728820 | 4330 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4331 | { |
5d0d9be8 HX |
4332 | switch (ret) { |
4333 | case GRO_NORMAL: | |
ae78dbfa | 4334 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4335 | ret = GRO_DROP; |
4336 | break; | |
5d38a079 | 4337 | |
5d0d9be8 | 4338 | case GRO_DROP: |
5d38a079 HX |
4339 | kfree_skb(skb); |
4340 | break; | |
5b252f0c | 4341 | |
daa86548 | 4342 | case GRO_MERGED_FREE: |
d7e8883c ED |
4343 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
4344 | kmem_cache_free(skbuff_head_cache, skb); | |
4345 | else | |
4346 | __kfree_skb(skb); | |
daa86548 ED |
4347 | break; |
4348 | ||
5b252f0c BH |
4349 | case GRO_HELD: |
4350 | case GRO_MERGED: | |
4351 | break; | |
5d38a079 HX |
4352 | } |
4353 | ||
c7c4b3b6 | 4354 | return ret; |
5d0d9be8 | 4355 | } |
5d0d9be8 | 4356 | |
c7c4b3b6 | 4357 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4358 | { |
93f93a44 | 4359 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 4360 | trace_napi_gro_receive_entry(skb); |
86911732 | 4361 | |
a50e233c ED |
4362 | skb_gro_reset_offset(skb); |
4363 | ||
89c5fa33 | 4364 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4365 | } |
4366 | EXPORT_SYMBOL(napi_gro_receive); | |
4367 | ||
d0c2b0d2 | 4368 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4369 | { |
93a35f59 ED |
4370 | if (unlikely(skb->pfmemalloc)) { |
4371 | consume_skb(skb); | |
4372 | return; | |
4373 | } | |
96e93eab | 4374 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4375 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4376 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4377 | skb->vlan_tci = 0; |
66c46d74 | 4378 | skb->dev = napi->dev; |
6d152e23 | 4379 | skb->skb_iif = 0; |
c3caf119 JC |
4380 | skb->encapsulation = 0; |
4381 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4382 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
96e93eab HX |
4383 | |
4384 | napi->skb = skb; | |
4385 | } | |
96e93eab | 4386 | |
76620aaf | 4387 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4388 | { |
5d38a079 | 4389 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4390 | |
4391 | if (!skb) { | |
fd11a83d | 4392 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
4393 | if (skb) { |
4394 | napi->skb = skb; | |
4395 | skb_mark_napi_id(skb, napi); | |
4396 | } | |
80595d59 | 4397 | } |
96e93eab HX |
4398 | return skb; |
4399 | } | |
76620aaf | 4400 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4401 | |
a50e233c ED |
4402 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4403 | struct sk_buff *skb, | |
4404 | gro_result_t ret) | |
96e93eab | 4405 | { |
5d0d9be8 HX |
4406 | switch (ret) { |
4407 | case GRO_NORMAL: | |
a50e233c ED |
4408 | case GRO_HELD: |
4409 | __skb_push(skb, ETH_HLEN); | |
4410 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4411 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4412 | ret = GRO_DROP; |
86911732 | 4413 | break; |
5d38a079 | 4414 | |
5d0d9be8 | 4415 | case GRO_DROP: |
5d0d9be8 HX |
4416 | case GRO_MERGED_FREE: |
4417 | napi_reuse_skb(napi, skb); | |
4418 | break; | |
5b252f0c BH |
4419 | |
4420 | case GRO_MERGED: | |
4421 | break; | |
5d0d9be8 | 4422 | } |
5d38a079 | 4423 | |
c7c4b3b6 | 4424 | return ret; |
5d38a079 | 4425 | } |
5d0d9be8 | 4426 | |
a50e233c ED |
4427 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
4428 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
4429 | * We copy ethernet header into skb->data to have a common layout. | |
4430 | */ | |
4adb9c4a | 4431 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4432 | { |
4433 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
4434 | const struct ethhdr *eth; |
4435 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
4436 | |
4437 | napi->skb = NULL; | |
4438 | ||
a50e233c ED |
4439 | skb_reset_mac_header(skb); |
4440 | skb_gro_reset_offset(skb); | |
4441 | ||
4442 | eth = skb_gro_header_fast(skb, 0); | |
4443 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
4444 | eth = skb_gro_header_slow(skb, hlen, 0); | |
4445 | if (unlikely(!eth)) { | |
4446 | napi_reuse_skb(napi, skb); | |
4447 | return NULL; | |
4448 | } | |
4449 | } else { | |
4450 | gro_pull_from_frag0(skb, hlen); | |
4451 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
4452 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 4453 | } |
a50e233c ED |
4454 | __skb_pull(skb, hlen); |
4455 | ||
4456 | /* | |
4457 | * This works because the only protocols we care about don't require | |
4458 | * special handling. | |
4459 | * We'll fix it up properly in napi_frags_finish() | |
4460 | */ | |
4461 | skb->protocol = eth->h_proto; | |
76620aaf | 4462 | |
76620aaf HX |
4463 | return skb; |
4464 | } | |
76620aaf | 4465 | |
c7c4b3b6 | 4466 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4467 | { |
76620aaf | 4468 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4469 | |
4470 | if (!skb) | |
c7c4b3b6 | 4471 | return GRO_DROP; |
5d0d9be8 | 4472 | |
ae78dbfa BH |
4473 | trace_napi_gro_frags_entry(skb); |
4474 | ||
89c5fa33 | 4475 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4476 | } |
5d38a079 HX |
4477 | EXPORT_SYMBOL(napi_gro_frags); |
4478 | ||
573e8fca TH |
4479 | /* Compute the checksum from gro_offset and return the folded value |
4480 | * after adding in any pseudo checksum. | |
4481 | */ | |
4482 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
4483 | { | |
4484 | __wsum wsum; | |
4485 | __sum16 sum; | |
4486 | ||
4487 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
4488 | ||
4489 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
4490 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
4491 | if (likely(!sum)) { | |
4492 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
4493 | !skb->csum_complete_sw) | |
4494 | netdev_rx_csum_fault(skb->dev); | |
4495 | } | |
4496 | ||
4497 | NAPI_GRO_CB(skb)->csum = wsum; | |
4498 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4499 | ||
4500 | return sum; | |
4501 | } | |
4502 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
4503 | ||
e326bed2 | 4504 | /* |
855abcf0 | 4505 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
4506 | * Note: called with local irq disabled, but exits with local irq enabled. |
4507 | */ | |
4508 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4509 | { | |
4510 | #ifdef CONFIG_RPS | |
4511 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4512 | ||
4513 | if (remsd) { | |
4514 | sd->rps_ipi_list = NULL; | |
4515 | ||
4516 | local_irq_enable(); | |
4517 | ||
4518 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
4519 | while (remsd) { | |
4520 | struct softnet_data *next = remsd->rps_ipi_next; | |
4521 | ||
4522 | if (cpu_online(remsd->cpu)) | |
c46fff2a | 4523 | smp_call_function_single_async(remsd->cpu, |
fce8ad15 | 4524 | &remsd->csd); |
e326bed2 ED |
4525 | remsd = next; |
4526 | } | |
4527 | } else | |
4528 | #endif | |
4529 | local_irq_enable(); | |
4530 | } | |
4531 | ||
d75b1ade ED |
4532 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
4533 | { | |
4534 | #ifdef CONFIG_RPS | |
4535 | return sd->rps_ipi_list != NULL; | |
4536 | #else | |
4537 | return false; | |
4538 | #endif | |
4539 | } | |
4540 | ||
bea3348e | 4541 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
4542 | { |
4543 | int work = 0; | |
eecfd7c4 | 4544 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 4545 | |
e326bed2 ED |
4546 | /* Check if we have pending ipi, its better to send them now, |
4547 | * not waiting net_rx_action() end. | |
4548 | */ | |
d75b1ade | 4549 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
4550 | local_irq_disable(); |
4551 | net_rps_action_and_irq_enable(sd); | |
4552 | } | |
d75b1ade | 4553 | |
bea3348e | 4554 | napi->weight = weight_p; |
6e7676c1 | 4555 | local_irq_disable(); |
11ef7a89 | 4556 | while (1) { |
1da177e4 | 4557 | struct sk_buff *skb; |
6e7676c1 CG |
4558 | |
4559 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 4560 | rcu_read_lock(); |
6e7676c1 CG |
4561 | local_irq_enable(); |
4562 | __netif_receive_skb(skb); | |
2c17d27c | 4563 | rcu_read_unlock(); |
6e7676c1 | 4564 | local_irq_disable(); |
76cc8b13 TH |
4565 | input_queue_head_incr(sd); |
4566 | if (++work >= quota) { | |
4567 | local_irq_enable(); | |
4568 | return work; | |
4569 | } | |
6e7676c1 | 4570 | } |
1da177e4 | 4571 | |
e36fa2f7 | 4572 | rps_lock(sd); |
11ef7a89 | 4573 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
4574 | /* |
4575 | * Inline a custom version of __napi_complete(). | |
4576 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
4577 | * and NAPI_STATE_SCHED is the only possible flag set |
4578 | * on backlog. | |
4579 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
4580 | * and we dont need an smp_mb() memory barrier. |
4581 | */ | |
eecfd7c4 | 4582 | napi->state = 0; |
11ef7a89 | 4583 | rps_unlock(sd); |
eecfd7c4 | 4584 | |
11ef7a89 | 4585 | break; |
bea3348e | 4586 | } |
11ef7a89 TH |
4587 | |
4588 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
4589 | &sd->process_queue); | |
e36fa2f7 | 4590 | rps_unlock(sd); |
6e7676c1 CG |
4591 | } |
4592 | local_irq_enable(); | |
1da177e4 | 4593 | |
bea3348e SH |
4594 | return work; |
4595 | } | |
1da177e4 | 4596 | |
bea3348e SH |
4597 | /** |
4598 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4599 | * @n: entry to schedule |
bea3348e | 4600 | * |
bc9ad166 ED |
4601 | * The entry's receive function will be scheduled to run. |
4602 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 4603 | */ |
b5606c2d | 4604 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4605 | { |
4606 | unsigned long flags; | |
1da177e4 | 4607 | |
bea3348e | 4608 | local_irq_save(flags); |
903ceff7 | 4609 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 4610 | local_irq_restore(flags); |
1da177e4 | 4611 | } |
bea3348e SH |
4612 | EXPORT_SYMBOL(__napi_schedule); |
4613 | ||
bc9ad166 ED |
4614 | /** |
4615 | * __napi_schedule_irqoff - schedule for receive | |
4616 | * @n: entry to schedule | |
4617 | * | |
4618 | * Variant of __napi_schedule() assuming hard irqs are masked | |
4619 | */ | |
4620 | void __napi_schedule_irqoff(struct napi_struct *n) | |
4621 | { | |
4622 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
4623 | } | |
4624 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
4625 | ||
d565b0a1 HX |
4626 | void __napi_complete(struct napi_struct *n) |
4627 | { | |
4628 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
d565b0a1 | 4629 | |
d75b1ade | 4630 | list_del_init(&n->poll_list); |
4e857c58 | 4631 | smp_mb__before_atomic(); |
d565b0a1 HX |
4632 | clear_bit(NAPI_STATE_SCHED, &n->state); |
4633 | } | |
4634 | EXPORT_SYMBOL(__napi_complete); | |
4635 | ||
3b47d303 | 4636 | void napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 HX |
4637 | { |
4638 | unsigned long flags; | |
4639 | ||
4640 | /* | |
4641 | * don't let napi dequeue from the cpu poll list | |
4642 | * just in case its running on a different cpu | |
4643 | */ | |
4644 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4645 | return; | |
4646 | ||
3b47d303 ED |
4647 | if (n->gro_list) { |
4648 | unsigned long timeout = 0; | |
d75b1ade | 4649 | |
3b47d303 ED |
4650 | if (work_done) |
4651 | timeout = n->dev->gro_flush_timeout; | |
4652 | ||
4653 | if (timeout) | |
4654 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
4655 | HRTIMER_MODE_REL_PINNED); | |
4656 | else | |
4657 | napi_gro_flush(n, false); | |
4658 | } | |
d75b1ade ED |
4659 | if (likely(list_empty(&n->poll_list))) { |
4660 | WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state)); | |
4661 | } else { | |
4662 | /* If n->poll_list is not empty, we need to mask irqs */ | |
4663 | local_irq_save(flags); | |
4664 | __napi_complete(n); | |
4665 | local_irq_restore(flags); | |
4666 | } | |
d565b0a1 | 4667 | } |
3b47d303 | 4668 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 4669 | |
af12fa6e | 4670 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 4671 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
4672 | { |
4673 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
4674 | struct napi_struct *napi; | |
4675 | ||
4676 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
4677 | if (napi->napi_id == napi_id) | |
4678 | return napi; | |
4679 | ||
4680 | return NULL; | |
4681 | } | |
02d62e86 ED |
4682 | |
4683 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
ce6aea93 | 4684 | #define BUSY_POLL_BUDGET 8 |
02d62e86 ED |
4685 | bool sk_busy_loop(struct sock *sk, int nonblock) |
4686 | { | |
4687 | unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0; | |
ce6aea93 | 4688 | int (*busy_poll)(struct napi_struct *dev); |
02d62e86 ED |
4689 | struct napi_struct *napi; |
4690 | int rc = false; | |
4691 | ||
2a028ecb | 4692 | rcu_read_lock(); |
02d62e86 ED |
4693 | |
4694 | napi = napi_by_id(sk->sk_napi_id); | |
4695 | if (!napi) | |
4696 | goto out; | |
4697 | ||
ce6aea93 ED |
4698 | /* Note: ndo_busy_poll method is optional in linux-4.5 */ |
4699 | busy_poll = napi->dev->netdev_ops->ndo_busy_poll; | |
02d62e86 ED |
4700 | |
4701 | do { | |
ce6aea93 | 4702 | rc = 0; |
2a028ecb | 4703 | local_bh_disable(); |
ce6aea93 ED |
4704 | if (busy_poll) { |
4705 | rc = busy_poll(napi); | |
4706 | } else if (napi_schedule_prep(napi)) { | |
4707 | void *have = netpoll_poll_lock(napi); | |
4708 | ||
4709 | if (test_bit(NAPI_STATE_SCHED, &napi->state)) { | |
4710 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
4711 | trace_napi_poll(napi); | |
4712 | if (rc == BUSY_POLL_BUDGET) { | |
4713 | napi_complete_done(napi, rc); | |
4714 | napi_schedule(napi); | |
4715 | } | |
4716 | } | |
4717 | netpoll_poll_unlock(have); | |
4718 | } | |
2a028ecb ED |
4719 | if (rc > 0) |
4720 | NET_ADD_STATS_BH(sock_net(sk), | |
4721 | LINUX_MIB_BUSYPOLLRXPACKETS, rc); | |
4722 | local_bh_enable(); | |
02d62e86 ED |
4723 | |
4724 | if (rc == LL_FLUSH_FAILED) | |
4725 | break; /* permanent failure */ | |
4726 | ||
02d62e86 | 4727 | cpu_relax(); |
02d62e86 ED |
4728 | } while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) && |
4729 | !need_resched() && !busy_loop_timeout(end_time)); | |
4730 | ||
4731 | rc = !skb_queue_empty(&sk->sk_receive_queue); | |
4732 | out: | |
2a028ecb | 4733 | rcu_read_unlock(); |
02d62e86 ED |
4734 | return rc; |
4735 | } | |
4736 | EXPORT_SYMBOL(sk_busy_loop); | |
4737 | ||
4738 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e ET |
4739 | |
4740 | void napi_hash_add(struct napi_struct *napi) | |
4741 | { | |
d64b5e85 ED |
4742 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
4743 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 4744 | return; |
af12fa6e | 4745 | |
52bd2d62 | 4746 | spin_lock(&napi_hash_lock); |
af12fa6e | 4747 | |
52bd2d62 ED |
4748 | /* 0..NR_CPUS+1 range is reserved for sender_cpu use */ |
4749 | do { | |
4750 | if (unlikely(++napi_gen_id < NR_CPUS + 1)) | |
4751 | napi_gen_id = NR_CPUS + 1; | |
4752 | } while (napi_by_id(napi_gen_id)); | |
4753 | napi->napi_id = napi_gen_id; | |
af12fa6e | 4754 | |
52bd2d62 ED |
4755 | hlist_add_head_rcu(&napi->napi_hash_node, |
4756 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 4757 | |
52bd2d62 | 4758 | spin_unlock(&napi_hash_lock); |
af12fa6e ET |
4759 | } |
4760 | EXPORT_SYMBOL_GPL(napi_hash_add); | |
4761 | ||
4762 | /* Warning : caller is responsible to make sure rcu grace period | |
4763 | * is respected before freeing memory containing @napi | |
4764 | */ | |
34cbe27e | 4765 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 4766 | { |
34cbe27e ED |
4767 | bool rcu_sync_needed = false; |
4768 | ||
af12fa6e ET |
4769 | spin_lock(&napi_hash_lock); |
4770 | ||
34cbe27e ED |
4771 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
4772 | rcu_sync_needed = true; | |
af12fa6e | 4773 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 4774 | } |
af12fa6e | 4775 | spin_unlock(&napi_hash_lock); |
34cbe27e | 4776 | return rcu_sync_needed; |
af12fa6e ET |
4777 | } |
4778 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
4779 | ||
3b47d303 ED |
4780 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
4781 | { | |
4782 | struct napi_struct *napi; | |
4783 | ||
4784 | napi = container_of(timer, struct napi_struct, timer); | |
4785 | if (napi->gro_list) | |
4786 | napi_schedule(napi); | |
4787 | ||
4788 | return HRTIMER_NORESTART; | |
4789 | } | |
4790 | ||
d565b0a1 HX |
4791 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
4792 | int (*poll)(struct napi_struct *, int), int weight) | |
4793 | { | |
4794 | INIT_LIST_HEAD(&napi->poll_list); | |
3b47d303 ED |
4795 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
4796 | napi->timer.function = napi_watchdog; | |
4ae5544f | 4797 | napi->gro_count = 0; |
d565b0a1 | 4798 | napi->gro_list = NULL; |
5d38a079 | 4799 | napi->skb = NULL; |
d565b0a1 | 4800 | napi->poll = poll; |
82dc3c63 ED |
4801 | if (weight > NAPI_POLL_WEIGHT) |
4802 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
4803 | weight, dev->name); | |
d565b0a1 HX |
4804 | napi->weight = weight; |
4805 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 4806 | napi->dev = dev; |
5d38a079 | 4807 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
4808 | spin_lock_init(&napi->poll_lock); |
4809 | napi->poll_owner = -1; | |
4810 | #endif | |
4811 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 4812 | napi_hash_add(napi); |
d565b0a1 HX |
4813 | } |
4814 | EXPORT_SYMBOL(netif_napi_add); | |
4815 | ||
3b47d303 ED |
4816 | void napi_disable(struct napi_struct *n) |
4817 | { | |
4818 | might_sleep(); | |
4819 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
4820 | ||
4821 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
4822 | msleep(1); | |
2d8bff12 NH |
4823 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
4824 | msleep(1); | |
3b47d303 ED |
4825 | |
4826 | hrtimer_cancel(&n->timer); | |
4827 | ||
4828 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
4829 | } | |
4830 | EXPORT_SYMBOL(napi_disable); | |
4831 | ||
93d05d4a | 4832 | /* Must be called in process context */ |
d565b0a1 HX |
4833 | void netif_napi_del(struct napi_struct *napi) |
4834 | { | |
93d05d4a ED |
4835 | might_sleep(); |
4836 | if (napi_hash_del(napi)) | |
4837 | synchronize_net(); | |
d7b06636 | 4838 | list_del_init(&napi->dev_list); |
76620aaf | 4839 | napi_free_frags(napi); |
d565b0a1 | 4840 | |
289dccbe | 4841 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 4842 | napi->gro_list = NULL; |
4ae5544f | 4843 | napi->gro_count = 0; |
d565b0a1 HX |
4844 | } |
4845 | EXPORT_SYMBOL(netif_napi_del); | |
4846 | ||
726ce70e HX |
4847 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
4848 | { | |
4849 | void *have; | |
4850 | int work, weight; | |
4851 | ||
4852 | list_del_init(&n->poll_list); | |
4853 | ||
4854 | have = netpoll_poll_lock(n); | |
4855 | ||
4856 | weight = n->weight; | |
4857 | ||
4858 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
4859 | * with netpoll's poll_napi(). Only the entity which | |
4860 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
4861 | * actually make the ->poll() call. Therefore we avoid | |
4862 | * accidentally calling ->poll() when NAPI is not scheduled. | |
4863 | */ | |
4864 | work = 0; | |
4865 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
4866 | work = n->poll(n, weight); | |
4867 | trace_napi_poll(n); | |
4868 | } | |
4869 | ||
4870 | WARN_ON_ONCE(work > weight); | |
4871 | ||
4872 | if (likely(work < weight)) | |
4873 | goto out_unlock; | |
4874 | ||
4875 | /* Drivers must not modify the NAPI state if they | |
4876 | * consume the entire weight. In such cases this code | |
4877 | * still "owns" the NAPI instance and therefore can | |
4878 | * move the instance around on the list at-will. | |
4879 | */ | |
4880 | if (unlikely(napi_disable_pending(n))) { | |
4881 | napi_complete(n); | |
4882 | goto out_unlock; | |
4883 | } | |
4884 | ||
4885 | if (n->gro_list) { | |
4886 | /* flush too old packets | |
4887 | * If HZ < 1000, flush all packets. | |
4888 | */ | |
4889 | napi_gro_flush(n, HZ >= 1000); | |
4890 | } | |
4891 | ||
001ce546 HX |
4892 | /* Some drivers may have called napi_schedule |
4893 | * prior to exhausting their budget. | |
4894 | */ | |
4895 | if (unlikely(!list_empty(&n->poll_list))) { | |
4896 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
4897 | n->dev ? n->dev->name : "backlog"); | |
4898 | goto out_unlock; | |
4899 | } | |
4900 | ||
726ce70e HX |
4901 | list_add_tail(&n->poll_list, repoll); |
4902 | ||
4903 | out_unlock: | |
4904 | netpoll_poll_unlock(have); | |
4905 | ||
4906 | return work; | |
4907 | } | |
4908 | ||
1da177e4 LT |
4909 | static void net_rx_action(struct softirq_action *h) |
4910 | { | |
903ceff7 | 4911 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
24f8b238 | 4912 | unsigned long time_limit = jiffies + 2; |
51b0bded | 4913 | int budget = netdev_budget; |
d75b1ade ED |
4914 | LIST_HEAD(list); |
4915 | LIST_HEAD(repoll); | |
53fb95d3 | 4916 | |
1da177e4 | 4917 | local_irq_disable(); |
d75b1ade ED |
4918 | list_splice_init(&sd->poll_list, &list); |
4919 | local_irq_enable(); | |
1da177e4 | 4920 | |
ceb8d5bf | 4921 | for (;;) { |
bea3348e | 4922 | struct napi_struct *n; |
1da177e4 | 4923 | |
ceb8d5bf HX |
4924 | if (list_empty(&list)) { |
4925 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
4926 | return; | |
4927 | break; | |
4928 | } | |
4929 | ||
6bd373eb HX |
4930 | n = list_first_entry(&list, struct napi_struct, poll_list); |
4931 | budget -= napi_poll(n, &repoll); | |
4932 | ||
d75b1ade | 4933 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
4934 | * Allow this to run for 2 jiffies since which will allow |
4935 | * an average latency of 1.5/HZ. | |
bea3348e | 4936 | */ |
ceb8d5bf HX |
4937 | if (unlikely(budget <= 0 || |
4938 | time_after_eq(jiffies, time_limit))) { | |
4939 | sd->time_squeeze++; | |
4940 | break; | |
4941 | } | |
1da177e4 | 4942 | } |
d75b1ade | 4943 | |
d75b1ade ED |
4944 | local_irq_disable(); |
4945 | ||
4946 | list_splice_tail_init(&sd->poll_list, &list); | |
4947 | list_splice_tail(&repoll, &list); | |
4948 | list_splice(&list, &sd->poll_list); | |
4949 | if (!list_empty(&sd->poll_list)) | |
4950 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4951 | ||
e326bed2 | 4952 | net_rps_action_and_irq_enable(sd); |
1da177e4 LT |
4953 | } |
4954 | ||
aa9d8560 | 4955 | struct netdev_adjacent { |
9ff162a8 | 4956 | struct net_device *dev; |
5d261913 VF |
4957 | |
4958 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 4959 | bool master; |
5d261913 | 4960 | |
5d261913 VF |
4961 | /* counter for the number of times this device was added to us */ |
4962 | u16 ref_nr; | |
4963 | ||
402dae96 VF |
4964 | /* private field for the users */ |
4965 | void *private; | |
4966 | ||
9ff162a8 JP |
4967 | struct list_head list; |
4968 | struct rcu_head rcu; | |
9ff162a8 JP |
4969 | }; |
4970 | ||
6ea29da1 | 4971 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 4972 | struct list_head *adj_list) |
9ff162a8 | 4973 | { |
5d261913 | 4974 | struct netdev_adjacent *adj; |
5d261913 | 4975 | |
2f268f12 | 4976 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
4977 | if (adj->dev == adj_dev) |
4978 | return adj; | |
9ff162a8 JP |
4979 | } |
4980 | return NULL; | |
4981 | } | |
4982 | ||
4983 | /** | |
4984 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
4985 | * @dev: device | |
4986 | * @upper_dev: upper device to check | |
4987 | * | |
4988 | * Find out if a device is linked to specified upper device and return true | |
4989 | * in case it is. Note that this checks only immediate upper device, | |
4990 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
4991 | */ | |
4992 | bool netdev_has_upper_dev(struct net_device *dev, | |
4993 | struct net_device *upper_dev) | |
4994 | { | |
4995 | ASSERT_RTNL(); | |
4996 | ||
6ea29da1 | 4997 | return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper); |
9ff162a8 JP |
4998 | } |
4999 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
5000 | ||
5001 | /** | |
5002 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
5003 | * @dev: device | |
5004 | * | |
5005 | * Find out if a device is linked to an upper device and return true in case | |
5006 | * it is. The caller must hold the RTNL lock. | |
5007 | */ | |
1d143d9f | 5008 | static bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
5009 | { |
5010 | ASSERT_RTNL(); | |
5011 | ||
2f268f12 | 5012 | return !list_empty(&dev->all_adj_list.upper); |
9ff162a8 | 5013 | } |
9ff162a8 JP |
5014 | |
5015 | /** | |
5016 | * netdev_master_upper_dev_get - Get master upper device | |
5017 | * @dev: device | |
5018 | * | |
5019 | * Find a master upper device and return pointer to it or NULL in case | |
5020 | * it's not there. The caller must hold the RTNL lock. | |
5021 | */ | |
5022 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
5023 | { | |
aa9d8560 | 5024 | struct netdev_adjacent *upper; |
9ff162a8 JP |
5025 | |
5026 | ASSERT_RTNL(); | |
5027 | ||
2f268f12 | 5028 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
5029 | return NULL; |
5030 | ||
2f268f12 | 5031 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 5032 | struct netdev_adjacent, list); |
9ff162a8 JP |
5033 | if (likely(upper->master)) |
5034 | return upper->dev; | |
5035 | return NULL; | |
5036 | } | |
5037 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
5038 | ||
b6ccba4c VF |
5039 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
5040 | { | |
5041 | struct netdev_adjacent *adj; | |
5042 | ||
5043 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
5044 | ||
5045 | return adj->private; | |
5046 | } | |
5047 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
5048 | ||
44a40855 VY |
5049 | /** |
5050 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
5051 | * @dev: device | |
5052 | * @iter: list_head ** of the current position | |
5053 | * | |
5054 | * Gets the next device from the dev's upper list, starting from iter | |
5055 | * position. The caller must hold RCU read lock. | |
5056 | */ | |
5057 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
5058 | struct list_head **iter) | |
5059 | { | |
5060 | struct netdev_adjacent *upper; | |
5061 | ||
5062 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5063 | ||
5064 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5065 | ||
5066 | if (&upper->list == &dev->adj_list.upper) | |
5067 | return NULL; | |
5068 | ||
5069 | *iter = &upper->list; | |
5070 | ||
5071 | return upper->dev; | |
5072 | } | |
5073 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
5074 | ||
31088a11 VF |
5075 | /** |
5076 | * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list | |
48311f46 VF |
5077 | * @dev: device |
5078 | * @iter: list_head ** of the current position | |
5079 | * | |
5080 | * Gets the next device from the dev's upper list, starting from iter | |
5081 | * position. The caller must hold RCU read lock. | |
5082 | */ | |
2f268f12 VF |
5083 | struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, |
5084 | struct list_head **iter) | |
48311f46 VF |
5085 | { |
5086 | struct netdev_adjacent *upper; | |
5087 | ||
85328240 | 5088 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); |
48311f46 VF |
5089 | |
5090 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5091 | ||
2f268f12 | 5092 | if (&upper->list == &dev->all_adj_list.upper) |
48311f46 VF |
5093 | return NULL; |
5094 | ||
5095 | *iter = &upper->list; | |
5096 | ||
5097 | return upper->dev; | |
5098 | } | |
2f268f12 | 5099 | EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); |
48311f46 | 5100 | |
31088a11 VF |
5101 | /** |
5102 | * netdev_lower_get_next_private - Get the next ->private from the | |
5103 | * lower neighbour list | |
5104 | * @dev: device | |
5105 | * @iter: list_head ** of the current position | |
5106 | * | |
5107 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5108 | * list, starting from iter position. The caller must hold either hold the | |
5109 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 5110 | * list will remain unchanged. |
31088a11 VF |
5111 | */ |
5112 | void *netdev_lower_get_next_private(struct net_device *dev, | |
5113 | struct list_head **iter) | |
5114 | { | |
5115 | struct netdev_adjacent *lower; | |
5116 | ||
5117 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
5118 | ||
5119 | if (&lower->list == &dev->adj_list.lower) | |
5120 | return NULL; | |
5121 | ||
6859e7df | 5122 | *iter = lower->list.next; |
31088a11 VF |
5123 | |
5124 | return lower->private; | |
5125 | } | |
5126 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
5127 | ||
5128 | /** | |
5129 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
5130 | * lower neighbour list, RCU | |
5131 | * variant | |
5132 | * @dev: device | |
5133 | * @iter: list_head ** of the current position | |
5134 | * | |
5135 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5136 | * list, starting from iter position. The caller must hold RCU read lock. | |
5137 | */ | |
5138 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
5139 | struct list_head **iter) | |
5140 | { | |
5141 | struct netdev_adjacent *lower; | |
5142 | ||
5143 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5144 | ||
5145 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5146 | ||
5147 | if (&lower->list == &dev->adj_list.lower) | |
5148 | return NULL; | |
5149 | ||
6859e7df | 5150 | *iter = &lower->list; |
31088a11 VF |
5151 | |
5152 | return lower->private; | |
5153 | } | |
5154 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
5155 | ||
4085ebe8 VY |
5156 | /** |
5157 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
5158 | * list | |
5159 | * @dev: device | |
5160 | * @iter: list_head ** of the current position | |
5161 | * | |
5162 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
5163 | * list, starting from iter position. The caller must hold RTNL lock or | |
5164 | * its own locking that guarantees that the neighbour lower | |
b469139e | 5165 | * list will remain unchanged. |
4085ebe8 VY |
5166 | */ |
5167 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
5168 | { | |
5169 | struct netdev_adjacent *lower; | |
5170 | ||
5171 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
5172 | ||
5173 | if (&lower->list == &dev->adj_list.lower) | |
5174 | return NULL; | |
5175 | ||
5176 | *iter = &lower->list; | |
5177 | ||
5178 | return lower->dev; | |
5179 | } | |
5180 | EXPORT_SYMBOL(netdev_lower_get_next); | |
5181 | ||
e001bfad | 5182 | /** |
5183 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
5184 | * lower neighbour list, RCU | |
5185 | * variant | |
5186 | * @dev: device | |
5187 | * | |
5188 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
5189 | * list. The caller must hold RCU read lock. | |
5190 | */ | |
5191 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
5192 | { | |
5193 | struct netdev_adjacent *lower; | |
5194 | ||
5195 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
5196 | struct netdev_adjacent, list); | |
5197 | if (lower) | |
5198 | return lower->private; | |
5199 | return NULL; | |
5200 | } | |
5201 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
5202 | ||
9ff162a8 JP |
5203 | /** |
5204 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
5205 | * @dev: device | |
5206 | * | |
5207 | * Find a master upper device and return pointer to it or NULL in case | |
5208 | * it's not there. The caller must hold the RCU read lock. | |
5209 | */ | |
5210 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
5211 | { | |
aa9d8560 | 5212 | struct netdev_adjacent *upper; |
9ff162a8 | 5213 | |
2f268f12 | 5214 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 5215 | struct netdev_adjacent, list); |
9ff162a8 JP |
5216 | if (upper && likely(upper->master)) |
5217 | return upper->dev; | |
5218 | return NULL; | |
5219 | } | |
5220 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
5221 | ||
0a59f3a9 | 5222 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
5223 | struct net_device *adj_dev, |
5224 | struct list_head *dev_list) | |
5225 | { | |
5226 | char linkname[IFNAMSIZ+7]; | |
5227 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
5228 | "upper_%s" : "lower_%s", adj_dev->name); | |
5229 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
5230 | linkname); | |
5231 | } | |
0a59f3a9 | 5232 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
5233 | char *name, |
5234 | struct list_head *dev_list) | |
5235 | { | |
5236 | char linkname[IFNAMSIZ+7]; | |
5237 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
5238 | "upper_%s" : "lower_%s", name); | |
5239 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
5240 | } | |
5241 | ||
7ce64c79 AF |
5242 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
5243 | struct net_device *adj_dev, | |
5244 | struct list_head *dev_list) | |
5245 | { | |
5246 | return (dev_list == &dev->adj_list.upper || | |
5247 | dev_list == &dev->adj_list.lower) && | |
5248 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
5249 | } | |
3ee32707 | 5250 | |
5d261913 VF |
5251 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
5252 | struct net_device *adj_dev, | |
7863c054 | 5253 | struct list_head *dev_list, |
402dae96 | 5254 | void *private, bool master) |
5d261913 VF |
5255 | { |
5256 | struct netdev_adjacent *adj; | |
842d67a7 | 5257 | int ret; |
5d261913 | 5258 | |
6ea29da1 | 5259 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
5260 | |
5261 | if (adj) { | |
5d261913 VF |
5262 | adj->ref_nr++; |
5263 | return 0; | |
5264 | } | |
5265 | ||
5266 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
5267 | if (!adj) | |
5268 | return -ENOMEM; | |
5269 | ||
5270 | adj->dev = adj_dev; | |
5271 | adj->master = master; | |
5d261913 | 5272 | adj->ref_nr = 1; |
402dae96 | 5273 | adj->private = private; |
5d261913 | 5274 | dev_hold(adj_dev); |
2f268f12 VF |
5275 | |
5276 | pr_debug("dev_hold for %s, because of link added from %s to %s\n", | |
5277 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 | 5278 | |
7ce64c79 | 5279 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 5280 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
5281 | if (ret) |
5282 | goto free_adj; | |
5283 | } | |
5284 | ||
7863c054 | 5285 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
5286 | if (master) { |
5287 | ret = sysfs_create_link(&(dev->dev.kobj), | |
5288 | &(adj_dev->dev.kobj), "master"); | |
5289 | if (ret) | |
5831d66e | 5290 | goto remove_symlinks; |
842d67a7 | 5291 | |
7863c054 | 5292 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 5293 | } else { |
7863c054 | 5294 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 5295 | } |
5d261913 VF |
5296 | |
5297 | return 0; | |
842d67a7 | 5298 | |
5831d66e | 5299 | remove_symlinks: |
7ce64c79 | 5300 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5301 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
5302 | free_adj: |
5303 | kfree(adj); | |
974daef7 | 5304 | dev_put(adj_dev); |
842d67a7 VF |
5305 | |
5306 | return ret; | |
5d261913 VF |
5307 | } |
5308 | ||
1d143d9f | 5309 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
5310 | struct net_device *adj_dev, | |
5311 | struct list_head *dev_list) | |
5d261913 VF |
5312 | { |
5313 | struct netdev_adjacent *adj; | |
5314 | ||
6ea29da1 | 5315 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 5316 | |
2f268f12 VF |
5317 | if (!adj) { |
5318 | pr_err("tried to remove device %s from %s\n", | |
5319 | dev->name, adj_dev->name); | |
5d261913 | 5320 | BUG(); |
2f268f12 | 5321 | } |
5d261913 VF |
5322 | |
5323 | if (adj->ref_nr > 1) { | |
2f268f12 VF |
5324 | pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name, |
5325 | adj->ref_nr-1); | |
5d261913 VF |
5326 | adj->ref_nr--; |
5327 | return; | |
5328 | } | |
5329 | ||
842d67a7 VF |
5330 | if (adj->master) |
5331 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
5332 | ||
7ce64c79 | 5333 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5334 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 5335 | |
5d261913 | 5336 | list_del_rcu(&adj->list); |
2f268f12 VF |
5337 | pr_debug("dev_put for %s, because link removed from %s to %s\n", |
5338 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 VF |
5339 | dev_put(adj_dev); |
5340 | kfree_rcu(adj, rcu); | |
5341 | } | |
5342 | ||
1d143d9f | 5343 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
5344 | struct net_device *upper_dev, | |
5345 | struct list_head *up_list, | |
5346 | struct list_head *down_list, | |
5347 | void *private, bool master) | |
5d261913 VF |
5348 | { |
5349 | int ret; | |
5350 | ||
402dae96 VF |
5351 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private, |
5352 | master); | |
5d261913 VF |
5353 | if (ret) |
5354 | return ret; | |
5355 | ||
402dae96 VF |
5356 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private, |
5357 | false); | |
5d261913 | 5358 | if (ret) { |
2f268f12 | 5359 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5d261913 VF |
5360 | return ret; |
5361 | } | |
5362 | ||
5363 | return 0; | |
5364 | } | |
5365 | ||
1d143d9f | 5366 | static int __netdev_adjacent_dev_link(struct net_device *dev, |
5367 | struct net_device *upper_dev) | |
5d261913 | 5368 | { |
2f268f12 VF |
5369 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
5370 | &dev->all_adj_list.upper, | |
5371 | &upper_dev->all_adj_list.lower, | |
402dae96 | 5372 | NULL, false); |
5d261913 VF |
5373 | } |
5374 | ||
1d143d9f | 5375 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
5376 | struct net_device *upper_dev, | |
5377 | struct list_head *up_list, | |
5378 | struct list_head *down_list) | |
5d261913 | 5379 | { |
2f268f12 VF |
5380 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5381 | __netdev_adjacent_dev_remove(upper_dev, dev, down_list); | |
5d261913 VF |
5382 | } |
5383 | ||
1d143d9f | 5384 | static void __netdev_adjacent_dev_unlink(struct net_device *dev, |
5385 | struct net_device *upper_dev) | |
5d261913 | 5386 | { |
2f268f12 VF |
5387 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, |
5388 | &dev->all_adj_list.upper, | |
5389 | &upper_dev->all_adj_list.lower); | |
5390 | } | |
5391 | ||
1d143d9f | 5392 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
5393 | struct net_device *upper_dev, | |
5394 | void *private, bool master) | |
2f268f12 VF |
5395 | { |
5396 | int ret = __netdev_adjacent_dev_link(dev, upper_dev); | |
5397 | ||
5398 | if (ret) | |
5399 | return ret; | |
5400 | ||
5401 | ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, | |
5402 | &dev->adj_list.upper, | |
5403 | &upper_dev->adj_list.lower, | |
402dae96 | 5404 | private, master); |
2f268f12 VF |
5405 | if (ret) { |
5406 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5407 | return ret; | |
5408 | } | |
5409 | ||
5410 | return 0; | |
5d261913 VF |
5411 | } |
5412 | ||
1d143d9f | 5413 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
5414 | struct net_device *upper_dev) | |
2f268f12 VF |
5415 | { |
5416 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5417 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, | |
5418 | &dev->adj_list.upper, | |
5419 | &upper_dev->adj_list.lower); | |
5420 | } | |
5d261913 | 5421 | |
9ff162a8 | 5422 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 5423 | struct net_device *upper_dev, bool master, |
6dffb044 | 5424 | void *upper_priv) |
9ff162a8 | 5425 | { |
0e4ead9d | 5426 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 VF |
5427 | struct netdev_adjacent *i, *j, *to_i, *to_j; |
5428 | int ret = 0; | |
9ff162a8 JP |
5429 | |
5430 | ASSERT_RTNL(); | |
5431 | ||
5432 | if (dev == upper_dev) | |
5433 | return -EBUSY; | |
5434 | ||
5435 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
6ea29da1 | 5436 | if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper)) |
9ff162a8 JP |
5437 | return -EBUSY; |
5438 | ||
6ea29da1 | 5439 | if (__netdev_find_adj(upper_dev, &dev->adj_list.upper)) |
9ff162a8 JP |
5440 | return -EEXIST; |
5441 | ||
5442 | if (master && netdev_master_upper_dev_get(dev)) | |
5443 | return -EBUSY; | |
5444 | ||
0e4ead9d JP |
5445 | changeupper_info.upper_dev = upper_dev; |
5446 | changeupper_info.master = master; | |
5447 | changeupper_info.linking = true; | |
5448 | ||
573c7ba0 JP |
5449 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
5450 | &changeupper_info.info); | |
5451 | ret = notifier_to_errno(ret); | |
5452 | if (ret) | |
5453 | return ret; | |
5454 | ||
6dffb044 | 5455 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 5456 | master); |
5d261913 VF |
5457 | if (ret) |
5458 | return ret; | |
9ff162a8 | 5459 | |
5d261913 | 5460 | /* Now that we linked these devs, make all the upper_dev's |
2f268f12 | 5461 | * all_adj_list.upper visible to every dev's all_adj_list.lower an |
5d261913 VF |
5462 | * versa, and don't forget the devices itself. All of these |
5463 | * links are non-neighbours. | |
5464 | */ | |
2f268f12 VF |
5465 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5466 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5467 | pr_debug("Interlinking %s with %s, non-neighbour\n", | |
5468 | i->dev->name, j->dev->name); | |
5d261913 VF |
5469 | ret = __netdev_adjacent_dev_link(i->dev, j->dev); |
5470 | if (ret) | |
5471 | goto rollback_mesh; | |
5472 | } | |
5473 | } | |
5474 | ||
5475 | /* add dev to every upper_dev's upper device */ | |
2f268f12 VF |
5476 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5477 | pr_debug("linking %s's upper device %s with %s\n", | |
5478 | upper_dev->name, i->dev->name, dev->name); | |
5d261913 VF |
5479 | ret = __netdev_adjacent_dev_link(dev, i->dev); |
5480 | if (ret) | |
5481 | goto rollback_upper_mesh; | |
5482 | } | |
5483 | ||
5484 | /* add upper_dev to every dev's lower device */ | |
2f268f12 VF |
5485 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5486 | pr_debug("linking %s's lower device %s with %s\n", dev->name, | |
5487 | i->dev->name, upper_dev->name); | |
5d261913 VF |
5488 | ret = __netdev_adjacent_dev_link(i->dev, upper_dev); |
5489 | if (ret) | |
5490 | goto rollback_lower_mesh; | |
5491 | } | |
9ff162a8 | 5492 | |
b03804e7 IS |
5493 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
5494 | &changeupper_info.info); | |
5495 | ret = notifier_to_errno(ret); | |
5496 | if (ret) | |
5497 | goto rollback_lower_mesh; | |
5498 | ||
9ff162a8 | 5499 | return 0; |
5d261913 VF |
5500 | |
5501 | rollback_lower_mesh: | |
5502 | to_i = i; | |
2f268f12 | 5503 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5d261913 VF |
5504 | if (i == to_i) |
5505 | break; | |
5506 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); | |
5507 | } | |
5508 | ||
5509 | i = NULL; | |
5510 | ||
5511 | rollback_upper_mesh: | |
5512 | to_i = i; | |
2f268f12 | 5513 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5d261913 VF |
5514 | if (i == to_i) |
5515 | break; | |
5516 | __netdev_adjacent_dev_unlink(dev, i->dev); | |
5517 | } | |
5518 | ||
5519 | i = j = NULL; | |
5520 | ||
5521 | rollback_mesh: | |
5522 | to_i = i; | |
5523 | to_j = j; | |
2f268f12 VF |
5524 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5525 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5d261913 VF |
5526 | if (i == to_i && j == to_j) |
5527 | break; | |
5528 | __netdev_adjacent_dev_unlink(i->dev, j->dev); | |
5529 | } | |
5530 | if (i == to_i) | |
5531 | break; | |
5532 | } | |
5533 | ||
2f268f12 | 5534 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5535 | |
5536 | return ret; | |
9ff162a8 JP |
5537 | } |
5538 | ||
5539 | /** | |
5540 | * netdev_upper_dev_link - Add a link to the upper device | |
5541 | * @dev: device | |
5542 | * @upper_dev: new upper device | |
5543 | * | |
5544 | * Adds a link to device which is upper to this one. The caller must hold | |
5545 | * the RTNL lock. On a failure a negative errno code is returned. | |
5546 | * On success the reference counts are adjusted and the function | |
5547 | * returns zero. | |
5548 | */ | |
5549 | int netdev_upper_dev_link(struct net_device *dev, | |
5550 | struct net_device *upper_dev) | |
5551 | { | |
402dae96 | 5552 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL); |
9ff162a8 JP |
5553 | } |
5554 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
5555 | ||
5556 | /** | |
5557 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
5558 | * @dev: device | |
5559 | * @upper_dev: new upper device | |
6dffb044 | 5560 | * @upper_priv: upper device private |
9ff162a8 JP |
5561 | * |
5562 | * Adds a link to device which is upper to this one. In this case, only | |
5563 | * one master upper device can be linked, although other non-master devices | |
5564 | * might be linked as well. The caller must hold the RTNL lock. | |
5565 | * On a failure a negative errno code is returned. On success the reference | |
5566 | * counts are adjusted and the function returns zero. | |
5567 | */ | |
5568 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 JP |
5569 | struct net_device *upper_dev, |
5570 | void *upper_priv) | |
9ff162a8 | 5571 | { |
6dffb044 | 5572 | return __netdev_upper_dev_link(dev, upper_dev, true, upper_priv); |
9ff162a8 JP |
5573 | } |
5574 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
5575 | ||
5576 | /** | |
5577 | * netdev_upper_dev_unlink - Removes a link to upper device | |
5578 | * @dev: device | |
5579 | * @upper_dev: new upper device | |
5580 | * | |
5581 | * Removes a link to device which is upper to this one. The caller must hold | |
5582 | * the RTNL lock. | |
5583 | */ | |
5584 | void netdev_upper_dev_unlink(struct net_device *dev, | |
5585 | struct net_device *upper_dev) | |
5586 | { | |
0e4ead9d | 5587 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 | 5588 | struct netdev_adjacent *i, *j; |
9ff162a8 JP |
5589 | ASSERT_RTNL(); |
5590 | ||
0e4ead9d JP |
5591 | changeupper_info.upper_dev = upper_dev; |
5592 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; | |
5593 | changeupper_info.linking = false; | |
5594 | ||
573c7ba0 JP |
5595 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
5596 | &changeupper_info.info); | |
5597 | ||
2f268f12 | 5598 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5599 | |
5600 | /* Here is the tricky part. We must remove all dev's lower | |
5601 | * devices from all upper_dev's upper devices and vice | |
5602 | * versa, to maintain the graph relationship. | |
5603 | */ | |
2f268f12 VF |
5604 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5605 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) | |
5d261913 VF |
5606 | __netdev_adjacent_dev_unlink(i->dev, j->dev); |
5607 | ||
5608 | /* remove also the devices itself from lower/upper device | |
5609 | * list | |
5610 | */ | |
2f268f12 | 5611 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5d261913 VF |
5612 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); |
5613 | ||
2f268f12 | 5614 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) |
5d261913 VF |
5615 | __netdev_adjacent_dev_unlink(dev, i->dev); |
5616 | ||
0e4ead9d JP |
5617 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
5618 | &changeupper_info.info); | |
9ff162a8 JP |
5619 | } |
5620 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
5621 | ||
61bd3857 MS |
5622 | /** |
5623 | * netdev_bonding_info_change - Dispatch event about slave change | |
5624 | * @dev: device | |
4a26e453 | 5625 | * @bonding_info: info to dispatch |
61bd3857 MS |
5626 | * |
5627 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
5628 | * The caller must hold the RTNL lock. | |
5629 | */ | |
5630 | void netdev_bonding_info_change(struct net_device *dev, | |
5631 | struct netdev_bonding_info *bonding_info) | |
5632 | { | |
5633 | struct netdev_notifier_bonding_info info; | |
5634 | ||
5635 | memcpy(&info.bonding_info, bonding_info, | |
5636 | sizeof(struct netdev_bonding_info)); | |
5637 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, | |
5638 | &info.info); | |
5639 | } | |
5640 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
5641 | ||
2ce1ee17 | 5642 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
5643 | { |
5644 | struct netdev_adjacent *iter; | |
5645 | ||
5646 | struct net *net = dev_net(dev); | |
5647 | ||
5648 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
5649 | if (!net_eq(net,dev_net(iter->dev))) | |
5650 | continue; | |
5651 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5652 | &iter->dev->adj_list.lower); | |
5653 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5654 | &dev->adj_list.upper); | |
5655 | } | |
5656 | ||
5657 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
5658 | if (!net_eq(net,dev_net(iter->dev))) | |
5659 | continue; | |
5660 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5661 | &iter->dev->adj_list.upper); | |
5662 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5663 | &dev->adj_list.lower); | |
5664 | } | |
5665 | } | |
5666 | ||
2ce1ee17 | 5667 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
5668 | { |
5669 | struct netdev_adjacent *iter; | |
5670 | ||
5671 | struct net *net = dev_net(dev); | |
5672 | ||
5673 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
5674 | if (!net_eq(net,dev_net(iter->dev))) | |
5675 | continue; | |
5676 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5677 | &iter->dev->adj_list.lower); | |
5678 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5679 | &dev->adj_list.upper); | |
5680 | } | |
5681 | ||
5682 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
5683 | if (!net_eq(net,dev_net(iter->dev))) | |
5684 | continue; | |
5685 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5686 | &iter->dev->adj_list.upper); | |
5687 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5688 | &dev->adj_list.lower); | |
5689 | } | |
5690 | } | |
5691 | ||
5bb025fa | 5692 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 5693 | { |
5bb025fa | 5694 | struct netdev_adjacent *iter; |
402dae96 | 5695 | |
4c75431a AF |
5696 | struct net *net = dev_net(dev); |
5697 | ||
5bb025fa | 5698 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
4c75431a AF |
5699 | if (!net_eq(net,dev_net(iter->dev))) |
5700 | continue; | |
5bb025fa VF |
5701 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
5702 | &iter->dev->adj_list.lower); | |
5703 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5704 | &iter->dev->adj_list.lower); | |
5705 | } | |
402dae96 | 5706 | |
5bb025fa | 5707 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
4c75431a AF |
5708 | if (!net_eq(net,dev_net(iter->dev))) |
5709 | continue; | |
5bb025fa VF |
5710 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
5711 | &iter->dev->adj_list.upper); | |
5712 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5713 | &iter->dev->adj_list.upper); | |
5714 | } | |
402dae96 | 5715 | } |
402dae96 VF |
5716 | |
5717 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
5718 | struct net_device *lower_dev) | |
5719 | { | |
5720 | struct netdev_adjacent *lower; | |
5721 | ||
5722 | if (!lower_dev) | |
5723 | return NULL; | |
6ea29da1 | 5724 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
5725 | if (!lower) |
5726 | return NULL; | |
5727 | ||
5728 | return lower->private; | |
5729 | } | |
5730 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
5731 | ||
4085ebe8 VY |
5732 | |
5733 | int dev_get_nest_level(struct net_device *dev, | |
5734 | bool (*type_check)(struct net_device *dev)) | |
5735 | { | |
5736 | struct net_device *lower = NULL; | |
5737 | struct list_head *iter; | |
5738 | int max_nest = -1; | |
5739 | int nest; | |
5740 | ||
5741 | ASSERT_RTNL(); | |
5742 | ||
5743 | netdev_for_each_lower_dev(dev, lower, iter) { | |
5744 | nest = dev_get_nest_level(lower, type_check); | |
5745 | if (max_nest < nest) | |
5746 | max_nest = nest; | |
5747 | } | |
5748 | ||
5749 | if (type_check(dev)) | |
5750 | max_nest++; | |
5751 | ||
5752 | return max_nest; | |
5753 | } | |
5754 | EXPORT_SYMBOL(dev_get_nest_level); | |
5755 | ||
b6c40d68 PM |
5756 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
5757 | { | |
d314774c SH |
5758 | const struct net_device_ops *ops = dev->netdev_ops; |
5759 | ||
d2615bf4 | 5760 | if (ops->ndo_change_rx_flags) |
d314774c | 5761 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
5762 | } |
5763 | ||
991fb3f7 | 5764 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5765 | { |
b536db93 | 5766 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
5767 | kuid_t uid; |
5768 | kgid_t gid; | |
1da177e4 | 5769 | |
24023451 PM |
5770 | ASSERT_RTNL(); |
5771 | ||
dad9b335 WC |
5772 | dev->flags |= IFF_PROMISC; |
5773 | dev->promiscuity += inc; | |
5774 | if (dev->promiscuity == 0) { | |
5775 | /* | |
5776 | * Avoid overflow. | |
5777 | * If inc causes overflow, untouch promisc and return error. | |
5778 | */ | |
5779 | if (inc < 0) | |
5780 | dev->flags &= ~IFF_PROMISC; | |
5781 | else { | |
5782 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
5783 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
5784 | dev->name); | |
dad9b335 WC |
5785 | return -EOVERFLOW; |
5786 | } | |
5787 | } | |
52609c0b | 5788 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
5789 | pr_info("device %s %s promiscuous mode\n", |
5790 | dev->name, | |
5791 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
5792 | if (audit_enabled) { |
5793 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
5794 | audit_log(current->audit_context, GFP_ATOMIC, |
5795 | AUDIT_ANOM_PROMISCUOUS, | |
5796 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
5797 | dev->name, (dev->flags & IFF_PROMISC), | |
5798 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 5799 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
5800 | from_kuid(&init_user_ns, uid), |
5801 | from_kgid(&init_user_ns, gid), | |
7759db82 | 5802 | audit_get_sessionid(current)); |
8192b0c4 | 5803 | } |
24023451 | 5804 | |
b6c40d68 | 5805 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 5806 | } |
991fb3f7 ND |
5807 | if (notify) |
5808 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 5809 | return 0; |
1da177e4 LT |
5810 | } |
5811 | ||
4417da66 PM |
5812 | /** |
5813 | * dev_set_promiscuity - update promiscuity count on a device | |
5814 | * @dev: device | |
5815 | * @inc: modifier | |
5816 | * | |
5817 | * Add or remove promiscuity from a device. While the count in the device | |
5818 | * remains above zero the interface remains promiscuous. Once it hits zero | |
5819 | * the device reverts back to normal filtering operation. A negative inc | |
5820 | * value is used to drop promiscuity on the device. | |
dad9b335 | 5821 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 5822 | */ |
dad9b335 | 5823 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 5824 | { |
b536db93 | 5825 | unsigned int old_flags = dev->flags; |
dad9b335 | 5826 | int err; |
4417da66 | 5827 | |
991fb3f7 | 5828 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 5829 | if (err < 0) |
dad9b335 | 5830 | return err; |
4417da66 PM |
5831 | if (dev->flags != old_flags) |
5832 | dev_set_rx_mode(dev); | |
dad9b335 | 5833 | return err; |
4417da66 | 5834 | } |
d1b19dff | 5835 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 5836 | |
991fb3f7 | 5837 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5838 | { |
991fb3f7 | 5839 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 5840 | |
24023451 PM |
5841 | ASSERT_RTNL(); |
5842 | ||
1da177e4 | 5843 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
5844 | dev->allmulti += inc; |
5845 | if (dev->allmulti == 0) { | |
5846 | /* | |
5847 | * Avoid overflow. | |
5848 | * If inc causes overflow, untouch allmulti and return error. | |
5849 | */ | |
5850 | if (inc < 0) | |
5851 | dev->flags &= ~IFF_ALLMULTI; | |
5852 | else { | |
5853 | dev->allmulti -= inc; | |
7b6cd1ce JP |
5854 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
5855 | dev->name); | |
dad9b335 WC |
5856 | return -EOVERFLOW; |
5857 | } | |
5858 | } | |
24023451 | 5859 | if (dev->flags ^ old_flags) { |
b6c40d68 | 5860 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 5861 | dev_set_rx_mode(dev); |
991fb3f7 ND |
5862 | if (notify) |
5863 | __dev_notify_flags(dev, old_flags, | |
5864 | dev->gflags ^ old_gflags); | |
24023451 | 5865 | } |
dad9b335 | 5866 | return 0; |
4417da66 | 5867 | } |
991fb3f7 ND |
5868 | |
5869 | /** | |
5870 | * dev_set_allmulti - update allmulti count on a device | |
5871 | * @dev: device | |
5872 | * @inc: modifier | |
5873 | * | |
5874 | * Add or remove reception of all multicast frames to a device. While the | |
5875 | * count in the device remains above zero the interface remains listening | |
5876 | * to all interfaces. Once it hits zero the device reverts back to normal | |
5877 | * filtering operation. A negative @inc value is used to drop the counter | |
5878 | * when releasing a resource needing all multicasts. | |
5879 | * Return 0 if successful or a negative errno code on error. | |
5880 | */ | |
5881 | ||
5882 | int dev_set_allmulti(struct net_device *dev, int inc) | |
5883 | { | |
5884 | return __dev_set_allmulti(dev, inc, true); | |
5885 | } | |
d1b19dff | 5886 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
5887 | |
5888 | /* | |
5889 | * Upload unicast and multicast address lists to device and | |
5890 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 5891 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
5892 | * are present. |
5893 | */ | |
5894 | void __dev_set_rx_mode(struct net_device *dev) | |
5895 | { | |
d314774c SH |
5896 | const struct net_device_ops *ops = dev->netdev_ops; |
5897 | ||
4417da66 PM |
5898 | /* dev_open will call this function so the list will stay sane. */ |
5899 | if (!(dev->flags&IFF_UP)) | |
5900 | return; | |
5901 | ||
5902 | if (!netif_device_present(dev)) | |
40b77c94 | 5903 | return; |
4417da66 | 5904 | |
01789349 | 5905 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
5906 | /* Unicast addresses changes may only happen under the rtnl, |
5907 | * therefore calling __dev_set_promiscuity here is safe. | |
5908 | */ | |
32e7bfc4 | 5909 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 5910 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 5911 | dev->uc_promisc = true; |
32e7bfc4 | 5912 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 5913 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 5914 | dev->uc_promisc = false; |
4417da66 | 5915 | } |
4417da66 | 5916 | } |
01789349 JP |
5917 | |
5918 | if (ops->ndo_set_rx_mode) | |
5919 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
5920 | } |
5921 | ||
5922 | void dev_set_rx_mode(struct net_device *dev) | |
5923 | { | |
b9e40857 | 5924 | netif_addr_lock_bh(dev); |
4417da66 | 5925 | __dev_set_rx_mode(dev); |
b9e40857 | 5926 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
5927 | } |
5928 | ||
f0db275a SH |
5929 | /** |
5930 | * dev_get_flags - get flags reported to userspace | |
5931 | * @dev: device | |
5932 | * | |
5933 | * Get the combination of flag bits exported through APIs to userspace. | |
5934 | */ | |
95c96174 | 5935 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 5936 | { |
95c96174 | 5937 | unsigned int flags; |
1da177e4 LT |
5938 | |
5939 | flags = (dev->flags & ~(IFF_PROMISC | | |
5940 | IFF_ALLMULTI | | |
b00055aa SR |
5941 | IFF_RUNNING | |
5942 | IFF_LOWER_UP | | |
5943 | IFF_DORMANT)) | | |
1da177e4 LT |
5944 | (dev->gflags & (IFF_PROMISC | |
5945 | IFF_ALLMULTI)); | |
5946 | ||
b00055aa SR |
5947 | if (netif_running(dev)) { |
5948 | if (netif_oper_up(dev)) | |
5949 | flags |= IFF_RUNNING; | |
5950 | if (netif_carrier_ok(dev)) | |
5951 | flags |= IFF_LOWER_UP; | |
5952 | if (netif_dormant(dev)) | |
5953 | flags |= IFF_DORMANT; | |
5954 | } | |
1da177e4 LT |
5955 | |
5956 | return flags; | |
5957 | } | |
d1b19dff | 5958 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 5959 | |
bd380811 | 5960 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 5961 | { |
b536db93 | 5962 | unsigned int old_flags = dev->flags; |
bd380811 | 5963 | int ret; |
1da177e4 | 5964 | |
24023451 PM |
5965 | ASSERT_RTNL(); |
5966 | ||
1da177e4 LT |
5967 | /* |
5968 | * Set the flags on our device. | |
5969 | */ | |
5970 | ||
5971 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
5972 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
5973 | IFF_AUTOMEDIA)) | | |
5974 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
5975 | IFF_ALLMULTI)); | |
5976 | ||
5977 | /* | |
5978 | * Load in the correct multicast list now the flags have changed. | |
5979 | */ | |
5980 | ||
b6c40d68 PM |
5981 | if ((old_flags ^ flags) & IFF_MULTICAST) |
5982 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 5983 | |
4417da66 | 5984 | dev_set_rx_mode(dev); |
1da177e4 LT |
5985 | |
5986 | /* | |
5987 | * Have we downed the interface. We handle IFF_UP ourselves | |
5988 | * according to user attempts to set it, rather than blindly | |
5989 | * setting it. | |
5990 | */ | |
5991 | ||
5992 | ret = 0; | |
d215d10f | 5993 | if ((old_flags ^ flags) & IFF_UP) |
bd380811 | 5994 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 | 5995 | |
1da177e4 | 5996 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 5997 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 5998 | unsigned int old_flags = dev->flags; |
d1b19dff | 5999 | |
1da177e4 | 6000 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
6001 | |
6002 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
6003 | if (dev->flags != old_flags) | |
6004 | dev_set_rx_mode(dev); | |
1da177e4 LT |
6005 | } |
6006 | ||
6007 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
6008 | is important. Some (broken) drivers set IFF_PROMISC, when | |
6009 | IFF_ALLMULTI is requested not asking us and not reporting. | |
6010 | */ | |
6011 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
6012 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
6013 | ||
1da177e4 | 6014 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 6015 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
6016 | } |
6017 | ||
bd380811 PM |
6018 | return ret; |
6019 | } | |
6020 | ||
a528c219 ND |
6021 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
6022 | unsigned int gchanges) | |
bd380811 PM |
6023 | { |
6024 | unsigned int changes = dev->flags ^ old_flags; | |
6025 | ||
a528c219 | 6026 | if (gchanges) |
7f294054 | 6027 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 6028 | |
bd380811 PM |
6029 | if (changes & IFF_UP) { |
6030 | if (dev->flags & IFF_UP) | |
6031 | call_netdevice_notifiers(NETDEV_UP, dev); | |
6032 | else | |
6033 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
6034 | } | |
6035 | ||
6036 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
6037 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
6038 | struct netdev_notifier_change_info change_info; | |
6039 | ||
6040 | change_info.flags_changed = changes; | |
6041 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
6042 | &change_info.info); | |
6043 | } | |
bd380811 PM |
6044 | } |
6045 | ||
6046 | /** | |
6047 | * dev_change_flags - change device settings | |
6048 | * @dev: device | |
6049 | * @flags: device state flags | |
6050 | * | |
6051 | * Change settings on device based state flags. The flags are | |
6052 | * in the userspace exported format. | |
6053 | */ | |
b536db93 | 6054 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 6055 | { |
b536db93 | 6056 | int ret; |
991fb3f7 | 6057 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
6058 | |
6059 | ret = __dev_change_flags(dev, flags); | |
6060 | if (ret < 0) | |
6061 | return ret; | |
6062 | ||
991fb3f7 | 6063 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 6064 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
6065 | return ret; |
6066 | } | |
d1b19dff | 6067 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 6068 | |
2315dc91 VF |
6069 | static int __dev_set_mtu(struct net_device *dev, int new_mtu) |
6070 | { | |
6071 | const struct net_device_ops *ops = dev->netdev_ops; | |
6072 | ||
6073 | if (ops->ndo_change_mtu) | |
6074 | return ops->ndo_change_mtu(dev, new_mtu); | |
6075 | ||
6076 | dev->mtu = new_mtu; | |
6077 | return 0; | |
6078 | } | |
6079 | ||
f0db275a SH |
6080 | /** |
6081 | * dev_set_mtu - Change maximum transfer unit | |
6082 | * @dev: device | |
6083 | * @new_mtu: new transfer unit | |
6084 | * | |
6085 | * Change the maximum transfer size of the network device. | |
6086 | */ | |
1da177e4 LT |
6087 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
6088 | { | |
2315dc91 | 6089 | int err, orig_mtu; |
1da177e4 LT |
6090 | |
6091 | if (new_mtu == dev->mtu) | |
6092 | return 0; | |
6093 | ||
6094 | /* MTU must be positive. */ | |
6095 | if (new_mtu < 0) | |
6096 | return -EINVAL; | |
6097 | ||
6098 | if (!netif_device_present(dev)) | |
6099 | return -ENODEV; | |
6100 | ||
1d486bfb VF |
6101 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
6102 | err = notifier_to_errno(err); | |
6103 | if (err) | |
6104 | return err; | |
d314774c | 6105 | |
2315dc91 VF |
6106 | orig_mtu = dev->mtu; |
6107 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 6108 | |
2315dc91 VF |
6109 | if (!err) { |
6110 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6111 | err = notifier_to_errno(err); | |
6112 | if (err) { | |
6113 | /* setting mtu back and notifying everyone again, | |
6114 | * so that they have a chance to revert changes. | |
6115 | */ | |
6116 | __dev_set_mtu(dev, orig_mtu); | |
6117 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6118 | } | |
6119 | } | |
1da177e4 LT |
6120 | return err; |
6121 | } | |
d1b19dff | 6122 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 6123 | |
cbda10fa VD |
6124 | /** |
6125 | * dev_set_group - Change group this device belongs to | |
6126 | * @dev: device | |
6127 | * @new_group: group this device should belong to | |
6128 | */ | |
6129 | void dev_set_group(struct net_device *dev, int new_group) | |
6130 | { | |
6131 | dev->group = new_group; | |
6132 | } | |
6133 | EXPORT_SYMBOL(dev_set_group); | |
6134 | ||
f0db275a SH |
6135 | /** |
6136 | * dev_set_mac_address - Change Media Access Control Address | |
6137 | * @dev: device | |
6138 | * @sa: new address | |
6139 | * | |
6140 | * Change the hardware (MAC) address of the device | |
6141 | */ | |
1da177e4 LT |
6142 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
6143 | { | |
d314774c | 6144 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
6145 | int err; |
6146 | ||
d314774c | 6147 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
6148 | return -EOPNOTSUPP; |
6149 | if (sa->sa_family != dev->type) | |
6150 | return -EINVAL; | |
6151 | if (!netif_device_present(dev)) | |
6152 | return -ENODEV; | |
d314774c | 6153 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
6154 | if (err) |
6155 | return err; | |
fbdeca2d | 6156 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 6157 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 6158 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 6159 | return 0; |
1da177e4 | 6160 | } |
d1b19dff | 6161 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 6162 | |
4bf84c35 JP |
6163 | /** |
6164 | * dev_change_carrier - Change device carrier | |
6165 | * @dev: device | |
691b3b7e | 6166 | * @new_carrier: new value |
4bf84c35 JP |
6167 | * |
6168 | * Change device carrier | |
6169 | */ | |
6170 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
6171 | { | |
6172 | const struct net_device_ops *ops = dev->netdev_ops; | |
6173 | ||
6174 | if (!ops->ndo_change_carrier) | |
6175 | return -EOPNOTSUPP; | |
6176 | if (!netif_device_present(dev)) | |
6177 | return -ENODEV; | |
6178 | return ops->ndo_change_carrier(dev, new_carrier); | |
6179 | } | |
6180 | EXPORT_SYMBOL(dev_change_carrier); | |
6181 | ||
66b52b0d JP |
6182 | /** |
6183 | * dev_get_phys_port_id - Get device physical port ID | |
6184 | * @dev: device | |
6185 | * @ppid: port ID | |
6186 | * | |
6187 | * Get device physical port ID | |
6188 | */ | |
6189 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 6190 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
6191 | { |
6192 | const struct net_device_ops *ops = dev->netdev_ops; | |
6193 | ||
6194 | if (!ops->ndo_get_phys_port_id) | |
6195 | return -EOPNOTSUPP; | |
6196 | return ops->ndo_get_phys_port_id(dev, ppid); | |
6197 | } | |
6198 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
6199 | ||
db24a904 DA |
6200 | /** |
6201 | * dev_get_phys_port_name - Get device physical port name | |
6202 | * @dev: device | |
6203 | * @name: port name | |
6204 | * | |
6205 | * Get device physical port name | |
6206 | */ | |
6207 | int dev_get_phys_port_name(struct net_device *dev, | |
6208 | char *name, size_t len) | |
6209 | { | |
6210 | const struct net_device_ops *ops = dev->netdev_ops; | |
6211 | ||
6212 | if (!ops->ndo_get_phys_port_name) | |
6213 | return -EOPNOTSUPP; | |
6214 | return ops->ndo_get_phys_port_name(dev, name, len); | |
6215 | } | |
6216 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
6217 | ||
d746d707 AK |
6218 | /** |
6219 | * dev_change_proto_down - update protocol port state information | |
6220 | * @dev: device | |
6221 | * @proto_down: new value | |
6222 | * | |
6223 | * This info can be used by switch drivers to set the phys state of the | |
6224 | * port. | |
6225 | */ | |
6226 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
6227 | { | |
6228 | const struct net_device_ops *ops = dev->netdev_ops; | |
6229 | ||
6230 | if (!ops->ndo_change_proto_down) | |
6231 | return -EOPNOTSUPP; | |
6232 | if (!netif_device_present(dev)) | |
6233 | return -ENODEV; | |
6234 | return ops->ndo_change_proto_down(dev, proto_down); | |
6235 | } | |
6236 | EXPORT_SYMBOL(dev_change_proto_down); | |
6237 | ||
1da177e4 LT |
6238 | /** |
6239 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 6240 | * @net: the applicable net namespace |
1da177e4 LT |
6241 | * |
6242 | * Returns a suitable unique value for a new device interface | |
6243 | * number. The caller must hold the rtnl semaphore or the | |
6244 | * dev_base_lock to be sure it remains unique. | |
6245 | */ | |
881d966b | 6246 | static int dev_new_index(struct net *net) |
1da177e4 | 6247 | { |
aa79e66e | 6248 | int ifindex = net->ifindex; |
1da177e4 LT |
6249 | for (;;) { |
6250 | if (++ifindex <= 0) | |
6251 | ifindex = 1; | |
881d966b | 6252 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 6253 | return net->ifindex = ifindex; |
1da177e4 LT |
6254 | } |
6255 | } | |
6256 | ||
1da177e4 | 6257 | /* Delayed registration/unregisteration */ |
3b5b34fd | 6258 | static LIST_HEAD(net_todo_list); |
200b916f | 6259 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 6260 | |
6f05f629 | 6261 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 6262 | { |
1da177e4 | 6263 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 6264 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
6265 | } |
6266 | ||
9b5e383c | 6267 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 6268 | { |
e93737b0 | 6269 | struct net_device *dev, *tmp; |
5cde2829 | 6270 | LIST_HEAD(close_head); |
9b5e383c | 6271 | |
93ee31f1 DL |
6272 | BUG_ON(dev_boot_phase); |
6273 | ASSERT_RTNL(); | |
6274 | ||
e93737b0 | 6275 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 6276 | /* Some devices call without registering |
e93737b0 KK |
6277 | * for initialization unwind. Remove those |
6278 | * devices and proceed with the remaining. | |
9b5e383c ED |
6279 | */ |
6280 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
6281 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
6282 | dev->name, dev); | |
93ee31f1 | 6283 | |
9b5e383c | 6284 | WARN_ON(1); |
e93737b0 KK |
6285 | list_del(&dev->unreg_list); |
6286 | continue; | |
9b5e383c | 6287 | } |
449f4544 | 6288 | dev->dismantle = true; |
9b5e383c | 6289 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 6290 | } |
93ee31f1 | 6291 | |
44345724 | 6292 | /* If device is running, close it first. */ |
5cde2829 EB |
6293 | list_for_each_entry(dev, head, unreg_list) |
6294 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 6295 | dev_close_many(&close_head, true); |
93ee31f1 | 6296 | |
44345724 | 6297 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
6298 | /* And unlink it from device chain. */ |
6299 | unlist_netdevice(dev); | |
93ee31f1 | 6300 | |
9b5e383c | 6301 | dev->reg_state = NETREG_UNREGISTERING; |
e9e4dd32 | 6302 | on_each_cpu(flush_backlog, dev, 1); |
9b5e383c | 6303 | } |
93ee31f1 DL |
6304 | |
6305 | synchronize_net(); | |
6306 | ||
9b5e383c | 6307 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
6308 | struct sk_buff *skb = NULL; |
6309 | ||
9b5e383c ED |
6310 | /* Shutdown queueing discipline. */ |
6311 | dev_shutdown(dev); | |
93ee31f1 DL |
6312 | |
6313 | ||
9b5e383c ED |
6314 | /* Notify protocols, that we are about to destroy |
6315 | this device. They should clean all the things. | |
6316 | */ | |
6317 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 6318 | |
395eea6c MB |
6319 | if (!dev->rtnl_link_ops || |
6320 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
6321 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, | |
6322 | GFP_KERNEL); | |
6323 | ||
9b5e383c ED |
6324 | /* |
6325 | * Flush the unicast and multicast chains | |
6326 | */ | |
a748ee24 | 6327 | dev_uc_flush(dev); |
22bedad3 | 6328 | dev_mc_flush(dev); |
93ee31f1 | 6329 | |
9b5e383c ED |
6330 | if (dev->netdev_ops->ndo_uninit) |
6331 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 6332 | |
395eea6c MB |
6333 | if (skb) |
6334 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 6335 | |
9ff162a8 JP |
6336 | /* Notifier chain MUST detach us all upper devices. */ |
6337 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 6338 | |
9b5e383c ED |
6339 | /* Remove entries from kobject tree */ |
6340 | netdev_unregister_kobject(dev); | |
024e9679 AD |
6341 | #ifdef CONFIG_XPS |
6342 | /* Remove XPS queueing entries */ | |
6343 | netif_reset_xps_queues_gt(dev, 0); | |
6344 | #endif | |
9b5e383c | 6345 | } |
93ee31f1 | 6346 | |
850a545b | 6347 | synchronize_net(); |
395264d5 | 6348 | |
a5ee1551 | 6349 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
6350 | dev_put(dev); |
6351 | } | |
6352 | ||
6353 | static void rollback_registered(struct net_device *dev) | |
6354 | { | |
6355 | LIST_HEAD(single); | |
6356 | ||
6357 | list_add(&dev->unreg_list, &single); | |
6358 | rollback_registered_many(&single); | |
ceaaec98 | 6359 | list_del(&single); |
93ee31f1 DL |
6360 | } |
6361 | ||
fd867d51 JW |
6362 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
6363 | struct net_device *upper, netdev_features_t features) | |
6364 | { | |
6365 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
6366 | netdev_features_t feature; | |
5ba3f7d6 | 6367 | int feature_bit; |
fd867d51 | 6368 | |
5ba3f7d6 JW |
6369 | for_each_netdev_feature(&upper_disables, feature_bit) { |
6370 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
6371 | if (!(upper->wanted_features & feature) |
6372 | && (features & feature)) { | |
6373 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
6374 | &feature, upper->name); | |
6375 | features &= ~feature; | |
6376 | } | |
6377 | } | |
6378 | ||
6379 | return features; | |
6380 | } | |
6381 | ||
6382 | static void netdev_sync_lower_features(struct net_device *upper, | |
6383 | struct net_device *lower, netdev_features_t features) | |
6384 | { | |
6385 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
6386 | netdev_features_t feature; | |
5ba3f7d6 | 6387 | int feature_bit; |
fd867d51 | 6388 | |
5ba3f7d6 JW |
6389 | for_each_netdev_feature(&upper_disables, feature_bit) { |
6390 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
6391 | if (!(features & feature) && (lower->features & feature)) { |
6392 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
6393 | &feature, lower->name); | |
6394 | lower->wanted_features &= ~feature; | |
6395 | netdev_update_features(lower); | |
6396 | ||
6397 | if (unlikely(lower->features & feature)) | |
6398 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
6399 | &feature, lower->name); | |
6400 | } | |
6401 | } | |
6402 | } | |
6403 | ||
c8f44aff MM |
6404 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
6405 | netdev_features_t features) | |
b63365a2 | 6406 | { |
57422dc5 MM |
6407 | /* Fix illegal checksum combinations */ |
6408 | if ((features & NETIF_F_HW_CSUM) && | |
6409 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 6410 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
6411 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
6412 | } | |
6413 | ||
b63365a2 | 6414 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 6415 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 6416 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 6417 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
6418 | } |
6419 | ||
ec5f0615 PS |
6420 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
6421 | !(features & NETIF_F_IP_CSUM)) { | |
6422 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
6423 | features &= ~NETIF_F_TSO; | |
6424 | features &= ~NETIF_F_TSO_ECN; | |
6425 | } | |
6426 | ||
6427 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
6428 | !(features & NETIF_F_IPV6_CSUM)) { | |
6429 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
6430 | features &= ~NETIF_F_TSO6; | |
6431 | } | |
6432 | ||
31d8b9e0 BH |
6433 | /* TSO ECN requires that TSO is present as well. */ |
6434 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
6435 | features &= ~NETIF_F_TSO_ECN; | |
6436 | ||
212b573f MM |
6437 | /* Software GSO depends on SG. */ |
6438 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 6439 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
6440 | features &= ~NETIF_F_GSO; |
6441 | } | |
6442 | ||
acd1130e | 6443 | /* UFO needs SG and checksumming */ |
b63365a2 | 6444 | if (features & NETIF_F_UFO) { |
79032644 MM |
6445 | /* maybe split UFO into V4 and V6? */ |
6446 | if (!((features & NETIF_F_GEN_CSUM) || | |
6447 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
6448 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 6449 | netdev_dbg(dev, |
acd1130e | 6450 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
6451 | features &= ~NETIF_F_UFO; |
6452 | } | |
6453 | ||
6454 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 6455 | netdev_dbg(dev, |
acd1130e | 6456 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
6457 | features &= ~NETIF_F_UFO; |
6458 | } | |
6459 | } | |
6460 | ||
d0290214 JP |
6461 | #ifdef CONFIG_NET_RX_BUSY_POLL |
6462 | if (dev->netdev_ops->ndo_busy_poll) | |
6463 | features |= NETIF_F_BUSY_POLL; | |
6464 | else | |
6465 | #endif | |
6466 | features &= ~NETIF_F_BUSY_POLL; | |
6467 | ||
b63365a2 HX |
6468 | return features; |
6469 | } | |
b63365a2 | 6470 | |
6cb6a27c | 6471 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 6472 | { |
fd867d51 | 6473 | struct net_device *upper, *lower; |
c8f44aff | 6474 | netdev_features_t features; |
fd867d51 | 6475 | struct list_head *iter; |
e7868a85 | 6476 | int err = -1; |
5455c699 | 6477 | |
87267485 MM |
6478 | ASSERT_RTNL(); |
6479 | ||
5455c699 MM |
6480 | features = netdev_get_wanted_features(dev); |
6481 | ||
6482 | if (dev->netdev_ops->ndo_fix_features) | |
6483 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
6484 | ||
6485 | /* driver might be less strict about feature dependencies */ | |
6486 | features = netdev_fix_features(dev, features); | |
6487 | ||
fd867d51 JW |
6488 | /* some features can't be enabled if they're off an an upper device */ |
6489 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
6490 | features = netdev_sync_upper_features(dev, upper, features); | |
6491 | ||
5455c699 | 6492 | if (dev->features == features) |
e7868a85 | 6493 | goto sync_lower; |
5455c699 | 6494 | |
c8f44aff MM |
6495 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
6496 | &dev->features, &features); | |
5455c699 MM |
6497 | |
6498 | if (dev->netdev_ops->ndo_set_features) | |
6499 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
6500 | else |
6501 | err = 0; | |
5455c699 | 6502 | |
6cb6a27c | 6503 | if (unlikely(err < 0)) { |
5455c699 | 6504 | netdev_err(dev, |
c8f44aff MM |
6505 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
6506 | err, &features, &dev->features); | |
17b85d29 NA |
6507 | /* return non-0 since some features might have changed and |
6508 | * it's better to fire a spurious notification than miss it | |
6509 | */ | |
6510 | return -1; | |
6cb6a27c MM |
6511 | } |
6512 | ||
e7868a85 | 6513 | sync_lower: |
fd867d51 JW |
6514 | /* some features must be disabled on lower devices when disabled |
6515 | * on an upper device (think: bonding master or bridge) | |
6516 | */ | |
6517 | netdev_for_each_lower_dev(dev, lower, iter) | |
6518 | netdev_sync_lower_features(dev, lower, features); | |
6519 | ||
6cb6a27c MM |
6520 | if (!err) |
6521 | dev->features = features; | |
6522 | ||
e7868a85 | 6523 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
6524 | } |
6525 | ||
afe12cc8 MM |
6526 | /** |
6527 | * netdev_update_features - recalculate device features | |
6528 | * @dev: the device to check | |
6529 | * | |
6530 | * Recalculate dev->features set and send notifications if it | |
6531 | * has changed. Should be called after driver or hardware dependent | |
6532 | * conditions might have changed that influence the features. | |
6533 | */ | |
6cb6a27c MM |
6534 | void netdev_update_features(struct net_device *dev) |
6535 | { | |
6536 | if (__netdev_update_features(dev)) | |
6537 | netdev_features_change(dev); | |
5455c699 MM |
6538 | } |
6539 | EXPORT_SYMBOL(netdev_update_features); | |
6540 | ||
afe12cc8 MM |
6541 | /** |
6542 | * netdev_change_features - recalculate device features | |
6543 | * @dev: the device to check | |
6544 | * | |
6545 | * Recalculate dev->features set and send notifications even | |
6546 | * if they have not changed. Should be called instead of | |
6547 | * netdev_update_features() if also dev->vlan_features might | |
6548 | * have changed to allow the changes to be propagated to stacked | |
6549 | * VLAN devices. | |
6550 | */ | |
6551 | void netdev_change_features(struct net_device *dev) | |
6552 | { | |
6553 | __netdev_update_features(dev); | |
6554 | netdev_features_change(dev); | |
6555 | } | |
6556 | EXPORT_SYMBOL(netdev_change_features); | |
6557 | ||
fc4a7489 PM |
6558 | /** |
6559 | * netif_stacked_transfer_operstate - transfer operstate | |
6560 | * @rootdev: the root or lower level device to transfer state from | |
6561 | * @dev: the device to transfer operstate to | |
6562 | * | |
6563 | * Transfer operational state from root to device. This is normally | |
6564 | * called when a stacking relationship exists between the root | |
6565 | * device and the device(a leaf device). | |
6566 | */ | |
6567 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
6568 | struct net_device *dev) | |
6569 | { | |
6570 | if (rootdev->operstate == IF_OPER_DORMANT) | |
6571 | netif_dormant_on(dev); | |
6572 | else | |
6573 | netif_dormant_off(dev); | |
6574 | ||
6575 | if (netif_carrier_ok(rootdev)) { | |
6576 | if (!netif_carrier_ok(dev)) | |
6577 | netif_carrier_on(dev); | |
6578 | } else { | |
6579 | if (netif_carrier_ok(dev)) | |
6580 | netif_carrier_off(dev); | |
6581 | } | |
6582 | } | |
6583 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
6584 | ||
a953be53 | 6585 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
6586 | static int netif_alloc_rx_queues(struct net_device *dev) |
6587 | { | |
1b4bf461 | 6588 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 6589 | struct netdev_rx_queue *rx; |
10595902 | 6590 | size_t sz = count * sizeof(*rx); |
1b4bf461 | 6591 | |
bd25fa7b | 6592 | BUG_ON(count < 1); |
1b4bf461 | 6593 | |
10595902 PG |
6594 | rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6595 | if (!rx) { | |
6596 | rx = vzalloc(sz); | |
6597 | if (!rx) | |
6598 | return -ENOMEM; | |
6599 | } | |
bd25fa7b TH |
6600 | dev->_rx = rx; |
6601 | ||
bd25fa7b | 6602 | for (i = 0; i < count; i++) |
fe822240 | 6603 | rx[i].dev = dev; |
1b4bf461 ED |
6604 | return 0; |
6605 | } | |
bf264145 | 6606 | #endif |
1b4bf461 | 6607 | |
aa942104 CG |
6608 | static void netdev_init_one_queue(struct net_device *dev, |
6609 | struct netdev_queue *queue, void *_unused) | |
6610 | { | |
6611 | /* Initialize queue lock */ | |
6612 | spin_lock_init(&queue->_xmit_lock); | |
6613 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
6614 | queue->xmit_lock_owner = -1; | |
b236da69 | 6615 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 6616 | queue->dev = dev; |
114cf580 TH |
6617 | #ifdef CONFIG_BQL |
6618 | dql_init(&queue->dql, HZ); | |
6619 | #endif | |
aa942104 CG |
6620 | } |
6621 | ||
60877a32 ED |
6622 | static void netif_free_tx_queues(struct net_device *dev) |
6623 | { | |
4cb28970 | 6624 | kvfree(dev->_tx); |
60877a32 ED |
6625 | } |
6626 | ||
e6484930 TH |
6627 | static int netif_alloc_netdev_queues(struct net_device *dev) |
6628 | { | |
6629 | unsigned int count = dev->num_tx_queues; | |
6630 | struct netdev_queue *tx; | |
60877a32 | 6631 | size_t sz = count * sizeof(*tx); |
e6484930 | 6632 | |
d339727c ED |
6633 | if (count < 1 || count > 0xffff) |
6634 | return -EINVAL; | |
62b5942a | 6635 | |
60877a32 ED |
6636 | tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6637 | if (!tx) { | |
6638 | tx = vzalloc(sz); | |
6639 | if (!tx) | |
6640 | return -ENOMEM; | |
6641 | } | |
e6484930 | 6642 | dev->_tx = tx; |
1d24eb48 | 6643 | |
e6484930 TH |
6644 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
6645 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
6646 | |
6647 | return 0; | |
e6484930 TH |
6648 | } |
6649 | ||
a2029240 DV |
6650 | void netif_tx_stop_all_queues(struct net_device *dev) |
6651 | { | |
6652 | unsigned int i; | |
6653 | ||
6654 | for (i = 0; i < dev->num_tx_queues; i++) { | |
6655 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
6656 | netif_tx_stop_queue(txq); | |
6657 | } | |
6658 | } | |
6659 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
6660 | ||
1da177e4 LT |
6661 | /** |
6662 | * register_netdevice - register a network device | |
6663 | * @dev: device to register | |
6664 | * | |
6665 | * Take a completed network device structure and add it to the kernel | |
6666 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6667 | * chain. 0 is returned on success. A negative errno code is returned | |
6668 | * on a failure to set up the device, or if the name is a duplicate. | |
6669 | * | |
6670 | * Callers must hold the rtnl semaphore. You may want | |
6671 | * register_netdev() instead of this. | |
6672 | * | |
6673 | * BUGS: | |
6674 | * The locking appears insufficient to guarantee two parallel registers | |
6675 | * will not get the same name. | |
6676 | */ | |
6677 | ||
6678 | int register_netdevice(struct net_device *dev) | |
6679 | { | |
1da177e4 | 6680 | int ret; |
d314774c | 6681 | struct net *net = dev_net(dev); |
1da177e4 LT |
6682 | |
6683 | BUG_ON(dev_boot_phase); | |
6684 | ASSERT_RTNL(); | |
6685 | ||
b17a7c17 SH |
6686 | might_sleep(); |
6687 | ||
1da177e4 LT |
6688 | /* When net_device's are persistent, this will be fatal. */ |
6689 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 6690 | BUG_ON(!net); |
1da177e4 | 6691 | |
f1f28aa3 | 6692 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 6693 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 6694 | |
828de4f6 | 6695 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
6696 | if (ret < 0) |
6697 | goto out; | |
6698 | ||
1da177e4 | 6699 | /* Init, if this function is available */ |
d314774c SH |
6700 | if (dev->netdev_ops->ndo_init) { |
6701 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
6702 | if (ret) { |
6703 | if (ret > 0) | |
6704 | ret = -EIO; | |
90833aa4 | 6705 | goto out; |
1da177e4 LT |
6706 | } |
6707 | } | |
4ec93edb | 6708 | |
f646968f PM |
6709 | if (((dev->hw_features | dev->features) & |
6710 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
6711 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
6712 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
6713 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
6714 | ret = -EINVAL; | |
6715 | goto err_uninit; | |
6716 | } | |
6717 | ||
9c7dafbf PE |
6718 | ret = -EBUSY; |
6719 | if (!dev->ifindex) | |
6720 | dev->ifindex = dev_new_index(net); | |
6721 | else if (__dev_get_by_index(net, dev->ifindex)) | |
6722 | goto err_uninit; | |
6723 | ||
5455c699 MM |
6724 | /* Transfer changeable features to wanted_features and enable |
6725 | * software offloads (GSO and GRO). | |
6726 | */ | |
6727 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
6728 | dev->features |= NETIF_F_SOFT_FEATURES; |
6729 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 6730 | |
34324dc2 MM |
6731 | if (!(dev->flags & IFF_LOOPBACK)) { |
6732 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
c6e1a0d1 TH |
6733 | } |
6734 | ||
1180e7d6 | 6735 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 6736 | */ |
1180e7d6 | 6737 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 6738 | |
ee579677 PS |
6739 | /* Make NETIF_F_SG inheritable to tunnel devices. |
6740 | */ | |
6741 | dev->hw_enc_features |= NETIF_F_SG; | |
6742 | ||
0d89d203 SH |
6743 | /* Make NETIF_F_SG inheritable to MPLS. |
6744 | */ | |
6745 | dev->mpls_features |= NETIF_F_SG; | |
6746 | ||
7ffbe3fd JB |
6747 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
6748 | ret = notifier_to_errno(ret); | |
6749 | if (ret) | |
6750 | goto err_uninit; | |
6751 | ||
8b41d188 | 6752 | ret = netdev_register_kobject(dev); |
b17a7c17 | 6753 | if (ret) |
7ce1b0ed | 6754 | goto err_uninit; |
b17a7c17 SH |
6755 | dev->reg_state = NETREG_REGISTERED; |
6756 | ||
6cb6a27c | 6757 | __netdev_update_features(dev); |
8e9b59b2 | 6758 | |
1da177e4 LT |
6759 | /* |
6760 | * Default initial state at registry is that the | |
6761 | * device is present. | |
6762 | */ | |
6763 | ||
6764 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6765 | ||
8f4cccbb BH |
6766 | linkwatch_init_dev(dev); |
6767 | ||
1da177e4 | 6768 | dev_init_scheduler(dev); |
1da177e4 | 6769 | dev_hold(dev); |
ce286d32 | 6770 | list_netdevice(dev); |
7bf23575 | 6771 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 6772 | |
948b337e JP |
6773 | /* If the device has permanent device address, driver should |
6774 | * set dev_addr and also addr_assign_type should be set to | |
6775 | * NET_ADDR_PERM (default value). | |
6776 | */ | |
6777 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
6778 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
6779 | ||
1da177e4 | 6780 | /* Notify protocols, that a new device appeared. */ |
056925ab | 6781 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 6782 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
6783 | if (ret) { |
6784 | rollback_registered(dev); | |
6785 | dev->reg_state = NETREG_UNREGISTERED; | |
6786 | } | |
d90a909e EB |
6787 | /* |
6788 | * Prevent userspace races by waiting until the network | |
6789 | * device is fully setup before sending notifications. | |
6790 | */ | |
a2835763 PM |
6791 | if (!dev->rtnl_link_ops || |
6792 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 6793 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
6794 | |
6795 | out: | |
6796 | return ret; | |
7ce1b0ed HX |
6797 | |
6798 | err_uninit: | |
d314774c SH |
6799 | if (dev->netdev_ops->ndo_uninit) |
6800 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 6801 | goto out; |
1da177e4 | 6802 | } |
d1b19dff | 6803 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 6804 | |
937f1ba5 BH |
6805 | /** |
6806 | * init_dummy_netdev - init a dummy network device for NAPI | |
6807 | * @dev: device to init | |
6808 | * | |
6809 | * This takes a network device structure and initialize the minimum | |
6810 | * amount of fields so it can be used to schedule NAPI polls without | |
6811 | * registering a full blown interface. This is to be used by drivers | |
6812 | * that need to tie several hardware interfaces to a single NAPI | |
6813 | * poll scheduler due to HW limitations. | |
6814 | */ | |
6815 | int init_dummy_netdev(struct net_device *dev) | |
6816 | { | |
6817 | /* Clear everything. Note we don't initialize spinlocks | |
6818 | * are they aren't supposed to be taken by any of the | |
6819 | * NAPI code and this dummy netdev is supposed to be | |
6820 | * only ever used for NAPI polls | |
6821 | */ | |
6822 | memset(dev, 0, sizeof(struct net_device)); | |
6823 | ||
6824 | /* make sure we BUG if trying to hit standard | |
6825 | * register/unregister code path | |
6826 | */ | |
6827 | dev->reg_state = NETREG_DUMMY; | |
6828 | ||
937f1ba5 BH |
6829 | /* NAPI wants this */ |
6830 | INIT_LIST_HEAD(&dev->napi_list); | |
6831 | ||
6832 | /* a dummy interface is started by default */ | |
6833 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6834 | set_bit(__LINK_STATE_START, &dev->state); | |
6835 | ||
29b4433d ED |
6836 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
6837 | * because users of this 'device' dont need to change | |
6838 | * its refcount. | |
6839 | */ | |
6840 | ||
937f1ba5 BH |
6841 | return 0; |
6842 | } | |
6843 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
6844 | ||
6845 | ||
1da177e4 LT |
6846 | /** |
6847 | * register_netdev - register a network device | |
6848 | * @dev: device to register | |
6849 | * | |
6850 | * Take a completed network device structure and add it to the kernel | |
6851 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6852 | * chain. 0 is returned on success. A negative errno code is returned | |
6853 | * on a failure to set up the device, or if the name is a duplicate. | |
6854 | * | |
38b4da38 | 6855 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
6856 | * and expands the device name if you passed a format string to |
6857 | * alloc_netdev. | |
6858 | */ | |
6859 | int register_netdev(struct net_device *dev) | |
6860 | { | |
6861 | int err; | |
6862 | ||
6863 | rtnl_lock(); | |
1da177e4 | 6864 | err = register_netdevice(dev); |
1da177e4 LT |
6865 | rtnl_unlock(); |
6866 | return err; | |
6867 | } | |
6868 | EXPORT_SYMBOL(register_netdev); | |
6869 | ||
29b4433d ED |
6870 | int netdev_refcnt_read(const struct net_device *dev) |
6871 | { | |
6872 | int i, refcnt = 0; | |
6873 | ||
6874 | for_each_possible_cpu(i) | |
6875 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
6876 | return refcnt; | |
6877 | } | |
6878 | EXPORT_SYMBOL(netdev_refcnt_read); | |
6879 | ||
2c53040f | 6880 | /** |
1da177e4 | 6881 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 6882 | * @dev: target net_device |
1da177e4 LT |
6883 | * |
6884 | * This is called when unregistering network devices. | |
6885 | * | |
6886 | * Any protocol or device that holds a reference should register | |
6887 | * for netdevice notification, and cleanup and put back the | |
6888 | * reference if they receive an UNREGISTER event. | |
6889 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 6890 | * call dev_put. |
1da177e4 LT |
6891 | */ |
6892 | static void netdev_wait_allrefs(struct net_device *dev) | |
6893 | { | |
6894 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 6895 | int refcnt; |
1da177e4 | 6896 | |
e014debe ED |
6897 | linkwatch_forget_dev(dev); |
6898 | ||
1da177e4 | 6899 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
6900 | refcnt = netdev_refcnt_read(dev); |
6901 | ||
6902 | while (refcnt != 0) { | |
1da177e4 | 6903 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 6904 | rtnl_lock(); |
1da177e4 LT |
6905 | |
6906 | /* Rebroadcast unregister notification */ | |
056925ab | 6907 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 6908 | |
748e2d93 | 6909 | __rtnl_unlock(); |
0115e8e3 | 6910 | rcu_barrier(); |
748e2d93 ED |
6911 | rtnl_lock(); |
6912 | ||
0115e8e3 | 6913 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
6914 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
6915 | &dev->state)) { | |
6916 | /* We must not have linkwatch events | |
6917 | * pending on unregister. If this | |
6918 | * happens, we simply run the queue | |
6919 | * unscheduled, resulting in a noop | |
6920 | * for this device. | |
6921 | */ | |
6922 | linkwatch_run_queue(); | |
6923 | } | |
6924 | ||
6756ae4b | 6925 | __rtnl_unlock(); |
1da177e4 LT |
6926 | |
6927 | rebroadcast_time = jiffies; | |
6928 | } | |
6929 | ||
6930 | msleep(250); | |
6931 | ||
29b4433d ED |
6932 | refcnt = netdev_refcnt_read(dev); |
6933 | ||
1da177e4 | 6934 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
6935 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
6936 | dev->name, refcnt); | |
1da177e4 LT |
6937 | warning_time = jiffies; |
6938 | } | |
6939 | } | |
6940 | } | |
6941 | ||
6942 | /* The sequence is: | |
6943 | * | |
6944 | * rtnl_lock(); | |
6945 | * ... | |
6946 | * register_netdevice(x1); | |
6947 | * register_netdevice(x2); | |
6948 | * ... | |
6949 | * unregister_netdevice(y1); | |
6950 | * unregister_netdevice(y2); | |
6951 | * ... | |
6952 | * rtnl_unlock(); | |
6953 | * free_netdev(y1); | |
6954 | * free_netdev(y2); | |
6955 | * | |
58ec3b4d | 6956 | * We are invoked by rtnl_unlock(). |
1da177e4 | 6957 | * This allows us to deal with problems: |
b17a7c17 | 6958 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
6959 | * without deadlocking with linkwatch via keventd. |
6960 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
6961 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
6962 | * |
6963 | * We must not return until all unregister events added during | |
6964 | * the interval the lock was held have been completed. | |
1da177e4 | 6965 | */ |
1da177e4 LT |
6966 | void netdev_run_todo(void) |
6967 | { | |
626ab0e6 | 6968 | struct list_head list; |
1da177e4 | 6969 | |
1da177e4 | 6970 | /* Snapshot list, allow later requests */ |
626ab0e6 | 6971 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
6972 | |
6973 | __rtnl_unlock(); | |
626ab0e6 | 6974 | |
0115e8e3 ED |
6975 | |
6976 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
6977 | if (!list_empty(&list)) |
6978 | rcu_barrier(); | |
6979 | ||
1da177e4 LT |
6980 | while (!list_empty(&list)) { |
6981 | struct net_device *dev | |
e5e26d75 | 6982 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
6983 | list_del(&dev->todo_list); |
6984 | ||
748e2d93 | 6985 | rtnl_lock(); |
0115e8e3 | 6986 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 6987 | __rtnl_unlock(); |
0115e8e3 | 6988 | |
b17a7c17 | 6989 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 6990 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
6991 | dev->name, dev->reg_state); |
6992 | dump_stack(); | |
6993 | continue; | |
6994 | } | |
1da177e4 | 6995 | |
b17a7c17 | 6996 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 6997 | |
b17a7c17 | 6998 | netdev_wait_allrefs(dev); |
1da177e4 | 6999 | |
b17a7c17 | 7000 | /* paranoia */ |
29b4433d | 7001 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
7002 | BUG_ON(!list_empty(&dev->ptype_all)); |
7003 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
7004 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
7005 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 7006 | WARN_ON(dev->dn_ptr); |
1da177e4 | 7007 | |
b17a7c17 SH |
7008 | if (dev->destructor) |
7009 | dev->destructor(dev); | |
9093bbb2 | 7010 | |
50624c93 EB |
7011 | /* Report a network device has been unregistered */ |
7012 | rtnl_lock(); | |
7013 | dev_net(dev)->dev_unreg_count--; | |
7014 | __rtnl_unlock(); | |
7015 | wake_up(&netdev_unregistering_wq); | |
7016 | ||
9093bbb2 SH |
7017 | /* Free network device */ |
7018 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 7019 | } |
1da177e4 LT |
7020 | } |
7021 | ||
3cfde79c BH |
7022 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
7023 | * fields in the same order, with only the type differing. | |
7024 | */ | |
77a1abf5 ED |
7025 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
7026 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
7027 | { |
7028 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
7029 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
7030 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
7031 | #else |
7032 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
7033 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
7034 | u64 *dst = (u64 *)stats64; | |
7035 | ||
7036 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
7037 | sizeof(*stats64) / sizeof(u64)); | |
7038 | for (i = 0; i < n; i++) | |
7039 | dst[i] = src[i]; | |
7040 | #endif | |
7041 | } | |
77a1abf5 | 7042 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 7043 | |
eeda3fd6 SH |
7044 | /** |
7045 | * dev_get_stats - get network device statistics | |
7046 | * @dev: device to get statistics from | |
28172739 | 7047 | * @storage: place to store stats |
eeda3fd6 | 7048 | * |
d7753516 BH |
7049 | * Get network statistics from device. Return @storage. |
7050 | * The device driver may provide its own method by setting | |
7051 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
7052 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 7053 | */ |
d7753516 BH |
7054 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
7055 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 7056 | { |
eeda3fd6 SH |
7057 | const struct net_device_ops *ops = dev->netdev_ops; |
7058 | ||
28172739 ED |
7059 | if (ops->ndo_get_stats64) { |
7060 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
7061 | ops->ndo_get_stats64(dev, storage); |
7062 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 7063 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
7064 | } else { |
7065 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 7066 | } |
caf586e5 | 7067 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
015f0688 | 7068 | storage->tx_dropped += atomic_long_read(&dev->tx_dropped); |
28172739 | 7069 | return storage; |
c45d286e | 7070 | } |
eeda3fd6 | 7071 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 7072 | |
24824a09 | 7073 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 7074 | { |
24824a09 | 7075 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 7076 | |
24824a09 ED |
7077 | #ifdef CONFIG_NET_CLS_ACT |
7078 | if (queue) | |
7079 | return queue; | |
7080 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
7081 | if (!queue) | |
7082 | return NULL; | |
7083 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 7084 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
7085 | queue->qdisc_sleeping = &noop_qdisc; |
7086 | rcu_assign_pointer(dev->ingress_queue, queue); | |
7087 | #endif | |
7088 | return queue; | |
bb949fbd DM |
7089 | } |
7090 | ||
2c60db03 ED |
7091 | static const struct ethtool_ops default_ethtool_ops; |
7092 | ||
d07d7507 SG |
7093 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
7094 | const struct ethtool_ops *ops) | |
7095 | { | |
7096 | if (dev->ethtool_ops == &default_ethtool_ops) | |
7097 | dev->ethtool_ops = ops; | |
7098 | } | |
7099 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
7100 | ||
74d332c1 ED |
7101 | void netdev_freemem(struct net_device *dev) |
7102 | { | |
7103 | char *addr = (char *)dev - dev->padded; | |
7104 | ||
4cb28970 | 7105 | kvfree(addr); |
74d332c1 ED |
7106 | } |
7107 | ||
1da177e4 | 7108 | /** |
36909ea4 | 7109 | * alloc_netdev_mqs - allocate network device |
c835a677 TG |
7110 | * @sizeof_priv: size of private data to allocate space for |
7111 | * @name: device name format string | |
7112 | * @name_assign_type: origin of device name | |
7113 | * @setup: callback to initialize device | |
7114 | * @txqs: the number of TX subqueues to allocate | |
7115 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
7116 | * |
7117 | * Allocates a struct net_device with private data area for driver use | |
90e51adf | 7118 | * and performs basic initialization. Also allocates subqueue structs |
36909ea4 | 7119 | * for each queue on the device. |
1da177e4 | 7120 | */ |
36909ea4 | 7121 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 7122 | unsigned char name_assign_type, |
36909ea4 TH |
7123 | void (*setup)(struct net_device *), |
7124 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 7125 | { |
1da177e4 | 7126 | struct net_device *dev; |
7943986c | 7127 | size_t alloc_size; |
1ce8e7b5 | 7128 | struct net_device *p; |
1da177e4 | 7129 | |
b6fe17d6 SH |
7130 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
7131 | ||
36909ea4 | 7132 | if (txqs < 1) { |
7b6cd1ce | 7133 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
7134 | return NULL; |
7135 | } | |
7136 | ||
a953be53 | 7137 | #ifdef CONFIG_SYSFS |
36909ea4 | 7138 | if (rxqs < 1) { |
7b6cd1ce | 7139 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
7140 | return NULL; |
7141 | } | |
7142 | #endif | |
7143 | ||
fd2ea0a7 | 7144 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
7145 | if (sizeof_priv) { |
7146 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 7147 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
7148 | alloc_size += sizeof_priv; |
7149 | } | |
7150 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 7151 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 7152 | |
74d332c1 ED |
7153 | p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
7154 | if (!p) | |
7155 | p = vzalloc(alloc_size); | |
62b5942a | 7156 | if (!p) |
1da177e4 | 7157 | return NULL; |
1da177e4 | 7158 | |
1ce8e7b5 | 7159 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 7160 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 7161 | |
29b4433d ED |
7162 | dev->pcpu_refcnt = alloc_percpu(int); |
7163 | if (!dev->pcpu_refcnt) | |
74d332c1 | 7164 | goto free_dev; |
ab9c73cc | 7165 | |
ab9c73cc | 7166 | if (dev_addr_init(dev)) |
29b4433d | 7167 | goto free_pcpu; |
ab9c73cc | 7168 | |
22bedad3 | 7169 | dev_mc_init(dev); |
a748ee24 | 7170 | dev_uc_init(dev); |
ccffad25 | 7171 | |
c346dca1 | 7172 | dev_net_set(dev, &init_net); |
1da177e4 | 7173 | |
8d3bdbd5 | 7174 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 7175 | dev->gso_max_segs = GSO_MAX_SEGS; |
fcbeb976 | 7176 | dev->gso_min_segs = 0; |
8d3bdbd5 | 7177 | |
8d3bdbd5 DM |
7178 | INIT_LIST_HEAD(&dev->napi_list); |
7179 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 7180 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 7181 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
7182 | INIT_LIST_HEAD(&dev->adj_list.upper); |
7183 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7184 | INIT_LIST_HEAD(&dev->all_adj_list.upper); | |
7185 | INIT_LIST_HEAD(&dev->all_adj_list.lower); | |
7866a621 SN |
7186 | INIT_LIST_HEAD(&dev->ptype_all); |
7187 | INIT_LIST_HEAD(&dev->ptype_specific); | |
02875878 | 7188 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
7189 | setup(dev); |
7190 | ||
906470c1 | 7191 | if (!dev->tx_queue_len) |
f84bb1ea | 7192 | dev->priv_flags |= IFF_NO_QUEUE; |
906470c1 | 7193 | |
36909ea4 TH |
7194 | dev->num_tx_queues = txqs; |
7195 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 7196 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 7197 | goto free_all; |
e8a0464c | 7198 | |
a953be53 | 7199 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
7200 | dev->num_rx_queues = rxqs; |
7201 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 7202 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 7203 | goto free_all; |
df334545 | 7204 | #endif |
0a9627f2 | 7205 | |
1da177e4 | 7206 | strcpy(dev->name, name); |
c835a677 | 7207 | dev->name_assign_type = name_assign_type; |
cbda10fa | 7208 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
7209 | if (!dev->ethtool_ops) |
7210 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 PN |
7211 | |
7212 | nf_hook_ingress_init(dev); | |
7213 | ||
1da177e4 | 7214 | return dev; |
ab9c73cc | 7215 | |
8d3bdbd5 DM |
7216 | free_all: |
7217 | free_netdev(dev); | |
7218 | return NULL; | |
7219 | ||
29b4433d ED |
7220 | free_pcpu: |
7221 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
7222 | free_dev: |
7223 | netdev_freemem(dev); | |
ab9c73cc | 7224 | return NULL; |
1da177e4 | 7225 | } |
36909ea4 | 7226 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
7227 | |
7228 | /** | |
7229 | * free_netdev - free network device | |
7230 | * @dev: device | |
7231 | * | |
4ec93edb YH |
7232 | * This function does the last stage of destroying an allocated device |
7233 | * interface. The reference to the device object is released. | |
1da177e4 | 7234 | * If this is the last reference then it will be freed. |
93d05d4a | 7235 | * Must be called in process context. |
1da177e4 LT |
7236 | */ |
7237 | void free_netdev(struct net_device *dev) | |
7238 | { | |
d565b0a1 HX |
7239 | struct napi_struct *p, *n; |
7240 | ||
93d05d4a | 7241 | might_sleep(); |
60877a32 | 7242 | netif_free_tx_queues(dev); |
a953be53 | 7243 | #ifdef CONFIG_SYSFS |
10595902 | 7244 | kvfree(dev->_rx); |
fe822240 | 7245 | #endif |
e8a0464c | 7246 | |
33d480ce | 7247 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 7248 | |
f001fde5 JP |
7249 | /* Flush device addresses */ |
7250 | dev_addr_flush(dev); | |
7251 | ||
d565b0a1 HX |
7252 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
7253 | netif_napi_del(p); | |
7254 | ||
29b4433d ED |
7255 | free_percpu(dev->pcpu_refcnt); |
7256 | dev->pcpu_refcnt = NULL; | |
7257 | ||
3041a069 | 7258 | /* Compatibility with error handling in drivers */ |
1da177e4 | 7259 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 7260 | netdev_freemem(dev); |
1da177e4 LT |
7261 | return; |
7262 | } | |
7263 | ||
7264 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
7265 | dev->reg_state = NETREG_RELEASED; | |
7266 | ||
43cb76d9 GKH |
7267 | /* will free via device release */ |
7268 | put_device(&dev->dev); | |
1da177e4 | 7269 | } |
d1b19dff | 7270 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 7271 | |
f0db275a SH |
7272 | /** |
7273 | * synchronize_net - Synchronize with packet receive processing | |
7274 | * | |
7275 | * Wait for packets currently being received to be done. | |
7276 | * Does not block later packets from starting. | |
7277 | */ | |
4ec93edb | 7278 | void synchronize_net(void) |
1da177e4 LT |
7279 | { |
7280 | might_sleep(); | |
be3fc413 ED |
7281 | if (rtnl_is_locked()) |
7282 | synchronize_rcu_expedited(); | |
7283 | else | |
7284 | synchronize_rcu(); | |
1da177e4 | 7285 | } |
d1b19dff | 7286 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
7287 | |
7288 | /** | |
44a0873d | 7289 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 7290 | * @dev: device |
44a0873d | 7291 | * @head: list |
6ebfbc06 | 7292 | * |
1da177e4 | 7293 | * This function shuts down a device interface and removes it |
d59b54b1 | 7294 | * from the kernel tables. |
44a0873d | 7295 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
7296 | * |
7297 | * Callers must hold the rtnl semaphore. You may want | |
7298 | * unregister_netdev() instead of this. | |
7299 | */ | |
7300 | ||
44a0873d | 7301 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 7302 | { |
a6620712 HX |
7303 | ASSERT_RTNL(); |
7304 | ||
44a0873d | 7305 | if (head) { |
9fdce099 | 7306 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
7307 | } else { |
7308 | rollback_registered(dev); | |
7309 | /* Finish processing unregister after unlock */ | |
7310 | net_set_todo(dev); | |
7311 | } | |
1da177e4 | 7312 | } |
44a0873d | 7313 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 7314 | |
9b5e383c ED |
7315 | /** |
7316 | * unregister_netdevice_many - unregister many devices | |
7317 | * @head: list of devices | |
87757a91 ED |
7318 | * |
7319 | * Note: As most callers use a stack allocated list_head, | |
7320 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
7321 | */ |
7322 | void unregister_netdevice_many(struct list_head *head) | |
7323 | { | |
7324 | struct net_device *dev; | |
7325 | ||
7326 | if (!list_empty(head)) { | |
7327 | rollback_registered_many(head); | |
7328 | list_for_each_entry(dev, head, unreg_list) | |
7329 | net_set_todo(dev); | |
87757a91 | 7330 | list_del(head); |
9b5e383c ED |
7331 | } |
7332 | } | |
63c8099d | 7333 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 7334 | |
1da177e4 LT |
7335 | /** |
7336 | * unregister_netdev - remove device from the kernel | |
7337 | * @dev: device | |
7338 | * | |
7339 | * This function shuts down a device interface and removes it | |
d59b54b1 | 7340 | * from the kernel tables. |
1da177e4 LT |
7341 | * |
7342 | * This is just a wrapper for unregister_netdevice that takes | |
7343 | * the rtnl semaphore. In general you want to use this and not | |
7344 | * unregister_netdevice. | |
7345 | */ | |
7346 | void unregister_netdev(struct net_device *dev) | |
7347 | { | |
7348 | rtnl_lock(); | |
7349 | unregister_netdevice(dev); | |
7350 | rtnl_unlock(); | |
7351 | } | |
1da177e4 LT |
7352 | EXPORT_SYMBOL(unregister_netdev); |
7353 | ||
ce286d32 EB |
7354 | /** |
7355 | * dev_change_net_namespace - move device to different nethost namespace | |
7356 | * @dev: device | |
7357 | * @net: network namespace | |
7358 | * @pat: If not NULL name pattern to try if the current device name | |
7359 | * is already taken in the destination network namespace. | |
7360 | * | |
7361 | * This function shuts down a device interface and moves it | |
7362 | * to a new network namespace. On success 0 is returned, on | |
7363 | * a failure a netagive errno code is returned. | |
7364 | * | |
7365 | * Callers must hold the rtnl semaphore. | |
7366 | */ | |
7367 | ||
7368 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
7369 | { | |
ce286d32 EB |
7370 | int err; |
7371 | ||
7372 | ASSERT_RTNL(); | |
7373 | ||
7374 | /* Don't allow namespace local devices to be moved. */ | |
7375 | err = -EINVAL; | |
7376 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7377 | goto out; | |
7378 | ||
7379 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
7380 | if (dev->reg_state != NETREG_REGISTERED) |
7381 | goto out; | |
7382 | ||
7383 | /* Get out if there is nothing todo */ | |
7384 | err = 0; | |
878628fb | 7385 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
7386 | goto out; |
7387 | ||
7388 | /* Pick the destination device name, and ensure | |
7389 | * we can use it in the destination network namespace. | |
7390 | */ | |
7391 | err = -EEXIST; | |
d9031024 | 7392 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
7393 | /* We get here if we can't use the current device name */ |
7394 | if (!pat) | |
7395 | goto out; | |
828de4f6 | 7396 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
7397 | goto out; |
7398 | } | |
7399 | ||
7400 | /* | |
7401 | * And now a mini version of register_netdevice unregister_netdevice. | |
7402 | */ | |
7403 | ||
7404 | /* If device is running close it first. */ | |
9b772652 | 7405 | dev_close(dev); |
ce286d32 EB |
7406 | |
7407 | /* And unlink it from device chain */ | |
7408 | err = -ENODEV; | |
7409 | unlist_netdevice(dev); | |
7410 | ||
7411 | synchronize_net(); | |
7412 | ||
7413 | /* Shutdown queueing discipline. */ | |
7414 | dev_shutdown(dev); | |
7415 | ||
7416 | /* Notify protocols, that we are about to destroy | |
7417 | this device. They should clean all the things. | |
3b27e105 DL |
7418 | |
7419 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
7420 | This is wanted because this way 8021q and macvlan know | |
7421 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
7422 | */ |
7423 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
7424 | rcu_barrier(); |
7425 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 7426 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
7427 | |
7428 | /* | |
7429 | * Flush the unicast and multicast chains | |
7430 | */ | |
a748ee24 | 7431 | dev_uc_flush(dev); |
22bedad3 | 7432 | dev_mc_flush(dev); |
ce286d32 | 7433 | |
4e66ae2e SH |
7434 | /* Send a netdev-removed uevent to the old namespace */ |
7435 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 7436 | netdev_adjacent_del_links(dev); |
4e66ae2e | 7437 | |
ce286d32 | 7438 | /* Actually switch the network namespace */ |
c346dca1 | 7439 | dev_net_set(dev, net); |
ce286d32 | 7440 | |
ce286d32 | 7441 | /* If there is an ifindex conflict assign a new one */ |
7a66bbc9 | 7442 | if (__dev_get_by_index(net, dev->ifindex)) |
ce286d32 | 7443 | dev->ifindex = dev_new_index(net); |
ce286d32 | 7444 | |
4e66ae2e SH |
7445 | /* Send a netdev-add uevent to the new namespace */ |
7446 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 7447 | netdev_adjacent_add_links(dev); |
4e66ae2e | 7448 | |
8b41d188 | 7449 | /* Fixup kobjects */ |
a1b3f594 | 7450 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 7451 | WARN_ON(err); |
ce286d32 EB |
7452 | |
7453 | /* Add the device back in the hashes */ | |
7454 | list_netdevice(dev); | |
7455 | ||
7456 | /* Notify protocols, that a new device appeared. */ | |
7457 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
7458 | ||
d90a909e EB |
7459 | /* |
7460 | * Prevent userspace races by waiting until the network | |
7461 | * device is fully setup before sending notifications. | |
7462 | */ | |
7f294054 | 7463 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 7464 | |
ce286d32 EB |
7465 | synchronize_net(); |
7466 | err = 0; | |
7467 | out: | |
7468 | return err; | |
7469 | } | |
463d0183 | 7470 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 7471 | |
1da177e4 LT |
7472 | static int dev_cpu_callback(struct notifier_block *nfb, |
7473 | unsigned long action, | |
7474 | void *ocpu) | |
7475 | { | |
7476 | struct sk_buff **list_skb; | |
1da177e4 LT |
7477 | struct sk_buff *skb; |
7478 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
7479 | struct softnet_data *sd, *oldsd; | |
7480 | ||
8bb78442 | 7481 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
7482 | return NOTIFY_OK; |
7483 | ||
7484 | local_irq_disable(); | |
7485 | cpu = smp_processor_id(); | |
7486 | sd = &per_cpu(softnet_data, cpu); | |
7487 | oldsd = &per_cpu(softnet_data, oldcpu); | |
7488 | ||
7489 | /* Find end of our completion_queue. */ | |
7490 | list_skb = &sd->completion_queue; | |
7491 | while (*list_skb) | |
7492 | list_skb = &(*list_skb)->next; | |
7493 | /* Append completion queue from offline CPU. */ | |
7494 | *list_skb = oldsd->completion_queue; | |
7495 | oldsd->completion_queue = NULL; | |
7496 | ||
1da177e4 | 7497 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
7498 | if (oldsd->output_queue) { |
7499 | *sd->output_queue_tailp = oldsd->output_queue; | |
7500 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
7501 | oldsd->output_queue = NULL; | |
7502 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
7503 | } | |
ac64da0b ED |
7504 | /* Append NAPI poll list from offline CPU, with one exception : |
7505 | * process_backlog() must be called by cpu owning percpu backlog. | |
7506 | * We properly handle process_queue & input_pkt_queue later. | |
7507 | */ | |
7508 | while (!list_empty(&oldsd->poll_list)) { | |
7509 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
7510 | struct napi_struct, | |
7511 | poll_list); | |
7512 | ||
7513 | list_del_init(&napi->poll_list); | |
7514 | if (napi->poll == process_backlog) | |
7515 | napi->state = 0; | |
7516 | else | |
7517 | ____napi_schedule(sd, napi); | |
264524d5 | 7518 | } |
1da177e4 LT |
7519 | |
7520 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
7521 | local_irq_enable(); | |
7522 | ||
7523 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 7524 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 7525 | netif_rx_ni(skb); |
76cc8b13 | 7526 | input_queue_head_incr(oldsd); |
fec5e652 | 7527 | } |
ac64da0b | 7528 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 7529 | netif_rx_ni(skb); |
76cc8b13 TH |
7530 | input_queue_head_incr(oldsd); |
7531 | } | |
1da177e4 LT |
7532 | |
7533 | return NOTIFY_OK; | |
7534 | } | |
1da177e4 LT |
7535 | |
7536 | ||
7f353bf2 | 7537 | /** |
b63365a2 HX |
7538 | * netdev_increment_features - increment feature set by one |
7539 | * @all: current feature set | |
7540 | * @one: new feature set | |
7541 | * @mask: mask feature set | |
7f353bf2 HX |
7542 | * |
7543 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
7544 | * @one to the master device with current feature set @all. Will not |
7545 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 7546 | */ |
c8f44aff MM |
7547 | netdev_features_t netdev_increment_features(netdev_features_t all, |
7548 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 7549 | { |
1742f183 MM |
7550 | if (mask & NETIF_F_GEN_CSUM) |
7551 | mask |= NETIF_F_ALL_CSUM; | |
7552 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 7553 | |
1742f183 MM |
7554 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
7555 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 7556 | |
1742f183 MM |
7557 | /* If one device supports hw checksumming, set for all. */ |
7558 | if (all & NETIF_F_GEN_CSUM) | |
7559 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
7560 | |
7561 | return all; | |
7562 | } | |
b63365a2 | 7563 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 7564 | |
430f03cd | 7565 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
7566 | { |
7567 | int i; | |
7568 | struct hlist_head *hash; | |
7569 | ||
7570 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
7571 | if (hash != NULL) | |
7572 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
7573 | INIT_HLIST_HEAD(&hash[i]); | |
7574 | ||
7575 | return hash; | |
7576 | } | |
7577 | ||
881d966b | 7578 | /* Initialize per network namespace state */ |
4665079c | 7579 | static int __net_init netdev_init(struct net *net) |
881d966b | 7580 | { |
734b6541 RM |
7581 | if (net != &init_net) |
7582 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 7583 | |
30d97d35 PE |
7584 | net->dev_name_head = netdev_create_hash(); |
7585 | if (net->dev_name_head == NULL) | |
7586 | goto err_name; | |
881d966b | 7587 | |
30d97d35 PE |
7588 | net->dev_index_head = netdev_create_hash(); |
7589 | if (net->dev_index_head == NULL) | |
7590 | goto err_idx; | |
881d966b EB |
7591 | |
7592 | return 0; | |
30d97d35 PE |
7593 | |
7594 | err_idx: | |
7595 | kfree(net->dev_name_head); | |
7596 | err_name: | |
7597 | return -ENOMEM; | |
881d966b EB |
7598 | } |
7599 | ||
f0db275a SH |
7600 | /** |
7601 | * netdev_drivername - network driver for the device | |
7602 | * @dev: network device | |
f0db275a SH |
7603 | * |
7604 | * Determine network driver for device. | |
7605 | */ | |
3019de12 | 7606 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 7607 | { |
cf04a4c7 SH |
7608 | const struct device_driver *driver; |
7609 | const struct device *parent; | |
3019de12 | 7610 | const char *empty = ""; |
6579e57b AV |
7611 | |
7612 | parent = dev->dev.parent; | |
6579e57b | 7613 | if (!parent) |
3019de12 | 7614 | return empty; |
6579e57b AV |
7615 | |
7616 | driver = parent->driver; | |
7617 | if (driver && driver->name) | |
3019de12 DM |
7618 | return driver->name; |
7619 | return empty; | |
6579e57b AV |
7620 | } |
7621 | ||
6ea754eb JP |
7622 | static void __netdev_printk(const char *level, const struct net_device *dev, |
7623 | struct va_format *vaf) | |
256df2f3 | 7624 | { |
b004ff49 | 7625 | if (dev && dev->dev.parent) { |
6ea754eb JP |
7626 | dev_printk_emit(level[1] - '0', |
7627 | dev->dev.parent, | |
7628 | "%s %s %s%s: %pV", | |
7629 | dev_driver_string(dev->dev.parent), | |
7630 | dev_name(dev->dev.parent), | |
7631 | netdev_name(dev), netdev_reg_state(dev), | |
7632 | vaf); | |
b004ff49 | 7633 | } else if (dev) { |
6ea754eb JP |
7634 | printk("%s%s%s: %pV", |
7635 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 7636 | } else { |
6ea754eb | 7637 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 7638 | } |
256df2f3 JP |
7639 | } |
7640 | ||
6ea754eb JP |
7641 | void netdev_printk(const char *level, const struct net_device *dev, |
7642 | const char *format, ...) | |
256df2f3 JP |
7643 | { |
7644 | struct va_format vaf; | |
7645 | va_list args; | |
256df2f3 JP |
7646 | |
7647 | va_start(args, format); | |
7648 | ||
7649 | vaf.fmt = format; | |
7650 | vaf.va = &args; | |
7651 | ||
6ea754eb | 7652 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 7653 | |
256df2f3 | 7654 | va_end(args); |
256df2f3 JP |
7655 | } |
7656 | EXPORT_SYMBOL(netdev_printk); | |
7657 | ||
7658 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 7659 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 7660 | { \ |
256df2f3 JP |
7661 | struct va_format vaf; \ |
7662 | va_list args; \ | |
7663 | \ | |
7664 | va_start(args, fmt); \ | |
7665 | \ | |
7666 | vaf.fmt = fmt; \ | |
7667 | vaf.va = &args; \ | |
7668 | \ | |
6ea754eb | 7669 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 7670 | \ |
256df2f3 | 7671 | va_end(args); \ |
256df2f3 JP |
7672 | } \ |
7673 | EXPORT_SYMBOL(func); | |
7674 | ||
7675 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
7676 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
7677 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
7678 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
7679 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
7680 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
7681 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
7682 | ||
4665079c | 7683 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
7684 | { |
7685 | kfree(net->dev_name_head); | |
7686 | kfree(net->dev_index_head); | |
7687 | } | |
7688 | ||
022cbae6 | 7689 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
7690 | .init = netdev_init, |
7691 | .exit = netdev_exit, | |
7692 | }; | |
7693 | ||
4665079c | 7694 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 7695 | { |
e008b5fc | 7696 | struct net_device *dev, *aux; |
ce286d32 | 7697 | /* |
e008b5fc | 7698 | * Push all migratable network devices back to the |
ce286d32 EB |
7699 | * initial network namespace |
7700 | */ | |
7701 | rtnl_lock(); | |
e008b5fc | 7702 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 7703 | int err; |
aca51397 | 7704 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
7705 | |
7706 | /* Ignore unmoveable devices (i.e. loopback) */ | |
7707 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7708 | continue; | |
7709 | ||
e008b5fc EB |
7710 | /* Leave virtual devices for the generic cleanup */ |
7711 | if (dev->rtnl_link_ops) | |
7712 | continue; | |
d0c082ce | 7713 | |
25985edc | 7714 | /* Push remaining network devices to init_net */ |
aca51397 PE |
7715 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
7716 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 7717 | if (err) { |
7b6cd1ce JP |
7718 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
7719 | __func__, dev->name, err); | |
aca51397 | 7720 | BUG(); |
ce286d32 EB |
7721 | } |
7722 | } | |
7723 | rtnl_unlock(); | |
7724 | } | |
7725 | ||
50624c93 EB |
7726 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
7727 | { | |
7728 | /* Return with the rtnl_lock held when there are no network | |
7729 | * devices unregistering in any network namespace in net_list. | |
7730 | */ | |
7731 | struct net *net; | |
7732 | bool unregistering; | |
ff960a73 | 7733 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 7734 | |
ff960a73 | 7735 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 7736 | for (;;) { |
50624c93 EB |
7737 | unregistering = false; |
7738 | rtnl_lock(); | |
7739 | list_for_each_entry(net, net_list, exit_list) { | |
7740 | if (net->dev_unreg_count > 0) { | |
7741 | unregistering = true; | |
7742 | break; | |
7743 | } | |
7744 | } | |
7745 | if (!unregistering) | |
7746 | break; | |
7747 | __rtnl_unlock(); | |
ff960a73 PZ |
7748 | |
7749 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 7750 | } |
ff960a73 | 7751 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
7752 | } |
7753 | ||
04dc7f6b EB |
7754 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
7755 | { | |
7756 | /* At exit all network devices most be removed from a network | |
b595076a | 7757 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
7758 | * Do this across as many network namespaces as possible to |
7759 | * improve batching efficiency. | |
7760 | */ | |
7761 | struct net_device *dev; | |
7762 | struct net *net; | |
7763 | LIST_HEAD(dev_kill_list); | |
7764 | ||
50624c93 EB |
7765 | /* To prevent network device cleanup code from dereferencing |
7766 | * loopback devices or network devices that have been freed | |
7767 | * wait here for all pending unregistrations to complete, | |
7768 | * before unregistring the loopback device and allowing the | |
7769 | * network namespace be freed. | |
7770 | * | |
7771 | * The netdev todo list containing all network devices | |
7772 | * unregistrations that happen in default_device_exit_batch | |
7773 | * will run in the rtnl_unlock() at the end of | |
7774 | * default_device_exit_batch. | |
7775 | */ | |
7776 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
7777 | list_for_each_entry(net, net_list, exit_list) { |
7778 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 7779 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
7780 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
7781 | else | |
7782 | unregister_netdevice_queue(dev, &dev_kill_list); | |
7783 | } | |
7784 | } | |
7785 | unregister_netdevice_many(&dev_kill_list); | |
7786 | rtnl_unlock(); | |
7787 | } | |
7788 | ||
022cbae6 | 7789 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 7790 | .exit = default_device_exit, |
04dc7f6b | 7791 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
7792 | }; |
7793 | ||
1da177e4 LT |
7794 | /* |
7795 | * Initialize the DEV module. At boot time this walks the device list and | |
7796 | * unhooks any devices that fail to initialise (normally hardware not | |
7797 | * present) and leaves us with a valid list of present and active devices. | |
7798 | * | |
7799 | */ | |
7800 | ||
7801 | /* | |
7802 | * This is called single threaded during boot, so no need | |
7803 | * to take the rtnl semaphore. | |
7804 | */ | |
7805 | static int __init net_dev_init(void) | |
7806 | { | |
7807 | int i, rc = -ENOMEM; | |
7808 | ||
7809 | BUG_ON(!dev_boot_phase); | |
7810 | ||
1da177e4 LT |
7811 | if (dev_proc_init()) |
7812 | goto out; | |
7813 | ||
8b41d188 | 7814 | if (netdev_kobject_init()) |
1da177e4 LT |
7815 | goto out; |
7816 | ||
7817 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 7818 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
7819 | INIT_LIST_HEAD(&ptype_base[i]); |
7820 | ||
62532da9 VY |
7821 | INIT_LIST_HEAD(&offload_base); |
7822 | ||
881d966b EB |
7823 | if (register_pernet_subsys(&netdev_net_ops)) |
7824 | goto out; | |
1da177e4 LT |
7825 | |
7826 | /* | |
7827 | * Initialise the packet receive queues. | |
7828 | */ | |
7829 | ||
6f912042 | 7830 | for_each_possible_cpu(i) { |
e36fa2f7 | 7831 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 7832 | |
e36fa2f7 | 7833 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 7834 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 7835 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 7836 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 7837 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
7838 | sd->csd.func = rps_trigger_softirq; |
7839 | sd->csd.info = sd; | |
e36fa2f7 | 7840 | sd->cpu = i; |
1e94d72f | 7841 | #endif |
0a9627f2 | 7842 | |
e36fa2f7 ED |
7843 | sd->backlog.poll = process_backlog; |
7844 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
7845 | } |
7846 | ||
1da177e4 LT |
7847 | dev_boot_phase = 0; |
7848 | ||
505d4f73 EB |
7849 | /* The loopback device is special if any other network devices |
7850 | * is present in a network namespace the loopback device must | |
7851 | * be present. Since we now dynamically allocate and free the | |
7852 | * loopback device ensure this invariant is maintained by | |
7853 | * keeping the loopback device as the first device on the | |
7854 | * list of network devices. Ensuring the loopback devices | |
7855 | * is the first device that appears and the last network device | |
7856 | * that disappears. | |
7857 | */ | |
7858 | if (register_pernet_device(&loopback_net_ops)) | |
7859 | goto out; | |
7860 | ||
7861 | if (register_pernet_device(&default_device_ops)) | |
7862 | goto out; | |
7863 | ||
962cf36c CM |
7864 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
7865 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
7866 | |
7867 | hotcpu_notifier(dev_cpu_callback, 0); | |
f38a9eb1 | 7868 | dst_subsys_init(); |
1da177e4 LT |
7869 | rc = 0; |
7870 | out: | |
7871 | return rc; | |
7872 | } | |
7873 | ||
7874 | subsys_initcall(net_dev_init); |