Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * NET3 Protocol independent device support routines. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Derived from the non IP parts of dev.c 1.0.19 | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * | |
14 | * Additional Authors: | |
15 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
16 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
17 | * David Hinds <dahinds@users.sourceforge.net> | |
18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
19 | * Adam Sulmicki <adam@cfar.umd.edu> | |
20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
21 | * | |
22 | * Changes: | |
23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
24 | * to 2 if register_netdev gets called | |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
27 | * Alan Cox : device private ioctl copies fields back. | |
28 | * Alan Cox : Transmit queue code does relevant | |
29 | * stunts to keep the queue safe. | |
30 | * Alan Cox : Fixed double lock. | |
31 | * Alan Cox : Fixed promisc NULL pointer trap | |
32 | * ???????? : Support the full private ioctl range | |
33 | * Alan Cox : Moved ioctl permission check into | |
34 | * drivers | |
35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
36 | * Alan Cox : 100 backlog just doesn't cut it when | |
37 | * you start doing multicast video 8) | |
38 | * Alan Cox : Rewrote net_bh and list manager. | |
39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. | |
40 | * Alan Cox : Took out transmit every packet pass | |
41 | * Saved a few bytes in the ioctl handler | |
42 | * Alan Cox : Network driver sets packet type before | |
43 | * calling netif_rx. Saves a function | |
44 | * call a packet. | |
45 | * Alan Cox : Hashed net_bh() | |
46 | * Richard Kooijman: Timestamp fixes. | |
47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
48 | * Alan Cox : Device lock protection. | |
49 | * Alan Cox : Fixed nasty side effect of device close | |
50 | * changes. | |
51 | * Rudi Cilibrasi : Pass the right thing to | |
52 | * set_mac_address() | |
53 | * Dave Miller : 32bit quantity for the device lock to | |
54 | * make it work out on a Sparc. | |
55 | * Bjorn Ekwall : Added KERNELD hack. | |
56 | * Alan Cox : Cleaned up the backlog initialise. | |
57 | * Craig Metz : SIOCGIFCONF fix if space for under | |
58 | * 1 device. | |
59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
60 | * is no device open function. | |
61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
63 | * Cyrus Durgin : Cleaned for KMOD | |
64 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
65 | * A network device unload needs to purge | |
66 | * the backlog queue. | |
67 | * Paul Rusty Russell : SIOCSIFNAME | |
68 | * Pekka Riikonen : Netdev boot-time settings code | |
69 | * Andrew Morton : Make unregister_netdevice wait | |
70 | * indefinitely on dev->refcnt | |
71 | * J Hadi Salim : - Backlog queue sampling | |
72 | * - netif_rx() feedback | |
73 | */ | |
74 | ||
75 | #include <asm/uaccess.h> | |
1da177e4 | 76 | #include <linux/bitops.h> |
4fc268d2 | 77 | #include <linux/capability.h> |
1da177e4 LT |
78 | #include <linux/cpu.h> |
79 | #include <linux/types.h> | |
80 | #include <linux/kernel.h> | |
08e9897d | 81 | #include <linux/hash.h> |
5a0e3ad6 | 82 | #include <linux/slab.h> |
1da177e4 | 83 | #include <linux/sched.h> |
4a3e2f71 | 84 | #include <linux/mutex.h> |
1da177e4 LT |
85 | #include <linux/string.h> |
86 | #include <linux/mm.h> | |
87 | #include <linux/socket.h> | |
88 | #include <linux/sockios.h> | |
89 | #include <linux/errno.h> | |
90 | #include <linux/interrupt.h> | |
91 | #include <linux/if_ether.h> | |
92 | #include <linux/netdevice.h> | |
93 | #include <linux/etherdevice.h> | |
0187bdfb | 94 | #include <linux/ethtool.h> |
1da177e4 LT |
95 | #include <linux/notifier.h> |
96 | #include <linux/skbuff.h> | |
457c4cbc | 97 | #include <net/net_namespace.h> |
1da177e4 LT |
98 | #include <net/sock.h> |
99 | #include <linux/rtnetlink.h> | |
1da177e4 | 100 | #include <linux/stat.h> |
1da177e4 LT |
101 | #include <net/dst.h> |
102 | #include <net/pkt_sched.h> | |
103 | #include <net/checksum.h> | |
44540960 | 104 | #include <net/xfrm.h> |
1da177e4 LT |
105 | #include <linux/highmem.h> |
106 | #include <linux/init.h> | |
1da177e4 | 107 | #include <linux/module.h> |
1da177e4 LT |
108 | #include <linux/netpoll.h> |
109 | #include <linux/rcupdate.h> | |
110 | #include <linux/delay.h> | |
1da177e4 | 111 | #include <net/iw_handler.h> |
1da177e4 | 112 | #include <asm/current.h> |
5bdb9886 | 113 | #include <linux/audit.h> |
db217334 | 114 | #include <linux/dmaengine.h> |
f6a78bfc | 115 | #include <linux/err.h> |
c7fa9d18 | 116 | #include <linux/ctype.h> |
723e98b7 | 117 | #include <linux/if_arp.h> |
6de329e2 | 118 | #include <linux/if_vlan.h> |
8f0f2223 | 119 | #include <linux/ip.h> |
ad55dcaf | 120 | #include <net/ip.h> |
25cd9ba0 | 121 | #include <net/mpls.h> |
8f0f2223 DM |
122 | #include <linux/ipv6.h> |
123 | #include <linux/in.h> | |
b6b2fed1 DM |
124 | #include <linux/jhash.h> |
125 | #include <linux/random.h> | |
9cbc1cb8 | 126 | #include <trace/events/napi.h> |
cf66ba58 | 127 | #include <trace/events/net.h> |
07dc22e7 | 128 | #include <trace/events/skb.h> |
5acbbd42 | 129 | #include <linux/pci.h> |
caeda9b9 | 130 | #include <linux/inetdevice.h> |
c445477d | 131 | #include <linux/cpu_rmap.h> |
c5905afb | 132 | #include <linux/static_key.h> |
af12fa6e | 133 | #include <linux/hashtable.h> |
60877a32 | 134 | #include <linux/vmalloc.h> |
529d0489 | 135 | #include <linux/if_macvlan.h> |
e7fd2885 | 136 | #include <linux/errqueue.h> |
3b47d303 | 137 | #include <linux/hrtimer.h> |
1da177e4 | 138 | |
342709ef PE |
139 | #include "net-sysfs.h" |
140 | ||
d565b0a1 HX |
141 | /* Instead of increasing this, you should create a hash table. */ |
142 | #define MAX_GRO_SKBS 8 | |
143 | ||
5d38a079 HX |
144 | /* This should be increased if a protocol with a bigger head is added. */ |
145 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
146 | ||
1da177e4 | 147 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 148 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
149 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
150 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 151 | static struct list_head offload_base __read_mostly; |
1da177e4 | 152 | |
ae78dbfa | 153 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
154 | static int call_netdevice_notifiers_info(unsigned long val, |
155 | struct net_device *dev, | |
156 | struct netdev_notifier_info *info); | |
ae78dbfa | 157 | |
1da177e4 | 158 | /* |
7562f876 | 159 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
160 | * semaphore. |
161 | * | |
c6d14c84 | 162 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
163 | * |
164 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 165 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
166 | * actual updates. This allows pure readers to access the list even |
167 | * while a writer is preparing to update it. | |
168 | * | |
169 | * To put it another way, dev_base_lock is held for writing only to | |
170 | * protect against pure readers; the rtnl semaphore provides the | |
171 | * protection against other writers. | |
172 | * | |
173 | * See, for example usages, register_netdevice() and | |
174 | * unregister_netdevice(), which must be called with the rtnl | |
175 | * semaphore held. | |
176 | */ | |
1da177e4 | 177 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
178 | EXPORT_SYMBOL(dev_base_lock); |
179 | ||
af12fa6e ET |
180 | /* protects napi_hash addition/deletion and napi_gen_id */ |
181 | static DEFINE_SPINLOCK(napi_hash_lock); | |
182 | ||
183 | static unsigned int napi_gen_id; | |
184 | static DEFINE_HASHTABLE(napi_hash, 8); | |
185 | ||
18afa4b0 | 186 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 187 | |
4e985ada TG |
188 | static inline void dev_base_seq_inc(struct net *net) |
189 | { | |
190 | while (++net->dev_base_seq == 0); | |
191 | } | |
192 | ||
881d966b | 193 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 194 | { |
95c96174 ED |
195 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
196 | ||
08e9897d | 197 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
198 | } |
199 | ||
881d966b | 200 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 201 | { |
7c28bd0b | 202 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
203 | } |
204 | ||
e36fa2f7 | 205 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
206 | { |
207 | #ifdef CONFIG_RPS | |
e36fa2f7 | 208 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
209 | #endif |
210 | } | |
211 | ||
e36fa2f7 | 212 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
213 | { |
214 | #ifdef CONFIG_RPS | |
e36fa2f7 | 215 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
216 | #endif |
217 | } | |
218 | ||
ce286d32 | 219 | /* Device list insertion */ |
53759be9 | 220 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 221 | { |
c346dca1 | 222 | struct net *net = dev_net(dev); |
ce286d32 EB |
223 | |
224 | ASSERT_RTNL(); | |
225 | ||
226 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 227 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 228 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
229 | hlist_add_head_rcu(&dev->index_hlist, |
230 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 231 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
232 | |
233 | dev_base_seq_inc(net); | |
ce286d32 EB |
234 | } |
235 | ||
fb699dfd ED |
236 | /* Device list removal |
237 | * caller must respect a RCU grace period before freeing/reusing dev | |
238 | */ | |
ce286d32 EB |
239 | static void unlist_netdevice(struct net_device *dev) |
240 | { | |
241 | ASSERT_RTNL(); | |
242 | ||
243 | /* Unlink dev from the device chain */ | |
244 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 245 | list_del_rcu(&dev->dev_list); |
72c9528b | 246 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 247 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 248 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
249 | |
250 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
251 | } |
252 | ||
1da177e4 LT |
253 | /* |
254 | * Our notifier list | |
255 | */ | |
256 | ||
f07d5b94 | 257 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
258 | |
259 | /* | |
260 | * Device drivers call our routines to queue packets here. We empty the | |
261 | * queue in the local softnet handler. | |
262 | */ | |
bea3348e | 263 | |
9958da05 | 264 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 265 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 266 | |
cf508b12 | 267 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 268 | /* |
c773e847 | 269 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
270 | * according to dev->type |
271 | */ | |
272 | static const unsigned short netdev_lock_type[] = | |
273 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
274 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
275 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
276 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
277 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
278 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
279 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
280 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
281 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
282 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
283 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
284 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
285 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
286 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
287 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 288 | |
36cbd3dc | 289 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
290 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
291 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
292 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
293 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
294 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
295 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
296 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
297 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
298 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
299 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
300 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
301 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
302 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
303 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
304 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
305 | |
306 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 307 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
308 | |
309 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
310 | { | |
311 | int i; | |
312 | ||
313 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
314 | if (netdev_lock_type[i] == dev_type) | |
315 | return i; | |
316 | /* the last key is used by default */ | |
317 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
318 | } | |
319 | ||
cf508b12 DM |
320 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
321 | unsigned short dev_type) | |
723e98b7 JP |
322 | { |
323 | int i; | |
324 | ||
325 | i = netdev_lock_pos(dev_type); | |
326 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
327 | netdev_lock_name[i]); | |
328 | } | |
cf508b12 DM |
329 | |
330 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
331 | { | |
332 | int i; | |
333 | ||
334 | i = netdev_lock_pos(dev->type); | |
335 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
336 | &netdev_addr_lock_key[i], | |
337 | netdev_lock_name[i]); | |
338 | } | |
723e98b7 | 339 | #else |
cf508b12 DM |
340 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
341 | unsigned short dev_type) | |
342 | { | |
343 | } | |
344 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
345 | { |
346 | } | |
347 | #endif | |
1da177e4 LT |
348 | |
349 | /******************************************************************************* | |
350 | ||
351 | Protocol management and registration routines | |
352 | ||
353 | *******************************************************************************/ | |
354 | ||
1da177e4 LT |
355 | /* |
356 | * Add a protocol ID to the list. Now that the input handler is | |
357 | * smarter we can dispense with all the messy stuff that used to be | |
358 | * here. | |
359 | * | |
360 | * BEWARE!!! Protocol handlers, mangling input packets, | |
361 | * MUST BE last in hash buckets and checking protocol handlers | |
362 | * MUST start from promiscuous ptype_all chain in net_bh. | |
363 | * It is true now, do not change it. | |
364 | * Explanation follows: if protocol handler, mangling packet, will | |
365 | * be the first on list, it is not able to sense, that packet | |
366 | * is cloned and should be copied-on-write, so that it will | |
367 | * change it and subsequent readers will get broken packet. | |
368 | * --ANK (980803) | |
369 | */ | |
370 | ||
c07b68e8 ED |
371 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
372 | { | |
373 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 374 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 375 | else |
7866a621 SN |
376 | return pt->dev ? &pt->dev->ptype_specific : |
377 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
378 | } |
379 | ||
1da177e4 LT |
380 | /** |
381 | * dev_add_pack - add packet handler | |
382 | * @pt: packet type declaration | |
383 | * | |
384 | * Add a protocol handler to the networking stack. The passed &packet_type | |
385 | * is linked into kernel lists and may not be freed until it has been | |
386 | * removed from the kernel lists. | |
387 | * | |
4ec93edb | 388 | * This call does not sleep therefore it can not |
1da177e4 LT |
389 | * guarantee all CPU's that are in middle of receiving packets |
390 | * will see the new packet type (until the next received packet). | |
391 | */ | |
392 | ||
393 | void dev_add_pack(struct packet_type *pt) | |
394 | { | |
c07b68e8 | 395 | struct list_head *head = ptype_head(pt); |
1da177e4 | 396 | |
c07b68e8 ED |
397 | spin_lock(&ptype_lock); |
398 | list_add_rcu(&pt->list, head); | |
399 | spin_unlock(&ptype_lock); | |
1da177e4 | 400 | } |
d1b19dff | 401 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 402 | |
1da177e4 LT |
403 | /** |
404 | * __dev_remove_pack - remove packet handler | |
405 | * @pt: packet type declaration | |
406 | * | |
407 | * Remove a protocol handler that was previously added to the kernel | |
408 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
409 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 410 | * returns. |
1da177e4 LT |
411 | * |
412 | * The packet type might still be in use by receivers | |
413 | * and must not be freed until after all the CPU's have gone | |
414 | * through a quiescent state. | |
415 | */ | |
416 | void __dev_remove_pack(struct packet_type *pt) | |
417 | { | |
c07b68e8 | 418 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
419 | struct packet_type *pt1; |
420 | ||
c07b68e8 | 421 | spin_lock(&ptype_lock); |
1da177e4 LT |
422 | |
423 | list_for_each_entry(pt1, head, list) { | |
424 | if (pt == pt1) { | |
425 | list_del_rcu(&pt->list); | |
426 | goto out; | |
427 | } | |
428 | } | |
429 | ||
7b6cd1ce | 430 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 431 | out: |
c07b68e8 | 432 | spin_unlock(&ptype_lock); |
1da177e4 | 433 | } |
d1b19dff ED |
434 | EXPORT_SYMBOL(__dev_remove_pack); |
435 | ||
1da177e4 LT |
436 | /** |
437 | * dev_remove_pack - remove packet handler | |
438 | * @pt: packet type declaration | |
439 | * | |
440 | * Remove a protocol handler that was previously added to the kernel | |
441 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
442 | * from the kernel lists and can be freed or reused once this function | |
443 | * returns. | |
444 | * | |
445 | * This call sleeps to guarantee that no CPU is looking at the packet | |
446 | * type after return. | |
447 | */ | |
448 | void dev_remove_pack(struct packet_type *pt) | |
449 | { | |
450 | __dev_remove_pack(pt); | |
4ec93edb | 451 | |
1da177e4 LT |
452 | synchronize_net(); |
453 | } | |
d1b19dff | 454 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 455 | |
62532da9 VY |
456 | |
457 | /** | |
458 | * dev_add_offload - register offload handlers | |
459 | * @po: protocol offload declaration | |
460 | * | |
461 | * Add protocol offload handlers to the networking stack. The passed | |
462 | * &proto_offload is linked into kernel lists and may not be freed until | |
463 | * it has been removed from the kernel lists. | |
464 | * | |
465 | * This call does not sleep therefore it can not | |
466 | * guarantee all CPU's that are in middle of receiving packets | |
467 | * will see the new offload handlers (until the next received packet). | |
468 | */ | |
469 | void dev_add_offload(struct packet_offload *po) | |
470 | { | |
471 | struct list_head *head = &offload_base; | |
472 | ||
473 | spin_lock(&offload_lock); | |
474 | list_add_rcu(&po->list, head); | |
475 | spin_unlock(&offload_lock); | |
476 | } | |
477 | EXPORT_SYMBOL(dev_add_offload); | |
478 | ||
479 | /** | |
480 | * __dev_remove_offload - remove offload handler | |
481 | * @po: packet offload declaration | |
482 | * | |
483 | * Remove a protocol offload handler that was previously added to the | |
484 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
485 | * is removed from the kernel lists and can be freed or reused once this | |
486 | * function returns. | |
487 | * | |
488 | * The packet type might still be in use by receivers | |
489 | * and must not be freed until after all the CPU's have gone | |
490 | * through a quiescent state. | |
491 | */ | |
1d143d9f | 492 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
493 | { |
494 | struct list_head *head = &offload_base; | |
495 | struct packet_offload *po1; | |
496 | ||
c53aa505 | 497 | spin_lock(&offload_lock); |
62532da9 VY |
498 | |
499 | list_for_each_entry(po1, head, list) { | |
500 | if (po == po1) { | |
501 | list_del_rcu(&po->list); | |
502 | goto out; | |
503 | } | |
504 | } | |
505 | ||
506 | pr_warn("dev_remove_offload: %p not found\n", po); | |
507 | out: | |
c53aa505 | 508 | spin_unlock(&offload_lock); |
62532da9 | 509 | } |
62532da9 VY |
510 | |
511 | /** | |
512 | * dev_remove_offload - remove packet offload handler | |
513 | * @po: packet offload declaration | |
514 | * | |
515 | * Remove a packet offload handler that was previously added to the kernel | |
516 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
517 | * removed from the kernel lists and can be freed or reused once this | |
518 | * function returns. | |
519 | * | |
520 | * This call sleeps to guarantee that no CPU is looking at the packet | |
521 | * type after return. | |
522 | */ | |
523 | void dev_remove_offload(struct packet_offload *po) | |
524 | { | |
525 | __dev_remove_offload(po); | |
526 | ||
527 | synchronize_net(); | |
528 | } | |
529 | EXPORT_SYMBOL(dev_remove_offload); | |
530 | ||
1da177e4 LT |
531 | /****************************************************************************** |
532 | ||
533 | Device Boot-time Settings Routines | |
534 | ||
535 | *******************************************************************************/ | |
536 | ||
537 | /* Boot time configuration table */ | |
538 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
539 | ||
540 | /** | |
541 | * netdev_boot_setup_add - add new setup entry | |
542 | * @name: name of the device | |
543 | * @map: configured settings for the device | |
544 | * | |
545 | * Adds new setup entry to the dev_boot_setup list. The function | |
546 | * returns 0 on error and 1 on success. This is a generic routine to | |
547 | * all netdevices. | |
548 | */ | |
549 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
550 | { | |
551 | struct netdev_boot_setup *s; | |
552 | int i; | |
553 | ||
554 | s = dev_boot_setup; | |
555 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
556 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
557 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 558 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
559 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
560 | break; | |
561 | } | |
562 | } | |
563 | ||
564 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
565 | } | |
566 | ||
567 | /** | |
568 | * netdev_boot_setup_check - check boot time settings | |
569 | * @dev: the netdevice | |
570 | * | |
571 | * Check boot time settings for the device. | |
572 | * The found settings are set for the device to be used | |
573 | * later in the device probing. | |
574 | * Returns 0 if no settings found, 1 if they are. | |
575 | */ | |
576 | int netdev_boot_setup_check(struct net_device *dev) | |
577 | { | |
578 | struct netdev_boot_setup *s = dev_boot_setup; | |
579 | int i; | |
580 | ||
581 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
582 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 583 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
584 | dev->irq = s[i].map.irq; |
585 | dev->base_addr = s[i].map.base_addr; | |
586 | dev->mem_start = s[i].map.mem_start; | |
587 | dev->mem_end = s[i].map.mem_end; | |
588 | return 1; | |
589 | } | |
590 | } | |
591 | return 0; | |
592 | } | |
d1b19dff | 593 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
594 | |
595 | ||
596 | /** | |
597 | * netdev_boot_base - get address from boot time settings | |
598 | * @prefix: prefix for network device | |
599 | * @unit: id for network device | |
600 | * | |
601 | * Check boot time settings for the base address of device. | |
602 | * The found settings are set for the device to be used | |
603 | * later in the device probing. | |
604 | * Returns 0 if no settings found. | |
605 | */ | |
606 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
607 | { | |
608 | const struct netdev_boot_setup *s = dev_boot_setup; | |
609 | char name[IFNAMSIZ]; | |
610 | int i; | |
611 | ||
612 | sprintf(name, "%s%d", prefix, unit); | |
613 | ||
614 | /* | |
615 | * If device already registered then return base of 1 | |
616 | * to indicate not to probe for this interface | |
617 | */ | |
881d966b | 618 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
619 | return 1; |
620 | ||
621 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
622 | if (!strcmp(name, s[i].name)) | |
623 | return s[i].map.base_addr; | |
624 | return 0; | |
625 | } | |
626 | ||
627 | /* | |
628 | * Saves at boot time configured settings for any netdevice. | |
629 | */ | |
630 | int __init netdev_boot_setup(char *str) | |
631 | { | |
632 | int ints[5]; | |
633 | struct ifmap map; | |
634 | ||
635 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
636 | if (!str || !*str) | |
637 | return 0; | |
638 | ||
639 | /* Save settings */ | |
640 | memset(&map, 0, sizeof(map)); | |
641 | if (ints[0] > 0) | |
642 | map.irq = ints[1]; | |
643 | if (ints[0] > 1) | |
644 | map.base_addr = ints[2]; | |
645 | if (ints[0] > 2) | |
646 | map.mem_start = ints[3]; | |
647 | if (ints[0] > 3) | |
648 | map.mem_end = ints[4]; | |
649 | ||
650 | /* Add new entry to the list */ | |
651 | return netdev_boot_setup_add(str, &map); | |
652 | } | |
653 | ||
654 | __setup("netdev=", netdev_boot_setup); | |
655 | ||
656 | /******************************************************************************* | |
657 | ||
658 | Device Interface Subroutines | |
659 | ||
660 | *******************************************************************************/ | |
661 | ||
a54acb3a ND |
662 | /** |
663 | * dev_get_iflink - get 'iflink' value of a interface | |
664 | * @dev: targeted interface | |
665 | * | |
666 | * Indicates the ifindex the interface is linked to. | |
667 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
668 | */ | |
669 | ||
670 | int dev_get_iflink(const struct net_device *dev) | |
671 | { | |
672 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
673 | return dev->netdev_ops->ndo_get_iflink(dev); | |
674 | ||
675 | return dev->iflink; | |
676 | } | |
677 | EXPORT_SYMBOL(dev_get_iflink); | |
678 | ||
1da177e4 LT |
679 | /** |
680 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 681 | * @net: the applicable net namespace |
1da177e4 LT |
682 | * @name: name to find |
683 | * | |
684 | * Find an interface by name. Must be called under RTNL semaphore | |
685 | * or @dev_base_lock. If the name is found a pointer to the device | |
686 | * is returned. If the name is not found then %NULL is returned. The | |
687 | * reference counters are not incremented so the caller must be | |
688 | * careful with locks. | |
689 | */ | |
690 | ||
881d966b | 691 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 692 | { |
0bd8d536 ED |
693 | struct net_device *dev; |
694 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 695 | |
b67bfe0d | 696 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
697 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
698 | return dev; | |
0bd8d536 | 699 | |
1da177e4 LT |
700 | return NULL; |
701 | } | |
d1b19dff | 702 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 703 | |
72c9528b ED |
704 | /** |
705 | * dev_get_by_name_rcu - find a device by its name | |
706 | * @net: the applicable net namespace | |
707 | * @name: name to find | |
708 | * | |
709 | * Find an interface by name. | |
710 | * If the name is found a pointer to the device is returned. | |
711 | * If the name is not found then %NULL is returned. | |
712 | * The reference counters are not incremented so the caller must be | |
713 | * careful with locks. The caller must hold RCU lock. | |
714 | */ | |
715 | ||
716 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
717 | { | |
72c9528b ED |
718 | struct net_device *dev; |
719 | struct hlist_head *head = dev_name_hash(net, name); | |
720 | ||
b67bfe0d | 721 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
722 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
723 | return dev; | |
724 | ||
725 | return NULL; | |
726 | } | |
727 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
728 | ||
1da177e4 LT |
729 | /** |
730 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 731 | * @net: the applicable net namespace |
1da177e4 LT |
732 | * @name: name to find |
733 | * | |
734 | * Find an interface by name. This can be called from any | |
735 | * context and does its own locking. The returned handle has | |
736 | * the usage count incremented and the caller must use dev_put() to | |
737 | * release it when it is no longer needed. %NULL is returned if no | |
738 | * matching device is found. | |
739 | */ | |
740 | ||
881d966b | 741 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
742 | { |
743 | struct net_device *dev; | |
744 | ||
72c9528b ED |
745 | rcu_read_lock(); |
746 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
747 | if (dev) |
748 | dev_hold(dev); | |
72c9528b | 749 | rcu_read_unlock(); |
1da177e4 LT |
750 | return dev; |
751 | } | |
d1b19dff | 752 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
753 | |
754 | /** | |
755 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 756 | * @net: the applicable net namespace |
1da177e4 LT |
757 | * @ifindex: index of device |
758 | * | |
759 | * Search for an interface by index. Returns %NULL if the device | |
760 | * is not found or a pointer to the device. The device has not | |
761 | * had its reference counter increased so the caller must be careful | |
762 | * about locking. The caller must hold either the RTNL semaphore | |
763 | * or @dev_base_lock. | |
764 | */ | |
765 | ||
881d966b | 766 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 767 | { |
0bd8d536 ED |
768 | struct net_device *dev; |
769 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 770 | |
b67bfe0d | 771 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
772 | if (dev->ifindex == ifindex) |
773 | return dev; | |
0bd8d536 | 774 | |
1da177e4 LT |
775 | return NULL; |
776 | } | |
d1b19dff | 777 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 778 | |
fb699dfd ED |
779 | /** |
780 | * dev_get_by_index_rcu - find a device by its ifindex | |
781 | * @net: the applicable net namespace | |
782 | * @ifindex: index of device | |
783 | * | |
784 | * Search for an interface by index. Returns %NULL if the device | |
785 | * is not found or a pointer to the device. The device has not | |
786 | * had its reference counter increased so the caller must be careful | |
787 | * about locking. The caller must hold RCU lock. | |
788 | */ | |
789 | ||
790 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
791 | { | |
fb699dfd ED |
792 | struct net_device *dev; |
793 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
794 | ||
b67bfe0d | 795 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
796 | if (dev->ifindex == ifindex) |
797 | return dev; | |
798 | ||
799 | return NULL; | |
800 | } | |
801 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
802 | ||
1da177e4 LT |
803 | |
804 | /** | |
805 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 806 | * @net: the applicable net namespace |
1da177e4 LT |
807 | * @ifindex: index of device |
808 | * | |
809 | * Search for an interface by index. Returns NULL if the device | |
810 | * is not found or a pointer to the device. The device returned has | |
811 | * had a reference added and the pointer is safe until the user calls | |
812 | * dev_put to indicate they have finished with it. | |
813 | */ | |
814 | ||
881d966b | 815 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
816 | { |
817 | struct net_device *dev; | |
818 | ||
fb699dfd ED |
819 | rcu_read_lock(); |
820 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
821 | if (dev) |
822 | dev_hold(dev); | |
fb699dfd | 823 | rcu_read_unlock(); |
1da177e4 LT |
824 | return dev; |
825 | } | |
d1b19dff | 826 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 827 | |
5dbe7c17 NS |
828 | /** |
829 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
830 | * @net: network namespace | |
831 | * @name: a pointer to the buffer where the name will be stored. | |
832 | * @ifindex: the ifindex of the interface to get the name from. | |
833 | * | |
834 | * The use of raw_seqcount_begin() and cond_resched() before | |
835 | * retrying is required as we want to give the writers a chance | |
836 | * to complete when CONFIG_PREEMPT is not set. | |
837 | */ | |
838 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
839 | { | |
840 | struct net_device *dev; | |
841 | unsigned int seq; | |
842 | ||
843 | retry: | |
844 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
845 | rcu_read_lock(); | |
846 | dev = dev_get_by_index_rcu(net, ifindex); | |
847 | if (!dev) { | |
848 | rcu_read_unlock(); | |
849 | return -ENODEV; | |
850 | } | |
851 | ||
852 | strcpy(name, dev->name); | |
853 | rcu_read_unlock(); | |
854 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
855 | cond_resched(); | |
856 | goto retry; | |
857 | } | |
858 | ||
859 | return 0; | |
860 | } | |
861 | ||
1da177e4 | 862 | /** |
941666c2 | 863 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 864 | * @net: the applicable net namespace |
1da177e4 LT |
865 | * @type: media type of device |
866 | * @ha: hardware address | |
867 | * | |
868 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
869 | * is not found or a pointer to the device. |
870 | * The caller must hold RCU or RTNL. | |
941666c2 | 871 | * The returned device has not had its ref count increased |
1da177e4 LT |
872 | * and the caller must therefore be careful about locking |
873 | * | |
1da177e4 LT |
874 | */ |
875 | ||
941666c2 ED |
876 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
877 | const char *ha) | |
1da177e4 LT |
878 | { |
879 | struct net_device *dev; | |
880 | ||
941666c2 | 881 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
882 | if (dev->type == type && |
883 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
884 | return dev; |
885 | ||
886 | return NULL; | |
1da177e4 | 887 | } |
941666c2 | 888 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 889 | |
881d966b | 890 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
891 | { |
892 | struct net_device *dev; | |
893 | ||
4e9cac2b | 894 | ASSERT_RTNL(); |
881d966b | 895 | for_each_netdev(net, dev) |
4e9cac2b | 896 | if (dev->type == type) |
7562f876 PE |
897 | return dev; |
898 | ||
899 | return NULL; | |
4e9cac2b | 900 | } |
4e9cac2b PM |
901 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
902 | ||
881d966b | 903 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 904 | { |
99fe3c39 | 905 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 906 | |
99fe3c39 ED |
907 | rcu_read_lock(); |
908 | for_each_netdev_rcu(net, dev) | |
909 | if (dev->type == type) { | |
910 | dev_hold(dev); | |
911 | ret = dev; | |
912 | break; | |
913 | } | |
914 | rcu_read_unlock(); | |
915 | return ret; | |
1da177e4 | 916 | } |
1da177e4 LT |
917 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
918 | ||
919 | /** | |
6c555490 | 920 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 921 | * @net: the applicable net namespace |
1da177e4 LT |
922 | * @if_flags: IFF_* values |
923 | * @mask: bitmask of bits in if_flags to check | |
924 | * | |
925 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 926 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 927 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
928 | */ |
929 | ||
6c555490 WC |
930 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
931 | unsigned short mask) | |
1da177e4 | 932 | { |
7562f876 | 933 | struct net_device *dev, *ret; |
1da177e4 | 934 | |
6c555490 WC |
935 | ASSERT_RTNL(); |
936 | ||
7562f876 | 937 | ret = NULL; |
6c555490 | 938 | for_each_netdev(net, dev) { |
1da177e4 | 939 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 940 | ret = dev; |
1da177e4 LT |
941 | break; |
942 | } | |
943 | } | |
7562f876 | 944 | return ret; |
1da177e4 | 945 | } |
6c555490 | 946 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
947 | |
948 | /** | |
949 | * dev_valid_name - check if name is okay for network device | |
950 | * @name: name string | |
951 | * | |
952 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
953 | * to allow sysfs to work. We also disallow any kind of |
954 | * whitespace. | |
1da177e4 | 955 | */ |
95f050bf | 956 | bool dev_valid_name(const char *name) |
1da177e4 | 957 | { |
c7fa9d18 | 958 | if (*name == '\0') |
95f050bf | 959 | return false; |
b6fe17d6 | 960 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 961 | return false; |
c7fa9d18 | 962 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 963 | return false; |
c7fa9d18 DM |
964 | |
965 | while (*name) { | |
a4176a93 | 966 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 967 | return false; |
c7fa9d18 DM |
968 | name++; |
969 | } | |
95f050bf | 970 | return true; |
1da177e4 | 971 | } |
d1b19dff | 972 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
973 | |
974 | /** | |
b267b179 EB |
975 | * __dev_alloc_name - allocate a name for a device |
976 | * @net: network namespace to allocate the device name in | |
1da177e4 | 977 | * @name: name format string |
b267b179 | 978 | * @buf: scratch buffer and result name string |
1da177e4 LT |
979 | * |
980 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
981 | * id. It scans list of devices to build up a free map, then chooses |
982 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
983 | * while allocating the name and adding the device in order to avoid | |
984 | * duplicates. | |
985 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
986 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
987 | */ |
988 | ||
b267b179 | 989 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
990 | { |
991 | int i = 0; | |
1da177e4 LT |
992 | const char *p; |
993 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 994 | unsigned long *inuse; |
1da177e4 LT |
995 | struct net_device *d; |
996 | ||
997 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
998 | if (p) { | |
999 | /* | |
1000 | * Verify the string as this thing may have come from | |
1001 | * the user. There must be either one "%d" and no other "%" | |
1002 | * characters. | |
1003 | */ | |
1004 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1005 | return -EINVAL; | |
1006 | ||
1007 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1008 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1009 | if (!inuse) |
1010 | return -ENOMEM; | |
1011 | ||
881d966b | 1012 | for_each_netdev(net, d) { |
1da177e4 LT |
1013 | if (!sscanf(d->name, name, &i)) |
1014 | continue; | |
1015 | if (i < 0 || i >= max_netdevices) | |
1016 | continue; | |
1017 | ||
1018 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1019 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1020 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1021 | set_bit(i, inuse); | |
1022 | } | |
1023 | ||
1024 | i = find_first_zero_bit(inuse, max_netdevices); | |
1025 | free_page((unsigned long) inuse); | |
1026 | } | |
1027 | ||
d9031024 OP |
1028 | if (buf != name) |
1029 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1030 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1031 | return i; |
1da177e4 LT |
1032 | |
1033 | /* It is possible to run out of possible slots | |
1034 | * when the name is long and there isn't enough space left | |
1035 | * for the digits, or if all bits are used. | |
1036 | */ | |
1037 | return -ENFILE; | |
1038 | } | |
1039 | ||
b267b179 EB |
1040 | /** |
1041 | * dev_alloc_name - allocate a name for a device | |
1042 | * @dev: device | |
1043 | * @name: name format string | |
1044 | * | |
1045 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1046 | * id. It scans list of devices to build up a free map, then chooses | |
1047 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1048 | * while allocating the name and adding the device in order to avoid | |
1049 | * duplicates. | |
1050 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1051 | * Returns the number of the unit assigned or a negative errno code. | |
1052 | */ | |
1053 | ||
1054 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1055 | { | |
1056 | char buf[IFNAMSIZ]; | |
1057 | struct net *net; | |
1058 | int ret; | |
1059 | ||
c346dca1 YH |
1060 | BUG_ON(!dev_net(dev)); |
1061 | net = dev_net(dev); | |
b267b179 EB |
1062 | ret = __dev_alloc_name(net, name, buf); |
1063 | if (ret >= 0) | |
1064 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1065 | return ret; | |
1066 | } | |
d1b19dff | 1067 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1068 | |
828de4f6 G |
1069 | static int dev_alloc_name_ns(struct net *net, |
1070 | struct net_device *dev, | |
1071 | const char *name) | |
d9031024 | 1072 | { |
828de4f6 G |
1073 | char buf[IFNAMSIZ]; |
1074 | int ret; | |
8ce6cebc | 1075 | |
828de4f6 G |
1076 | ret = __dev_alloc_name(net, name, buf); |
1077 | if (ret >= 0) | |
1078 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1079 | return ret; | |
1080 | } | |
1081 | ||
1082 | static int dev_get_valid_name(struct net *net, | |
1083 | struct net_device *dev, | |
1084 | const char *name) | |
1085 | { | |
1086 | BUG_ON(!net); | |
8ce6cebc | 1087 | |
d9031024 OP |
1088 | if (!dev_valid_name(name)) |
1089 | return -EINVAL; | |
1090 | ||
1c5cae81 | 1091 | if (strchr(name, '%')) |
828de4f6 | 1092 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1093 | else if (__dev_get_by_name(net, name)) |
1094 | return -EEXIST; | |
8ce6cebc DL |
1095 | else if (dev->name != name) |
1096 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1097 | |
1098 | return 0; | |
1099 | } | |
1da177e4 LT |
1100 | |
1101 | /** | |
1102 | * dev_change_name - change name of a device | |
1103 | * @dev: device | |
1104 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1105 | * | |
1106 | * Change name of a device, can pass format strings "eth%d". | |
1107 | * for wildcarding. | |
1108 | */ | |
cf04a4c7 | 1109 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1110 | { |
238fa362 | 1111 | unsigned char old_assign_type; |
fcc5a03a | 1112 | char oldname[IFNAMSIZ]; |
1da177e4 | 1113 | int err = 0; |
fcc5a03a | 1114 | int ret; |
881d966b | 1115 | struct net *net; |
1da177e4 LT |
1116 | |
1117 | ASSERT_RTNL(); | |
c346dca1 | 1118 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1119 | |
c346dca1 | 1120 | net = dev_net(dev); |
1da177e4 LT |
1121 | if (dev->flags & IFF_UP) |
1122 | return -EBUSY; | |
1123 | ||
30e6c9fa | 1124 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1125 | |
1126 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1127 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1128 | return 0; |
c91f6df2 | 1129 | } |
c8d90dca | 1130 | |
fcc5a03a HX |
1131 | memcpy(oldname, dev->name, IFNAMSIZ); |
1132 | ||
828de4f6 | 1133 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1134 | if (err < 0) { |
30e6c9fa | 1135 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1136 | return err; |
c91f6df2 | 1137 | } |
1da177e4 | 1138 | |
6fe82a39 VF |
1139 | if (oldname[0] && !strchr(oldname, '%')) |
1140 | netdev_info(dev, "renamed from %s\n", oldname); | |
1141 | ||
238fa362 TG |
1142 | old_assign_type = dev->name_assign_type; |
1143 | dev->name_assign_type = NET_NAME_RENAMED; | |
1144 | ||
fcc5a03a | 1145 | rollback: |
a1b3f594 EB |
1146 | ret = device_rename(&dev->dev, dev->name); |
1147 | if (ret) { | |
1148 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1149 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1150 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1151 | return ret; |
dcc99773 | 1152 | } |
7f988eab | 1153 | |
30e6c9fa | 1154 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1155 | |
5bb025fa VF |
1156 | netdev_adjacent_rename_links(dev, oldname); |
1157 | ||
7f988eab | 1158 | write_lock_bh(&dev_base_lock); |
372b2312 | 1159 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1160 | write_unlock_bh(&dev_base_lock); |
1161 | ||
1162 | synchronize_rcu(); | |
1163 | ||
1164 | write_lock_bh(&dev_base_lock); | |
1165 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1166 | write_unlock_bh(&dev_base_lock); |
1167 | ||
056925ab | 1168 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1169 | ret = notifier_to_errno(ret); |
1170 | ||
1171 | if (ret) { | |
91e9c07b ED |
1172 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1173 | if (err >= 0) { | |
fcc5a03a | 1174 | err = ret; |
30e6c9fa | 1175 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1176 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1177 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1178 | dev->name_assign_type = old_assign_type; |
1179 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1180 | goto rollback; |
91e9c07b | 1181 | } else { |
7b6cd1ce | 1182 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1183 | dev->name, ret); |
fcc5a03a HX |
1184 | } |
1185 | } | |
1da177e4 LT |
1186 | |
1187 | return err; | |
1188 | } | |
1189 | ||
0b815a1a SH |
1190 | /** |
1191 | * dev_set_alias - change ifalias of a device | |
1192 | * @dev: device | |
1193 | * @alias: name up to IFALIASZ | |
f0db275a | 1194 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1195 | * |
1196 | * Set ifalias for a device, | |
1197 | */ | |
1198 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1199 | { | |
7364e445 AK |
1200 | char *new_ifalias; |
1201 | ||
0b815a1a SH |
1202 | ASSERT_RTNL(); |
1203 | ||
1204 | if (len >= IFALIASZ) | |
1205 | return -EINVAL; | |
1206 | ||
96ca4a2c | 1207 | if (!len) { |
388dfc2d SK |
1208 | kfree(dev->ifalias); |
1209 | dev->ifalias = NULL; | |
96ca4a2c OH |
1210 | return 0; |
1211 | } | |
1212 | ||
7364e445 AK |
1213 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1214 | if (!new_ifalias) | |
0b815a1a | 1215 | return -ENOMEM; |
7364e445 | 1216 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1217 | |
1218 | strlcpy(dev->ifalias, alias, len+1); | |
1219 | return len; | |
1220 | } | |
1221 | ||
1222 | ||
d8a33ac4 | 1223 | /** |
3041a069 | 1224 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1225 | * @dev: device to cause notification |
1226 | * | |
1227 | * Called to indicate a device has changed features. | |
1228 | */ | |
1229 | void netdev_features_change(struct net_device *dev) | |
1230 | { | |
056925ab | 1231 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1232 | } |
1233 | EXPORT_SYMBOL(netdev_features_change); | |
1234 | ||
1da177e4 LT |
1235 | /** |
1236 | * netdev_state_change - device changes state | |
1237 | * @dev: device to cause notification | |
1238 | * | |
1239 | * Called to indicate a device has changed state. This function calls | |
1240 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1241 | * to the routing socket. | |
1242 | */ | |
1243 | void netdev_state_change(struct net_device *dev) | |
1244 | { | |
1245 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1246 | struct netdev_notifier_change_info change_info; |
1247 | ||
1248 | change_info.flags_changed = 0; | |
1249 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1250 | &change_info.info); | |
7f294054 | 1251 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1252 | } |
1253 | } | |
d1b19dff | 1254 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1255 | |
ee89bab1 AW |
1256 | /** |
1257 | * netdev_notify_peers - notify network peers about existence of @dev | |
1258 | * @dev: network device | |
1259 | * | |
1260 | * Generate traffic such that interested network peers are aware of | |
1261 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1262 | * a device wants to inform the rest of the network about some sort of | |
1263 | * reconfiguration such as a failover event or virtual machine | |
1264 | * migration. | |
1265 | */ | |
1266 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1267 | { |
ee89bab1 AW |
1268 | rtnl_lock(); |
1269 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1270 | rtnl_unlock(); | |
c1da4ac7 | 1271 | } |
ee89bab1 | 1272 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1273 | |
bd380811 | 1274 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1275 | { |
d314774c | 1276 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1277 | int ret; |
1da177e4 | 1278 | |
e46b66bc BH |
1279 | ASSERT_RTNL(); |
1280 | ||
1da177e4 LT |
1281 | if (!netif_device_present(dev)) |
1282 | return -ENODEV; | |
1283 | ||
ca99ca14 NH |
1284 | /* Block netpoll from trying to do any rx path servicing. |
1285 | * If we don't do this there is a chance ndo_poll_controller | |
1286 | * or ndo_poll may be running while we open the device | |
1287 | */ | |
66b5552f | 1288 | netpoll_poll_disable(dev); |
ca99ca14 | 1289 | |
3b8bcfd5 JB |
1290 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1291 | ret = notifier_to_errno(ret); | |
1292 | if (ret) | |
1293 | return ret; | |
1294 | ||
1da177e4 | 1295 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1296 | |
d314774c SH |
1297 | if (ops->ndo_validate_addr) |
1298 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1299 | |
d314774c SH |
1300 | if (!ret && ops->ndo_open) |
1301 | ret = ops->ndo_open(dev); | |
1da177e4 | 1302 | |
66b5552f | 1303 | netpoll_poll_enable(dev); |
ca99ca14 | 1304 | |
bada339b JG |
1305 | if (ret) |
1306 | clear_bit(__LINK_STATE_START, &dev->state); | |
1307 | else { | |
1da177e4 | 1308 | dev->flags |= IFF_UP; |
4417da66 | 1309 | dev_set_rx_mode(dev); |
1da177e4 | 1310 | dev_activate(dev); |
7bf23575 | 1311 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1312 | } |
bada339b | 1313 | |
1da177e4 LT |
1314 | return ret; |
1315 | } | |
1316 | ||
1317 | /** | |
bd380811 PM |
1318 | * dev_open - prepare an interface for use. |
1319 | * @dev: device to open | |
1da177e4 | 1320 | * |
bd380811 PM |
1321 | * Takes a device from down to up state. The device's private open |
1322 | * function is invoked and then the multicast lists are loaded. Finally | |
1323 | * the device is moved into the up state and a %NETDEV_UP message is | |
1324 | * sent to the netdev notifier chain. | |
1325 | * | |
1326 | * Calling this function on an active interface is a nop. On a failure | |
1327 | * a negative errno code is returned. | |
1da177e4 | 1328 | */ |
bd380811 PM |
1329 | int dev_open(struct net_device *dev) |
1330 | { | |
1331 | int ret; | |
1332 | ||
bd380811 PM |
1333 | if (dev->flags & IFF_UP) |
1334 | return 0; | |
1335 | ||
bd380811 PM |
1336 | ret = __dev_open(dev); |
1337 | if (ret < 0) | |
1338 | return ret; | |
1339 | ||
7f294054 | 1340 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1341 | call_netdevice_notifiers(NETDEV_UP, dev); |
1342 | ||
1343 | return ret; | |
1344 | } | |
1345 | EXPORT_SYMBOL(dev_open); | |
1346 | ||
44345724 | 1347 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1348 | { |
44345724 | 1349 | struct net_device *dev; |
e46b66bc | 1350 | |
bd380811 | 1351 | ASSERT_RTNL(); |
9d5010db DM |
1352 | might_sleep(); |
1353 | ||
5cde2829 | 1354 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1355 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1356 | netpoll_poll_disable(dev); |
3f4df206 | 1357 | |
44345724 | 1358 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1359 | |
44345724 | 1360 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1361 | |
44345724 OP |
1362 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1363 | * can be even on different cpu. So just clear netif_running(). | |
1364 | * | |
1365 | * dev->stop() will invoke napi_disable() on all of it's | |
1366 | * napi_struct instances on this device. | |
1367 | */ | |
4e857c58 | 1368 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1369 | } |
1da177e4 | 1370 | |
44345724 | 1371 | dev_deactivate_many(head); |
d8b2a4d2 | 1372 | |
5cde2829 | 1373 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1374 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1375 | |
44345724 OP |
1376 | /* |
1377 | * Call the device specific close. This cannot fail. | |
1378 | * Only if device is UP | |
1379 | * | |
1380 | * We allow it to be called even after a DETACH hot-plug | |
1381 | * event. | |
1382 | */ | |
1383 | if (ops->ndo_stop) | |
1384 | ops->ndo_stop(dev); | |
1385 | ||
44345724 | 1386 | dev->flags &= ~IFF_UP; |
66b5552f | 1387 | netpoll_poll_enable(dev); |
44345724 OP |
1388 | } |
1389 | ||
1390 | return 0; | |
1391 | } | |
1392 | ||
1393 | static int __dev_close(struct net_device *dev) | |
1394 | { | |
f87e6f47 | 1395 | int retval; |
44345724 OP |
1396 | LIST_HEAD(single); |
1397 | ||
5cde2829 | 1398 | list_add(&dev->close_list, &single); |
f87e6f47 LT |
1399 | retval = __dev_close_many(&single); |
1400 | list_del(&single); | |
ca99ca14 | 1401 | |
f87e6f47 | 1402 | return retval; |
44345724 OP |
1403 | } |
1404 | ||
99c4a26a | 1405 | int dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1406 | { |
1407 | struct net_device *dev, *tmp; | |
1da177e4 | 1408 | |
5cde2829 EB |
1409 | /* Remove the devices that don't need to be closed */ |
1410 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1411 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1412 | list_del_init(&dev->close_list); |
44345724 OP |
1413 | |
1414 | __dev_close_many(head); | |
1da177e4 | 1415 | |
5cde2829 | 1416 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1417 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1418 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1419 | if (unlink) |
1420 | list_del_init(&dev->close_list); | |
44345724 | 1421 | } |
bd380811 PM |
1422 | |
1423 | return 0; | |
1424 | } | |
99c4a26a | 1425 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1426 | |
1427 | /** | |
1428 | * dev_close - shutdown an interface. | |
1429 | * @dev: device to shutdown | |
1430 | * | |
1431 | * This function moves an active device into down state. A | |
1432 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1433 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1434 | * chain. | |
1435 | */ | |
1436 | int dev_close(struct net_device *dev) | |
1437 | { | |
e14a5993 ED |
1438 | if (dev->flags & IFF_UP) { |
1439 | LIST_HEAD(single); | |
1da177e4 | 1440 | |
5cde2829 | 1441 | list_add(&dev->close_list, &single); |
99c4a26a | 1442 | dev_close_many(&single, true); |
e14a5993 ED |
1443 | list_del(&single); |
1444 | } | |
da6e378b | 1445 | return 0; |
1da177e4 | 1446 | } |
d1b19dff | 1447 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1448 | |
1449 | ||
0187bdfb BH |
1450 | /** |
1451 | * dev_disable_lro - disable Large Receive Offload on a device | |
1452 | * @dev: device | |
1453 | * | |
1454 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1455 | * called under RTNL. This is needed if received packets may be | |
1456 | * forwarded to another interface. | |
1457 | */ | |
1458 | void dev_disable_lro(struct net_device *dev) | |
1459 | { | |
fbe168ba MK |
1460 | struct net_device *lower_dev; |
1461 | struct list_head *iter; | |
529d0489 | 1462 | |
bc5787c6 MM |
1463 | dev->wanted_features &= ~NETIF_F_LRO; |
1464 | netdev_update_features(dev); | |
27660515 | 1465 | |
22d5969f MM |
1466 | if (unlikely(dev->features & NETIF_F_LRO)) |
1467 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1468 | |
1469 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1470 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1471 | } |
1472 | EXPORT_SYMBOL(dev_disable_lro); | |
1473 | ||
351638e7 JP |
1474 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1475 | struct net_device *dev) | |
1476 | { | |
1477 | struct netdev_notifier_info info; | |
1478 | ||
1479 | netdev_notifier_info_init(&info, dev); | |
1480 | return nb->notifier_call(nb, val, &info); | |
1481 | } | |
0187bdfb | 1482 | |
881d966b EB |
1483 | static int dev_boot_phase = 1; |
1484 | ||
1da177e4 LT |
1485 | /** |
1486 | * register_netdevice_notifier - register a network notifier block | |
1487 | * @nb: notifier | |
1488 | * | |
1489 | * Register a notifier to be called when network device events occur. | |
1490 | * The notifier passed is linked into the kernel structures and must | |
1491 | * not be reused until it has been unregistered. A negative errno code | |
1492 | * is returned on a failure. | |
1493 | * | |
1494 | * When registered all registration and up events are replayed | |
4ec93edb | 1495 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1496 | * view of the network device list. |
1497 | */ | |
1498 | ||
1499 | int register_netdevice_notifier(struct notifier_block *nb) | |
1500 | { | |
1501 | struct net_device *dev; | |
fcc5a03a | 1502 | struct net_device *last; |
881d966b | 1503 | struct net *net; |
1da177e4 LT |
1504 | int err; |
1505 | ||
1506 | rtnl_lock(); | |
f07d5b94 | 1507 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1508 | if (err) |
1509 | goto unlock; | |
881d966b EB |
1510 | if (dev_boot_phase) |
1511 | goto unlock; | |
1512 | for_each_net(net) { | |
1513 | for_each_netdev(net, dev) { | |
351638e7 | 1514 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1515 | err = notifier_to_errno(err); |
1516 | if (err) | |
1517 | goto rollback; | |
1518 | ||
1519 | if (!(dev->flags & IFF_UP)) | |
1520 | continue; | |
1da177e4 | 1521 | |
351638e7 | 1522 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1523 | } |
1da177e4 | 1524 | } |
fcc5a03a HX |
1525 | |
1526 | unlock: | |
1da177e4 LT |
1527 | rtnl_unlock(); |
1528 | return err; | |
fcc5a03a HX |
1529 | |
1530 | rollback: | |
1531 | last = dev; | |
881d966b EB |
1532 | for_each_net(net) { |
1533 | for_each_netdev(net, dev) { | |
1534 | if (dev == last) | |
8f891489 | 1535 | goto outroll; |
fcc5a03a | 1536 | |
881d966b | 1537 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1538 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1539 | dev); | |
1540 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1541 | } |
351638e7 | 1542 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1543 | } |
fcc5a03a | 1544 | } |
c67625a1 | 1545 | |
8f891489 | 1546 | outroll: |
c67625a1 | 1547 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1548 | goto unlock; |
1da177e4 | 1549 | } |
d1b19dff | 1550 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1551 | |
1552 | /** | |
1553 | * unregister_netdevice_notifier - unregister a network notifier block | |
1554 | * @nb: notifier | |
1555 | * | |
1556 | * Unregister a notifier previously registered by | |
1557 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1558 | * kernel structures and may then be reused. A negative errno code | |
1559 | * is returned on a failure. | |
7d3d43da EB |
1560 | * |
1561 | * After unregistering unregister and down device events are synthesized | |
1562 | * for all devices on the device list to the removed notifier to remove | |
1563 | * the need for special case cleanup code. | |
1da177e4 LT |
1564 | */ |
1565 | ||
1566 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1567 | { | |
7d3d43da EB |
1568 | struct net_device *dev; |
1569 | struct net *net; | |
9f514950 HX |
1570 | int err; |
1571 | ||
1572 | rtnl_lock(); | |
f07d5b94 | 1573 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1574 | if (err) |
1575 | goto unlock; | |
1576 | ||
1577 | for_each_net(net) { | |
1578 | for_each_netdev(net, dev) { | |
1579 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1580 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1581 | dev); | |
1582 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1583 | } |
351638e7 | 1584 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1585 | } |
1586 | } | |
1587 | unlock: | |
9f514950 HX |
1588 | rtnl_unlock(); |
1589 | return err; | |
1da177e4 | 1590 | } |
d1b19dff | 1591 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1592 | |
351638e7 JP |
1593 | /** |
1594 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1595 | * @val: value passed unmodified to notifier function | |
1596 | * @dev: net_device pointer passed unmodified to notifier function | |
1597 | * @info: notifier information data | |
1598 | * | |
1599 | * Call all network notifier blocks. Parameters and return value | |
1600 | * are as for raw_notifier_call_chain(). | |
1601 | */ | |
1602 | ||
1d143d9f | 1603 | static int call_netdevice_notifiers_info(unsigned long val, |
1604 | struct net_device *dev, | |
1605 | struct netdev_notifier_info *info) | |
351638e7 JP |
1606 | { |
1607 | ASSERT_RTNL(); | |
1608 | netdev_notifier_info_init(info, dev); | |
1609 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1610 | } | |
351638e7 | 1611 | |
1da177e4 LT |
1612 | /** |
1613 | * call_netdevice_notifiers - call all network notifier blocks | |
1614 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1615 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1616 | * |
1617 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1618 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1619 | */ |
1620 | ||
ad7379d4 | 1621 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1622 | { |
351638e7 JP |
1623 | struct netdev_notifier_info info; |
1624 | ||
1625 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1626 | } |
edf947f1 | 1627 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1628 | |
c5905afb | 1629 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1630 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1631 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1632 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1633 | * static_key_slow_dec() calls. |
b90e5794 ED |
1634 | */ |
1635 | static atomic_t netstamp_needed_deferred; | |
1636 | #endif | |
1da177e4 LT |
1637 | |
1638 | void net_enable_timestamp(void) | |
1639 | { | |
b90e5794 ED |
1640 | #ifdef HAVE_JUMP_LABEL |
1641 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1642 | ||
1643 | if (deferred) { | |
1644 | while (--deferred) | |
c5905afb | 1645 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1646 | return; |
1647 | } | |
1648 | #endif | |
c5905afb | 1649 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1650 | } |
d1b19dff | 1651 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1652 | |
1653 | void net_disable_timestamp(void) | |
1654 | { | |
b90e5794 ED |
1655 | #ifdef HAVE_JUMP_LABEL |
1656 | if (in_interrupt()) { | |
1657 | atomic_inc(&netstamp_needed_deferred); | |
1658 | return; | |
1659 | } | |
1660 | #endif | |
c5905afb | 1661 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1662 | } |
d1b19dff | 1663 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1664 | |
3b098e2d | 1665 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1666 | { |
588f0330 | 1667 | skb->tstamp.tv64 = 0; |
c5905afb | 1668 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1669 | __net_timestamp(skb); |
1da177e4 LT |
1670 | } |
1671 | ||
588f0330 | 1672 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1673 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1674 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1675 | __net_timestamp(SKB); \ | |
1676 | } \ | |
3b098e2d | 1677 | |
1ee481fb | 1678 | bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb) |
79b569f0 DL |
1679 | { |
1680 | unsigned int len; | |
1681 | ||
1682 | if (!(dev->flags & IFF_UP)) | |
1683 | return false; | |
1684 | ||
1685 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1686 | if (skb->len <= len) | |
1687 | return true; | |
1688 | ||
1689 | /* if TSO is enabled, we don't care about the length as the packet | |
1690 | * could be forwarded without being segmented before | |
1691 | */ | |
1692 | if (skb_is_gso(skb)) | |
1693 | return true; | |
1694 | ||
1695 | return false; | |
1696 | } | |
1ee481fb | 1697 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1698 | |
a0265d28 HX |
1699 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1700 | { | |
1701 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { | |
1702 | if (skb_copy_ubufs(skb, GFP_ATOMIC)) { | |
1703 | atomic_long_inc(&dev->rx_dropped); | |
1704 | kfree_skb(skb); | |
1705 | return NET_RX_DROP; | |
1706 | } | |
1707 | } | |
1708 | ||
1709 | if (unlikely(!is_skb_forwardable(dev, skb))) { | |
1710 | atomic_long_inc(&dev->rx_dropped); | |
1711 | kfree_skb(skb); | |
1712 | return NET_RX_DROP; | |
1713 | } | |
1714 | ||
1715 | skb_scrub_packet(skb, true); | |
08b4b8ea | 1716 | skb->priority = 0; |
a0265d28 | 1717 | skb->protocol = eth_type_trans(skb, dev); |
2c26d34b | 1718 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); |
a0265d28 HX |
1719 | |
1720 | return 0; | |
1721 | } | |
1722 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1723 | ||
44540960 AB |
1724 | /** |
1725 | * dev_forward_skb - loopback an skb to another netif | |
1726 | * | |
1727 | * @dev: destination network device | |
1728 | * @skb: buffer to forward | |
1729 | * | |
1730 | * return values: | |
1731 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1732 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1733 | * |
1734 | * dev_forward_skb can be used for injecting an skb from the | |
1735 | * start_xmit function of one device into the receive queue | |
1736 | * of another device. | |
1737 | * | |
1738 | * The receiving device may be in another namespace, so | |
1739 | * we have to clear all information in the skb that could | |
1740 | * impact namespace isolation. | |
1741 | */ | |
1742 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1743 | { | |
a0265d28 | 1744 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1745 | } |
1746 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1747 | ||
71d9dec2 CG |
1748 | static inline int deliver_skb(struct sk_buff *skb, |
1749 | struct packet_type *pt_prev, | |
1750 | struct net_device *orig_dev) | |
1751 | { | |
1080e512 MT |
1752 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1753 | return -ENOMEM; | |
71d9dec2 CG |
1754 | atomic_inc(&skb->users); |
1755 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1756 | } | |
1757 | ||
7866a621 SN |
1758 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
1759 | struct packet_type **pt, | |
fbcb2170 JP |
1760 | struct net_device *orig_dev, |
1761 | __be16 type, | |
7866a621 SN |
1762 | struct list_head *ptype_list) |
1763 | { | |
1764 | struct packet_type *ptype, *pt_prev = *pt; | |
1765 | ||
1766 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1767 | if (ptype->type != type) | |
1768 | continue; | |
1769 | if (pt_prev) | |
fbcb2170 | 1770 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
1771 | pt_prev = ptype; |
1772 | } | |
1773 | *pt = pt_prev; | |
1774 | } | |
1775 | ||
c0de08d0 EL |
1776 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1777 | { | |
a3d744e9 | 1778 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1779 | return false; |
1780 | ||
1781 | if (ptype->id_match) | |
1782 | return ptype->id_match(ptype, skb->sk); | |
1783 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1784 | return true; | |
1785 | ||
1786 | return false; | |
1787 | } | |
1788 | ||
1da177e4 LT |
1789 | /* |
1790 | * Support routine. Sends outgoing frames to any network | |
1791 | * taps currently in use. | |
1792 | */ | |
1793 | ||
f6a78bfc | 1794 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1795 | { |
1796 | struct packet_type *ptype; | |
71d9dec2 CG |
1797 | struct sk_buff *skb2 = NULL; |
1798 | struct packet_type *pt_prev = NULL; | |
7866a621 | 1799 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 1800 | |
1da177e4 | 1801 | rcu_read_lock(); |
7866a621 SN |
1802 | again: |
1803 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1da177e4 LT |
1804 | /* Never send packets back to the socket |
1805 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1806 | */ | |
7866a621 SN |
1807 | if (skb_loop_sk(ptype, skb)) |
1808 | continue; | |
71d9dec2 | 1809 | |
7866a621 SN |
1810 | if (pt_prev) { |
1811 | deliver_skb(skb2, pt_prev, skb->dev); | |
1812 | pt_prev = ptype; | |
1813 | continue; | |
1814 | } | |
1da177e4 | 1815 | |
7866a621 SN |
1816 | /* need to clone skb, done only once */ |
1817 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1818 | if (!skb2) | |
1819 | goto out_unlock; | |
70978182 | 1820 | |
7866a621 | 1821 | net_timestamp_set(skb2); |
1da177e4 | 1822 | |
7866a621 SN |
1823 | /* skb->nh should be correctly |
1824 | * set by sender, so that the second statement is | |
1825 | * just protection against buggy protocols. | |
1826 | */ | |
1827 | skb_reset_mac_header(skb2); | |
1828 | ||
1829 | if (skb_network_header(skb2) < skb2->data || | |
1830 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
1831 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
1832 | ntohs(skb2->protocol), | |
1833 | dev->name); | |
1834 | skb_reset_network_header(skb2); | |
1da177e4 | 1835 | } |
7866a621 SN |
1836 | |
1837 | skb2->transport_header = skb2->network_header; | |
1838 | skb2->pkt_type = PACKET_OUTGOING; | |
1839 | pt_prev = ptype; | |
1840 | } | |
1841 | ||
1842 | if (ptype_list == &ptype_all) { | |
1843 | ptype_list = &dev->ptype_all; | |
1844 | goto again; | |
1da177e4 | 1845 | } |
7866a621 | 1846 | out_unlock: |
71d9dec2 CG |
1847 | if (pt_prev) |
1848 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1849 | rcu_read_unlock(); |
1850 | } | |
1851 | ||
2c53040f BH |
1852 | /** |
1853 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1854 | * @dev: Network device |
1855 | * @txq: number of queues available | |
1856 | * | |
1857 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1858 | * valid. To resolve this verify the tc mapping remains valid and if | |
1859 | * not NULL the mapping. With no priorities mapping to this | |
1860 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1861 | * is invalid nothing can be done so disable priority mappings. If is | |
1862 | * expected that drivers will fix this mapping if they can before | |
1863 | * calling netif_set_real_num_tx_queues. | |
1864 | */ | |
bb134d22 | 1865 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1866 | { |
1867 | int i; | |
1868 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1869 | ||
1870 | /* If TC0 is invalidated disable TC mapping */ | |
1871 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1872 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1873 | dev->num_tc = 0; |
1874 | return; | |
1875 | } | |
1876 | ||
1877 | /* Invalidated prio to tc mappings set to TC0 */ | |
1878 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1879 | int q = netdev_get_prio_tc_map(dev, i); | |
1880 | ||
1881 | tc = &dev->tc_to_txq[q]; | |
1882 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1883 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1884 | i, q); | |
4f57c087 JF |
1885 | netdev_set_prio_tc_map(dev, i, 0); |
1886 | } | |
1887 | } | |
1888 | } | |
1889 | ||
537c00de AD |
1890 | #ifdef CONFIG_XPS |
1891 | static DEFINE_MUTEX(xps_map_mutex); | |
1892 | #define xmap_dereference(P) \ | |
1893 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1894 | ||
10cdc3f3 AD |
1895 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1896 | int cpu, u16 index) | |
537c00de | 1897 | { |
10cdc3f3 AD |
1898 | struct xps_map *map = NULL; |
1899 | int pos; | |
537c00de | 1900 | |
10cdc3f3 AD |
1901 | if (dev_maps) |
1902 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1903 | |
10cdc3f3 AD |
1904 | for (pos = 0; map && pos < map->len; pos++) { |
1905 | if (map->queues[pos] == index) { | |
537c00de AD |
1906 | if (map->len > 1) { |
1907 | map->queues[pos] = map->queues[--map->len]; | |
1908 | } else { | |
10cdc3f3 | 1909 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1910 | kfree_rcu(map, rcu); |
1911 | map = NULL; | |
1912 | } | |
10cdc3f3 | 1913 | break; |
537c00de | 1914 | } |
537c00de AD |
1915 | } |
1916 | ||
10cdc3f3 AD |
1917 | return map; |
1918 | } | |
1919 | ||
024e9679 | 1920 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
10cdc3f3 AD |
1921 | { |
1922 | struct xps_dev_maps *dev_maps; | |
024e9679 | 1923 | int cpu, i; |
10cdc3f3 AD |
1924 | bool active = false; |
1925 | ||
1926 | mutex_lock(&xps_map_mutex); | |
1927 | dev_maps = xmap_dereference(dev->xps_maps); | |
1928 | ||
1929 | if (!dev_maps) | |
1930 | goto out_no_maps; | |
1931 | ||
1932 | for_each_possible_cpu(cpu) { | |
024e9679 AD |
1933 | for (i = index; i < dev->num_tx_queues; i++) { |
1934 | if (!remove_xps_queue(dev_maps, cpu, i)) | |
1935 | break; | |
1936 | } | |
1937 | if (i == dev->num_tx_queues) | |
10cdc3f3 AD |
1938 | active = true; |
1939 | } | |
1940 | ||
1941 | if (!active) { | |
537c00de AD |
1942 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
1943 | kfree_rcu(dev_maps, rcu); | |
1944 | } | |
1945 | ||
024e9679 AD |
1946 | for (i = index; i < dev->num_tx_queues; i++) |
1947 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), | |
1948 | NUMA_NO_NODE); | |
1949 | ||
537c00de AD |
1950 | out_no_maps: |
1951 | mutex_unlock(&xps_map_mutex); | |
1952 | } | |
1953 | ||
01c5f864 AD |
1954 | static struct xps_map *expand_xps_map(struct xps_map *map, |
1955 | int cpu, u16 index) | |
1956 | { | |
1957 | struct xps_map *new_map; | |
1958 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
1959 | int i, pos; | |
1960 | ||
1961 | for (pos = 0; map && pos < map->len; pos++) { | |
1962 | if (map->queues[pos] != index) | |
1963 | continue; | |
1964 | return map; | |
1965 | } | |
1966 | ||
1967 | /* Need to add queue to this CPU's existing map */ | |
1968 | if (map) { | |
1969 | if (pos < map->alloc_len) | |
1970 | return map; | |
1971 | ||
1972 | alloc_len = map->alloc_len * 2; | |
1973 | } | |
1974 | ||
1975 | /* Need to allocate new map to store queue on this CPU's map */ | |
1976 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
1977 | cpu_to_node(cpu)); | |
1978 | if (!new_map) | |
1979 | return NULL; | |
1980 | ||
1981 | for (i = 0; i < pos; i++) | |
1982 | new_map->queues[i] = map->queues[i]; | |
1983 | new_map->alloc_len = alloc_len; | |
1984 | new_map->len = pos; | |
1985 | ||
1986 | return new_map; | |
1987 | } | |
1988 | ||
3573540c MT |
1989 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
1990 | u16 index) | |
537c00de | 1991 | { |
01c5f864 | 1992 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 1993 | struct xps_map *map, *new_map; |
537c00de | 1994 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
1995 | int cpu, numa_node_id = -2; |
1996 | bool active = false; | |
537c00de AD |
1997 | |
1998 | mutex_lock(&xps_map_mutex); | |
1999 | ||
2000 | dev_maps = xmap_dereference(dev->xps_maps); | |
2001 | ||
01c5f864 AD |
2002 | /* allocate memory for queue storage */ |
2003 | for_each_online_cpu(cpu) { | |
2004 | if (!cpumask_test_cpu(cpu, mask)) | |
2005 | continue; | |
2006 | ||
2007 | if (!new_dev_maps) | |
2008 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2009 | if (!new_dev_maps) { |
2010 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2011 | return -ENOMEM; |
2bb60cb9 | 2012 | } |
01c5f864 AD |
2013 | |
2014 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2015 | NULL; | |
2016 | ||
2017 | map = expand_xps_map(map, cpu, index); | |
2018 | if (!map) | |
2019 | goto error; | |
2020 | ||
2021 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
2022 | } | |
2023 | ||
2024 | if (!new_dev_maps) | |
2025 | goto out_no_new_maps; | |
2026 | ||
537c00de | 2027 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
2028 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
2029 | /* add queue to CPU maps */ | |
2030 | int pos = 0; | |
2031 | ||
2032 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2033 | while ((pos < map->len) && (map->queues[pos] != index)) | |
2034 | pos++; | |
2035 | ||
2036 | if (pos == map->len) | |
2037 | map->queues[map->len++] = index; | |
537c00de | 2038 | #ifdef CONFIG_NUMA |
537c00de AD |
2039 | if (numa_node_id == -2) |
2040 | numa_node_id = cpu_to_node(cpu); | |
2041 | else if (numa_node_id != cpu_to_node(cpu)) | |
2042 | numa_node_id = -1; | |
537c00de | 2043 | #endif |
01c5f864 AD |
2044 | } else if (dev_maps) { |
2045 | /* fill in the new device map from the old device map */ | |
2046 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2047 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 2048 | } |
01c5f864 | 2049 | |
537c00de AD |
2050 | } |
2051 | ||
01c5f864 AD |
2052 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2053 | ||
537c00de | 2054 | /* Cleanup old maps */ |
01c5f864 AD |
2055 | if (dev_maps) { |
2056 | for_each_possible_cpu(cpu) { | |
2057 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2058 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2059 | if (map && map != new_map) | |
2060 | kfree_rcu(map, rcu); | |
2061 | } | |
537c00de | 2062 | |
01c5f864 | 2063 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
2064 | } |
2065 | ||
01c5f864 AD |
2066 | dev_maps = new_dev_maps; |
2067 | active = true; | |
537c00de | 2068 | |
01c5f864 AD |
2069 | out_no_new_maps: |
2070 | /* update Tx queue numa node */ | |
537c00de AD |
2071 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2072 | (numa_node_id >= 0) ? numa_node_id : | |
2073 | NUMA_NO_NODE); | |
2074 | ||
01c5f864 AD |
2075 | if (!dev_maps) |
2076 | goto out_no_maps; | |
2077 | ||
2078 | /* removes queue from unused CPUs */ | |
2079 | for_each_possible_cpu(cpu) { | |
2080 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
2081 | continue; | |
2082 | ||
2083 | if (remove_xps_queue(dev_maps, cpu, index)) | |
2084 | active = true; | |
2085 | } | |
2086 | ||
2087 | /* free map if not active */ | |
2088 | if (!active) { | |
2089 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2090 | kfree_rcu(dev_maps, rcu); | |
2091 | } | |
2092 | ||
2093 | out_no_maps: | |
537c00de AD |
2094 | mutex_unlock(&xps_map_mutex); |
2095 | ||
2096 | return 0; | |
2097 | error: | |
01c5f864 AD |
2098 | /* remove any maps that we added */ |
2099 | for_each_possible_cpu(cpu) { | |
2100 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2101 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2102 | NULL; | |
2103 | if (new_map && new_map != map) | |
2104 | kfree(new_map); | |
2105 | } | |
2106 | ||
537c00de AD |
2107 | mutex_unlock(&xps_map_mutex); |
2108 | ||
537c00de AD |
2109 | kfree(new_dev_maps); |
2110 | return -ENOMEM; | |
2111 | } | |
2112 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2113 | ||
2114 | #endif | |
f0796d5c JF |
2115 | /* |
2116 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2117 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2118 | */ | |
e6484930 | 2119 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2120 | { |
1d24eb48 TH |
2121 | int rc; |
2122 | ||
e6484930 TH |
2123 | if (txq < 1 || txq > dev->num_tx_queues) |
2124 | return -EINVAL; | |
f0796d5c | 2125 | |
5c56580b BH |
2126 | if (dev->reg_state == NETREG_REGISTERED || |
2127 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2128 | ASSERT_RTNL(); |
2129 | ||
1d24eb48 TH |
2130 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2131 | txq); | |
bf264145 TH |
2132 | if (rc) |
2133 | return rc; | |
2134 | ||
4f57c087 JF |
2135 | if (dev->num_tc) |
2136 | netif_setup_tc(dev, txq); | |
2137 | ||
024e9679 | 2138 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2139 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2140 | #ifdef CONFIG_XPS |
2141 | netif_reset_xps_queues_gt(dev, txq); | |
2142 | #endif | |
2143 | } | |
f0796d5c | 2144 | } |
e6484930 TH |
2145 | |
2146 | dev->real_num_tx_queues = txq; | |
2147 | return 0; | |
f0796d5c JF |
2148 | } |
2149 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2150 | |
a953be53 | 2151 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2152 | /** |
2153 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2154 | * @dev: Network device | |
2155 | * @rxq: Actual number of RX queues | |
2156 | * | |
2157 | * This must be called either with the rtnl_lock held or before | |
2158 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2159 | * negative error code. If called before registration, it always |
2160 | * succeeds. | |
62fe0b40 BH |
2161 | */ |
2162 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2163 | { | |
2164 | int rc; | |
2165 | ||
bd25fa7b TH |
2166 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2167 | return -EINVAL; | |
2168 | ||
62fe0b40 BH |
2169 | if (dev->reg_state == NETREG_REGISTERED) { |
2170 | ASSERT_RTNL(); | |
2171 | ||
62fe0b40 BH |
2172 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2173 | rxq); | |
2174 | if (rc) | |
2175 | return rc; | |
62fe0b40 BH |
2176 | } |
2177 | ||
2178 | dev->real_num_rx_queues = rxq; | |
2179 | return 0; | |
2180 | } | |
2181 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2182 | #endif | |
2183 | ||
2c53040f BH |
2184 | /** |
2185 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2186 | * |
2187 | * This routine should set an upper limit on the number of RSS queues | |
2188 | * used by default by multiqueue devices. | |
2189 | */ | |
a55b138b | 2190 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
2191 | { |
2192 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
2193 | } | |
2194 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2195 | ||
def82a1d | 2196 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 2197 | { |
def82a1d JP |
2198 | struct softnet_data *sd; |
2199 | unsigned long flags; | |
56079431 | 2200 | |
def82a1d | 2201 | local_irq_save(flags); |
903ceff7 | 2202 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2203 | q->next_sched = NULL; |
2204 | *sd->output_queue_tailp = q; | |
2205 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2206 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2207 | local_irq_restore(flags); | |
2208 | } | |
2209 | ||
2210 | void __netif_schedule(struct Qdisc *q) | |
2211 | { | |
2212 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2213 | __netif_reschedule(q); | |
56079431 DV |
2214 | } |
2215 | EXPORT_SYMBOL(__netif_schedule); | |
2216 | ||
e6247027 ED |
2217 | struct dev_kfree_skb_cb { |
2218 | enum skb_free_reason reason; | |
2219 | }; | |
2220 | ||
2221 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2222 | { |
e6247027 ED |
2223 | return (struct dev_kfree_skb_cb *)skb->cb; |
2224 | } | |
2225 | ||
46e5da40 JF |
2226 | void netif_schedule_queue(struct netdev_queue *txq) |
2227 | { | |
2228 | rcu_read_lock(); | |
2229 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2230 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2231 | ||
2232 | __netif_schedule(q); | |
2233 | } | |
2234 | rcu_read_unlock(); | |
2235 | } | |
2236 | EXPORT_SYMBOL(netif_schedule_queue); | |
2237 | ||
2238 | /** | |
2239 | * netif_wake_subqueue - allow sending packets on subqueue | |
2240 | * @dev: network device | |
2241 | * @queue_index: sub queue index | |
2242 | * | |
2243 | * Resume individual transmit queue of a device with multiple transmit queues. | |
2244 | */ | |
2245 | void netif_wake_subqueue(struct net_device *dev, u16 queue_index) | |
2246 | { | |
2247 | struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index); | |
2248 | ||
2249 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) { | |
2250 | struct Qdisc *q; | |
2251 | ||
2252 | rcu_read_lock(); | |
2253 | q = rcu_dereference(txq->qdisc); | |
2254 | __netif_schedule(q); | |
2255 | rcu_read_unlock(); | |
2256 | } | |
2257 | } | |
2258 | EXPORT_SYMBOL(netif_wake_subqueue); | |
2259 | ||
2260 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) | |
2261 | { | |
2262 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2263 | struct Qdisc *q; | |
2264 | ||
2265 | rcu_read_lock(); | |
2266 | q = rcu_dereference(dev_queue->qdisc); | |
2267 | __netif_schedule(q); | |
2268 | rcu_read_unlock(); | |
2269 | } | |
2270 | } | |
2271 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2272 | ||
e6247027 | 2273 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2274 | { |
e6247027 | 2275 | unsigned long flags; |
56079431 | 2276 | |
e6247027 ED |
2277 | if (likely(atomic_read(&skb->users) == 1)) { |
2278 | smp_rmb(); | |
2279 | atomic_set(&skb->users, 0); | |
2280 | } else if (likely(!atomic_dec_and_test(&skb->users))) { | |
2281 | return; | |
bea3348e | 2282 | } |
e6247027 ED |
2283 | get_kfree_skb_cb(skb)->reason = reason; |
2284 | local_irq_save(flags); | |
2285 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2286 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2287 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2288 | local_irq_restore(flags); | |
56079431 | 2289 | } |
e6247027 | 2290 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2291 | |
e6247027 | 2292 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2293 | { |
2294 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2295 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2296 | else |
2297 | dev_kfree_skb(skb); | |
2298 | } | |
e6247027 | 2299 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2300 | |
2301 | ||
bea3348e SH |
2302 | /** |
2303 | * netif_device_detach - mark device as removed | |
2304 | * @dev: network device | |
2305 | * | |
2306 | * Mark device as removed from system and therefore no longer available. | |
2307 | */ | |
56079431 DV |
2308 | void netif_device_detach(struct net_device *dev) |
2309 | { | |
2310 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2311 | netif_running(dev)) { | |
d543103a | 2312 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2313 | } |
2314 | } | |
2315 | EXPORT_SYMBOL(netif_device_detach); | |
2316 | ||
bea3348e SH |
2317 | /** |
2318 | * netif_device_attach - mark device as attached | |
2319 | * @dev: network device | |
2320 | * | |
2321 | * Mark device as attached from system and restart if needed. | |
2322 | */ | |
56079431 DV |
2323 | void netif_device_attach(struct net_device *dev) |
2324 | { | |
2325 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2326 | netif_running(dev)) { | |
d543103a | 2327 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2328 | __netdev_watchdog_up(dev); |
56079431 DV |
2329 | } |
2330 | } | |
2331 | EXPORT_SYMBOL(netif_device_attach); | |
2332 | ||
36c92474 BH |
2333 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2334 | { | |
65e9d2fa | 2335 | static const netdev_features_t null_features = 0; |
36c92474 BH |
2336 | struct net_device *dev = skb->dev; |
2337 | const char *driver = ""; | |
2338 | ||
c846ad9b BG |
2339 | if (!net_ratelimit()) |
2340 | return; | |
2341 | ||
36c92474 BH |
2342 | if (dev && dev->dev.parent) |
2343 | driver = dev_driver_string(dev->dev.parent); | |
2344 | ||
2345 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " | |
2346 | "gso_type=%d ip_summed=%d\n", | |
65e9d2fa MM |
2347 | driver, dev ? &dev->features : &null_features, |
2348 | skb->sk ? &skb->sk->sk_route_caps : &null_features, | |
36c92474 BH |
2349 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2350 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2351 | } | |
2352 | ||
1da177e4 LT |
2353 | /* |
2354 | * Invalidate hardware checksum when packet is to be mangled, and | |
2355 | * complete checksum manually on outgoing path. | |
2356 | */ | |
84fa7933 | 2357 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2358 | { |
d3bc23e7 | 2359 | __wsum csum; |
663ead3b | 2360 | int ret = 0, offset; |
1da177e4 | 2361 | |
84fa7933 | 2362 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2363 | goto out_set_summed; |
2364 | ||
2365 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2366 | skb_warn_bad_offload(skb); |
2367 | return -EINVAL; | |
1da177e4 LT |
2368 | } |
2369 | ||
cef401de ED |
2370 | /* Before computing a checksum, we should make sure no frag could |
2371 | * be modified by an external entity : checksum could be wrong. | |
2372 | */ | |
2373 | if (skb_has_shared_frag(skb)) { | |
2374 | ret = __skb_linearize(skb); | |
2375 | if (ret) | |
2376 | goto out; | |
2377 | } | |
2378 | ||
55508d60 | 2379 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2380 | BUG_ON(offset >= skb_headlen(skb)); |
2381 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2382 | ||
2383 | offset += skb->csum_offset; | |
2384 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2385 | ||
2386 | if (skb_cloned(skb) && | |
2387 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2388 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2389 | if (ret) | |
2390 | goto out; | |
2391 | } | |
2392 | ||
a030847e | 2393 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2394 | out_set_summed: |
1da177e4 | 2395 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2396 | out: |
1da177e4 LT |
2397 | return ret; |
2398 | } | |
d1b19dff | 2399 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2400 | |
53d6471c | 2401 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2402 | { |
252e3346 | 2403 | __be16 type = skb->protocol; |
f6a78bfc | 2404 | |
19acc327 PS |
2405 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2406 | if (type == htons(ETH_P_TEB)) { | |
2407 | struct ethhdr *eth; | |
2408 | ||
2409 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2410 | return 0; | |
2411 | ||
2412 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2413 | type = eth->h_proto; | |
2414 | } | |
2415 | ||
d4bcef3f | 2416 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
2417 | } |
2418 | ||
2419 | /** | |
2420 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2421 | * @skb: buffer to segment | |
2422 | * @features: features for the output path (see dev->features) | |
2423 | */ | |
2424 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2425 | netdev_features_t features) | |
2426 | { | |
2427 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2428 | struct packet_offload *ptype; | |
53d6471c VY |
2429 | int vlan_depth = skb->mac_len; |
2430 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2431 | |
2432 | if (unlikely(!type)) | |
2433 | return ERR_PTR(-EINVAL); | |
2434 | ||
53d6471c | 2435 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2436 | |
2437 | rcu_read_lock(); | |
22061d80 | 2438 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2439 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2440 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2441 | break; |
2442 | } | |
2443 | } | |
2444 | rcu_read_unlock(); | |
2445 | ||
98e399f8 | 2446 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2447 | |
f6a78bfc HX |
2448 | return segs; |
2449 | } | |
05e8ef4a PS |
2450 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2451 | ||
2452 | ||
2453 | /* openvswitch calls this on rx path, so we need a different check. | |
2454 | */ | |
2455 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2456 | { | |
2457 | if (tx_path) | |
2458 | return skb->ip_summed != CHECKSUM_PARTIAL; | |
2459 | else | |
2460 | return skb->ip_summed == CHECKSUM_NONE; | |
2461 | } | |
2462 | ||
2463 | /** | |
2464 | * __skb_gso_segment - Perform segmentation on skb. | |
2465 | * @skb: buffer to segment | |
2466 | * @features: features for the output path (see dev->features) | |
2467 | * @tx_path: whether it is called in TX path | |
2468 | * | |
2469 | * This function segments the given skb and returns a list of segments. | |
2470 | * | |
2471 | * It may return NULL if the skb requires no segmentation. This is | |
2472 | * only possible when GSO is used for verifying header integrity. | |
2473 | */ | |
2474 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2475 | netdev_features_t features, bool tx_path) | |
2476 | { | |
2477 | if (unlikely(skb_needs_check(skb, tx_path))) { | |
2478 | int err; | |
2479 | ||
2480 | skb_warn_bad_offload(skb); | |
2481 | ||
a40e0a66 | 2482 | err = skb_cow_head(skb, 0); |
2483 | if (err < 0) | |
05e8ef4a PS |
2484 | return ERR_PTR(err); |
2485 | } | |
2486 | ||
68c33163 | 2487 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2488 | SKB_GSO_CB(skb)->encap_level = 0; |
2489 | ||
05e8ef4a PS |
2490 | skb_reset_mac_header(skb); |
2491 | skb_reset_mac_len(skb); | |
2492 | ||
2493 | return skb_mac_gso_segment(skb, features); | |
2494 | } | |
12b0004d | 2495 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2496 | |
fb286bb2 HX |
2497 | /* Take action when hardware reception checksum errors are detected. */ |
2498 | #ifdef CONFIG_BUG | |
2499 | void netdev_rx_csum_fault(struct net_device *dev) | |
2500 | { | |
2501 | if (net_ratelimit()) { | |
7b6cd1ce | 2502 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2503 | dump_stack(); |
2504 | } | |
2505 | } | |
2506 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2507 | #endif | |
2508 | ||
1da177e4 LT |
2509 | /* Actually, we should eliminate this check as soon as we know, that: |
2510 | * 1. IOMMU is present and allows to map all the memory. | |
2511 | * 2. No high memory really exists on this machine. | |
2512 | */ | |
2513 | ||
c1e756bf | 2514 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2515 | { |
3d3a8533 | 2516 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2517 | int i; |
5acbbd42 | 2518 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2519 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2520 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2521 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2522 | return 1; |
ea2ab693 | 2523 | } |
5acbbd42 | 2524 | } |
1da177e4 | 2525 | |
5acbbd42 FT |
2526 | if (PCI_DMA_BUS_IS_PHYS) { |
2527 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2528 | |
9092c658 ED |
2529 | if (!pdev) |
2530 | return 0; | |
5acbbd42 | 2531 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2532 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2533 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2534 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2535 | return 1; | |
2536 | } | |
2537 | } | |
3d3a8533 | 2538 | #endif |
1da177e4 LT |
2539 | return 0; |
2540 | } | |
1da177e4 | 2541 | |
3b392ddb SH |
2542 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2543 | * instead of standard features for the netdev. | |
2544 | */ | |
d0edc7bf | 2545 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
2546 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
2547 | netdev_features_t features, | |
2548 | __be16 type) | |
2549 | { | |
25cd9ba0 | 2550 | if (eth_p_mpls(type)) |
3b392ddb SH |
2551 | features &= skb->dev->mpls_features; |
2552 | ||
2553 | return features; | |
2554 | } | |
2555 | #else | |
2556 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2557 | netdev_features_t features, | |
2558 | __be16 type) | |
2559 | { | |
2560 | return features; | |
2561 | } | |
2562 | #endif | |
2563 | ||
c8f44aff | 2564 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2565 | netdev_features_t features) |
f01a5236 | 2566 | { |
53d6471c | 2567 | int tmp; |
3b392ddb SH |
2568 | __be16 type; |
2569 | ||
2570 | type = skb_network_protocol(skb, &tmp); | |
2571 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2572 | |
c0d680e5 | 2573 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2574 | !can_checksum_protocol(features, type)) { |
f01a5236 | 2575 | features &= ~NETIF_F_ALL_CSUM; |
c1e756bf | 2576 | } else if (illegal_highdma(skb->dev, skb)) { |
f01a5236 JG |
2577 | features &= ~NETIF_F_SG; |
2578 | } | |
2579 | ||
2580 | return features; | |
2581 | } | |
2582 | ||
e38f3025 TM |
2583 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
2584 | struct net_device *dev, | |
2585 | netdev_features_t features) | |
2586 | { | |
2587 | return features; | |
2588 | } | |
2589 | EXPORT_SYMBOL(passthru_features_check); | |
2590 | ||
8cb65d00 TM |
2591 | static netdev_features_t dflt_features_check(const struct sk_buff *skb, |
2592 | struct net_device *dev, | |
2593 | netdev_features_t features) | |
2594 | { | |
2595 | return vlan_features_check(skb, features); | |
2596 | } | |
2597 | ||
c1e756bf | 2598 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 2599 | { |
5f35227e | 2600 | struct net_device *dev = skb->dev; |
fcbeb976 ED |
2601 | netdev_features_t features = dev->features; |
2602 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
58e998c6 | 2603 | |
fcbeb976 | 2604 | if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs) |
30b678d8 BH |
2605 | features &= ~NETIF_F_GSO_MASK; |
2606 | ||
5f35227e JG |
2607 | /* If encapsulation offload request, verify we are testing |
2608 | * hardware encapsulation features instead of standard | |
2609 | * features for the netdev | |
2610 | */ | |
2611 | if (skb->encapsulation) | |
2612 | features &= dev->hw_enc_features; | |
2613 | ||
f5a7fb88 TM |
2614 | if (skb_vlan_tagged(skb)) |
2615 | features = netdev_intersect_features(features, | |
2616 | dev->vlan_features | | |
2617 | NETIF_F_HW_VLAN_CTAG_TX | | |
2618 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2619 | |
5f35227e JG |
2620 | if (dev->netdev_ops->ndo_features_check) |
2621 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
2622 | features); | |
8cb65d00 TM |
2623 | else |
2624 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 2625 | |
c1e756bf | 2626 | return harmonize_features(skb, features); |
58e998c6 | 2627 | } |
c1e756bf | 2628 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2629 | |
2ea25513 | 2630 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2631 | struct netdev_queue *txq, bool more) |
f6a78bfc | 2632 | { |
2ea25513 DM |
2633 | unsigned int len; |
2634 | int rc; | |
00829823 | 2635 | |
7866a621 | 2636 | if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) |
2ea25513 | 2637 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 2638 | |
2ea25513 DM |
2639 | len = skb->len; |
2640 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 2641 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 2642 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 2643 | |
2ea25513 DM |
2644 | return rc; |
2645 | } | |
7b9c6090 | 2646 | |
8dcda22a DM |
2647 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
2648 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
2649 | { |
2650 | struct sk_buff *skb = first; | |
2651 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 2652 | |
7f2e870f DM |
2653 | while (skb) { |
2654 | struct sk_buff *next = skb->next; | |
fc70fb64 | 2655 | |
7f2e870f | 2656 | skb->next = NULL; |
95f6b3dd | 2657 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
2658 | if (unlikely(!dev_xmit_complete(rc))) { |
2659 | skb->next = next; | |
2660 | goto out; | |
2661 | } | |
6afff0ca | 2662 | |
7f2e870f DM |
2663 | skb = next; |
2664 | if (netif_xmit_stopped(txq) && skb) { | |
2665 | rc = NETDEV_TX_BUSY; | |
2666 | break; | |
9ccb8975 | 2667 | } |
7f2e870f | 2668 | } |
9ccb8975 | 2669 | |
7f2e870f DM |
2670 | out: |
2671 | *ret = rc; | |
2672 | return skb; | |
2673 | } | |
b40863c6 | 2674 | |
1ff0dc94 ED |
2675 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
2676 | netdev_features_t features) | |
f6a78bfc | 2677 | { |
df8a39de | 2678 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
2679 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
2680 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
2681 | return skb; |
2682 | } | |
f6a78bfc | 2683 | |
55a93b3e | 2684 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
2685 | { |
2686 | netdev_features_t features; | |
f6a78bfc | 2687 | |
eae3f88e DM |
2688 | if (skb->next) |
2689 | return skb; | |
068a2de5 | 2690 | |
eae3f88e DM |
2691 | features = netif_skb_features(skb); |
2692 | skb = validate_xmit_vlan(skb, features); | |
2693 | if (unlikely(!skb)) | |
2694 | goto out_null; | |
7b9c6090 | 2695 | |
04ffcb25 | 2696 | if (netif_needs_gso(dev, skb, features)) { |
ce93718f DM |
2697 | struct sk_buff *segs; |
2698 | ||
2699 | segs = skb_gso_segment(skb, features); | |
cecda693 | 2700 | if (IS_ERR(segs)) { |
af6dabc9 | 2701 | goto out_kfree_skb; |
cecda693 JW |
2702 | } else if (segs) { |
2703 | consume_skb(skb); | |
2704 | skb = segs; | |
f6a78bfc | 2705 | } |
eae3f88e DM |
2706 | } else { |
2707 | if (skb_needs_linearize(skb, features) && | |
2708 | __skb_linearize(skb)) | |
2709 | goto out_kfree_skb; | |
4ec93edb | 2710 | |
eae3f88e DM |
2711 | /* If packet is not checksummed and device does not |
2712 | * support checksumming for this protocol, complete | |
2713 | * checksumming here. | |
2714 | */ | |
2715 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
2716 | if (skb->encapsulation) | |
2717 | skb_set_inner_transport_header(skb, | |
2718 | skb_checksum_start_offset(skb)); | |
2719 | else | |
2720 | skb_set_transport_header(skb, | |
2721 | skb_checksum_start_offset(skb)); | |
2722 | if (!(features & NETIF_F_ALL_CSUM) && | |
2723 | skb_checksum_help(skb)) | |
2724 | goto out_kfree_skb; | |
7b9c6090 | 2725 | } |
0c772159 | 2726 | } |
7b9c6090 | 2727 | |
eae3f88e | 2728 | return skb; |
fc70fb64 | 2729 | |
f6a78bfc HX |
2730 | out_kfree_skb: |
2731 | kfree_skb(skb); | |
eae3f88e DM |
2732 | out_null: |
2733 | return NULL; | |
2734 | } | |
6afff0ca | 2735 | |
55a93b3e ED |
2736 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
2737 | { | |
2738 | struct sk_buff *next, *head = NULL, *tail; | |
2739 | ||
bec3cfdc | 2740 | for (; skb != NULL; skb = next) { |
55a93b3e ED |
2741 | next = skb->next; |
2742 | skb->next = NULL; | |
bec3cfdc ED |
2743 | |
2744 | /* in case skb wont be segmented, point to itself */ | |
2745 | skb->prev = skb; | |
2746 | ||
55a93b3e | 2747 | skb = validate_xmit_skb(skb, dev); |
bec3cfdc ED |
2748 | if (!skb) |
2749 | continue; | |
55a93b3e | 2750 | |
bec3cfdc ED |
2751 | if (!head) |
2752 | head = skb; | |
2753 | else | |
2754 | tail->next = skb; | |
2755 | /* If skb was segmented, skb->prev points to | |
2756 | * the last segment. If not, it still contains skb. | |
2757 | */ | |
2758 | tail = skb->prev; | |
55a93b3e ED |
2759 | } |
2760 | return head; | |
f6a78bfc HX |
2761 | } |
2762 | ||
1def9238 ED |
2763 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
2764 | { | |
2765 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2766 | ||
2767 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2768 | ||
2769 | /* To get more precise estimation of bytes sent on wire, | |
2770 | * we add to pkt_len the headers size of all segments | |
2771 | */ | |
2772 | if (shinfo->gso_size) { | |
757b8b1d | 2773 | unsigned int hdr_len; |
15e5a030 | 2774 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 2775 | |
757b8b1d ED |
2776 | /* mac layer + network layer */ |
2777 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
2778 | ||
2779 | /* + transport layer */ | |
1def9238 ED |
2780 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
2781 | hdr_len += tcp_hdrlen(skb); | |
2782 | else | |
2783 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
2784 | |
2785 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
2786 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
2787 | shinfo->gso_size); | |
2788 | ||
2789 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
2790 | } |
2791 | } | |
2792 | ||
bbd8a0d3 KK |
2793 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2794 | struct net_device *dev, | |
2795 | struct netdev_queue *txq) | |
2796 | { | |
2797 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2798 | bool contended; |
bbd8a0d3 KK |
2799 | int rc; |
2800 | ||
1def9238 | 2801 | qdisc_pkt_len_init(skb); |
a2da570d | 2802 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
2803 | /* |
2804 | * Heuristic to force contended enqueues to serialize on a | |
2805 | * separate lock before trying to get qdisc main lock. | |
9bf2b8c2 YX |
2806 | * This permits __QDISC___STATE_RUNNING owner to get the lock more |
2807 | * often and dequeue packets faster. | |
79640a4c | 2808 | */ |
a2da570d | 2809 | contended = qdisc_is_running(q); |
79640a4c ED |
2810 | if (unlikely(contended)) |
2811 | spin_lock(&q->busylock); | |
2812 | ||
bbd8a0d3 KK |
2813 | spin_lock(root_lock); |
2814 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2815 | kfree_skb(skb); | |
2816 | rc = NET_XMIT_DROP; | |
2817 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2818 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2819 | /* |
2820 | * This is a work-conserving queue; there are no old skbs | |
2821 | * waiting to be sent out; and the qdisc is not running - | |
2822 | * xmit the skb directly. | |
2823 | */ | |
bfe0d029 | 2824 | |
bfe0d029 ED |
2825 | qdisc_bstats_update(q, skb); |
2826 | ||
55a93b3e | 2827 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
2828 | if (unlikely(contended)) { |
2829 | spin_unlock(&q->busylock); | |
2830 | contended = false; | |
2831 | } | |
bbd8a0d3 | 2832 | __qdisc_run(q); |
79640a4c | 2833 | } else |
bc135b23 | 2834 | qdisc_run_end(q); |
bbd8a0d3 KK |
2835 | |
2836 | rc = NET_XMIT_SUCCESS; | |
2837 | } else { | |
a2da570d | 2838 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2839 | if (qdisc_run_begin(q)) { |
2840 | if (unlikely(contended)) { | |
2841 | spin_unlock(&q->busylock); | |
2842 | contended = false; | |
2843 | } | |
2844 | __qdisc_run(q); | |
2845 | } | |
bbd8a0d3 KK |
2846 | } |
2847 | spin_unlock(root_lock); | |
79640a4c ED |
2848 | if (unlikely(contended)) |
2849 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2850 | return rc; |
2851 | } | |
2852 | ||
86f8515f | 2853 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
2854 | static void skb_update_prio(struct sk_buff *skb) |
2855 | { | |
6977a79d | 2856 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 2857 | |
91c68ce2 ED |
2858 | if (!skb->priority && skb->sk && map) { |
2859 | unsigned int prioidx = skb->sk->sk_cgrp_prioidx; | |
2860 | ||
2861 | if (prioidx < map->priomap_len) | |
2862 | skb->priority = map->priomap[prioidx]; | |
2863 | } | |
5bc1421e NH |
2864 | } |
2865 | #else | |
2866 | #define skb_update_prio(skb) | |
2867 | #endif | |
2868 | ||
745e20f1 | 2869 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2870 | #define RECURSION_LIMIT 10 |
745e20f1 | 2871 | |
95603e22 MM |
2872 | /** |
2873 | * dev_loopback_xmit - loop back @skb | |
2874 | * @skb: buffer to transmit | |
2875 | */ | |
2876 | int dev_loopback_xmit(struct sk_buff *skb) | |
2877 | { | |
2878 | skb_reset_mac_header(skb); | |
2879 | __skb_pull(skb, skb_network_offset(skb)); | |
2880 | skb->pkt_type = PACKET_LOOPBACK; | |
2881 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2882 | WARN_ON(!skb_dst(skb)); | |
2883 | skb_dst_force(skb); | |
2884 | netif_rx_ni(skb); | |
2885 | return 0; | |
2886 | } | |
2887 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2888 | ||
d29f749e | 2889 | /** |
9d08dd3d | 2890 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 2891 | * @skb: buffer to transmit |
9d08dd3d | 2892 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
2893 | * |
2894 | * Queue a buffer for transmission to a network device. The caller must | |
2895 | * have set the device and priority and built the buffer before calling | |
2896 | * this function. The function can be called from an interrupt. | |
2897 | * | |
2898 | * A negative errno code is returned on a failure. A success does not | |
2899 | * guarantee the frame will be transmitted as it may be dropped due | |
2900 | * to congestion or traffic shaping. | |
2901 | * | |
2902 | * ----------------------------------------------------------------------------------- | |
2903 | * I notice this method can also return errors from the queue disciplines, | |
2904 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2905 | * be positive. | |
2906 | * | |
2907 | * Regardless of the return value, the skb is consumed, so it is currently | |
2908 | * difficult to retry a send to this method. (You can bump the ref count | |
2909 | * before sending to hold a reference for retry if you are careful.) | |
2910 | * | |
2911 | * When calling this method, interrupts MUST be enabled. This is because | |
2912 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2913 | * --BLG | |
2914 | */ | |
0a59f3a9 | 2915 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
2916 | { |
2917 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2918 | struct netdev_queue *txq; |
1da177e4 LT |
2919 | struct Qdisc *q; |
2920 | int rc = -ENOMEM; | |
2921 | ||
6d1ccff6 ED |
2922 | skb_reset_mac_header(skb); |
2923 | ||
e7fd2885 WB |
2924 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
2925 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
2926 | ||
4ec93edb YH |
2927 | /* Disable soft irqs for various locks below. Also |
2928 | * stops preemption for RCU. | |
1da177e4 | 2929 | */ |
4ec93edb | 2930 | rcu_read_lock_bh(); |
1da177e4 | 2931 | |
5bc1421e NH |
2932 | skb_update_prio(skb); |
2933 | ||
02875878 ED |
2934 | /* If device/qdisc don't need skb->dst, release it right now while |
2935 | * its hot in this cpu cache. | |
2936 | */ | |
2937 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2938 | skb_dst_drop(skb); | |
2939 | else | |
2940 | skb_dst_force(skb); | |
2941 | ||
f663dd9a | 2942 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 2943 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2944 | |
1da177e4 | 2945 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2946 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2947 | #endif |
cf66ba58 | 2948 | trace_net_dev_queue(skb); |
1da177e4 | 2949 | if (q->enqueue) { |
bbd8a0d3 | 2950 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2951 | goto out; |
1da177e4 LT |
2952 | } |
2953 | ||
2954 | /* The device has no queue. Common case for software devices: | |
2955 | loopback, all the sorts of tunnels... | |
2956 | ||
932ff279 HX |
2957 | Really, it is unlikely that netif_tx_lock protection is necessary |
2958 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2959 | counters.) |
2960 | However, it is possible, that they rely on protection | |
2961 | made by us here. | |
2962 | ||
2963 | Check this and shot the lock. It is not prone from deadlocks. | |
2964 | Either shot noqueue qdisc, it is even simpler 8) | |
2965 | */ | |
2966 | if (dev->flags & IFF_UP) { | |
2967 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2968 | ||
c773e847 | 2969 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2970 | |
745e20f1 ED |
2971 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2972 | goto recursion_alert; | |
2973 | ||
1f59533f JDB |
2974 | skb = validate_xmit_skb(skb, dev); |
2975 | if (!skb) | |
2976 | goto drop; | |
2977 | ||
c773e847 | 2978 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2979 | |
73466498 | 2980 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 2981 | __this_cpu_inc(xmit_recursion); |
ce93718f | 2982 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 2983 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2984 | if (dev_xmit_complete(rc)) { |
c773e847 | 2985 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2986 | goto out; |
2987 | } | |
2988 | } | |
c773e847 | 2989 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
2990 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
2991 | dev->name); | |
1da177e4 LT |
2992 | } else { |
2993 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2994 | * unfortunately |
2995 | */ | |
2996 | recursion_alert: | |
e87cc472 JP |
2997 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
2998 | dev->name); | |
1da177e4 LT |
2999 | } |
3000 | } | |
3001 | ||
3002 | rc = -ENETDOWN; | |
1f59533f | 3003 | drop: |
d4828d85 | 3004 | rcu_read_unlock_bh(); |
1da177e4 | 3005 | |
015f0688 | 3006 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 3007 | kfree_skb_list(skb); |
1da177e4 LT |
3008 | return rc; |
3009 | out: | |
d4828d85 | 3010 | rcu_read_unlock_bh(); |
1da177e4 LT |
3011 | return rc; |
3012 | } | |
f663dd9a JW |
3013 | |
3014 | int dev_queue_xmit(struct sk_buff *skb) | |
3015 | { | |
3016 | return __dev_queue_xmit(skb, NULL); | |
3017 | } | |
d1b19dff | 3018 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3019 | |
f663dd9a JW |
3020 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3021 | { | |
3022 | return __dev_queue_xmit(skb, accel_priv); | |
3023 | } | |
3024 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3025 | ||
1da177e4 LT |
3026 | |
3027 | /*======================================================================= | |
3028 | Receiver routines | |
3029 | =======================================================================*/ | |
3030 | ||
6b2bedc3 | 3031 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3032 | EXPORT_SYMBOL(netdev_max_backlog); |
3033 | ||
3b098e2d | 3034 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
3035 | int netdev_budget __read_mostly = 300; |
3036 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 3037 | |
eecfd7c4 ED |
3038 | /* Called with irq disabled */ |
3039 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3040 | struct napi_struct *napi) | |
3041 | { | |
3042 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3043 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3044 | } | |
3045 | ||
bfb564e7 KK |
3046 | #ifdef CONFIG_RPS |
3047 | ||
3048 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3049 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 3050 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
3051 | u32 rps_cpu_mask __read_mostly; |
3052 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 3053 | |
c5905afb | 3054 | struct static_key rps_needed __read_mostly; |
adc9300e | 3055 | |
c445477d BH |
3056 | static struct rps_dev_flow * |
3057 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3058 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3059 | { | |
09994d1b | 3060 | if (next_cpu != RPS_NO_CPU) { |
c445477d BH |
3061 | #ifdef CONFIG_RFS_ACCEL |
3062 | struct netdev_rx_queue *rxqueue; | |
3063 | struct rps_dev_flow_table *flow_table; | |
3064 | struct rps_dev_flow *old_rflow; | |
3065 | u32 flow_id; | |
3066 | u16 rxq_index; | |
3067 | int rc; | |
3068 | ||
3069 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3070 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3071 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3072 | goto out; |
3073 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3074 | if (rxq_index == skb_get_rx_queue(skb)) | |
3075 | goto out; | |
3076 | ||
3077 | rxqueue = dev->_rx + rxq_index; | |
3078 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3079 | if (!flow_table) | |
3080 | goto out; | |
61b905da | 3081 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3082 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3083 | rxq_index, flow_id); | |
3084 | if (rc < 0) | |
3085 | goto out; | |
3086 | old_rflow = rflow; | |
3087 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3088 | rflow->filter = rc; |
3089 | if (old_rflow->filter == rflow->filter) | |
3090 | old_rflow->filter = RPS_NO_FILTER; | |
3091 | out: | |
3092 | #endif | |
3093 | rflow->last_qtail = | |
09994d1b | 3094 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3095 | } |
3096 | ||
09994d1b | 3097 | rflow->cpu = next_cpu; |
c445477d BH |
3098 | return rflow; |
3099 | } | |
3100 | ||
bfb564e7 KK |
3101 | /* |
3102 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3103 | * CPU from the RPS map of the receiving queue for a given skb. | |
3104 | * rcu_read_lock must be held on entry. | |
3105 | */ | |
3106 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3107 | struct rps_dev_flow **rflowp) | |
3108 | { | |
567e4b79 ED |
3109 | const struct rps_sock_flow_table *sock_flow_table; |
3110 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 3111 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 3112 | struct rps_map *map; |
bfb564e7 | 3113 | int cpu = -1; |
567e4b79 | 3114 | u32 tcpu; |
61b905da | 3115 | u32 hash; |
bfb564e7 KK |
3116 | |
3117 | if (skb_rx_queue_recorded(skb)) { | |
3118 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 3119 | |
62fe0b40 BH |
3120 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3121 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3122 | "%s received packet on queue %u, but number " | |
3123 | "of RX queues is %u\n", | |
3124 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3125 | goto done; |
3126 | } | |
567e4b79 ED |
3127 | rxqueue += index; |
3128 | } | |
bfb564e7 | 3129 | |
567e4b79 ED |
3130 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
3131 | ||
3132 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 3133 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 3134 | if (!flow_table && !map) |
bfb564e7 KK |
3135 | goto done; |
3136 | ||
2d47b459 | 3137 | skb_reset_network_header(skb); |
61b905da TH |
3138 | hash = skb_get_hash(skb); |
3139 | if (!hash) | |
bfb564e7 KK |
3140 | goto done; |
3141 | ||
fec5e652 TH |
3142 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
3143 | if (flow_table && sock_flow_table) { | |
fec5e652 | 3144 | struct rps_dev_flow *rflow; |
567e4b79 ED |
3145 | u32 next_cpu; |
3146 | u32 ident; | |
3147 | ||
3148 | /* First check into global flow table if there is a match */ | |
3149 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
3150 | if ((ident ^ hash) & ~rps_cpu_mask) | |
3151 | goto try_rps; | |
fec5e652 | 3152 | |
567e4b79 ED |
3153 | next_cpu = ident & rps_cpu_mask; |
3154 | ||
3155 | /* OK, now we know there is a match, | |
3156 | * we can look at the local (per receive queue) flow table | |
3157 | */ | |
61b905da | 3158 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3159 | tcpu = rflow->cpu; |
3160 | ||
fec5e652 TH |
3161 | /* |
3162 | * If the desired CPU (where last recvmsg was done) is | |
3163 | * different from current CPU (one in the rx-queue flow | |
3164 | * table entry), switch if one of the following holds: | |
3165 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
3166 | * - Current CPU is offline. | |
3167 | * - The current CPU's queue tail has advanced beyond the | |
3168 | * last packet that was enqueued using this table entry. | |
3169 | * This guarantees that all previous packets for the flow | |
3170 | * have been dequeued, thus preserving in order delivery. | |
3171 | */ | |
3172 | if (unlikely(tcpu != next_cpu) && | |
3173 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
3174 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
baefa31d TH |
3175 | rflow->last_qtail)) >= 0)) { |
3176 | tcpu = next_cpu; | |
c445477d | 3177 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3178 | } |
c445477d | 3179 | |
fec5e652 TH |
3180 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
3181 | *rflowp = rflow; | |
3182 | cpu = tcpu; | |
3183 | goto done; | |
3184 | } | |
3185 | } | |
3186 | ||
567e4b79 ED |
3187 | try_rps: |
3188 | ||
0a9627f2 | 3189 | if (map) { |
8fc54f68 | 3190 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3191 | if (cpu_online(tcpu)) { |
3192 | cpu = tcpu; | |
3193 | goto done; | |
3194 | } | |
3195 | } | |
3196 | ||
3197 | done: | |
0a9627f2 TH |
3198 | return cpu; |
3199 | } | |
3200 | ||
c445477d BH |
3201 | #ifdef CONFIG_RFS_ACCEL |
3202 | ||
3203 | /** | |
3204 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3205 | * @dev: Device on which the filter was set | |
3206 | * @rxq_index: RX queue index | |
3207 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3208 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3209 | * | |
3210 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3211 | * this function for each installed filter and remove the filters for | |
3212 | * which it returns %true. | |
3213 | */ | |
3214 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3215 | u32 flow_id, u16 filter_id) | |
3216 | { | |
3217 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3218 | struct rps_dev_flow_table *flow_table; | |
3219 | struct rps_dev_flow *rflow; | |
3220 | bool expire = true; | |
3221 | int cpu; | |
3222 | ||
3223 | rcu_read_lock(); | |
3224 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3225 | if (flow_table && flow_id <= flow_table->mask) { | |
3226 | rflow = &flow_table->flows[flow_id]; | |
3227 | cpu = ACCESS_ONCE(rflow->cpu); | |
3228 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
3229 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
3230 | rflow->last_qtail) < | |
3231 | (int)(10 * flow_table->mask))) | |
3232 | expire = false; | |
3233 | } | |
3234 | rcu_read_unlock(); | |
3235 | return expire; | |
3236 | } | |
3237 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3238 | ||
3239 | #endif /* CONFIG_RFS_ACCEL */ | |
3240 | ||
0a9627f2 | 3241 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3242 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3243 | { |
e36fa2f7 ED |
3244 | struct softnet_data *sd = data; |
3245 | ||
eecfd7c4 | 3246 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3247 | sd->received_rps++; |
0a9627f2 | 3248 | } |
e36fa2f7 | 3249 | |
fec5e652 | 3250 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3251 | |
e36fa2f7 ED |
3252 | /* |
3253 | * Check if this softnet_data structure is another cpu one | |
3254 | * If yes, queue it to our IPI list and return 1 | |
3255 | * If no, return 0 | |
3256 | */ | |
3257 | static int rps_ipi_queued(struct softnet_data *sd) | |
3258 | { | |
3259 | #ifdef CONFIG_RPS | |
903ceff7 | 3260 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
3261 | |
3262 | if (sd != mysd) { | |
3263 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3264 | mysd->rps_ipi_list = sd; | |
3265 | ||
3266 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3267 | return 1; | |
3268 | } | |
3269 | #endif /* CONFIG_RPS */ | |
3270 | return 0; | |
3271 | } | |
3272 | ||
99bbc707 WB |
3273 | #ifdef CONFIG_NET_FLOW_LIMIT |
3274 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3275 | #endif | |
3276 | ||
3277 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3278 | { | |
3279 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3280 | struct sd_flow_limit *fl; | |
3281 | struct softnet_data *sd; | |
3282 | unsigned int old_flow, new_flow; | |
3283 | ||
3284 | if (qlen < (netdev_max_backlog >> 1)) | |
3285 | return false; | |
3286 | ||
903ceff7 | 3287 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
3288 | |
3289 | rcu_read_lock(); | |
3290 | fl = rcu_dereference(sd->flow_limit); | |
3291 | if (fl) { | |
3958afa1 | 3292 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3293 | old_flow = fl->history[fl->history_head]; |
3294 | fl->history[fl->history_head] = new_flow; | |
3295 | ||
3296 | fl->history_head++; | |
3297 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3298 | ||
3299 | if (likely(fl->buckets[old_flow])) | |
3300 | fl->buckets[old_flow]--; | |
3301 | ||
3302 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3303 | fl->count++; | |
3304 | rcu_read_unlock(); | |
3305 | return true; | |
3306 | } | |
3307 | } | |
3308 | rcu_read_unlock(); | |
3309 | #endif | |
3310 | return false; | |
3311 | } | |
3312 | ||
0a9627f2 TH |
3313 | /* |
3314 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3315 | * queue (may be a remote CPU queue). | |
3316 | */ | |
fec5e652 TH |
3317 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3318 | unsigned int *qtail) | |
0a9627f2 | 3319 | { |
e36fa2f7 | 3320 | struct softnet_data *sd; |
0a9627f2 | 3321 | unsigned long flags; |
99bbc707 | 3322 | unsigned int qlen; |
0a9627f2 | 3323 | |
e36fa2f7 | 3324 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3325 | |
3326 | local_irq_save(flags); | |
0a9627f2 | 3327 | |
e36fa2f7 | 3328 | rps_lock(sd); |
99bbc707 WB |
3329 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3330 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 3331 | if (qlen) { |
0a9627f2 | 3332 | enqueue: |
e36fa2f7 | 3333 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3334 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3335 | rps_unlock(sd); |
152102c7 | 3336 | local_irq_restore(flags); |
0a9627f2 TH |
3337 | return NET_RX_SUCCESS; |
3338 | } | |
3339 | ||
ebda37c2 ED |
3340 | /* Schedule NAPI for backlog device |
3341 | * We can use non atomic operation since we own the queue lock | |
3342 | */ | |
3343 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3344 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3345 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3346 | } |
3347 | goto enqueue; | |
3348 | } | |
3349 | ||
dee42870 | 3350 | sd->dropped++; |
e36fa2f7 | 3351 | rps_unlock(sd); |
0a9627f2 | 3352 | |
0a9627f2 TH |
3353 | local_irq_restore(flags); |
3354 | ||
caf586e5 | 3355 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3356 | kfree_skb(skb); |
3357 | return NET_RX_DROP; | |
3358 | } | |
1da177e4 | 3359 | |
ae78dbfa | 3360 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3361 | { |
b0e28f1e | 3362 | int ret; |
1da177e4 | 3363 | |
588f0330 | 3364 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3365 | |
cf66ba58 | 3366 | trace_netif_rx(skb); |
df334545 | 3367 | #ifdef CONFIG_RPS |
c5905afb | 3368 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3369 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3370 | int cpu; |
3371 | ||
cece1945 | 3372 | preempt_disable(); |
b0e28f1e | 3373 | rcu_read_lock(); |
fec5e652 TH |
3374 | |
3375 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3376 | if (cpu < 0) |
3377 | cpu = smp_processor_id(); | |
fec5e652 TH |
3378 | |
3379 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3380 | ||
b0e28f1e | 3381 | rcu_read_unlock(); |
cece1945 | 3382 | preempt_enable(); |
adc9300e ED |
3383 | } else |
3384 | #endif | |
fec5e652 TH |
3385 | { |
3386 | unsigned int qtail; | |
3387 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3388 | put_cpu(); | |
3389 | } | |
b0e28f1e | 3390 | return ret; |
1da177e4 | 3391 | } |
ae78dbfa BH |
3392 | |
3393 | /** | |
3394 | * netif_rx - post buffer to the network code | |
3395 | * @skb: buffer to post | |
3396 | * | |
3397 | * This function receives a packet from a device driver and queues it for | |
3398 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3399 | * may be dropped during processing for congestion control or by the | |
3400 | * protocol layers. | |
3401 | * | |
3402 | * return values: | |
3403 | * NET_RX_SUCCESS (no congestion) | |
3404 | * NET_RX_DROP (packet was dropped) | |
3405 | * | |
3406 | */ | |
3407 | ||
3408 | int netif_rx(struct sk_buff *skb) | |
3409 | { | |
3410 | trace_netif_rx_entry(skb); | |
3411 | ||
3412 | return netif_rx_internal(skb); | |
3413 | } | |
d1b19dff | 3414 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3415 | |
3416 | int netif_rx_ni(struct sk_buff *skb) | |
3417 | { | |
3418 | int err; | |
3419 | ||
ae78dbfa BH |
3420 | trace_netif_rx_ni_entry(skb); |
3421 | ||
1da177e4 | 3422 | preempt_disable(); |
ae78dbfa | 3423 | err = netif_rx_internal(skb); |
1da177e4 LT |
3424 | if (local_softirq_pending()) |
3425 | do_softirq(); | |
3426 | preempt_enable(); | |
3427 | ||
3428 | return err; | |
3429 | } | |
1da177e4 LT |
3430 | EXPORT_SYMBOL(netif_rx_ni); |
3431 | ||
1da177e4 LT |
3432 | static void net_tx_action(struct softirq_action *h) |
3433 | { | |
903ceff7 | 3434 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
3435 | |
3436 | if (sd->completion_queue) { | |
3437 | struct sk_buff *clist; | |
3438 | ||
3439 | local_irq_disable(); | |
3440 | clist = sd->completion_queue; | |
3441 | sd->completion_queue = NULL; | |
3442 | local_irq_enable(); | |
3443 | ||
3444 | while (clist) { | |
3445 | struct sk_buff *skb = clist; | |
3446 | clist = clist->next; | |
3447 | ||
547b792c | 3448 | WARN_ON(atomic_read(&skb->users)); |
e6247027 ED |
3449 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
3450 | trace_consume_skb(skb); | |
3451 | else | |
3452 | trace_kfree_skb(skb, net_tx_action); | |
1da177e4 LT |
3453 | __kfree_skb(skb); |
3454 | } | |
3455 | } | |
3456 | ||
3457 | if (sd->output_queue) { | |
37437bb2 | 3458 | struct Qdisc *head; |
1da177e4 LT |
3459 | |
3460 | local_irq_disable(); | |
3461 | head = sd->output_queue; | |
3462 | sd->output_queue = NULL; | |
a9cbd588 | 3463 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3464 | local_irq_enable(); |
3465 | ||
3466 | while (head) { | |
37437bb2 DM |
3467 | struct Qdisc *q = head; |
3468 | spinlock_t *root_lock; | |
3469 | ||
1da177e4 LT |
3470 | head = head->next_sched; |
3471 | ||
5fb66229 | 3472 | root_lock = qdisc_lock(q); |
37437bb2 | 3473 | if (spin_trylock(root_lock)) { |
4e857c58 | 3474 | smp_mb__before_atomic(); |
def82a1d JP |
3475 | clear_bit(__QDISC_STATE_SCHED, |
3476 | &q->state); | |
37437bb2 DM |
3477 | qdisc_run(q); |
3478 | spin_unlock(root_lock); | |
1da177e4 | 3479 | } else { |
195648bb | 3480 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3481 | &q->state)) { |
195648bb | 3482 | __netif_reschedule(q); |
e8a83e10 | 3483 | } else { |
4e857c58 | 3484 | smp_mb__before_atomic(); |
e8a83e10 JP |
3485 | clear_bit(__QDISC_STATE_SCHED, |
3486 | &q->state); | |
3487 | } | |
1da177e4 LT |
3488 | } |
3489 | } | |
3490 | } | |
3491 | } | |
3492 | ||
ab95bfe0 JP |
3493 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3494 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3495 | /* This hook is defined here for ATM LANE */ |
3496 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3497 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3498 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3499 | #endif |
1da177e4 | 3500 | |
1da177e4 LT |
3501 | #ifdef CONFIG_NET_CLS_ACT |
3502 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3503 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3504 | * a compare and 2 stores extra right now if we dont have it on | |
3505 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3506 | * NOTE: This doesn't stop any functionality; if you dont have |
3507 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3508 | * |
3509 | */ | |
24824a09 | 3510 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3511 | { |
1da177e4 | 3512 | struct net_device *dev = skb->dev; |
f697c3e8 | 3513 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3514 | int result = TC_ACT_OK; |
3515 | struct Qdisc *q; | |
4ec93edb | 3516 | |
de384830 | 3517 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
e87cc472 JP |
3518 | net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n", |
3519 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3520 | return TC_ACT_SHOT; |
3521 | } | |
1da177e4 | 3522 | |
f697c3e8 HX |
3523 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3524 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3525 | |
46e5da40 | 3526 | q = rcu_dereference(rxq->qdisc); |
8d50b53d | 3527 | if (q != &noop_qdisc) { |
83874000 | 3528 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3529 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3530 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3531 | spin_unlock(qdisc_lock(q)); |
3532 | } | |
f697c3e8 HX |
3533 | |
3534 | return result; | |
3535 | } | |
86e65da9 | 3536 | |
f697c3e8 HX |
3537 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3538 | struct packet_type **pt_prev, | |
3539 | int *ret, struct net_device *orig_dev) | |
3540 | { | |
24824a09 ED |
3541 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3542 | ||
46e5da40 | 3543 | if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc) |
f697c3e8 | 3544 | goto out; |
1da177e4 | 3545 | |
f697c3e8 HX |
3546 | if (*pt_prev) { |
3547 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3548 | *pt_prev = NULL; | |
1da177e4 LT |
3549 | } |
3550 | ||
24824a09 | 3551 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3552 | case TC_ACT_SHOT: |
3553 | case TC_ACT_STOLEN: | |
3554 | kfree_skb(skb); | |
3555 | return NULL; | |
3556 | } | |
3557 | ||
3558 | out: | |
3559 | skb->tc_verd = 0; | |
3560 | return skb; | |
1da177e4 LT |
3561 | } |
3562 | #endif | |
3563 | ||
ab95bfe0 JP |
3564 | /** |
3565 | * netdev_rx_handler_register - register receive handler | |
3566 | * @dev: device to register a handler for | |
3567 | * @rx_handler: receive handler to register | |
93e2c32b | 3568 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 3569 | * |
e227867f | 3570 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
3571 | * called from __netif_receive_skb. A negative errno code is returned |
3572 | * on a failure. | |
3573 | * | |
3574 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3575 | * |
3576 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3577 | */ |
3578 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3579 | rx_handler_func_t *rx_handler, |
3580 | void *rx_handler_data) | |
ab95bfe0 JP |
3581 | { |
3582 | ASSERT_RTNL(); | |
3583 | ||
3584 | if (dev->rx_handler) | |
3585 | return -EBUSY; | |
3586 | ||
00cfec37 | 3587 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 3588 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3589 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3590 | ||
3591 | return 0; | |
3592 | } | |
3593 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3594 | ||
3595 | /** | |
3596 | * netdev_rx_handler_unregister - unregister receive handler | |
3597 | * @dev: device to unregister a handler from | |
3598 | * | |
166ec369 | 3599 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
3600 | * |
3601 | * The caller must hold the rtnl_mutex. | |
3602 | */ | |
3603 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3604 | { | |
3605 | ||
3606 | ASSERT_RTNL(); | |
a9b3cd7f | 3607 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
3608 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
3609 | * section has a guarantee to see a non NULL rx_handler_data | |
3610 | * as well. | |
3611 | */ | |
3612 | synchronize_net(); | |
a9b3cd7f | 3613 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
3614 | } |
3615 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3616 | ||
b4b9e355 MG |
3617 | /* |
3618 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3619 | * the special handling of PFMEMALLOC skbs. | |
3620 | */ | |
3621 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3622 | { | |
3623 | switch (skb->protocol) { | |
2b8837ae JP |
3624 | case htons(ETH_P_ARP): |
3625 | case htons(ETH_P_IP): | |
3626 | case htons(ETH_P_IPV6): | |
3627 | case htons(ETH_P_8021Q): | |
3628 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
3629 | return true; |
3630 | default: | |
3631 | return false; | |
3632 | } | |
3633 | } | |
3634 | ||
9754e293 | 3635 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
3636 | { |
3637 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3638 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3639 | struct net_device *orig_dev; |
8a4eb573 | 3640 | bool deliver_exact = false; |
1da177e4 | 3641 | int ret = NET_RX_DROP; |
252e3346 | 3642 | __be16 type; |
1da177e4 | 3643 | |
588f0330 | 3644 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3645 | |
cf66ba58 | 3646 | trace_netif_receive_skb(skb); |
9b22ea56 | 3647 | |
cc9bd5ce | 3648 | orig_dev = skb->dev; |
8f903c70 | 3649 | |
c1d2bbe1 | 3650 | skb_reset_network_header(skb); |
fda55eca ED |
3651 | if (!skb_transport_header_was_set(skb)) |
3652 | skb_reset_transport_header(skb); | |
0b5c9db1 | 3653 | skb_reset_mac_len(skb); |
1da177e4 LT |
3654 | |
3655 | pt_prev = NULL; | |
3656 | ||
3657 | rcu_read_lock(); | |
3658 | ||
63d8ea7f | 3659 | another_round: |
b6858177 | 3660 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
3661 | |
3662 | __this_cpu_inc(softnet_data.processed); | |
3663 | ||
8ad227ff PM |
3664 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
3665 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 3666 | skb = skb_vlan_untag(skb); |
bcc6d479 | 3667 | if (unlikely(!skb)) |
b4b9e355 | 3668 | goto unlock; |
bcc6d479 JP |
3669 | } |
3670 | ||
1da177e4 LT |
3671 | #ifdef CONFIG_NET_CLS_ACT |
3672 | if (skb->tc_verd & TC_NCLS) { | |
3673 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3674 | goto ncls; | |
3675 | } | |
3676 | #endif | |
3677 | ||
9754e293 | 3678 | if (pfmemalloc) |
b4b9e355 MG |
3679 | goto skip_taps; |
3680 | ||
1da177e4 | 3681 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
3682 | if (pt_prev) |
3683 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3684 | pt_prev = ptype; | |
3685 | } | |
3686 | ||
3687 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
3688 | if (pt_prev) | |
3689 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3690 | pt_prev = ptype; | |
1da177e4 LT |
3691 | } |
3692 | ||
b4b9e355 | 3693 | skip_taps: |
1da177e4 | 3694 | #ifdef CONFIG_NET_CLS_ACT |
f697c3e8 HX |
3695 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3696 | if (!skb) | |
b4b9e355 | 3697 | goto unlock; |
1da177e4 LT |
3698 | ncls: |
3699 | #endif | |
3700 | ||
9754e293 | 3701 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
3702 | goto drop; |
3703 | ||
df8a39de | 3704 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
3705 | if (pt_prev) { |
3706 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3707 | pt_prev = NULL; | |
3708 | } | |
48cc32d3 | 3709 | if (vlan_do_receive(&skb)) |
2425717b JF |
3710 | goto another_round; |
3711 | else if (unlikely(!skb)) | |
b4b9e355 | 3712 | goto unlock; |
2425717b JF |
3713 | } |
3714 | ||
48cc32d3 | 3715 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
3716 | if (rx_handler) { |
3717 | if (pt_prev) { | |
3718 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3719 | pt_prev = NULL; | |
3720 | } | |
8a4eb573 JP |
3721 | switch (rx_handler(&skb)) { |
3722 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 3723 | ret = NET_RX_SUCCESS; |
b4b9e355 | 3724 | goto unlock; |
8a4eb573 | 3725 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3726 | goto another_round; |
8a4eb573 JP |
3727 | case RX_HANDLER_EXACT: |
3728 | deliver_exact = true; | |
3729 | case RX_HANDLER_PASS: | |
3730 | break; | |
3731 | default: | |
3732 | BUG(); | |
3733 | } | |
ab95bfe0 | 3734 | } |
1da177e4 | 3735 | |
df8a39de JP |
3736 | if (unlikely(skb_vlan_tag_present(skb))) { |
3737 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
3738 | skb->pkt_type = PACKET_OTHERHOST; |
3739 | /* Note: we might in the future use prio bits | |
3740 | * and set skb->priority like in vlan_do_receive() | |
3741 | * For the time being, just ignore Priority Code Point | |
3742 | */ | |
3743 | skb->vlan_tci = 0; | |
3744 | } | |
48cc32d3 | 3745 | |
7866a621 SN |
3746 | type = skb->protocol; |
3747 | ||
63d8ea7f | 3748 | /* deliver only exact match when indicated */ |
7866a621 SN |
3749 | if (likely(!deliver_exact)) { |
3750 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
3751 | &ptype_base[ntohs(type) & | |
3752 | PTYPE_HASH_MASK]); | |
3753 | } | |
1f3c8804 | 3754 | |
7866a621 SN |
3755 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
3756 | &orig_dev->ptype_specific); | |
3757 | ||
3758 | if (unlikely(skb->dev != orig_dev)) { | |
3759 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
3760 | &skb->dev->ptype_specific); | |
1da177e4 LT |
3761 | } |
3762 | ||
3763 | if (pt_prev) { | |
1080e512 | 3764 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 3765 | goto drop; |
1080e512 MT |
3766 | else |
3767 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 3768 | } else { |
b4b9e355 | 3769 | drop: |
caf586e5 | 3770 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3771 | kfree_skb(skb); |
3772 | /* Jamal, now you will not able to escape explaining | |
3773 | * me how you were going to use this. :-) | |
3774 | */ | |
3775 | ret = NET_RX_DROP; | |
3776 | } | |
3777 | ||
b4b9e355 | 3778 | unlock: |
1da177e4 | 3779 | rcu_read_unlock(); |
9754e293 DM |
3780 | return ret; |
3781 | } | |
3782 | ||
3783 | static int __netif_receive_skb(struct sk_buff *skb) | |
3784 | { | |
3785 | int ret; | |
3786 | ||
3787 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
3788 | unsigned long pflags = current->flags; | |
3789 | ||
3790 | /* | |
3791 | * PFMEMALLOC skbs are special, they should | |
3792 | * - be delivered to SOCK_MEMALLOC sockets only | |
3793 | * - stay away from userspace | |
3794 | * - have bounded memory usage | |
3795 | * | |
3796 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
3797 | * context down to all allocation sites. | |
3798 | */ | |
3799 | current->flags |= PF_MEMALLOC; | |
3800 | ret = __netif_receive_skb_core(skb, true); | |
3801 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
3802 | } else | |
3803 | ret = __netif_receive_skb_core(skb, false); | |
3804 | ||
1da177e4 LT |
3805 | return ret; |
3806 | } | |
0a9627f2 | 3807 | |
ae78dbfa | 3808 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 3809 | { |
588f0330 | 3810 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3811 | |
c1f19b51 RC |
3812 | if (skb_defer_rx_timestamp(skb)) |
3813 | return NET_RX_SUCCESS; | |
3814 | ||
df334545 | 3815 | #ifdef CONFIG_RPS |
c5905afb | 3816 | if (static_key_false(&rps_needed)) { |
3b098e2d ED |
3817 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
3818 | int cpu, ret; | |
fec5e652 | 3819 | |
3b098e2d ED |
3820 | rcu_read_lock(); |
3821 | ||
3822 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3823 | |
3b098e2d ED |
3824 | if (cpu >= 0) { |
3825 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3826 | rcu_read_unlock(); | |
adc9300e | 3827 | return ret; |
3b098e2d | 3828 | } |
adc9300e | 3829 | rcu_read_unlock(); |
fec5e652 | 3830 | } |
1e94d72f | 3831 | #endif |
adc9300e | 3832 | return __netif_receive_skb(skb); |
0a9627f2 | 3833 | } |
ae78dbfa BH |
3834 | |
3835 | /** | |
3836 | * netif_receive_skb - process receive buffer from network | |
3837 | * @skb: buffer to process | |
3838 | * | |
3839 | * netif_receive_skb() is the main receive data processing function. | |
3840 | * It always succeeds. The buffer may be dropped during processing | |
3841 | * for congestion control or by the protocol layers. | |
3842 | * | |
3843 | * This function may only be called from softirq context and interrupts | |
3844 | * should be enabled. | |
3845 | * | |
3846 | * Return values (usually ignored): | |
3847 | * NET_RX_SUCCESS: no congestion | |
3848 | * NET_RX_DROP: packet was dropped | |
3849 | */ | |
3850 | int netif_receive_skb(struct sk_buff *skb) | |
3851 | { | |
3852 | trace_netif_receive_skb_entry(skb); | |
3853 | ||
3854 | return netif_receive_skb_internal(skb); | |
3855 | } | |
d1b19dff | 3856 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3857 | |
88751275 ED |
3858 | /* Network device is going away, flush any packets still pending |
3859 | * Called with irqs disabled. | |
3860 | */ | |
152102c7 | 3861 | static void flush_backlog(void *arg) |
6e583ce5 | 3862 | { |
152102c7 | 3863 | struct net_device *dev = arg; |
903ceff7 | 3864 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
6e583ce5 SH |
3865 | struct sk_buff *skb, *tmp; |
3866 | ||
e36fa2f7 | 3867 | rps_lock(sd); |
6e7676c1 | 3868 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3869 | if (skb->dev == dev) { |
e36fa2f7 | 3870 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3871 | kfree_skb(skb); |
76cc8b13 | 3872 | input_queue_head_incr(sd); |
6e583ce5 | 3873 | } |
6e7676c1 | 3874 | } |
e36fa2f7 | 3875 | rps_unlock(sd); |
6e7676c1 CG |
3876 | |
3877 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3878 | if (skb->dev == dev) { | |
3879 | __skb_unlink(skb, &sd->process_queue); | |
3880 | kfree_skb(skb); | |
76cc8b13 | 3881 | input_queue_head_incr(sd); |
6e7676c1 CG |
3882 | } |
3883 | } | |
6e583ce5 SH |
3884 | } |
3885 | ||
d565b0a1 HX |
3886 | static int napi_gro_complete(struct sk_buff *skb) |
3887 | { | |
22061d80 | 3888 | struct packet_offload *ptype; |
d565b0a1 | 3889 | __be16 type = skb->protocol; |
22061d80 | 3890 | struct list_head *head = &offload_base; |
d565b0a1 HX |
3891 | int err = -ENOENT; |
3892 | ||
c3c7c254 ED |
3893 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
3894 | ||
fc59f9a3 HX |
3895 | if (NAPI_GRO_CB(skb)->count == 1) { |
3896 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3897 | goto out; |
fc59f9a3 | 3898 | } |
d565b0a1 HX |
3899 | |
3900 | rcu_read_lock(); | |
3901 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3902 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
3903 | continue; |
3904 | ||
299603e8 | 3905 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
3906 | break; |
3907 | } | |
3908 | rcu_read_unlock(); | |
3909 | ||
3910 | if (err) { | |
3911 | WARN_ON(&ptype->list == head); | |
3912 | kfree_skb(skb); | |
3913 | return NET_RX_SUCCESS; | |
3914 | } | |
3915 | ||
3916 | out: | |
ae78dbfa | 3917 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
3918 | } |
3919 | ||
2e71a6f8 ED |
3920 | /* napi->gro_list contains packets ordered by age. |
3921 | * youngest packets at the head of it. | |
3922 | * Complete skbs in reverse order to reduce latencies. | |
3923 | */ | |
3924 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 3925 | { |
2e71a6f8 | 3926 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 3927 | |
2e71a6f8 ED |
3928 | /* scan list and build reverse chain */ |
3929 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
3930 | skb->prev = prev; | |
3931 | prev = skb; | |
3932 | } | |
3933 | ||
3934 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 3935 | skb->next = NULL; |
2e71a6f8 ED |
3936 | |
3937 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
3938 | return; | |
3939 | ||
3940 | prev = skb->prev; | |
d565b0a1 | 3941 | napi_gro_complete(skb); |
2e71a6f8 | 3942 | napi->gro_count--; |
d565b0a1 HX |
3943 | } |
3944 | ||
3945 | napi->gro_list = NULL; | |
3946 | } | |
86cac58b | 3947 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3948 | |
89c5fa33 ED |
3949 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
3950 | { | |
3951 | struct sk_buff *p; | |
3952 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 3953 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
3954 | |
3955 | for (p = napi->gro_list; p; p = p->next) { | |
3956 | unsigned long diffs; | |
3957 | ||
0b4cec8c TH |
3958 | NAPI_GRO_CB(p)->flush = 0; |
3959 | ||
3960 | if (hash != skb_get_hash_raw(p)) { | |
3961 | NAPI_GRO_CB(p)->same_flow = 0; | |
3962 | continue; | |
3963 | } | |
3964 | ||
89c5fa33 ED |
3965 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
3966 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
3967 | if (maclen == ETH_HLEN) | |
3968 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 3969 | skb_mac_header(skb)); |
89c5fa33 ED |
3970 | else if (!diffs) |
3971 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 3972 | skb_mac_header(skb), |
89c5fa33 ED |
3973 | maclen); |
3974 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
3975 | } |
3976 | } | |
3977 | ||
299603e8 JC |
3978 | static void skb_gro_reset_offset(struct sk_buff *skb) |
3979 | { | |
3980 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
3981 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
3982 | ||
3983 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3984 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
3985 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
3986 | ||
3987 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
3988 | pinfo->nr_frags && | |
3989 | !PageHighMem(skb_frag_page(frag0))) { | |
3990 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
3991 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
89c5fa33 ED |
3992 | } |
3993 | } | |
3994 | ||
a50e233c ED |
3995 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
3996 | { | |
3997 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
3998 | ||
3999 | BUG_ON(skb->end - skb->tail < grow); | |
4000 | ||
4001 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
4002 | ||
4003 | skb->data_len -= grow; | |
4004 | skb->tail += grow; | |
4005 | ||
4006 | pinfo->frags[0].page_offset += grow; | |
4007 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
4008 | ||
4009 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
4010 | skb_frag_unref(skb, 0); | |
4011 | memmove(pinfo->frags, pinfo->frags + 1, | |
4012 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
4013 | } | |
4014 | } | |
4015 | ||
bb728820 | 4016 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
4017 | { |
4018 | struct sk_buff **pp = NULL; | |
22061d80 | 4019 | struct packet_offload *ptype; |
d565b0a1 | 4020 | __be16 type = skb->protocol; |
22061d80 | 4021 | struct list_head *head = &offload_base; |
0da2afd5 | 4022 | int same_flow; |
5b252f0c | 4023 | enum gro_result ret; |
a50e233c | 4024 | int grow; |
d565b0a1 | 4025 | |
9c62a68d | 4026 | if (!(skb->dev->features & NETIF_F_GRO)) |
d565b0a1 HX |
4027 | goto normal; |
4028 | ||
5a212329 | 4029 | if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad) |
f17f5c91 HX |
4030 | goto normal; |
4031 | ||
89c5fa33 ED |
4032 | gro_list_prepare(napi, skb); |
4033 | ||
d565b0a1 HX |
4034 | rcu_read_lock(); |
4035 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4036 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4037 | continue; |
4038 | ||
86911732 | 4039 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4040 | skb_reset_mac_len(skb); |
d565b0a1 HX |
4041 | NAPI_GRO_CB(skb)->same_flow = 0; |
4042 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 4043 | NAPI_GRO_CB(skb)->free = 0; |
b582ef09 | 4044 | NAPI_GRO_CB(skb)->udp_mark = 0; |
15e2396d | 4045 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 4046 | |
662880f4 TH |
4047 | /* Setup for GRO checksum validation */ |
4048 | switch (skb->ip_summed) { | |
4049 | case CHECKSUM_COMPLETE: | |
4050 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4051 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4052 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4053 | break; | |
4054 | case CHECKSUM_UNNECESSARY: | |
4055 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4056 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4057 | break; | |
4058 | default: | |
4059 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4060 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4061 | } | |
d565b0a1 | 4062 | |
f191a1d1 | 4063 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4064 | break; |
4065 | } | |
4066 | rcu_read_unlock(); | |
4067 | ||
4068 | if (&ptype->list == head) | |
4069 | goto normal; | |
4070 | ||
0da2afd5 | 4071 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4072 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4073 | |
d565b0a1 HX |
4074 | if (pp) { |
4075 | struct sk_buff *nskb = *pp; | |
4076 | ||
4077 | *pp = nskb->next; | |
4078 | nskb->next = NULL; | |
4079 | napi_gro_complete(nskb); | |
4ae5544f | 4080 | napi->gro_count--; |
d565b0a1 HX |
4081 | } |
4082 | ||
0da2afd5 | 4083 | if (same_flow) |
d565b0a1 HX |
4084 | goto ok; |
4085 | ||
600adc18 | 4086 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4087 | goto normal; |
d565b0a1 | 4088 | |
600adc18 ED |
4089 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4090 | struct sk_buff *nskb = napi->gro_list; | |
4091 | ||
4092 | /* locate the end of the list to select the 'oldest' flow */ | |
4093 | while (nskb->next) { | |
4094 | pp = &nskb->next; | |
4095 | nskb = *pp; | |
4096 | } | |
4097 | *pp = NULL; | |
4098 | nskb->next = NULL; | |
4099 | napi_gro_complete(nskb); | |
4100 | } else { | |
4101 | napi->gro_count++; | |
4102 | } | |
d565b0a1 | 4103 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4104 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4105 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4106 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4107 | skb->next = napi->gro_list; |
4108 | napi->gro_list = skb; | |
5d0d9be8 | 4109 | ret = GRO_HELD; |
d565b0a1 | 4110 | |
ad0f9904 | 4111 | pull: |
a50e233c ED |
4112 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4113 | if (grow > 0) | |
4114 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4115 | ok: |
5d0d9be8 | 4116 | return ret; |
d565b0a1 HX |
4117 | |
4118 | normal: | |
ad0f9904 HX |
4119 | ret = GRO_NORMAL; |
4120 | goto pull; | |
5d38a079 | 4121 | } |
96e93eab | 4122 | |
bf5a755f JC |
4123 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4124 | { | |
4125 | struct list_head *offload_head = &offload_base; | |
4126 | struct packet_offload *ptype; | |
4127 | ||
4128 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4129 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4130 | continue; | |
4131 | return ptype; | |
4132 | } | |
4133 | return NULL; | |
4134 | } | |
e27a2f83 | 4135 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4136 | |
4137 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4138 | { | |
4139 | struct list_head *offload_head = &offload_base; | |
4140 | struct packet_offload *ptype; | |
4141 | ||
4142 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4143 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4144 | continue; | |
4145 | return ptype; | |
4146 | } | |
4147 | return NULL; | |
4148 | } | |
e27a2f83 | 4149 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4150 | |
bb728820 | 4151 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4152 | { |
5d0d9be8 HX |
4153 | switch (ret) { |
4154 | case GRO_NORMAL: | |
ae78dbfa | 4155 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4156 | ret = GRO_DROP; |
4157 | break; | |
5d38a079 | 4158 | |
5d0d9be8 | 4159 | case GRO_DROP: |
5d38a079 HX |
4160 | kfree_skb(skb); |
4161 | break; | |
5b252f0c | 4162 | |
daa86548 | 4163 | case GRO_MERGED_FREE: |
d7e8883c ED |
4164 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
4165 | kmem_cache_free(skbuff_head_cache, skb); | |
4166 | else | |
4167 | __kfree_skb(skb); | |
daa86548 ED |
4168 | break; |
4169 | ||
5b252f0c BH |
4170 | case GRO_HELD: |
4171 | case GRO_MERGED: | |
4172 | break; | |
5d38a079 HX |
4173 | } |
4174 | ||
c7c4b3b6 | 4175 | return ret; |
5d0d9be8 | 4176 | } |
5d0d9be8 | 4177 | |
c7c4b3b6 | 4178 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4179 | { |
ae78dbfa | 4180 | trace_napi_gro_receive_entry(skb); |
86911732 | 4181 | |
a50e233c ED |
4182 | skb_gro_reset_offset(skb); |
4183 | ||
89c5fa33 | 4184 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4185 | } |
4186 | EXPORT_SYMBOL(napi_gro_receive); | |
4187 | ||
d0c2b0d2 | 4188 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4189 | { |
93a35f59 ED |
4190 | if (unlikely(skb->pfmemalloc)) { |
4191 | consume_skb(skb); | |
4192 | return; | |
4193 | } | |
96e93eab | 4194 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4195 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4196 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4197 | skb->vlan_tci = 0; |
66c46d74 | 4198 | skb->dev = napi->dev; |
6d152e23 | 4199 | skb->skb_iif = 0; |
c3caf119 JC |
4200 | skb->encapsulation = 0; |
4201 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4202 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
96e93eab HX |
4203 | |
4204 | napi->skb = skb; | |
4205 | } | |
96e93eab | 4206 | |
76620aaf | 4207 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4208 | { |
5d38a079 | 4209 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4210 | |
4211 | if (!skb) { | |
fd11a83d | 4212 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
84b9cd63 | 4213 | napi->skb = skb; |
80595d59 | 4214 | } |
96e93eab HX |
4215 | return skb; |
4216 | } | |
76620aaf | 4217 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4218 | |
a50e233c ED |
4219 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4220 | struct sk_buff *skb, | |
4221 | gro_result_t ret) | |
96e93eab | 4222 | { |
5d0d9be8 HX |
4223 | switch (ret) { |
4224 | case GRO_NORMAL: | |
a50e233c ED |
4225 | case GRO_HELD: |
4226 | __skb_push(skb, ETH_HLEN); | |
4227 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4228 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4229 | ret = GRO_DROP; |
86911732 | 4230 | break; |
5d38a079 | 4231 | |
5d0d9be8 | 4232 | case GRO_DROP: |
5d0d9be8 HX |
4233 | case GRO_MERGED_FREE: |
4234 | napi_reuse_skb(napi, skb); | |
4235 | break; | |
5b252f0c BH |
4236 | |
4237 | case GRO_MERGED: | |
4238 | break; | |
5d0d9be8 | 4239 | } |
5d38a079 | 4240 | |
c7c4b3b6 | 4241 | return ret; |
5d38a079 | 4242 | } |
5d0d9be8 | 4243 | |
a50e233c ED |
4244 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
4245 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
4246 | * We copy ethernet header into skb->data to have a common layout. | |
4247 | */ | |
4adb9c4a | 4248 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4249 | { |
4250 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
4251 | const struct ethhdr *eth; |
4252 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
4253 | |
4254 | napi->skb = NULL; | |
4255 | ||
a50e233c ED |
4256 | skb_reset_mac_header(skb); |
4257 | skb_gro_reset_offset(skb); | |
4258 | ||
4259 | eth = skb_gro_header_fast(skb, 0); | |
4260 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
4261 | eth = skb_gro_header_slow(skb, hlen, 0); | |
4262 | if (unlikely(!eth)) { | |
4263 | napi_reuse_skb(napi, skb); | |
4264 | return NULL; | |
4265 | } | |
4266 | } else { | |
4267 | gro_pull_from_frag0(skb, hlen); | |
4268 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
4269 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 4270 | } |
a50e233c ED |
4271 | __skb_pull(skb, hlen); |
4272 | ||
4273 | /* | |
4274 | * This works because the only protocols we care about don't require | |
4275 | * special handling. | |
4276 | * We'll fix it up properly in napi_frags_finish() | |
4277 | */ | |
4278 | skb->protocol = eth->h_proto; | |
76620aaf | 4279 | |
76620aaf HX |
4280 | return skb; |
4281 | } | |
76620aaf | 4282 | |
c7c4b3b6 | 4283 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4284 | { |
76620aaf | 4285 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4286 | |
4287 | if (!skb) | |
c7c4b3b6 | 4288 | return GRO_DROP; |
5d0d9be8 | 4289 | |
ae78dbfa BH |
4290 | trace_napi_gro_frags_entry(skb); |
4291 | ||
89c5fa33 | 4292 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4293 | } |
5d38a079 HX |
4294 | EXPORT_SYMBOL(napi_gro_frags); |
4295 | ||
573e8fca TH |
4296 | /* Compute the checksum from gro_offset and return the folded value |
4297 | * after adding in any pseudo checksum. | |
4298 | */ | |
4299 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
4300 | { | |
4301 | __wsum wsum; | |
4302 | __sum16 sum; | |
4303 | ||
4304 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
4305 | ||
4306 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
4307 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
4308 | if (likely(!sum)) { | |
4309 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
4310 | !skb->csum_complete_sw) | |
4311 | netdev_rx_csum_fault(skb->dev); | |
4312 | } | |
4313 | ||
4314 | NAPI_GRO_CB(skb)->csum = wsum; | |
4315 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4316 | ||
4317 | return sum; | |
4318 | } | |
4319 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
4320 | ||
e326bed2 | 4321 | /* |
855abcf0 | 4322 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
4323 | * Note: called with local irq disabled, but exits with local irq enabled. |
4324 | */ | |
4325 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4326 | { | |
4327 | #ifdef CONFIG_RPS | |
4328 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4329 | ||
4330 | if (remsd) { | |
4331 | sd->rps_ipi_list = NULL; | |
4332 | ||
4333 | local_irq_enable(); | |
4334 | ||
4335 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
4336 | while (remsd) { | |
4337 | struct softnet_data *next = remsd->rps_ipi_next; | |
4338 | ||
4339 | if (cpu_online(remsd->cpu)) | |
c46fff2a | 4340 | smp_call_function_single_async(remsd->cpu, |
fce8ad15 | 4341 | &remsd->csd); |
e326bed2 ED |
4342 | remsd = next; |
4343 | } | |
4344 | } else | |
4345 | #endif | |
4346 | local_irq_enable(); | |
4347 | } | |
4348 | ||
d75b1ade ED |
4349 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
4350 | { | |
4351 | #ifdef CONFIG_RPS | |
4352 | return sd->rps_ipi_list != NULL; | |
4353 | #else | |
4354 | return false; | |
4355 | #endif | |
4356 | } | |
4357 | ||
bea3348e | 4358 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
4359 | { |
4360 | int work = 0; | |
eecfd7c4 | 4361 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 4362 | |
e326bed2 ED |
4363 | /* Check if we have pending ipi, its better to send them now, |
4364 | * not waiting net_rx_action() end. | |
4365 | */ | |
d75b1ade | 4366 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
4367 | local_irq_disable(); |
4368 | net_rps_action_and_irq_enable(sd); | |
4369 | } | |
d75b1ade | 4370 | |
bea3348e | 4371 | napi->weight = weight_p; |
6e7676c1 | 4372 | local_irq_disable(); |
11ef7a89 | 4373 | while (1) { |
1da177e4 | 4374 | struct sk_buff *skb; |
6e7676c1 CG |
4375 | |
4376 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
4377 | local_irq_enable(); | |
4378 | __netif_receive_skb(skb); | |
6e7676c1 | 4379 | local_irq_disable(); |
76cc8b13 TH |
4380 | input_queue_head_incr(sd); |
4381 | if (++work >= quota) { | |
4382 | local_irq_enable(); | |
4383 | return work; | |
4384 | } | |
6e7676c1 | 4385 | } |
1da177e4 | 4386 | |
e36fa2f7 | 4387 | rps_lock(sd); |
11ef7a89 | 4388 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
4389 | /* |
4390 | * Inline a custom version of __napi_complete(). | |
4391 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
4392 | * and NAPI_STATE_SCHED is the only possible flag set |
4393 | * on backlog. | |
4394 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
4395 | * and we dont need an smp_mb() memory barrier. |
4396 | */ | |
eecfd7c4 | 4397 | napi->state = 0; |
11ef7a89 | 4398 | rps_unlock(sd); |
eecfd7c4 | 4399 | |
11ef7a89 | 4400 | break; |
bea3348e | 4401 | } |
11ef7a89 TH |
4402 | |
4403 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
4404 | &sd->process_queue); | |
e36fa2f7 | 4405 | rps_unlock(sd); |
6e7676c1 CG |
4406 | } |
4407 | local_irq_enable(); | |
1da177e4 | 4408 | |
bea3348e SH |
4409 | return work; |
4410 | } | |
1da177e4 | 4411 | |
bea3348e SH |
4412 | /** |
4413 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4414 | * @n: entry to schedule |
bea3348e | 4415 | * |
bc9ad166 ED |
4416 | * The entry's receive function will be scheduled to run. |
4417 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 4418 | */ |
b5606c2d | 4419 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4420 | { |
4421 | unsigned long flags; | |
1da177e4 | 4422 | |
bea3348e | 4423 | local_irq_save(flags); |
903ceff7 | 4424 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 4425 | local_irq_restore(flags); |
1da177e4 | 4426 | } |
bea3348e SH |
4427 | EXPORT_SYMBOL(__napi_schedule); |
4428 | ||
bc9ad166 ED |
4429 | /** |
4430 | * __napi_schedule_irqoff - schedule for receive | |
4431 | * @n: entry to schedule | |
4432 | * | |
4433 | * Variant of __napi_schedule() assuming hard irqs are masked | |
4434 | */ | |
4435 | void __napi_schedule_irqoff(struct napi_struct *n) | |
4436 | { | |
4437 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
4438 | } | |
4439 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
4440 | ||
d565b0a1 HX |
4441 | void __napi_complete(struct napi_struct *n) |
4442 | { | |
4443 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
d565b0a1 | 4444 | |
d75b1ade | 4445 | list_del_init(&n->poll_list); |
4e857c58 | 4446 | smp_mb__before_atomic(); |
d565b0a1 HX |
4447 | clear_bit(NAPI_STATE_SCHED, &n->state); |
4448 | } | |
4449 | EXPORT_SYMBOL(__napi_complete); | |
4450 | ||
3b47d303 | 4451 | void napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 HX |
4452 | { |
4453 | unsigned long flags; | |
4454 | ||
4455 | /* | |
4456 | * don't let napi dequeue from the cpu poll list | |
4457 | * just in case its running on a different cpu | |
4458 | */ | |
4459 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4460 | return; | |
4461 | ||
3b47d303 ED |
4462 | if (n->gro_list) { |
4463 | unsigned long timeout = 0; | |
d75b1ade | 4464 | |
3b47d303 ED |
4465 | if (work_done) |
4466 | timeout = n->dev->gro_flush_timeout; | |
4467 | ||
4468 | if (timeout) | |
4469 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
4470 | HRTIMER_MODE_REL_PINNED); | |
4471 | else | |
4472 | napi_gro_flush(n, false); | |
4473 | } | |
d75b1ade ED |
4474 | if (likely(list_empty(&n->poll_list))) { |
4475 | WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state)); | |
4476 | } else { | |
4477 | /* If n->poll_list is not empty, we need to mask irqs */ | |
4478 | local_irq_save(flags); | |
4479 | __napi_complete(n); | |
4480 | local_irq_restore(flags); | |
4481 | } | |
d565b0a1 | 4482 | } |
3b47d303 | 4483 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 4484 | |
af12fa6e ET |
4485 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
4486 | struct napi_struct *napi_by_id(unsigned int napi_id) | |
4487 | { | |
4488 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
4489 | struct napi_struct *napi; | |
4490 | ||
4491 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
4492 | if (napi->napi_id == napi_id) | |
4493 | return napi; | |
4494 | ||
4495 | return NULL; | |
4496 | } | |
4497 | EXPORT_SYMBOL_GPL(napi_by_id); | |
4498 | ||
4499 | void napi_hash_add(struct napi_struct *napi) | |
4500 | { | |
4501 | if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) { | |
4502 | ||
4503 | spin_lock(&napi_hash_lock); | |
4504 | ||
4505 | /* 0 is not a valid id, we also skip an id that is taken | |
4506 | * we expect both events to be extremely rare | |
4507 | */ | |
4508 | napi->napi_id = 0; | |
4509 | while (!napi->napi_id) { | |
4510 | napi->napi_id = ++napi_gen_id; | |
4511 | if (napi_by_id(napi->napi_id)) | |
4512 | napi->napi_id = 0; | |
4513 | } | |
4514 | ||
4515 | hlist_add_head_rcu(&napi->napi_hash_node, | |
4516 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
4517 | ||
4518 | spin_unlock(&napi_hash_lock); | |
4519 | } | |
4520 | } | |
4521 | EXPORT_SYMBOL_GPL(napi_hash_add); | |
4522 | ||
4523 | /* Warning : caller is responsible to make sure rcu grace period | |
4524 | * is respected before freeing memory containing @napi | |
4525 | */ | |
4526 | void napi_hash_del(struct napi_struct *napi) | |
4527 | { | |
4528 | spin_lock(&napi_hash_lock); | |
4529 | ||
4530 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) | |
4531 | hlist_del_rcu(&napi->napi_hash_node); | |
4532 | ||
4533 | spin_unlock(&napi_hash_lock); | |
4534 | } | |
4535 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
4536 | ||
3b47d303 ED |
4537 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
4538 | { | |
4539 | struct napi_struct *napi; | |
4540 | ||
4541 | napi = container_of(timer, struct napi_struct, timer); | |
4542 | if (napi->gro_list) | |
4543 | napi_schedule(napi); | |
4544 | ||
4545 | return HRTIMER_NORESTART; | |
4546 | } | |
4547 | ||
d565b0a1 HX |
4548 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
4549 | int (*poll)(struct napi_struct *, int), int weight) | |
4550 | { | |
4551 | INIT_LIST_HEAD(&napi->poll_list); | |
3b47d303 ED |
4552 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
4553 | napi->timer.function = napi_watchdog; | |
4ae5544f | 4554 | napi->gro_count = 0; |
d565b0a1 | 4555 | napi->gro_list = NULL; |
5d38a079 | 4556 | napi->skb = NULL; |
d565b0a1 | 4557 | napi->poll = poll; |
82dc3c63 ED |
4558 | if (weight > NAPI_POLL_WEIGHT) |
4559 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
4560 | weight, dev->name); | |
d565b0a1 HX |
4561 | napi->weight = weight; |
4562 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 4563 | napi->dev = dev; |
5d38a079 | 4564 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
4565 | spin_lock_init(&napi->poll_lock); |
4566 | napi->poll_owner = -1; | |
4567 | #endif | |
4568 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
4569 | } | |
4570 | EXPORT_SYMBOL(netif_napi_add); | |
4571 | ||
3b47d303 ED |
4572 | void napi_disable(struct napi_struct *n) |
4573 | { | |
4574 | might_sleep(); | |
4575 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
4576 | ||
4577 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
4578 | msleep(1); | |
4579 | ||
4580 | hrtimer_cancel(&n->timer); | |
4581 | ||
4582 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
4583 | } | |
4584 | EXPORT_SYMBOL(napi_disable); | |
4585 | ||
d565b0a1 HX |
4586 | void netif_napi_del(struct napi_struct *napi) |
4587 | { | |
d7b06636 | 4588 | list_del_init(&napi->dev_list); |
76620aaf | 4589 | napi_free_frags(napi); |
d565b0a1 | 4590 | |
289dccbe | 4591 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 4592 | napi->gro_list = NULL; |
4ae5544f | 4593 | napi->gro_count = 0; |
d565b0a1 HX |
4594 | } |
4595 | EXPORT_SYMBOL(netif_napi_del); | |
4596 | ||
726ce70e HX |
4597 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
4598 | { | |
4599 | void *have; | |
4600 | int work, weight; | |
4601 | ||
4602 | list_del_init(&n->poll_list); | |
4603 | ||
4604 | have = netpoll_poll_lock(n); | |
4605 | ||
4606 | weight = n->weight; | |
4607 | ||
4608 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
4609 | * with netpoll's poll_napi(). Only the entity which | |
4610 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
4611 | * actually make the ->poll() call. Therefore we avoid | |
4612 | * accidentally calling ->poll() when NAPI is not scheduled. | |
4613 | */ | |
4614 | work = 0; | |
4615 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
4616 | work = n->poll(n, weight); | |
4617 | trace_napi_poll(n); | |
4618 | } | |
4619 | ||
4620 | WARN_ON_ONCE(work > weight); | |
4621 | ||
4622 | if (likely(work < weight)) | |
4623 | goto out_unlock; | |
4624 | ||
4625 | /* Drivers must not modify the NAPI state if they | |
4626 | * consume the entire weight. In such cases this code | |
4627 | * still "owns" the NAPI instance and therefore can | |
4628 | * move the instance around on the list at-will. | |
4629 | */ | |
4630 | if (unlikely(napi_disable_pending(n))) { | |
4631 | napi_complete(n); | |
4632 | goto out_unlock; | |
4633 | } | |
4634 | ||
4635 | if (n->gro_list) { | |
4636 | /* flush too old packets | |
4637 | * If HZ < 1000, flush all packets. | |
4638 | */ | |
4639 | napi_gro_flush(n, HZ >= 1000); | |
4640 | } | |
4641 | ||
001ce546 HX |
4642 | /* Some drivers may have called napi_schedule |
4643 | * prior to exhausting their budget. | |
4644 | */ | |
4645 | if (unlikely(!list_empty(&n->poll_list))) { | |
4646 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
4647 | n->dev ? n->dev->name : "backlog"); | |
4648 | goto out_unlock; | |
4649 | } | |
4650 | ||
726ce70e HX |
4651 | list_add_tail(&n->poll_list, repoll); |
4652 | ||
4653 | out_unlock: | |
4654 | netpoll_poll_unlock(have); | |
4655 | ||
4656 | return work; | |
4657 | } | |
4658 | ||
1da177e4 LT |
4659 | static void net_rx_action(struct softirq_action *h) |
4660 | { | |
903ceff7 | 4661 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
24f8b238 | 4662 | unsigned long time_limit = jiffies + 2; |
51b0bded | 4663 | int budget = netdev_budget; |
d75b1ade ED |
4664 | LIST_HEAD(list); |
4665 | LIST_HEAD(repoll); | |
53fb95d3 | 4666 | |
1da177e4 | 4667 | local_irq_disable(); |
d75b1ade ED |
4668 | list_splice_init(&sd->poll_list, &list); |
4669 | local_irq_enable(); | |
1da177e4 | 4670 | |
ceb8d5bf | 4671 | for (;;) { |
bea3348e | 4672 | struct napi_struct *n; |
1da177e4 | 4673 | |
ceb8d5bf HX |
4674 | if (list_empty(&list)) { |
4675 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
4676 | return; | |
4677 | break; | |
4678 | } | |
4679 | ||
6bd373eb HX |
4680 | n = list_first_entry(&list, struct napi_struct, poll_list); |
4681 | budget -= napi_poll(n, &repoll); | |
4682 | ||
d75b1ade | 4683 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
4684 | * Allow this to run for 2 jiffies since which will allow |
4685 | * an average latency of 1.5/HZ. | |
bea3348e | 4686 | */ |
ceb8d5bf HX |
4687 | if (unlikely(budget <= 0 || |
4688 | time_after_eq(jiffies, time_limit))) { | |
4689 | sd->time_squeeze++; | |
4690 | break; | |
4691 | } | |
1da177e4 | 4692 | } |
d75b1ade | 4693 | |
d75b1ade ED |
4694 | local_irq_disable(); |
4695 | ||
4696 | list_splice_tail_init(&sd->poll_list, &list); | |
4697 | list_splice_tail(&repoll, &list); | |
4698 | list_splice(&list, &sd->poll_list); | |
4699 | if (!list_empty(&sd->poll_list)) | |
4700 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4701 | ||
e326bed2 | 4702 | net_rps_action_and_irq_enable(sd); |
1da177e4 LT |
4703 | } |
4704 | ||
aa9d8560 | 4705 | struct netdev_adjacent { |
9ff162a8 | 4706 | struct net_device *dev; |
5d261913 VF |
4707 | |
4708 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 4709 | bool master; |
5d261913 | 4710 | |
5d261913 VF |
4711 | /* counter for the number of times this device was added to us */ |
4712 | u16 ref_nr; | |
4713 | ||
402dae96 VF |
4714 | /* private field for the users */ |
4715 | void *private; | |
4716 | ||
9ff162a8 JP |
4717 | struct list_head list; |
4718 | struct rcu_head rcu; | |
9ff162a8 JP |
4719 | }; |
4720 | ||
5d261913 VF |
4721 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev, |
4722 | struct net_device *adj_dev, | |
2f268f12 | 4723 | struct list_head *adj_list) |
9ff162a8 | 4724 | { |
5d261913 | 4725 | struct netdev_adjacent *adj; |
5d261913 | 4726 | |
2f268f12 | 4727 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
4728 | if (adj->dev == adj_dev) |
4729 | return adj; | |
9ff162a8 JP |
4730 | } |
4731 | return NULL; | |
4732 | } | |
4733 | ||
4734 | /** | |
4735 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
4736 | * @dev: device | |
4737 | * @upper_dev: upper device to check | |
4738 | * | |
4739 | * Find out if a device is linked to specified upper device and return true | |
4740 | * in case it is. Note that this checks only immediate upper device, | |
4741 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
4742 | */ | |
4743 | bool netdev_has_upper_dev(struct net_device *dev, | |
4744 | struct net_device *upper_dev) | |
4745 | { | |
4746 | ASSERT_RTNL(); | |
4747 | ||
2f268f12 | 4748 | return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper); |
9ff162a8 JP |
4749 | } |
4750 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
4751 | ||
4752 | /** | |
4753 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
4754 | * @dev: device | |
4755 | * | |
4756 | * Find out if a device is linked to an upper device and return true in case | |
4757 | * it is. The caller must hold the RTNL lock. | |
4758 | */ | |
1d143d9f | 4759 | static bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
4760 | { |
4761 | ASSERT_RTNL(); | |
4762 | ||
2f268f12 | 4763 | return !list_empty(&dev->all_adj_list.upper); |
9ff162a8 | 4764 | } |
9ff162a8 JP |
4765 | |
4766 | /** | |
4767 | * netdev_master_upper_dev_get - Get master upper device | |
4768 | * @dev: device | |
4769 | * | |
4770 | * Find a master upper device and return pointer to it or NULL in case | |
4771 | * it's not there. The caller must hold the RTNL lock. | |
4772 | */ | |
4773 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
4774 | { | |
aa9d8560 | 4775 | struct netdev_adjacent *upper; |
9ff162a8 JP |
4776 | |
4777 | ASSERT_RTNL(); | |
4778 | ||
2f268f12 | 4779 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
4780 | return NULL; |
4781 | ||
2f268f12 | 4782 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 4783 | struct netdev_adjacent, list); |
9ff162a8 JP |
4784 | if (likely(upper->master)) |
4785 | return upper->dev; | |
4786 | return NULL; | |
4787 | } | |
4788 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
4789 | ||
b6ccba4c VF |
4790 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
4791 | { | |
4792 | struct netdev_adjacent *adj; | |
4793 | ||
4794 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
4795 | ||
4796 | return adj->private; | |
4797 | } | |
4798 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
4799 | ||
44a40855 VY |
4800 | /** |
4801 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
4802 | * @dev: device | |
4803 | * @iter: list_head ** of the current position | |
4804 | * | |
4805 | * Gets the next device from the dev's upper list, starting from iter | |
4806 | * position. The caller must hold RCU read lock. | |
4807 | */ | |
4808 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
4809 | struct list_head **iter) | |
4810 | { | |
4811 | struct netdev_adjacent *upper; | |
4812 | ||
4813 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
4814 | ||
4815 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4816 | ||
4817 | if (&upper->list == &dev->adj_list.upper) | |
4818 | return NULL; | |
4819 | ||
4820 | *iter = &upper->list; | |
4821 | ||
4822 | return upper->dev; | |
4823 | } | |
4824 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
4825 | ||
31088a11 VF |
4826 | /** |
4827 | * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list | |
48311f46 VF |
4828 | * @dev: device |
4829 | * @iter: list_head ** of the current position | |
4830 | * | |
4831 | * Gets the next device from the dev's upper list, starting from iter | |
4832 | * position. The caller must hold RCU read lock. | |
4833 | */ | |
2f268f12 VF |
4834 | struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, |
4835 | struct list_head **iter) | |
48311f46 VF |
4836 | { |
4837 | struct netdev_adjacent *upper; | |
4838 | ||
85328240 | 4839 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); |
48311f46 VF |
4840 | |
4841 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4842 | ||
2f268f12 | 4843 | if (&upper->list == &dev->all_adj_list.upper) |
48311f46 VF |
4844 | return NULL; |
4845 | ||
4846 | *iter = &upper->list; | |
4847 | ||
4848 | return upper->dev; | |
4849 | } | |
2f268f12 | 4850 | EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); |
48311f46 | 4851 | |
31088a11 VF |
4852 | /** |
4853 | * netdev_lower_get_next_private - Get the next ->private from the | |
4854 | * lower neighbour list | |
4855 | * @dev: device | |
4856 | * @iter: list_head ** of the current position | |
4857 | * | |
4858 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
4859 | * list, starting from iter position. The caller must hold either hold the | |
4860 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
4861 | * list will remain unchainged. | |
4862 | */ | |
4863 | void *netdev_lower_get_next_private(struct net_device *dev, | |
4864 | struct list_head **iter) | |
4865 | { | |
4866 | struct netdev_adjacent *lower; | |
4867 | ||
4868 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
4869 | ||
4870 | if (&lower->list == &dev->adj_list.lower) | |
4871 | return NULL; | |
4872 | ||
6859e7df | 4873 | *iter = lower->list.next; |
31088a11 VF |
4874 | |
4875 | return lower->private; | |
4876 | } | |
4877 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
4878 | ||
4879 | /** | |
4880 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
4881 | * lower neighbour list, RCU | |
4882 | * variant | |
4883 | * @dev: device | |
4884 | * @iter: list_head ** of the current position | |
4885 | * | |
4886 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
4887 | * list, starting from iter position. The caller must hold RCU read lock. | |
4888 | */ | |
4889 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
4890 | struct list_head **iter) | |
4891 | { | |
4892 | struct netdev_adjacent *lower; | |
4893 | ||
4894 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4895 | ||
4896 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
4897 | ||
4898 | if (&lower->list == &dev->adj_list.lower) | |
4899 | return NULL; | |
4900 | ||
6859e7df | 4901 | *iter = &lower->list; |
31088a11 VF |
4902 | |
4903 | return lower->private; | |
4904 | } | |
4905 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
4906 | ||
4085ebe8 VY |
4907 | /** |
4908 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
4909 | * list | |
4910 | * @dev: device | |
4911 | * @iter: list_head ** of the current position | |
4912 | * | |
4913 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
4914 | * list, starting from iter position. The caller must hold RTNL lock or | |
4915 | * its own locking that guarantees that the neighbour lower | |
4916 | * list will remain unchainged. | |
4917 | */ | |
4918 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
4919 | { | |
4920 | struct netdev_adjacent *lower; | |
4921 | ||
4922 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
4923 | ||
4924 | if (&lower->list == &dev->adj_list.lower) | |
4925 | return NULL; | |
4926 | ||
4927 | *iter = &lower->list; | |
4928 | ||
4929 | return lower->dev; | |
4930 | } | |
4931 | EXPORT_SYMBOL(netdev_lower_get_next); | |
4932 | ||
e001bfad | 4933 | /** |
4934 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
4935 | * lower neighbour list, RCU | |
4936 | * variant | |
4937 | * @dev: device | |
4938 | * | |
4939 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
4940 | * list. The caller must hold RCU read lock. | |
4941 | */ | |
4942 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
4943 | { | |
4944 | struct netdev_adjacent *lower; | |
4945 | ||
4946 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
4947 | struct netdev_adjacent, list); | |
4948 | if (lower) | |
4949 | return lower->private; | |
4950 | return NULL; | |
4951 | } | |
4952 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
4953 | ||
9ff162a8 JP |
4954 | /** |
4955 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
4956 | * @dev: device | |
4957 | * | |
4958 | * Find a master upper device and return pointer to it or NULL in case | |
4959 | * it's not there. The caller must hold the RCU read lock. | |
4960 | */ | |
4961 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
4962 | { | |
aa9d8560 | 4963 | struct netdev_adjacent *upper; |
9ff162a8 | 4964 | |
2f268f12 | 4965 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 4966 | struct netdev_adjacent, list); |
9ff162a8 JP |
4967 | if (upper && likely(upper->master)) |
4968 | return upper->dev; | |
4969 | return NULL; | |
4970 | } | |
4971 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
4972 | ||
0a59f3a9 | 4973 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
4974 | struct net_device *adj_dev, |
4975 | struct list_head *dev_list) | |
4976 | { | |
4977 | char linkname[IFNAMSIZ+7]; | |
4978 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
4979 | "upper_%s" : "lower_%s", adj_dev->name); | |
4980 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
4981 | linkname); | |
4982 | } | |
0a59f3a9 | 4983 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
4984 | char *name, |
4985 | struct list_head *dev_list) | |
4986 | { | |
4987 | char linkname[IFNAMSIZ+7]; | |
4988 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
4989 | "upper_%s" : "lower_%s", name); | |
4990 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
4991 | } | |
4992 | ||
7ce64c79 AF |
4993 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
4994 | struct net_device *adj_dev, | |
4995 | struct list_head *dev_list) | |
4996 | { | |
4997 | return (dev_list == &dev->adj_list.upper || | |
4998 | dev_list == &dev->adj_list.lower) && | |
4999 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
5000 | } | |
3ee32707 | 5001 | |
5d261913 VF |
5002 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
5003 | struct net_device *adj_dev, | |
7863c054 | 5004 | struct list_head *dev_list, |
402dae96 | 5005 | void *private, bool master) |
5d261913 VF |
5006 | { |
5007 | struct netdev_adjacent *adj; | |
842d67a7 | 5008 | int ret; |
5d261913 | 5009 | |
7863c054 | 5010 | adj = __netdev_find_adj(dev, adj_dev, dev_list); |
5d261913 VF |
5011 | |
5012 | if (adj) { | |
5d261913 VF |
5013 | adj->ref_nr++; |
5014 | return 0; | |
5015 | } | |
5016 | ||
5017 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
5018 | if (!adj) | |
5019 | return -ENOMEM; | |
5020 | ||
5021 | adj->dev = adj_dev; | |
5022 | adj->master = master; | |
5d261913 | 5023 | adj->ref_nr = 1; |
402dae96 | 5024 | adj->private = private; |
5d261913 | 5025 | dev_hold(adj_dev); |
2f268f12 VF |
5026 | |
5027 | pr_debug("dev_hold for %s, because of link added from %s to %s\n", | |
5028 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 | 5029 | |
7ce64c79 | 5030 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 5031 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
5032 | if (ret) |
5033 | goto free_adj; | |
5034 | } | |
5035 | ||
7863c054 | 5036 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
5037 | if (master) { |
5038 | ret = sysfs_create_link(&(dev->dev.kobj), | |
5039 | &(adj_dev->dev.kobj), "master"); | |
5040 | if (ret) | |
5831d66e | 5041 | goto remove_symlinks; |
842d67a7 | 5042 | |
7863c054 | 5043 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 5044 | } else { |
7863c054 | 5045 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 5046 | } |
5d261913 VF |
5047 | |
5048 | return 0; | |
842d67a7 | 5049 | |
5831d66e | 5050 | remove_symlinks: |
7ce64c79 | 5051 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5052 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
5053 | free_adj: |
5054 | kfree(adj); | |
974daef7 | 5055 | dev_put(adj_dev); |
842d67a7 VF |
5056 | |
5057 | return ret; | |
5d261913 VF |
5058 | } |
5059 | ||
1d143d9f | 5060 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
5061 | struct net_device *adj_dev, | |
5062 | struct list_head *dev_list) | |
5d261913 VF |
5063 | { |
5064 | struct netdev_adjacent *adj; | |
5065 | ||
7863c054 | 5066 | adj = __netdev_find_adj(dev, adj_dev, dev_list); |
5d261913 | 5067 | |
2f268f12 VF |
5068 | if (!adj) { |
5069 | pr_err("tried to remove device %s from %s\n", | |
5070 | dev->name, adj_dev->name); | |
5d261913 | 5071 | BUG(); |
2f268f12 | 5072 | } |
5d261913 VF |
5073 | |
5074 | if (adj->ref_nr > 1) { | |
2f268f12 VF |
5075 | pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name, |
5076 | adj->ref_nr-1); | |
5d261913 VF |
5077 | adj->ref_nr--; |
5078 | return; | |
5079 | } | |
5080 | ||
842d67a7 VF |
5081 | if (adj->master) |
5082 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
5083 | ||
7ce64c79 | 5084 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5085 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 5086 | |
5d261913 | 5087 | list_del_rcu(&adj->list); |
2f268f12 VF |
5088 | pr_debug("dev_put for %s, because link removed from %s to %s\n", |
5089 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 VF |
5090 | dev_put(adj_dev); |
5091 | kfree_rcu(adj, rcu); | |
5092 | } | |
5093 | ||
1d143d9f | 5094 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
5095 | struct net_device *upper_dev, | |
5096 | struct list_head *up_list, | |
5097 | struct list_head *down_list, | |
5098 | void *private, bool master) | |
5d261913 VF |
5099 | { |
5100 | int ret; | |
5101 | ||
402dae96 VF |
5102 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private, |
5103 | master); | |
5d261913 VF |
5104 | if (ret) |
5105 | return ret; | |
5106 | ||
402dae96 VF |
5107 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private, |
5108 | false); | |
5d261913 | 5109 | if (ret) { |
2f268f12 | 5110 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5d261913 VF |
5111 | return ret; |
5112 | } | |
5113 | ||
5114 | return 0; | |
5115 | } | |
5116 | ||
1d143d9f | 5117 | static int __netdev_adjacent_dev_link(struct net_device *dev, |
5118 | struct net_device *upper_dev) | |
5d261913 | 5119 | { |
2f268f12 VF |
5120 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
5121 | &dev->all_adj_list.upper, | |
5122 | &upper_dev->all_adj_list.lower, | |
402dae96 | 5123 | NULL, false); |
5d261913 VF |
5124 | } |
5125 | ||
1d143d9f | 5126 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
5127 | struct net_device *upper_dev, | |
5128 | struct list_head *up_list, | |
5129 | struct list_head *down_list) | |
5d261913 | 5130 | { |
2f268f12 VF |
5131 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5132 | __netdev_adjacent_dev_remove(upper_dev, dev, down_list); | |
5d261913 VF |
5133 | } |
5134 | ||
1d143d9f | 5135 | static void __netdev_adjacent_dev_unlink(struct net_device *dev, |
5136 | struct net_device *upper_dev) | |
5d261913 | 5137 | { |
2f268f12 VF |
5138 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, |
5139 | &dev->all_adj_list.upper, | |
5140 | &upper_dev->all_adj_list.lower); | |
5141 | } | |
5142 | ||
1d143d9f | 5143 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
5144 | struct net_device *upper_dev, | |
5145 | void *private, bool master) | |
2f268f12 VF |
5146 | { |
5147 | int ret = __netdev_adjacent_dev_link(dev, upper_dev); | |
5148 | ||
5149 | if (ret) | |
5150 | return ret; | |
5151 | ||
5152 | ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, | |
5153 | &dev->adj_list.upper, | |
5154 | &upper_dev->adj_list.lower, | |
402dae96 | 5155 | private, master); |
2f268f12 VF |
5156 | if (ret) { |
5157 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5158 | return ret; | |
5159 | } | |
5160 | ||
5161 | return 0; | |
5d261913 VF |
5162 | } |
5163 | ||
1d143d9f | 5164 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
5165 | struct net_device *upper_dev) | |
2f268f12 VF |
5166 | { |
5167 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5168 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, | |
5169 | &dev->adj_list.upper, | |
5170 | &upper_dev->adj_list.lower); | |
5171 | } | |
5d261913 | 5172 | |
9ff162a8 | 5173 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 VF |
5174 | struct net_device *upper_dev, bool master, |
5175 | void *private) | |
9ff162a8 | 5176 | { |
5d261913 VF |
5177 | struct netdev_adjacent *i, *j, *to_i, *to_j; |
5178 | int ret = 0; | |
9ff162a8 JP |
5179 | |
5180 | ASSERT_RTNL(); | |
5181 | ||
5182 | if (dev == upper_dev) | |
5183 | return -EBUSY; | |
5184 | ||
5185 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
2f268f12 | 5186 | if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper)) |
9ff162a8 JP |
5187 | return -EBUSY; |
5188 | ||
2f268f12 | 5189 | if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper)) |
9ff162a8 JP |
5190 | return -EEXIST; |
5191 | ||
5192 | if (master && netdev_master_upper_dev_get(dev)) | |
5193 | return -EBUSY; | |
5194 | ||
402dae96 VF |
5195 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private, |
5196 | master); | |
5d261913 VF |
5197 | if (ret) |
5198 | return ret; | |
9ff162a8 | 5199 | |
5d261913 | 5200 | /* Now that we linked these devs, make all the upper_dev's |
2f268f12 | 5201 | * all_adj_list.upper visible to every dev's all_adj_list.lower an |
5d261913 VF |
5202 | * versa, and don't forget the devices itself. All of these |
5203 | * links are non-neighbours. | |
5204 | */ | |
2f268f12 VF |
5205 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5206 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5207 | pr_debug("Interlinking %s with %s, non-neighbour\n", | |
5208 | i->dev->name, j->dev->name); | |
5d261913 VF |
5209 | ret = __netdev_adjacent_dev_link(i->dev, j->dev); |
5210 | if (ret) | |
5211 | goto rollback_mesh; | |
5212 | } | |
5213 | } | |
5214 | ||
5215 | /* add dev to every upper_dev's upper device */ | |
2f268f12 VF |
5216 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5217 | pr_debug("linking %s's upper device %s with %s\n", | |
5218 | upper_dev->name, i->dev->name, dev->name); | |
5d261913 VF |
5219 | ret = __netdev_adjacent_dev_link(dev, i->dev); |
5220 | if (ret) | |
5221 | goto rollback_upper_mesh; | |
5222 | } | |
5223 | ||
5224 | /* add upper_dev to every dev's lower device */ | |
2f268f12 VF |
5225 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5226 | pr_debug("linking %s's lower device %s with %s\n", dev->name, | |
5227 | i->dev->name, upper_dev->name); | |
5d261913 VF |
5228 | ret = __netdev_adjacent_dev_link(i->dev, upper_dev); |
5229 | if (ret) | |
5230 | goto rollback_lower_mesh; | |
5231 | } | |
9ff162a8 | 5232 | |
42e52bf9 | 5233 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 | 5234 | return 0; |
5d261913 VF |
5235 | |
5236 | rollback_lower_mesh: | |
5237 | to_i = i; | |
2f268f12 | 5238 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5d261913 VF |
5239 | if (i == to_i) |
5240 | break; | |
5241 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); | |
5242 | } | |
5243 | ||
5244 | i = NULL; | |
5245 | ||
5246 | rollback_upper_mesh: | |
5247 | to_i = i; | |
2f268f12 | 5248 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5d261913 VF |
5249 | if (i == to_i) |
5250 | break; | |
5251 | __netdev_adjacent_dev_unlink(dev, i->dev); | |
5252 | } | |
5253 | ||
5254 | i = j = NULL; | |
5255 | ||
5256 | rollback_mesh: | |
5257 | to_i = i; | |
5258 | to_j = j; | |
2f268f12 VF |
5259 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5260 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5d261913 VF |
5261 | if (i == to_i && j == to_j) |
5262 | break; | |
5263 | __netdev_adjacent_dev_unlink(i->dev, j->dev); | |
5264 | } | |
5265 | if (i == to_i) | |
5266 | break; | |
5267 | } | |
5268 | ||
2f268f12 | 5269 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5270 | |
5271 | return ret; | |
9ff162a8 JP |
5272 | } |
5273 | ||
5274 | /** | |
5275 | * netdev_upper_dev_link - Add a link to the upper device | |
5276 | * @dev: device | |
5277 | * @upper_dev: new upper device | |
5278 | * | |
5279 | * Adds a link to device which is upper to this one. The caller must hold | |
5280 | * the RTNL lock. On a failure a negative errno code is returned. | |
5281 | * On success the reference counts are adjusted and the function | |
5282 | * returns zero. | |
5283 | */ | |
5284 | int netdev_upper_dev_link(struct net_device *dev, | |
5285 | struct net_device *upper_dev) | |
5286 | { | |
402dae96 | 5287 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL); |
9ff162a8 JP |
5288 | } |
5289 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
5290 | ||
5291 | /** | |
5292 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
5293 | * @dev: device | |
5294 | * @upper_dev: new upper device | |
5295 | * | |
5296 | * Adds a link to device which is upper to this one. In this case, only | |
5297 | * one master upper device can be linked, although other non-master devices | |
5298 | * might be linked as well. The caller must hold the RTNL lock. | |
5299 | * On a failure a negative errno code is returned. On success the reference | |
5300 | * counts are adjusted and the function returns zero. | |
5301 | */ | |
5302 | int netdev_master_upper_dev_link(struct net_device *dev, | |
5303 | struct net_device *upper_dev) | |
5304 | { | |
402dae96 | 5305 | return __netdev_upper_dev_link(dev, upper_dev, true, NULL); |
9ff162a8 JP |
5306 | } |
5307 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
5308 | ||
402dae96 VF |
5309 | int netdev_master_upper_dev_link_private(struct net_device *dev, |
5310 | struct net_device *upper_dev, | |
5311 | void *private) | |
5312 | { | |
5313 | return __netdev_upper_dev_link(dev, upper_dev, true, private); | |
5314 | } | |
5315 | EXPORT_SYMBOL(netdev_master_upper_dev_link_private); | |
5316 | ||
9ff162a8 JP |
5317 | /** |
5318 | * netdev_upper_dev_unlink - Removes a link to upper device | |
5319 | * @dev: device | |
5320 | * @upper_dev: new upper device | |
5321 | * | |
5322 | * Removes a link to device which is upper to this one. The caller must hold | |
5323 | * the RTNL lock. | |
5324 | */ | |
5325 | void netdev_upper_dev_unlink(struct net_device *dev, | |
5326 | struct net_device *upper_dev) | |
5327 | { | |
5d261913 | 5328 | struct netdev_adjacent *i, *j; |
9ff162a8 JP |
5329 | ASSERT_RTNL(); |
5330 | ||
2f268f12 | 5331 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5332 | |
5333 | /* Here is the tricky part. We must remove all dev's lower | |
5334 | * devices from all upper_dev's upper devices and vice | |
5335 | * versa, to maintain the graph relationship. | |
5336 | */ | |
2f268f12 VF |
5337 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5338 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) | |
5d261913 VF |
5339 | __netdev_adjacent_dev_unlink(i->dev, j->dev); |
5340 | ||
5341 | /* remove also the devices itself from lower/upper device | |
5342 | * list | |
5343 | */ | |
2f268f12 | 5344 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5d261913 VF |
5345 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); |
5346 | ||
2f268f12 | 5347 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) |
5d261913 VF |
5348 | __netdev_adjacent_dev_unlink(dev, i->dev); |
5349 | ||
42e52bf9 | 5350 | call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); |
9ff162a8 JP |
5351 | } |
5352 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
5353 | ||
61bd3857 MS |
5354 | /** |
5355 | * netdev_bonding_info_change - Dispatch event about slave change | |
5356 | * @dev: device | |
4a26e453 | 5357 | * @bonding_info: info to dispatch |
61bd3857 MS |
5358 | * |
5359 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
5360 | * The caller must hold the RTNL lock. | |
5361 | */ | |
5362 | void netdev_bonding_info_change(struct net_device *dev, | |
5363 | struct netdev_bonding_info *bonding_info) | |
5364 | { | |
5365 | struct netdev_notifier_bonding_info info; | |
5366 | ||
5367 | memcpy(&info.bonding_info, bonding_info, | |
5368 | sizeof(struct netdev_bonding_info)); | |
5369 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, | |
5370 | &info.info); | |
5371 | } | |
5372 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
5373 | ||
2ce1ee17 | 5374 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
5375 | { |
5376 | struct netdev_adjacent *iter; | |
5377 | ||
5378 | struct net *net = dev_net(dev); | |
5379 | ||
5380 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
5381 | if (!net_eq(net,dev_net(iter->dev))) | |
5382 | continue; | |
5383 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5384 | &iter->dev->adj_list.lower); | |
5385 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5386 | &dev->adj_list.upper); | |
5387 | } | |
5388 | ||
5389 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
5390 | if (!net_eq(net,dev_net(iter->dev))) | |
5391 | continue; | |
5392 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5393 | &iter->dev->adj_list.upper); | |
5394 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5395 | &dev->adj_list.lower); | |
5396 | } | |
5397 | } | |
5398 | ||
2ce1ee17 | 5399 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
5400 | { |
5401 | struct netdev_adjacent *iter; | |
5402 | ||
5403 | struct net *net = dev_net(dev); | |
5404 | ||
5405 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
5406 | if (!net_eq(net,dev_net(iter->dev))) | |
5407 | continue; | |
5408 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5409 | &iter->dev->adj_list.lower); | |
5410 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5411 | &dev->adj_list.upper); | |
5412 | } | |
5413 | ||
5414 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
5415 | if (!net_eq(net,dev_net(iter->dev))) | |
5416 | continue; | |
5417 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5418 | &iter->dev->adj_list.upper); | |
5419 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5420 | &dev->adj_list.lower); | |
5421 | } | |
5422 | } | |
5423 | ||
5bb025fa | 5424 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 5425 | { |
5bb025fa | 5426 | struct netdev_adjacent *iter; |
402dae96 | 5427 | |
4c75431a AF |
5428 | struct net *net = dev_net(dev); |
5429 | ||
5bb025fa | 5430 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
4c75431a AF |
5431 | if (!net_eq(net,dev_net(iter->dev))) |
5432 | continue; | |
5bb025fa VF |
5433 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
5434 | &iter->dev->adj_list.lower); | |
5435 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5436 | &iter->dev->adj_list.lower); | |
5437 | } | |
402dae96 | 5438 | |
5bb025fa | 5439 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
4c75431a AF |
5440 | if (!net_eq(net,dev_net(iter->dev))) |
5441 | continue; | |
5bb025fa VF |
5442 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
5443 | &iter->dev->adj_list.upper); | |
5444 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5445 | &iter->dev->adj_list.upper); | |
5446 | } | |
402dae96 | 5447 | } |
402dae96 VF |
5448 | |
5449 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
5450 | struct net_device *lower_dev) | |
5451 | { | |
5452 | struct netdev_adjacent *lower; | |
5453 | ||
5454 | if (!lower_dev) | |
5455 | return NULL; | |
5456 | lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower); | |
5457 | if (!lower) | |
5458 | return NULL; | |
5459 | ||
5460 | return lower->private; | |
5461 | } | |
5462 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
5463 | ||
4085ebe8 VY |
5464 | |
5465 | int dev_get_nest_level(struct net_device *dev, | |
5466 | bool (*type_check)(struct net_device *dev)) | |
5467 | { | |
5468 | struct net_device *lower = NULL; | |
5469 | struct list_head *iter; | |
5470 | int max_nest = -1; | |
5471 | int nest; | |
5472 | ||
5473 | ASSERT_RTNL(); | |
5474 | ||
5475 | netdev_for_each_lower_dev(dev, lower, iter) { | |
5476 | nest = dev_get_nest_level(lower, type_check); | |
5477 | if (max_nest < nest) | |
5478 | max_nest = nest; | |
5479 | } | |
5480 | ||
5481 | if (type_check(dev)) | |
5482 | max_nest++; | |
5483 | ||
5484 | return max_nest; | |
5485 | } | |
5486 | EXPORT_SYMBOL(dev_get_nest_level); | |
5487 | ||
b6c40d68 PM |
5488 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
5489 | { | |
d314774c SH |
5490 | const struct net_device_ops *ops = dev->netdev_ops; |
5491 | ||
d2615bf4 | 5492 | if (ops->ndo_change_rx_flags) |
d314774c | 5493 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
5494 | } |
5495 | ||
991fb3f7 | 5496 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5497 | { |
b536db93 | 5498 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
5499 | kuid_t uid; |
5500 | kgid_t gid; | |
1da177e4 | 5501 | |
24023451 PM |
5502 | ASSERT_RTNL(); |
5503 | ||
dad9b335 WC |
5504 | dev->flags |= IFF_PROMISC; |
5505 | dev->promiscuity += inc; | |
5506 | if (dev->promiscuity == 0) { | |
5507 | /* | |
5508 | * Avoid overflow. | |
5509 | * If inc causes overflow, untouch promisc and return error. | |
5510 | */ | |
5511 | if (inc < 0) | |
5512 | dev->flags &= ~IFF_PROMISC; | |
5513 | else { | |
5514 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
5515 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
5516 | dev->name); | |
dad9b335 WC |
5517 | return -EOVERFLOW; |
5518 | } | |
5519 | } | |
52609c0b | 5520 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
5521 | pr_info("device %s %s promiscuous mode\n", |
5522 | dev->name, | |
5523 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
5524 | if (audit_enabled) { |
5525 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
5526 | audit_log(current->audit_context, GFP_ATOMIC, |
5527 | AUDIT_ANOM_PROMISCUOUS, | |
5528 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
5529 | dev->name, (dev->flags & IFF_PROMISC), | |
5530 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 5531 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
5532 | from_kuid(&init_user_ns, uid), |
5533 | from_kgid(&init_user_ns, gid), | |
7759db82 | 5534 | audit_get_sessionid(current)); |
8192b0c4 | 5535 | } |
24023451 | 5536 | |
b6c40d68 | 5537 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 5538 | } |
991fb3f7 ND |
5539 | if (notify) |
5540 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 5541 | return 0; |
1da177e4 LT |
5542 | } |
5543 | ||
4417da66 PM |
5544 | /** |
5545 | * dev_set_promiscuity - update promiscuity count on a device | |
5546 | * @dev: device | |
5547 | * @inc: modifier | |
5548 | * | |
5549 | * Add or remove promiscuity from a device. While the count in the device | |
5550 | * remains above zero the interface remains promiscuous. Once it hits zero | |
5551 | * the device reverts back to normal filtering operation. A negative inc | |
5552 | * value is used to drop promiscuity on the device. | |
dad9b335 | 5553 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 5554 | */ |
dad9b335 | 5555 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 5556 | { |
b536db93 | 5557 | unsigned int old_flags = dev->flags; |
dad9b335 | 5558 | int err; |
4417da66 | 5559 | |
991fb3f7 | 5560 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 5561 | if (err < 0) |
dad9b335 | 5562 | return err; |
4417da66 PM |
5563 | if (dev->flags != old_flags) |
5564 | dev_set_rx_mode(dev); | |
dad9b335 | 5565 | return err; |
4417da66 | 5566 | } |
d1b19dff | 5567 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 5568 | |
991fb3f7 | 5569 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 5570 | { |
991fb3f7 | 5571 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 5572 | |
24023451 PM |
5573 | ASSERT_RTNL(); |
5574 | ||
1da177e4 | 5575 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
5576 | dev->allmulti += inc; |
5577 | if (dev->allmulti == 0) { | |
5578 | /* | |
5579 | * Avoid overflow. | |
5580 | * If inc causes overflow, untouch allmulti and return error. | |
5581 | */ | |
5582 | if (inc < 0) | |
5583 | dev->flags &= ~IFF_ALLMULTI; | |
5584 | else { | |
5585 | dev->allmulti -= inc; | |
7b6cd1ce JP |
5586 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
5587 | dev->name); | |
dad9b335 WC |
5588 | return -EOVERFLOW; |
5589 | } | |
5590 | } | |
24023451 | 5591 | if (dev->flags ^ old_flags) { |
b6c40d68 | 5592 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 5593 | dev_set_rx_mode(dev); |
991fb3f7 ND |
5594 | if (notify) |
5595 | __dev_notify_flags(dev, old_flags, | |
5596 | dev->gflags ^ old_gflags); | |
24023451 | 5597 | } |
dad9b335 | 5598 | return 0; |
4417da66 | 5599 | } |
991fb3f7 ND |
5600 | |
5601 | /** | |
5602 | * dev_set_allmulti - update allmulti count on a device | |
5603 | * @dev: device | |
5604 | * @inc: modifier | |
5605 | * | |
5606 | * Add or remove reception of all multicast frames to a device. While the | |
5607 | * count in the device remains above zero the interface remains listening | |
5608 | * to all interfaces. Once it hits zero the device reverts back to normal | |
5609 | * filtering operation. A negative @inc value is used to drop the counter | |
5610 | * when releasing a resource needing all multicasts. | |
5611 | * Return 0 if successful or a negative errno code on error. | |
5612 | */ | |
5613 | ||
5614 | int dev_set_allmulti(struct net_device *dev, int inc) | |
5615 | { | |
5616 | return __dev_set_allmulti(dev, inc, true); | |
5617 | } | |
d1b19dff | 5618 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
5619 | |
5620 | /* | |
5621 | * Upload unicast and multicast address lists to device and | |
5622 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 5623 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
5624 | * are present. |
5625 | */ | |
5626 | void __dev_set_rx_mode(struct net_device *dev) | |
5627 | { | |
d314774c SH |
5628 | const struct net_device_ops *ops = dev->netdev_ops; |
5629 | ||
4417da66 PM |
5630 | /* dev_open will call this function so the list will stay sane. */ |
5631 | if (!(dev->flags&IFF_UP)) | |
5632 | return; | |
5633 | ||
5634 | if (!netif_device_present(dev)) | |
40b77c94 | 5635 | return; |
4417da66 | 5636 | |
01789349 | 5637 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
5638 | /* Unicast addresses changes may only happen under the rtnl, |
5639 | * therefore calling __dev_set_promiscuity here is safe. | |
5640 | */ | |
32e7bfc4 | 5641 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 5642 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 5643 | dev->uc_promisc = true; |
32e7bfc4 | 5644 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 5645 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 5646 | dev->uc_promisc = false; |
4417da66 | 5647 | } |
4417da66 | 5648 | } |
01789349 JP |
5649 | |
5650 | if (ops->ndo_set_rx_mode) | |
5651 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
5652 | } |
5653 | ||
5654 | void dev_set_rx_mode(struct net_device *dev) | |
5655 | { | |
b9e40857 | 5656 | netif_addr_lock_bh(dev); |
4417da66 | 5657 | __dev_set_rx_mode(dev); |
b9e40857 | 5658 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
5659 | } |
5660 | ||
f0db275a SH |
5661 | /** |
5662 | * dev_get_flags - get flags reported to userspace | |
5663 | * @dev: device | |
5664 | * | |
5665 | * Get the combination of flag bits exported through APIs to userspace. | |
5666 | */ | |
95c96174 | 5667 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 5668 | { |
95c96174 | 5669 | unsigned int flags; |
1da177e4 LT |
5670 | |
5671 | flags = (dev->flags & ~(IFF_PROMISC | | |
5672 | IFF_ALLMULTI | | |
b00055aa SR |
5673 | IFF_RUNNING | |
5674 | IFF_LOWER_UP | | |
5675 | IFF_DORMANT)) | | |
1da177e4 LT |
5676 | (dev->gflags & (IFF_PROMISC | |
5677 | IFF_ALLMULTI)); | |
5678 | ||
b00055aa SR |
5679 | if (netif_running(dev)) { |
5680 | if (netif_oper_up(dev)) | |
5681 | flags |= IFF_RUNNING; | |
5682 | if (netif_carrier_ok(dev)) | |
5683 | flags |= IFF_LOWER_UP; | |
5684 | if (netif_dormant(dev)) | |
5685 | flags |= IFF_DORMANT; | |
5686 | } | |
1da177e4 LT |
5687 | |
5688 | return flags; | |
5689 | } | |
d1b19dff | 5690 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 5691 | |
bd380811 | 5692 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 5693 | { |
b536db93 | 5694 | unsigned int old_flags = dev->flags; |
bd380811 | 5695 | int ret; |
1da177e4 | 5696 | |
24023451 PM |
5697 | ASSERT_RTNL(); |
5698 | ||
1da177e4 LT |
5699 | /* |
5700 | * Set the flags on our device. | |
5701 | */ | |
5702 | ||
5703 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
5704 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
5705 | IFF_AUTOMEDIA)) | | |
5706 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
5707 | IFF_ALLMULTI)); | |
5708 | ||
5709 | /* | |
5710 | * Load in the correct multicast list now the flags have changed. | |
5711 | */ | |
5712 | ||
b6c40d68 PM |
5713 | if ((old_flags ^ flags) & IFF_MULTICAST) |
5714 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 5715 | |
4417da66 | 5716 | dev_set_rx_mode(dev); |
1da177e4 LT |
5717 | |
5718 | /* | |
5719 | * Have we downed the interface. We handle IFF_UP ourselves | |
5720 | * according to user attempts to set it, rather than blindly | |
5721 | * setting it. | |
5722 | */ | |
5723 | ||
5724 | ret = 0; | |
d215d10f | 5725 | if ((old_flags ^ flags) & IFF_UP) |
bd380811 | 5726 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 | 5727 | |
1da177e4 | 5728 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 5729 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 5730 | unsigned int old_flags = dev->flags; |
d1b19dff | 5731 | |
1da177e4 | 5732 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
5733 | |
5734 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
5735 | if (dev->flags != old_flags) | |
5736 | dev_set_rx_mode(dev); | |
1da177e4 LT |
5737 | } |
5738 | ||
5739 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
5740 | is important. Some (broken) drivers set IFF_PROMISC, when | |
5741 | IFF_ALLMULTI is requested not asking us and not reporting. | |
5742 | */ | |
5743 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
5744 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
5745 | ||
1da177e4 | 5746 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 5747 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
5748 | } |
5749 | ||
bd380811 PM |
5750 | return ret; |
5751 | } | |
5752 | ||
a528c219 ND |
5753 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
5754 | unsigned int gchanges) | |
bd380811 PM |
5755 | { |
5756 | unsigned int changes = dev->flags ^ old_flags; | |
5757 | ||
a528c219 | 5758 | if (gchanges) |
7f294054 | 5759 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 5760 | |
bd380811 PM |
5761 | if (changes & IFF_UP) { |
5762 | if (dev->flags & IFF_UP) | |
5763 | call_netdevice_notifiers(NETDEV_UP, dev); | |
5764 | else | |
5765 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
5766 | } | |
5767 | ||
5768 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
5769 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
5770 | struct netdev_notifier_change_info change_info; | |
5771 | ||
5772 | change_info.flags_changed = changes; | |
5773 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
5774 | &change_info.info); | |
5775 | } | |
bd380811 PM |
5776 | } |
5777 | ||
5778 | /** | |
5779 | * dev_change_flags - change device settings | |
5780 | * @dev: device | |
5781 | * @flags: device state flags | |
5782 | * | |
5783 | * Change settings on device based state flags. The flags are | |
5784 | * in the userspace exported format. | |
5785 | */ | |
b536db93 | 5786 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 5787 | { |
b536db93 | 5788 | int ret; |
991fb3f7 | 5789 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
5790 | |
5791 | ret = __dev_change_flags(dev, flags); | |
5792 | if (ret < 0) | |
5793 | return ret; | |
5794 | ||
991fb3f7 | 5795 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 5796 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
5797 | return ret; |
5798 | } | |
d1b19dff | 5799 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 5800 | |
2315dc91 VF |
5801 | static int __dev_set_mtu(struct net_device *dev, int new_mtu) |
5802 | { | |
5803 | const struct net_device_ops *ops = dev->netdev_ops; | |
5804 | ||
5805 | if (ops->ndo_change_mtu) | |
5806 | return ops->ndo_change_mtu(dev, new_mtu); | |
5807 | ||
5808 | dev->mtu = new_mtu; | |
5809 | return 0; | |
5810 | } | |
5811 | ||
f0db275a SH |
5812 | /** |
5813 | * dev_set_mtu - Change maximum transfer unit | |
5814 | * @dev: device | |
5815 | * @new_mtu: new transfer unit | |
5816 | * | |
5817 | * Change the maximum transfer size of the network device. | |
5818 | */ | |
1da177e4 LT |
5819 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
5820 | { | |
2315dc91 | 5821 | int err, orig_mtu; |
1da177e4 LT |
5822 | |
5823 | if (new_mtu == dev->mtu) | |
5824 | return 0; | |
5825 | ||
5826 | /* MTU must be positive. */ | |
5827 | if (new_mtu < 0) | |
5828 | return -EINVAL; | |
5829 | ||
5830 | if (!netif_device_present(dev)) | |
5831 | return -ENODEV; | |
5832 | ||
1d486bfb VF |
5833 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
5834 | err = notifier_to_errno(err); | |
5835 | if (err) | |
5836 | return err; | |
d314774c | 5837 | |
2315dc91 VF |
5838 | orig_mtu = dev->mtu; |
5839 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 5840 | |
2315dc91 VF |
5841 | if (!err) { |
5842 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
5843 | err = notifier_to_errno(err); | |
5844 | if (err) { | |
5845 | /* setting mtu back and notifying everyone again, | |
5846 | * so that they have a chance to revert changes. | |
5847 | */ | |
5848 | __dev_set_mtu(dev, orig_mtu); | |
5849 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
5850 | } | |
5851 | } | |
1da177e4 LT |
5852 | return err; |
5853 | } | |
d1b19dff | 5854 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 5855 | |
cbda10fa VD |
5856 | /** |
5857 | * dev_set_group - Change group this device belongs to | |
5858 | * @dev: device | |
5859 | * @new_group: group this device should belong to | |
5860 | */ | |
5861 | void dev_set_group(struct net_device *dev, int new_group) | |
5862 | { | |
5863 | dev->group = new_group; | |
5864 | } | |
5865 | EXPORT_SYMBOL(dev_set_group); | |
5866 | ||
f0db275a SH |
5867 | /** |
5868 | * dev_set_mac_address - Change Media Access Control Address | |
5869 | * @dev: device | |
5870 | * @sa: new address | |
5871 | * | |
5872 | * Change the hardware (MAC) address of the device | |
5873 | */ | |
1da177e4 LT |
5874 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
5875 | { | |
d314774c | 5876 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
5877 | int err; |
5878 | ||
d314774c | 5879 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
5880 | return -EOPNOTSUPP; |
5881 | if (sa->sa_family != dev->type) | |
5882 | return -EINVAL; | |
5883 | if (!netif_device_present(dev)) | |
5884 | return -ENODEV; | |
d314774c | 5885 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
5886 | if (err) |
5887 | return err; | |
fbdeca2d | 5888 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 5889 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 5890 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 5891 | return 0; |
1da177e4 | 5892 | } |
d1b19dff | 5893 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 5894 | |
4bf84c35 JP |
5895 | /** |
5896 | * dev_change_carrier - Change device carrier | |
5897 | * @dev: device | |
691b3b7e | 5898 | * @new_carrier: new value |
4bf84c35 JP |
5899 | * |
5900 | * Change device carrier | |
5901 | */ | |
5902 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
5903 | { | |
5904 | const struct net_device_ops *ops = dev->netdev_ops; | |
5905 | ||
5906 | if (!ops->ndo_change_carrier) | |
5907 | return -EOPNOTSUPP; | |
5908 | if (!netif_device_present(dev)) | |
5909 | return -ENODEV; | |
5910 | return ops->ndo_change_carrier(dev, new_carrier); | |
5911 | } | |
5912 | EXPORT_SYMBOL(dev_change_carrier); | |
5913 | ||
66b52b0d JP |
5914 | /** |
5915 | * dev_get_phys_port_id - Get device physical port ID | |
5916 | * @dev: device | |
5917 | * @ppid: port ID | |
5918 | * | |
5919 | * Get device physical port ID | |
5920 | */ | |
5921 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 5922 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
5923 | { |
5924 | const struct net_device_ops *ops = dev->netdev_ops; | |
5925 | ||
5926 | if (!ops->ndo_get_phys_port_id) | |
5927 | return -EOPNOTSUPP; | |
5928 | return ops->ndo_get_phys_port_id(dev, ppid); | |
5929 | } | |
5930 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
5931 | ||
db24a904 DA |
5932 | /** |
5933 | * dev_get_phys_port_name - Get device physical port name | |
5934 | * @dev: device | |
5935 | * @name: port name | |
5936 | * | |
5937 | * Get device physical port name | |
5938 | */ | |
5939 | int dev_get_phys_port_name(struct net_device *dev, | |
5940 | char *name, size_t len) | |
5941 | { | |
5942 | const struct net_device_ops *ops = dev->netdev_ops; | |
5943 | ||
5944 | if (!ops->ndo_get_phys_port_name) | |
5945 | return -EOPNOTSUPP; | |
5946 | return ops->ndo_get_phys_port_name(dev, name, len); | |
5947 | } | |
5948 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
5949 | ||
1da177e4 LT |
5950 | /** |
5951 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5952 | * @net: the applicable net namespace |
1da177e4 LT |
5953 | * |
5954 | * Returns a suitable unique value for a new device interface | |
5955 | * number. The caller must hold the rtnl semaphore or the | |
5956 | * dev_base_lock to be sure it remains unique. | |
5957 | */ | |
881d966b | 5958 | static int dev_new_index(struct net *net) |
1da177e4 | 5959 | { |
aa79e66e | 5960 | int ifindex = net->ifindex; |
1da177e4 LT |
5961 | for (;;) { |
5962 | if (++ifindex <= 0) | |
5963 | ifindex = 1; | |
881d966b | 5964 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 5965 | return net->ifindex = ifindex; |
1da177e4 LT |
5966 | } |
5967 | } | |
5968 | ||
1da177e4 | 5969 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5970 | static LIST_HEAD(net_todo_list); |
200b916f | 5971 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 5972 | |
6f05f629 | 5973 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5974 | { |
1da177e4 | 5975 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 5976 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
5977 | } |
5978 | ||
9b5e383c | 5979 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5980 | { |
e93737b0 | 5981 | struct net_device *dev, *tmp; |
5cde2829 | 5982 | LIST_HEAD(close_head); |
9b5e383c | 5983 | |
93ee31f1 DL |
5984 | BUG_ON(dev_boot_phase); |
5985 | ASSERT_RTNL(); | |
5986 | ||
e93737b0 | 5987 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5988 | /* Some devices call without registering |
e93737b0 KK |
5989 | * for initialization unwind. Remove those |
5990 | * devices and proceed with the remaining. | |
9b5e383c ED |
5991 | */ |
5992 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
5993 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
5994 | dev->name, dev); | |
93ee31f1 | 5995 | |
9b5e383c | 5996 | WARN_ON(1); |
e93737b0 KK |
5997 | list_del(&dev->unreg_list); |
5998 | continue; | |
9b5e383c | 5999 | } |
449f4544 | 6000 | dev->dismantle = true; |
9b5e383c | 6001 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 6002 | } |
93ee31f1 | 6003 | |
44345724 | 6004 | /* If device is running, close it first. */ |
5cde2829 EB |
6005 | list_for_each_entry(dev, head, unreg_list) |
6006 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 6007 | dev_close_many(&close_head, true); |
93ee31f1 | 6008 | |
44345724 | 6009 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
6010 | /* And unlink it from device chain. */ |
6011 | unlist_netdevice(dev); | |
93ee31f1 | 6012 | |
9b5e383c ED |
6013 | dev->reg_state = NETREG_UNREGISTERING; |
6014 | } | |
93ee31f1 DL |
6015 | |
6016 | synchronize_net(); | |
6017 | ||
9b5e383c | 6018 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
6019 | struct sk_buff *skb = NULL; |
6020 | ||
9b5e383c ED |
6021 | /* Shutdown queueing discipline. */ |
6022 | dev_shutdown(dev); | |
93ee31f1 DL |
6023 | |
6024 | ||
9b5e383c ED |
6025 | /* Notify protocols, that we are about to destroy |
6026 | this device. They should clean all the things. | |
6027 | */ | |
6028 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 6029 | |
395eea6c MB |
6030 | if (!dev->rtnl_link_ops || |
6031 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
6032 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, | |
6033 | GFP_KERNEL); | |
6034 | ||
9b5e383c ED |
6035 | /* |
6036 | * Flush the unicast and multicast chains | |
6037 | */ | |
a748ee24 | 6038 | dev_uc_flush(dev); |
22bedad3 | 6039 | dev_mc_flush(dev); |
93ee31f1 | 6040 | |
9b5e383c ED |
6041 | if (dev->netdev_ops->ndo_uninit) |
6042 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 6043 | |
395eea6c MB |
6044 | if (skb) |
6045 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 6046 | |
9ff162a8 JP |
6047 | /* Notifier chain MUST detach us all upper devices. */ |
6048 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 6049 | |
9b5e383c ED |
6050 | /* Remove entries from kobject tree */ |
6051 | netdev_unregister_kobject(dev); | |
024e9679 AD |
6052 | #ifdef CONFIG_XPS |
6053 | /* Remove XPS queueing entries */ | |
6054 | netif_reset_xps_queues_gt(dev, 0); | |
6055 | #endif | |
9b5e383c | 6056 | } |
93ee31f1 | 6057 | |
850a545b | 6058 | synchronize_net(); |
395264d5 | 6059 | |
a5ee1551 | 6060 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
6061 | dev_put(dev); |
6062 | } | |
6063 | ||
6064 | static void rollback_registered(struct net_device *dev) | |
6065 | { | |
6066 | LIST_HEAD(single); | |
6067 | ||
6068 | list_add(&dev->unreg_list, &single); | |
6069 | rollback_registered_many(&single); | |
ceaaec98 | 6070 | list_del(&single); |
93ee31f1 DL |
6071 | } |
6072 | ||
c8f44aff MM |
6073 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
6074 | netdev_features_t features) | |
b63365a2 | 6075 | { |
57422dc5 MM |
6076 | /* Fix illegal checksum combinations */ |
6077 | if ((features & NETIF_F_HW_CSUM) && | |
6078 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 6079 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
6080 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
6081 | } | |
6082 | ||
b63365a2 | 6083 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 6084 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 6085 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 6086 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
6087 | } |
6088 | ||
ec5f0615 PS |
6089 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
6090 | !(features & NETIF_F_IP_CSUM)) { | |
6091 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
6092 | features &= ~NETIF_F_TSO; | |
6093 | features &= ~NETIF_F_TSO_ECN; | |
6094 | } | |
6095 | ||
6096 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
6097 | !(features & NETIF_F_IPV6_CSUM)) { | |
6098 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
6099 | features &= ~NETIF_F_TSO6; | |
6100 | } | |
6101 | ||
31d8b9e0 BH |
6102 | /* TSO ECN requires that TSO is present as well. */ |
6103 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
6104 | features &= ~NETIF_F_TSO_ECN; | |
6105 | ||
212b573f MM |
6106 | /* Software GSO depends on SG. */ |
6107 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 6108 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
6109 | features &= ~NETIF_F_GSO; |
6110 | } | |
6111 | ||
acd1130e | 6112 | /* UFO needs SG and checksumming */ |
b63365a2 | 6113 | if (features & NETIF_F_UFO) { |
79032644 MM |
6114 | /* maybe split UFO into V4 and V6? */ |
6115 | if (!((features & NETIF_F_GEN_CSUM) || | |
6116 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
6117 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 6118 | netdev_dbg(dev, |
acd1130e | 6119 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
6120 | features &= ~NETIF_F_UFO; |
6121 | } | |
6122 | ||
6123 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 6124 | netdev_dbg(dev, |
acd1130e | 6125 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
6126 | features &= ~NETIF_F_UFO; |
6127 | } | |
6128 | } | |
6129 | ||
d0290214 JP |
6130 | #ifdef CONFIG_NET_RX_BUSY_POLL |
6131 | if (dev->netdev_ops->ndo_busy_poll) | |
6132 | features |= NETIF_F_BUSY_POLL; | |
6133 | else | |
6134 | #endif | |
6135 | features &= ~NETIF_F_BUSY_POLL; | |
6136 | ||
b63365a2 HX |
6137 | return features; |
6138 | } | |
b63365a2 | 6139 | |
6cb6a27c | 6140 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 6141 | { |
c8f44aff | 6142 | netdev_features_t features; |
5455c699 MM |
6143 | int err = 0; |
6144 | ||
87267485 MM |
6145 | ASSERT_RTNL(); |
6146 | ||
5455c699 MM |
6147 | features = netdev_get_wanted_features(dev); |
6148 | ||
6149 | if (dev->netdev_ops->ndo_fix_features) | |
6150 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
6151 | ||
6152 | /* driver might be less strict about feature dependencies */ | |
6153 | features = netdev_fix_features(dev, features); | |
6154 | ||
6155 | if (dev->features == features) | |
6cb6a27c | 6156 | return 0; |
5455c699 | 6157 | |
c8f44aff MM |
6158 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
6159 | &dev->features, &features); | |
5455c699 MM |
6160 | |
6161 | if (dev->netdev_ops->ndo_set_features) | |
6162 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
6163 | ||
6cb6a27c | 6164 | if (unlikely(err < 0)) { |
5455c699 | 6165 | netdev_err(dev, |
c8f44aff MM |
6166 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
6167 | err, &features, &dev->features); | |
6cb6a27c MM |
6168 | return -1; |
6169 | } | |
6170 | ||
6171 | if (!err) | |
6172 | dev->features = features; | |
6173 | ||
6174 | return 1; | |
6175 | } | |
6176 | ||
afe12cc8 MM |
6177 | /** |
6178 | * netdev_update_features - recalculate device features | |
6179 | * @dev: the device to check | |
6180 | * | |
6181 | * Recalculate dev->features set and send notifications if it | |
6182 | * has changed. Should be called after driver or hardware dependent | |
6183 | * conditions might have changed that influence the features. | |
6184 | */ | |
6cb6a27c MM |
6185 | void netdev_update_features(struct net_device *dev) |
6186 | { | |
6187 | if (__netdev_update_features(dev)) | |
6188 | netdev_features_change(dev); | |
5455c699 MM |
6189 | } |
6190 | EXPORT_SYMBOL(netdev_update_features); | |
6191 | ||
afe12cc8 MM |
6192 | /** |
6193 | * netdev_change_features - recalculate device features | |
6194 | * @dev: the device to check | |
6195 | * | |
6196 | * Recalculate dev->features set and send notifications even | |
6197 | * if they have not changed. Should be called instead of | |
6198 | * netdev_update_features() if also dev->vlan_features might | |
6199 | * have changed to allow the changes to be propagated to stacked | |
6200 | * VLAN devices. | |
6201 | */ | |
6202 | void netdev_change_features(struct net_device *dev) | |
6203 | { | |
6204 | __netdev_update_features(dev); | |
6205 | netdev_features_change(dev); | |
6206 | } | |
6207 | EXPORT_SYMBOL(netdev_change_features); | |
6208 | ||
fc4a7489 PM |
6209 | /** |
6210 | * netif_stacked_transfer_operstate - transfer operstate | |
6211 | * @rootdev: the root or lower level device to transfer state from | |
6212 | * @dev: the device to transfer operstate to | |
6213 | * | |
6214 | * Transfer operational state from root to device. This is normally | |
6215 | * called when a stacking relationship exists between the root | |
6216 | * device and the device(a leaf device). | |
6217 | */ | |
6218 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
6219 | struct net_device *dev) | |
6220 | { | |
6221 | if (rootdev->operstate == IF_OPER_DORMANT) | |
6222 | netif_dormant_on(dev); | |
6223 | else | |
6224 | netif_dormant_off(dev); | |
6225 | ||
6226 | if (netif_carrier_ok(rootdev)) { | |
6227 | if (!netif_carrier_ok(dev)) | |
6228 | netif_carrier_on(dev); | |
6229 | } else { | |
6230 | if (netif_carrier_ok(dev)) | |
6231 | netif_carrier_off(dev); | |
6232 | } | |
6233 | } | |
6234 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
6235 | ||
a953be53 | 6236 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
6237 | static int netif_alloc_rx_queues(struct net_device *dev) |
6238 | { | |
1b4bf461 | 6239 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 6240 | struct netdev_rx_queue *rx; |
10595902 | 6241 | size_t sz = count * sizeof(*rx); |
1b4bf461 | 6242 | |
bd25fa7b | 6243 | BUG_ON(count < 1); |
1b4bf461 | 6244 | |
10595902 PG |
6245 | rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6246 | if (!rx) { | |
6247 | rx = vzalloc(sz); | |
6248 | if (!rx) | |
6249 | return -ENOMEM; | |
6250 | } | |
bd25fa7b TH |
6251 | dev->_rx = rx; |
6252 | ||
bd25fa7b | 6253 | for (i = 0; i < count; i++) |
fe822240 | 6254 | rx[i].dev = dev; |
1b4bf461 ED |
6255 | return 0; |
6256 | } | |
bf264145 | 6257 | #endif |
1b4bf461 | 6258 | |
aa942104 CG |
6259 | static void netdev_init_one_queue(struct net_device *dev, |
6260 | struct netdev_queue *queue, void *_unused) | |
6261 | { | |
6262 | /* Initialize queue lock */ | |
6263 | spin_lock_init(&queue->_xmit_lock); | |
6264 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
6265 | queue->xmit_lock_owner = -1; | |
b236da69 | 6266 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 6267 | queue->dev = dev; |
114cf580 TH |
6268 | #ifdef CONFIG_BQL |
6269 | dql_init(&queue->dql, HZ); | |
6270 | #endif | |
aa942104 CG |
6271 | } |
6272 | ||
60877a32 ED |
6273 | static void netif_free_tx_queues(struct net_device *dev) |
6274 | { | |
4cb28970 | 6275 | kvfree(dev->_tx); |
60877a32 ED |
6276 | } |
6277 | ||
e6484930 TH |
6278 | static int netif_alloc_netdev_queues(struct net_device *dev) |
6279 | { | |
6280 | unsigned int count = dev->num_tx_queues; | |
6281 | struct netdev_queue *tx; | |
60877a32 | 6282 | size_t sz = count * sizeof(*tx); |
e6484930 | 6283 | |
60877a32 | 6284 | BUG_ON(count < 1 || count > 0xffff); |
62b5942a | 6285 | |
60877a32 ED |
6286 | tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6287 | if (!tx) { | |
6288 | tx = vzalloc(sz); | |
6289 | if (!tx) | |
6290 | return -ENOMEM; | |
6291 | } | |
e6484930 | 6292 | dev->_tx = tx; |
1d24eb48 | 6293 | |
e6484930 TH |
6294 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
6295 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
6296 | |
6297 | return 0; | |
e6484930 TH |
6298 | } |
6299 | ||
1da177e4 LT |
6300 | /** |
6301 | * register_netdevice - register a network device | |
6302 | * @dev: device to register | |
6303 | * | |
6304 | * Take a completed network device structure and add it to the kernel | |
6305 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6306 | * chain. 0 is returned on success. A negative errno code is returned | |
6307 | * on a failure to set up the device, or if the name is a duplicate. | |
6308 | * | |
6309 | * Callers must hold the rtnl semaphore. You may want | |
6310 | * register_netdev() instead of this. | |
6311 | * | |
6312 | * BUGS: | |
6313 | * The locking appears insufficient to guarantee two parallel registers | |
6314 | * will not get the same name. | |
6315 | */ | |
6316 | ||
6317 | int register_netdevice(struct net_device *dev) | |
6318 | { | |
1da177e4 | 6319 | int ret; |
d314774c | 6320 | struct net *net = dev_net(dev); |
1da177e4 LT |
6321 | |
6322 | BUG_ON(dev_boot_phase); | |
6323 | ASSERT_RTNL(); | |
6324 | ||
b17a7c17 SH |
6325 | might_sleep(); |
6326 | ||
1da177e4 LT |
6327 | /* When net_device's are persistent, this will be fatal. */ |
6328 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 6329 | BUG_ON(!net); |
1da177e4 | 6330 | |
f1f28aa3 | 6331 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 6332 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 6333 | |
1da177e4 LT |
6334 | dev->iflink = -1; |
6335 | ||
828de4f6 | 6336 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
6337 | if (ret < 0) |
6338 | goto out; | |
6339 | ||
1da177e4 | 6340 | /* Init, if this function is available */ |
d314774c SH |
6341 | if (dev->netdev_ops->ndo_init) { |
6342 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
6343 | if (ret) { |
6344 | if (ret > 0) | |
6345 | ret = -EIO; | |
90833aa4 | 6346 | goto out; |
1da177e4 LT |
6347 | } |
6348 | } | |
4ec93edb | 6349 | |
f646968f PM |
6350 | if (((dev->hw_features | dev->features) & |
6351 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
6352 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
6353 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
6354 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
6355 | ret = -EINVAL; | |
6356 | goto err_uninit; | |
6357 | } | |
6358 | ||
9c7dafbf PE |
6359 | ret = -EBUSY; |
6360 | if (!dev->ifindex) | |
6361 | dev->ifindex = dev_new_index(net); | |
6362 | else if (__dev_get_by_index(net, dev->ifindex)) | |
6363 | goto err_uninit; | |
6364 | ||
a54acb3a | 6365 | if (dev_get_iflink(dev) == -1) |
1da177e4 LT |
6366 | dev->iflink = dev->ifindex; |
6367 | ||
5455c699 MM |
6368 | /* Transfer changeable features to wanted_features and enable |
6369 | * software offloads (GSO and GRO). | |
6370 | */ | |
6371 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
6372 | dev->features |= NETIF_F_SOFT_FEATURES; |
6373 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 6374 | |
34324dc2 MM |
6375 | if (!(dev->flags & IFF_LOOPBACK)) { |
6376 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
c6e1a0d1 TH |
6377 | } |
6378 | ||
1180e7d6 | 6379 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 6380 | */ |
1180e7d6 | 6381 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 6382 | |
ee579677 PS |
6383 | /* Make NETIF_F_SG inheritable to tunnel devices. |
6384 | */ | |
6385 | dev->hw_enc_features |= NETIF_F_SG; | |
6386 | ||
0d89d203 SH |
6387 | /* Make NETIF_F_SG inheritable to MPLS. |
6388 | */ | |
6389 | dev->mpls_features |= NETIF_F_SG; | |
6390 | ||
7ffbe3fd JB |
6391 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
6392 | ret = notifier_to_errno(ret); | |
6393 | if (ret) | |
6394 | goto err_uninit; | |
6395 | ||
8b41d188 | 6396 | ret = netdev_register_kobject(dev); |
b17a7c17 | 6397 | if (ret) |
7ce1b0ed | 6398 | goto err_uninit; |
b17a7c17 SH |
6399 | dev->reg_state = NETREG_REGISTERED; |
6400 | ||
6cb6a27c | 6401 | __netdev_update_features(dev); |
8e9b59b2 | 6402 | |
1da177e4 LT |
6403 | /* |
6404 | * Default initial state at registry is that the | |
6405 | * device is present. | |
6406 | */ | |
6407 | ||
6408 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6409 | ||
8f4cccbb BH |
6410 | linkwatch_init_dev(dev); |
6411 | ||
1da177e4 | 6412 | dev_init_scheduler(dev); |
1da177e4 | 6413 | dev_hold(dev); |
ce286d32 | 6414 | list_netdevice(dev); |
7bf23575 | 6415 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 6416 | |
948b337e JP |
6417 | /* If the device has permanent device address, driver should |
6418 | * set dev_addr and also addr_assign_type should be set to | |
6419 | * NET_ADDR_PERM (default value). | |
6420 | */ | |
6421 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
6422 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
6423 | ||
1da177e4 | 6424 | /* Notify protocols, that a new device appeared. */ |
056925ab | 6425 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 6426 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
6427 | if (ret) { |
6428 | rollback_registered(dev); | |
6429 | dev->reg_state = NETREG_UNREGISTERED; | |
6430 | } | |
d90a909e EB |
6431 | /* |
6432 | * Prevent userspace races by waiting until the network | |
6433 | * device is fully setup before sending notifications. | |
6434 | */ | |
a2835763 PM |
6435 | if (!dev->rtnl_link_ops || |
6436 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 6437 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
6438 | |
6439 | out: | |
6440 | return ret; | |
7ce1b0ed HX |
6441 | |
6442 | err_uninit: | |
d314774c SH |
6443 | if (dev->netdev_ops->ndo_uninit) |
6444 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 6445 | goto out; |
1da177e4 | 6446 | } |
d1b19dff | 6447 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 6448 | |
937f1ba5 BH |
6449 | /** |
6450 | * init_dummy_netdev - init a dummy network device for NAPI | |
6451 | * @dev: device to init | |
6452 | * | |
6453 | * This takes a network device structure and initialize the minimum | |
6454 | * amount of fields so it can be used to schedule NAPI polls without | |
6455 | * registering a full blown interface. This is to be used by drivers | |
6456 | * that need to tie several hardware interfaces to a single NAPI | |
6457 | * poll scheduler due to HW limitations. | |
6458 | */ | |
6459 | int init_dummy_netdev(struct net_device *dev) | |
6460 | { | |
6461 | /* Clear everything. Note we don't initialize spinlocks | |
6462 | * are they aren't supposed to be taken by any of the | |
6463 | * NAPI code and this dummy netdev is supposed to be | |
6464 | * only ever used for NAPI polls | |
6465 | */ | |
6466 | memset(dev, 0, sizeof(struct net_device)); | |
6467 | ||
6468 | /* make sure we BUG if trying to hit standard | |
6469 | * register/unregister code path | |
6470 | */ | |
6471 | dev->reg_state = NETREG_DUMMY; | |
6472 | ||
937f1ba5 BH |
6473 | /* NAPI wants this */ |
6474 | INIT_LIST_HEAD(&dev->napi_list); | |
6475 | ||
6476 | /* a dummy interface is started by default */ | |
6477 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6478 | set_bit(__LINK_STATE_START, &dev->state); | |
6479 | ||
29b4433d ED |
6480 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
6481 | * because users of this 'device' dont need to change | |
6482 | * its refcount. | |
6483 | */ | |
6484 | ||
937f1ba5 BH |
6485 | return 0; |
6486 | } | |
6487 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
6488 | ||
6489 | ||
1da177e4 LT |
6490 | /** |
6491 | * register_netdev - register a network device | |
6492 | * @dev: device to register | |
6493 | * | |
6494 | * Take a completed network device structure and add it to the kernel | |
6495 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6496 | * chain. 0 is returned on success. A negative errno code is returned | |
6497 | * on a failure to set up the device, or if the name is a duplicate. | |
6498 | * | |
38b4da38 | 6499 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
6500 | * and expands the device name if you passed a format string to |
6501 | * alloc_netdev. | |
6502 | */ | |
6503 | int register_netdev(struct net_device *dev) | |
6504 | { | |
6505 | int err; | |
6506 | ||
6507 | rtnl_lock(); | |
1da177e4 | 6508 | err = register_netdevice(dev); |
1da177e4 LT |
6509 | rtnl_unlock(); |
6510 | return err; | |
6511 | } | |
6512 | EXPORT_SYMBOL(register_netdev); | |
6513 | ||
29b4433d ED |
6514 | int netdev_refcnt_read(const struct net_device *dev) |
6515 | { | |
6516 | int i, refcnt = 0; | |
6517 | ||
6518 | for_each_possible_cpu(i) | |
6519 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
6520 | return refcnt; | |
6521 | } | |
6522 | EXPORT_SYMBOL(netdev_refcnt_read); | |
6523 | ||
2c53040f | 6524 | /** |
1da177e4 | 6525 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 6526 | * @dev: target net_device |
1da177e4 LT |
6527 | * |
6528 | * This is called when unregistering network devices. | |
6529 | * | |
6530 | * Any protocol or device that holds a reference should register | |
6531 | * for netdevice notification, and cleanup and put back the | |
6532 | * reference if they receive an UNREGISTER event. | |
6533 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 6534 | * call dev_put. |
1da177e4 LT |
6535 | */ |
6536 | static void netdev_wait_allrefs(struct net_device *dev) | |
6537 | { | |
6538 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 6539 | int refcnt; |
1da177e4 | 6540 | |
e014debe ED |
6541 | linkwatch_forget_dev(dev); |
6542 | ||
1da177e4 | 6543 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
6544 | refcnt = netdev_refcnt_read(dev); |
6545 | ||
6546 | while (refcnt != 0) { | |
1da177e4 | 6547 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 6548 | rtnl_lock(); |
1da177e4 LT |
6549 | |
6550 | /* Rebroadcast unregister notification */ | |
056925ab | 6551 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 6552 | |
748e2d93 | 6553 | __rtnl_unlock(); |
0115e8e3 | 6554 | rcu_barrier(); |
748e2d93 ED |
6555 | rtnl_lock(); |
6556 | ||
0115e8e3 | 6557 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
6558 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
6559 | &dev->state)) { | |
6560 | /* We must not have linkwatch events | |
6561 | * pending on unregister. If this | |
6562 | * happens, we simply run the queue | |
6563 | * unscheduled, resulting in a noop | |
6564 | * for this device. | |
6565 | */ | |
6566 | linkwatch_run_queue(); | |
6567 | } | |
6568 | ||
6756ae4b | 6569 | __rtnl_unlock(); |
1da177e4 LT |
6570 | |
6571 | rebroadcast_time = jiffies; | |
6572 | } | |
6573 | ||
6574 | msleep(250); | |
6575 | ||
29b4433d ED |
6576 | refcnt = netdev_refcnt_read(dev); |
6577 | ||
1da177e4 | 6578 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
6579 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
6580 | dev->name, refcnt); | |
1da177e4 LT |
6581 | warning_time = jiffies; |
6582 | } | |
6583 | } | |
6584 | } | |
6585 | ||
6586 | /* The sequence is: | |
6587 | * | |
6588 | * rtnl_lock(); | |
6589 | * ... | |
6590 | * register_netdevice(x1); | |
6591 | * register_netdevice(x2); | |
6592 | * ... | |
6593 | * unregister_netdevice(y1); | |
6594 | * unregister_netdevice(y2); | |
6595 | * ... | |
6596 | * rtnl_unlock(); | |
6597 | * free_netdev(y1); | |
6598 | * free_netdev(y2); | |
6599 | * | |
58ec3b4d | 6600 | * We are invoked by rtnl_unlock(). |
1da177e4 | 6601 | * This allows us to deal with problems: |
b17a7c17 | 6602 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
6603 | * without deadlocking with linkwatch via keventd. |
6604 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
6605 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
6606 | * |
6607 | * We must not return until all unregister events added during | |
6608 | * the interval the lock was held have been completed. | |
1da177e4 | 6609 | */ |
1da177e4 LT |
6610 | void netdev_run_todo(void) |
6611 | { | |
626ab0e6 | 6612 | struct list_head list; |
1da177e4 | 6613 | |
1da177e4 | 6614 | /* Snapshot list, allow later requests */ |
626ab0e6 | 6615 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
6616 | |
6617 | __rtnl_unlock(); | |
626ab0e6 | 6618 | |
0115e8e3 ED |
6619 | |
6620 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
6621 | if (!list_empty(&list)) |
6622 | rcu_barrier(); | |
6623 | ||
1da177e4 LT |
6624 | while (!list_empty(&list)) { |
6625 | struct net_device *dev | |
e5e26d75 | 6626 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
6627 | list_del(&dev->todo_list); |
6628 | ||
748e2d93 | 6629 | rtnl_lock(); |
0115e8e3 | 6630 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 6631 | __rtnl_unlock(); |
0115e8e3 | 6632 | |
b17a7c17 | 6633 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 6634 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
6635 | dev->name, dev->reg_state); |
6636 | dump_stack(); | |
6637 | continue; | |
6638 | } | |
1da177e4 | 6639 | |
b17a7c17 | 6640 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 6641 | |
152102c7 | 6642 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 6643 | |
b17a7c17 | 6644 | netdev_wait_allrefs(dev); |
1da177e4 | 6645 | |
b17a7c17 | 6646 | /* paranoia */ |
29b4433d | 6647 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
6648 | BUG_ON(!list_empty(&dev->ptype_all)); |
6649 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
6650 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
6651 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 6652 | WARN_ON(dev->dn_ptr); |
1da177e4 | 6653 | |
b17a7c17 SH |
6654 | if (dev->destructor) |
6655 | dev->destructor(dev); | |
9093bbb2 | 6656 | |
50624c93 EB |
6657 | /* Report a network device has been unregistered */ |
6658 | rtnl_lock(); | |
6659 | dev_net(dev)->dev_unreg_count--; | |
6660 | __rtnl_unlock(); | |
6661 | wake_up(&netdev_unregistering_wq); | |
6662 | ||
9093bbb2 SH |
6663 | /* Free network device */ |
6664 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 6665 | } |
1da177e4 LT |
6666 | } |
6667 | ||
3cfde79c BH |
6668 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
6669 | * fields in the same order, with only the type differing. | |
6670 | */ | |
77a1abf5 ED |
6671 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
6672 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
6673 | { |
6674 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
6675 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
6676 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
6677 | #else |
6678 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
6679 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
6680 | u64 *dst = (u64 *)stats64; | |
6681 | ||
6682 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
6683 | sizeof(*stats64) / sizeof(u64)); | |
6684 | for (i = 0; i < n; i++) | |
6685 | dst[i] = src[i]; | |
6686 | #endif | |
6687 | } | |
77a1abf5 | 6688 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 6689 | |
eeda3fd6 SH |
6690 | /** |
6691 | * dev_get_stats - get network device statistics | |
6692 | * @dev: device to get statistics from | |
28172739 | 6693 | * @storage: place to store stats |
eeda3fd6 | 6694 | * |
d7753516 BH |
6695 | * Get network statistics from device. Return @storage. |
6696 | * The device driver may provide its own method by setting | |
6697 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
6698 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 6699 | */ |
d7753516 BH |
6700 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
6701 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 6702 | { |
eeda3fd6 SH |
6703 | const struct net_device_ops *ops = dev->netdev_ops; |
6704 | ||
28172739 ED |
6705 | if (ops->ndo_get_stats64) { |
6706 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
6707 | ops->ndo_get_stats64(dev, storage); |
6708 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 6709 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
6710 | } else { |
6711 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 6712 | } |
caf586e5 | 6713 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
015f0688 | 6714 | storage->tx_dropped += atomic_long_read(&dev->tx_dropped); |
28172739 | 6715 | return storage; |
c45d286e | 6716 | } |
eeda3fd6 | 6717 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 6718 | |
24824a09 | 6719 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 6720 | { |
24824a09 | 6721 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 6722 | |
24824a09 ED |
6723 | #ifdef CONFIG_NET_CLS_ACT |
6724 | if (queue) | |
6725 | return queue; | |
6726 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
6727 | if (!queue) | |
6728 | return NULL; | |
6729 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 6730 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
6731 | queue->qdisc_sleeping = &noop_qdisc; |
6732 | rcu_assign_pointer(dev->ingress_queue, queue); | |
6733 | #endif | |
6734 | return queue; | |
bb949fbd DM |
6735 | } |
6736 | ||
2c60db03 ED |
6737 | static const struct ethtool_ops default_ethtool_ops; |
6738 | ||
d07d7507 SG |
6739 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
6740 | const struct ethtool_ops *ops) | |
6741 | { | |
6742 | if (dev->ethtool_ops == &default_ethtool_ops) | |
6743 | dev->ethtool_ops = ops; | |
6744 | } | |
6745 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
6746 | ||
74d332c1 ED |
6747 | void netdev_freemem(struct net_device *dev) |
6748 | { | |
6749 | char *addr = (char *)dev - dev->padded; | |
6750 | ||
4cb28970 | 6751 | kvfree(addr); |
74d332c1 ED |
6752 | } |
6753 | ||
1da177e4 | 6754 | /** |
36909ea4 | 6755 | * alloc_netdev_mqs - allocate network device |
c835a677 TG |
6756 | * @sizeof_priv: size of private data to allocate space for |
6757 | * @name: device name format string | |
6758 | * @name_assign_type: origin of device name | |
6759 | * @setup: callback to initialize device | |
6760 | * @txqs: the number of TX subqueues to allocate | |
6761 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
6762 | * |
6763 | * Allocates a struct net_device with private data area for driver use | |
90e51adf | 6764 | * and performs basic initialization. Also allocates subqueue structs |
36909ea4 | 6765 | * for each queue on the device. |
1da177e4 | 6766 | */ |
36909ea4 | 6767 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 6768 | unsigned char name_assign_type, |
36909ea4 TH |
6769 | void (*setup)(struct net_device *), |
6770 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 6771 | { |
1da177e4 | 6772 | struct net_device *dev; |
7943986c | 6773 | size_t alloc_size; |
1ce8e7b5 | 6774 | struct net_device *p; |
1da177e4 | 6775 | |
b6fe17d6 SH |
6776 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
6777 | ||
36909ea4 | 6778 | if (txqs < 1) { |
7b6cd1ce | 6779 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
6780 | return NULL; |
6781 | } | |
6782 | ||
a953be53 | 6783 | #ifdef CONFIG_SYSFS |
36909ea4 | 6784 | if (rxqs < 1) { |
7b6cd1ce | 6785 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
6786 | return NULL; |
6787 | } | |
6788 | #endif | |
6789 | ||
fd2ea0a7 | 6790 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
6791 | if (sizeof_priv) { |
6792 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 6793 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
6794 | alloc_size += sizeof_priv; |
6795 | } | |
6796 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 6797 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 6798 | |
74d332c1 ED |
6799 | p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6800 | if (!p) | |
6801 | p = vzalloc(alloc_size); | |
62b5942a | 6802 | if (!p) |
1da177e4 | 6803 | return NULL; |
1da177e4 | 6804 | |
1ce8e7b5 | 6805 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 6806 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 6807 | |
29b4433d ED |
6808 | dev->pcpu_refcnt = alloc_percpu(int); |
6809 | if (!dev->pcpu_refcnt) | |
74d332c1 | 6810 | goto free_dev; |
ab9c73cc | 6811 | |
ab9c73cc | 6812 | if (dev_addr_init(dev)) |
29b4433d | 6813 | goto free_pcpu; |
ab9c73cc | 6814 | |
22bedad3 | 6815 | dev_mc_init(dev); |
a748ee24 | 6816 | dev_uc_init(dev); |
ccffad25 | 6817 | |
c346dca1 | 6818 | dev_net_set(dev, &init_net); |
1da177e4 | 6819 | |
8d3bdbd5 | 6820 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 6821 | dev->gso_max_segs = GSO_MAX_SEGS; |
fcbeb976 | 6822 | dev->gso_min_segs = 0; |
8d3bdbd5 | 6823 | |
8d3bdbd5 DM |
6824 | INIT_LIST_HEAD(&dev->napi_list); |
6825 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 6826 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 6827 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
6828 | INIT_LIST_HEAD(&dev->adj_list.upper); |
6829 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
6830 | INIT_LIST_HEAD(&dev->all_adj_list.upper); | |
6831 | INIT_LIST_HEAD(&dev->all_adj_list.lower); | |
7866a621 SN |
6832 | INIT_LIST_HEAD(&dev->ptype_all); |
6833 | INIT_LIST_HEAD(&dev->ptype_specific); | |
02875878 | 6834 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
6835 | setup(dev); |
6836 | ||
36909ea4 TH |
6837 | dev->num_tx_queues = txqs; |
6838 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 6839 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 6840 | goto free_all; |
e8a0464c | 6841 | |
a953be53 | 6842 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
6843 | dev->num_rx_queues = rxqs; |
6844 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 6845 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 6846 | goto free_all; |
df334545 | 6847 | #endif |
0a9627f2 | 6848 | |
1da177e4 | 6849 | strcpy(dev->name, name); |
c835a677 | 6850 | dev->name_assign_type = name_assign_type; |
cbda10fa | 6851 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
6852 | if (!dev->ethtool_ops) |
6853 | dev->ethtool_ops = &default_ethtool_ops; | |
1da177e4 | 6854 | return dev; |
ab9c73cc | 6855 | |
8d3bdbd5 DM |
6856 | free_all: |
6857 | free_netdev(dev); | |
6858 | return NULL; | |
6859 | ||
29b4433d ED |
6860 | free_pcpu: |
6861 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
6862 | free_dev: |
6863 | netdev_freemem(dev); | |
ab9c73cc | 6864 | return NULL; |
1da177e4 | 6865 | } |
36909ea4 | 6866 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
6867 | |
6868 | /** | |
6869 | * free_netdev - free network device | |
6870 | * @dev: device | |
6871 | * | |
4ec93edb YH |
6872 | * This function does the last stage of destroying an allocated device |
6873 | * interface. The reference to the device object is released. | |
1da177e4 LT |
6874 | * If this is the last reference then it will be freed. |
6875 | */ | |
6876 | void free_netdev(struct net_device *dev) | |
6877 | { | |
d565b0a1 HX |
6878 | struct napi_struct *p, *n; |
6879 | ||
60877a32 | 6880 | netif_free_tx_queues(dev); |
a953be53 | 6881 | #ifdef CONFIG_SYSFS |
10595902 | 6882 | kvfree(dev->_rx); |
fe822240 | 6883 | #endif |
e8a0464c | 6884 | |
33d480ce | 6885 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 6886 | |
f001fde5 JP |
6887 | /* Flush device addresses */ |
6888 | dev_addr_flush(dev); | |
6889 | ||
d565b0a1 HX |
6890 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
6891 | netif_napi_del(p); | |
6892 | ||
29b4433d ED |
6893 | free_percpu(dev->pcpu_refcnt); |
6894 | dev->pcpu_refcnt = NULL; | |
6895 | ||
3041a069 | 6896 | /* Compatibility with error handling in drivers */ |
1da177e4 | 6897 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 6898 | netdev_freemem(dev); |
1da177e4 LT |
6899 | return; |
6900 | } | |
6901 | ||
6902 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
6903 | dev->reg_state = NETREG_RELEASED; | |
6904 | ||
43cb76d9 GKH |
6905 | /* will free via device release */ |
6906 | put_device(&dev->dev); | |
1da177e4 | 6907 | } |
d1b19dff | 6908 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 6909 | |
f0db275a SH |
6910 | /** |
6911 | * synchronize_net - Synchronize with packet receive processing | |
6912 | * | |
6913 | * Wait for packets currently being received to be done. | |
6914 | * Does not block later packets from starting. | |
6915 | */ | |
4ec93edb | 6916 | void synchronize_net(void) |
1da177e4 LT |
6917 | { |
6918 | might_sleep(); | |
be3fc413 ED |
6919 | if (rtnl_is_locked()) |
6920 | synchronize_rcu_expedited(); | |
6921 | else | |
6922 | synchronize_rcu(); | |
1da177e4 | 6923 | } |
d1b19dff | 6924 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
6925 | |
6926 | /** | |
44a0873d | 6927 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 6928 | * @dev: device |
44a0873d | 6929 | * @head: list |
6ebfbc06 | 6930 | * |
1da177e4 | 6931 | * This function shuts down a device interface and removes it |
d59b54b1 | 6932 | * from the kernel tables. |
44a0873d | 6933 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
6934 | * |
6935 | * Callers must hold the rtnl semaphore. You may want | |
6936 | * unregister_netdev() instead of this. | |
6937 | */ | |
6938 | ||
44a0873d | 6939 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 6940 | { |
a6620712 HX |
6941 | ASSERT_RTNL(); |
6942 | ||
44a0873d | 6943 | if (head) { |
9fdce099 | 6944 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
6945 | } else { |
6946 | rollback_registered(dev); | |
6947 | /* Finish processing unregister after unlock */ | |
6948 | net_set_todo(dev); | |
6949 | } | |
1da177e4 | 6950 | } |
44a0873d | 6951 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 6952 | |
9b5e383c ED |
6953 | /** |
6954 | * unregister_netdevice_many - unregister many devices | |
6955 | * @head: list of devices | |
87757a91 ED |
6956 | * |
6957 | * Note: As most callers use a stack allocated list_head, | |
6958 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
6959 | */ |
6960 | void unregister_netdevice_many(struct list_head *head) | |
6961 | { | |
6962 | struct net_device *dev; | |
6963 | ||
6964 | if (!list_empty(head)) { | |
6965 | rollback_registered_many(head); | |
6966 | list_for_each_entry(dev, head, unreg_list) | |
6967 | net_set_todo(dev); | |
87757a91 | 6968 | list_del(head); |
9b5e383c ED |
6969 | } |
6970 | } | |
63c8099d | 6971 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 6972 | |
1da177e4 LT |
6973 | /** |
6974 | * unregister_netdev - remove device from the kernel | |
6975 | * @dev: device | |
6976 | * | |
6977 | * This function shuts down a device interface and removes it | |
d59b54b1 | 6978 | * from the kernel tables. |
1da177e4 LT |
6979 | * |
6980 | * This is just a wrapper for unregister_netdevice that takes | |
6981 | * the rtnl semaphore. In general you want to use this and not | |
6982 | * unregister_netdevice. | |
6983 | */ | |
6984 | void unregister_netdev(struct net_device *dev) | |
6985 | { | |
6986 | rtnl_lock(); | |
6987 | unregister_netdevice(dev); | |
6988 | rtnl_unlock(); | |
6989 | } | |
1da177e4 LT |
6990 | EXPORT_SYMBOL(unregister_netdev); |
6991 | ||
ce286d32 EB |
6992 | /** |
6993 | * dev_change_net_namespace - move device to different nethost namespace | |
6994 | * @dev: device | |
6995 | * @net: network namespace | |
6996 | * @pat: If not NULL name pattern to try if the current device name | |
6997 | * is already taken in the destination network namespace. | |
6998 | * | |
6999 | * This function shuts down a device interface and moves it | |
7000 | * to a new network namespace. On success 0 is returned, on | |
7001 | * a failure a netagive errno code is returned. | |
7002 | * | |
7003 | * Callers must hold the rtnl semaphore. | |
7004 | */ | |
7005 | ||
7006 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
7007 | { | |
ce286d32 EB |
7008 | int err; |
7009 | ||
7010 | ASSERT_RTNL(); | |
7011 | ||
7012 | /* Don't allow namespace local devices to be moved. */ | |
7013 | err = -EINVAL; | |
7014 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7015 | goto out; | |
7016 | ||
7017 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
7018 | if (dev->reg_state != NETREG_REGISTERED) |
7019 | goto out; | |
7020 | ||
7021 | /* Get out if there is nothing todo */ | |
7022 | err = 0; | |
878628fb | 7023 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
7024 | goto out; |
7025 | ||
7026 | /* Pick the destination device name, and ensure | |
7027 | * we can use it in the destination network namespace. | |
7028 | */ | |
7029 | err = -EEXIST; | |
d9031024 | 7030 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
7031 | /* We get here if we can't use the current device name */ |
7032 | if (!pat) | |
7033 | goto out; | |
828de4f6 | 7034 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
7035 | goto out; |
7036 | } | |
7037 | ||
7038 | /* | |
7039 | * And now a mini version of register_netdevice unregister_netdevice. | |
7040 | */ | |
7041 | ||
7042 | /* If device is running close it first. */ | |
9b772652 | 7043 | dev_close(dev); |
ce286d32 EB |
7044 | |
7045 | /* And unlink it from device chain */ | |
7046 | err = -ENODEV; | |
7047 | unlist_netdevice(dev); | |
7048 | ||
7049 | synchronize_net(); | |
7050 | ||
7051 | /* Shutdown queueing discipline. */ | |
7052 | dev_shutdown(dev); | |
7053 | ||
7054 | /* Notify protocols, that we are about to destroy | |
7055 | this device. They should clean all the things. | |
3b27e105 DL |
7056 | |
7057 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
7058 | This is wanted because this way 8021q and macvlan know | |
7059 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
7060 | */ |
7061 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
7062 | rcu_barrier(); |
7063 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 7064 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
7065 | |
7066 | /* | |
7067 | * Flush the unicast and multicast chains | |
7068 | */ | |
a748ee24 | 7069 | dev_uc_flush(dev); |
22bedad3 | 7070 | dev_mc_flush(dev); |
ce286d32 | 7071 | |
4e66ae2e SH |
7072 | /* Send a netdev-removed uevent to the old namespace */ |
7073 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 7074 | netdev_adjacent_del_links(dev); |
4e66ae2e | 7075 | |
ce286d32 | 7076 | /* Actually switch the network namespace */ |
c346dca1 | 7077 | dev_net_set(dev, net); |
ce286d32 | 7078 | |
ce286d32 EB |
7079 | /* If there is an ifindex conflict assign a new one */ |
7080 | if (__dev_get_by_index(net, dev->ifindex)) { | |
a54acb3a | 7081 | int iflink = (dev_get_iflink(dev) == dev->ifindex); |
ce286d32 EB |
7082 | dev->ifindex = dev_new_index(net); |
7083 | if (iflink) | |
7084 | dev->iflink = dev->ifindex; | |
7085 | } | |
7086 | ||
4e66ae2e SH |
7087 | /* Send a netdev-add uevent to the new namespace */ |
7088 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 7089 | netdev_adjacent_add_links(dev); |
4e66ae2e | 7090 | |
8b41d188 | 7091 | /* Fixup kobjects */ |
a1b3f594 | 7092 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 7093 | WARN_ON(err); |
ce286d32 EB |
7094 | |
7095 | /* Add the device back in the hashes */ | |
7096 | list_netdevice(dev); | |
7097 | ||
7098 | /* Notify protocols, that a new device appeared. */ | |
7099 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
7100 | ||
d90a909e EB |
7101 | /* |
7102 | * Prevent userspace races by waiting until the network | |
7103 | * device is fully setup before sending notifications. | |
7104 | */ | |
7f294054 | 7105 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 7106 | |
ce286d32 EB |
7107 | synchronize_net(); |
7108 | err = 0; | |
7109 | out: | |
7110 | return err; | |
7111 | } | |
463d0183 | 7112 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 7113 | |
1da177e4 LT |
7114 | static int dev_cpu_callback(struct notifier_block *nfb, |
7115 | unsigned long action, | |
7116 | void *ocpu) | |
7117 | { | |
7118 | struct sk_buff **list_skb; | |
1da177e4 LT |
7119 | struct sk_buff *skb; |
7120 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
7121 | struct softnet_data *sd, *oldsd; | |
7122 | ||
8bb78442 | 7123 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
7124 | return NOTIFY_OK; |
7125 | ||
7126 | local_irq_disable(); | |
7127 | cpu = smp_processor_id(); | |
7128 | sd = &per_cpu(softnet_data, cpu); | |
7129 | oldsd = &per_cpu(softnet_data, oldcpu); | |
7130 | ||
7131 | /* Find end of our completion_queue. */ | |
7132 | list_skb = &sd->completion_queue; | |
7133 | while (*list_skb) | |
7134 | list_skb = &(*list_skb)->next; | |
7135 | /* Append completion queue from offline CPU. */ | |
7136 | *list_skb = oldsd->completion_queue; | |
7137 | oldsd->completion_queue = NULL; | |
7138 | ||
1da177e4 | 7139 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
7140 | if (oldsd->output_queue) { |
7141 | *sd->output_queue_tailp = oldsd->output_queue; | |
7142 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
7143 | oldsd->output_queue = NULL; | |
7144 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
7145 | } | |
ac64da0b ED |
7146 | /* Append NAPI poll list from offline CPU, with one exception : |
7147 | * process_backlog() must be called by cpu owning percpu backlog. | |
7148 | * We properly handle process_queue & input_pkt_queue later. | |
7149 | */ | |
7150 | while (!list_empty(&oldsd->poll_list)) { | |
7151 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
7152 | struct napi_struct, | |
7153 | poll_list); | |
7154 | ||
7155 | list_del_init(&napi->poll_list); | |
7156 | if (napi->poll == process_backlog) | |
7157 | napi->state = 0; | |
7158 | else | |
7159 | ____napi_schedule(sd, napi); | |
264524d5 | 7160 | } |
1da177e4 LT |
7161 | |
7162 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
7163 | local_irq_enable(); | |
7164 | ||
7165 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 7166 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 7167 | netif_rx_ni(skb); |
76cc8b13 | 7168 | input_queue_head_incr(oldsd); |
fec5e652 | 7169 | } |
ac64da0b | 7170 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 7171 | netif_rx_ni(skb); |
76cc8b13 TH |
7172 | input_queue_head_incr(oldsd); |
7173 | } | |
1da177e4 LT |
7174 | |
7175 | return NOTIFY_OK; | |
7176 | } | |
1da177e4 LT |
7177 | |
7178 | ||
7f353bf2 | 7179 | /** |
b63365a2 HX |
7180 | * netdev_increment_features - increment feature set by one |
7181 | * @all: current feature set | |
7182 | * @one: new feature set | |
7183 | * @mask: mask feature set | |
7f353bf2 HX |
7184 | * |
7185 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
7186 | * @one to the master device with current feature set @all. Will not |
7187 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 7188 | */ |
c8f44aff MM |
7189 | netdev_features_t netdev_increment_features(netdev_features_t all, |
7190 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 7191 | { |
1742f183 MM |
7192 | if (mask & NETIF_F_GEN_CSUM) |
7193 | mask |= NETIF_F_ALL_CSUM; | |
7194 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 7195 | |
1742f183 MM |
7196 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
7197 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 7198 | |
1742f183 MM |
7199 | /* If one device supports hw checksumming, set for all. */ |
7200 | if (all & NETIF_F_GEN_CSUM) | |
7201 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
7202 | |
7203 | return all; | |
7204 | } | |
b63365a2 | 7205 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 7206 | |
430f03cd | 7207 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
7208 | { |
7209 | int i; | |
7210 | struct hlist_head *hash; | |
7211 | ||
7212 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
7213 | if (hash != NULL) | |
7214 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
7215 | INIT_HLIST_HEAD(&hash[i]); | |
7216 | ||
7217 | return hash; | |
7218 | } | |
7219 | ||
881d966b | 7220 | /* Initialize per network namespace state */ |
4665079c | 7221 | static int __net_init netdev_init(struct net *net) |
881d966b | 7222 | { |
734b6541 RM |
7223 | if (net != &init_net) |
7224 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 7225 | |
30d97d35 PE |
7226 | net->dev_name_head = netdev_create_hash(); |
7227 | if (net->dev_name_head == NULL) | |
7228 | goto err_name; | |
881d966b | 7229 | |
30d97d35 PE |
7230 | net->dev_index_head = netdev_create_hash(); |
7231 | if (net->dev_index_head == NULL) | |
7232 | goto err_idx; | |
881d966b EB |
7233 | |
7234 | return 0; | |
30d97d35 PE |
7235 | |
7236 | err_idx: | |
7237 | kfree(net->dev_name_head); | |
7238 | err_name: | |
7239 | return -ENOMEM; | |
881d966b EB |
7240 | } |
7241 | ||
f0db275a SH |
7242 | /** |
7243 | * netdev_drivername - network driver for the device | |
7244 | * @dev: network device | |
f0db275a SH |
7245 | * |
7246 | * Determine network driver for device. | |
7247 | */ | |
3019de12 | 7248 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 7249 | { |
cf04a4c7 SH |
7250 | const struct device_driver *driver; |
7251 | const struct device *parent; | |
3019de12 | 7252 | const char *empty = ""; |
6579e57b AV |
7253 | |
7254 | parent = dev->dev.parent; | |
6579e57b | 7255 | if (!parent) |
3019de12 | 7256 | return empty; |
6579e57b AV |
7257 | |
7258 | driver = parent->driver; | |
7259 | if (driver && driver->name) | |
3019de12 DM |
7260 | return driver->name; |
7261 | return empty; | |
6579e57b AV |
7262 | } |
7263 | ||
6ea754eb JP |
7264 | static void __netdev_printk(const char *level, const struct net_device *dev, |
7265 | struct va_format *vaf) | |
256df2f3 | 7266 | { |
b004ff49 | 7267 | if (dev && dev->dev.parent) { |
6ea754eb JP |
7268 | dev_printk_emit(level[1] - '0', |
7269 | dev->dev.parent, | |
7270 | "%s %s %s%s: %pV", | |
7271 | dev_driver_string(dev->dev.parent), | |
7272 | dev_name(dev->dev.parent), | |
7273 | netdev_name(dev), netdev_reg_state(dev), | |
7274 | vaf); | |
b004ff49 | 7275 | } else if (dev) { |
6ea754eb JP |
7276 | printk("%s%s%s: %pV", |
7277 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 7278 | } else { |
6ea754eb | 7279 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 7280 | } |
256df2f3 JP |
7281 | } |
7282 | ||
6ea754eb JP |
7283 | void netdev_printk(const char *level, const struct net_device *dev, |
7284 | const char *format, ...) | |
256df2f3 JP |
7285 | { |
7286 | struct va_format vaf; | |
7287 | va_list args; | |
256df2f3 JP |
7288 | |
7289 | va_start(args, format); | |
7290 | ||
7291 | vaf.fmt = format; | |
7292 | vaf.va = &args; | |
7293 | ||
6ea754eb | 7294 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 7295 | |
256df2f3 | 7296 | va_end(args); |
256df2f3 JP |
7297 | } |
7298 | EXPORT_SYMBOL(netdev_printk); | |
7299 | ||
7300 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 7301 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 7302 | { \ |
256df2f3 JP |
7303 | struct va_format vaf; \ |
7304 | va_list args; \ | |
7305 | \ | |
7306 | va_start(args, fmt); \ | |
7307 | \ | |
7308 | vaf.fmt = fmt; \ | |
7309 | vaf.va = &args; \ | |
7310 | \ | |
6ea754eb | 7311 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 7312 | \ |
256df2f3 | 7313 | va_end(args); \ |
256df2f3 JP |
7314 | } \ |
7315 | EXPORT_SYMBOL(func); | |
7316 | ||
7317 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
7318 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
7319 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
7320 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
7321 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
7322 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
7323 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
7324 | ||
4665079c | 7325 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
7326 | { |
7327 | kfree(net->dev_name_head); | |
7328 | kfree(net->dev_index_head); | |
7329 | } | |
7330 | ||
022cbae6 | 7331 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
7332 | .init = netdev_init, |
7333 | .exit = netdev_exit, | |
7334 | }; | |
7335 | ||
4665079c | 7336 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 7337 | { |
e008b5fc | 7338 | struct net_device *dev, *aux; |
ce286d32 | 7339 | /* |
e008b5fc | 7340 | * Push all migratable network devices back to the |
ce286d32 EB |
7341 | * initial network namespace |
7342 | */ | |
7343 | rtnl_lock(); | |
e008b5fc | 7344 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 7345 | int err; |
aca51397 | 7346 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
7347 | |
7348 | /* Ignore unmoveable devices (i.e. loopback) */ | |
7349 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7350 | continue; | |
7351 | ||
e008b5fc EB |
7352 | /* Leave virtual devices for the generic cleanup */ |
7353 | if (dev->rtnl_link_ops) | |
7354 | continue; | |
d0c082ce | 7355 | |
25985edc | 7356 | /* Push remaining network devices to init_net */ |
aca51397 PE |
7357 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
7358 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 7359 | if (err) { |
7b6cd1ce JP |
7360 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
7361 | __func__, dev->name, err); | |
aca51397 | 7362 | BUG(); |
ce286d32 EB |
7363 | } |
7364 | } | |
7365 | rtnl_unlock(); | |
7366 | } | |
7367 | ||
50624c93 EB |
7368 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
7369 | { | |
7370 | /* Return with the rtnl_lock held when there are no network | |
7371 | * devices unregistering in any network namespace in net_list. | |
7372 | */ | |
7373 | struct net *net; | |
7374 | bool unregistering; | |
ff960a73 | 7375 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 7376 | |
ff960a73 | 7377 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 7378 | for (;;) { |
50624c93 EB |
7379 | unregistering = false; |
7380 | rtnl_lock(); | |
7381 | list_for_each_entry(net, net_list, exit_list) { | |
7382 | if (net->dev_unreg_count > 0) { | |
7383 | unregistering = true; | |
7384 | break; | |
7385 | } | |
7386 | } | |
7387 | if (!unregistering) | |
7388 | break; | |
7389 | __rtnl_unlock(); | |
ff960a73 PZ |
7390 | |
7391 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 7392 | } |
ff960a73 | 7393 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
7394 | } |
7395 | ||
04dc7f6b EB |
7396 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
7397 | { | |
7398 | /* At exit all network devices most be removed from a network | |
b595076a | 7399 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
7400 | * Do this across as many network namespaces as possible to |
7401 | * improve batching efficiency. | |
7402 | */ | |
7403 | struct net_device *dev; | |
7404 | struct net *net; | |
7405 | LIST_HEAD(dev_kill_list); | |
7406 | ||
50624c93 EB |
7407 | /* To prevent network device cleanup code from dereferencing |
7408 | * loopback devices or network devices that have been freed | |
7409 | * wait here for all pending unregistrations to complete, | |
7410 | * before unregistring the loopback device and allowing the | |
7411 | * network namespace be freed. | |
7412 | * | |
7413 | * The netdev todo list containing all network devices | |
7414 | * unregistrations that happen in default_device_exit_batch | |
7415 | * will run in the rtnl_unlock() at the end of | |
7416 | * default_device_exit_batch. | |
7417 | */ | |
7418 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
7419 | list_for_each_entry(net, net_list, exit_list) { |
7420 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 7421 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
7422 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
7423 | else | |
7424 | unregister_netdevice_queue(dev, &dev_kill_list); | |
7425 | } | |
7426 | } | |
7427 | unregister_netdevice_many(&dev_kill_list); | |
7428 | rtnl_unlock(); | |
7429 | } | |
7430 | ||
022cbae6 | 7431 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 7432 | .exit = default_device_exit, |
04dc7f6b | 7433 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
7434 | }; |
7435 | ||
1da177e4 LT |
7436 | /* |
7437 | * Initialize the DEV module. At boot time this walks the device list and | |
7438 | * unhooks any devices that fail to initialise (normally hardware not | |
7439 | * present) and leaves us with a valid list of present and active devices. | |
7440 | * | |
7441 | */ | |
7442 | ||
7443 | /* | |
7444 | * This is called single threaded during boot, so no need | |
7445 | * to take the rtnl semaphore. | |
7446 | */ | |
7447 | static int __init net_dev_init(void) | |
7448 | { | |
7449 | int i, rc = -ENOMEM; | |
7450 | ||
7451 | BUG_ON(!dev_boot_phase); | |
7452 | ||
1da177e4 LT |
7453 | if (dev_proc_init()) |
7454 | goto out; | |
7455 | ||
8b41d188 | 7456 | if (netdev_kobject_init()) |
1da177e4 LT |
7457 | goto out; |
7458 | ||
7459 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 7460 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
7461 | INIT_LIST_HEAD(&ptype_base[i]); |
7462 | ||
62532da9 VY |
7463 | INIT_LIST_HEAD(&offload_base); |
7464 | ||
881d966b EB |
7465 | if (register_pernet_subsys(&netdev_net_ops)) |
7466 | goto out; | |
1da177e4 LT |
7467 | |
7468 | /* | |
7469 | * Initialise the packet receive queues. | |
7470 | */ | |
7471 | ||
6f912042 | 7472 | for_each_possible_cpu(i) { |
e36fa2f7 | 7473 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 7474 | |
e36fa2f7 | 7475 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 7476 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 7477 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 7478 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 7479 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
7480 | sd->csd.func = rps_trigger_softirq; |
7481 | sd->csd.info = sd; | |
e36fa2f7 | 7482 | sd->cpu = i; |
1e94d72f | 7483 | #endif |
0a9627f2 | 7484 | |
e36fa2f7 ED |
7485 | sd->backlog.poll = process_backlog; |
7486 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
7487 | } |
7488 | ||
1da177e4 LT |
7489 | dev_boot_phase = 0; |
7490 | ||
505d4f73 EB |
7491 | /* The loopback device is special if any other network devices |
7492 | * is present in a network namespace the loopback device must | |
7493 | * be present. Since we now dynamically allocate and free the | |
7494 | * loopback device ensure this invariant is maintained by | |
7495 | * keeping the loopback device as the first device on the | |
7496 | * list of network devices. Ensuring the loopback devices | |
7497 | * is the first device that appears and the last network device | |
7498 | * that disappears. | |
7499 | */ | |
7500 | if (register_pernet_device(&loopback_net_ops)) | |
7501 | goto out; | |
7502 | ||
7503 | if (register_pernet_device(&default_device_ops)) | |
7504 | goto out; | |
7505 | ||
962cf36c CM |
7506 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
7507 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
7508 | |
7509 | hotcpu_notifier(dev_cpu_callback, 0); | |
7510 | dst_init(); | |
1da177e4 LT |
7511 | rc = 0; |
7512 | out: | |
7513 | return rc; | |
7514 | } | |
7515 | ||
7516 | subsys_initcall(net_dev_init); |