]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
722c9a0c | 2 | * NET3 Protocol independent device support routines. |
1da177e4 LT |
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 | |
722c9a0c | 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 | |
722c9a0c | 24 | * to 2 if register_netdev gets called |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
1da177e4 LT |
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. | |
722c9a0c | 39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. |
1da177e4 LT |
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. | |
722c9a0c | 49 | * Alan Cox : Fixed nasty side effect of device close |
1da177e4 LT |
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 | |
722c9a0c | 70 | * indefinitely on dev->refcnt |
71 | * J Hadi Salim : - Backlog queue sampling | |
1da177e4 LT |
72 | * - netif_rx() feedback |
73 | */ | |
74 | ||
7c0f6ba6 | 75 | #include <linux/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> |
f1083048 | 84 | #include <linux/sched/mm.h> |
4a3e2f71 | 85 | #include <linux/mutex.h> |
1da177e4 LT |
86 | #include <linux/string.h> |
87 | #include <linux/mm.h> | |
88 | #include <linux/socket.h> | |
89 | #include <linux/sockios.h> | |
90 | #include <linux/errno.h> | |
91 | #include <linux/interrupt.h> | |
92 | #include <linux/if_ether.h> | |
93 | #include <linux/netdevice.h> | |
94 | #include <linux/etherdevice.h> | |
0187bdfb | 95 | #include <linux/ethtool.h> |
1da177e4 LT |
96 | #include <linux/notifier.h> |
97 | #include <linux/skbuff.h> | |
a7862b45 | 98 | #include <linux/bpf.h> |
b5cdae32 | 99 | #include <linux/bpf_trace.h> |
457c4cbc | 100 | #include <net/net_namespace.h> |
1da177e4 | 101 | #include <net/sock.h> |
02d62e86 | 102 | #include <net/busy_poll.h> |
1da177e4 | 103 | #include <linux/rtnetlink.h> |
1da177e4 | 104 | #include <linux/stat.h> |
1da177e4 | 105 | #include <net/dst.h> |
fc4099f1 | 106 | #include <net/dst_metadata.h> |
1da177e4 | 107 | #include <net/pkt_sched.h> |
87d83093 | 108 | #include <net/pkt_cls.h> |
1da177e4 | 109 | #include <net/checksum.h> |
44540960 | 110 | #include <net/xfrm.h> |
1da177e4 LT |
111 | #include <linux/highmem.h> |
112 | #include <linux/init.h> | |
1da177e4 | 113 | #include <linux/module.h> |
1da177e4 LT |
114 | #include <linux/netpoll.h> |
115 | #include <linux/rcupdate.h> | |
116 | #include <linux/delay.h> | |
1da177e4 | 117 | #include <net/iw_handler.h> |
1da177e4 | 118 | #include <asm/current.h> |
5bdb9886 | 119 | #include <linux/audit.h> |
db217334 | 120 | #include <linux/dmaengine.h> |
f6a78bfc | 121 | #include <linux/err.h> |
c7fa9d18 | 122 | #include <linux/ctype.h> |
723e98b7 | 123 | #include <linux/if_arp.h> |
6de329e2 | 124 | #include <linux/if_vlan.h> |
8f0f2223 | 125 | #include <linux/ip.h> |
ad55dcaf | 126 | #include <net/ip.h> |
25cd9ba0 | 127 | #include <net/mpls.h> |
8f0f2223 DM |
128 | #include <linux/ipv6.h> |
129 | #include <linux/in.h> | |
b6b2fed1 DM |
130 | #include <linux/jhash.h> |
131 | #include <linux/random.h> | |
9cbc1cb8 | 132 | #include <trace/events/napi.h> |
cf66ba58 | 133 | #include <trace/events/net.h> |
07dc22e7 | 134 | #include <trace/events/skb.h> |
5acbbd42 | 135 | #include <linux/pci.h> |
caeda9b9 | 136 | #include <linux/inetdevice.h> |
c445477d | 137 | #include <linux/cpu_rmap.h> |
c5905afb | 138 | #include <linux/static_key.h> |
af12fa6e | 139 | #include <linux/hashtable.h> |
60877a32 | 140 | #include <linux/vmalloc.h> |
529d0489 | 141 | #include <linux/if_macvlan.h> |
e7fd2885 | 142 | #include <linux/errqueue.h> |
3b47d303 | 143 | #include <linux/hrtimer.h> |
e687ad60 | 144 | #include <linux/netfilter_ingress.h> |
40e4e713 | 145 | #include <linux/crash_dump.h> |
b72b5bf6 | 146 | #include <linux/sctp.h> |
1da177e4 | 147 | |
342709ef PE |
148 | #include "net-sysfs.h" |
149 | ||
d565b0a1 HX |
150 | /* Instead of increasing this, you should create a hash table. */ |
151 | #define MAX_GRO_SKBS 8 | |
152 | ||
5d38a079 HX |
153 | /* This should be increased if a protocol with a bigger head is added. */ |
154 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
155 | ||
1da177e4 | 156 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 157 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
158 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
159 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 160 | static struct list_head offload_base __read_mostly; |
1da177e4 | 161 | |
ae78dbfa | 162 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
163 | static int call_netdevice_notifiers_info(unsigned long val, |
164 | struct net_device *dev, | |
165 | struct netdev_notifier_info *info); | |
90b602f8 | 166 | static struct napi_struct *napi_by_id(unsigned int napi_id); |
ae78dbfa | 167 | |
1da177e4 | 168 | /* |
7562f876 | 169 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
170 | * semaphore. |
171 | * | |
c6d14c84 | 172 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
173 | * |
174 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 175 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
176 | * actual updates. This allows pure readers to access the list even |
177 | * while a writer is preparing to update it. | |
178 | * | |
179 | * To put it another way, dev_base_lock is held for writing only to | |
180 | * protect against pure readers; the rtnl semaphore provides the | |
181 | * protection against other writers. | |
182 | * | |
183 | * See, for example usages, register_netdevice() and | |
184 | * unregister_netdevice(), which must be called with the rtnl | |
185 | * semaphore held. | |
186 | */ | |
1da177e4 | 187 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
188 | EXPORT_SYMBOL(dev_base_lock); |
189 | ||
af12fa6e ET |
190 | /* protects napi_hash addition/deletion and napi_gen_id */ |
191 | static DEFINE_SPINLOCK(napi_hash_lock); | |
192 | ||
52bd2d62 | 193 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 194 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 195 | |
18afa4b0 | 196 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 197 | |
4e985ada TG |
198 | static inline void dev_base_seq_inc(struct net *net) |
199 | { | |
643aa9cb | 200 | while (++net->dev_base_seq == 0) |
201 | ; | |
4e985ada TG |
202 | } |
203 | ||
881d966b | 204 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 205 | { |
8387ff25 | 206 | unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); |
95c96174 | 207 | |
08e9897d | 208 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
209 | } |
210 | ||
881d966b | 211 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 212 | { |
7c28bd0b | 213 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
214 | } |
215 | ||
e36fa2f7 | 216 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
217 | { |
218 | #ifdef CONFIG_RPS | |
e36fa2f7 | 219 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
220 | #endif |
221 | } | |
222 | ||
e36fa2f7 | 223 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
224 | { |
225 | #ifdef CONFIG_RPS | |
e36fa2f7 | 226 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
227 | #endif |
228 | } | |
229 | ||
ce286d32 | 230 | /* Device list insertion */ |
53759be9 | 231 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 232 | { |
c346dca1 | 233 | struct net *net = dev_net(dev); |
ce286d32 EB |
234 | |
235 | ASSERT_RTNL(); | |
236 | ||
237 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 238 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 239 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
240 | hlist_add_head_rcu(&dev->index_hlist, |
241 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 242 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
243 | |
244 | dev_base_seq_inc(net); | |
ce286d32 EB |
245 | } |
246 | ||
fb699dfd ED |
247 | /* Device list removal |
248 | * caller must respect a RCU grace period before freeing/reusing dev | |
249 | */ | |
ce286d32 EB |
250 | static void unlist_netdevice(struct net_device *dev) |
251 | { | |
252 | ASSERT_RTNL(); | |
253 | ||
254 | /* Unlink dev from the device chain */ | |
255 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 256 | list_del_rcu(&dev->dev_list); |
72c9528b | 257 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 258 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 259 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
260 | |
261 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
262 | } |
263 | ||
1da177e4 LT |
264 | /* |
265 | * Our notifier list | |
266 | */ | |
267 | ||
f07d5b94 | 268 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
269 | |
270 | /* | |
271 | * Device drivers call our routines to queue packets here. We empty the | |
272 | * queue in the local softnet handler. | |
273 | */ | |
bea3348e | 274 | |
9958da05 | 275 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 276 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 277 | |
cf508b12 | 278 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 279 | /* |
c773e847 | 280 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
281 | * according to dev->type |
282 | */ | |
643aa9cb | 283 | static const unsigned short netdev_lock_type[] = { |
284 | ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
723e98b7 JP |
285 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, |
286 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
287 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
288 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
289 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
290 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
291 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
292 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
293 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
294 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
295 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
296 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
297 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
298 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 299 | |
643aa9cb | 300 | static const char *const netdev_lock_name[] = { |
301 | "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
302 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
303 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
304 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
305 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
306 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
307 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
308 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
309 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
310 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
311 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
312 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
313 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", | |
314 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
315 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
316 | |
317 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 318 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
319 | |
320 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
321 | { | |
322 | int i; | |
323 | ||
324 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
325 | if (netdev_lock_type[i] == dev_type) | |
326 | return i; | |
327 | /* the last key is used by default */ | |
328 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
329 | } | |
330 | ||
cf508b12 DM |
331 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
332 | unsigned short dev_type) | |
723e98b7 JP |
333 | { |
334 | int i; | |
335 | ||
336 | i = netdev_lock_pos(dev_type); | |
337 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
338 | netdev_lock_name[i]); | |
339 | } | |
cf508b12 DM |
340 | |
341 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
342 | { | |
343 | int i; | |
344 | ||
345 | i = netdev_lock_pos(dev->type); | |
346 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
347 | &netdev_addr_lock_key[i], | |
348 | netdev_lock_name[i]); | |
349 | } | |
723e98b7 | 350 | #else |
cf508b12 DM |
351 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
352 | unsigned short dev_type) | |
353 | { | |
354 | } | |
355 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
356 | { |
357 | } | |
358 | #endif | |
1da177e4 LT |
359 | |
360 | /******************************************************************************* | |
eb13da1a | 361 | * |
362 | * Protocol management and registration routines | |
363 | * | |
364 | *******************************************************************************/ | |
1da177e4 | 365 | |
1da177e4 | 366 | |
1da177e4 LT |
367 | /* |
368 | * Add a protocol ID to the list. Now that the input handler is | |
369 | * smarter we can dispense with all the messy stuff that used to be | |
370 | * here. | |
371 | * | |
372 | * BEWARE!!! Protocol handlers, mangling input packets, | |
373 | * MUST BE last in hash buckets and checking protocol handlers | |
374 | * MUST start from promiscuous ptype_all chain in net_bh. | |
375 | * It is true now, do not change it. | |
376 | * Explanation follows: if protocol handler, mangling packet, will | |
377 | * be the first on list, it is not able to sense, that packet | |
378 | * is cloned and should be copied-on-write, so that it will | |
379 | * change it and subsequent readers will get broken packet. | |
380 | * --ANK (980803) | |
381 | */ | |
382 | ||
c07b68e8 ED |
383 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
384 | { | |
385 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 386 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 387 | else |
7866a621 SN |
388 | return pt->dev ? &pt->dev->ptype_specific : |
389 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
390 | } |
391 | ||
1da177e4 LT |
392 | /** |
393 | * dev_add_pack - add packet handler | |
394 | * @pt: packet type declaration | |
395 | * | |
396 | * Add a protocol handler to the networking stack. The passed &packet_type | |
397 | * is linked into kernel lists and may not be freed until it has been | |
398 | * removed from the kernel lists. | |
399 | * | |
4ec93edb | 400 | * This call does not sleep therefore it can not |
1da177e4 LT |
401 | * guarantee all CPU's that are in middle of receiving packets |
402 | * will see the new packet type (until the next received packet). | |
403 | */ | |
404 | ||
405 | void dev_add_pack(struct packet_type *pt) | |
406 | { | |
c07b68e8 | 407 | struct list_head *head = ptype_head(pt); |
1da177e4 | 408 | |
c07b68e8 ED |
409 | spin_lock(&ptype_lock); |
410 | list_add_rcu(&pt->list, head); | |
411 | spin_unlock(&ptype_lock); | |
1da177e4 | 412 | } |
d1b19dff | 413 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 414 | |
1da177e4 LT |
415 | /** |
416 | * __dev_remove_pack - remove packet handler | |
417 | * @pt: packet type declaration | |
418 | * | |
419 | * Remove a protocol handler that was previously added to the kernel | |
420 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
421 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 422 | * returns. |
1da177e4 LT |
423 | * |
424 | * The packet type might still be in use by receivers | |
425 | * and must not be freed until after all the CPU's have gone | |
426 | * through a quiescent state. | |
427 | */ | |
428 | void __dev_remove_pack(struct packet_type *pt) | |
429 | { | |
c07b68e8 | 430 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
431 | struct packet_type *pt1; |
432 | ||
c07b68e8 | 433 | spin_lock(&ptype_lock); |
1da177e4 LT |
434 | |
435 | list_for_each_entry(pt1, head, list) { | |
436 | if (pt == pt1) { | |
437 | list_del_rcu(&pt->list); | |
438 | goto out; | |
439 | } | |
440 | } | |
441 | ||
7b6cd1ce | 442 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 443 | out: |
c07b68e8 | 444 | spin_unlock(&ptype_lock); |
1da177e4 | 445 | } |
d1b19dff ED |
446 | EXPORT_SYMBOL(__dev_remove_pack); |
447 | ||
1da177e4 LT |
448 | /** |
449 | * dev_remove_pack - remove packet handler | |
450 | * @pt: packet type declaration | |
451 | * | |
452 | * Remove a protocol handler that was previously added to the kernel | |
453 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
454 | * from the kernel lists and can be freed or reused once this function | |
455 | * returns. | |
456 | * | |
457 | * This call sleeps to guarantee that no CPU is looking at the packet | |
458 | * type after return. | |
459 | */ | |
460 | void dev_remove_pack(struct packet_type *pt) | |
461 | { | |
462 | __dev_remove_pack(pt); | |
4ec93edb | 463 | |
1da177e4 LT |
464 | synchronize_net(); |
465 | } | |
d1b19dff | 466 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 467 | |
62532da9 VY |
468 | |
469 | /** | |
470 | * dev_add_offload - register offload handlers | |
471 | * @po: protocol offload declaration | |
472 | * | |
473 | * Add protocol offload handlers to the networking stack. The passed | |
474 | * &proto_offload is linked into kernel lists and may not be freed until | |
475 | * it has been removed from the kernel lists. | |
476 | * | |
477 | * This call does not sleep therefore it can not | |
478 | * guarantee all CPU's that are in middle of receiving packets | |
479 | * will see the new offload handlers (until the next received packet). | |
480 | */ | |
481 | void dev_add_offload(struct packet_offload *po) | |
482 | { | |
bdef7de4 | 483 | struct packet_offload *elem; |
62532da9 VY |
484 | |
485 | spin_lock(&offload_lock); | |
bdef7de4 DM |
486 | list_for_each_entry(elem, &offload_base, list) { |
487 | if (po->priority < elem->priority) | |
488 | break; | |
489 | } | |
490 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
491 | spin_unlock(&offload_lock); |
492 | } | |
493 | EXPORT_SYMBOL(dev_add_offload); | |
494 | ||
495 | /** | |
496 | * __dev_remove_offload - remove offload handler | |
497 | * @po: packet offload declaration | |
498 | * | |
499 | * Remove a protocol offload handler that was previously added to the | |
500 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
501 | * is removed from the kernel lists and can be freed or reused once this | |
502 | * function returns. | |
503 | * | |
504 | * The packet type might still be in use by receivers | |
505 | * and must not be freed until after all the CPU's have gone | |
506 | * through a quiescent state. | |
507 | */ | |
1d143d9f | 508 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
509 | { |
510 | struct list_head *head = &offload_base; | |
511 | struct packet_offload *po1; | |
512 | ||
c53aa505 | 513 | spin_lock(&offload_lock); |
62532da9 VY |
514 | |
515 | list_for_each_entry(po1, head, list) { | |
516 | if (po == po1) { | |
517 | list_del_rcu(&po->list); | |
518 | goto out; | |
519 | } | |
520 | } | |
521 | ||
522 | pr_warn("dev_remove_offload: %p not found\n", po); | |
523 | out: | |
c53aa505 | 524 | spin_unlock(&offload_lock); |
62532da9 | 525 | } |
62532da9 VY |
526 | |
527 | /** | |
528 | * dev_remove_offload - remove packet offload handler | |
529 | * @po: packet offload declaration | |
530 | * | |
531 | * Remove a packet offload handler that was previously added to the kernel | |
532 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
533 | * removed from the kernel lists and can be freed or reused once this | |
534 | * function returns. | |
535 | * | |
536 | * This call sleeps to guarantee that no CPU is looking at the packet | |
537 | * type after return. | |
538 | */ | |
539 | void dev_remove_offload(struct packet_offload *po) | |
540 | { | |
541 | __dev_remove_offload(po); | |
542 | ||
543 | synchronize_net(); | |
544 | } | |
545 | EXPORT_SYMBOL(dev_remove_offload); | |
546 | ||
1da177e4 | 547 | /****************************************************************************** |
eb13da1a | 548 | * |
549 | * Device Boot-time Settings Routines | |
550 | * | |
551 | ******************************************************************************/ | |
1da177e4 LT |
552 | |
553 | /* Boot time configuration table */ | |
554 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
555 | ||
556 | /** | |
557 | * netdev_boot_setup_add - add new setup entry | |
558 | * @name: name of the device | |
559 | * @map: configured settings for the device | |
560 | * | |
561 | * Adds new setup entry to the dev_boot_setup list. The function | |
562 | * returns 0 on error and 1 on success. This is a generic routine to | |
563 | * all netdevices. | |
564 | */ | |
565 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
566 | { | |
567 | struct netdev_boot_setup *s; | |
568 | int i; | |
569 | ||
570 | s = dev_boot_setup; | |
571 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
572 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
573 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 574 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
575 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
576 | break; | |
577 | } | |
578 | } | |
579 | ||
580 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
581 | } | |
582 | ||
583 | /** | |
722c9a0c | 584 | * netdev_boot_setup_check - check boot time settings |
585 | * @dev: the netdevice | |
1da177e4 | 586 | * |
722c9a0c | 587 | * Check boot time settings for the device. |
588 | * The found settings are set for the device to be used | |
589 | * later in the device probing. | |
590 | * Returns 0 if no settings found, 1 if they are. | |
1da177e4 LT |
591 | */ |
592 | int netdev_boot_setup_check(struct net_device *dev) | |
593 | { | |
594 | struct netdev_boot_setup *s = dev_boot_setup; | |
595 | int i; | |
596 | ||
597 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
598 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 599 | !strcmp(dev->name, s[i].name)) { |
722c9a0c | 600 | dev->irq = s[i].map.irq; |
601 | dev->base_addr = s[i].map.base_addr; | |
602 | dev->mem_start = s[i].map.mem_start; | |
603 | dev->mem_end = s[i].map.mem_end; | |
1da177e4 LT |
604 | return 1; |
605 | } | |
606 | } | |
607 | return 0; | |
608 | } | |
d1b19dff | 609 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
610 | |
611 | ||
612 | /** | |
722c9a0c | 613 | * netdev_boot_base - get address from boot time settings |
614 | * @prefix: prefix for network device | |
615 | * @unit: id for network device | |
616 | * | |
617 | * Check boot time settings for the base address of device. | |
618 | * The found settings are set for the device to be used | |
619 | * later in the device probing. | |
620 | * Returns 0 if no settings found. | |
1da177e4 LT |
621 | */ |
622 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
623 | { | |
624 | const struct netdev_boot_setup *s = dev_boot_setup; | |
625 | char name[IFNAMSIZ]; | |
626 | int i; | |
627 | ||
628 | sprintf(name, "%s%d", prefix, unit); | |
629 | ||
630 | /* | |
631 | * If device already registered then return base of 1 | |
632 | * to indicate not to probe for this interface | |
633 | */ | |
881d966b | 634 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
635 | return 1; |
636 | ||
637 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
638 | if (!strcmp(name, s[i].name)) | |
639 | return s[i].map.base_addr; | |
640 | return 0; | |
641 | } | |
642 | ||
643 | /* | |
644 | * Saves at boot time configured settings for any netdevice. | |
645 | */ | |
646 | int __init netdev_boot_setup(char *str) | |
647 | { | |
648 | int ints[5]; | |
649 | struct ifmap map; | |
650 | ||
651 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
652 | if (!str || !*str) | |
653 | return 0; | |
654 | ||
655 | /* Save settings */ | |
656 | memset(&map, 0, sizeof(map)); | |
657 | if (ints[0] > 0) | |
658 | map.irq = ints[1]; | |
659 | if (ints[0] > 1) | |
660 | map.base_addr = ints[2]; | |
661 | if (ints[0] > 2) | |
662 | map.mem_start = ints[3]; | |
663 | if (ints[0] > 3) | |
664 | map.mem_end = ints[4]; | |
665 | ||
666 | /* Add new entry to the list */ | |
667 | return netdev_boot_setup_add(str, &map); | |
668 | } | |
669 | ||
670 | __setup("netdev=", netdev_boot_setup); | |
671 | ||
672 | /******************************************************************************* | |
eb13da1a | 673 | * |
674 | * Device Interface Subroutines | |
675 | * | |
676 | *******************************************************************************/ | |
1da177e4 | 677 | |
a54acb3a ND |
678 | /** |
679 | * dev_get_iflink - get 'iflink' value of a interface | |
680 | * @dev: targeted interface | |
681 | * | |
682 | * Indicates the ifindex the interface is linked to. | |
683 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
684 | */ | |
685 | ||
686 | int dev_get_iflink(const struct net_device *dev) | |
687 | { | |
688 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
689 | return dev->netdev_ops->ndo_get_iflink(dev); | |
690 | ||
7a66bbc9 | 691 | return dev->ifindex; |
a54acb3a ND |
692 | } |
693 | EXPORT_SYMBOL(dev_get_iflink); | |
694 | ||
fc4099f1 PS |
695 | /** |
696 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
697 | * @dev: targeted interface | |
698 | * @skb: The packet. | |
699 | * | |
700 | * For better visibility of tunnel traffic OVS needs to retrieve | |
701 | * egress tunnel information for a packet. Following API allows | |
702 | * user to get this info. | |
703 | */ | |
704 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
705 | { | |
706 | struct ip_tunnel_info *info; | |
707 | ||
708 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
709 | return -EINVAL; | |
710 | ||
711 | info = skb_tunnel_info_unclone(skb); | |
712 | if (!info) | |
713 | return -ENOMEM; | |
714 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
715 | return -EINVAL; | |
716 | ||
717 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
718 | } | |
719 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
720 | ||
1da177e4 LT |
721 | /** |
722 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 723 | * @net: the applicable net namespace |
1da177e4 LT |
724 | * @name: name to find |
725 | * | |
726 | * Find an interface by name. Must be called under RTNL semaphore | |
727 | * or @dev_base_lock. If the name is found a pointer to the device | |
728 | * is returned. If the name is not found then %NULL is returned. The | |
729 | * reference counters are not incremented so the caller must be | |
730 | * careful with locks. | |
731 | */ | |
732 | ||
881d966b | 733 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 734 | { |
0bd8d536 ED |
735 | struct net_device *dev; |
736 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 737 | |
b67bfe0d | 738 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
739 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
740 | return dev; | |
0bd8d536 | 741 | |
1da177e4 LT |
742 | return NULL; |
743 | } | |
d1b19dff | 744 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 745 | |
72c9528b | 746 | /** |
722c9a0c | 747 | * dev_get_by_name_rcu - find a device by its name |
748 | * @net: the applicable net namespace | |
749 | * @name: name to find | |
750 | * | |
751 | * Find an interface by name. | |
752 | * If the name is found a pointer to the device is returned. | |
753 | * If the name is not found then %NULL is returned. | |
754 | * The reference counters are not incremented so the caller must be | |
755 | * careful with locks. The caller must hold RCU lock. | |
72c9528b ED |
756 | */ |
757 | ||
758 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
759 | { | |
72c9528b ED |
760 | struct net_device *dev; |
761 | struct hlist_head *head = dev_name_hash(net, name); | |
762 | ||
b67bfe0d | 763 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
764 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
765 | return dev; | |
766 | ||
767 | return NULL; | |
768 | } | |
769 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
770 | ||
1da177e4 LT |
771 | /** |
772 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 773 | * @net: the applicable net namespace |
1da177e4 LT |
774 | * @name: name to find |
775 | * | |
776 | * Find an interface by name. This can be called from any | |
777 | * context and does its own locking. The returned handle has | |
778 | * the usage count incremented and the caller must use dev_put() to | |
779 | * release it when it is no longer needed. %NULL is returned if no | |
780 | * matching device is found. | |
781 | */ | |
782 | ||
881d966b | 783 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
784 | { |
785 | struct net_device *dev; | |
786 | ||
72c9528b ED |
787 | rcu_read_lock(); |
788 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
789 | if (dev) |
790 | dev_hold(dev); | |
72c9528b | 791 | rcu_read_unlock(); |
1da177e4 LT |
792 | return dev; |
793 | } | |
d1b19dff | 794 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
795 | |
796 | /** | |
797 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 798 | * @net: the applicable net namespace |
1da177e4 LT |
799 | * @ifindex: index of device |
800 | * | |
801 | * Search for an interface by index. Returns %NULL if the device | |
802 | * is not found or a pointer to the device. The device has not | |
803 | * had its reference counter increased so the caller must be careful | |
804 | * about locking. The caller must hold either the RTNL semaphore | |
805 | * or @dev_base_lock. | |
806 | */ | |
807 | ||
881d966b | 808 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 809 | { |
0bd8d536 ED |
810 | struct net_device *dev; |
811 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 812 | |
b67bfe0d | 813 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
814 | if (dev->ifindex == ifindex) |
815 | return dev; | |
0bd8d536 | 816 | |
1da177e4 LT |
817 | return NULL; |
818 | } | |
d1b19dff | 819 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 820 | |
fb699dfd ED |
821 | /** |
822 | * dev_get_by_index_rcu - find a device by its ifindex | |
823 | * @net: the applicable net namespace | |
824 | * @ifindex: index of device | |
825 | * | |
826 | * Search for an interface by index. Returns %NULL if the device | |
827 | * is not found or a pointer to the device. The device has not | |
828 | * had its reference counter increased so the caller must be careful | |
829 | * about locking. The caller must hold RCU lock. | |
830 | */ | |
831 | ||
832 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
833 | { | |
fb699dfd ED |
834 | struct net_device *dev; |
835 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
836 | ||
b67bfe0d | 837 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
838 | if (dev->ifindex == ifindex) |
839 | return dev; | |
840 | ||
841 | return NULL; | |
842 | } | |
843 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
844 | ||
1da177e4 LT |
845 | |
846 | /** | |
847 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 848 | * @net: the applicable net namespace |
1da177e4 LT |
849 | * @ifindex: index of device |
850 | * | |
851 | * Search for an interface by index. Returns NULL if the device | |
852 | * is not found or a pointer to the device. The device returned has | |
853 | * had a reference added and the pointer is safe until the user calls | |
854 | * dev_put to indicate they have finished with it. | |
855 | */ | |
856 | ||
881d966b | 857 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
858 | { |
859 | struct net_device *dev; | |
860 | ||
fb699dfd ED |
861 | rcu_read_lock(); |
862 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
863 | if (dev) |
864 | dev_hold(dev); | |
fb699dfd | 865 | rcu_read_unlock(); |
1da177e4 LT |
866 | return dev; |
867 | } | |
d1b19dff | 868 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 869 | |
90b602f8 ML |
870 | /** |
871 | * dev_get_by_napi_id - find a device by napi_id | |
872 | * @napi_id: ID of the NAPI struct | |
873 | * | |
874 | * Search for an interface by NAPI ID. Returns %NULL if the device | |
875 | * is not found or a pointer to the device. The device has not had | |
876 | * its reference counter increased so the caller must be careful | |
877 | * about locking. The caller must hold RCU lock. | |
878 | */ | |
879 | ||
880 | struct net_device *dev_get_by_napi_id(unsigned int napi_id) | |
881 | { | |
882 | struct napi_struct *napi; | |
883 | ||
884 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
885 | ||
886 | if (napi_id < MIN_NAPI_ID) | |
887 | return NULL; | |
888 | ||
889 | napi = napi_by_id(napi_id); | |
890 | ||
891 | return napi ? napi->dev : NULL; | |
892 | } | |
893 | EXPORT_SYMBOL(dev_get_by_napi_id); | |
894 | ||
5dbe7c17 NS |
895 | /** |
896 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
897 | * @net: network namespace | |
898 | * @name: a pointer to the buffer where the name will be stored. | |
899 | * @ifindex: the ifindex of the interface to get the name from. | |
900 | * | |
901 | * The use of raw_seqcount_begin() and cond_resched() before | |
902 | * retrying is required as we want to give the writers a chance | |
903 | * to complete when CONFIG_PREEMPT is not set. | |
904 | */ | |
905 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
906 | { | |
907 | struct net_device *dev; | |
908 | unsigned int seq; | |
909 | ||
910 | retry: | |
911 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
912 | rcu_read_lock(); | |
913 | dev = dev_get_by_index_rcu(net, ifindex); | |
914 | if (!dev) { | |
915 | rcu_read_unlock(); | |
916 | return -ENODEV; | |
917 | } | |
918 | ||
919 | strcpy(name, dev->name); | |
920 | rcu_read_unlock(); | |
921 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
922 | cond_resched(); | |
923 | goto retry; | |
924 | } | |
925 | ||
926 | return 0; | |
927 | } | |
928 | ||
1da177e4 | 929 | /** |
941666c2 | 930 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 931 | * @net: the applicable net namespace |
1da177e4 LT |
932 | * @type: media type of device |
933 | * @ha: hardware address | |
934 | * | |
935 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
936 | * is not found or a pointer to the device. |
937 | * The caller must hold RCU or RTNL. | |
941666c2 | 938 | * The returned device has not had its ref count increased |
1da177e4 LT |
939 | * and the caller must therefore be careful about locking |
940 | * | |
1da177e4 LT |
941 | */ |
942 | ||
941666c2 ED |
943 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
944 | const char *ha) | |
1da177e4 LT |
945 | { |
946 | struct net_device *dev; | |
947 | ||
941666c2 | 948 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
949 | if (dev->type == type && |
950 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
951 | return dev; |
952 | ||
953 | return NULL; | |
1da177e4 | 954 | } |
941666c2 | 955 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 956 | |
881d966b | 957 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
958 | { |
959 | struct net_device *dev; | |
960 | ||
4e9cac2b | 961 | ASSERT_RTNL(); |
881d966b | 962 | for_each_netdev(net, dev) |
4e9cac2b | 963 | if (dev->type == type) |
7562f876 PE |
964 | return dev; |
965 | ||
966 | return NULL; | |
4e9cac2b | 967 | } |
4e9cac2b PM |
968 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
969 | ||
881d966b | 970 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 971 | { |
99fe3c39 | 972 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 973 | |
99fe3c39 ED |
974 | rcu_read_lock(); |
975 | for_each_netdev_rcu(net, dev) | |
976 | if (dev->type == type) { | |
977 | dev_hold(dev); | |
978 | ret = dev; | |
979 | break; | |
980 | } | |
981 | rcu_read_unlock(); | |
982 | return ret; | |
1da177e4 | 983 | } |
1da177e4 LT |
984 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
985 | ||
986 | /** | |
6c555490 | 987 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 988 | * @net: the applicable net namespace |
1da177e4 LT |
989 | * @if_flags: IFF_* values |
990 | * @mask: bitmask of bits in if_flags to check | |
991 | * | |
992 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 993 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 994 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
995 | */ |
996 | ||
6c555490 WC |
997 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
998 | unsigned short mask) | |
1da177e4 | 999 | { |
7562f876 | 1000 | struct net_device *dev, *ret; |
1da177e4 | 1001 | |
6c555490 WC |
1002 | ASSERT_RTNL(); |
1003 | ||
7562f876 | 1004 | ret = NULL; |
6c555490 | 1005 | for_each_netdev(net, dev) { |
1da177e4 | 1006 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 1007 | ret = dev; |
1da177e4 LT |
1008 | break; |
1009 | } | |
1010 | } | |
7562f876 | 1011 | return ret; |
1da177e4 | 1012 | } |
6c555490 | 1013 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
1014 | |
1015 | /** | |
1016 | * dev_valid_name - check if name is okay for network device | |
1017 | * @name: name string | |
1018 | * | |
1019 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
1020 | * to allow sysfs to work. We also disallow any kind of |
1021 | * whitespace. | |
1da177e4 | 1022 | */ |
95f050bf | 1023 | bool dev_valid_name(const char *name) |
1da177e4 | 1024 | { |
c7fa9d18 | 1025 | if (*name == '\0') |
95f050bf | 1026 | return false; |
b6fe17d6 | 1027 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 1028 | return false; |
c7fa9d18 | 1029 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1030 | return false; |
c7fa9d18 DM |
1031 | |
1032 | while (*name) { | |
a4176a93 | 1033 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1034 | return false; |
c7fa9d18 DM |
1035 | name++; |
1036 | } | |
95f050bf | 1037 | return true; |
1da177e4 | 1038 | } |
d1b19dff | 1039 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1040 | |
1041 | /** | |
b267b179 EB |
1042 | * __dev_alloc_name - allocate a name for a device |
1043 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1044 | * @name: name format string |
b267b179 | 1045 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1046 | * |
1047 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1048 | * id. It scans list of devices to build up a free map, then chooses |
1049 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1050 | * while allocating the name and adding the device in order to avoid | |
1051 | * duplicates. | |
1052 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1053 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1054 | */ |
1055 | ||
b267b179 | 1056 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1057 | { |
1058 | int i = 0; | |
1da177e4 LT |
1059 | const char *p; |
1060 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1061 | unsigned long *inuse; |
1da177e4 LT |
1062 | struct net_device *d; |
1063 | ||
1064 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
1065 | if (p) { | |
1066 | /* | |
1067 | * Verify the string as this thing may have come from | |
1068 | * the user. There must be either one "%d" and no other "%" | |
1069 | * characters. | |
1070 | */ | |
1071 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1072 | return -EINVAL; | |
1073 | ||
1074 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1075 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1076 | if (!inuse) |
1077 | return -ENOMEM; | |
1078 | ||
881d966b | 1079 | for_each_netdev(net, d) { |
1da177e4 LT |
1080 | if (!sscanf(d->name, name, &i)) |
1081 | continue; | |
1082 | if (i < 0 || i >= max_netdevices) | |
1083 | continue; | |
1084 | ||
1085 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1086 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1087 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1088 | set_bit(i, inuse); | |
1089 | } | |
1090 | ||
1091 | i = find_first_zero_bit(inuse, max_netdevices); | |
1092 | free_page((unsigned long) inuse); | |
1093 | } | |
1094 | ||
d9031024 OP |
1095 | if (buf != name) |
1096 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1097 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1098 | return i; |
1da177e4 LT |
1099 | |
1100 | /* It is possible to run out of possible slots | |
1101 | * when the name is long and there isn't enough space left | |
1102 | * for the digits, or if all bits are used. | |
1103 | */ | |
1104 | return -ENFILE; | |
1105 | } | |
1106 | ||
b267b179 EB |
1107 | /** |
1108 | * dev_alloc_name - allocate a name for a device | |
1109 | * @dev: device | |
1110 | * @name: name format string | |
1111 | * | |
1112 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1113 | * id. It scans list of devices to build up a free map, then chooses | |
1114 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1115 | * while allocating the name and adding the device in order to avoid | |
1116 | * duplicates. | |
1117 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1118 | * Returns the number of the unit assigned or a negative errno code. | |
1119 | */ | |
1120 | ||
1121 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1122 | { | |
1123 | char buf[IFNAMSIZ]; | |
1124 | struct net *net; | |
1125 | int ret; | |
1126 | ||
c346dca1 YH |
1127 | BUG_ON(!dev_net(dev)); |
1128 | net = dev_net(dev); | |
b267b179 EB |
1129 | ret = __dev_alloc_name(net, name, buf); |
1130 | if (ret >= 0) | |
1131 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1132 | return ret; | |
1133 | } | |
d1b19dff | 1134 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1135 | |
828de4f6 G |
1136 | static int dev_alloc_name_ns(struct net *net, |
1137 | struct net_device *dev, | |
1138 | const char *name) | |
d9031024 | 1139 | { |
828de4f6 G |
1140 | char buf[IFNAMSIZ]; |
1141 | int ret; | |
8ce6cebc | 1142 | |
828de4f6 G |
1143 | ret = __dev_alloc_name(net, name, buf); |
1144 | if (ret >= 0) | |
1145 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1146 | return ret; | |
1147 | } | |
1148 | ||
1149 | static int dev_get_valid_name(struct net *net, | |
1150 | struct net_device *dev, | |
1151 | const char *name) | |
1152 | { | |
1153 | BUG_ON(!net); | |
8ce6cebc | 1154 | |
d9031024 OP |
1155 | if (!dev_valid_name(name)) |
1156 | return -EINVAL; | |
1157 | ||
1c5cae81 | 1158 | if (strchr(name, '%')) |
828de4f6 | 1159 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1160 | else if (__dev_get_by_name(net, name)) |
1161 | return -EEXIST; | |
8ce6cebc DL |
1162 | else if (dev->name != name) |
1163 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1164 | |
1165 | return 0; | |
1166 | } | |
1da177e4 LT |
1167 | |
1168 | /** | |
1169 | * dev_change_name - change name of a device | |
1170 | * @dev: device | |
1171 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1172 | * | |
1173 | * Change name of a device, can pass format strings "eth%d". | |
1174 | * for wildcarding. | |
1175 | */ | |
cf04a4c7 | 1176 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1177 | { |
238fa362 | 1178 | unsigned char old_assign_type; |
fcc5a03a | 1179 | char oldname[IFNAMSIZ]; |
1da177e4 | 1180 | int err = 0; |
fcc5a03a | 1181 | int ret; |
881d966b | 1182 | struct net *net; |
1da177e4 LT |
1183 | |
1184 | ASSERT_RTNL(); | |
c346dca1 | 1185 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1186 | |
c346dca1 | 1187 | net = dev_net(dev); |
1da177e4 LT |
1188 | if (dev->flags & IFF_UP) |
1189 | return -EBUSY; | |
1190 | ||
30e6c9fa | 1191 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1192 | |
1193 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1194 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1195 | return 0; |
c91f6df2 | 1196 | } |
c8d90dca | 1197 | |
fcc5a03a HX |
1198 | memcpy(oldname, dev->name, IFNAMSIZ); |
1199 | ||
828de4f6 | 1200 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1201 | if (err < 0) { |
30e6c9fa | 1202 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1203 | return err; |
c91f6df2 | 1204 | } |
1da177e4 | 1205 | |
6fe82a39 VF |
1206 | if (oldname[0] && !strchr(oldname, '%')) |
1207 | netdev_info(dev, "renamed from %s\n", oldname); | |
1208 | ||
238fa362 TG |
1209 | old_assign_type = dev->name_assign_type; |
1210 | dev->name_assign_type = NET_NAME_RENAMED; | |
1211 | ||
fcc5a03a | 1212 | rollback: |
a1b3f594 EB |
1213 | ret = device_rename(&dev->dev, dev->name); |
1214 | if (ret) { | |
1215 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1216 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1217 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1218 | return ret; |
dcc99773 | 1219 | } |
7f988eab | 1220 | |
30e6c9fa | 1221 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1222 | |
5bb025fa VF |
1223 | netdev_adjacent_rename_links(dev, oldname); |
1224 | ||
7f988eab | 1225 | write_lock_bh(&dev_base_lock); |
372b2312 | 1226 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1227 | write_unlock_bh(&dev_base_lock); |
1228 | ||
1229 | synchronize_rcu(); | |
1230 | ||
1231 | write_lock_bh(&dev_base_lock); | |
1232 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1233 | write_unlock_bh(&dev_base_lock); |
1234 | ||
056925ab | 1235 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1236 | ret = notifier_to_errno(ret); |
1237 | ||
1238 | if (ret) { | |
91e9c07b ED |
1239 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1240 | if (err >= 0) { | |
fcc5a03a | 1241 | err = ret; |
30e6c9fa | 1242 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1243 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1244 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1245 | dev->name_assign_type = old_assign_type; |
1246 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1247 | goto rollback; |
91e9c07b | 1248 | } else { |
7b6cd1ce | 1249 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1250 | dev->name, ret); |
fcc5a03a HX |
1251 | } |
1252 | } | |
1da177e4 LT |
1253 | |
1254 | return err; | |
1255 | } | |
1256 | ||
0b815a1a SH |
1257 | /** |
1258 | * dev_set_alias - change ifalias of a device | |
1259 | * @dev: device | |
1260 | * @alias: name up to IFALIASZ | |
f0db275a | 1261 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1262 | * |
1263 | * Set ifalias for a device, | |
1264 | */ | |
1265 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1266 | { | |
7364e445 AK |
1267 | char *new_ifalias; |
1268 | ||
0b815a1a SH |
1269 | ASSERT_RTNL(); |
1270 | ||
1271 | if (len >= IFALIASZ) | |
1272 | return -EINVAL; | |
1273 | ||
96ca4a2c | 1274 | if (!len) { |
388dfc2d SK |
1275 | kfree(dev->ifalias); |
1276 | dev->ifalias = NULL; | |
96ca4a2c OH |
1277 | return 0; |
1278 | } | |
1279 | ||
7364e445 AK |
1280 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1281 | if (!new_ifalias) | |
0b815a1a | 1282 | return -ENOMEM; |
7364e445 | 1283 | dev->ifalias = new_ifalias; |
c28294b9 AP |
1284 | memcpy(dev->ifalias, alias, len); |
1285 | dev->ifalias[len] = 0; | |
0b815a1a | 1286 | |
0b815a1a SH |
1287 | return len; |
1288 | } | |
1289 | ||
1290 | ||
d8a33ac4 | 1291 | /** |
3041a069 | 1292 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1293 | * @dev: device to cause notification |
1294 | * | |
1295 | * Called to indicate a device has changed features. | |
1296 | */ | |
1297 | void netdev_features_change(struct net_device *dev) | |
1298 | { | |
056925ab | 1299 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1300 | } |
1301 | EXPORT_SYMBOL(netdev_features_change); | |
1302 | ||
1da177e4 LT |
1303 | /** |
1304 | * netdev_state_change - device changes state | |
1305 | * @dev: device to cause notification | |
1306 | * | |
1307 | * Called to indicate a device has changed state. This function calls | |
1308 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1309 | * to the routing socket. | |
1310 | */ | |
1311 | void netdev_state_change(struct net_device *dev) | |
1312 | { | |
1313 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1314 | struct netdev_notifier_change_info change_info; |
1315 | ||
1316 | change_info.flags_changed = 0; | |
1317 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1318 | &change_info.info); | |
7f294054 | 1319 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1320 | } |
1321 | } | |
d1b19dff | 1322 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1323 | |
ee89bab1 | 1324 | /** |
722c9a0c | 1325 | * netdev_notify_peers - notify network peers about existence of @dev |
1326 | * @dev: network device | |
ee89bab1 AW |
1327 | * |
1328 | * Generate traffic such that interested network peers are aware of | |
1329 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1330 | * a device wants to inform the rest of the network about some sort of | |
1331 | * reconfiguration such as a failover event or virtual machine | |
1332 | * migration. | |
1333 | */ | |
1334 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1335 | { |
ee89bab1 AW |
1336 | rtnl_lock(); |
1337 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
37c343b4 | 1338 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); |
ee89bab1 | 1339 | rtnl_unlock(); |
c1da4ac7 | 1340 | } |
ee89bab1 | 1341 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1342 | |
bd380811 | 1343 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1344 | { |
d314774c | 1345 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1346 | int ret; |
1da177e4 | 1347 | |
e46b66bc BH |
1348 | ASSERT_RTNL(); |
1349 | ||
1da177e4 LT |
1350 | if (!netif_device_present(dev)) |
1351 | return -ENODEV; | |
1352 | ||
ca99ca14 NH |
1353 | /* Block netpoll from trying to do any rx path servicing. |
1354 | * If we don't do this there is a chance ndo_poll_controller | |
1355 | * or ndo_poll may be running while we open the device | |
1356 | */ | |
66b5552f | 1357 | netpoll_poll_disable(dev); |
ca99ca14 | 1358 | |
3b8bcfd5 JB |
1359 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1360 | ret = notifier_to_errno(ret); | |
1361 | if (ret) | |
1362 | return ret; | |
1363 | ||
1da177e4 | 1364 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1365 | |
d314774c SH |
1366 | if (ops->ndo_validate_addr) |
1367 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1368 | |
d314774c SH |
1369 | if (!ret && ops->ndo_open) |
1370 | ret = ops->ndo_open(dev); | |
1da177e4 | 1371 | |
66b5552f | 1372 | netpoll_poll_enable(dev); |
ca99ca14 | 1373 | |
bada339b JG |
1374 | if (ret) |
1375 | clear_bit(__LINK_STATE_START, &dev->state); | |
1376 | else { | |
1da177e4 | 1377 | dev->flags |= IFF_UP; |
4417da66 | 1378 | dev_set_rx_mode(dev); |
1da177e4 | 1379 | dev_activate(dev); |
7bf23575 | 1380 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1381 | } |
bada339b | 1382 | |
1da177e4 LT |
1383 | return ret; |
1384 | } | |
1385 | ||
1386 | /** | |
bd380811 PM |
1387 | * dev_open - prepare an interface for use. |
1388 | * @dev: device to open | |
1da177e4 | 1389 | * |
bd380811 PM |
1390 | * Takes a device from down to up state. The device's private open |
1391 | * function is invoked and then the multicast lists are loaded. Finally | |
1392 | * the device is moved into the up state and a %NETDEV_UP message is | |
1393 | * sent to the netdev notifier chain. | |
1394 | * | |
1395 | * Calling this function on an active interface is a nop. On a failure | |
1396 | * a negative errno code is returned. | |
1da177e4 | 1397 | */ |
bd380811 PM |
1398 | int dev_open(struct net_device *dev) |
1399 | { | |
1400 | int ret; | |
1401 | ||
bd380811 PM |
1402 | if (dev->flags & IFF_UP) |
1403 | return 0; | |
1404 | ||
bd380811 PM |
1405 | ret = __dev_open(dev); |
1406 | if (ret < 0) | |
1407 | return ret; | |
1408 | ||
7f294054 | 1409 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1410 | call_netdevice_notifiers(NETDEV_UP, dev); |
1411 | ||
1412 | return ret; | |
1413 | } | |
1414 | EXPORT_SYMBOL(dev_open); | |
1415 | ||
44345724 | 1416 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1417 | { |
44345724 | 1418 | struct net_device *dev; |
e46b66bc | 1419 | |
bd380811 | 1420 | ASSERT_RTNL(); |
9d5010db DM |
1421 | might_sleep(); |
1422 | ||
5cde2829 | 1423 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1424 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1425 | netpoll_poll_disable(dev); |
3f4df206 | 1426 | |
44345724 | 1427 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1428 | |
44345724 | 1429 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1430 | |
44345724 OP |
1431 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1432 | * can be even on different cpu. So just clear netif_running(). | |
1433 | * | |
1434 | * dev->stop() will invoke napi_disable() on all of it's | |
1435 | * napi_struct instances on this device. | |
1436 | */ | |
4e857c58 | 1437 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1438 | } |
1da177e4 | 1439 | |
44345724 | 1440 | dev_deactivate_many(head); |
d8b2a4d2 | 1441 | |
5cde2829 | 1442 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1443 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1444 | |
44345724 OP |
1445 | /* |
1446 | * Call the device specific close. This cannot fail. | |
1447 | * Only if device is UP | |
1448 | * | |
1449 | * We allow it to be called even after a DETACH hot-plug | |
1450 | * event. | |
1451 | */ | |
1452 | if (ops->ndo_stop) | |
1453 | ops->ndo_stop(dev); | |
1454 | ||
44345724 | 1455 | dev->flags &= ~IFF_UP; |
66b5552f | 1456 | netpoll_poll_enable(dev); |
44345724 OP |
1457 | } |
1458 | ||
1459 | return 0; | |
1460 | } | |
1461 | ||
1462 | static int __dev_close(struct net_device *dev) | |
1463 | { | |
f87e6f47 | 1464 | int retval; |
44345724 OP |
1465 | LIST_HEAD(single); |
1466 | ||
5cde2829 | 1467 | list_add(&dev->close_list, &single); |
f87e6f47 LT |
1468 | retval = __dev_close_many(&single); |
1469 | list_del(&single); | |
ca99ca14 | 1470 | |
f87e6f47 | 1471 | return retval; |
44345724 OP |
1472 | } |
1473 | ||
99c4a26a | 1474 | int dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1475 | { |
1476 | struct net_device *dev, *tmp; | |
1da177e4 | 1477 | |
5cde2829 EB |
1478 | /* Remove the devices that don't need to be closed */ |
1479 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1480 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1481 | list_del_init(&dev->close_list); |
44345724 OP |
1482 | |
1483 | __dev_close_many(head); | |
1da177e4 | 1484 | |
5cde2829 | 1485 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1486 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1487 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1488 | if (unlink) |
1489 | list_del_init(&dev->close_list); | |
44345724 | 1490 | } |
bd380811 PM |
1491 | |
1492 | return 0; | |
1493 | } | |
99c4a26a | 1494 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1495 | |
1496 | /** | |
1497 | * dev_close - shutdown an interface. | |
1498 | * @dev: device to shutdown | |
1499 | * | |
1500 | * This function moves an active device into down state. A | |
1501 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1502 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1503 | * chain. | |
1504 | */ | |
1505 | int dev_close(struct net_device *dev) | |
1506 | { | |
e14a5993 ED |
1507 | if (dev->flags & IFF_UP) { |
1508 | LIST_HEAD(single); | |
1da177e4 | 1509 | |
5cde2829 | 1510 | list_add(&dev->close_list, &single); |
99c4a26a | 1511 | dev_close_many(&single, true); |
e14a5993 ED |
1512 | list_del(&single); |
1513 | } | |
da6e378b | 1514 | return 0; |
1da177e4 | 1515 | } |
d1b19dff | 1516 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1517 | |
1518 | ||
0187bdfb BH |
1519 | /** |
1520 | * dev_disable_lro - disable Large Receive Offload on a device | |
1521 | * @dev: device | |
1522 | * | |
1523 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1524 | * called under RTNL. This is needed if received packets may be | |
1525 | * forwarded to another interface. | |
1526 | */ | |
1527 | void dev_disable_lro(struct net_device *dev) | |
1528 | { | |
fbe168ba MK |
1529 | struct net_device *lower_dev; |
1530 | struct list_head *iter; | |
529d0489 | 1531 | |
bc5787c6 MM |
1532 | dev->wanted_features &= ~NETIF_F_LRO; |
1533 | netdev_update_features(dev); | |
27660515 | 1534 | |
22d5969f MM |
1535 | if (unlikely(dev->features & NETIF_F_LRO)) |
1536 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1537 | |
1538 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1539 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1540 | } |
1541 | EXPORT_SYMBOL(dev_disable_lro); | |
1542 | ||
351638e7 JP |
1543 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1544 | struct net_device *dev) | |
1545 | { | |
1546 | struct netdev_notifier_info info; | |
1547 | ||
1548 | netdev_notifier_info_init(&info, dev); | |
1549 | return nb->notifier_call(nb, val, &info); | |
1550 | } | |
0187bdfb | 1551 | |
881d966b EB |
1552 | static int dev_boot_phase = 1; |
1553 | ||
1da177e4 | 1554 | /** |
722c9a0c | 1555 | * register_netdevice_notifier - register a network notifier block |
1556 | * @nb: notifier | |
1da177e4 | 1557 | * |
722c9a0c | 1558 | * Register a notifier to be called when network device events occur. |
1559 | * The notifier passed is linked into the kernel structures and must | |
1560 | * not be reused until it has been unregistered. A negative errno code | |
1561 | * is returned on a failure. | |
1da177e4 | 1562 | * |
722c9a0c | 1563 | * When registered all registration and up events are replayed |
1564 | * to the new notifier to allow device to have a race free | |
1565 | * view of the network device list. | |
1da177e4 LT |
1566 | */ |
1567 | ||
1568 | int register_netdevice_notifier(struct notifier_block *nb) | |
1569 | { | |
1570 | struct net_device *dev; | |
fcc5a03a | 1571 | struct net_device *last; |
881d966b | 1572 | struct net *net; |
1da177e4 LT |
1573 | int err; |
1574 | ||
1575 | rtnl_lock(); | |
f07d5b94 | 1576 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1577 | if (err) |
1578 | goto unlock; | |
881d966b EB |
1579 | if (dev_boot_phase) |
1580 | goto unlock; | |
1581 | for_each_net(net) { | |
1582 | for_each_netdev(net, dev) { | |
351638e7 | 1583 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1584 | err = notifier_to_errno(err); |
1585 | if (err) | |
1586 | goto rollback; | |
1587 | ||
1588 | if (!(dev->flags & IFF_UP)) | |
1589 | continue; | |
1da177e4 | 1590 | |
351638e7 | 1591 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1592 | } |
1da177e4 | 1593 | } |
fcc5a03a HX |
1594 | |
1595 | unlock: | |
1da177e4 LT |
1596 | rtnl_unlock(); |
1597 | return err; | |
fcc5a03a HX |
1598 | |
1599 | rollback: | |
1600 | last = dev; | |
881d966b EB |
1601 | for_each_net(net) { |
1602 | for_each_netdev(net, dev) { | |
1603 | if (dev == last) | |
8f891489 | 1604 | goto outroll; |
fcc5a03a | 1605 | |
881d966b | 1606 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1607 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1608 | dev); | |
1609 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1610 | } |
351638e7 | 1611 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1612 | } |
fcc5a03a | 1613 | } |
c67625a1 | 1614 | |
8f891489 | 1615 | outroll: |
c67625a1 | 1616 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1617 | goto unlock; |
1da177e4 | 1618 | } |
d1b19dff | 1619 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1620 | |
1621 | /** | |
722c9a0c | 1622 | * unregister_netdevice_notifier - unregister a network notifier block |
1623 | * @nb: notifier | |
1da177e4 | 1624 | * |
722c9a0c | 1625 | * Unregister a notifier previously registered by |
1626 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1627 | * kernel structures and may then be reused. A negative errno code | |
1628 | * is returned on a failure. | |
7d3d43da | 1629 | * |
722c9a0c | 1630 | * After unregistering unregister and down device events are synthesized |
1631 | * for all devices on the device list to the removed notifier to remove | |
1632 | * the need for special case cleanup code. | |
1da177e4 LT |
1633 | */ |
1634 | ||
1635 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1636 | { | |
7d3d43da EB |
1637 | struct net_device *dev; |
1638 | struct net *net; | |
9f514950 HX |
1639 | int err; |
1640 | ||
1641 | rtnl_lock(); | |
f07d5b94 | 1642 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1643 | if (err) |
1644 | goto unlock; | |
1645 | ||
1646 | for_each_net(net) { | |
1647 | for_each_netdev(net, dev) { | |
1648 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1649 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1650 | dev); | |
1651 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1652 | } |
351638e7 | 1653 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1654 | } |
1655 | } | |
1656 | unlock: | |
9f514950 HX |
1657 | rtnl_unlock(); |
1658 | return err; | |
1da177e4 | 1659 | } |
d1b19dff | 1660 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1661 | |
351638e7 JP |
1662 | /** |
1663 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1664 | * @val: value passed unmodified to notifier function | |
1665 | * @dev: net_device pointer passed unmodified to notifier function | |
1666 | * @info: notifier information data | |
1667 | * | |
1668 | * Call all network notifier blocks. Parameters and return value | |
1669 | * are as for raw_notifier_call_chain(). | |
1670 | */ | |
1671 | ||
1d143d9f | 1672 | static int call_netdevice_notifiers_info(unsigned long val, |
1673 | struct net_device *dev, | |
1674 | struct netdev_notifier_info *info) | |
351638e7 JP |
1675 | { |
1676 | ASSERT_RTNL(); | |
1677 | netdev_notifier_info_init(info, dev); | |
1678 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1679 | } | |
351638e7 | 1680 | |
1da177e4 LT |
1681 | /** |
1682 | * call_netdevice_notifiers - call all network notifier blocks | |
1683 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1684 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1685 | * |
1686 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1687 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1688 | */ |
1689 | ||
ad7379d4 | 1690 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1691 | { |
351638e7 JP |
1692 | struct netdev_notifier_info info; |
1693 | ||
1694 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1695 | } |
edf947f1 | 1696 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1697 | |
1cf51900 | 1698 | #ifdef CONFIG_NET_INGRESS |
4577139b DB |
1699 | static struct static_key ingress_needed __read_mostly; |
1700 | ||
1701 | void net_inc_ingress_queue(void) | |
1702 | { | |
1703 | static_key_slow_inc(&ingress_needed); | |
1704 | } | |
1705 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
1706 | ||
1707 | void net_dec_ingress_queue(void) | |
1708 | { | |
1709 | static_key_slow_dec(&ingress_needed); | |
1710 | } | |
1711 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
1712 | #endif | |
1713 | ||
1f211a1b DB |
1714 | #ifdef CONFIG_NET_EGRESS |
1715 | static struct static_key egress_needed __read_mostly; | |
1716 | ||
1717 | void net_inc_egress_queue(void) | |
1718 | { | |
1719 | static_key_slow_inc(&egress_needed); | |
1720 | } | |
1721 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
1722 | ||
1723 | void net_dec_egress_queue(void) | |
1724 | { | |
1725 | static_key_slow_dec(&egress_needed); | |
1726 | } | |
1727 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
1728 | #endif | |
1729 | ||
c5905afb | 1730 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1731 | #ifdef HAVE_JUMP_LABEL |
b90e5794 | 1732 | static atomic_t netstamp_needed_deferred; |
13baa00a | 1733 | static atomic_t netstamp_wanted; |
5fa8bbda | 1734 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 1735 | { |
b90e5794 | 1736 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 1737 | int wanted; |
b90e5794 | 1738 | |
13baa00a ED |
1739 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
1740 | if (wanted > 0) | |
1741 | static_key_enable(&netstamp_needed); | |
1742 | else | |
1743 | static_key_disable(&netstamp_needed); | |
5fa8bbda ED |
1744 | } |
1745 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 1746 | #endif |
5fa8bbda ED |
1747 | |
1748 | void net_enable_timestamp(void) | |
1749 | { | |
13baa00a ED |
1750 | #ifdef HAVE_JUMP_LABEL |
1751 | int wanted; | |
1752 | ||
1753 | while (1) { | |
1754 | wanted = atomic_read(&netstamp_wanted); | |
1755 | if (wanted <= 0) | |
1756 | break; | |
1757 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) | |
1758 | return; | |
1759 | } | |
1760 | atomic_inc(&netstamp_needed_deferred); | |
1761 | schedule_work(&netstamp_work); | |
1762 | #else | |
c5905afb | 1763 | static_key_slow_inc(&netstamp_needed); |
13baa00a | 1764 | #endif |
1da177e4 | 1765 | } |
d1b19dff | 1766 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1767 | |
1768 | void net_disable_timestamp(void) | |
1769 | { | |
b90e5794 | 1770 | #ifdef HAVE_JUMP_LABEL |
13baa00a ED |
1771 | int wanted; |
1772 | ||
1773 | while (1) { | |
1774 | wanted = atomic_read(&netstamp_wanted); | |
1775 | if (wanted <= 1) | |
1776 | break; | |
1777 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) | |
1778 | return; | |
1779 | } | |
1780 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
1781 | schedule_work(&netstamp_work); |
1782 | #else | |
c5905afb | 1783 | static_key_slow_dec(&netstamp_needed); |
5fa8bbda | 1784 | #endif |
1da177e4 | 1785 | } |
d1b19dff | 1786 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1787 | |
3b098e2d | 1788 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1789 | { |
2456e855 | 1790 | skb->tstamp = 0; |
c5905afb | 1791 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1792 | __net_timestamp(skb); |
1da177e4 LT |
1793 | } |
1794 | ||
588f0330 | 1795 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1796 | if (static_key_false(&netstamp_needed)) { \ |
2456e855 | 1797 | if ((COND) && !(SKB)->tstamp) \ |
588f0330 ED |
1798 | __net_timestamp(SKB); \ |
1799 | } \ | |
3b098e2d | 1800 | |
f4b05d27 | 1801 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 DL |
1802 | { |
1803 | unsigned int len; | |
1804 | ||
1805 | if (!(dev->flags & IFF_UP)) | |
1806 | return false; | |
1807 | ||
1808 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1809 | if (skb->len <= len) | |
1810 | return true; | |
1811 | ||
1812 | /* if TSO is enabled, we don't care about the length as the packet | |
1813 | * could be forwarded without being segmented before | |
1814 | */ | |
1815 | if (skb_is_gso(skb)) | |
1816 | return true; | |
1817 | ||
1818 | return false; | |
1819 | } | |
1ee481fb | 1820 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1821 | |
a0265d28 HX |
1822 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1823 | { | |
4e3264d2 | 1824 | int ret = ____dev_forward_skb(dev, skb); |
a0265d28 | 1825 | |
4e3264d2 MKL |
1826 | if (likely(!ret)) { |
1827 | skb->protocol = eth_type_trans(skb, dev); | |
1828 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
1829 | } | |
a0265d28 | 1830 | |
4e3264d2 | 1831 | return ret; |
a0265d28 HX |
1832 | } |
1833 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1834 | ||
44540960 AB |
1835 | /** |
1836 | * dev_forward_skb - loopback an skb to another netif | |
1837 | * | |
1838 | * @dev: destination network device | |
1839 | * @skb: buffer to forward | |
1840 | * | |
1841 | * return values: | |
1842 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1843 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1844 | * |
1845 | * dev_forward_skb can be used for injecting an skb from the | |
1846 | * start_xmit function of one device into the receive queue | |
1847 | * of another device. | |
1848 | * | |
1849 | * The receiving device may be in another namespace, so | |
1850 | * we have to clear all information in the skb that could | |
1851 | * impact namespace isolation. | |
1852 | */ | |
1853 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1854 | { | |
a0265d28 | 1855 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1856 | } |
1857 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1858 | ||
71d9dec2 CG |
1859 | static inline int deliver_skb(struct sk_buff *skb, |
1860 | struct packet_type *pt_prev, | |
1861 | struct net_device *orig_dev) | |
1862 | { | |
1080e512 MT |
1863 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1864 | return -ENOMEM; | |
63354797 | 1865 | refcount_inc(&skb->users); |
71d9dec2 CG |
1866 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1867 | } | |
1868 | ||
7866a621 SN |
1869 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
1870 | struct packet_type **pt, | |
fbcb2170 JP |
1871 | struct net_device *orig_dev, |
1872 | __be16 type, | |
7866a621 SN |
1873 | struct list_head *ptype_list) |
1874 | { | |
1875 | struct packet_type *ptype, *pt_prev = *pt; | |
1876 | ||
1877 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1878 | if (ptype->type != type) | |
1879 | continue; | |
1880 | if (pt_prev) | |
fbcb2170 | 1881 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
1882 | pt_prev = ptype; |
1883 | } | |
1884 | *pt = pt_prev; | |
1885 | } | |
1886 | ||
c0de08d0 EL |
1887 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1888 | { | |
a3d744e9 | 1889 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1890 | return false; |
1891 | ||
1892 | if (ptype->id_match) | |
1893 | return ptype->id_match(ptype, skb->sk); | |
1894 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1895 | return true; | |
1896 | ||
1897 | return false; | |
1898 | } | |
1899 | ||
1da177e4 LT |
1900 | /* |
1901 | * Support routine. Sends outgoing frames to any network | |
1902 | * taps currently in use. | |
1903 | */ | |
1904 | ||
74b20582 | 1905 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1906 | { |
1907 | struct packet_type *ptype; | |
71d9dec2 CG |
1908 | struct sk_buff *skb2 = NULL; |
1909 | struct packet_type *pt_prev = NULL; | |
7866a621 | 1910 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 1911 | |
1da177e4 | 1912 | rcu_read_lock(); |
7866a621 SN |
1913 | again: |
1914 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1da177e4 LT |
1915 | /* Never send packets back to the socket |
1916 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1917 | */ | |
7866a621 SN |
1918 | if (skb_loop_sk(ptype, skb)) |
1919 | continue; | |
71d9dec2 | 1920 | |
7866a621 SN |
1921 | if (pt_prev) { |
1922 | deliver_skb(skb2, pt_prev, skb->dev); | |
1923 | pt_prev = ptype; | |
1924 | continue; | |
1925 | } | |
1da177e4 | 1926 | |
7866a621 SN |
1927 | /* need to clone skb, done only once */ |
1928 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1929 | if (!skb2) | |
1930 | goto out_unlock; | |
70978182 | 1931 | |
7866a621 | 1932 | net_timestamp_set(skb2); |
1da177e4 | 1933 | |
7866a621 SN |
1934 | /* skb->nh should be correctly |
1935 | * set by sender, so that the second statement is | |
1936 | * just protection against buggy protocols. | |
1937 | */ | |
1938 | skb_reset_mac_header(skb2); | |
1939 | ||
1940 | if (skb_network_header(skb2) < skb2->data || | |
1941 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
1942 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
1943 | ntohs(skb2->protocol), | |
1944 | dev->name); | |
1945 | skb_reset_network_header(skb2); | |
1da177e4 | 1946 | } |
7866a621 SN |
1947 | |
1948 | skb2->transport_header = skb2->network_header; | |
1949 | skb2->pkt_type = PACKET_OUTGOING; | |
1950 | pt_prev = ptype; | |
1951 | } | |
1952 | ||
1953 | if (ptype_list == &ptype_all) { | |
1954 | ptype_list = &dev->ptype_all; | |
1955 | goto again; | |
1da177e4 | 1956 | } |
7866a621 | 1957 | out_unlock: |
71d9dec2 CG |
1958 | if (pt_prev) |
1959 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1960 | rcu_read_unlock(); |
1961 | } | |
74b20582 | 1962 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 1963 | |
2c53040f BH |
1964 | /** |
1965 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1966 | * @dev: Network device |
1967 | * @txq: number of queues available | |
1968 | * | |
1969 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1970 | * valid. To resolve this verify the tc mapping remains valid and if | |
1971 | * not NULL the mapping. With no priorities mapping to this | |
1972 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1973 | * is invalid nothing can be done so disable priority mappings. If is | |
1974 | * expected that drivers will fix this mapping if they can before | |
1975 | * calling netif_set_real_num_tx_queues. | |
1976 | */ | |
bb134d22 | 1977 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1978 | { |
1979 | int i; | |
1980 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1981 | ||
1982 | /* If TC0 is invalidated disable TC mapping */ | |
1983 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1984 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1985 | dev->num_tc = 0; |
1986 | return; | |
1987 | } | |
1988 | ||
1989 | /* Invalidated prio to tc mappings set to TC0 */ | |
1990 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1991 | int q = netdev_get_prio_tc_map(dev, i); | |
1992 | ||
1993 | tc = &dev->tc_to_txq[q]; | |
1994 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1995 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1996 | i, q); | |
4f57c087 JF |
1997 | netdev_set_prio_tc_map(dev, i, 0); |
1998 | } | |
1999 | } | |
2000 | } | |
2001 | ||
8d059b0f AD |
2002 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
2003 | { | |
2004 | if (dev->num_tc) { | |
2005 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2006 | int i; | |
2007 | ||
2008 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { | |
2009 | if ((txq - tc->offset) < tc->count) | |
2010 | return i; | |
2011 | } | |
2012 | ||
2013 | return -1; | |
2014 | } | |
2015 | ||
2016 | return 0; | |
2017 | } | |
2018 | ||
537c00de AD |
2019 | #ifdef CONFIG_XPS |
2020 | static DEFINE_MUTEX(xps_map_mutex); | |
2021 | #define xmap_dereference(P) \ | |
2022 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
2023 | ||
6234f874 AD |
2024 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
2025 | int tci, u16 index) | |
537c00de | 2026 | { |
10cdc3f3 AD |
2027 | struct xps_map *map = NULL; |
2028 | int pos; | |
537c00de | 2029 | |
10cdc3f3 | 2030 | if (dev_maps) |
6234f874 AD |
2031 | map = xmap_dereference(dev_maps->cpu_map[tci]); |
2032 | if (!map) | |
2033 | return false; | |
537c00de | 2034 | |
6234f874 AD |
2035 | for (pos = map->len; pos--;) { |
2036 | if (map->queues[pos] != index) | |
2037 | continue; | |
2038 | ||
2039 | if (map->len > 1) { | |
2040 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2041 | break; |
537c00de | 2042 | } |
6234f874 AD |
2043 | |
2044 | RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL); | |
2045 | kfree_rcu(map, rcu); | |
2046 | return false; | |
537c00de AD |
2047 | } |
2048 | ||
6234f874 | 2049 | return true; |
10cdc3f3 AD |
2050 | } |
2051 | ||
6234f874 AD |
2052 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2053 | struct xps_dev_maps *dev_maps, | |
2054 | int cpu, u16 offset, u16 count) | |
2055 | { | |
184c449f AD |
2056 | int num_tc = dev->num_tc ? : 1; |
2057 | bool active = false; | |
2058 | int tci; | |
6234f874 | 2059 | |
184c449f AD |
2060 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2061 | int i, j; | |
2062 | ||
2063 | for (i = count, j = offset; i--; j++) { | |
2064 | if (!remove_xps_queue(dev_maps, cpu, j)) | |
2065 | break; | |
2066 | } | |
2067 | ||
2068 | active |= i < 0; | |
6234f874 AD |
2069 | } |
2070 | ||
184c449f | 2071 | return active; |
6234f874 AD |
2072 | } |
2073 | ||
2074 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, | |
2075 | u16 count) | |
10cdc3f3 AD |
2076 | { |
2077 | struct xps_dev_maps *dev_maps; | |
024e9679 | 2078 | int cpu, i; |
10cdc3f3 AD |
2079 | bool active = false; |
2080 | ||
2081 | mutex_lock(&xps_map_mutex); | |
2082 | dev_maps = xmap_dereference(dev->xps_maps); | |
2083 | ||
2084 | if (!dev_maps) | |
2085 | goto out_no_maps; | |
2086 | ||
6234f874 AD |
2087 | for_each_possible_cpu(cpu) |
2088 | active |= remove_xps_queue_cpu(dev, dev_maps, cpu, | |
2089 | offset, count); | |
10cdc3f3 AD |
2090 | |
2091 | if (!active) { | |
537c00de AD |
2092 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
2093 | kfree_rcu(dev_maps, rcu); | |
2094 | } | |
2095 | ||
6234f874 | 2096 | for (i = offset + (count - 1); count--; i--) |
024e9679 AD |
2097 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), |
2098 | NUMA_NO_NODE); | |
2099 | ||
537c00de AD |
2100 | out_no_maps: |
2101 | mutex_unlock(&xps_map_mutex); | |
2102 | } | |
2103 | ||
6234f874 AD |
2104 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2105 | { | |
2106 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2107 | } | |
2108 | ||
01c5f864 AD |
2109 | static struct xps_map *expand_xps_map(struct xps_map *map, |
2110 | int cpu, u16 index) | |
2111 | { | |
2112 | struct xps_map *new_map; | |
2113 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2114 | int i, pos; | |
2115 | ||
2116 | for (pos = 0; map && pos < map->len; pos++) { | |
2117 | if (map->queues[pos] != index) | |
2118 | continue; | |
2119 | return map; | |
2120 | } | |
2121 | ||
2122 | /* Need to add queue to this CPU's existing map */ | |
2123 | if (map) { | |
2124 | if (pos < map->alloc_len) | |
2125 | return map; | |
2126 | ||
2127 | alloc_len = map->alloc_len * 2; | |
2128 | } | |
2129 | ||
2130 | /* Need to allocate new map to store queue on this CPU's map */ | |
2131 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2132 | cpu_to_node(cpu)); | |
2133 | if (!new_map) | |
2134 | return NULL; | |
2135 | ||
2136 | for (i = 0; i < pos; i++) | |
2137 | new_map->queues[i] = map->queues[i]; | |
2138 | new_map->alloc_len = alloc_len; | |
2139 | new_map->len = pos; | |
2140 | ||
2141 | return new_map; | |
2142 | } | |
2143 | ||
3573540c MT |
2144 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
2145 | u16 index) | |
537c00de | 2146 | { |
01c5f864 | 2147 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
184c449f AD |
2148 | int i, cpu, tci, numa_node_id = -2; |
2149 | int maps_sz, num_tc = 1, tc = 0; | |
537c00de | 2150 | struct xps_map *map, *new_map; |
01c5f864 | 2151 | bool active = false; |
537c00de | 2152 | |
184c449f AD |
2153 | if (dev->num_tc) { |
2154 | num_tc = dev->num_tc; | |
2155 | tc = netdev_txq_to_tc(dev, index); | |
2156 | if (tc < 0) | |
2157 | return -EINVAL; | |
2158 | } | |
2159 | ||
2160 | maps_sz = XPS_DEV_MAPS_SIZE(num_tc); | |
2161 | if (maps_sz < L1_CACHE_BYTES) | |
2162 | maps_sz = L1_CACHE_BYTES; | |
2163 | ||
537c00de AD |
2164 | mutex_lock(&xps_map_mutex); |
2165 | ||
2166 | dev_maps = xmap_dereference(dev->xps_maps); | |
2167 | ||
01c5f864 | 2168 | /* allocate memory for queue storage */ |
184c449f | 2169 | for_each_cpu_and(cpu, cpu_online_mask, mask) { |
01c5f864 AD |
2170 | if (!new_dev_maps) |
2171 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2172 | if (!new_dev_maps) { |
2173 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2174 | return -ENOMEM; |
2bb60cb9 | 2175 | } |
01c5f864 | 2176 | |
184c449f AD |
2177 | tci = cpu * num_tc + tc; |
2178 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) : | |
01c5f864 AD |
2179 | NULL; |
2180 | ||
2181 | map = expand_xps_map(map, cpu, index); | |
2182 | if (!map) | |
2183 | goto error; | |
2184 | ||
184c449f | 2185 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); |
01c5f864 AD |
2186 | } |
2187 | ||
2188 | if (!new_dev_maps) | |
2189 | goto out_no_new_maps; | |
2190 | ||
537c00de | 2191 | for_each_possible_cpu(cpu) { |
184c449f AD |
2192 | /* copy maps belonging to foreign traffic classes */ |
2193 | for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) { | |
2194 | /* fill in the new device map from the old device map */ | |
2195 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
2196 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
2197 | } | |
2198 | ||
2199 | /* We need to explicitly update tci as prevous loop | |
2200 | * could break out early if dev_maps is NULL. | |
2201 | */ | |
2202 | tci = cpu * num_tc + tc; | |
2203 | ||
01c5f864 AD |
2204 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
2205 | /* add queue to CPU maps */ | |
2206 | int pos = 0; | |
2207 | ||
184c449f | 2208 | map = xmap_dereference(new_dev_maps->cpu_map[tci]); |
01c5f864 AD |
2209 | while ((pos < map->len) && (map->queues[pos] != index)) |
2210 | pos++; | |
2211 | ||
2212 | if (pos == map->len) | |
2213 | map->queues[map->len++] = index; | |
537c00de | 2214 | #ifdef CONFIG_NUMA |
537c00de AD |
2215 | if (numa_node_id == -2) |
2216 | numa_node_id = cpu_to_node(cpu); | |
2217 | else if (numa_node_id != cpu_to_node(cpu)) | |
2218 | numa_node_id = -1; | |
537c00de | 2219 | #endif |
01c5f864 AD |
2220 | } else if (dev_maps) { |
2221 | /* fill in the new device map from the old device map */ | |
184c449f AD |
2222 | map = xmap_dereference(dev_maps->cpu_map[tci]); |
2223 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
537c00de | 2224 | } |
01c5f864 | 2225 | |
184c449f AD |
2226 | /* copy maps belonging to foreign traffic classes */ |
2227 | for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { | |
2228 | /* fill in the new device map from the old device map */ | |
2229 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
2230 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
2231 | } | |
537c00de AD |
2232 | } |
2233 | ||
01c5f864 AD |
2234 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2235 | ||
537c00de | 2236 | /* Cleanup old maps */ |
184c449f AD |
2237 | if (!dev_maps) |
2238 | goto out_no_old_maps; | |
2239 | ||
2240 | for_each_possible_cpu(cpu) { | |
2241 | for (i = num_tc, tci = cpu * num_tc; i--; tci++) { | |
2242 | new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); | |
2243 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
01c5f864 AD |
2244 | if (map && map != new_map) |
2245 | kfree_rcu(map, rcu); | |
2246 | } | |
537c00de AD |
2247 | } |
2248 | ||
184c449f AD |
2249 | kfree_rcu(dev_maps, rcu); |
2250 | ||
2251 | out_no_old_maps: | |
01c5f864 AD |
2252 | dev_maps = new_dev_maps; |
2253 | active = true; | |
537c00de | 2254 | |
01c5f864 AD |
2255 | out_no_new_maps: |
2256 | /* update Tx queue numa node */ | |
537c00de AD |
2257 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2258 | (numa_node_id >= 0) ? numa_node_id : | |
2259 | NUMA_NO_NODE); | |
2260 | ||
01c5f864 AD |
2261 | if (!dev_maps) |
2262 | goto out_no_maps; | |
2263 | ||
2264 | /* removes queue from unused CPUs */ | |
2265 | for_each_possible_cpu(cpu) { | |
184c449f AD |
2266 | for (i = tc, tci = cpu * num_tc; i--; tci++) |
2267 | active |= remove_xps_queue(dev_maps, tci, index); | |
2268 | if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu)) | |
2269 | active |= remove_xps_queue(dev_maps, tci, index); | |
2270 | for (i = num_tc - tc, tci++; --i; tci++) | |
2271 | active |= remove_xps_queue(dev_maps, tci, index); | |
01c5f864 AD |
2272 | } |
2273 | ||
2274 | /* free map if not active */ | |
2275 | if (!active) { | |
2276 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2277 | kfree_rcu(dev_maps, rcu); | |
2278 | } | |
2279 | ||
2280 | out_no_maps: | |
537c00de AD |
2281 | mutex_unlock(&xps_map_mutex); |
2282 | ||
2283 | return 0; | |
2284 | error: | |
01c5f864 AD |
2285 | /* remove any maps that we added */ |
2286 | for_each_possible_cpu(cpu) { | |
184c449f AD |
2287 | for (i = num_tc, tci = cpu * num_tc; i--; tci++) { |
2288 | new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); | |
2289 | map = dev_maps ? | |
2290 | xmap_dereference(dev_maps->cpu_map[tci]) : | |
2291 | NULL; | |
2292 | if (new_map && new_map != map) | |
2293 | kfree(new_map); | |
2294 | } | |
01c5f864 AD |
2295 | } |
2296 | ||
537c00de AD |
2297 | mutex_unlock(&xps_map_mutex); |
2298 | ||
537c00de AD |
2299 | kfree(new_dev_maps); |
2300 | return -ENOMEM; | |
2301 | } | |
2302 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2303 | ||
2304 | #endif | |
9cf1f6a8 AD |
2305 | void netdev_reset_tc(struct net_device *dev) |
2306 | { | |
6234f874 AD |
2307 | #ifdef CONFIG_XPS |
2308 | netif_reset_xps_queues_gt(dev, 0); | |
2309 | #endif | |
9cf1f6a8 AD |
2310 | dev->num_tc = 0; |
2311 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2312 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2313 | } | |
2314 | EXPORT_SYMBOL(netdev_reset_tc); | |
2315 | ||
2316 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2317 | { | |
2318 | if (tc >= dev->num_tc) | |
2319 | return -EINVAL; | |
2320 | ||
6234f874 AD |
2321 | #ifdef CONFIG_XPS |
2322 | netif_reset_xps_queues(dev, offset, count); | |
2323 | #endif | |
9cf1f6a8 AD |
2324 | dev->tc_to_txq[tc].count = count; |
2325 | dev->tc_to_txq[tc].offset = offset; | |
2326 | return 0; | |
2327 | } | |
2328 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2329 | ||
2330 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2331 | { | |
2332 | if (num_tc > TC_MAX_QUEUE) | |
2333 | return -EINVAL; | |
2334 | ||
6234f874 AD |
2335 | #ifdef CONFIG_XPS |
2336 | netif_reset_xps_queues_gt(dev, 0); | |
2337 | #endif | |
9cf1f6a8 AD |
2338 | dev->num_tc = num_tc; |
2339 | return 0; | |
2340 | } | |
2341 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2342 | ||
f0796d5c JF |
2343 | /* |
2344 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2345 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2346 | */ | |
e6484930 | 2347 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2348 | { |
1d24eb48 TH |
2349 | int rc; |
2350 | ||
e6484930 TH |
2351 | if (txq < 1 || txq > dev->num_tx_queues) |
2352 | return -EINVAL; | |
f0796d5c | 2353 | |
5c56580b BH |
2354 | if (dev->reg_state == NETREG_REGISTERED || |
2355 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2356 | ASSERT_RTNL(); |
2357 | ||
1d24eb48 TH |
2358 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2359 | txq); | |
bf264145 TH |
2360 | if (rc) |
2361 | return rc; | |
2362 | ||
4f57c087 JF |
2363 | if (dev->num_tc) |
2364 | netif_setup_tc(dev, txq); | |
2365 | ||
024e9679 | 2366 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2367 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2368 | #ifdef CONFIG_XPS |
2369 | netif_reset_xps_queues_gt(dev, txq); | |
2370 | #endif | |
2371 | } | |
f0796d5c | 2372 | } |
e6484930 TH |
2373 | |
2374 | dev->real_num_tx_queues = txq; | |
2375 | return 0; | |
f0796d5c JF |
2376 | } |
2377 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2378 | |
a953be53 | 2379 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2380 | /** |
2381 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2382 | * @dev: Network device | |
2383 | * @rxq: Actual number of RX queues | |
2384 | * | |
2385 | * This must be called either with the rtnl_lock held or before | |
2386 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2387 | * negative error code. If called before registration, it always |
2388 | * succeeds. | |
62fe0b40 BH |
2389 | */ |
2390 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2391 | { | |
2392 | int rc; | |
2393 | ||
bd25fa7b TH |
2394 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2395 | return -EINVAL; | |
2396 | ||
62fe0b40 BH |
2397 | if (dev->reg_state == NETREG_REGISTERED) { |
2398 | ASSERT_RTNL(); | |
2399 | ||
62fe0b40 BH |
2400 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2401 | rxq); | |
2402 | if (rc) | |
2403 | return rc; | |
62fe0b40 BH |
2404 | } |
2405 | ||
2406 | dev->real_num_rx_queues = rxq; | |
2407 | return 0; | |
2408 | } | |
2409 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2410 | #endif | |
2411 | ||
2c53040f BH |
2412 | /** |
2413 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2414 | * |
2415 | * This routine should set an upper limit on the number of RSS queues | |
2416 | * used by default by multiqueue devices. | |
2417 | */ | |
a55b138b | 2418 | int netif_get_num_default_rss_queues(void) |
16917b87 | 2419 | { |
40e4e713 HS |
2420 | return is_kdump_kernel() ? |
2421 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
2422 | } |
2423 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2424 | ||
3bcb846c | 2425 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 2426 | { |
def82a1d JP |
2427 | struct softnet_data *sd; |
2428 | unsigned long flags; | |
56079431 | 2429 | |
def82a1d | 2430 | local_irq_save(flags); |
903ceff7 | 2431 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2432 | q->next_sched = NULL; |
2433 | *sd->output_queue_tailp = q; | |
2434 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2435 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2436 | local_irq_restore(flags); | |
2437 | } | |
2438 | ||
2439 | void __netif_schedule(struct Qdisc *q) | |
2440 | { | |
2441 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2442 | __netif_reschedule(q); | |
56079431 DV |
2443 | } |
2444 | EXPORT_SYMBOL(__netif_schedule); | |
2445 | ||
e6247027 ED |
2446 | struct dev_kfree_skb_cb { |
2447 | enum skb_free_reason reason; | |
2448 | }; | |
2449 | ||
2450 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2451 | { |
e6247027 ED |
2452 | return (struct dev_kfree_skb_cb *)skb->cb; |
2453 | } | |
2454 | ||
46e5da40 JF |
2455 | void netif_schedule_queue(struct netdev_queue *txq) |
2456 | { | |
2457 | rcu_read_lock(); | |
2458 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2459 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2460 | ||
2461 | __netif_schedule(q); | |
2462 | } | |
2463 | rcu_read_unlock(); | |
2464 | } | |
2465 | EXPORT_SYMBOL(netif_schedule_queue); | |
2466 | ||
46e5da40 JF |
2467 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
2468 | { | |
2469 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2470 | struct Qdisc *q; | |
2471 | ||
2472 | rcu_read_lock(); | |
2473 | q = rcu_dereference(dev_queue->qdisc); | |
2474 | __netif_schedule(q); | |
2475 | rcu_read_unlock(); | |
2476 | } | |
2477 | } | |
2478 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2479 | ||
e6247027 | 2480 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2481 | { |
e6247027 | 2482 | unsigned long flags; |
56079431 | 2483 | |
9899886d MJ |
2484 | if (unlikely(!skb)) |
2485 | return; | |
2486 | ||
63354797 | 2487 | if (likely(refcount_read(&skb->users) == 1)) { |
e6247027 | 2488 | smp_rmb(); |
63354797 RE |
2489 | refcount_set(&skb->users, 0); |
2490 | } else if (likely(!refcount_dec_and_test(&skb->users))) { | |
e6247027 | 2491 | return; |
bea3348e | 2492 | } |
e6247027 ED |
2493 | get_kfree_skb_cb(skb)->reason = reason; |
2494 | local_irq_save(flags); | |
2495 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2496 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2497 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2498 | local_irq_restore(flags); | |
56079431 | 2499 | } |
e6247027 | 2500 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2501 | |
e6247027 | 2502 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2503 | { |
2504 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2505 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2506 | else |
2507 | dev_kfree_skb(skb); | |
2508 | } | |
e6247027 | 2509 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2510 | |
2511 | ||
bea3348e SH |
2512 | /** |
2513 | * netif_device_detach - mark device as removed | |
2514 | * @dev: network device | |
2515 | * | |
2516 | * Mark device as removed from system and therefore no longer available. | |
2517 | */ | |
56079431 DV |
2518 | void netif_device_detach(struct net_device *dev) |
2519 | { | |
2520 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2521 | netif_running(dev)) { | |
d543103a | 2522 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2523 | } |
2524 | } | |
2525 | EXPORT_SYMBOL(netif_device_detach); | |
2526 | ||
bea3348e SH |
2527 | /** |
2528 | * netif_device_attach - mark device as attached | |
2529 | * @dev: network device | |
2530 | * | |
2531 | * Mark device as attached from system and restart if needed. | |
2532 | */ | |
56079431 DV |
2533 | void netif_device_attach(struct net_device *dev) |
2534 | { | |
2535 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2536 | netif_running(dev)) { | |
d543103a | 2537 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2538 | __netdev_watchdog_up(dev); |
56079431 DV |
2539 | } |
2540 | } | |
2541 | EXPORT_SYMBOL(netif_device_attach); | |
2542 | ||
5605c762 JP |
2543 | /* |
2544 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2545 | * to be used as a distribution range. | |
2546 | */ | |
2547 | u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, | |
2548 | unsigned int num_tx_queues) | |
2549 | { | |
2550 | u32 hash; | |
2551 | u16 qoffset = 0; | |
2552 | u16 qcount = num_tx_queues; | |
2553 | ||
2554 | if (skb_rx_queue_recorded(skb)) { | |
2555 | hash = skb_get_rx_queue(skb); | |
2556 | while (unlikely(hash >= num_tx_queues)) | |
2557 | hash -= num_tx_queues; | |
2558 | return hash; | |
2559 | } | |
2560 | ||
2561 | if (dev->num_tc) { | |
2562 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
f4563a75 | 2563 | |
5605c762 JP |
2564 | qoffset = dev->tc_to_txq[tc].offset; |
2565 | qcount = dev->tc_to_txq[tc].count; | |
2566 | } | |
2567 | ||
2568 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
2569 | } | |
2570 | EXPORT_SYMBOL(__skb_tx_hash); | |
2571 | ||
36c92474 BH |
2572 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2573 | { | |
84d15ae5 | 2574 | static const netdev_features_t null_features; |
36c92474 | 2575 | struct net_device *dev = skb->dev; |
88ad4175 | 2576 | const char *name = ""; |
36c92474 | 2577 | |
c846ad9b BG |
2578 | if (!net_ratelimit()) |
2579 | return; | |
2580 | ||
88ad4175 BM |
2581 | if (dev) { |
2582 | if (dev->dev.parent) | |
2583 | name = dev_driver_string(dev->dev.parent); | |
2584 | else | |
2585 | name = netdev_name(dev); | |
2586 | } | |
36c92474 BH |
2587 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " |
2588 | "gso_type=%d ip_summed=%d\n", | |
88ad4175 | 2589 | name, dev ? &dev->features : &null_features, |
65e9d2fa | 2590 | skb->sk ? &skb->sk->sk_route_caps : &null_features, |
36c92474 BH |
2591 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2592 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2593 | } | |
2594 | ||
1da177e4 LT |
2595 | /* |
2596 | * Invalidate hardware checksum when packet is to be mangled, and | |
2597 | * complete checksum manually on outgoing path. | |
2598 | */ | |
84fa7933 | 2599 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2600 | { |
d3bc23e7 | 2601 | __wsum csum; |
663ead3b | 2602 | int ret = 0, offset; |
1da177e4 | 2603 | |
84fa7933 | 2604 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2605 | goto out_set_summed; |
2606 | ||
2607 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2608 | skb_warn_bad_offload(skb); |
2609 | return -EINVAL; | |
1da177e4 LT |
2610 | } |
2611 | ||
cef401de ED |
2612 | /* Before computing a checksum, we should make sure no frag could |
2613 | * be modified by an external entity : checksum could be wrong. | |
2614 | */ | |
2615 | if (skb_has_shared_frag(skb)) { | |
2616 | ret = __skb_linearize(skb); | |
2617 | if (ret) | |
2618 | goto out; | |
2619 | } | |
2620 | ||
55508d60 | 2621 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2622 | BUG_ON(offset >= skb_headlen(skb)); |
2623 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2624 | ||
2625 | offset += skb->csum_offset; | |
2626 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2627 | ||
2628 | if (skb_cloned(skb) && | |
2629 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2630 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2631 | if (ret) | |
2632 | goto out; | |
2633 | } | |
2634 | ||
4f2e4ad5 | 2635 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 2636 | out_set_summed: |
1da177e4 | 2637 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2638 | out: |
1da177e4 LT |
2639 | return ret; |
2640 | } | |
d1b19dff | 2641 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2642 | |
b72b5bf6 DC |
2643 | int skb_crc32c_csum_help(struct sk_buff *skb) |
2644 | { | |
2645 | __le32 crc32c_csum; | |
2646 | int ret = 0, offset, start; | |
2647 | ||
2648 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
2649 | goto out; | |
2650 | ||
2651 | if (unlikely(skb_is_gso(skb))) | |
2652 | goto out; | |
2653 | ||
2654 | /* Before computing a checksum, we should make sure no frag could | |
2655 | * be modified by an external entity : checksum could be wrong. | |
2656 | */ | |
2657 | if (unlikely(skb_has_shared_frag(skb))) { | |
2658 | ret = __skb_linearize(skb); | |
2659 | if (ret) | |
2660 | goto out; | |
2661 | } | |
2662 | start = skb_checksum_start_offset(skb); | |
2663 | offset = start + offsetof(struct sctphdr, checksum); | |
2664 | if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { | |
2665 | ret = -EINVAL; | |
2666 | goto out; | |
2667 | } | |
2668 | if (skb_cloned(skb) && | |
2669 | !skb_clone_writable(skb, offset + sizeof(__le32))) { | |
2670 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); | |
2671 | if (ret) | |
2672 | goto out; | |
2673 | } | |
2674 | crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, | |
2675 | skb->len - start, ~(__u32)0, | |
2676 | crc32c_csum_stub)); | |
2677 | *(__le32 *)(skb->data + offset) = crc32c_csum; | |
2678 | skb->ip_summed = CHECKSUM_NONE; | |
dba00306 | 2679 | skb->csum_not_inet = 0; |
b72b5bf6 DC |
2680 | out: |
2681 | return ret; | |
2682 | } | |
2683 | ||
53d6471c | 2684 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2685 | { |
252e3346 | 2686 | __be16 type = skb->protocol; |
f6a78bfc | 2687 | |
19acc327 PS |
2688 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2689 | if (type == htons(ETH_P_TEB)) { | |
2690 | struct ethhdr *eth; | |
2691 | ||
2692 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2693 | return 0; | |
2694 | ||
2695 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2696 | type = eth->h_proto; | |
2697 | } | |
2698 | ||
d4bcef3f | 2699 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
2700 | } |
2701 | ||
2702 | /** | |
2703 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2704 | * @skb: buffer to segment | |
2705 | * @features: features for the output path (see dev->features) | |
2706 | */ | |
2707 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2708 | netdev_features_t features) | |
2709 | { | |
2710 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2711 | struct packet_offload *ptype; | |
53d6471c VY |
2712 | int vlan_depth = skb->mac_len; |
2713 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2714 | |
2715 | if (unlikely(!type)) | |
2716 | return ERR_PTR(-EINVAL); | |
2717 | ||
53d6471c | 2718 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2719 | |
2720 | rcu_read_lock(); | |
22061d80 | 2721 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2722 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2723 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2724 | break; |
2725 | } | |
2726 | } | |
2727 | rcu_read_unlock(); | |
2728 | ||
98e399f8 | 2729 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2730 | |
f6a78bfc HX |
2731 | return segs; |
2732 | } | |
05e8ef4a PS |
2733 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2734 | ||
2735 | ||
2736 | /* openvswitch calls this on rx path, so we need a different check. | |
2737 | */ | |
2738 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2739 | { | |
2740 | if (tx_path) | |
6e7bc478 | 2741 | return skb->ip_summed != CHECKSUM_PARTIAL && |
8d63bee6 | 2742 | skb->ip_summed != CHECKSUM_UNNECESSARY; |
6e7bc478 ED |
2743 | |
2744 | return skb->ip_summed == CHECKSUM_NONE; | |
05e8ef4a PS |
2745 | } |
2746 | ||
2747 | /** | |
2748 | * __skb_gso_segment - Perform segmentation on skb. | |
2749 | * @skb: buffer to segment | |
2750 | * @features: features for the output path (see dev->features) | |
2751 | * @tx_path: whether it is called in TX path | |
2752 | * | |
2753 | * This function segments the given skb and returns a list of segments. | |
2754 | * | |
2755 | * It may return NULL if the skb requires no segmentation. This is | |
2756 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 KK |
2757 | * |
2758 | * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. | |
05e8ef4a PS |
2759 | */ |
2760 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2761 | netdev_features_t features, bool tx_path) | |
2762 | { | |
b2504a5d ED |
2763 | struct sk_buff *segs; |
2764 | ||
05e8ef4a PS |
2765 | if (unlikely(skb_needs_check(skb, tx_path))) { |
2766 | int err; | |
2767 | ||
b2504a5d | 2768 | /* We're going to init ->check field in TCP or UDP header */ |
a40e0a66 | 2769 | err = skb_cow_head(skb, 0); |
2770 | if (err < 0) | |
05e8ef4a PS |
2771 | return ERR_PTR(err); |
2772 | } | |
2773 | ||
802ab55a AD |
2774 | /* Only report GSO partial support if it will enable us to |
2775 | * support segmentation on this frame without needing additional | |
2776 | * work. | |
2777 | */ | |
2778 | if (features & NETIF_F_GSO_PARTIAL) { | |
2779 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
2780 | struct net_device *dev = skb->dev; | |
2781 | ||
2782 | partial_features |= dev->features & dev->gso_partial_features; | |
2783 | if (!skb_gso_ok(skb, features | partial_features)) | |
2784 | features &= ~NETIF_F_GSO_PARTIAL; | |
2785 | } | |
2786 | ||
9207f9d4 KK |
2787 | BUILD_BUG_ON(SKB_SGO_CB_OFFSET + |
2788 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); | |
2789 | ||
68c33163 | 2790 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2791 | SKB_GSO_CB(skb)->encap_level = 0; |
2792 | ||
05e8ef4a PS |
2793 | skb_reset_mac_header(skb); |
2794 | skb_reset_mac_len(skb); | |
2795 | ||
b2504a5d ED |
2796 | segs = skb_mac_gso_segment(skb, features); |
2797 | ||
2798 | if (unlikely(skb_needs_check(skb, tx_path))) | |
2799 | skb_warn_bad_offload(skb); | |
2800 | ||
2801 | return segs; | |
05e8ef4a | 2802 | } |
12b0004d | 2803 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2804 | |
fb286bb2 HX |
2805 | /* Take action when hardware reception checksum errors are detected. */ |
2806 | #ifdef CONFIG_BUG | |
2807 | void netdev_rx_csum_fault(struct net_device *dev) | |
2808 | { | |
2809 | if (net_ratelimit()) { | |
7b6cd1ce | 2810 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2811 | dump_stack(); |
2812 | } | |
2813 | } | |
2814 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2815 | #endif | |
2816 | ||
1da177e4 LT |
2817 | /* Actually, we should eliminate this check as soon as we know, that: |
2818 | * 1. IOMMU is present and allows to map all the memory. | |
2819 | * 2. No high memory really exists on this machine. | |
2820 | */ | |
2821 | ||
c1e756bf | 2822 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2823 | { |
3d3a8533 | 2824 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2825 | int i; |
f4563a75 | 2826 | |
5acbbd42 | 2827 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2828 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2829 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 2830 | |
ea2ab693 | 2831 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 2832 | return 1; |
ea2ab693 | 2833 | } |
5acbbd42 | 2834 | } |
1da177e4 | 2835 | |
5acbbd42 FT |
2836 | if (PCI_DMA_BUS_IS_PHYS) { |
2837 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2838 | |
9092c658 ED |
2839 | if (!pdev) |
2840 | return 0; | |
5acbbd42 | 2841 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2842 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2843 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
f4563a75 | 2844 | |
5acbbd42 FT |
2845 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2846 | return 1; | |
2847 | } | |
2848 | } | |
3d3a8533 | 2849 | #endif |
1da177e4 LT |
2850 | return 0; |
2851 | } | |
1da177e4 | 2852 | |
3b392ddb SH |
2853 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2854 | * instead of standard features for the netdev. | |
2855 | */ | |
d0edc7bf | 2856 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
2857 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
2858 | netdev_features_t features, | |
2859 | __be16 type) | |
2860 | { | |
25cd9ba0 | 2861 | if (eth_p_mpls(type)) |
3b392ddb SH |
2862 | features &= skb->dev->mpls_features; |
2863 | ||
2864 | return features; | |
2865 | } | |
2866 | #else | |
2867 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2868 | netdev_features_t features, | |
2869 | __be16 type) | |
2870 | { | |
2871 | return features; | |
2872 | } | |
2873 | #endif | |
2874 | ||
c8f44aff | 2875 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2876 | netdev_features_t features) |
f01a5236 | 2877 | { |
53d6471c | 2878 | int tmp; |
3b392ddb SH |
2879 | __be16 type; |
2880 | ||
2881 | type = skb_network_protocol(skb, &tmp); | |
2882 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2883 | |
c0d680e5 | 2884 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2885 | !can_checksum_protocol(features, type)) { |
996e8021 | 2886 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 2887 | } |
7be2c82c ED |
2888 | if (illegal_highdma(skb->dev, skb)) |
2889 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
2890 | |
2891 | return features; | |
2892 | } | |
2893 | ||
e38f3025 TM |
2894 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
2895 | struct net_device *dev, | |
2896 | netdev_features_t features) | |
2897 | { | |
2898 | return features; | |
2899 | } | |
2900 | EXPORT_SYMBOL(passthru_features_check); | |
2901 | ||
8cb65d00 TM |
2902 | static netdev_features_t dflt_features_check(const struct sk_buff *skb, |
2903 | struct net_device *dev, | |
2904 | netdev_features_t features) | |
2905 | { | |
2906 | return vlan_features_check(skb, features); | |
2907 | } | |
2908 | ||
cbc53e08 AD |
2909 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
2910 | struct net_device *dev, | |
2911 | netdev_features_t features) | |
2912 | { | |
2913 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
2914 | ||
2915 | if (gso_segs > dev->gso_max_segs) | |
2916 | return features & ~NETIF_F_GSO_MASK; | |
2917 | ||
802ab55a AD |
2918 | /* Support for GSO partial features requires software |
2919 | * intervention before we can actually process the packets | |
2920 | * so we need to strip support for any partial features now | |
2921 | * and we can pull them back in after we have partially | |
2922 | * segmented the frame. | |
2923 | */ | |
2924 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
2925 | features &= ~dev->gso_partial_features; | |
2926 | ||
2927 | /* Make sure to clear the IPv4 ID mangling feature if the | |
2928 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
2929 | */ |
2930 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
2931 | struct iphdr *iph = skb->encapsulation ? | |
2932 | inner_ip_hdr(skb) : ip_hdr(skb); | |
2933 | ||
2934 | if (!(iph->frag_off & htons(IP_DF))) | |
2935 | features &= ~NETIF_F_TSO_MANGLEID; | |
2936 | } | |
2937 | ||
2938 | return features; | |
2939 | } | |
2940 | ||
c1e756bf | 2941 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 2942 | { |
5f35227e | 2943 | struct net_device *dev = skb->dev; |
fcbeb976 | 2944 | netdev_features_t features = dev->features; |
58e998c6 | 2945 | |
cbc53e08 AD |
2946 | if (skb_is_gso(skb)) |
2947 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 2948 | |
5f35227e JG |
2949 | /* If encapsulation offload request, verify we are testing |
2950 | * hardware encapsulation features instead of standard | |
2951 | * features for the netdev | |
2952 | */ | |
2953 | if (skb->encapsulation) | |
2954 | features &= dev->hw_enc_features; | |
2955 | ||
f5a7fb88 TM |
2956 | if (skb_vlan_tagged(skb)) |
2957 | features = netdev_intersect_features(features, | |
2958 | dev->vlan_features | | |
2959 | NETIF_F_HW_VLAN_CTAG_TX | | |
2960 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2961 | |
5f35227e JG |
2962 | if (dev->netdev_ops->ndo_features_check) |
2963 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
2964 | features); | |
8cb65d00 TM |
2965 | else |
2966 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 2967 | |
c1e756bf | 2968 | return harmonize_features(skb, features); |
58e998c6 | 2969 | } |
c1e756bf | 2970 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2971 | |
2ea25513 | 2972 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2973 | struct netdev_queue *txq, bool more) |
f6a78bfc | 2974 | { |
2ea25513 DM |
2975 | unsigned int len; |
2976 | int rc; | |
00829823 | 2977 | |
7866a621 | 2978 | if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) |
2ea25513 | 2979 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 2980 | |
2ea25513 DM |
2981 | len = skb->len; |
2982 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 2983 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 2984 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 2985 | |
2ea25513 DM |
2986 | return rc; |
2987 | } | |
7b9c6090 | 2988 | |
8dcda22a DM |
2989 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
2990 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
2991 | { |
2992 | struct sk_buff *skb = first; | |
2993 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 2994 | |
7f2e870f DM |
2995 | while (skb) { |
2996 | struct sk_buff *next = skb->next; | |
fc70fb64 | 2997 | |
7f2e870f | 2998 | skb->next = NULL; |
95f6b3dd | 2999 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
3000 | if (unlikely(!dev_xmit_complete(rc))) { |
3001 | skb->next = next; | |
3002 | goto out; | |
3003 | } | |
6afff0ca | 3004 | |
7f2e870f DM |
3005 | skb = next; |
3006 | if (netif_xmit_stopped(txq) && skb) { | |
3007 | rc = NETDEV_TX_BUSY; | |
3008 | break; | |
9ccb8975 | 3009 | } |
7f2e870f | 3010 | } |
9ccb8975 | 3011 | |
7f2e870f DM |
3012 | out: |
3013 | *ret = rc; | |
3014 | return skb; | |
3015 | } | |
b40863c6 | 3016 | |
1ff0dc94 ED |
3017 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
3018 | netdev_features_t features) | |
f6a78bfc | 3019 | { |
df8a39de | 3020 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
3021 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
3022 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
3023 | return skb; |
3024 | } | |
f6a78bfc | 3025 | |
43c26a1a DC |
3026 | int skb_csum_hwoffload_help(struct sk_buff *skb, |
3027 | const netdev_features_t features) | |
3028 | { | |
3029 | if (unlikely(skb->csum_not_inet)) | |
3030 | return !!(features & NETIF_F_SCTP_CRC) ? 0 : | |
3031 | skb_crc32c_csum_help(skb); | |
3032 | ||
3033 | return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb); | |
3034 | } | |
3035 | EXPORT_SYMBOL(skb_csum_hwoffload_help); | |
3036 | ||
55a93b3e | 3037 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
3038 | { |
3039 | netdev_features_t features; | |
f6a78bfc | 3040 | |
eae3f88e DM |
3041 | features = netif_skb_features(skb); |
3042 | skb = validate_xmit_vlan(skb, features); | |
3043 | if (unlikely(!skb)) | |
3044 | goto out_null; | |
7b9c6090 | 3045 | |
8b86a61d | 3046 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
3047 | struct sk_buff *segs; |
3048 | ||
3049 | segs = skb_gso_segment(skb, features); | |
cecda693 | 3050 | if (IS_ERR(segs)) { |
af6dabc9 | 3051 | goto out_kfree_skb; |
cecda693 JW |
3052 | } else if (segs) { |
3053 | consume_skb(skb); | |
3054 | skb = segs; | |
f6a78bfc | 3055 | } |
eae3f88e DM |
3056 | } else { |
3057 | if (skb_needs_linearize(skb, features) && | |
3058 | __skb_linearize(skb)) | |
3059 | goto out_kfree_skb; | |
4ec93edb | 3060 | |
f6e27114 SK |
3061 | if (validate_xmit_xfrm(skb, features)) |
3062 | goto out_kfree_skb; | |
3063 | ||
eae3f88e DM |
3064 | /* If packet is not checksummed and device does not |
3065 | * support checksumming for this protocol, complete | |
3066 | * checksumming here. | |
3067 | */ | |
3068 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
3069 | if (skb->encapsulation) | |
3070 | skb_set_inner_transport_header(skb, | |
3071 | skb_checksum_start_offset(skb)); | |
3072 | else | |
3073 | skb_set_transport_header(skb, | |
3074 | skb_checksum_start_offset(skb)); | |
43c26a1a | 3075 | if (skb_csum_hwoffload_help(skb, features)) |
eae3f88e | 3076 | goto out_kfree_skb; |
7b9c6090 | 3077 | } |
0c772159 | 3078 | } |
7b9c6090 | 3079 | |
eae3f88e | 3080 | return skb; |
fc70fb64 | 3081 | |
f6a78bfc HX |
3082 | out_kfree_skb: |
3083 | kfree_skb(skb); | |
eae3f88e | 3084 | out_null: |
d21fd63e | 3085 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3086 | return NULL; |
3087 | } | |
6afff0ca | 3088 | |
55a93b3e ED |
3089 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
3090 | { | |
3091 | struct sk_buff *next, *head = NULL, *tail; | |
3092 | ||
bec3cfdc | 3093 | for (; skb != NULL; skb = next) { |
55a93b3e ED |
3094 | next = skb->next; |
3095 | skb->next = NULL; | |
bec3cfdc ED |
3096 | |
3097 | /* in case skb wont be segmented, point to itself */ | |
3098 | skb->prev = skb; | |
3099 | ||
55a93b3e | 3100 | skb = validate_xmit_skb(skb, dev); |
bec3cfdc ED |
3101 | if (!skb) |
3102 | continue; | |
55a93b3e | 3103 | |
bec3cfdc ED |
3104 | if (!head) |
3105 | head = skb; | |
3106 | else | |
3107 | tail->next = skb; | |
3108 | /* If skb was segmented, skb->prev points to | |
3109 | * the last segment. If not, it still contains skb. | |
3110 | */ | |
3111 | tail = skb->prev; | |
55a93b3e ED |
3112 | } |
3113 | return head; | |
f6a78bfc | 3114 | } |
104ba78c | 3115 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3116 | |
1def9238 ED |
3117 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3118 | { | |
3119 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3120 | ||
3121 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3122 | ||
3123 | /* To get more precise estimation of bytes sent on wire, | |
3124 | * we add to pkt_len the headers size of all segments | |
3125 | */ | |
3126 | if (shinfo->gso_size) { | |
757b8b1d | 3127 | unsigned int hdr_len; |
15e5a030 | 3128 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3129 | |
757b8b1d ED |
3130 | /* mac layer + network layer */ |
3131 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3132 | ||
3133 | /* + transport layer */ | |
1def9238 ED |
3134 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
3135 | hdr_len += tcp_hdrlen(skb); | |
3136 | else | |
3137 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
3138 | |
3139 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3140 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3141 | shinfo->gso_size); | |
3142 | ||
3143 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3144 | } |
3145 | } | |
3146 | ||
bbd8a0d3 KK |
3147 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3148 | struct net_device *dev, | |
3149 | struct netdev_queue *txq) | |
3150 | { | |
3151 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3152 | struct sk_buff *to_free = NULL; |
a2da570d | 3153 | bool contended; |
bbd8a0d3 KK |
3154 | int rc; |
3155 | ||
a2da570d | 3156 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
3157 | /* |
3158 | * Heuristic to force contended enqueues to serialize on a | |
3159 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3160 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3161 | * often and dequeue packets faster. |
79640a4c | 3162 | */ |
a2da570d | 3163 | contended = qdisc_is_running(q); |
79640a4c ED |
3164 | if (unlikely(contended)) |
3165 | spin_lock(&q->busylock); | |
3166 | ||
bbd8a0d3 KK |
3167 | spin_lock(root_lock); |
3168 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3169 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3170 | rc = NET_XMIT_DROP; |
3171 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3172 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3173 | /* |
3174 | * This is a work-conserving queue; there are no old skbs | |
3175 | * waiting to be sent out; and the qdisc is not running - | |
3176 | * xmit the skb directly. | |
3177 | */ | |
bfe0d029 | 3178 | |
bfe0d029 ED |
3179 | qdisc_bstats_update(q, skb); |
3180 | ||
55a93b3e | 3181 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3182 | if (unlikely(contended)) { |
3183 | spin_unlock(&q->busylock); | |
3184 | contended = false; | |
3185 | } | |
bbd8a0d3 | 3186 | __qdisc_run(q); |
79640a4c | 3187 | } else |
bc135b23 | 3188 | qdisc_run_end(q); |
bbd8a0d3 KK |
3189 | |
3190 | rc = NET_XMIT_SUCCESS; | |
3191 | } else { | |
520ac30f | 3192 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3193 | if (qdisc_run_begin(q)) { |
3194 | if (unlikely(contended)) { | |
3195 | spin_unlock(&q->busylock); | |
3196 | contended = false; | |
3197 | } | |
3198 | __qdisc_run(q); | |
3199 | } | |
bbd8a0d3 KK |
3200 | } |
3201 | spin_unlock(root_lock); | |
520ac30f ED |
3202 | if (unlikely(to_free)) |
3203 | kfree_skb_list(to_free); | |
79640a4c ED |
3204 | if (unlikely(contended)) |
3205 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3206 | return rc; |
3207 | } | |
3208 | ||
86f8515f | 3209 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3210 | static void skb_update_prio(struct sk_buff *skb) |
3211 | { | |
6977a79d | 3212 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 3213 | |
91c68ce2 | 3214 | if (!skb->priority && skb->sk && map) { |
2a56a1fe TH |
3215 | unsigned int prioidx = |
3216 | sock_cgroup_prioidx(&skb->sk->sk_cgrp_data); | |
91c68ce2 ED |
3217 | |
3218 | if (prioidx < map->priomap_len) | |
3219 | skb->priority = map->priomap[prioidx]; | |
3220 | } | |
5bc1421e NH |
3221 | } |
3222 | #else | |
3223 | #define skb_update_prio(skb) | |
3224 | #endif | |
3225 | ||
f60e5990 | 3226 | DEFINE_PER_CPU(int, xmit_recursion); |
3227 | EXPORT_SYMBOL(xmit_recursion); | |
3228 | ||
95603e22 MM |
3229 | /** |
3230 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3231 | * @net: network namespace this loopback is happening in |
3232 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3233 | * @skb: buffer to transmit |
3234 | */ | |
0c4b51f0 | 3235 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3236 | { |
3237 | skb_reset_mac_header(skb); | |
3238 | __skb_pull(skb, skb_network_offset(skb)); | |
3239 | skb->pkt_type = PACKET_LOOPBACK; | |
3240 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3241 | WARN_ON(!skb_dst(skb)); | |
3242 | skb_dst_force(skb); | |
3243 | netif_rx_ni(skb); | |
3244 | return 0; | |
3245 | } | |
3246 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3247 | ||
1f211a1b DB |
3248 | #ifdef CONFIG_NET_EGRESS |
3249 | static struct sk_buff * | |
3250 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3251 | { | |
3252 | struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list); | |
3253 | struct tcf_result cl_res; | |
3254 | ||
3255 | if (!cl) | |
3256 | return skb; | |
3257 | ||
8dc07fdb | 3258 | /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ |
1f211a1b DB |
3259 | qdisc_bstats_cpu_update(cl->q, skb); |
3260 | ||
87d83093 | 3261 | switch (tcf_classify(skb, cl, &cl_res, false)) { |
1f211a1b DB |
3262 | case TC_ACT_OK: |
3263 | case TC_ACT_RECLASSIFY: | |
3264 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3265 | break; | |
3266 | case TC_ACT_SHOT: | |
3267 | qdisc_qstats_cpu_drop(cl->q); | |
3268 | *ret = NET_XMIT_DROP; | |
7e2c3aea DB |
3269 | kfree_skb(skb); |
3270 | return NULL; | |
1f211a1b DB |
3271 | case TC_ACT_STOLEN: |
3272 | case TC_ACT_QUEUED: | |
e25ea21f | 3273 | case TC_ACT_TRAP: |
1f211a1b | 3274 | *ret = NET_XMIT_SUCCESS; |
7e2c3aea | 3275 | consume_skb(skb); |
1f211a1b DB |
3276 | return NULL; |
3277 | case TC_ACT_REDIRECT: | |
3278 | /* No need to push/pop skb's mac_header here on egress! */ | |
3279 | skb_do_redirect(skb); | |
3280 | *ret = NET_XMIT_SUCCESS; | |
3281 | return NULL; | |
3282 | default: | |
3283 | break; | |
3284 | } | |
3285 | ||
3286 | return skb; | |
3287 | } | |
3288 | #endif /* CONFIG_NET_EGRESS */ | |
3289 | ||
638b2a69 JP |
3290 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
3291 | { | |
3292 | #ifdef CONFIG_XPS | |
3293 | struct xps_dev_maps *dev_maps; | |
3294 | struct xps_map *map; | |
3295 | int queue_index = -1; | |
3296 | ||
3297 | rcu_read_lock(); | |
3298 | dev_maps = rcu_dereference(dev->xps_maps); | |
3299 | if (dev_maps) { | |
184c449f AD |
3300 | unsigned int tci = skb->sender_cpu - 1; |
3301 | ||
3302 | if (dev->num_tc) { | |
3303 | tci *= dev->num_tc; | |
3304 | tci += netdev_get_prio_tc_map(dev, skb->priority); | |
3305 | } | |
3306 | ||
3307 | map = rcu_dereference(dev_maps->cpu_map[tci]); | |
638b2a69 JP |
3308 | if (map) { |
3309 | if (map->len == 1) | |
3310 | queue_index = map->queues[0]; | |
3311 | else | |
3312 | queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), | |
3313 | map->len)]; | |
3314 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3315 | queue_index = -1; | |
3316 | } | |
3317 | } | |
3318 | rcu_read_unlock(); | |
3319 | ||
3320 | return queue_index; | |
3321 | #else | |
3322 | return -1; | |
3323 | #endif | |
3324 | } | |
3325 | ||
3326 | static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) | |
3327 | { | |
3328 | struct sock *sk = skb->sk; | |
3329 | int queue_index = sk_tx_queue_get(sk); | |
3330 | ||
3331 | if (queue_index < 0 || skb->ooo_okay || | |
3332 | queue_index >= dev->real_num_tx_queues) { | |
3333 | int new_index = get_xps_queue(dev, skb); | |
f4563a75 | 3334 | |
638b2a69 JP |
3335 | if (new_index < 0) |
3336 | new_index = skb_tx_hash(dev, skb); | |
3337 | ||
3338 | if (queue_index != new_index && sk && | |
004a5d01 | 3339 | sk_fullsock(sk) && |
638b2a69 JP |
3340 | rcu_access_pointer(sk->sk_dst_cache)) |
3341 | sk_tx_queue_set(sk, new_index); | |
3342 | ||
3343 | queue_index = new_index; | |
3344 | } | |
3345 | ||
3346 | return queue_index; | |
3347 | } | |
3348 | ||
3349 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, | |
3350 | struct sk_buff *skb, | |
3351 | void *accel_priv) | |
3352 | { | |
3353 | int queue_index = 0; | |
3354 | ||
3355 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
3356 | u32 sender_cpu = skb->sender_cpu - 1; |
3357 | ||
3358 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
3359 | skb->sender_cpu = raw_smp_processor_id() + 1; |
3360 | #endif | |
3361 | ||
3362 | if (dev->real_num_tx_queues != 1) { | |
3363 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 3364 | |
638b2a69 JP |
3365 | if (ops->ndo_select_queue) |
3366 | queue_index = ops->ndo_select_queue(dev, skb, accel_priv, | |
3367 | __netdev_pick_tx); | |
3368 | else | |
3369 | queue_index = __netdev_pick_tx(dev, skb); | |
3370 | ||
3371 | if (!accel_priv) | |
3372 | queue_index = netdev_cap_txqueue(dev, queue_index); | |
3373 | } | |
3374 | ||
3375 | skb_set_queue_mapping(skb, queue_index); | |
3376 | return netdev_get_tx_queue(dev, queue_index); | |
3377 | } | |
3378 | ||
d29f749e | 3379 | /** |
9d08dd3d | 3380 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3381 | * @skb: buffer to transmit |
9d08dd3d | 3382 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
3383 | * |
3384 | * Queue a buffer for transmission to a network device. The caller must | |
3385 | * have set the device and priority and built the buffer before calling | |
3386 | * this function. The function can be called from an interrupt. | |
3387 | * | |
3388 | * A negative errno code is returned on a failure. A success does not | |
3389 | * guarantee the frame will be transmitted as it may be dropped due | |
3390 | * to congestion or traffic shaping. | |
3391 | * | |
3392 | * ----------------------------------------------------------------------------------- | |
3393 | * I notice this method can also return errors from the queue disciplines, | |
3394 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3395 | * be positive. | |
3396 | * | |
3397 | * Regardless of the return value, the skb is consumed, so it is currently | |
3398 | * difficult to retry a send to this method. (You can bump the ref count | |
3399 | * before sending to hold a reference for retry if you are careful.) | |
3400 | * | |
3401 | * When calling this method, interrupts MUST be enabled. This is because | |
3402 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3403 | * --BLG | |
3404 | */ | |
0a59f3a9 | 3405 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
3406 | { |
3407 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3408 | struct netdev_queue *txq; |
1da177e4 LT |
3409 | struct Qdisc *q; |
3410 | int rc = -ENOMEM; | |
3411 | ||
6d1ccff6 ED |
3412 | skb_reset_mac_header(skb); |
3413 | ||
e7fd2885 WB |
3414 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3415 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3416 | ||
4ec93edb YH |
3417 | /* Disable soft irqs for various locks below. Also |
3418 | * stops preemption for RCU. | |
1da177e4 | 3419 | */ |
4ec93edb | 3420 | rcu_read_lock_bh(); |
1da177e4 | 3421 | |
5bc1421e NH |
3422 | skb_update_prio(skb); |
3423 | ||
1f211a1b DB |
3424 | qdisc_pkt_len_init(skb); |
3425 | #ifdef CONFIG_NET_CLS_ACT | |
8dc07fdb | 3426 | skb->tc_at_ingress = 0; |
1f211a1b DB |
3427 | # ifdef CONFIG_NET_EGRESS |
3428 | if (static_key_false(&egress_needed)) { | |
3429 | skb = sch_handle_egress(skb, &rc, dev); | |
3430 | if (!skb) | |
3431 | goto out; | |
3432 | } | |
3433 | # endif | |
3434 | #endif | |
02875878 ED |
3435 | /* If device/qdisc don't need skb->dst, release it right now while |
3436 | * its hot in this cpu cache. | |
3437 | */ | |
3438 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
3439 | skb_dst_drop(skb); | |
3440 | else | |
3441 | skb_dst_force(skb); | |
3442 | ||
f663dd9a | 3443 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 3444 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 3445 | |
cf66ba58 | 3446 | trace_net_dev_queue(skb); |
1da177e4 | 3447 | if (q->enqueue) { |
bbd8a0d3 | 3448 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 3449 | goto out; |
1da177e4 LT |
3450 | } |
3451 | ||
3452 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 3453 | * loopback, all the sorts of tunnels... |
1da177e4 | 3454 | |
eb13da1a | 3455 | * Really, it is unlikely that netif_tx_lock protection is necessary |
3456 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
3457 | * counters.) | |
3458 | * However, it is possible, that they rely on protection | |
3459 | * made by us here. | |
1da177e4 | 3460 | |
eb13da1a | 3461 | * Check this and shot the lock. It is not prone from deadlocks. |
3462 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
3463 | */ |
3464 | if (dev->flags & IFF_UP) { | |
3465 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
3466 | ||
c773e847 | 3467 | if (txq->xmit_lock_owner != cpu) { |
a70b506e DB |
3468 | if (unlikely(__this_cpu_read(xmit_recursion) > |
3469 | XMIT_RECURSION_LIMIT)) | |
745e20f1 ED |
3470 | goto recursion_alert; |
3471 | ||
1f59533f JDB |
3472 | skb = validate_xmit_skb(skb, dev); |
3473 | if (!skb) | |
d21fd63e | 3474 | goto out; |
1f59533f | 3475 | |
c773e847 | 3476 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 3477 | |
73466498 | 3478 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 3479 | __this_cpu_inc(xmit_recursion); |
ce93718f | 3480 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 3481 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 3482 | if (dev_xmit_complete(rc)) { |
c773e847 | 3483 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
3484 | goto out; |
3485 | } | |
3486 | } | |
c773e847 | 3487 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
3488 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
3489 | dev->name); | |
1da177e4 LT |
3490 | } else { |
3491 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
3492 | * unfortunately |
3493 | */ | |
3494 | recursion_alert: | |
e87cc472 JP |
3495 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
3496 | dev->name); | |
1da177e4 LT |
3497 | } |
3498 | } | |
3499 | ||
3500 | rc = -ENETDOWN; | |
d4828d85 | 3501 | rcu_read_unlock_bh(); |
1da177e4 | 3502 | |
015f0688 | 3503 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 3504 | kfree_skb_list(skb); |
1da177e4 LT |
3505 | return rc; |
3506 | out: | |
d4828d85 | 3507 | rcu_read_unlock_bh(); |
1da177e4 LT |
3508 | return rc; |
3509 | } | |
f663dd9a | 3510 | |
2b4aa3ce | 3511 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
3512 | { |
3513 | return __dev_queue_xmit(skb, NULL); | |
3514 | } | |
2b4aa3ce | 3515 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3516 | |
f663dd9a JW |
3517 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3518 | { | |
3519 | return __dev_queue_xmit(skb, accel_priv); | |
3520 | } | |
3521 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3522 | ||
1da177e4 | 3523 | |
eb13da1a | 3524 | /************************************************************************* |
3525 | * Receiver routines | |
3526 | *************************************************************************/ | |
1da177e4 | 3527 | |
6b2bedc3 | 3528 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3529 | EXPORT_SYMBOL(netdev_max_backlog); |
3530 | ||
3b098e2d | 3531 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 | 3532 | int netdev_budget __read_mostly = 300; |
7acf8a1e | 3533 | unsigned int __read_mostly netdev_budget_usecs = 2000; |
3d48b53f MT |
3534 | int weight_p __read_mostly = 64; /* old backlog weight */ |
3535 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
3536 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
3537 | int dev_rx_weight __read_mostly = 64; | |
3538 | int dev_tx_weight __read_mostly = 64; | |
1da177e4 | 3539 | |
eecfd7c4 ED |
3540 | /* Called with irq disabled */ |
3541 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3542 | struct napi_struct *napi) | |
3543 | { | |
3544 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3545 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3546 | } | |
3547 | ||
bfb564e7 KK |
3548 | #ifdef CONFIG_RPS |
3549 | ||
3550 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3551 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 3552 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
3553 | u32 rps_cpu_mask __read_mostly; |
3554 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 3555 | |
c5905afb | 3556 | struct static_key rps_needed __read_mostly; |
3df97ba8 | 3557 | EXPORT_SYMBOL(rps_needed); |
13bfff25 ED |
3558 | struct static_key rfs_needed __read_mostly; |
3559 | EXPORT_SYMBOL(rfs_needed); | |
adc9300e | 3560 | |
c445477d BH |
3561 | static struct rps_dev_flow * |
3562 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3563 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3564 | { | |
a31196b0 | 3565 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
3566 | #ifdef CONFIG_RFS_ACCEL |
3567 | struct netdev_rx_queue *rxqueue; | |
3568 | struct rps_dev_flow_table *flow_table; | |
3569 | struct rps_dev_flow *old_rflow; | |
3570 | u32 flow_id; | |
3571 | u16 rxq_index; | |
3572 | int rc; | |
3573 | ||
3574 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3575 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3576 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3577 | goto out; |
3578 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3579 | if (rxq_index == skb_get_rx_queue(skb)) | |
3580 | goto out; | |
3581 | ||
3582 | rxqueue = dev->_rx + rxq_index; | |
3583 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3584 | if (!flow_table) | |
3585 | goto out; | |
61b905da | 3586 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3587 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3588 | rxq_index, flow_id); | |
3589 | if (rc < 0) | |
3590 | goto out; | |
3591 | old_rflow = rflow; | |
3592 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3593 | rflow->filter = rc; |
3594 | if (old_rflow->filter == rflow->filter) | |
3595 | old_rflow->filter = RPS_NO_FILTER; | |
3596 | out: | |
3597 | #endif | |
3598 | rflow->last_qtail = | |
09994d1b | 3599 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3600 | } |
3601 | ||
09994d1b | 3602 | rflow->cpu = next_cpu; |
c445477d BH |
3603 | return rflow; |
3604 | } | |
3605 | ||
bfb564e7 KK |
3606 | /* |
3607 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3608 | * CPU from the RPS map of the receiving queue for a given skb. | |
3609 | * rcu_read_lock must be held on entry. | |
3610 | */ | |
3611 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3612 | struct rps_dev_flow **rflowp) | |
3613 | { | |
567e4b79 ED |
3614 | const struct rps_sock_flow_table *sock_flow_table; |
3615 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 3616 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 3617 | struct rps_map *map; |
bfb564e7 | 3618 | int cpu = -1; |
567e4b79 | 3619 | u32 tcpu; |
61b905da | 3620 | u32 hash; |
bfb564e7 KK |
3621 | |
3622 | if (skb_rx_queue_recorded(skb)) { | |
3623 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 3624 | |
62fe0b40 BH |
3625 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3626 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3627 | "%s received packet on queue %u, but number " | |
3628 | "of RX queues is %u\n", | |
3629 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3630 | goto done; |
3631 | } | |
567e4b79 ED |
3632 | rxqueue += index; |
3633 | } | |
bfb564e7 | 3634 | |
567e4b79 ED |
3635 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
3636 | ||
3637 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 3638 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 3639 | if (!flow_table && !map) |
bfb564e7 KK |
3640 | goto done; |
3641 | ||
2d47b459 | 3642 | skb_reset_network_header(skb); |
61b905da TH |
3643 | hash = skb_get_hash(skb); |
3644 | if (!hash) | |
bfb564e7 KK |
3645 | goto done; |
3646 | ||
fec5e652 TH |
3647 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
3648 | if (flow_table && sock_flow_table) { | |
fec5e652 | 3649 | struct rps_dev_flow *rflow; |
567e4b79 ED |
3650 | u32 next_cpu; |
3651 | u32 ident; | |
3652 | ||
3653 | /* First check into global flow table if there is a match */ | |
3654 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
3655 | if ((ident ^ hash) & ~rps_cpu_mask) | |
3656 | goto try_rps; | |
fec5e652 | 3657 | |
567e4b79 ED |
3658 | next_cpu = ident & rps_cpu_mask; |
3659 | ||
3660 | /* OK, now we know there is a match, | |
3661 | * we can look at the local (per receive queue) flow table | |
3662 | */ | |
61b905da | 3663 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3664 | tcpu = rflow->cpu; |
3665 | ||
fec5e652 TH |
3666 | /* |
3667 | * If the desired CPU (where last recvmsg was done) is | |
3668 | * different from current CPU (one in the rx-queue flow | |
3669 | * table entry), switch if one of the following holds: | |
a31196b0 | 3670 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
3671 | * - Current CPU is offline. |
3672 | * - The current CPU's queue tail has advanced beyond the | |
3673 | * last packet that was enqueued using this table entry. | |
3674 | * This guarantees that all previous packets for the flow | |
3675 | * have been dequeued, thus preserving in order delivery. | |
3676 | */ | |
3677 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 3678 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 3679 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
3680 | rflow->last_qtail)) >= 0)) { |
3681 | tcpu = next_cpu; | |
c445477d | 3682 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3683 | } |
c445477d | 3684 | |
a31196b0 | 3685 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
3686 | *rflowp = rflow; |
3687 | cpu = tcpu; | |
3688 | goto done; | |
3689 | } | |
3690 | } | |
3691 | ||
567e4b79 ED |
3692 | try_rps: |
3693 | ||
0a9627f2 | 3694 | if (map) { |
8fc54f68 | 3695 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3696 | if (cpu_online(tcpu)) { |
3697 | cpu = tcpu; | |
3698 | goto done; | |
3699 | } | |
3700 | } | |
3701 | ||
3702 | done: | |
0a9627f2 TH |
3703 | return cpu; |
3704 | } | |
3705 | ||
c445477d BH |
3706 | #ifdef CONFIG_RFS_ACCEL |
3707 | ||
3708 | /** | |
3709 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3710 | * @dev: Device on which the filter was set | |
3711 | * @rxq_index: RX queue index | |
3712 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3713 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3714 | * | |
3715 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3716 | * this function for each installed filter and remove the filters for | |
3717 | * which it returns %true. | |
3718 | */ | |
3719 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3720 | u32 flow_id, u16 filter_id) | |
3721 | { | |
3722 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3723 | struct rps_dev_flow_table *flow_table; | |
3724 | struct rps_dev_flow *rflow; | |
3725 | bool expire = true; | |
a31196b0 | 3726 | unsigned int cpu; |
c445477d BH |
3727 | |
3728 | rcu_read_lock(); | |
3729 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3730 | if (flow_table && flow_id <= flow_table->mask) { | |
3731 | rflow = &flow_table->flows[flow_id]; | |
3732 | cpu = ACCESS_ONCE(rflow->cpu); | |
a31196b0 | 3733 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
3734 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
3735 | rflow->last_qtail) < | |
3736 | (int)(10 * flow_table->mask))) | |
3737 | expire = false; | |
3738 | } | |
3739 | rcu_read_unlock(); | |
3740 | return expire; | |
3741 | } | |
3742 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3743 | ||
3744 | #endif /* CONFIG_RFS_ACCEL */ | |
3745 | ||
0a9627f2 | 3746 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3747 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3748 | { |
e36fa2f7 ED |
3749 | struct softnet_data *sd = data; |
3750 | ||
eecfd7c4 | 3751 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3752 | sd->received_rps++; |
0a9627f2 | 3753 | } |
e36fa2f7 | 3754 | |
fec5e652 | 3755 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3756 | |
e36fa2f7 ED |
3757 | /* |
3758 | * Check if this softnet_data structure is another cpu one | |
3759 | * If yes, queue it to our IPI list and return 1 | |
3760 | * If no, return 0 | |
3761 | */ | |
3762 | static int rps_ipi_queued(struct softnet_data *sd) | |
3763 | { | |
3764 | #ifdef CONFIG_RPS | |
903ceff7 | 3765 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
3766 | |
3767 | if (sd != mysd) { | |
3768 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3769 | mysd->rps_ipi_list = sd; | |
3770 | ||
3771 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3772 | return 1; | |
3773 | } | |
3774 | #endif /* CONFIG_RPS */ | |
3775 | return 0; | |
3776 | } | |
3777 | ||
99bbc707 WB |
3778 | #ifdef CONFIG_NET_FLOW_LIMIT |
3779 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3780 | #endif | |
3781 | ||
3782 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3783 | { | |
3784 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3785 | struct sd_flow_limit *fl; | |
3786 | struct softnet_data *sd; | |
3787 | unsigned int old_flow, new_flow; | |
3788 | ||
3789 | if (qlen < (netdev_max_backlog >> 1)) | |
3790 | return false; | |
3791 | ||
903ceff7 | 3792 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
3793 | |
3794 | rcu_read_lock(); | |
3795 | fl = rcu_dereference(sd->flow_limit); | |
3796 | if (fl) { | |
3958afa1 | 3797 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3798 | old_flow = fl->history[fl->history_head]; |
3799 | fl->history[fl->history_head] = new_flow; | |
3800 | ||
3801 | fl->history_head++; | |
3802 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3803 | ||
3804 | if (likely(fl->buckets[old_flow])) | |
3805 | fl->buckets[old_flow]--; | |
3806 | ||
3807 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3808 | fl->count++; | |
3809 | rcu_read_unlock(); | |
3810 | return true; | |
3811 | } | |
3812 | } | |
3813 | rcu_read_unlock(); | |
3814 | #endif | |
3815 | return false; | |
3816 | } | |
3817 | ||
0a9627f2 TH |
3818 | /* |
3819 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3820 | * queue (may be a remote CPU queue). | |
3821 | */ | |
fec5e652 TH |
3822 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3823 | unsigned int *qtail) | |
0a9627f2 | 3824 | { |
e36fa2f7 | 3825 | struct softnet_data *sd; |
0a9627f2 | 3826 | unsigned long flags; |
99bbc707 | 3827 | unsigned int qlen; |
0a9627f2 | 3828 | |
e36fa2f7 | 3829 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3830 | |
3831 | local_irq_save(flags); | |
0a9627f2 | 3832 | |
e36fa2f7 | 3833 | rps_lock(sd); |
e9e4dd32 JA |
3834 | if (!netif_running(skb->dev)) |
3835 | goto drop; | |
99bbc707 WB |
3836 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3837 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 3838 | if (qlen) { |
0a9627f2 | 3839 | enqueue: |
e36fa2f7 | 3840 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3841 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3842 | rps_unlock(sd); |
152102c7 | 3843 | local_irq_restore(flags); |
0a9627f2 TH |
3844 | return NET_RX_SUCCESS; |
3845 | } | |
3846 | ||
ebda37c2 ED |
3847 | /* Schedule NAPI for backlog device |
3848 | * We can use non atomic operation since we own the queue lock | |
3849 | */ | |
3850 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3851 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3852 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3853 | } |
3854 | goto enqueue; | |
3855 | } | |
3856 | ||
e9e4dd32 | 3857 | drop: |
dee42870 | 3858 | sd->dropped++; |
e36fa2f7 | 3859 | rps_unlock(sd); |
0a9627f2 | 3860 | |
0a9627f2 TH |
3861 | local_irq_restore(flags); |
3862 | ||
caf586e5 | 3863 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3864 | kfree_skb(skb); |
3865 | return NET_RX_DROP; | |
3866 | } | |
1da177e4 | 3867 | |
ae78dbfa | 3868 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3869 | { |
b0e28f1e | 3870 | int ret; |
1da177e4 | 3871 | |
588f0330 | 3872 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3873 | |
cf66ba58 | 3874 | trace_netif_rx(skb); |
df334545 | 3875 | #ifdef CONFIG_RPS |
c5905afb | 3876 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3877 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3878 | int cpu; |
3879 | ||
cece1945 | 3880 | preempt_disable(); |
b0e28f1e | 3881 | rcu_read_lock(); |
fec5e652 TH |
3882 | |
3883 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3884 | if (cpu < 0) |
3885 | cpu = smp_processor_id(); | |
fec5e652 TH |
3886 | |
3887 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3888 | ||
b0e28f1e | 3889 | rcu_read_unlock(); |
cece1945 | 3890 | preempt_enable(); |
adc9300e ED |
3891 | } else |
3892 | #endif | |
fec5e652 TH |
3893 | { |
3894 | unsigned int qtail; | |
f4563a75 | 3895 | |
fec5e652 TH |
3896 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); |
3897 | put_cpu(); | |
3898 | } | |
b0e28f1e | 3899 | return ret; |
1da177e4 | 3900 | } |
ae78dbfa BH |
3901 | |
3902 | /** | |
3903 | * netif_rx - post buffer to the network code | |
3904 | * @skb: buffer to post | |
3905 | * | |
3906 | * This function receives a packet from a device driver and queues it for | |
3907 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3908 | * may be dropped during processing for congestion control or by the | |
3909 | * protocol layers. | |
3910 | * | |
3911 | * return values: | |
3912 | * NET_RX_SUCCESS (no congestion) | |
3913 | * NET_RX_DROP (packet was dropped) | |
3914 | * | |
3915 | */ | |
3916 | ||
3917 | int netif_rx(struct sk_buff *skb) | |
3918 | { | |
3919 | trace_netif_rx_entry(skb); | |
3920 | ||
3921 | return netif_rx_internal(skb); | |
3922 | } | |
d1b19dff | 3923 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3924 | |
3925 | int netif_rx_ni(struct sk_buff *skb) | |
3926 | { | |
3927 | int err; | |
3928 | ||
ae78dbfa BH |
3929 | trace_netif_rx_ni_entry(skb); |
3930 | ||
1da177e4 | 3931 | preempt_disable(); |
ae78dbfa | 3932 | err = netif_rx_internal(skb); |
1da177e4 LT |
3933 | if (local_softirq_pending()) |
3934 | do_softirq(); | |
3935 | preempt_enable(); | |
3936 | ||
3937 | return err; | |
3938 | } | |
1da177e4 LT |
3939 | EXPORT_SYMBOL(netif_rx_ni); |
3940 | ||
0766f788 | 3941 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 3942 | { |
903ceff7 | 3943 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
3944 | |
3945 | if (sd->completion_queue) { | |
3946 | struct sk_buff *clist; | |
3947 | ||
3948 | local_irq_disable(); | |
3949 | clist = sd->completion_queue; | |
3950 | sd->completion_queue = NULL; | |
3951 | local_irq_enable(); | |
3952 | ||
3953 | while (clist) { | |
3954 | struct sk_buff *skb = clist; | |
f4563a75 | 3955 | |
1da177e4 LT |
3956 | clist = clist->next; |
3957 | ||
63354797 | 3958 | WARN_ON(refcount_read(&skb->users)); |
e6247027 ED |
3959 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
3960 | trace_consume_skb(skb); | |
3961 | else | |
3962 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
3963 | |
3964 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
3965 | __kfree_skb(skb); | |
3966 | else | |
3967 | __kfree_skb_defer(skb); | |
1da177e4 | 3968 | } |
15fad714 JDB |
3969 | |
3970 | __kfree_skb_flush(); | |
1da177e4 LT |
3971 | } |
3972 | ||
3973 | if (sd->output_queue) { | |
37437bb2 | 3974 | struct Qdisc *head; |
1da177e4 LT |
3975 | |
3976 | local_irq_disable(); | |
3977 | head = sd->output_queue; | |
3978 | sd->output_queue = NULL; | |
a9cbd588 | 3979 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3980 | local_irq_enable(); |
3981 | ||
3982 | while (head) { | |
37437bb2 DM |
3983 | struct Qdisc *q = head; |
3984 | spinlock_t *root_lock; | |
3985 | ||
1da177e4 LT |
3986 | head = head->next_sched; |
3987 | ||
5fb66229 | 3988 | root_lock = qdisc_lock(q); |
3bcb846c ED |
3989 | spin_lock(root_lock); |
3990 | /* We need to make sure head->next_sched is read | |
3991 | * before clearing __QDISC_STATE_SCHED | |
3992 | */ | |
3993 | smp_mb__before_atomic(); | |
3994 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
3995 | qdisc_run(q); | |
3996 | spin_unlock(root_lock); | |
1da177e4 LT |
3997 | } |
3998 | } | |
3999 | } | |
4000 | ||
181402a5 | 4001 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
4002 | /* This hook is defined here for ATM LANE */ |
4003 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
4004 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 4005 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 4006 | #endif |
1da177e4 | 4007 | |
1f211a1b DB |
4008 | static inline struct sk_buff * |
4009 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
4010 | struct net_device *orig_dev) | |
f697c3e8 | 4011 | { |
e7582bab | 4012 | #ifdef CONFIG_NET_CLS_ACT |
d2788d34 DB |
4013 | struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list); |
4014 | struct tcf_result cl_res; | |
24824a09 | 4015 | |
c9e99fd0 DB |
4016 | /* If there's at least one ingress present somewhere (so |
4017 | * we get here via enabled static key), remaining devices | |
4018 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 4019 | * out here. |
c9e99fd0 | 4020 | */ |
d2788d34 | 4021 | if (!cl) |
4577139b | 4022 | return skb; |
f697c3e8 HX |
4023 | if (*pt_prev) { |
4024 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4025 | *pt_prev = NULL; | |
1da177e4 LT |
4026 | } |
4027 | ||
3365495c | 4028 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
8dc07fdb | 4029 | skb->tc_at_ingress = 1; |
24ea591d | 4030 | qdisc_bstats_cpu_update(cl->q, skb); |
c9e99fd0 | 4031 | |
87d83093 | 4032 | switch (tcf_classify(skb, cl, &cl_res, false)) { |
d2788d34 DB |
4033 | case TC_ACT_OK: |
4034 | case TC_ACT_RECLASSIFY: | |
4035 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
4036 | break; | |
4037 | case TC_ACT_SHOT: | |
24ea591d | 4038 | qdisc_qstats_cpu_drop(cl->q); |
8a3a4c6e ED |
4039 | kfree_skb(skb); |
4040 | return NULL; | |
d2788d34 DB |
4041 | case TC_ACT_STOLEN: |
4042 | case TC_ACT_QUEUED: | |
e25ea21f | 4043 | case TC_ACT_TRAP: |
8a3a4c6e | 4044 | consume_skb(skb); |
d2788d34 | 4045 | return NULL; |
27b29f63 AS |
4046 | case TC_ACT_REDIRECT: |
4047 | /* skb_mac_header check was done by cls/act_bpf, so | |
4048 | * we can safely push the L2 header back before | |
4049 | * redirecting to another netdev | |
4050 | */ | |
4051 | __skb_push(skb, skb->mac_len); | |
4052 | skb_do_redirect(skb); | |
4053 | return NULL; | |
d2788d34 DB |
4054 | default: |
4055 | break; | |
f697c3e8 | 4056 | } |
e7582bab | 4057 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
4058 | return skb; |
4059 | } | |
1da177e4 | 4060 | |
24b27fc4 MB |
4061 | /** |
4062 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
4063 | * @dev: device to check | |
4064 | * | |
4065 | * Check if a receive handler is already registered for a given device. | |
4066 | * Return true if there one. | |
4067 | * | |
4068 | * The caller must hold the rtnl_mutex. | |
4069 | */ | |
4070 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
4071 | { | |
4072 | ASSERT_RTNL(); | |
4073 | return dev && rtnl_dereference(dev->rx_handler); | |
4074 | } | |
4075 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
4076 | ||
ab95bfe0 JP |
4077 | /** |
4078 | * netdev_rx_handler_register - register receive handler | |
4079 | * @dev: device to register a handler for | |
4080 | * @rx_handler: receive handler to register | |
93e2c32b | 4081 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 4082 | * |
e227867f | 4083 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
4084 | * called from __netif_receive_skb. A negative errno code is returned |
4085 | * on a failure. | |
4086 | * | |
4087 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
4088 | * |
4089 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
4090 | */ |
4091 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
4092 | rx_handler_func_t *rx_handler, |
4093 | void *rx_handler_data) | |
ab95bfe0 | 4094 | { |
1b7cd004 | 4095 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
4096 | return -EBUSY; |
4097 | ||
00cfec37 | 4098 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 4099 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
4100 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
4101 | ||
4102 | return 0; | |
4103 | } | |
4104 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
4105 | ||
4106 | /** | |
4107 | * netdev_rx_handler_unregister - unregister receive handler | |
4108 | * @dev: device to unregister a handler from | |
4109 | * | |
166ec369 | 4110 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
4111 | * |
4112 | * The caller must hold the rtnl_mutex. | |
4113 | */ | |
4114 | void netdev_rx_handler_unregister(struct net_device *dev) | |
4115 | { | |
4116 | ||
4117 | ASSERT_RTNL(); | |
a9b3cd7f | 4118 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
4119 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
4120 | * section has a guarantee to see a non NULL rx_handler_data | |
4121 | * as well. | |
4122 | */ | |
4123 | synchronize_net(); | |
a9b3cd7f | 4124 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
4125 | } |
4126 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
4127 | ||
b4b9e355 MG |
4128 | /* |
4129 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
4130 | * the special handling of PFMEMALLOC skbs. | |
4131 | */ | |
4132 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
4133 | { | |
4134 | switch (skb->protocol) { | |
2b8837ae JP |
4135 | case htons(ETH_P_ARP): |
4136 | case htons(ETH_P_IP): | |
4137 | case htons(ETH_P_IPV6): | |
4138 | case htons(ETH_P_8021Q): | |
4139 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
4140 | return true; |
4141 | default: | |
4142 | return false; | |
4143 | } | |
4144 | } | |
4145 | ||
e687ad60 PN |
4146 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
4147 | int *ret, struct net_device *orig_dev) | |
4148 | { | |
e7582bab | 4149 | #ifdef CONFIG_NETFILTER_INGRESS |
e687ad60 | 4150 | if (nf_hook_ingress_active(skb)) { |
2c1e2703 AC |
4151 | int ingress_retval; |
4152 | ||
e687ad60 PN |
4153 | if (*pt_prev) { |
4154 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4155 | *pt_prev = NULL; | |
4156 | } | |
4157 | ||
2c1e2703 AC |
4158 | rcu_read_lock(); |
4159 | ingress_retval = nf_hook_ingress(skb); | |
4160 | rcu_read_unlock(); | |
4161 | return ingress_retval; | |
e687ad60 | 4162 | } |
e7582bab | 4163 | #endif /* CONFIG_NETFILTER_INGRESS */ |
e687ad60 PN |
4164 | return 0; |
4165 | } | |
e687ad60 | 4166 | |
9754e293 | 4167 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
4168 | { |
4169 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 4170 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 4171 | struct net_device *orig_dev; |
8a4eb573 | 4172 | bool deliver_exact = false; |
1da177e4 | 4173 | int ret = NET_RX_DROP; |
252e3346 | 4174 | __be16 type; |
1da177e4 | 4175 | |
588f0330 | 4176 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 4177 | |
cf66ba58 | 4178 | trace_netif_receive_skb(skb); |
9b22ea56 | 4179 | |
cc9bd5ce | 4180 | orig_dev = skb->dev; |
8f903c70 | 4181 | |
c1d2bbe1 | 4182 | skb_reset_network_header(skb); |
fda55eca ED |
4183 | if (!skb_transport_header_was_set(skb)) |
4184 | skb_reset_transport_header(skb); | |
0b5c9db1 | 4185 | skb_reset_mac_len(skb); |
1da177e4 LT |
4186 | |
4187 | pt_prev = NULL; | |
4188 | ||
63d8ea7f | 4189 | another_round: |
b6858177 | 4190 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
4191 | |
4192 | __this_cpu_inc(softnet_data.processed); | |
4193 | ||
8ad227ff PM |
4194 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
4195 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 4196 | skb = skb_vlan_untag(skb); |
bcc6d479 | 4197 | if (unlikely(!skb)) |
2c17d27c | 4198 | goto out; |
bcc6d479 JP |
4199 | } |
4200 | ||
e7246e12 WB |
4201 | if (skb_skip_tc_classify(skb)) |
4202 | goto skip_classify; | |
1da177e4 | 4203 | |
9754e293 | 4204 | if (pfmemalloc) |
b4b9e355 MG |
4205 | goto skip_taps; |
4206 | ||
1da177e4 | 4207 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
4208 | if (pt_prev) |
4209 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4210 | pt_prev = ptype; | |
4211 | } | |
4212 | ||
4213 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
4214 | if (pt_prev) | |
4215 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4216 | pt_prev = ptype; | |
1da177e4 LT |
4217 | } |
4218 | ||
b4b9e355 | 4219 | skip_taps: |
1cf51900 | 4220 | #ifdef CONFIG_NET_INGRESS |
4577139b | 4221 | if (static_key_false(&ingress_needed)) { |
1f211a1b | 4222 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); |
4577139b | 4223 | if (!skb) |
2c17d27c | 4224 | goto out; |
e687ad60 PN |
4225 | |
4226 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 4227 | goto out; |
4577139b | 4228 | } |
1cf51900 | 4229 | #endif |
a5135bcf | 4230 | skb_reset_tc(skb); |
e7246e12 | 4231 | skip_classify: |
9754e293 | 4232 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
4233 | goto drop; |
4234 | ||
df8a39de | 4235 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
4236 | if (pt_prev) { |
4237 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4238 | pt_prev = NULL; | |
4239 | } | |
48cc32d3 | 4240 | if (vlan_do_receive(&skb)) |
2425717b JF |
4241 | goto another_round; |
4242 | else if (unlikely(!skb)) | |
2c17d27c | 4243 | goto out; |
2425717b JF |
4244 | } |
4245 | ||
48cc32d3 | 4246 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
4247 | if (rx_handler) { |
4248 | if (pt_prev) { | |
4249 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4250 | pt_prev = NULL; | |
4251 | } | |
8a4eb573 JP |
4252 | switch (rx_handler(&skb)) { |
4253 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 4254 | ret = NET_RX_SUCCESS; |
2c17d27c | 4255 | goto out; |
8a4eb573 | 4256 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 4257 | goto another_round; |
8a4eb573 JP |
4258 | case RX_HANDLER_EXACT: |
4259 | deliver_exact = true; | |
4260 | case RX_HANDLER_PASS: | |
4261 | break; | |
4262 | default: | |
4263 | BUG(); | |
4264 | } | |
ab95bfe0 | 4265 | } |
1da177e4 | 4266 | |
df8a39de JP |
4267 | if (unlikely(skb_vlan_tag_present(skb))) { |
4268 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
4269 | skb->pkt_type = PACKET_OTHERHOST; |
4270 | /* Note: we might in the future use prio bits | |
4271 | * and set skb->priority like in vlan_do_receive() | |
4272 | * For the time being, just ignore Priority Code Point | |
4273 | */ | |
4274 | skb->vlan_tci = 0; | |
4275 | } | |
48cc32d3 | 4276 | |
7866a621 SN |
4277 | type = skb->protocol; |
4278 | ||
63d8ea7f | 4279 | /* deliver only exact match when indicated */ |
7866a621 SN |
4280 | if (likely(!deliver_exact)) { |
4281 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4282 | &ptype_base[ntohs(type) & | |
4283 | PTYPE_HASH_MASK]); | |
4284 | } | |
1f3c8804 | 4285 | |
7866a621 SN |
4286 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
4287 | &orig_dev->ptype_specific); | |
4288 | ||
4289 | if (unlikely(skb->dev != orig_dev)) { | |
4290 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4291 | &skb->dev->ptype_specific); | |
1da177e4 LT |
4292 | } |
4293 | ||
4294 | if (pt_prev) { | |
1080e512 | 4295 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 4296 | goto drop; |
1080e512 MT |
4297 | else |
4298 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 4299 | } else { |
b4b9e355 | 4300 | drop: |
6e7333d3 JW |
4301 | if (!deliver_exact) |
4302 | atomic_long_inc(&skb->dev->rx_dropped); | |
4303 | else | |
4304 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
4305 | kfree_skb(skb); |
4306 | /* Jamal, now you will not able to escape explaining | |
4307 | * me how you were going to use this. :-) | |
4308 | */ | |
4309 | ret = NET_RX_DROP; | |
4310 | } | |
4311 | ||
2c17d27c | 4312 | out: |
9754e293 DM |
4313 | return ret; |
4314 | } | |
4315 | ||
4316 | static int __netif_receive_skb(struct sk_buff *skb) | |
4317 | { | |
4318 | int ret; | |
4319 | ||
4320 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 4321 | unsigned int noreclaim_flag; |
9754e293 DM |
4322 | |
4323 | /* | |
4324 | * PFMEMALLOC skbs are special, they should | |
4325 | * - be delivered to SOCK_MEMALLOC sockets only | |
4326 | * - stay away from userspace | |
4327 | * - have bounded memory usage | |
4328 | * | |
4329 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
4330 | * context down to all allocation sites. | |
4331 | */ | |
f1083048 | 4332 | noreclaim_flag = memalloc_noreclaim_save(); |
9754e293 | 4333 | ret = __netif_receive_skb_core(skb, true); |
f1083048 | 4334 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 DM |
4335 | } else |
4336 | ret = __netif_receive_skb_core(skb, false); | |
4337 | ||
1da177e4 LT |
4338 | return ret; |
4339 | } | |
0a9627f2 | 4340 | |
b5cdae32 DM |
4341 | static struct static_key generic_xdp_needed __read_mostly; |
4342 | ||
4343 | static int generic_xdp_install(struct net_device *dev, struct netdev_xdp *xdp) | |
4344 | { | |
58038695 | 4345 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); |
b5cdae32 DM |
4346 | struct bpf_prog *new = xdp->prog; |
4347 | int ret = 0; | |
4348 | ||
4349 | switch (xdp->command) { | |
58038695 | 4350 | case XDP_SETUP_PROG: |
b5cdae32 DM |
4351 | rcu_assign_pointer(dev->xdp_prog, new); |
4352 | if (old) | |
4353 | bpf_prog_put(old); | |
4354 | ||
4355 | if (old && !new) { | |
4356 | static_key_slow_dec(&generic_xdp_needed); | |
4357 | } else if (new && !old) { | |
4358 | static_key_slow_inc(&generic_xdp_needed); | |
4359 | dev_disable_lro(dev); | |
4360 | } | |
4361 | break; | |
b5cdae32 DM |
4362 | |
4363 | case XDP_QUERY_PROG: | |
58038695 MKL |
4364 | xdp->prog_attached = !!old; |
4365 | xdp->prog_id = old ? old->aux->id : 0; | |
b5cdae32 DM |
4366 | break; |
4367 | ||
4368 | default: | |
4369 | ret = -EINVAL; | |
4370 | break; | |
4371 | } | |
4372 | ||
4373 | return ret; | |
4374 | } | |
4375 | ||
4376 | static u32 netif_receive_generic_xdp(struct sk_buff *skb, | |
4377 | struct bpf_prog *xdp_prog) | |
4378 | { | |
4379 | struct xdp_buff xdp; | |
4380 | u32 act = XDP_DROP; | |
4381 | void *orig_data; | |
4382 | int hlen, off; | |
4383 | u32 mac_len; | |
4384 | ||
4385 | /* Reinjected packets coming from act_mirred or similar should | |
4386 | * not get XDP generic processing. | |
4387 | */ | |
4388 | if (skb_cloned(skb)) | |
4389 | return XDP_PASS; | |
4390 | ||
4391 | if (skb_linearize(skb)) | |
4392 | goto do_drop; | |
4393 | ||
4394 | /* The XDP program wants to see the packet starting at the MAC | |
4395 | * header. | |
4396 | */ | |
4397 | mac_len = skb->data - skb_mac_header(skb); | |
4398 | hlen = skb_headlen(skb) + mac_len; | |
4399 | xdp.data = skb->data - mac_len; | |
4400 | xdp.data_end = xdp.data + hlen; | |
4401 | xdp.data_hard_start = skb->data - skb_headroom(skb); | |
4402 | orig_data = xdp.data; | |
4403 | ||
4404 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
4405 | ||
4406 | off = xdp.data - orig_data; | |
4407 | if (off > 0) | |
4408 | __skb_pull(skb, off); | |
4409 | else if (off < 0) | |
4410 | __skb_push(skb, -off); | |
4411 | ||
4412 | switch (act) { | |
4413 | case XDP_TX: | |
4414 | __skb_push(skb, mac_len); | |
4415 | /* fall through */ | |
4416 | case XDP_PASS: | |
4417 | break; | |
4418 | ||
4419 | default: | |
4420 | bpf_warn_invalid_xdp_action(act); | |
4421 | /* fall through */ | |
4422 | case XDP_ABORTED: | |
4423 | trace_xdp_exception(skb->dev, xdp_prog, act); | |
4424 | /* fall through */ | |
4425 | case XDP_DROP: | |
4426 | do_drop: | |
4427 | kfree_skb(skb); | |
4428 | break; | |
4429 | } | |
4430 | ||
4431 | return act; | |
4432 | } | |
4433 | ||
4434 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
4435 | * network taps in order to match in-driver-XDP behavior. | |
4436 | */ | |
4437 | static void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) | |
4438 | { | |
4439 | struct net_device *dev = skb->dev; | |
4440 | struct netdev_queue *txq; | |
4441 | bool free_skb = true; | |
4442 | int cpu, rc; | |
4443 | ||
4444 | txq = netdev_pick_tx(dev, skb, NULL); | |
4445 | cpu = smp_processor_id(); | |
4446 | HARD_TX_LOCK(dev, txq, cpu); | |
4447 | if (!netif_xmit_stopped(txq)) { | |
4448 | rc = netdev_start_xmit(skb, dev, txq, 0); | |
4449 | if (dev_xmit_complete(rc)) | |
4450 | free_skb = false; | |
4451 | } | |
4452 | HARD_TX_UNLOCK(dev, txq); | |
4453 | if (free_skb) { | |
4454 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
4455 | kfree_skb(skb); | |
4456 | } | |
4457 | } | |
4458 | ||
ae78dbfa | 4459 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 4460 | { |
2c17d27c JA |
4461 | int ret; |
4462 | ||
588f0330 | 4463 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 4464 | |
c1f19b51 RC |
4465 | if (skb_defer_rx_timestamp(skb)) |
4466 | return NET_RX_SUCCESS; | |
4467 | ||
2c17d27c JA |
4468 | rcu_read_lock(); |
4469 | ||
b5cdae32 DM |
4470 | if (static_key_false(&generic_xdp_needed)) { |
4471 | struct bpf_prog *xdp_prog = rcu_dereference(skb->dev->xdp_prog); | |
4472 | ||
4473 | if (xdp_prog) { | |
4474 | u32 act = netif_receive_generic_xdp(skb, xdp_prog); | |
4475 | ||
4476 | if (act != XDP_PASS) { | |
4477 | rcu_read_unlock(); | |
4478 | if (act == XDP_TX) | |
4479 | generic_xdp_tx(skb, xdp_prog); | |
4480 | return NET_RX_DROP; | |
4481 | } | |
4482 | } | |
4483 | } | |
4484 | ||
df334545 | 4485 | #ifdef CONFIG_RPS |
c5905afb | 4486 | if (static_key_false(&rps_needed)) { |
3b098e2d | 4487 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 4488 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 4489 | |
3b098e2d ED |
4490 | if (cpu >= 0) { |
4491 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4492 | rcu_read_unlock(); | |
adc9300e | 4493 | return ret; |
3b098e2d | 4494 | } |
fec5e652 | 4495 | } |
1e94d72f | 4496 | #endif |
2c17d27c JA |
4497 | ret = __netif_receive_skb(skb); |
4498 | rcu_read_unlock(); | |
4499 | return ret; | |
0a9627f2 | 4500 | } |
ae78dbfa BH |
4501 | |
4502 | /** | |
4503 | * netif_receive_skb - process receive buffer from network | |
4504 | * @skb: buffer to process | |
4505 | * | |
4506 | * netif_receive_skb() is the main receive data processing function. | |
4507 | * It always succeeds. The buffer may be dropped during processing | |
4508 | * for congestion control or by the protocol layers. | |
4509 | * | |
4510 | * This function may only be called from softirq context and interrupts | |
4511 | * should be enabled. | |
4512 | * | |
4513 | * Return values (usually ignored): | |
4514 | * NET_RX_SUCCESS: no congestion | |
4515 | * NET_RX_DROP: packet was dropped | |
4516 | */ | |
04eb4489 | 4517 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa BH |
4518 | { |
4519 | trace_netif_receive_skb_entry(skb); | |
4520 | ||
4521 | return netif_receive_skb_internal(skb); | |
4522 | } | |
04eb4489 | 4523 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 4524 | |
41852497 | 4525 | DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
4526 | |
4527 | /* Network device is going away, flush any packets still pending */ | |
4528 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 4529 | { |
6e583ce5 | 4530 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
4531 | struct softnet_data *sd; |
4532 | ||
4533 | local_bh_disable(); | |
4534 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 4535 | |
145dd5f9 | 4536 | local_irq_disable(); |
e36fa2f7 | 4537 | rps_lock(sd); |
6e7676c1 | 4538 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 4539 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 4540 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 4541 | kfree_skb(skb); |
76cc8b13 | 4542 | input_queue_head_incr(sd); |
6e583ce5 | 4543 | } |
6e7676c1 | 4544 | } |
e36fa2f7 | 4545 | rps_unlock(sd); |
145dd5f9 | 4546 | local_irq_enable(); |
6e7676c1 CG |
4547 | |
4548 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 4549 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
4550 | __skb_unlink(skb, &sd->process_queue); |
4551 | kfree_skb(skb); | |
76cc8b13 | 4552 | input_queue_head_incr(sd); |
6e7676c1 CG |
4553 | } |
4554 | } | |
145dd5f9 PA |
4555 | local_bh_enable(); |
4556 | } | |
4557 | ||
41852497 | 4558 | static void flush_all_backlogs(void) |
145dd5f9 PA |
4559 | { |
4560 | unsigned int cpu; | |
4561 | ||
4562 | get_online_cpus(); | |
4563 | ||
41852497 ED |
4564 | for_each_online_cpu(cpu) |
4565 | queue_work_on(cpu, system_highpri_wq, | |
4566 | per_cpu_ptr(&flush_works, cpu)); | |
145dd5f9 PA |
4567 | |
4568 | for_each_online_cpu(cpu) | |
41852497 | 4569 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 PA |
4570 | |
4571 | put_online_cpus(); | |
6e583ce5 SH |
4572 | } |
4573 | ||
d565b0a1 HX |
4574 | static int napi_gro_complete(struct sk_buff *skb) |
4575 | { | |
22061d80 | 4576 | struct packet_offload *ptype; |
d565b0a1 | 4577 | __be16 type = skb->protocol; |
22061d80 | 4578 | struct list_head *head = &offload_base; |
d565b0a1 HX |
4579 | int err = -ENOENT; |
4580 | ||
c3c7c254 ED |
4581 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
4582 | ||
fc59f9a3 HX |
4583 | if (NAPI_GRO_CB(skb)->count == 1) { |
4584 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 4585 | goto out; |
fc59f9a3 | 4586 | } |
d565b0a1 HX |
4587 | |
4588 | rcu_read_lock(); | |
4589 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4590 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
4591 | continue; |
4592 | ||
299603e8 | 4593 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
4594 | break; |
4595 | } | |
4596 | rcu_read_unlock(); | |
4597 | ||
4598 | if (err) { | |
4599 | WARN_ON(&ptype->list == head); | |
4600 | kfree_skb(skb); | |
4601 | return NET_RX_SUCCESS; | |
4602 | } | |
4603 | ||
4604 | out: | |
ae78dbfa | 4605 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
4606 | } |
4607 | ||
2e71a6f8 ED |
4608 | /* napi->gro_list contains packets ordered by age. |
4609 | * youngest packets at the head of it. | |
4610 | * Complete skbs in reverse order to reduce latencies. | |
4611 | */ | |
4612 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 4613 | { |
2e71a6f8 | 4614 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 4615 | |
2e71a6f8 ED |
4616 | /* scan list and build reverse chain */ |
4617 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
4618 | skb->prev = prev; | |
4619 | prev = skb; | |
4620 | } | |
4621 | ||
4622 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 4623 | skb->next = NULL; |
2e71a6f8 ED |
4624 | |
4625 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
4626 | return; | |
4627 | ||
4628 | prev = skb->prev; | |
d565b0a1 | 4629 | napi_gro_complete(skb); |
2e71a6f8 | 4630 | napi->gro_count--; |
d565b0a1 HX |
4631 | } |
4632 | ||
4633 | napi->gro_list = NULL; | |
4634 | } | |
86cac58b | 4635 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 4636 | |
89c5fa33 ED |
4637 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
4638 | { | |
4639 | struct sk_buff *p; | |
4640 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 4641 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
4642 | |
4643 | for (p = napi->gro_list; p; p = p->next) { | |
4644 | unsigned long diffs; | |
4645 | ||
0b4cec8c TH |
4646 | NAPI_GRO_CB(p)->flush = 0; |
4647 | ||
4648 | if (hash != skb_get_hash_raw(p)) { | |
4649 | NAPI_GRO_CB(p)->same_flow = 0; | |
4650 | continue; | |
4651 | } | |
4652 | ||
89c5fa33 ED |
4653 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
4654 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
ce87fc6c | 4655 | diffs |= skb_metadata_dst_cmp(p, skb); |
89c5fa33 ED |
4656 | if (maclen == ETH_HLEN) |
4657 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 4658 | skb_mac_header(skb)); |
89c5fa33 ED |
4659 | else if (!diffs) |
4660 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 4661 | skb_mac_header(skb), |
89c5fa33 ED |
4662 | maclen); |
4663 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
4664 | } |
4665 | } | |
4666 | ||
299603e8 JC |
4667 | static void skb_gro_reset_offset(struct sk_buff *skb) |
4668 | { | |
4669 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4670 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
4671 | ||
4672 | NAPI_GRO_CB(skb)->data_offset = 0; | |
4673 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
4674 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
4675 | ||
4676 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
4677 | pinfo->nr_frags && | |
4678 | !PageHighMem(skb_frag_page(frag0))) { | |
4679 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
7cfd5fd5 ED |
4680 | NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, |
4681 | skb_frag_size(frag0), | |
4682 | skb->end - skb->tail); | |
89c5fa33 ED |
4683 | } |
4684 | } | |
4685 | ||
a50e233c ED |
4686 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
4687 | { | |
4688 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4689 | ||
4690 | BUG_ON(skb->end - skb->tail < grow); | |
4691 | ||
4692 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
4693 | ||
4694 | skb->data_len -= grow; | |
4695 | skb->tail += grow; | |
4696 | ||
4697 | pinfo->frags[0].page_offset += grow; | |
4698 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
4699 | ||
4700 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
4701 | skb_frag_unref(skb, 0); | |
4702 | memmove(pinfo->frags, pinfo->frags + 1, | |
4703 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
4704 | } | |
4705 | } | |
4706 | ||
bb728820 | 4707 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
4708 | { |
4709 | struct sk_buff **pp = NULL; | |
22061d80 | 4710 | struct packet_offload *ptype; |
d565b0a1 | 4711 | __be16 type = skb->protocol; |
22061d80 | 4712 | struct list_head *head = &offload_base; |
0da2afd5 | 4713 | int same_flow; |
5b252f0c | 4714 | enum gro_result ret; |
a50e233c | 4715 | int grow; |
d565b0a1 | 4716 | |
b5cdae32 | 4717 | if (netif_elide_gro(skb->dev)) |
d565b0a1 HX |
4718 | goto normal; |
4719 | ||
89c5fa33 ED |
4720 | gro_list_prepare(napi, skb); |
4721 | ||
d565b0a1 HX |
4722 | rcu_read_lock(); |
4723 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4724 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4725 | continue; |
4726 | ||
86911732 | 4727 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4728 | skb_reset_mac_len(skb); |
d565b0a1 | 4729 | NAPI_GRO_CB(skb)->same_flow = 0; |
d61d072e | 4730 | NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); |
5d38a079 | 4731 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 4732 | NAPI_GRO_CB(skb)->encap_mark = 0; |
fcd91dd4 | 4733 | NAPI_GRO_CB(skb)->recursion_counter = 0; |
a0ca153f | 4734 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 4735 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 4736 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 4737 | |
662880f4 TH |
4738 | /* Setup for GRO checksum validation */ |
4739 | switch (skb->ip_summed) { | |
4740 | case CHECKSUM_COMPLETE: | |
4741 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4742 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4743 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4744 | break; | |
4745 | case CHECKSUM_UNNECESSARY: | |
4746 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4747 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4748 | break; | |
4749 | default: | |
4750 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4751 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4752 | } | |
d565b0a1 | 4753 | |
f191a1d1 | 4754 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4755 | break; |
4756 | } | |
4757 | rcu_read_unlock(); | |
4758 | ||
4759 | if (&ptype->list == head) | |
4760 | goto normal; | |
4761 | ||
25393d3f SK |
4762 | if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) { |
4763 | ret = GRO_CONSUMED; | |
4764 | goto ok; | |
4765 | } | |
4766 | ||
0da2afd5 | 4767 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4768 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4769 | |
d565b0a1 HX |
4770 | if (pp) { |
4771 | struct sk_buff *nskb = *pp; | |
4772 | ||
4773 | *pp = nskb->next; | |
4774 | nskb->next = NULL; | |
4775 | napi_gro_complete(nskb); | |
4ae5544f | 4776 | napi->gro_count--; |
d565b0a1 HX |
4777 | } |
4778 | ||
0da2afd5 | 4779 | if (same_flow) |
d565b0a1 HX |
4780 | goto ok; |
4781 | ||
600adc18 | 4782 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4783 | goto normal; |
d565b0a1 | 4784 | |
600adc18 ED |
4785 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4786 | struct sk_buff *nskb = napi->gro_list; | |
4787 | ||
4788 | /* locate the end of the list to select the 'oldest' flow */ | |
4789 | while (nskb->next) { | |
4790 | pp = &nskb->next; | |
4791 | nskb = *pp; | |
4792 | } | |
4793 | *pp = NULL; | |
4794 | nskb->next = NULL; | |
4795 | napi_gro_complete(nskb); | |
4796 | } else { | |
4797 | napi->gro_count++; | |
4798 | } | |
d565b0a1 | 4799 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4800 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4801 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4802 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4803 | skb->next = napi->gro_list; |
4804 | napi->gro_list = skb; | |
5d0d9be8 | 4805 | ret = GRO_HELD; |
d565b0a1 | 4806 | |
ad0f9904 | 4807 | pull: |
a50e233c ED |
4808 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4809 | if (grow > 0) | |
4810 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4811 | ok: |
5d0d9be8 | 4812 | return ret; |
d565b0a1 HX |
4813 | |
4814 | normal: | |
ad0f9904 HX |
4815 | ret = GRO_NORMAL; |
4816 | goto pull; | |
5d38a079 | 4817 | } |
96e93eab | 4818 | |
bf5a755f JC |
4819 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4820 | { | |
4821 | struct list_head *offload_head = &offload_base; | |
4822 | struct packet_offload *ptype; | |
4823 | ||
4824 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4825 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4826 | continue; | |
4827 | return ptype; | |
4828 | } | |
4829 | return NULL; | |
4830 | } | |
e27a2f83 | 4831 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4832 | |
4833 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4834 | { | |
4835 | struct list_head *offload_head = &offload_base; | |
4836 | struct packet_offload *ptype; | |
4837 | ||
4838 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4839 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4840 | continue; | |
4841 | return ptype; | |
4842 | } | |
4843 | return NULL; | |
4844 | } | |
e27a2f83 | 4845 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4846 | |
e44699d2 MK |
4847 | static void napi_skb_free_stolen_head(struct sk_buff *skb) |
4848 | { | |
4849 | skb_dst_drop(skb); | |
4850 | secpath_reset(skb); | |
4851 | kmem_cache_free(skbuff_head_cache, skb); | |
4852 | } | |
4853 | ||
bb728820 | 4854 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4855 | { |
5d0d9be8 HX |
4856 | switch (ret) { |
4857 | case GRO_NORMAL: | |
ae78dbfa | 4858 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4859 | ret = GRO_DROP; |
4860 | break; | |
5d38a079 | 4861 | |
5d0d9be8 | 4862 | case GRO_DROP: |
5d38a079 HX |
4863 | kfree_skb(skb); |
4864 | break; | |
5b252f0c | 4865 | |
daa86548 | 4866 | case GRO_MERGED_FREE: |
e44699d2 MK |
4867 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
4868 | napi_skb_free_stolen_head(skb); | |
4869 | else | |
d7e8883c | 4870 | __kfree_skb(skb); |
daa86548 ED |
4871 | break; |
4872 | ||
5b252f0c BH |
4873 | case GRO_HELD: |
4874 | case GRO_MERGED: | |
25393d3f | 4875 | case GRO_CONSUMED: |
5b252f0c | 4876 | break; |
5d38a079 HX |
4877 | } |
4878 | ||
c7c4b3b6 | 4879 | return ret; |
5d0d9be8 | 4880 | } |
5d0d9be8 | 4881 | |
c7c4b3b6 | 4882 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4883 | { |
93f93a44 | 4884 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 4885 | trace_napi_gro_receive_entry(skb); |
86911732 | 4886 | |
a50e233c ED |
4887 | skb_gro_reset_offset(skb); |
4888 | ||
89c5fa33 | 4889 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4890 | } |
4891 | EXPORT_SYMBOL(napi_gro_receive); | |
4892 | ||
d0c2b0d2 | 4893 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4894 | { |
93a35f59 ED |
4895 | if (unlikely(skb->pfmemalloc)) { |
4896 | consume_skb(skb); | |
4897 | return; | |
4898 | } | |
96e93eab | 4899 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4900 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4901 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4902 | skb->vlan_tci = 0; |
66c46d74 | 4903 | skb->dev = napi->dev; |
6d152e23 | 4904 | skb->skb_iif = 0; |
c3caf119 JC |
4905 | skb->encapsulation = 0; |
4906 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4907 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
f991bb9d | 4908 | secpath_reset(skb); |
96e93eab HX |
4909 | |
4910 | napi->skb = skb; | |
4911 | } | |
96e93eab | 4912 | |
76620aaf | 4913 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4914 | { |
5d38a079 | 4915 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4916 | |
4917 | if (!skb) { | |
fd11a83d | 4918 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
4919 | if (skb) { |
4920 | napi->skb = skb; | |
4921 | skb_mark_napi_id(skb, napi); | |
4922 | } | |
80595d59 | 4923 | } |
96e93eab HX |
4924 | return skb; |
4925 | } | |
76620aaf | 4926 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4927 | |
a50e233c ED |
4928 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4929 | struct sk_buff *skb, | |
4930 | gro_result_t ret) | |
96e93eab | 4931 | { |
5d0d9be8 HX |
4932 | switch (ret) { |
4933 | case GRO_NORMAL: | |
a50e233c ED |
4934 | case GRO_HELD: |
4935 | __skb_push(skb, ETH_HLEN); | |
4936 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4937 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4938 | ret = GRO_DROP; |
86911732 | 4939 | break; |
5d38a079 | 4940 | |
5d0d9be8 | 4941 | case GRO_DROP: |
5d0d9be8 HX |
4942 | napi_reuse_skb(napi, skb); |
4943 | break; | |
5b252f0c | 4944 | |
e44699d2 MK |
4945 | case GRO_MERGED_FREE: |
4946 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) | |
4947 | napi_skb_free_stolen_head(skb); | |
4948 | else | |
4949 | napi_reuse_skb(napi, skb); | |
4950 | break; | |
4951 | ||
5b252f0c | 4952 | case GRO_MERGED: |
25393d3f | 4953 | case GRO_CONSUMED: |
5b252f0c | 4954 | break; |
5d0d9be8 | 4955 | } |
5d38a079 | 4956 | |
c7c4b3b6 | 4957 | return ret; |
5d38a079 | 4958 | } |
5d0d9be8 | 4959 | |
a50e233c ED |
4960 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
4961 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
4962 | * We copy ethernet header into skb->data to have a common layout. | |
4963 | */ | |
4adb9c4a | 4964 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4965 | { |
4966 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
4967 | const struct ethhdr *eth; |
4968 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
4969 | |
4970 | napi->skb = NULL; | |
4971 | ||
a50e233c ED |
4972 | skb_reset_mac_header(skb); |
4973 | skb_gro_reset_offset(skb); | |
4974 | ||
4975 | eth = skb_gro_header_fast(skb, 0); | |
4976 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
4977 | eth = skb_gro_header_slow(skb, hlen, 0); | |
4978 | if (unlikely(!eth)) { | |
4da46ceb AC |
4979 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
4980 | __func__, napi->dev->name); | |
a50e233c ED |
4981 | napi_reuse_skb(napi, skb); |
4982 | return NULL; | |
4983 | } | |
4984 | } else { | |
4985 | gro_pull_from_frag0(skb, hlen); | |
4986 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
4987 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 4988 | } |
a50e233c ED |
4989 | __skb_pull(skb, hlen); |
4990 | ||
4991 | /* | |
4992 | * This works because the only protocols we care about don't require | |
4993 | * special handling. | |
4994 | * We'll fix it up properly in napi_frags_finish() | |
4995 | */ | |
4996 | skb->protocol = eth->h_proto; | |
76620aaf | 4997 | |
76620aaf HX |
4998 | return skb; |
4999 | } | |
76620aaf | 5000 | |
c7c4b3b6 | 5001 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 5002 | { |
76620aaf | 5003 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
5004 | |
5005 | if (!skb) | |
c7c4b3b6 | 5006 | return GRO_DROP; |
5d0d9be8 | 5007 | |
ae78dbfa BH |
5008 | trace_napi_gro_frags_entry(skb); |
5009 | ||
89c5fa33 | 5010 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 5011 | } |
5d38a079 HX |
5012 | EXPORT_SYMBOL(napi_gro_frags); |
5013 | ||
573e8fca TH |
5014 | /* Compute the checksum from gro_offset and return the folded value |
5015 | * after adding in any pseudo checksum. | |
5016 | */ | |
5017 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
5018 | { | |
5019 | __wsum wsum; | |
5020 | __sum16 sum; | |
5021 | ||
5022 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
5023 | ||
5024 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
5025 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
5026 | if (likely(!sum)) { | |
5027 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
5028 | !skb->csum_complete_sw) | |
5029 | netdev_rx_csum_fault(skb->dev); | |
5030 | } | |
5031 | ||
5032 | NAPI_GRO_CB(skb)->csum = wsum; | |
5033 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
5034 | ||
5035 | return sum; | |
5036 | } | |
5037 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
5038 | ||
773fc8f6 | 5039 | static void net_rps_send_ipi(struct softnet_data *remsd) |
5040 | { | |
5041 | #ifdef CONFIG_RPS | |
5042 | while (remsd) { | |
5043 | struct softnet_data *next = remsd->rps_ipi_next; | |
5044 | ||
5045 | if (cpu_online(remsd->cpu)) | |
5046 | smp_call_function_single_async(remsd->cpu, &remsd->csd); | |
5047 | remsd = next; | |
5048 | } | |
5049 | #endif | |
5050 | } | |
5051 | ||
e326bed2 | 5052 | /* |
855abcf0 | 5053 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
5054 | * Note: called with local irq disabled, but exits with local irq enabled. |
5055 | */ | |
5056 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
5057 | { | |
5058 | #ifdef CONFIG_RPS | |
5059 | struct softnet_data *remsd = sd->rps_ipi_list; | |
5060 | ||
5061 | if (remsd) { | |
5062 | sd->rps_ipi_list = NULL; | |
5063 | ||
5064 | local_irq_enable(); | |
5065 | ||
5066 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
773fc8f6 | 5067 | net_rps_send_ipi(remsd); |
e326bed2 ED |
5068 | } else |
5069 | #endif | |
5070 | local_irq_enable(); | |
5071 | } | |
5072 | ||
d75b1ade ED |
5073 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
5074 | { | |
5075 | #ifdef CONFIG_RPS | |
5076 | return sd->rps_ipi_list != NULL; | |
5077 | #else | |
5078 | return false; | |
5079 | #endif | |
5080 | } | |
5081 | ||
bea3348e | 5082 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 5083 | { |
eecfd7c4 | 5084 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
5085 | bool again = true; |
5086 | int work = 0; | |
1da177e4 | 5087 | |
e326bed2 ED |
5088 | /* Check if we have pending ipi, its better to send them now, |
5089 | * not waiting net_rx_action() end. | |
5090 | */ | |
d75b1ade | 5091 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
5092 | local_irq_disable(); |
5093 | net_rps_action_and_irq_enable(sd); | |
5094 | } | |
d75b1ade | 5095 | |
3d48b53f | 5096 | napi->weight = dev_rx_weight; |
145dd5f9 | 5097 | while (again) { |
1da177e4 | 5098 | struct sk_buff *skb; |
6e7676c1 CG |
5099 | |
5100 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 5101 | rcu_read_lock(); |
6e7676c1 | 5102 | __netif_receive_skb(skb); |
2c17d27c | 5103 | rcu_read_unlock(); |
76cc8b13 | 5104 | input_queue_head_incr(sd); |
145dd5f9 | 5105 | if (++work >= quota) |
76cc8b13 | 5106 | return work; |
145dd5f9 | 5107 | |
6e7676c1 | 5108 | } |
1da177e4 | 5109 | |
145dd5f9 | 5110 | local_irq_disable(); |
e36fa2f7 | 5111 | rps_lock(sd); |
11ef7a89 | 5112 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
5113 | /* |
5114 | * Inline a custom version of __napi_complete(). | |
5115 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
5116 | * and NAPI_STATE_SCHED is the only possible flag set |
5117 | * on backlog. | |
5118 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
5119 | * and we dont need an smp_mb() memory barrier. |
5120 | */ | |
eecfd7c4 | 5121 | napi->state = 0; |
145dd5f9 PA |
5122 | again = false; |
5123 | } else { | |
5124 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
5125 | &sd->process_queue); | |
bea3348e | 5126 | } |
e36fa2f7 | 5127 | rps_unlock(sd); |
145dd5f9 | 5128 | local_irq_enable(); |
6e7676c1 | 5129 | } |
1da177e4 | 5130 | |
bea3348e SH |
5131 | return work; |
5132 | } | |
1da177e4 | 5133 | |
bea3348e SH |
5134 | /** |
5135 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 5136 | * @n: entry to schedule |
bea3348e | 5137 | * |
bc9ad166 ED |
5138 | * The entry's receive function will be scheduled to run. |
5139 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 5140 | */ |
b5606c2d | 5141 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
5142 | { |
5143 | unsigned long flags; | |
1da177e4 | 5144 | |
bea3348e | 5145 | local_irq_save(flags); |
903ceff7 | 5146 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 5147 | local_irq_restore(flags); |
1da177e4 | 5148 | } |
bea3348e SH |
5149 | EXPORT_SYMBOL(__napi_schedule); |
5150 | ||
39e6c820 ED |
5151 | /** |
5152 | * napi_schedule_prep - check if napi can be scheduled | |
5153 | * @n: napi context | |
5154 | * | |
5155 | * Test if NAPI routine is already running, and if not mark | |
5156 | * it as running. This is used as a condition variable | |
5157 | * insure only one NAPI poll instance runs. We also make | |
5158 | * sure there is no pending NAPI disable. | |
5159 | */ | |
5160 | bool napi_schedule_prep(struct napi_struct *n) | |
5161 | { | |
5162 | unsigned long val, new; | |
5163 | ||
5164 | do { | |
5165 | val = READ_ONCE(n->state); | |
5166 | if (unlikely(val & NAPIF_STATE_DISABLE)) | |
5167 | return false; | |
5168 | new = val | NAPIF_STATE_SCHED; | |
5169 | ||
5170 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
5171 | * This was suggested by Alexander Duyck, as compiler | |
5172 | * emits better code than : | |
5173 | * if (val & NAPIF_STATE_SCHED) | |
5174 | * new |= NAPIF_STATE_MISSED; | |
5175 | */ | |
5176 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
5177 | NAPIF_STATE_MISSED; | |
5178 | } while (cmpxchg(&n->state, val, new) != val); | |
5179 | ||
5180 | return !(val & NAPIF_STATE_SCHED); | |
5181 | } | |
5182 | EXPORT_SYMBOL(napi_schedule_prep); | |
5183 | ||
bc9ad166 ED |
5184 | /** |
5185 | * __napi_schedule_irqoff - schedule for receive | |
5186 | * @n: entry to schedule | |
5187 | * | |
5188 | * Variant of __napi_schedule() assuming hard irqs are masked | |
5189 | */ | |
5190 | void __napi_schedule_irqoff(struct napi_struct *n) | |
5191 | { | |
5192 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
5193 | } | |
5194 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
5195 | ||
364b6055 | 5196 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 5197 | { |
39e6c820 | 5198 | unsigned long flags, val, new; |
d565b0a1 HX |
5199 | |
5200 | /* | |
217f6974 ED |
5201 | * 1) Don't let napi dequeue from the cpu poll list |
5202 | * just in case its running on a different cpu. | |
5203 | * 2) If we are busy polling, do nothing here, we have | |
5204 | * the guarantee we will be called later. | |
d565b0a1 | 5205 | */ |
217f6974 ED |
5206 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
5207 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 5208 | return false; |
d565b0a1 | 5209 | |
3b47d303 ED |
5210 | if (n->gro_list) { |
5211 | unsigned long timeout = 0; | |
d75b1ade | 5212 | |
3b47d303 ED |
5213 | if (work_done) |
5214 | timeout = n->dev->gro_flush_timeout; | |
5215 | ||
5216 | if (timeout) | |
5217 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
5218 | HRTIMER_MODE_REL_PINNED); | |
5219 | else | |
5220 | napi_gro_flush(n, false); | |
5221 | } | |
02c1602e | 5222 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
5223 | /* If n->poll_list is not empty, we need to mask irqs */ |
5224 | local_irq_save(flags); | |
02c1602e | 5225 | list_del_init(&n->poll_list); |
d75b1ade ED |
5226 | local_irq_restore(flags); |
5227 | } | |
39e6c820 ED |
5228 | |
5229 | do { | |
5230 | val = READ_ONCE(n->state); | |
5231 | ||
5232 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); | |
5233 | ||
5234 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); | |
5235 | ||
5236 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
5237 | * because we will call napi->poll() one more time. | |
5238 | * This C code was suggested by Alexander Duyck to help gcc. | |
5239 | */ | |
5240 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
5241 | NAPIF_STATE_SCHED; | |
5242 | } while (cmpxchg(&n->state, val, new) != val); | |
5243 | ||
5244 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
5245 | __napi_schedule(n); | |
5246 | return false; | |
5247 | } | |
5248 | ||
364b6055 | 5249 | return true; |
d565b0a1 | 5250 | } |
3b47d303 | 5251 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 5252 | |
af12fa6e | 5253 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 5254 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
5255 | { |
5256 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
5257 | struct napi_struct *napi; | |
5258 | ||
5259 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
5260 | if (napi->napi_id == napi_id) | |
5261 | return napi; | |
5262 | ||
5263 | return NULL; | |
5264 | } | |
02d62e86 ED |
5265 | |
5266 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
217f6974 | 5267 | |
ce6aea93 | 5268 | #define BUSY_POLL_BUDGET 8 |
217f6974 ED |
5269 | |
5270 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) | |
5271 | { | |
5272 | int rc; | |
5273 | ||
39e6c820 ED |
5274 | /* Busy polling means there is a high chance device driver hard irq |
5275 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
5276 | * set in napi_schedule_prep(). | |
5277 | * Since we are about to call napi->poll() once more, we can safely | |
5278 | * clear NAPI_STATE_MISSED. | |
5279 | * | |
5280 | * Note: x86 could use a single "lock and ..." instruction | |
5281 | * to perform these two clear_bit() | |
5282 | */ | |
5283 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
5284 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
5285 | ||
5286 | local_bh_disable(); | |
5287 | ||
5288 | /* All we really want here is to re-enable device interrupts. | |
5289 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
5290 | */ | |
5291 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
5292 | netpoll_poll_unlock(have_poll_lock); | |
5293 | if (rc == BUSY_POLL_BUDGET) | |
5294 | __napi_schedule(napi); | |
5295 | local_bh_enable(); | |
217f6974 ED |
5296 | } |
5297 | ||
7db6b048 SS |
5298 | void napi_busy_loop(unsigned int napi_id, |
5299 | bool (*loop_end)(void *, unsigned long), | |
5300 | void *loop_end_arg) | |
02d62e86 | 5301 | { |
7db6b048 | 5302 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 5303 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 5304 | void *have_poll_lock = NULL; |
02d62e86 | 5305 | struct napi_struct *napi; |
217f6974 ED |
5306 | |
5307 | restart: | |
217f6974 | 5308 | napi_poll = NULL; |
02d62e86 | 5309 | |
2a028ecb | 5310 | rcu_read_lock(); |
02d62e86 | 5311 | |
545cd5e5 | 5312 | napi = napi_by_id(napi_id); |
02d62e86 ED |
5313 | if (!napi) |
5314 | goto out; | |
5315 | ||
217f6974 ED |
5316 | preempt_disable(); |
5317 | for (;;) { | |
2b5cd0df AD |
5318 | int work = 0; |
5319 | ||
2a028ecb | 5320 | local_bh_disable(); |
217f6974 ED |
5321 | if (!napi_poll) { |
5322 | unsigned long val = READ_ONCE(napi->state); | |
5323 | ||
5324 | /* If multiple threads are competing for this napi, | |
5325 | * we avoid dirtying napi->state as much as we can. | |
5326 | */ | |
5327 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
5328 | NAPIF_STATE_IN_BUSY_POLL)) | |
5329 | goto count; | |
5330 | if (cmpxchg(&napi->state, val, | |
5331 | val | NAPIF_STATE_IN_BUSY_POLL | | |
5332 | NAPIF_STATE_SCHED) != val) | |
5333 | goto count; | |
5334 | have_poll_lock = netpoll_poll_lock(napi); | |
5335 | napi_poll = napi->poll; | |
5336 | } | |
2b5cd0df AD |
5337 | work = napi_poll(napi, BUSY_POLL_BUDGET); |
5338 | trace_napi_poll(napi, work, BUSY_POLL_BUDGET); | |
217f6974 | 5339 | count: |
2b5cd0df | 5340 | if (work > 0) |
7db6b048 | 5341 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 5342 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
2a028ecb | 5343 | local_bh_enable(); |
02d62e86 | 5344 | |
7db6b048 | 5345 | if (!loop_end || loop_end(loop_end_arg, start_time)) |
217f6974 | 5346 | break; |
02d62e86 | 5347 | |
217f6974 ED |
5348 | if (unlikely(need_resched())) { |
5349 | if (napi_poll) | |
5350 | busy_poll_stop(napi, have_poll_lock); | |
5351 | preempt_enable(); | |
5352 | rcu_read_unlock(); | |
5353 | cond_resched(); | |
7db6b048 | 5354 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 5355 | return; |
217f6974 ED |
5356 | goto restart; |
5357 | } | |
6cdf89b1 | 5358 | cpu_relax(); |
217f6974 ED |
5359 | } |
5360 | if (napi_poll) | |
5361 | busy_poll_stop(napi, have_poll_lock); | |
5362 | preempt_enable(); | |
02d62e86 | 5363 | out: |
2a028ecb | 5364 | rcu_read_unlock(); |
02d62e86 | 5365 | } |
7db6b048 | 5366 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
5367 | |
5368 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 5369 | |
149d6ad8 | 5370 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 5371 | { |
d64b5e85 ED |
5372 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
5373 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 5374 | return; |
af12fa6e | 5375 | |
52bd2d62 | 5376 | spin_lock(&napi_hash_lock); |
af12fa6e | 5377 | |
545cd5e5 | 5378 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 5379 | do { |
545cd5e5 AD |
5380 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
5381 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
5382 | } while (napi_by_id(napi_gen_id)); |
5383 | napi->napi_id = napi_gen_id; | |
af12fa6e | 5384 | |
52bd2d62 ED |
5385 | hlist_add_head_rcu(&napi->napi_hash_node, |
5386 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 5387 | |
52bd2d62 | 5388 | spin_unlock(&napi_hash_lock); |
af12fa6e | 5389 | } |
af12fa6e ET |
5390 | |
5391 | /* Warning : caller is responsible to make sure rcu grace period | |
5392 | * is respected before freeing memory containing @napi | |
5393 | */ | |
34cbe27e | 5394 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 5395 | { |
34cbe27e ED |
5396 | bool rcu_sync_needed = false; |
5397 | ||
af12fa6e ET |
5398 | spin_lock(&napi_hash_lock); |
5399 | ||
34cbe27e ED |
5400 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
5401 | rcu_sync_needed = true; | |
af12fa6e | 5402 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 5403 | } |
af12fa6e | 5404 | spin_unlock(&napi_hash_lock); |
34cbe27e | 5405 | return rcu_sync_needed; |
af12fa6e ET |
5406 | } |
5407 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
5408 | ||
3b47d303 ED |
5409 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
5410 | { | |
5411 | struct napi_struct *napi; | |
5412 | ||
5413 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
5414 | |
5415 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
5416 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
5417 | */ | |
5418 | if (napi->gro_list && !napi_disable_pending(napi) && | |
5419 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) | |
5420 | __napi_schedule_irqoff(napi); | |
3b47d303 ED |
5421 | |
5422 | return HRTIMER_NORESTART; | |
5423 | } | |
5424 | ||
d565b0a1 HX |
5425 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
5426 | int (*poll)(struct napi_struct *, int), int weight) | |
5427 | { | |
5428 | INIT_LIST_HEAD(&napi->poll_list); | |
3b47d303 ED |
5429 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
5430 | napi->timer.function = napi_watchdog; | |
4ae5544f | 5431 | napi->gro_count = 0; |
d565b0a1 | 5432 | napi->gro_list = NULL; |
5d38a079 | 5433 | napi->skb = NULL; |
d565b0a1 | 5434 | napi->poll = poll; |
82dc3c63 ED |
5435 | if (weight > NAPI_POLL_WEIGHT) |
5436 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
5437 | weight, dev->name); | |
d565b0a1 HX |
5438 | napi->weight = weight; |
5439 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 5440 | napi->dev = dev; |
5d38a079 | 5441 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
5442 | napi->poll_owner = -1; |
5443 | #endif | |
5444 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 5445 | napi_hash_add(napi); |
d565b0a1 HX |
5446 | } |
5447 | EXPORT_SYMBOL(netif_napi_add); | |
5448 | ||
3b47d303 ED |
5449 | void napi_disable(struct napi_struct *n) |
5450 | { | |
5451 | might_sleep(); | |
5452 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
5453 | ||
5454 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
5455 | msleep(1); | |
2d8bff12 NH |
5456 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
5457 | msleep(1); | |
3b47d303 ED |
5458 | |
5459 | hrtimer_cancel(&n->timer); | |
5460 | ||
5461 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
5462 | } | |
5463 | EXPORT_SYMBOL(napi_disable); | |
5464 | ||
93d05d4a | 5465 | /* Must be called in process context */ |
d565b0a1 HX |
5466 | void netif_napi_del(struct napi_struct *napi) |
5467 | { | |
93d05d4a ED |
5468 | might_sleep(); |
5469 | if (napi_hash_del(napi)) | |
5470 | synchronize_net(); | |
d7b06636 | 5471 | list_del_init(&napi->dev_list); |
76620aaf | 5472 | napi_free_frags(napi); |
d565b0a1 | 5473 | |
289dccbe | 5474 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 5475 | napi->gro_list = NULL; |
4ae5544f | 5476 | napi->gro_count = 0; |
d565b0a1 HX |
5477 | } |
5478 | EXPORT_SYMBOL(netif_napi_del); | |
5479 | ||
726ce70e HX |
5480 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
5481 | { | |
5482 | void *have; | |
5483 | int work, weight; | |
5484 | ||
5485 | list_del_init(&n->poll_list); | |
5486 | ||
5487 | have = netpoll_poll_lock(n); | |
5488 | ||
5489 | weight = n->weight; | |
5490 | ||
5491 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
5492 | * with netpoll's poll_napi(). Only the entity which | |
5493 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
5494 | * actually make the ->poll() call. Therefore we avoid | |
5495 | * accidentally calling ->poll() when NAPI is not scheduled. | |
5496 | */ | |
5497 | work = 0; | |
5498 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
5499 | work = n->poll(n, weight); | |
1db19db7 | 5500 | trace_napi_poll(n, work, weight); |
726ce70e HX |
5501 | } |
5502 | ||
5503 | WARN_ON_ONCE(work > weight); | |
5504 | ||
5505 | if (likely(work < weight)) | |
5506 | goto out_unlock; | |
5507 | ||
5508 | /* Drivers must not modify the NAPI state if they | |
5509 | * consume the entire weight. In such cases this code | |
5510 | * still "owns" the NAPI instance and therefore can | |
5511 | * move the instance around on the list at-will. | |
5512 | */ | |
5513 | if (unlikely(napi_disable_pending(n))) { | |
5514 | napi_complete(n); | |
5515 | goto out_unlock; | |
5516 | } | |
5517 | ||
5518 | if (n->gro_list) { | |
5519 | /* flush too old packets | |
5520 | * If HZ < 1000, flush all packets. | |
5521 | */ | |
5522 | napi_gro_flush(n, HZ >= 1000); | |
5523 | } | |
5524 | ||
001ce546 HX |
5525 | /* Some drivers may have called napi_schedule |
5526 | * prior to exhausting their budget. | |
5527 | */ | |
5528 | if (unlikely(!list_empty(&n->poll_list))) { | |
5529 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
5530 | n->dev ? n->dev->name : "backlog"); | |
5531 | goto out_unlock; | |
5532 | } | |
5533 | ||
726ce70e HX |
5534 | list_add_tail(&n->poll_list, repoll); |
5535 | ||
5536 | out_unlock: | |
5537 | netpoll_poll_unlock(have); | |
5538 | ||
5539 | return work; | |
5540 | } | |
5541 | ||
0766f788 | 5542 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 5543 | { |
903ceff7 | 5544 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e MW |
5545 | unsigned long time_limit = jiffies + |
5546 | usecs_to_jiffies(netdev_budget_usecs); | |
51b0bded | 5547 | int budget = netdev_budget; |
d75b1ade ED |
5548 | LIST_HEAD(list); |
5549 | LIST_HEAD(repoll); | |
53fb95d3 | 5550 | |
1da177e4 | 5551 | local_irq_disable(); |
d75b1ade ED |
5552 | list_splice_init(&sd->poll_list, &list); |
5553 | local_irq_enable(); | |
1da177e4 | 5554 | |
ceb8d5bf | 5555 | for (;;) { |
bea3348e | 5556 | struct napi_struct *n; |
1da177e4 | 5557 | |
ceb8d5bf HX |
5558 | if (list_empty(&list)) { |
5559 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
f52dffe0 | 5560 | goto out; |
ceb8d5bf HX |
5561 | break; |
5562 | } | |
5563 | ||
6bd373eb HX |
5564 | n = list_first_entry(&list, struct napi_struct, poll_list); |
5565 | budget -= napi_poll(n, &repoll); | |
5566 | ||
d75b1ade | 5567 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
5568 | * Allow this to run for 2 jiffies since which will allow |
5569 | * an average latency of 1.5/HZ. | |
bea3348e | 5570 | */ |
ceb8d5bf HX |
5571 | if (unlikely(budget <= 0 || |
5572 | time_after_eq(jiffies, time_limit))) { | |
5573 | sd->time_squeeze++; | |
5574 | break; | |
5575 | } | |
1da177e4 | 5576 | } |
d75b1ade | 5577 | |
d75b1ade ED |
5578 | local_irq_disable(); |
5579 | ||
5580 | list_splice_tail_init(&sd->poll_list, &list); | |
5581 | list_splice_tail(&repoll, &list); | |
5582 | list_splice(&list, &sd->poll_list); | |
5583 | if (!list_empty(&sd->poll_list)) | |
5584 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
5585 | ||
e326bed2 | 5586 | net_rps_action_and_irq_enable(sd); |
f52dffe0 ED |
5587 | out: |
5588 | __kfree_skb_flush(); | |
1da177e4 LT |
5589 | } |
5590 | ||
aa9d8560 | 5591 | struct netdev_adjacent { |
9ff162a8 | 5592 | struct net_device *dev; |
5d261913 VF |
5593 | |
5594 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 5595 | bool master; |
5d261913 | 5596 | |
5d261913 VF |
5597 | /* counter for the number of times this device was added to us */ |
5598 | u16 ref_nr; | |
5599 | ||
402dae96 VF |
5600 | /* private field for the users */ |
5601 | void *private; | |
5602 | ||
9ff162a8 JP |
5603 | struct list_head list; |
5604 | struct rcu_head rcu; | |
9ff162a8 JP |
5605 | }; |
5606 | ||
6ea29da1 | 5607 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 5608 | struct list_head *adj_list) |
9ff162a8 | 5609 | { |
5d261913 | 5610 | struct netdev_adjacent *adj; |
5d261913 | 5611 | |
2f268f12 | 5612 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
5613 | if (adj->dev == adj_dev) |
5614 | return adj; | |
9ff162a8 JP |
5615 | } |
5616 | return NULL; | |
5617 | } | |
5618 | ||
f1170fd4 DA |
5619 | static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data) |
5620 | { | |
5621 | struct net_device *dev = data; | |
5622 | ||
5623 | return upper_dev == dev; | |
5624 | } | |
5625 | ||
9ff162a8 JP |
5626 | /** |
5627 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
5628 | * @dev: device | |
5629 | * @upper_dev: upper device to check | |
5630 | * | |
5631 | * Find out if a device is linked to specified upper device and return true | |
5632 | * in case it is. Note that this checks only immediate upper device, | |
5633 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
5634 | */ | |
5635 | bool netdev_has_upper_dev(struct net_device *dev, | |
5636 | struct net_device *upper_dev) | |
5637 | { | |
5638 | ASSERT_RTNL(); | |
5639 | ||
f1170fd4 DA |
5640 | return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, |
5641 | upper_dev); | |
9ff162a8 JP |
5642 | } |
5643 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
5644 | ||
1a3f060c DA |
5645 | /** |
5646 | * netdev_has_upper_dev_all - Check if device is linked to an upper device | |
5647 | * @dev: device | |
5648 | * @upper_dev: upper device to check | |
5649 | * | |
5650 | * Find out if a device is linked to specified upper device and return true | |
5651 | * in case it is. Note that this checks the entire upper device chain. | |
5652 | * The caller must hold rcu lock. | |
5653 | */ | |
5654 | ||
1a3f060c DA |
5655 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
5656 | struct net_device *upper_dev) | |
5657 | { | |
5658 | return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, | |
5659 | upper_dev); | |
5660 | } | |
5661 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
5662 | ||
9ff162a8 JP |
5663 | /** |
5664 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
5665 | * @dev: device | |
5666 | * | |
5667 | * Find out if a device is linked to an upper device and return true in case | |
5668 | * it is. The caller must hold the RTNL lock. | |
5669 | */ | |
1d143d9f | 5670 | static bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
5671 | { |
5672 | ASSERT_RTNL(); | |
5673 | ||
f1170fd4 | 5674 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 5675 | } |
9ff162a8 JP |
5676 | |
5677 | /** | |
5678 | * netdev_master_upper_dev_get - Get master upper device | |
5679 | * @dev: device | |
5680 | * | |
5681 | * Find a master upper device and return pointer to it or NULL in case | |
5682 | * it's not there. The caller must hold the RTNL lock. | |
5683 | */ | |
5684 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
5685 | { | |
aa9d8560 | 5686 | struct netdev_adjacent *upper; |
9ff162a8 JP |
5687 | |
5688 | ASSERT_RTNL(); | |
5689 | ||
2f268f12 | 5690 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
5691 | return NULL; |
5692 | ||
2f268f12 | 5693 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 5694 | struct netdev_adjacent, list); |
9ff162a8 JP |
5695 | if (likely(upper->master)) |
5696 | return upper->dev; | |
5697 | return NULL; | |
5698 | } | |
5699 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
5700 | ||
0f524a80 DA |
5701 | /** |
5702 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
5703 | * @dev: device | |
5704 | * | |
5705 | * Find out if a device is linked to a lower device and return true in case | |
5706 | * it is. The caller must hold the RTNL lock. | |
5707 | */ | |
5708 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
5709 | { | |
5710 | ASSERT_RTNL(); | |
5711 | ||
5712 | return !list_empty(&dev->adj_list.lower); | |
5713 | } | |
5714 | ||
b6ccba4c VF |
5715 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
5716 | { | |
5717 | struct netdev_adjacent *adj; | |
5718 | ||
5719 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
5720 | ||
5721 | return adj->private; | |
5722 | } | |
5723 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
5724 | ||
44a40855 VY |
5725 | /** |
5726 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
5727 | * @dev: device | |
5728 | * @iter: list_head ** of the current position | |
5729 | * | |
5730 | * Gets the next device from the dev's upper list, starting from iter | |
5731 | * position. The caller must hold RCU read lock. | |
5732 | */ | |
5733 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
5734 | struct list_head **iter) | |
5735 | { | |
5736 | struct netdev_adjacent *upper; | |
5737 | ||
5738 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5739 | ||
5740 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5741 | ||
5742 | if (&upper->list == &dev->adj_list.upper) | |
5743 | return NULL; | |
5744 | ||
5745 | *iter = &upper->list; | |
5746 | ||
5747 | return upper->dev; | |
5748 | } | |
5749 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
5750 | ||
1a3f060c DA |
5751 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
5752 | struct list_head **iter) | |
5753 | { | |
5754 | struct netdev_adjacent *upper; | |
5755 | ||
5756 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5757 | ||
5758 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5759 | ||
5760 | if (&upper->list == &dev->adj_list.upper) | |
5761 | return NULL; | |
5762 | ||
5763 | *iter = &upper->list; | |
5764 | ||
5765 | return upper->dev; | |
5766 | } | |
5767 | ||
5768 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, | |
5769 | int (*fn)(struct net_device *dev, | |
5770 | void *data), | |
5771 | void *data) | |
5772 | { | |
5773 | struct net_device *udev; | |
5774 | struct list_head *iter; | |
5775 | int ret; | |
5776 | ||
5777 | for (iter = &dev->adj_list.upper, | |
5778 | udev = netdev_next_upper_dev_rcu(dev, &iter); | |
5779 | udev; | |
5780 | udev = netdev_next_upper_dev_rcu(dev, &iter)) { | |
5781 | /* first is the upper device itself */ | |
5782 | ret = fn(udev, data); | |
5783 | if (ret) | |
5784 | return ret; | |
5785 | ||
5786 | /* then look at all of its upper devices */ | |
5787 | ret = netdev_walk_all_upper_dev_rcu(udev, fn, data); | |
5788 | if (ret) | |
5789 | return ret; | |
5790 | } | |
5791 | ||
5792 | return 0; | |
5793 | } | |
5794 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
5795 | ||
31088a11 VF |
5796 | /** |
5797 | * netdev_lower_get_next_private - Get the next ->private from the | |
5798 | * lower neighbour list | |
5799 | * @dev: device | |
5800 | * @iter: list_head ** of the current position | |
5801 | * | |
5802 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5803 | * list, starting from iter position. The caller must hold either hold the | |
5804 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 5805 | * list will remain unchanged. |
31088a11 VF |
5806 | */ |
5807 | void *netdev_lower_get_next_private(struct net_device *dev, | |
5808 | struct list_head **iter) | |
5809 | { | |
5810 | struct netdev_adjacent *lower; | |
5811 | ||
5812 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
5813 | ||
5814 | if (&lower->list == &dev->adj_list.lower) | |
5815 | return NULL; | |
5816 | ||
6859e7df | 5817 | *iter = lower->list.next; |
31088a11 VF |
5818 | |
5819 | return lower->private; | |
5820 | } | |
5821 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
5822 | ||
5823 | /** | |
5824 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
5825 | * lower neighbour list, RCU | |
5826 | * variant | |
5827 | * @dev: device | |
5828 | * @iter: list_head ** of the current position | |
5829 | * | |
5830 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5831 | * list, starting from iter position. The caller must hold RCU read lock. | |
5832 | */ | |
5833 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
5834 | struct list_head **iter) | |
5835 | { | |
5836 | struct netdev_adjacent *lower; | |
5837 | ||
5838 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5839 | ||
5840 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5841 | ||
5842 | if (&lower->list == &dev->adj_list.lower) | |
5843 | return NULL; | |
5844 | ||
6859e7df | 5845 | *iter = &lower->list; |
31088a11 VF |
5846 | |
5847 | return lower->private; | |
5848 | } | |
5849 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
5850 | ||
4085ebe8 VY |
5851 | /** |
5852 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
5853 | * list | |
5854 | * @dev: device | |
5855 | * @iter: list_head ** of the current position | |
5856 | * | |
5857 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
5858 | * list, starting from iter position. The caller must hold RTNL lock or | |
5859 | * its own locking that guarantees that the neighbour lower | |
b469139e | 5860 | * list will remain unchanged. |
4085ebe8 VY |
5861 | */ |
5862 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
5863 | { | |
5864 | struct netdev_adjacent *lower; | |
5865 | ||
cfdd28be | 5866 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
5867 | |
5868 | if (&lower->list == &dev->adj_list.lower) | |
5869 | return NULL; | |
5870 | ||
cfdd28be | 5871 | *iter = lower->list.next; |
4085ebe8 VY |
5872 | |
5873 | return lower->dev; | |
5874 | } | |
5875 | EXPORT_SYMBOL(netdev_lower_get_next); | |
5876 | ||
1a3f060c DA |
5877 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
5878 | struct list_head **iter) | |
5879 | { | |
5880 | struct netdev_adjacent *lower; | |
5881 | ||
46b5ab1a | 5882 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
5883 | |
5884 | if (&lower->list == &dev->adj_list.lower) | |
5885 | return NULL; | |
5886 | ||
46b5ab1a | 5887 | *iter = &lower->list; |
1a3f060c DA |
5888 | |
5889 | return lower->dev; | |
5890 | } | |
5891 | ||
5892 | int netdev_walk_all_lower_dev(struct net_device *dev, | |
5893 | int (*fn)(struct net_device *dev, | |
5894 | void *data), | |
5895 | void *data) | |
5896 | { | |
5897 | struct net_device *ldev; | |
5898 | struct list_head *iter; | |
5899 | int ret; | |
5900 | ||
5901 | for (iter = &dev->adj_list.lower, | |
5902 | ldev = netdev_next_lower_dev(dev, &iter); | |
5903 | ldev; | |
5904 | ldev = netdev_next_lower_dev(dev, &iter)) { | |
5905 | /* first is the lower device itself */ | |
5906 | ret = fn(ldev, data); | |
5907 | if (ret) | |
5908 | return ret; | |
5909 | ||
5910 | /* then look at all of its lower devices */ | |
5911 | ret = netdev_walk_all_lower_dev(ldev, fn, data); | |
5912 | if (ret) | |
5913 | return ret; | |
5914 | } | |
5915 | ||
5916 | return 0; | |
5917 | } | |
5918 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
5919 | ||
1a3f060c DA |
5920 | static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
5921 | struct list_head **iter) | |
5922 | { | |
5923 | struct netdev_adjacent *lower; | |
5924 | ||
5925 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5926 | if (&lower->list == &dev->adj_list.lower) | |
5927 | return NULL; | |
5928 | ||
5929 | *iter = &lower->list; | |
5930 | ||
5931 | return lower->dev; | |
5932 | } | |
5933 | ||
5934 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
5935 | int (*fn)(struct net_device *dev, | |
5936 | void *data), | |
5937 | void *data) | |
5938 | { | |
5939 | struct net_device *ldev; | |
5940 | struct list_head *iter; | |
5941 | int ret; | |
5942 | ||
5943 | for (iter = &dev->adj_list.lower, | |
5944 | ldev = netdev_next_lower_dev_rcu(dev, &iter); | |
5945 | ldev; | |
5946 | ldev = netdev_next_lower_dev_rcu(dev, &iter)) { | |
5947 | /* first is the lower device itself */ | |
5948 | ret = fn(ldev, data); | |
5949 | if (ret) | |
5950 | return ret; | |
5951 | ||
5952 | /* then look at all of its lower devices */ | |
5953 | ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data); | |
5954 | if (ret) | |
5955 | return ret; | |
5956 | } | |
5957 | ||
5958 | return 0; | |
5959 | } | |
5960 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
5961 | ||
e001bfad | 5962 | /** |
5963 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
5964 | * lower neighbour list, RCU | |
5965 | * variant | |
5966 | * @dev: device | |
5967 | * | |
5968 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
5969 | * list. The caller must hold RCU read lock. | |
5970 | */ | |
5971 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
5972 | { | |
5973 | struct netdev_adjacent *lower; | |
5974 | ||
5975 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
5976 | struct netdev_adjacent, list); | |
5977 | if (lower) | |
5978 | return lower->private; | |
5979 | return NULL; | |
5980 | } | |
5981 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
5982 | ||
9ff162a8 JP |
5983 | /** |
5984 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
5985 | * @dev: device | |
5986 | * | |
5987 | * Find a master upper device and return pointer to it or NULL in case | |
5988 | * it's not there. The caller must hold the RCU read lock. | |
5989 | */ | |
5990 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
5991 | { | |
aa9d8560 | 5992 | struct netdev_adjacent *upper; |
9ff162a8 | 5993 | |
2f268f12 | 5994 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 5995 | struct netdev_adjacent, list); |
9ff162a8 JP |
5996 | if (upper && likely(upper->master)) |
5997 | return upper->dev; | |
5998 | return NULL; | |
5999 | } | |
6000 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
6001 | ||
0a59f3a9 | 6002 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
6003 | struct net_device *adj_dev, |
6004 | struct list_head *dev_list) | |
6005 | { | |
6006 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 6007 | |
3ee32707 VF |
6008 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
6009 | "upper_%s" : "lower_%s", adj_dev->name); | |
6010 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
6011 | linkname); | |
6012 | } | |
0a59f3a9 | 6013 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
6014 | char *name, |
6015 | struct list_head *dev_list) | |
6016 | { | |
6017 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 6018 | |
3ee32707 VF |
6019 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
6020 | "upper_%s" : "lower_%s", name); | |
6021 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
6022 | } | |
6023 | ||
7ce64c79 AF |
6024 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
6025 | struct net_device *adj_dev, | |
6026 | struct list_head *dev_list) | |
6027 | { | |
6028 | return (dev_list == &dev->adj_list.upper || | |
6029 | dev_list == &dev->adj_list.lower) && | |
6030 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
6031 | } | |
3ee32707 | 6032 | |
5d261913 VF |
6033 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
6034 | struct net_device *adj_dev, | |
7863c054 | 6035 | struct list_head *dev_list, |
402dae96 | 6036 | void *private, bool master) |
5d261913 VF |
6037 | { |
6038 | struct netdev_adjacent *adj; | |
842d67a7 | 6039 | int ret; |
5d261913 | 6040 | |
6ea29da1 | 6041 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
6042 | |
6043 | if (adj) { | |
790510d9 | 6044 | adj->ref_nr += 1; |
67b62f98 DA |
6045 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
6046 | dev->name, adj_dev->name, adj->ref_nr); | |
6047 | ||
5d261913 VF |
6048 | return 0; |
6049 | } | |
6050 | ||
6051 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
6052 | if (!adj) | |
6053 | return -ENOMEM; | |
6054 | ||
6055 | adj->dev = adj_dev; | |
6056 | adj->master = master; | |
790510d9 | 6057 | adj->ref_nr = 1; |
402dae96 | 6058 | adj->private = private; |
5d261913 | 6059 | dev_hold(adj_dev); |
2f268f12 | 6060 | |
67b62f98 DA |
6061 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
6062 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 6063 | |
7ce64c79 | 6064 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 6065 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
6066 | if (ret) |
6067 | goto free_adj; | |
6068 | } | |
6069 | ||
7863c054 | 6070 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
6071 | if (master) { |
6072 | ret = sysfs_create_link(&(dev->dev.kobj), | |
6073 | &(adj_dev->dev.kobj), "master"); | |
6074 | if (ret) | |
5831d66e | 6075 | goto remove_symlinks; |
842d67a7 | 6076 | |
7863c054 | 6077 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 6078 | } else { |
7863c054 | 6079 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 6080 | } |
5d261913 VF |
6081 | |
6082 | return 0; | |
842d67a7 | 6083 | |
5831d66e | 6084 | remove_symlinks: |
7ce64c79 | 6085 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6086 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
6087 | free_adj: |
6088 | kfree(adj); | |
974daef7 | 6089 | dev_put(adj_dev); |
842d67a7 VF |
6090 | |
6091 | return ret; | |
5d261913 VF |
6092 | } |
6093 | ||
1d143d9f | 6094 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
6095 | struct net_device *adj_dev, | |
93409033 | 6096 | u16 ref_nr, |
1d143d9f | 6097 | struct list_head *dev_list) |
5d261913 VF |
6098 | { |
6099 | struct netdev_adjacent *adj; | |
6100 | ||
67b62f98 DA |
6101 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
6102 | dev->name, adj_dev->name, ref_nr); | |
6103 | ||
6ea29da1 | 6104 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 6105 | |
2f268f12 | 6106 | if (!adj) { |
67b62f98 | 6107 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 6108 | dev->name, adj_dev->name); |
67b62f98 DA |
6109 | WARN_ON(1); |
6110 | return; | |
2f268f12 | 6111 | } |
5d261913 | 6112 | |
93409033 | 6113 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
6114 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
6115 | dev->name, adj_dev->name, ref_nr, | |
6116 | adj->ref_nr - ref_nr); | |
93409033 | 6117 | adj->ref_nr -= ref_nr; |
5d261913 VF |
6118 | return; |
6119 | } | |
6120 | ||
842d67a7 VF |
6121 | if (adj->master) |
6122 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
6123 | ||
7ce64c79 | 6124 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6125 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 6126 | |
5d261913 | 6127 | list_del_rcu(&adj->list); |
67b62f98 | 6128 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 6129 | adj_dev->name, dev->name, adj_dev->name); |
5d261913 VF |
6130 | dev_put(adj_dev); |
6131 | kfree_rcu(adj, rcu); | |
6132 | } | |
6133 | ||
1d143d9f | 6134 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
6135 | struct net_device *upper_dev, | |
6136 | struct list_head *up_list, | |
6137 | struct list_head *down_list, | |
6138 | void *private, bool master) | |
5d261913 VF |
6139 | { |
6140 | int ret; | |
6141 | ||
790510d9 | 6142 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 6143 | private, master); |
5d261913 VF |
6144 | if (ret) |
6145 | return ret; | |
6146 | ||
790510d9 | 6147 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 6148 | private, false); |
5d261913 | 6149 | if (ret) { |
790510d9 | 6150 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
6151 | return ret; |
6152 | } | |
6153 | ||
6154 | return 0; | |
6155 | } | |
6156 | ||
1d143d9f | 6157 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
6158 | struct net_device *upper_dev, | |
93409033 | 6159 | u16 ref_nr, |
1d143d9f | 6160 | struct list_head *up_list, |
6161 | struct list_head *down_list) | |
5d261913 | 6162 | { |
93409033 AC |
6163 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
6164 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
6165 | } |
6166 | ||
1d143d9f | 6167 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
6168 | struct net_device *upper_dev, | |
6169 | void *private, bool master) | |
2f268f12 | 6170 | { |
f1170fd4 DA |
6171 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
6172 | &dev->adj_list.upper, | |
6173 | &upper_dev->adj_list.lower, | |
6174 | private, master); | |
5d261913 VF |
6175 | } |
6176 | ||
1d143d9f | 6177 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
6178 | struct net_device *upper_dev) | |
2f268f12 | 6179 | { |
93409033 | 6180 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
6181 | &dev->adj_list.upper, |
6182 | &upper_dev->adj_list.lower); | |
6183 | } | |
5d261913 | 6184 | |
9ff162a8 | 6185 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 6186 | struct net_device *upper_dev, bool master, |
29bf24af | 6187 | void *upper_priv, void *upper_info) |
9ff162a8 | 6188 | { |
0e4ead9d | 6189 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 | 6190 | int ret = 0; |
9ff162a8 JP |
6191 | |
6192 | ASSERT_RTNL(); | |
6193 | ||
6194 | if (dev == upper_dev) | |
6195 | return -EBUSY; | |
6196 | ||
6197 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
f1170fd4 | 6198 | if (netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
6199 | return -EBUSY; |
6200 | ||
f1170fd4 | 6201 | if (netdev_has_upper_dev(dev, upper_dev)) |
9ff162a8 JP |
6202 | return -EEXIST; |
6203 | ||
6204 | if (master && netdev_master_upper_dev_get(dev)) | |
6205 | return -EBUSY; | |
6206 | ||
0e4ead9d JP |
6207 | changeupper_info.upper_dev = upper_dev; |
6208 | changeupper_info.master = master; | |
6209 | changeupper_info.linking = true; | |
29bf24af | 6210 | changeupper_info.upper_info = upper_info; |
0e4ead9d | 6211 | |
573c7ba0 JP |
6212 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
6213 | &changeupper_info.info); | |
6214 | ret = notifier_to_errno(ret); | |
6215 | if (ret) | |
6216 | return ret; | |
6217 | ||
6dffb044 | 6218 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 6219 | master); |
5d261913 VF |
6220 | if (ret) |
6221 | return ret; | |
9ff162a8 | 6222 | |
b03804e7 IS |
6223 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
6224 | &changeupper_info.info); | |
6225 | ret = notifier_to_errno(ret); | |
6226 | if (ret) | |
f1170fd4 | 6227 | goto rollback; |
b03804e7 | 6228 | |
9ff162a8 | 6229 | return 0; |
5d261913 | 6230 | |
f1170fd4 | 6231 | rollback: |
2f268f12 | 6232 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
6233 | |
6234 | return ret; | |
9ff162a8 JP |
6235 | } |
6236 | ||
6237 | /** | |
6238 | * netdev_upper_dev_link - Add a link to the upper device | |
6239 | * @dev: device | |
6240 | * @upper_dev: new upper device | |
6241 | * | |
6242 | * Adds a link to device which is upper to this one. The caller must hold | |
6243 | * the RTNL lock. On a failure a negative errno code is returned. | |
6244 | * On success the reference counts are adjusted and the function | |
6245 | * returns zero. | |
6246 | */ | |
6247 | int netdev_upper_dev_link(struct net_device *dev, | |
6248 | struct net_device *upper_dev) | |
6249 | { | |
29bf24af | 6250 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL); |
9ff162a8 JP |
6251 | } |
6252 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
6253 | ||
6254 | /** | |
6255 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
6256 | * @dev: device | |
6257 | * @upper_dev: new upper device | |
6dffb044 | 6258 | * @upper_priv: upper device private |
29bf24af | 6259 | * @upper_info: upper info to be passed down via notifier |
9ff162a8 JP |
6260 | * |
6261 | * Adds a link to device which is upper to this one. In this case, only | |
6262 | * one master upper device can be linked, although other non-master devices | |
6263 | * might be linked as well. The caller must hold the RTNL lock. | |
6264 | * On a failure a negative errno code is returned. On success the reference | |
6265 | * counts are adjusted and the function returns zero. | |
6266 | */ | |
6267 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 6268 | struct net_device *upper_dev, |
29bf24af | 6269 | void *upper_priv, void *upper_info) |
9ff162a8 | 6270 | { |
29bf24af JP |
6271 | return __netdev_upper_dev_link(dev, upper_dev, true, |
6272 | upper_priv, upper_info); | |
9ff162a8 JP |
6273 | } |
6274 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
6275 | ||
6276 | /** | |
6277 | * netdev_upper_dev_unlink - Removes a link to upper device | |
6278 | * @dev: device | |
6279 | * @upper_dev: new upper device | |
6280 | * | |
6281 | * Removes a link to device which is upper to this one. The caller must hold | |
6282 | * the RTNL lock. | |
6283 | */ | |
6284 | void netdev_upper_dev_unlink(struct net_device *dev, | |
6285 | struct net_device *upper_dev) | |
6286 | { | |
0e4ead9d | 6287 | struct netdev_notifier_changeupper_info changeupper_info; |
f4563a75 | 6288 | |
9ff162a8 JP |
6289 | ASSERT_RTNL(); |
6290 | ||
0e4ead9d JP |
6291 | changeupper_info.upper_dev = upper_dev; |
6292 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; | |
6293 | changeupper_info.linking = false; | |
6294 | ||
573c7ba0 JP |
6295 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
6296 | &changeupper_info.info); | |
6297 | ||
2f268f12 | 6298 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 6299 | |
0e4ead9d JP |
6300 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
6301 | &changeupper_info.info); | |
9ff162a8 JP |
6302 | } |
6303 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
6304 | ||
61bd3857 MS |
6305 | /** |
6306 | * netdev_bonding_info_change - Dispatch event about slave change | |
6307 | * @dev: device | |
4a26e453 | 6308 | * @bonding_info: info to dispatch |
61bd3857 MS |
6309 | * |
6310 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
6311 | * The caller must hold the RTNL lock. | |
6312 | */ | |
6313 | void netdev_bonding_info_change(struct net_device *dev, | |
6314 | struct netdev_bonding_info *bonding_info) | |
6315 | { | |
6316 | struct netdev_notifier_bonding_info info; | |
6317 | ||
6318 | memcpy(&info.bonding_info, bonding_info, | |
6319 | sizeof(struct netdev_bonding_info)); | |
6320 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, | |
6321 | &info.info); | |
6322 | } | |
6323 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
6324 | ||
2ce1ee17 | 6325 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
6326 | { |
6327 | struct netdev_adjacent *iter; | |
6328 | ||
6329 | struct net *net = dev_net(dev); | |
6330 | ||
6331 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 6332 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6333 | continue; |
6334 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6335 | &iter->dev->adj_list.lower); | |
6336 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
6337 | &dev->adj_list.upper); | |
6338 | } | |
6339 | ||
6340 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 6341 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6342 | continue; |
6343 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6344 | &iter->dev->adj_list.upper); | |
6345 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
6346 | &dev->adj_list.lower); | |
6347 | } | |
6348 | } | |
6349 | ||
2ce1ee17 | 6350 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
6351 | { |
6352 | struct netdev_adjacent *iter; | |
6353 | ||
6354 | struct net *net = dev_net(dev); | |
6355 | ||
6356 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 6357 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6358 | continue; |
6359 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6360 | &iter->dev->adj_list.lower); | |
6361 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6362 | &dev->adj_list.upper); | |
6363 | } | |
6364 | ||
6365 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 6366 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6367 | continue; |
6368 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6369 | &iter->dev->adj_list.upper); | |
6370 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6371 | &dev->adj_list.lower); | |
6372 | } | |
6373 | } | |
6374 | ||
5bb025fa | 6375 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 6376 | { |
5bb025fa | 6377 | struct netdev_adjacent *iter; |
402dae96 | 6378 | |
4c75431a AF |
6379 | struct net *net = dev_net(dev); |
6380 | ||
5bb025fa | 6381 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 6382 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6383 | continue; |
5bb025fa VF |
6384 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6385 | &iter->dev->adj_list.lower); | |
6386 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6387 | &iter->dev->adj_list.lower); | |
6388 | } | |
402dae96 | 6389 | |
5bb025fa | 6390 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 6391 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6392 | continue; |
5bb025fa VF |
6393 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6394 | &iter->dev->adj_list.upper); | |
6395 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6396 | &iter->dev->adj_list.upper); | |
6397 | } | |
402dae96 | 6398 | } |
402dae96 VF |
6399 | |
6400 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
6401 | struct net_device *lower_dev) | |
6402 | { | |
6403 | struct netdev_adjacent *lower; | |
6404 | ||
6405 | if (!lower_dev) | |
6406 | return NULL; | |
6ea29da1 | 6407 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
6408 | if (!lower) |
6409 | return NULL; | |
6410 | ||
6411 | return lower->private; | |
6412 | } | |
6413 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
6414 | ||
4085ebe8 | 6415 | |
952fcfd0 | 6416 | int dev_get_nest_level(struct net_device *dev) |
4085ebe8 VY |
6417 | { |
6418 | struct net_device *lower = NULL; | |
6419 | struct list_head *iter; | |
6420 | int max_nest = -1; | |
6421 | int nest; | |
6422 | ||
6423 | ASSERT_RTNL(); | |
6424 | ||
6425 | netdev_for_each_lower_dev(dev, lower, iter) { | |
952fcfd0 | 6426 | nest = dev_get_nest_level(lower); |
4085ebe8 VY |
6427 | if (max_nest < nest) |
6428 | max_nest = nest; | |
6429 | } | |
6430 | ||
952fcfd0 | 6431 | return max_nest + 1; |
4085ebe8 VY |
6432 | } |
6433 | EXPORT_SYMBOL(dev_get_nest_level); | |
6434 | ||
04d48266 JP |
6435 | /** |
6436 | * netdev_lower_change - Dispatch event about lower device state change | |
6437 | * @lower_dev: device | |
6438 | * @lower_state_info: state to dispatch | |
6439 | * | |
6440 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
6441 | * The caller must hold the RTNL lock. | |
6442 | */ | |
6443 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
6444 | void *lower_state_info) | |
6445 | { | |
6446 | struct netdev_notifier_changelowerstate_info changelowerstate_info; | |
6447 | ||
6448 | ASSERT_RTNL(); | |
6449 | changelowerstate_info.lower_state_info = lower_state_info; | |
6450 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev, | |
6451 | &changelowerstate_info.info); | |
6452 | } | |
6453 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
6454 | ||
b6c40d68 PM |
6455 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
6456 | { | |
d314774c SH |
6457 | const struct net_device_ops *ops = dev->netdev_ops; |
6458 | ||
d2615bf4 | 6459 | if (ops->ndo_change_rx_flags) |
d314774c | 6460 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
6461 | } |
6462 | ||
991fb3f7 | 6463 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6464 | { |
b536db93 | 6465 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
6466 | kuid_t uid; |
6467 | kgid_t gid; | |
1da177e4 | 6468 | |
24023451 PM |
6469 | ASSERT_RTNL(); |
6470 | ||
dad9b335 WC |
6471 | dev->flags |= IFF_PROMISC; |
6472 | dev->promiscuity += inc; | |
6473 | if (dev->promiscuity == 0) { | |
6474 | /* | |
6475 | * Avoid overflow. | |
6476 | * If inc causes overflow, untouch promisc and return error. | |
6477 | */ | |
6478 | if (inc < 0) | |
6479 | dev->flags &= ~IFF_PROMISC; | |
6480 | else { | |
6481 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
6482 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
6483 | dev->name); | |
dad9b335 WC |
6484 | return -EOVERFLOW; |
6485 | } | |
6486 | } | |
52609c0b | 6487 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
6488 | pr_info("device %s %s promiscuous mode\n", |
6489 | dev->name, | |
6490 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
6491 | if (audit_enabled) { |
6492 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
6493 | audit_log(current->audit_context, GFP_ATOMIC, |
6494 | AUDIT_ANOM_PROMISCUOUS, | |
6495 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
6496 | dev->name, (dev->flags & IFF_PROMISC), | |
6497 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 6498 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
6499 | from_kuid(&init_user_ns, uid), |
6500 | from_kgid(&init_user_ns, gid), | |
7759db82 | 6501 | audit_get_sessionid(current)); |
8192b0c4 | 6502 | } |
24023451 | 6503 | |
b6c40d68 | 6504 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 6505 | } |
991fb3f7 ND |
6506 | if (notify) |
6507 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 6508 | return 0; |
1da177e4 LT |
6509 | } |
6510 | ||
4417da66 PM |
6511 | /** |
6512 | * dev_set_promiscuity - update promiscuity count on a device | |
6513 | * @dev: device | |
6514 | * @inc: modifier | |
6515 | * | |
6516 | * Add or remove promiscuity from a device. While the count in the device | |
6517 | * remains above zero the interface remains promiscuous. Once it hits zero | |
6518 | * the device reverts back to normal filtering operation. A negative inc | |
6519 | * value is used to drop promiscuity on the device. | |
dad9b335 | 6520 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 6521 | */ |
dad9b335 | 6522 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 6523 | { |
b536db93 | 6524 | unsigned int old_flags = dev->flags; |
dad9b335 | 6525 | int err; |
4417da66 | 6526 | |
991fb3f7 | 6527 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 6528 | if (err < 0) |
dad9b335 | 6529 | return err; |
4417da66 PM |
6530 | if (dev->flags != old_flags) |
6531 | dev_set_rx_mode(dev); | |
dad9b335 | 6532 | return err; |
4417da66 | 6533 | } |
d1b19dff | 6534 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 6535 | |
991fb3f7 | 6536 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6537 | { |
991fb3f7 | 6538 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 6539 | |
24023451 PM |
6540 | ASSERT_RTNL(); |
6541 | ||
1da177e4 | 6542 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
6543 | dev->allmulti += inc; |
6544 | if (dev->allmulti == 0) { | |
6545 | /* | |
6546 | * Avoid overflow. | |
6547 | * If inc causes overflow, untouch allmulti and return error. | |
6548 | */ | |
6549 | if (inc < 0) | |
6550 | dev->flags &= ~IFF_ALLMULTI; | |
6551 | else { | |
6552 | dev->allmulti -= inc; | |
7b6cd1ce JP |
6553 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
6554 | dev->name); | |
dad9b335 WC |
6555 | return -EOVERFLOW; |
6556 | } | |
6557 | } | |
24023451 | 6558 | if (dev->flags ^ old_flags) { |
b6c40d68 | 6559 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 6560 | dev_set_rx_mode(dev); |
991fb3f7 ND |
6561 | if (notify) |
6562 | __dev_notify_flags(dev, old_flags, | |
6563 | dev->gflags ^ old_gflags); | |
24023451 | 6564 | } |
dad9b335 | 6565 | return 0; |
4417da66 | 6566 | } |
991fb3f7 ND |
6567 | |
6568 | /** | |
6569 | * dev_set_allmulti - update allmulti count on a device | |
6570 | * @dev: device | |
6571 | * @inc: modifier | |
6572 | * | |
6573 | * Add or remove reception of all multicast frames to a device. While the | |
6574 | * count in the device remains above zero the interface remains listening | |
6575 | * to all interfaces. Once it hits zero the device reverts back to normal | |
6576 | * filtering operation. A negative @inc value is used to drop the counter | |
6577 | * when releasing a resource needing all multicasts. | |
6578 | * Return 0 if successful or a negative errno code on error. | |
6579 | */ | |
6580 | ||
6581 | int dev_set_allmulti(struct net_device *dev, int inc) | |
6582 | { | |
6583 | return __dev_set_allmulti(dev, inc, true); | |
6584 | } | |
d1b19dff | 6585 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
6586 | |
6587 | /* | |
6588 | * Upload unicast and multicast address lists to device and | |
6589 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 6590 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
6591 | * are present. |
6592 | */ | |
6593 | void __dev_set_rx_mode(struct net_device *dev) | |
6594 | { | |
d314774c SH |
6595 | const struct net_device_ops *ops = dev->netdev_ops; |
6596 | ||
4417da66 PM |
6597 | /* dev_open will call this function so the list will stay sane. */ |
6598 | if (!(dev->flags&IFF_UP)) | |
6599 | return; | |
6600 | ||
6601 | if (!netif_device_present(dev)) | |
40b77c94 | 6602 | return; |
4417da66 | 6603 | |
01789349 | 6604 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
6605 | /* Unicast addresses changes may only happen under the rtnl, |
6606 | * therefore calling __dev_set_promiscuity here is safe. | |
6607 | */ | |
32e7bfc4 | 6608 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 6609 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 6610 | dev->uc_promisc = true; |
32e7bfc4 | 6611 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 6612 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 6613 | dev->uc_promisc = false; |
4417da66 | 6614 | } |
4417da66 | 6615 | } |
01789349 JP |
6616 | |
6617 | if (ops->ndo_set_rx_mode) | |
6618 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
6619 | } |
6620 | ||
6621 | void dev_set_rx_mode(struct net_device *dev) | |
6622 | { | |
b9e40857 | 6623 | netif_addr_lock_bh(dev); |
4417da66 | 6624 | __dev_set_rx_mode(dev); |
b9e40857 | 6625 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
6626 | } |
6627 | ||
f0db275a SH |
6628 | /** |
6629 | * dev_get_flags - get flags reported to userspace | |
6630 | * @dev: device | |
6631 | * | |
6632 | * Get the combination of flag bits exported through APIs to userspace. | |
6633 | */ | |
95c96174 | 6634 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 6635 | { |
95c96174 | 6636 | unsigned int flags; |
1da177e4 LT |
6637 | |
6638 | flags = (dev->flags & ~(IFF_PROMISC | | |
6639 | IFF_ALLMULTI | | |
b00055aa SR |
6640 | IFF_RUNNING | |
6641 | IFF_LOWER_UP | | |
6642 | IFF_DORMANT)) | | |
1da177e4 LT |
6643 | (dev->gflags & (IFF_PROMISC | |
6644 | IFF_ALLMULTI)); | |
6645 | ||
b00055aa SR |
6646 | if (netif_running(dev)) { |
6647 | if (netif_oper_up(dev)) | |
6648 | flags |= IFF_RUNNING; | |
6649 | if (netif_carrier_ok(dev)) | |
6650 | flags |= IFF_LOWER_UP; | |
6651 | if (netif_dormant(dev)) | |
6652 | flags |= IFF_DORMANT; | |
6653 | } | |
1da177e4 LT |
6654 | |
6655 | return flags; | |
6656 | } | |
d1b19dff | 6657 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 6658 | |
bd380811 | 6659 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 6660 | { |
b536db93 | 6661 | unsigned int old_flags = dev->flags; |
bd380811 | 6662 | int ret; |
1da177e4 | 6663 | |
24023451 PM |
6664 | ASSERT_RTNL(); |
6665 | ||
1da177e4 LT |
6666 | /* |
6667 | * Set the flags on our device. | |
6668 | */ | |
6669 | ||
6670 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
6671 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
6672 | IFF_AUTOMEDIA)) | | |
6673 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
6674 | IFF_ALLMULTI)); | |
6675 | ||
6676 | /* | |
6677 | * Load in the correct multicast list now the flags have changed. | |
6678 | */ | |
6679 | ||
b6c40d68 PM |
6680 | if ((old_flags ^ flags) & IFF_MULTICAST) |
6681 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 6682 | |
4417da66 | 6683 | dev_set_rx_mode(dev); |
1da177e4 LT |
6684 | |
6685 | /* | |
6686 | * Have we downed the interface. We handle IFF_UP ourselves | |
6687 | * according to user attempts to set it, rather than blindly | |
6688 | * setting it. | |
6689 | */ | |
6690 | ||
6691 | ret = 0; | |
d215d10f | 6692 | if ((old_flags ^ flags) & IFF_UP) |
bd380811 | 6693 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 | 6694 | |
1da177e4 | 6695 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 6696 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 6697 | unsigned int old_flags = dev->flags; |
d1b19dff | 6698 | |
1da177e4 | 6699 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
6700 | |
6701 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
6702 | if (dev->flags != old_flags) | |
6703 | dev_set_rx_mode(dev); | |
1da177e4 LT |
6704 | } |
6705 | ||
6706 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 6707 | * is important. Some (broken) drivers set IFF_PROMISC, when |
6708 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
6709 | */ |
6710 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
6711 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
6712 | ||
1da177e4 | 6713 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 6714 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
6715 | } |
6716 | ||
bd380811 PM |
6717 | return ret; |
6718 | } | |
6719 | ||
a528c219 ND |
6720 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
6721 | unsigned int gchanges) | |
bd380811 PM |
6722 | { |
6723 | unsigned int changes = dev->flags ^ old_flags; | |
6724 | ||
a528c219 | 6725 | if (gchanges) |
7f294054 | 6726 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 6727 | |
bd380811 PM |
6728 | if (changes & IFF_UP) { |
6729 | if (dev->flags & IFF_UP) | |
6730 | call_netdevice_notifiers(NETDEV_UP, dev); | |
6731 | else | |
6732 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
6733 | } | |
6734 | ||
6735 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
6736 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
6737 | struct netdev_notifier_change_info change_info; | |
6738 | ||
6739 | change_info.flags_changed = changes; | |
6740 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
6741 | &change_info.info); | |
6742 | } | |
bd380811 PM |
6743 | } |
6744 | ||
6745 | /** | |
6746 | * dev_change_flags - change device settings | |
6747 | * @dev: device | |
6748 | * @flags: device state flags | |
6749 | * | |
6750 | * Change settings on device based state flags. The flags are | |
6751 | * in the userspace exported format. | |
6752 | */ | |
b536db93 | 6753 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 6754 | { |
b536db93 | 6755 | int ret; |
991fb3f7 | 6756 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
6757 | |
6758 | ret = __dev_change_flags(dev, flags); | |
6759 | if (ret < 0) | |
6760 | return ret; | |
6761 | ||
991fb3f7 | 6762 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 6763 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
6764 | return ret; |
6765 | } | |
d1b19dff | 6766 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 6767 | |
f51048c3 | 6768 | int __dev_set_mtu(struct net_device *dev, int new_mtu) |
2315dc91 VF |
6769 | { |
6770 | const struct net_device_ops *ops = dev->netdev_ops; | |
6771 | ||
6772 | if (ops->ndo_change_mtu) | |
6773 | return ops->ndo_change_mtu(dev, new_mtu); | |
6774 | ||
6775 | dev->mtu = new_mtu; | |
6776 | return 0; | |
6777 | } | |
f51048c3 | 6778 | EXPORT_SYMBOL(__dev_set_mtu); |
2315dc91 | 6779 | |
f0db275a SH |
6780 | /** |
6781 | * dev_set_mtu - Change maximum transfer unit | |
6782 | * @dev: device | |
6783 | * @new_mtu: new transfer unit | |
6784 | * | |
6785 | * Change the maximum transfer size of the network device. | |
6786 | */ | |
1da177e4 LT |
6787 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
6788 | { | |
2315dc91 | 6789 | int err, orig_mtu; |
1da177e4 LT |
6790 | |
6791 | if (new_mtu == dev->mtu) | |
6792 | return 0; | |
6793 | ||
61e84623 JW |
6794 | /* MTU must be positive, and in range */ |
6795 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
6796 | net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n", | |
6797 | dev->name, new_mtu, dev->min_mtu); | |
1da177e4 | 6798 | return -EINVAL; |
61e84623 JW |
6799 | } |
6800 | ||
6801 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
6802 | net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n", | |
a0e65de7 | 6803 | dev->name, new_mtu, dev->max_mtu); |
61e84623 JW |
6804 | return -EINVAL; |
6805 | } | |
1da177e4 LT |
6806 | |
6807 | if (!netif_device_present(dev)) | |
6808 | return -ENODEV; | |
6809 | ||
1d486bfb VF |
6810 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
6811 | err = notifier_to_errno(err); | |
6812 | if (err) | |
6813 | return err; | |
d314774c | 6814 | |
2315dc91 VF |
6815 | orig_mtu = dev->mtu; |
6816 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 6817 | |
2315dc91 VF |
6818 | if (!err) { |
6819 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6820 | err = notifier_to_errno(err); | |
6821 | if (err) { | |
6822 | /* setting mtu back and notifying everyone again, | |
6823 | * so that they have a chance to revert changes. | |
6824 | */ | |
6825 | __dev_set_mtu(dev, orig_mtu); | |
6826 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6827 | } | |
6828 | } | |
1da177e4 LT |
6829 | return err; |
6830 | } | |
d1b19dff | 6831 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 6832 | |
cbda10fa VD |
6833 | /** |
6834 | * dev_set_group - Change group this device belongs to | |
6835 | * @dev: device | |
6836 | * @new_group: group this device should belong to | |
6837 | */ | |
6838 | void dev_set_group(struct net_device *dev, int new_group) | |
6839 | { | |
6840 | dev->group = new_group; | |
6841 | } | |
6842 | EXPORT_SYMBOL(dev_set_group); | |
6843 | ||
f0db275a SH |
6844 | /** |
6845 | * dev_set_mac_address - Change Media Access Control Address | |
6846 | * @dev: device | |
6847 | * @sa: new address | |
6848 | * | |
6849 | * Change the hardware (MAC) address of the device | |
6850 | */ | |
1da177e4 LT |
6851 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
6852 | { | |
d314774c | 6853 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
6854 | int err; |
6855 | ||
d314774c | 6856 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
6857 | return -EOPNOTSUPP; |
6858 | if (sa->sa_family != dev->type) | |
6859 | return -EINVAL; | |
6860 | if (!netif_device_present(dev)) | |
6861 | return -ENODEV; | |
d314774c | 6862 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
6863 | if (err) |
6864 | return err; | |
fbdeca2d | 6865 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 6866 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 6867 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 6868 | return 0; |
1da177e4 | 6869 | } |
d1b19dff | 6870 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 6871 | |
4bf84c35 JP |
6872 | /** |
6873 | * dev_change_carrier - Change device carrier | |
6874 | * @dev: device | |
691b3b7e | 6875 | * @new_carrier: new value |
4bf84c35 JP |
6876 | * |
6877 | * Change device carrier | |
6878 | */ | |
6879 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
6880 | { | |
6881 | const struct net_device_ops *ops = dev->netdev_ops; | |
6882 | ||
6883 | if (!ops->ndo_change_carrier) | |
6884 | return -EOPNOTSUPP; | |
6885 | if (!netif_device_present(dev)) | |
6886 | return -ENODEV; | |
6887 | return ops->ndo_change_carrier(dev, new_carrier); | |
6888 | } | |
6889 | EXPORT_SYMBOL(dev_change_carrier); | |
6890 | ||
66b52b0d JP |
6891 | /** |
6892 | * dev_get_phys_port_id - Get device physical port ID | |
6893 | * @dev: device | |
6894 | * @ppid: port ID | |
6895 | * | |
6896 | * Get device physical port ID | |
6897 | */ | |
6898 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 6899 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
6900 | { |
6901 | const struct net_device_ops *ops = dev->netdev_ops; | |
6902 | ||
6903 | if (!ops->ndo_get_phys_port_id) | |
6904 | return -EOPNOTSUPP; | |
6905 | return ops->ndo_get_phys_port_id(dev, ppid); | |
6906 | } | |
6907 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
6908 | ||
db24a904 DA |
6909 | /** |
6910 | * dev_get_phys_port_name - Get device physical port name | |
6911 | * @dev: device | |
6912 | * @name: port name | |
ed49e650 | 6913 | * @len: limit of bytes to copy to name |
db24a904 DA |
6914 | * |
6915 | * Get device physical port name | |
6916 | */ | |
6917 | int dev_get_phys_port_name(struct net_device *dev, | |
6918 | char *name, size_t len) | |
6919 | { | |
6920 | const struct net_device_ops *ops = dev->netdev_ops; | |
6921 | ||
6922 | if (!ops->ndo_get_phys_port_name) | |
6923 | return -EOPNOTSUPP; | |
6924 | return ops->ndo_get_phys_port_name(dev, name, len); | |
6925 | } | |
6926 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
6927 | ||
d746d707 AK |
6928 | /** |
6929 | * dev_change_proto_down - update protocol port state information | |
6930 | * @dev: device | |
6931 | * @proto_down: new value | |
6932 | * | |
6933 | * This info can be used by switch drivers to set the phys state of the | |
6934 | * port. | |
6935 | */ | |
6936 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
6937 | { | |
6938 | const struct net_device_ops *ops = dev->netdev_ops; | |
6939 | ||
6940 | if (!ops->ndo_change_proto_down) | |
6941 | return -EOPNOTSUPP; | |
6942 | if (!netif_device_present(dev)) | |
6943 | return -ENODEV; | |
6944 | return ops->ndo_change_proto_down(dev, proto_down); | |
6945 | } | |
6946 | EXPORT_SYMBOL(dev_change_proto_down); | |
6947 | ||
ce158e58 | 6948 | u8 __dev_xdp_attached(struct net_device *dev, xdp_op_t xdp_op, u32 *prog_id) |
d67b9cd2 DB |
6949 | { |
6950 | struct netdev_xdp xdp; | |
6951 | ||
6952 | memset(&xdp, 0, sizeof(xdp)); | |
6953 | xdp.command = XDP_QUERY_PROG; | |
6954 | ||
6955 | /* Query must always succeed. */ | |
6956 | WARN_ON(xdp_op(dev, &xdp) < 0); | |
58038695 MKL |
6957 | if (prog_id) |
6958 | *prog_id = xdp.prog_id; | |
6959 | ||
d67b9cd2 DB |
6960 | return xdp.prog_attached; |
6961 | } | |
6962 | ||
6963 | static int dev_xdp_install(struct net_device *dev, xdp_op_t xdp_op, | |
32d60277 | 6964 | struct netlink_ext_ack *extack, u32 flags, |
d67b9cd2 DB |
6965 | struct bpf_prog *prog) |
6966 | { | |
6967 | struct netdev_xdp xdp; | |
6968 | ||
6969 | memset(&xdp, 0, sizeof(xdp)); | |
ee5d032f JK |
6970 | if (flags & XDP_FLAGS_HW_MODE) |
6971 | xdp.command = XDP_SETUP_PROG_HW; | |
6972 | else | |
6973 | xdp.command = XDP_SETUP_PROG; | |
d67b9cd2 | 6974 | xdp.extack = extack; |
32d60277 | 6975 | xdp.flags = flags; |
d67b9cd2 DB |
6976 | xdp.prog = prog; |
6977 | ||
6978 | return xdp_op(dev, &xdp); | |
6979 | } | |
6980 | ||
a7862b45 BB |
6981 | /** |
6982 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
6983 | * @dev: device | |
b5d60989 | 6984 | * @extack: netlink extended ack |
a7862b45 | 6985 | * @fd: new program fd or negative value to clear |
85de8576 | 6986 | * @flags: xdp-related flags |
a7862b45 BB |
6987 | * |
6988 | * Set or clear a bpf program for a device | |
6989 | */ | |
ddf9f970 JK |
6990 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, |
6991 | int fd, u32 flags) | |
a7862b45 BB |
6992 | { |
6993 | const struct net_device_ops *ops = dev->netdev_ops; | |
6994 | struct bpf_prog *prog = NULL; | |
d67b9cd2 | 6995 | xdp_op_t xdp_op, xdp_chk; |
a7862b45 BB |
6996 | int err; |
6997 | ||
85de8576 DB |
6998 | ASSERT_RTNL(); |
6999 | ||
d67b9cd2 | 7000 | xdp_op = xdp_chk = ops->ndo_xdp; |
ee5d032f | 7001 | if (!xdp_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) |
0489df9a | 7002 | return -EOPNOTSUPP; |
b5cdae32 DM |
7003 | if (!xdp_op || (flags & XDP_FLAGS_SKB_MODE)) |
7004 | xdp_op = generic_xdp_install; | |
d67b9cd2 DB |
7005 | if (xdp_op == xdp_chk) |
7006 | xdp_chk = generic_xdp_install; | |
b5cdae32 | 7007 | |
a7862b45 | 7008 | if (fd >= 0) { |
58038695 | 7009 | if (xdp_chk && __dev_xdp_attached(dev, xdp_chk, NULL)) |
d67b9cd2 DB |
7010 | return -EEXIST; |
7011 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && | |
58038695 | 7012 | __dev_xdp_attached(dev, xdp_op, NULL)) |
d67b9cd2 | 7013 | return -EBUSY; |
85de8576 | 7014 | |
a7862b45 BB |
7015 | prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP); |
7016 | if (IS_ERR(prog)) | |
7017 | return PTR_ERR(prog); | |
7018 | } | |
7019 | ||
32d60277 | 7020 | err = dev_xdp_install(dev, xdp_op, extack, flags, prog); |
a7862b45 BB |
7021 | if (err < 0 && prog) |
7022 | bpf_prog_put(prog); | |
7023 | ||
7024 | return err; | |
7025 | } | |
a7862b45 | 7026 | |
1da177e4 LT |
7027 | /** |
7028 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 7029 | * @net: the applicable net namespace |
1da177e4 LT |
7030 | * |
7031 | * Returns a suitable unique value for a new device interface | |
7032 | * number. The caller must hold the rtnl semaphore or the | |
7033 | * dev_base_lock to be sure it remains unique. | |
7034 | */ | |
881d966b | 7035 | static int dev_new_index(struct net *net) |
1da177e4 | 7036 | { |
aa79e66e | 7037 | int ifindex = net->ifindex; |
f4563a75 | 7038 | |
1da177e4 LT |
7039 | for (;;) { |
7040 | if (++ifindex <= 0) | |
7041 | ifindex = 1; | |
881d966b | 7042 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 7043 | return net->ifindex = ifindex; |
1da177e4 LT |
7044 | } |
7045 | } | |
7046 | ||
1da177e4 | 7047 | /* Delayed registration/unregisteration */ |
3b5b34fd | 7048 | static LIST_HEAD(net_todo_list); |
200b916f | 7049 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 7050 | |
6f05f629 | 7051 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 7052 | { |
1da177e4 | 7053 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 7054 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
7055 | } |
7056 | ||
9b5e383c | 7057 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 7058 | { |
e93737b0 | 7059 | struct net_device *dev, *tmp; |
5cde2829 | 7060 | LIST_HEAD(close_head); |
9b5e383c | 7061 | |
93ee31f1 DL |
7062 | BUG_ON(dev_boot_phase); |
7063 | ASSERT_RTNL(); | |
7064 | ||
e93737b0 | 7065 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 7066 | /* Some devices call without registering |
e93737b0 KK |
7067 | * for initialization unwind. Remove those |
7068 | * devices and proceed with the remaining. | |
9b5e383c ED |
7069 | */ |
7070 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
7071 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
7072 | dev->name, dev); | |
93ee31f1 | 7073 | |
9b5e383c | 7074 | WARN_ON(1); |
e93737b0 KK |
7075 | list_del(&dev->unreg_list); |
7076 | continue; | |
9b5e383c | 7077 | } |
449f4544 | 7078 | dev->dismantle = true; |
9b5e383c | 7079 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 7080 | } |
93ee31f1 | 7081 | |
44345724 | 7082 | /* If device is running, close it first. */ |
5cde2829 EB |
7083 | list_for_each_entry(dev, head, unreg_list) |
7084 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 7085 | dev_close_many(&close_head, true); |
93ee31f1 | 7086 | |
44345724 | 7087 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
7088 | /* And unlink it from device chain. */ |
7089 | unlist_netdevice(dev); | |
93ee31f1 | 7090 | |
9b5e383c ED |
7091 | dev->reg_state = NETREG_UNREGISTERING; |
7092 | } | |
41852497 | 7093 | flush_all_backlogs(); |
93ee31f1 DL |
7094 | |
7095 | synchronize_net(); | |
7096 | ||
9b5e383c | 7097 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
7098 | struct sk_buff *skb = NULL; |
7099 | ||
9b5e383c ED |
7100 | /* Shutdown queueing discipline. */ |
7101 | dev_shutdown(dev); | |
93ee31f1 DL |
7102 | |
7103 | ||
9b5e383c | 7104 | /* Notify protocols, that we are about to destroy |
eb13da1a | 7105 | * this device. They should clean all the things. |
7106 | */ | |
9b5e383c | 7107 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
93ee31f1 | 7108 | |
395eea6c MB |
7109 | if (!dev->rtnl_link_ops || |
7110 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
3d3ea5af | 7111 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, |
395eea6c MB |
7112 | GFP_KERNEL); |
7113 | ||
9b5e383c ED |
7114 | /* |
7115 | * Flush the unicast and multicast chains | |
7116 | */ | |
a748ee24 | 7117 | dev_uc_flush(dev); |
22bedad3 | 7118 | dev_mc_flush(dev); |
93ee31f1 | 7119 | |
9b5e383c ED |
7120 | if (dev->netdev_ops->ndo_uninit) |
7121 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 7122 | |
395eea6c MB |
7123 | if (skb) |
7124 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 7125 | |
9ff162a8 JP |
7126 | /* Notifier chain MUST detach us all upper devices. */ |
7127 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
0f524a80 | 7128 | WARN_ON(netdev_has_any_lower_dev(dev)); |
93ee31f1 | 7129 | |
9b5e383c ED |
7130 | /* Remove entries from kobject tree */ |
7131 | netdev_unregister_kobject(dev); | |
024e9679 AD |
7132 | #ifdef CONFIG_XPS |
7133 | /* Remove XPS queueing entries */ | |
7134 | netif_reset_xps_queues_gt(dev, 0); | |
7135 | #endif | |
9b5e383c | 7136 | } |
93ee31f1 | 7137 | |
850a545b | 7138 | synchronize_net(); |
395264d5 | 7139 | |
a5ee1551 | 7140 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
7141 | dev_put(dev); |
7142 | } | |
7143 | ||
7144 | static void rollback_registered(struct net_device *dev) | |
7145 | { | |
7146 | LIST_HEAD(single); | |
7147 | ||
7148 | list_add(&dev->unreg_list, &single); | |
7149 | rollback_registered_many(&single); | |
ceaaec98 | 7150 | list_del(&single); |
93ee31f1 DL |
7151 | } |
7152 | ||
fd867d51 JW |
7153 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
7154 | struct net_device *upper, netdev_features_t features) | |
7155 | { | |
7156 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
7157 | netdev_features_t feature; | |
5ba3f7d6 | 7158 | int feature_bit; |
fd867d51 | 7159 | |
5ba3f7d6 JW |
7160 | for_each_netdev_feature(&upper_disables, feature_bit) { |
7161 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
7162 | if (!(upper->wanted_features & feature) |
7163 | && (features & feature)) { | |
7164 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
7165 | &feature, upper->name); | |
7166 | features &= ~feature; | |
7167 | } | |
7168 | } | |
7169 | ||
7170 | return features; | |
7171 | } | |
7172 | ||
7173 | static void netdev_sync_lower_features(struct net_device *upper, | |
7174 | struct net_device *lower, netdev_features_t features) | |
7175 | { | |
7176 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
7177 | netdev_features_t feature; | |
5ba3f7d6 | 7178 | int feature_bit; |
fd867d51 | 7179 | |
5ba3f7d6 JW |
7180 | for_each_netdev_feature(&upper_disables, feature_bit) { |
7181 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
7182 | if (!(features & feature) && (lower->features & feature)) { |
7183 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
7184 | &feature, lower->name); | |
7185 | lower->wanted_features &= ~feature; | |
7186 | netdev_update_features(lower); | |
7187 | ||
7188 | if (unlikely(lower->features & feature)) | |
7189 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
7190 | &feature, lower->name); | |
7191 | } | |
7192 | } | |
7193 | } | |
7194 | ||
c8f44aff MM |
7195 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
7196 | netdev_features_t features) | |
b63365a2 | 7197 | { |
57422dc5 MM |
7198 | /* Fix illegal checksum combinations */ |
7199 | if ((features & NETIF_F_HW_CSUM) && | |
7200 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 7201 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
7202 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
7203 | } | |
7204 | ||
b63365a2 | 7205 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 7206 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 7207 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 7208 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
7209 | } |
7210 | ||
ec5f0615 PS |
7211 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
7212 | !(features & NETIF_F_IP_CSUM)) { | |
7213 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
7214 | features &= ~NETIF_F_TSO; | |
7215 | features &= ~NETIF_F_TSO_ECN; | |
7216 | } | |
7217 | ||
7218 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
7219 | !(features & NETIF_F_IPV6_CSUM)) { | |
7220 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
7221 | features &= ~NETIF_F_TSO6; | |
7222 | } | |
7223 | ||
b1dc497b AD |
7224 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
7225 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
7226 | features &= ~NETIF_F_TSO_MANGLEID; | |
7227 | ||
31d8b9e0 BH |
7228 | /* TSO ECN requires that TSO is present as well. */ |
7229 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
7230 | features &= ~NETIF_F_TSO_ECN; | |
7231 | ||
212b573f MM |
7232 | /* Software GSO depends on SG. */ |
7233 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 7234 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
7235 | features &= ~NETIF_F_GSO; |
7236 | } | |
7237 | ||
acd1130e | 7238 | /* UFO needs SG and checksumming */ |
b63365a2 | 7239 | if (features & NETIF_F_UFO) { |
79032644 | 7240 | /* maybe split UFO into V4 and V6? */ |
c8cd0989 TH |
7241 | if (!(features & NETIF_F_HW_CSUM) && |
7242 | ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) != | |
7243 | (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 7244 | netdev_dbg(dev, |
acd1130e | 7245 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
7246 | features &= ~NETIF_F_UFO; |
7247 | } | |
7248 | ||
7249 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 7250 | netdev_dbg(dev, |
acd1130e | 7251 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
7252 | features &= ~NETIF_F_UFO; |
7253 | } | |
7254 | } | |
7255 | ||
802ab55a AD |
7256 | /* GSO partial features require GSO partial be set */ |
7257 | if ((features & dev->gso_partial_features) && | |
7258 | !(features & NETIF_F_GSO_PARTIAL)) { | |
7259 | netdev_dbg(dev, | |
7260 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
7261 | features &= ~dev->gso_partial_features; | |
7262 | } | |
7263 | ||
b63365a2 HX |
7264 | return features; |
7265 | } | |
b63365a2 | 7266 | |
6cb6a27c | 7267 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 7268 | { |
fd867d51 | 7269 | struct net_device *upper, *lower; |
c8f44aff | 7270 | netdev_features_t features; |
fd867d51 | 7271 | struct list_head *iter; |
e7868a85 | 7272 | int err = -1; |
5455c699 | 7273 | |
87267485 MM |
7274 | ASSERT_RTNL(); |
7275 | ||
5455c699 MM |
7276 | features = netdev_get_wanted_features(dev); |
7277 | ||
7278 | if (dev->netdev_ops->ndo_fix_features) | |
7279 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
7280 | ||
7281 | /* driver might be less strict about feature dependencies */ | |
7282 | features = netdev_fix_features(dev, features); | |
7283 | ||
fd867d51 JW |
7284 | /* some features can't be enabled if they're off an an upper device */ |
7285 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
7286 | features = netdev_sync_upper_features(dev, upper, features); | |
7287 | ||
5455c699 | 7288 | if (dev->features == features) |
e7868a85 | 7289 | goto sync_lower; |
5455c699 | 7290 | |
c8f44aff MM |
7291 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
7292 | &dev->features, &features); | |
5455c699 MM |
7293 | |
7294 | if (dev->netdev_ops->ndo_set_features) | |
7295 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
7296 | else |
7297 | err = 0; | |
5455c699 | 7298 | |
6cb6a27c | 7299 | if (unlikely(err < 0)) { |
5455c699 | 7300 | netdev_err(dev, |
c8f44aff MM |
7301 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
7302 | err, &features, &dev->features); | |
17b85d29 NA |
7303 | /* return non-0 since some features might have changed and |
7304 | * it's better to fire a spurious notification than miss it | |
7305 | */ | |
7306 | return -1; | |
6cb6a27c MM |
7307 | } |
7308 | ||
e7868a85 | 7309 | sync_lower: |
fd867d51 JW |
7310 | /* some features must be disabled on lower devices when disabled |
7311 | * on an upper device (think: bonding master or bridge) | |
7312 | */ | |
7313 | netdev_for_each_lower_dev(dev, lower, iter) | |
7314 | netdev_sync_lower_features(dev, lower, features); | |
7315 | ||
6cb6a27c MM |
7316 | if (!err) |
7317 | dev->features = features; | |
7318 | ||
e7868a85 | 7319 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
7320 | } |
7321 | ||
afe12cc8 MM |
7322 | /** |
7323 | * netdev_update_features - recalculate device features | |
7324 | * @dev: the device to check | |
7325 | * | |
7326 | * Recalculate dev->features set and send notifications if it | |
7327 | * has changed. Should be called after driver or hardware dependent | |
7328 | * conditions might have changed that influence the features. | |
7329 | */ | |
6cb6a27c MM |
7330 | void netdev_update_features(struct net_device *dev) |
7331 | { | |
7332 | if (__netdev_update_features(dev)) | |
7333 | netdev_features_change(dev); | |
5455c699 MM |
7334 | } |
7335 | EXPORT_SYMBOL(netdev_update_features); | |
7336 | ||
afe12cc8 MM |
7337 | /** |
7338 | * netdev_change_features - recalculate device features | |
7339 | * @dev: the device to check | |
7340 | * | |
7341 | * Recalculate dev->features set and send notifications even | |
7342 | * if they have not changed. Should be called instead of | |
7343 | * netdev_update_features() if also dev->vlan_features might | |
7344 | * have changed to allow the changes to be propagated to stacked | |
7345 | * VLAN devices. | |
7346 | */ | |
7347 | void netdev_change_features(struct net_device *dev) | |
7348 | { | |
7349 | __netdev_update_features(dev); | |
7350 | netdev_features_change(dev); | |
7351 | } | |
7352 | EXPORT_SYMBOL(netdev_change_features); | |
7353 | ||
fc4a7489 PM |
7354 | /** |
7355 | * netif_stacked_transfer_operstate - transfer operstate | |
7356 | * @rootdev: the root or lower level device to transfer state from | |
7357 | * @dev: the device to transfer operstate to | |
7358 | * | |
7359 | * Transfer operational state from root to device. This is normally | |
7360 | * called when a stacking relationship exists between the root | |
7361 | * device and the device(a leaf device). | |
7362 | */ | |
7363 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
7364 | struct net_device *dev) | |
7365 | { | |
7366 | if (rootdev->operstate == IF_OPER_DORMANT) | |
7367 | netif_dormant_on(dev); | |
7368 | else | |
7369 | netif_dormant_off(dev); | |
7370 | ||
0575c86b ZS |
7371 | if (netif_carrier_ok(rootdev)) |
7372 | netif_carrier_on(dev); | |
7373 | else | |
7374 | netif_carrier_off(dev); | |
fc4a7489 PM |
7375 | } |
7376 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
7377 | ||
a953be53 | 7378 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
7379 | static int netif_alloc_rx_queues(struct net_device *dev) |
7380 | { | |
1b4bf461 | 7381 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 7382 | struct netdev_rx_queue *rx; |
10595902 | 7383 | size_t sz = count * sizeof(*rx); |
1b4bf461 | 7384 | |
bd25fa7b | 7385 | BUG_ON(count < 1); |
1b4bf461 | 7386 | |
dcda9b04 | 7387 | rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
7388 | if (!rx) |
7389 | return -ENOMEM; | |
7390 | ||
bd25fa7b TH |
7391 | dev->_rx = rx; |
7392 | ||
bd25fa7b | 7393 | for (i = 0; i < count; i++) |
fe822240 | 7394 | rx[i].dev = dev; |
1b4bf461 ED |
7395 | return 0; |
7396 | } | |
bf264145 | 7397 | #endif |
1b4bf461 | 7398 | |
aa942104 CG |
7399 | static void netdev_init_one_queue(struct net_device *dev, |
7400 | struct netdev_queue *queue, void *_unused) | |
7401 | { | |
7402 | /* Initialize queue lock */ | |
7403 | spin_lock_init(&queue->_xmit_lock); | |
7404 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
7405 | queue->xmit_lock_owner = -1; | |
b236da69 | 7406 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 7407 | queue->dev = dev; |
114cf580 TH |
7408 | #ifdef CONFIG_BQL |
7409 | dql_init(&queue->dql, HZ); | |
7410 | #endif | |
aa942104 CG |
7411 | } |
7412 | ||
60877a32 ED |
7413 | static void netif_free_tx_queues(struct net_device *dev) |
7414 | { | |
4cb28970 | 7415 | kvfree(dev->_tx); |
60877a32 ED |
7416 | } |
7417 | ||
e6484930 TH |
7418 | static int netif_alloc_netdev_queues(struct net_device *dev) |
7419 | { | |
7420 | unsigned int count = dev->num_tx_queues; | |
7421 | struct netdev_queue *tx; | |
60877a32 | 7422 | size_t sz = count * sizeof(*tx); |
e6484930 | 7423 | |
d339727c ED |
7424 | if (count < 1 || count > 0xffff) |
7425 | return -EINVAL; | |
62b5942a | 7426 | |
dcda9b04 | 7427 | tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
7428 | if (!tx) |
7429 | return -ENOMEM; | |
7430 | ||
e6484930 | 7431 | dev->_tx = tx; |
1d24eb48 | 7432 | |
e6484930 TH |
7433 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
7434 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
7435 | |
7436 | return 0; | |
e6484930 TH |
7437 | } |
7438 | ||
a2029240 DV |
7439 | void netif_tx_stop_all_queues(struct net_device *dev) |
7440 | { | |
7441 | unsigned int i; | |
7442 | ||
7443 | for (i = 0; i < dev->num_tx_queues; i++) { | |
7444 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 7445 | |
a2029240 DV |
7446 | netif_tx_stop_queue(txq); |
7447 | } | |
7448 | } | |
7449 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
7450 | ||
1da177e4 LT |
7451 | /** |
7452 | * register_netdevice - register a network device | |
7453 | * @dev: device to register | |
7454 | * | |
7455 | * Take a completed network device structure and add it to the kernel | |
7456 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7457 | * chain. 0 is returned on success. A negative errno code is returned | |
7458 | * on a failure to set up the device, or if the name is a duplicate. | |
7459 | * | |
7460 | * Callers must hold the rtnl semaphore. You may want | |
7461 | * register_netdev() instead of this. | |
7462 | * | |
7463 | * BUGS: | |
7464 | * The locking appears insufficient to guarantee two parallel registers | |
7465 | * will not get the same name. | |
7466 | */ | |
7467 | ||
7468 | int register_netdevice(struct net_device *dev) | |
7469 | { | |
1da177e4 | 7470 | int ret; |
d314774c | 7471 | struct net *net = dev_net(dev); |
1da177e4 LT |
7472 | |
7473 | BUG_ON(dev_boot_phase); | |
7474 | ASSERT_RTNL(); | |
7475 | ||
b17a7c17 SH |
7476 | might_sleep(); |
7477 | ||
1da177e4 LT |
7478 | /* When net_device's are persistent, this will be fatal. */ |
7479 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 7480 | BUG_ON(!net); |
1da177e4 | 7481 | |
f1f28aa3 | 7482 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 7483 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 7484 | |
828de4f6 | 7485 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
7486 | if (ret < 0) |
7487 | goto out; | |
7488 | ||
1da177e4 | 7489 | /* Init, if this function is available */ |
d314774c SH |
7490 | if (dev->netdev_ops->ndo_init) { |
7491 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
7492 | if (ret) { |
7493 | if (ret > 0) | |
7494 | ret = -EIO; | |
90833aa4 | 7495 | goto out; |
1da177e4 LT |
7496 | } |
7497 | } | |
4ec93edb | 7498 | |
f646968f PM |
7499 | if (((dev->hw_features | dev->features) & |
7500 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
7501 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
7502 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
7503 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
7504 | ret = -EINVAL; | |
7505 | goto err_uninit; | |
7506 | } | |
7507 | ||
9c7dafbf PE |
7508 | ret = -EBUSY; |
7509 | if (!dev->ifindex) | |
7510 | dev->ifindex = dev_new_index(net); | |
7511 | else if (__dev_get_by_index(net, dev->ifindex)) | |
7512 | goto err_uninit; | |
7513 | ||
5455c699 MM |
7514 | /* Transfer changeable features to wanted_features and enable |
7515 | * software offloads (GSO and GRO). | |
7516 | */ | |
7517 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
7518 | dev->features |= NETIF_F_SOFT_FEATURES; |
7519 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 7520 | |
cbc53e08 | 7521 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 7522 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 7523 | |
7f348a60 AD |
7524 | /* If IPv4 TCP segmentation offload is supported we should also |
7525 | * allow the device to enable segmenting the frame with the option | |
7526 | * of ignoring a static IP ID value. This doesn't enable the | |
7527 | * feature itself but allows the user to enable it later. | |
7528 | */ | |
cbc53e08 AD |
7529 | if (dev->hw_features & NETIF_F_TSO) |
7530 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
7531 | if (dev->vlan_features & NETIF_F_TSO) |
7532 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
7533 | if (dev->mpls_features & NETIF_F_TSO) | |
7534 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
7535 | if (dev->hw_enc_features & NETIF_F_TSO) | |
7536 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 7537 | |
1180e7d6 | 7538 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 7539 | */ |
1180e7d6 | 7540 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 7541 | |
ee579677 PS |
7542 | /* Make NETIF_F_SG inheritable to tunnel devices. |
7543 | */ | |
802ab55a | 7544 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 7545 | |
0d89d203 SH |
7546 | /* Make NETIF_F_SG inheritable to MPLS. |
7547 | */ | |
7548 | dev->mpls_features |= NETIF_F_SG; | |
7549 | ||
7ffbe3fd JB |
7550 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
7551 | ret = notifier_to_errno(ret); | |
7552 | if (ret) | |
7553 | goto err_uninit; | |
7554 | ||
8b41d188 | 7555 | ret = netdev_register_kobject(dev); |
b17a7c17 | 7556 | if (ret) |
7ce1b0ed | 7557 | goto err_uninit; |
b17a7c17 SH |
7558 | dev->reg_state = NETREG_REGISTERED; |
7559 | ||
6cb6a27c | 7560 | __netdev_update_features(dev); |
8e9b59b2 | 7561 | |
1da177e4 LT |
7562 | /* |
7563 | * Default initial state at registry is that the | |
7564 | * device is present. | |
7565 | */ | |
7566 | ||
7567 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7568 | ||
8f4cccbb BH |
7569 | linkwatch_init_dev(dev); |
7570 | ||
1da177e4 | 7571 | dev_init_scheduler(dev); |
1da177e4 | 7572 | dev_hold(dev); |
ce286d32 | 7573 | list_netdevice(dev); |
7bf23575 | 7574 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 7575 | |
948b337e JP |
7576 | /* If the device has permanent device address, driver should |
7577 | * set dev_addr and also addr_assign_type should be set to | |
7578 | * NET_ADDR_PERM (default value). | |
7579 | */ | |
7580 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
7581 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
7582 | ||
1da177e4 | 7583 | /* Notify protocols, that a new device appeared. */ |
056925ab | 7584 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 7585 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
7586 | if (ret) { |
7587 | rollback_registered(dev); | |
7588 | dev->reg_state = NETREG_UNREGISTERED; | |
7589 | } | |
d90a909e EB |
7590 | /* |
7591 | * Prevent userspace races by waiting until the network | |
7592 | * device is fully setup before sending notifications. | |
7593 | */ | |
a2835763 PM |
7594 | if (!dev->rtnl_link_ops || |
7595 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 7596 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
7597 | |
7598 | out: | |
7599 | return ret; | |
7ce1b0ed HX |
7600 | |
7601 | err_uninit: | |
d314774c SH |
7602 | if (dev->netdev_ops->ndo_uninit) |
7603 | dev->netdev_ops->ndo_uninit(dev); | |
cf124db5 DM |
7604 | if (dev->priv_destructor) |
7605 | dev->priv_destructor(dev); | |
7ce1b0ed | 7606 | goto out; |
1da177e4 | 7607 | } |
d1b19dff | 7608 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 7609 | |
937f1ba5 BH |
7610 | /** |
7611 | * init_dummy_netdev - init a dummy network device for NAPI | |
7612 | * @dev: device to init | |
7613 | * | |
7614 | * This takes a network device structure and initialize the minimum | |
7615 | * amount of fields so it can be used to schedule NAPI polls without | |
7616 | * registering a full blown interface. This is to be used by drivers | |
7617 | * that need to tie several hardware interfaces to a single NAPI | |
7618 | * poll scheduler due to HW limitations. | |
7619 | */ | |
7620 | int init_dummy_netdev(struct net_device *dev) | |
7621 | { | |
7622 | /* Clear everything. Note we don't initialize spinlocks | |
7623 | * are they aren't supposed to be taken by any of the | |
7624 | * NAPI code and this dummy netdev is supposed to be | |
7625 | * only ever used for NAPI polls | |
7626 | */ | |
7627 | memset(dev, 0, sizeof(struct net_device)); | |
7628 | ||
7629 | /* make sure we BUG if trying to hit standard | |
7630 | * register/unregister code path | |
7631 | */ | |
7632 | dev->reg_state = NETREG_DUMMY; | |
7633 | ||
937f1ba5 BH |
7634 | /* NAPI wants this */ |
7635 | INIT_LIST_HEAD(&dev->napi_list); | |
7636 | ||
7637 | /* a dummy interface is started by default */ | |
7638 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7639 | set_bit(__LINK_STATE_START, &dev->state); | |
7640 | ||
29b4433d ED |
7641 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
7642 | * because users of this 'device' dont need to change | |
7643 | * its refcount. | |
7644 | */ | |
7645 | ||
937f1ba5 BH |
7646 | return 0; |
7647 | } | |
7648 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
7649 | ||
7650 | ||
1da177e4 LT |
7651 | /** |
7652 | * register_netdev - register a network device | |
7653 | * @dev: device to register | |
7654 | * | |
7655 | * Take a completed network device structure and add it to the kernel | |
7656 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7657 | * chain. 0 is returned on success. A negative errno code is returned | |
7658 | * on a failure to set up the device, or if the name is a duplicate. | |
7659 | * | |
38b4da38 | 7660 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
7661 | * and expands the device name if you passed a format string to |
7662 | * alloc_netdev. | |
7663 | */ | |
7664 | int register_netdev(struct net_device *dev) | |
7665 | { | |
7666 | int err; | |
7667 | ||
7668 | rtnl_lock(); | |
1da177e4 | 7669 | err = register_netdevice(dev); |
1da177e4 LT |
7670 | rtnl_unlock(); |
7671 | return err; | |
7672 | } | |
7673 | EXPORT_SYMBOL(register_netdev); | |
7674 | ||
29b4433d ED |
7675 | int netdev_refcnt_read(const struct net_device *dev) |
7676 | { | |
7677 | int i, refcnt = 0; | |
7678 | ||
7679 | for_each_possible_cpu(i) | |
7680 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
7681 | return refcnt; | |
7682 | } | |
7683 | EXPORT_SYMBOL(netdev_refcnt_read); | |
7684 | ||
2c53040f | 7685 | /** |
1da177e4 | 7686 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 7687 | * @dev: target net_device |
1da177e4 LT |
7688 | * |
7689 | * This is called when unregistering network devices. | |
7690 | * | |
7691 | * Any protocol or device that holds a reference should register | |
7692 | * for netdevice notification, and cleanup and put back the | |
7693 | * reference if they receive an UNREGISTER event. | |
7694 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 7695 | * call dev_put. |
1da177e4 LT |
7696 | */ |
7697 | static void netdev_wait_allrefs(struct net_device *dev) | |
7698 | { | |
7699 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 7700 | int refcnt; |
1da177e4 | 7701 | |
e014debe ED |
7702 | linkwatch_forget_dev(dev); |
7703 | ||
1da177e4 | 7704 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
7705 | refcnt = netdev_refcnt_read(dev); |
7706 | ||
7707 | while (refcnt != 0) { | |
1da177e4 | 7708 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 7709 | rtnl_lock(); |
1da177e4 LT |
7710 | |
7711 | /* Rebroadcast unregister notification */ | |
056925ab | 7712 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 7713 | |
748e2d93 | 7714 | __rtnl_unlock(); |
0115e8e3 | 7715 | rcu_barrier(); |
748e2d93 ED |
7716 | rtnl_lock(); |
7717 | ||
0115e8e3 | 7718 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
7719 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
7720 | &dev->state)) { | |
7721 | /* We must not have linkwatch events | |
7722 | * pending on unregister. If this | |
7723 | * happens, we simply run the queue | |
7724 | * unscheduled, resulting in a noop | |
7725 | * for this device. | |
7726 | */ | |
7727 | linkwatch_run_queue(); | |
7728 | } | |
7729 | ||
6756ae4b | 7730 | __rtnl_unlock(); |
1da177e4 LT |
7731 | |
7732 | rebroadcast_time = jiffies; | |
7733 | } | |
7734 | ||
7735 | msleep(250); | |
7736 | ||
29b4433d ED |
7737 | refcnt = netdev_refcnt_read(dev); |
7738 | ||
1da177e4 | 7739 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
7740 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
7741 | dev->name, refcnt); | |
1da177e4 LT |
7742 | warning_time = jiffies; |
7743 | } | |
7744 | } | |
7745 | } | |
7746 | ||
7747 | /* The sequence is: | |
7748 | * | |
7749 | * rtnl_lock(); | |
7750 | * ... | |
7751 | * register_netdevice(x1); | |
7752 | * register_netdevice(x2); | |
7753 | * ... | |
7754 | * unregister_netdevice(y1); | |
7755 | * unregister_netdevice(y2); | |
7756 | * ... | |
7757 | * rtnl_unlock(); | |
7758 | * free_netdev(y1); | |
7759 | * free_netdev(y2); | |
7760 | * | |
58ec3b4d | 7761 | * We are invoked by rtnl_unlock(). |
1da177e4 | 7762 | * This allows us to deal with problems: |
b17a7c17 | 7763 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
7764 | * without deadlocking with linkwatch via keventd. |
7765 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
7766 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
7767 | * |
7768 | * We must not return until all unregister events added during | |
7769 | * the interval the lock was held have been completed. | |
1da177e4 | 7770 | */ |
1da177e4 LT |
7771 | void netdev_run_todo(void) |
7772 | { | |
626ab0e6 | 7773 | struct list_head list; |
1da177e4 | 7774 | |
1da177e4 | 7775 | /* Snapshot list, allow later requests */ |
626ab0e6 | 7776 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
7777 | |
7778 | __rtnl_unlock(); | |
626ab0e6 | 7779 | |
0115e8e3 ED |
7780 | |
7781 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
7782 | if (!list_empty(&list)) |
7783 | rcu_barrier(); | |
7784 | ||
1da177e4 LT |
7785 | while (!list_empty(&list)) { |
7786 | struct net_device *dev | |
e5e26d75 | 7787 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
7788 | list_del(&dev->todo_list); |
7789 | ||
748e2d93 | 7790 | rtnl_lock(); |
0115e8e3 | 7791 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 7792 | __rtnl_unlock(); |
0115e8e3 | 7793 | |
b17a7c17 | 7794 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 7795 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
7796 | dev->name, dev->reg_state); |
7797 | dump_stack(); | |
7798 | continue; | |
7799 | } | |
1da177e4 | 7800 | |
b17a7c17 | 7801 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 7802 | |
b17a7c17 | 7803 | netdev_wait_allrefs(dev); |
1da177e4 | 7804 | |
b17a7c17 | 7805 | /* paranoia */ |
29b4433d | 7806 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
7807 | BUG_ON(!list_empty(&dev->ptype_all)); |
7808 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
7809 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
7810 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 7811 | WARN_ON(dev->dn_ptr); |
1da177e4 | 7812 | |
cf124db5 DM |
7813 | if (dev->priv_destructor) |
7814 | dev->priv_destructor(dev); | |
7815 | if (dev->needs_free_netdev) | |
7816 | free_netdev(dev); | |
9093bbb2 | 7817 | |
50624c93 EB |
7818 | /* Report a network device has been unregistered */ |
7819 | rtnl_lock(); | |
7820 | dev_net(dev)->dev_unreg_count--; | |
7821 | __rtnl_unlock(); | |
7822 | wake_up(&netdev_unregistering_wq); | |
7823 | ||
9093bbb2 SH |
7824 | /* Free network device */ |
7825 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 7826 | } |
1da177e4 LT |
7827 | } |
7828 | ||
9256645a JW |
7829 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
7830 | * all the same fields in the same order as net_device_stats, with only | |
7831 | * the type differing, but rtnl_link_stats64 may have additional fields | |
7832 | * at the end for newer counters. | |
3cfde79c | 7833 | */ |
77a1abf5 ED |
7834 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
7835 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
7836 | { |
7837 | #if BITS_PER_LONG == 64 | |
9256645a | 7838 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
9af9959e | 7839 | memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); |
9256645a JW |
7840 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7841 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
7842 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 7843 | #else |
9256645a | 7844 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
7845 | const unsigned long *src = (const unsigned long *)netdev_stats; |
7846 | u64 *dst = (u64 *)stats64; | |
7847 | ||
9256645a | 7848 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
7849 | for (i = 0; i < n; i++) |
7850 | dst[i] = src[i]; | |
9256645a JW |
7851 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7852 | memset((char *)stats64 + n * sizeof(u64), 0, | |
7853 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
7854 | #endif |
7855 | } | |
77a1abf5 | 7856 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 7857 | |
eeda3fd6 SH |
7858 | /** |
7859 | * dev_get_stats - get network device statistics | |
7860 | * @dev: device to get statistics from | |
28172739 | 7861 | * @storage: place to store stats |
eeda3fd6 | 7862 | * |
d7753516 BH |
7863 | * Get network statistics from device. Return @storage. |
7864 | * The device driver may provide its own method by setting | |
7865 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
7866 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 7867 | */ |
d7753516 BH |
7868 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
7869 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 7870 | { |
eeda3fd6 SH |
7871 | const struct net_device_ops *ops = dev->netdev_ops; |
7872 | ||
28172739 ED |
7873 | if (ops->ndo_get_stats64) { |
7874 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
7875 | ops->ndo_get_stats64(dev, storage); |
7876 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 7877 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
7878 | } else { |
7879 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 7880 | } |
6f64ec74 ED |
7881 | storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); |
7882 | storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); | |
7883 | storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); | |
28172739 | 7884 | return storage; |
c45d286e | 7885 | } |
eeda3fd6 | 7886 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 7887 | |
24824a09 | 7888 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 7889 | { |
24824a09 | 7890 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 7891 | |
24824a09 ED |
7892 | #ifdef CONFIG_NET_CLS_ACT |
7893 | if (queue) | |
7894 | return queue; | |
7895 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
7896 | if (!queue) | |
7897 | return NULL; | |
7898 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 7899 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
7900 | queue->qdisc_sleeping = &noop_qdisc; |
7901 | rcu_assign_pointer(dev->ingress_queue, queue); | |
7902 | #endif | |
7903 | return queue; | |
bb949fbd DM |
7904 | } |
7905 | ||
2c60db03 ED |
7906 | static const struct ethtool_ops default_ethtool_ops; |
7907 | ||
d07d7507 SG |
7908 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
7909 | const struct ethtool_ops *ops) | |
7910 | { | |
7911 | if (dev->ethtool_ops == &default_ethtool_ops) | |
7912 | dev->ethtool_ops = ops; | |
7913 | } | |
7914 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
7915 | ||
74d332c1 ED |
7916 | void netdev_freemem(struct net_device *dev) |
7917 | { | |
7918 | char *addr = (char *)dev - dev->padded; | |
7919 | ||
4cb28970 | 7920 | kvfree(addr); |
74d332c1 ED |
7921 | } |
7922 | ||
1da177e4 | 7923 | /** |
722c9a0c | 7924 | * alloc_netdev_mqs - allocate network device |
7925 | * @sizeof_priv: size of private data to allocate space for | |
7926 | * @name: device name format string | |
7927 | * @name_assign_type: origin of device name | |
7928 | * @setup: callback to initialize device | |
7929 | * @txqs: the number of TX subqueues to allocate | |
7930 | * @rxqs: the number of RX subqueues to allocate | |
7931 | * | |
7932 | * Allocates a struct net_device with private data area for driver use | |
7933 | * and performs basic initialization. Also allocates subqueue structs | |
7934 | * for each queue on the device. | |
1da177e4 | 7935 | */ |
36909ea4 | 7936 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 7937 | unsigned char name_assign_type, |
36909ea4 TH |
7938 | void (*setup)(struct net_device *), |
7939 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 7940 | { |
1da177e4 | 7941 | struct net_device *dev; |
7943986c | 7942 | size_t alloc_size; |
1ce8e7b5 | 7943 | struct net_device *p; |
1da177e4 | 7944 | |
b6fe17d6 SH |
7945 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
7946 | ||
36909ea4 | 7947 | if (txqs < 1) { |
7b6cd1ce | 7948 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
7949 | return NULL; |
7950 | } | |
7951 | ||
a953be53 | 7952 | #ifdef CONFIG_SYSFS |
36909ea4 | 7953 | if (rxqs < 1) { |
7b6cd1ce | 7954 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
7955 | return NULL; |
7956 | } | |
7957 | #endif | |
7958 | ||
fd2ea0a7 | 7959 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
7960 | if (sizeof_priv) { |
7961 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 7962 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
7963 | alloc_size += sizeof_priv; |
7964 | } | |
7965 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 7966 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 7967 | |
dcda9b04 | 7968 | p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
62b5942a | 7969 | if (!p) |
1da177e4 | 7970 | return NULL; |
1da177e4 | 7971 | |
1ce8e7b5 | 7972 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 7973 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 7974 | |
29b4433d ED |
7975 | dev->pcpu_refcnt = alloc_percpu(int); |
7976 | if (!dev->pcpu_refcnt) | |
74d332c1 | 7977 | goto free_dev; |
ab9c73cc | 7978 | |
ab9c73cc | 7979 | if (dev_addr_init(dev)) |
29b4433d | 7980 | goto free_pcpu; |
ab9c73cc | 7981 | |
22bedad3 | 7982 | dev_mc_init(dev); |
a748ee24 | 7983 | dev_uc_init(dev); |
ccffad25 | 7984 | |
c346dca1 | 7985 | dev_net_set(dev, &init_net); |
1da177e4 | 7986 | |
8d3bdbd5 | 7987 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 7988 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 7989 | |
8d3bdbd5 DM |
7990 | INIT_LIST_HEAD(&dev->napi_list); |
7991 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 7992 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 7993 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
7994 | INIT_LIST_HEAD(&dev->adj_list.upper); |
7995 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
7996 | INIT_LIST_HEAD(&dev->ptype_all); |
7997 | INIT_LIST_HEAD(&dev->ptype_specific); | |
59cc1f61 JK |
7998 | #ifdef CONFIG_NET_SCHED |
7999 | hash_init(dev->qdisc_hash); | |
8000 | #endif | |
02875878 | 8001 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
8002 | setup(dev); |
8003 | ||
a813104d | 8004 | if (!dev->tx_queue_len) { |
f84bb1ea | 8005 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 8006 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 8007 | } |
906470c1 | 8008 | |
36909ea4 TH |
8009 | dev->num_tx_queues = txqs; |
8010 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 8011 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 8012 | goto free_all; |
e8a0464c | 8013 | |
a953be53 | 8014 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
8015 | dev->num_rx_queues = rxqs; |
8016 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 8017 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 8018 | goto free_all; |
df334545 | 8019 | #endif |
0a9627f2 | 8020 | |
1da177e4 | 8021 | strcpy(dev->name, name); |
c835a677 | 8022 | dev->name_assign_type = name_assign_type; |
cbda10fa | 8023 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
8024 | if (!dev->ethtool_ops) |
8025 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 PN |
8026 | |
8027 | nf_hook_ingress_init(dev); | |
8028 | ||
1da177e4 | 8029 | return dev; |
ab9c73cc | 8030 | |
8d3bdbd5 DM |
8031 | free_all: |
8032 | free_netdev(dev); | |
8033 | return NULL; | |
8034 | ||
29b4433d ED |
8035 | free_pcpu: |
8036 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
8037 | free_dev: |
8038 | netdev_freemem(dev); | |
ab9c73cc | 8039 | return NULL; |
1da177e4 | 8040 | } |
36909ea4 | 8041 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
8042 | |
8043 | /** | |
722c9a0c | 8044 | * free_netdev - free network device |
8045 | * @dev: device | |
1da177e4 | 8046 | * |
722c9a0c | 8047 | * This function does the last stage of destroying an allocated device |
8048 | * interface. The reference to the device object is released. If this | |
8049 | * is the last reference then it will be freed.Must be called in process | |
8050 | * context. | |
1da177e4 LT |
8051 | */ |
8052 | void free_netdev(struct net_device *dev) | |
8053 | { | |
d565b0a1 | 8054 | struct napi_struct *p, *n; |
b5cdae32 | 8055 | struct bpf_prog *prog; |
d565b0a1 | 8056 | |
93d05d4a | 8057 | might_sleep(); |
60877a32 | 8058 | netif_free_tx_queues(dev); |
a953be53 | 8059 | #ifdef CONFIG_SYSFS |
10595902 | 8060 | kvfree(dev->_rx); |
fe822240 | 8061 | #endif |
e8a0464c | 8062 | |
33d480ce | 8063 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 8064 | |
f001fde5 JP |
8065 | /* Flush device addresses */ |
8066 | dev_addr_flush(dev); | |
8067 | ||
d565b0a1 HX |
8068 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
8069 | netif_napi_del(p); | |
8070 | ||
29b4433d ED |
8071 | free_percpu(dev->pcpu_refcnt); |
8072 | dev->pcpu_refcnt = NULL; | |
8073 | ||
b5cdae32 DM |
8074 | prog = rcu_dereference_protected(dev->xdp_prog, 1); |
8075 | if (prog) { | |
8076 | bpf_prog_put(prog); | |
8077 | static_key_slow_dec(&generic_xdp_needed); | |
8078 | } | |
8079 | ||
3041a069 | 8080 | /* Compatibility with error handling in drivers */ |
1da177e4 | 8081 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 8082 | netdev_freemem(dev); |
1da177e4 LT |
8083 | return; |
8084 | } | |
8085 | ||
8086 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
8087 | dev->reg_state = NETREG_RELEASED; | |
8088 | ||
43cb76d9 GKH |
8089 | /* will free via device release */ |
8090 | put_device(&dev->dev); | |
1da177e4 | 8091 | } |
d1b19dff | 8092 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 8093 | |
f0db275a SH |
8094 | /** |
8095 | * synchronize_net - Synchronize with packet receive processing | |
8096 | * | |
8097 | * Wait for packets currently being received to be done. | |
8098 | * Does not block later packets from starting. | |
8099 | */ | |
4ec93edb | 8100 | void synchronize_net(void) |
1da177e4 LT |
8101 | { |
8102 | might_sleep(); | |
be3fc413 ED |
8103 | if (rtnl_is_locked()) |
8104 | synchronize_rcu_expedited(); | |
8105 | else | |
8106 | synchronize_rcu(); | |
1da177e4 | 8107 | } |
d1b19dff | 8108 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
8109 | |
8110 | /** | |
44a0873d | 8111 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 8112 | * @dev: device |
44a0873d | 8113 | * @head: list |
6ebfbc06 | 8114 | * |
1da177e4 | 8115 | * This function shuts down a device interface and removes it |
d59b54b1 | 8116 | * from the kernel tables. |
44a0873d | 8117 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
8118 | * |
8119 | * Callers must hold the rtnl semaphore. You may want | |
8120 | * unregister_netdev() instead of this. | |
8121 | */ | |
8122 | ||
44a0873d | 8123 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 8124 | { |
a6620712 HX |
8125 | ASSERT_RTNL(); |
8126 | ||
44a0873d | 8127 | if (head) { |
9fdce099 | 8128 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
8129 | } else { |
8130 | rollback_registered(dev); | |
8131 | /* Finish processing unregister after unlock */ | |
8132 | net_set_todo(dev); | |
8133 | } | |
1da177e4 | 8134 | } |
44a0873d | 8135 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 8136 | |
9b5e383c ED |
8137 | /** |
8138 | * unregister_netdevice_many - unregister many devices | |
8139 | * @head: list of devices | |
87757a91 ED |
8140 | * |
8141 | * Note: As most callers use a stack allocated list_head, | |
8142 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
8143 | */ |
8144 | void unregister_netdevice_many(struct list_head *head) | |
8145 | { | |
8146 | struct net_device *dev; | |
8147 | ||
8148 | if (!list_empty(head)) { | |
8149 | rollback_registered_many(head); | |
8150 | list_for_each_entry(dev, head, unreg_list) | |
8151 | net_set_todo(dev); | |
87757a91 | 8152 | list_del(head); |
9b5e383c ED |
8153 | } |
8154 | } | |
63c8099d | 8155 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 8156 | |
1da177e4 LT |
8157 | /** |
8158 | * unregister_netdev - remove device from the kernel | |
8159 | * @dev: device | |
8160 | * | |
8161 | * This function shuts down a device interface and removes it | |
d59b54b1 | 8162 | * from the kernel tables. |
1da177e4 LT |
8163 | * |
8164 | * This is just a wrapper for unregister_netdevice that takes | |
8165 | * the rtnl semaphore. In general you want to use this and not | |
8166 | * unregister_netdevice. | |
8167 | */ | |
8168 | void unregister_netdev(struct net_device *dev) | |
8169 | { | |
8170 | rtnl_lock(); | |
8171 | unregister_netdevice(dev); | |
8172 | rtnl_unlock(); | |
8173 | } | |
1da177e4 LT |
8174 | EXPORT_SYMBOL(unregister_netdev); |
8175 | ||
ce286d32 EB |
8176 | /** |
8177 | * dev_change_net_namespace - move device to different nethost namespace | |
8178 | * @dev: device | |
8179 | * @net: network namespace | |
8180 | * @pat: If not NULL name pattern to try if the current device name | |
8181 | * is already taken in the destination network namespace. | |
8182 | * | |
8183 | * This function shuts down a device interface and moves it | |
8184 | * to a new network namespace. On success 0 is returned, on | |
8185 | * a failure a netagive errno code is returned. | |
8186 | * | |
8187 | * Callers must hold the rtnl semaphore. | |
8188 | */ | |
8189 | ||
8190 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
8191 | { | |
ce286d32 EB |
8192 | int err; |
8193 | ||
8194 | ASSERT_RTNL(); | |
8195 | ||
8196 | /* Don't allow namespace local devices to be moved. */ | |
8197 | err = -EINVAL; | |
8198 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8199 | goto out; | |
8200 | ||
8201 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
8202 | if (dev->reg_state != NETREG_REGISTERED) |
8203 | goto out; | |
8204 | ||
8205 | /* Get out if there is nothing todo */ | |
8206 | err = 0; | |
878628fb | 8207 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
8208 | goto out; |
8209 | ||
8210 | /* Pick the destination device name, and ensure | |
8211 | * we can use it in the destination network namespace. | |
8212 | */ | |
8213 | err = -EEXIST; | |
d9031024 | 8214 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
8215 | /* We get here if we can't use the current device name */ |
8216 | if (!pat) | |
8217 | goto out; | |
828de4f6 | 8218 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
8219 | goto out; |
8220 | } | |
8221 | ||
8222 | /* | |
8223 | * And now a mini version of register_netdevice unregister_netdevice. | |
8224 | */ | |
8225 | ||
8226 | /* If device is running close it first. */ | |
9b772652 | 8227 | dev_close(dev); |
ce286d32 EB |
8228 | |
8229 | /* And unlink it from device chain */ | |
8230 | err = -ENODEV; | |
8231 | unlist_netdevice(dev); | |
8232 | ||
8233 | synchronize_net(); | |
8234 | ||
8235 | /* Shutdown queueing discipline. */ | |
8236 | dev_shutdown(dev); | |
8237 | ||
8238 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 8239 | * this device. They should clean all the things. |
8240 | * | |
8241 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
8242 | * This is wanted because this way 8021q and macvlan know | |
8243 | * the device is just moving and can keep their slaves up. | |
8244 | */ | |
ce286d32 | 8245 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 G |
8246 | rcu_barrier(); |
8247 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 8248 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
8249 | |
8250 | /* | |
8251 | * Flush the unicast and multicast chains | |
8252 | */ | |
a748ee24 | 8253 | dev_uc_flush(dev); |
22bedad3 | 8254 | dev_mc_flush(dev); |
ce286d32 | 8255 | |
4e66ae2e SH |
8256 | /* Send a netdev-removed uevent to the old namespace */ |
8257 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 8258 | netdev_adjacent_del_links(dev); |
4e66ae2e | 8259 | |
ce286d32 | 8260 | /* Actually switch the network namespace */ |
c346dca1 | 8261 | dev_net_set(dev, net); |
ce286d32 | 8262 | |
ce286d32 | 8263 | /* If there is an ifindex conflict assign a new one */ |
7a66bbc9 | 8264 | if (__dev_get_by_index(net, dev->ifindex)) |
ce286d32 | 8265 | dev->ifindex = dev_new_index(net); |
ce286d32 | 8266 | |
4e66ae2e SH |
8267 | /* Send a netdev-add uevent to the new namespace */ |
8268 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 8269 | netdev_adjacent_add_links(dev); |
4e66ae2e | 8270 | |
8b41d188 | 8271 | /* Fixup kobjects */ |
a1b3f594 | 8272 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 8273 | WARN_ON(err); |
ce286d32 EB |
8274 | |
8275 | /* Add the device back in the hashes */ | |
8276 | list_netdevice(dev); | |
8277 | ||
8278 | /* Notify protocols, that a new device appeared. */ | |
8279 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
8280 | ||
d90a909e EB |
8281 | /* |
8282 | * Prevent userspace races by waiting until the network | |
8283 | * device is fully setup before sending notifications. | |
8284 | */ | |
7f294054 | 8285 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 8286 | |
ce286d32 EB |
8287 | synchronize_net(); |
8288 | err = 0; | |
8289 | out: | |
8290 | return err; | |
8291 | } | |
463d0183 | 8292 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 8293 | |
f0bf90de | 8294 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
8295 | { |
8296 | struct sk_buff **list_skb; | |
1da177e4 | 8297 | struct sk_buff *skb; |
f0bf90de | 8298 | unsigned int cpu; |
97d8b6e3 | 8299 | struct softnet_data *sd, *oldsd, *remsd = NULL; |
1da177e4 | 8300 | |
1da177e4 LT |
8301 | local_irq_disable(); |
8302 | cpu = smp_processor_id(); | |
8303 | sd = &per_cpu(softnet_data, cpu); | |
8304 | oldsd = &per_cpu(softnet_data, oldcpu); | |
8305 | ||
8306 | /* Find end of our completion_queue. */ | |
8307 | list_skb = &sd->completion_queue; | |
8308 | while (*list_skb) | |
8309 | list_skb = &(*list_skb)->next; | |
8310 | /* Append completion queue from offline CPU. */ | |
8311 | *list_skb = oldsd->completion_queue; | |
8312 | oldsd->completion_queue = NULL; | |
8313 | ||
1da177e4 | 8314 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
8315 | if (oldsd->output_queue) { |
8316 | *sd->output_queue_tailp = oldsd->output_queue; | |
8317 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
8318 | oldsd->output_queue = NULL; | |
8319 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
8320 | } | |
ac64da0b ED |
8321 | /* Append NAPI poll list from offline CPU, with one exception : |
8322 | * process_backlog() must be called by cpu owning percpu backlog. | |
8323 | * We properly handle process_queue & input_pkt_queue later. | |
8324 | */ | |
8325 | while (!list_empty(&oldsd->poll_list)) { | |
8326 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
8327 | struct napi_struct, | |
8328 | poll_list); | |
8329 | ||
8330 | list_del_init(&napi->poll_list); | |
8331 | if (napi->poll == process_backlog) | |
8332 | napi->state = 0; | |
8333 | else | |
8334 | ____napi_schedule(sd, napi); | |
264524d5 | 8335 | } |
1da177e4 LT |
8336 | |
8337 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
8338 | local_irq_enable(); | |
8339 | ||
773fc8f6 | 8340 | #ifdef CONFIG_RPS |
8341 | remsd = oldsd->rps_ipi_list; | |
8342 | oldsd->rps_ipi_list = NULL; | |
8343 | #endif | |
8344 | /* send out pending IPI's on offline CPU */ | |
8345 | net_rps_send_ipi(remsd); | |
8346 | ||
1da177e4 | 8347 | /* Process offline CPU's input_pkt_queue */ |
76cc8b13 | 8348 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 8349 | netif_rx_ni(skb); |
76cc8b13 | 8350 | input_queue_head_incr(oldsd); |
fec5e652 | 8351 | } |
ac64da0b | 8352 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 8353 | netif_rx_ni(skb); |
76cc8b13 TH |
8354 | input_queue_head_incr(oldsd); |
8355 | } | |
1da177e4 | 8356 | |
f0bf90de | 8357 | return 0; |
1da177e4 | 8358 | } |
1da177e4 | 8359 | |
7f353bf2 | 8360 | /** |
b63365a2 HX |
8361 | * netdev_increment_features - increment feature set by one |
8362 | * @all: current feature set | |
8363 | * @one: new feature set | |
8364 | * @mask: mask feature set | |
7f353bf2 HX |
8365 | * |
8366 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
8367 | * @one to the master device with current feature set @all. Will not |
8368 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 8369 | */ |
c8f44aff MM |
8370 | netdev_features_t netdev_increment_features(netdev_features_t all, |
8371 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 8372 | { |
c8cd0989 | 8373 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 8374 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 8375 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 8376 | |
a188222b | 8377 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 8378 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 8379 | |
1742f183 | 8380 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
8381 | if (all & NETIF_F_HW_CSUM) |
8382 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
8383 | |
8384 | return all; | |
8385 | } | |
b63365a2 | 8386 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 8387 | |
430f03cd | 8388 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
8389 | { |
8390 | int i; | |
8391 | struct hlist_head *hash; | |
8392 | ||
8393 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
8394 | if (hash != NULL) | |
8395 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
8396 | INIT_HLIST_HEAD(&hash[i]); | |
8397 | ||
8398 | return hash; | |
8399 | } | |
8400 | ||
881d966b | 8401 | /* Initialize per network namespace state */ |
4665079c | 8402 | static int __net_init netdev_init(struct net *net) |
881d966b | 8403 | { |
734b6541 RM |
8404 | if (net != &init_net) |
8405 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 8406 | |
30d97d35 PE |
8407 | net->dev_name_head = netdev_create_hash(); |
8408 | if (net->dev_name_head == NULL) | |
8409 | goto err_name; | |
881d966b | 8410 | |
30d97d35 PE |
8411 | net->dev_index_head = netdev_create_hash(); |
8412 | if (net->dev_index_head == NULL) | |
8413 | goto err_idx; | |
881d966b EB |
8414 | |
8415 | return 0; | |
30d97d35 PE |
8416 | |
8417 | err_idx: | |
8418 | kfree(net->dev_name_head); | |
8419 | err_name: | |
8420 | return -ENOMEM; | |
881d966b EB |
8421 | } |
8422 | ||
f0db275a SH |
8423 | /** |
8424 | * netdev_drivername - network driver for the device | |
8425 | * @dev: network device | |
f0db275a SH |
8426 | * |
8427 | * Determine network driver for device. | |
8428 | */ | |
3019de12 | 8429 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 8430 | { |
cf04a4c7 SH |
8431 | const struct device_driver *driver; |
8432 | const struct device *parent; | |
3019de12 | 8433 | const char *empty = ""; |
6579e57b AV |
8434 | |
8435 | parent = dev->dev.parent; | |
6579e57b | 8436 | if (!parent) |
3019de12 | 8437 | return empty; |
6579e57b AV |
8438 | |
8439 | driver = parent->driver; | |
8440 | if (driver && driver->name) | |
3019de12 DM |
8441 | return driver->name; |
8442 | return empty; | |
6579e57b AV |
8443 | } |
8444 | ||
6ea754eb JP |
8445 | static void __netdev_printk(const char *level, const struct net_device *dev, |
8446 | struct va_format *vaf) | |
256df2f3 | 8447 | { |
b004ff49 | 8448 | if (dev && dev->dev.parent) { |
6ea754eb JP |
8449 | dev_printk_emit(level[1] - '0', |
8450 | dev->dev.parent, | |
8451 | "%s %s %s%s: %pV", | |
8452 | dev_driver_string(dev->dev.parent), | |
8453 | dev_name(dev->dev.parent), | |
8454 | netdev_name(dev), netdev_reg_state(dev), | |
8455 | vaf); | |
b004ff49 | 8456 | } else if (dev) { |
6ea754eb JP |
8457 | printk("%s%s%s: %pV", |
8458 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 8459 | } else { |
6ea754eb | 8460 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 8461 | } |
256df2f3 JP |
8462 | } |
8463 | ||
6ea754eb JP |
8464 | void netdev_printk(const char *level, const struct net_device *dev, |
8465 | const char *format, ...) | |
256df2f3 JP |
8466 | { |
8467 | struct va_format vaf; | |
8468 | va_list args; | |
256df2f3 JP |
8469 | |
8470 | va_start(args, format); | |
8471 | ||
8472 | vaf.fmt = format; | |
8473 | vaf.va = &args; | |
8474 | ||
6ea754eb | 8475 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 8476 | |
256df2f3 | 8477 | va_end(args); |
256df2f3 JP |
8478 | } |
8479 | EXPORT_SYMBOL(netdev_printk); | |
8480 | ||
8481 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 8482 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 8483 | { \ |
256df2f3 JP |
8484 | struct va_format vaf; \ |
8485 | va_list args; \ | |
8486 | \ | |
8487 | va_start(args, fmt); \ | |
8488 | \ | |
8489 | vaf.fmt = fmt; \ | |
8490 | vaf.va = &args; \ | |
8491 | \ | |
6ea754eb | 8492 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 8493 | \ |
256df2f3 | 8494 | va_end(args); \ |
256df2f3 JP |
8495 | } \ |
8496 | EXPORT_SYMBOL(func); | |
8497 | ||
8498 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
8499 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
8500 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
8501 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
8502 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
8503 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
8504 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
8505 | ||
4665079c | 8506 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
8507 | { |
8508 | kfree(net->dev_name_head); | |
8509 | kfree(net->dev_index_head); | |
8510 | } | |
8511 | ||
022cbae6 | 8512 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
8513 | .init = netdev_init, |
8514 | .exit = netdev_exit, | |
8515 | }; | |
8516 | ||
4665079c | 8517 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 8518 | { |
e008b5fc | 8519 | struct net_device *dev, *aux; |
ce286d32 | 8520 | /* |
e008b5fc | 8521 | * Push all migratable network devices back to the |
ce286d32 EB |
8522 | * initial network namespace |
8523 | */ | |
8524 | rtnl_lock(); | |
e008b5fc | 8525 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 8526 | int err; |
aca51397 | 8527 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
8528 | |
8529 | /* Ignore unmoveable devices (i.e. loopback) */ | |
8530 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8531 | continue; | |
8532 | ||
e008b5fc EB |
8533 | /* Leave virtual devices for the generic cleanup */ |
8534 | if (dev->rtnl_link_ops) | |
8535 | continue; | |
d0c082ce | 8536 | |
25985edc | 8537 | /* Push remaining network devices to init_net */ |
aca51397 PE |
8538 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
8539 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 8540 | if (err) { |
7b6cd1ce JP |
8541 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
8542 | __func__, dev->name, err); | |
aca51397 | 8543 | BUG(); |
ce286d32 EB |
8544 | } |
8545 | } | |
8546 | rtnl_unlock(); | |
8547 | } | |
8548 | ||
50624c93 EB |
8549 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
8550 | { | |
8551 | /* Return with the rtnl_lock held when there are no network | |
8552 | * devices unregistering in any network namespace in net_list. | |
8553 | */ | |
8554 | struct net *net; | |
8555 | bool unregistering; | |
ff960a73 | 8556 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 8557 | |
ff960a73 | 8558 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 8559 | for (;;) { |
50624c93 EB |
8560 | unregistering = false; |
8561 | rtnl_lock(); | |
8562 | list_for_each_entry(net, net_list, exit_list) { | |
8563 | if (net->dev_unreg_count > 0) { | |
8564 | unregistering = true; | |
8565 | break; | |
8566 | } | |
8567 | } | |
8568 | if (!unregistering) | |
8569 | break; | |
8570 | __rtnl_unlock(); | |
ff960a73 PZ |
8571 | |
8572 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 8573 | } |
ff960a73 | 8574 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
8575 | } |
8576 | ||
04dc7f6b EB |
8577 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
8578 | { | |
8579 | /* At exit all network devices most be removed from a network | |
b595076a | 8580 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
8581 | * Do this across as many network namespaces as possible to |
8582 | * improve batching efficiency. | |
8583 | */ | |
8584 | struct net_device *dev; | |
8585 | struct net *net; | |
8586 | LIST_HEAD(dev_kill_list); | |
8587 | ||
50624c93 EB |
8588 | /* To prevent network device cleanup code from dereferencing |
8589 | * loopback devices or network devices that have been freed | |
8590 | * wait here for all pending unregistrations to complete, | |
8591 | * before unregistring the loopback device and allowing the | |
8592 | * network namespace be freed. | |
8593 | * | |
8594 | * The netdev todo list containing all network devices | |
8595 | * unregistrations that happen in default_device_exit_batch | |
8596 | * will run in the rtnl_unlock() at the end of | |
8597 | * default_device_exit_batch. | |
8598 | */ | |
8599 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
8600 | list_for_each_entry(net, net_list, exit_list) { |
8601 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 8602 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
8603 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
8604 | else | |
8605 | unregister_netdevice_queue(dev, &dev_kill_list); | |
8606 | } | |
8607 | } | |
8608 | unregister_netdevice_many(&dev_kill_list); | |
8609 | rtnl_unlock(); | |
8610 | } | |
8611 | ||
022cbae6 | 8612 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 8613 | .exit = default_device_exit, |
04dc7f6b | 8614 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
8615 | }; |
8616 | ||
1da177e4 LT |
8617 | /* |
8618 | * Initialize the DEV module. At boot time this walks the device list and | |
8619 | * unhooks any devices that fail to initialise (normally hardware not | |
8620 | * present) and leaves us with a valid list of present and active devices. | |
8621 | * | |
8622 | */ | |
8623 | ||
8624 | /* | |
8625 | * This is called single threaded during boot, so no need | |
8626 | * to take the rtnl semaphore. | |
8627 | */ | |
8628 | static int __init net_dev_init(void) | |
8629 | { | |
8630 | int i, rc = -ENOMEM; | |
8631 | ||
8632 | BUG_ON(!dev_boot_phase); | |
8633 | ||
1da177e4 LT |
8634 | if (dev_proc_init()) |
8635 | goto out; | |
8636 | ||
8b41d188 | 8637 | if (netdev_kobject_init()) |
1da177e4 LT |
8638 | goto out; |
8639 | ||
8640 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 8641 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
8642 | INIT_LIST_HEAD(&ptype_base[i]); |
8643 | ||
62532da9 VY |
8644 | INIT_LIST_HEAD(&offload_base); |
8645 | ||
881d966b EB |
8646 | if (register_pernet_subsys(&netdev_net_ops)) |
8647 | goto out; | |
1da177e4 LT |
8648 | |
8649 | /* | |
8650 | * Initialise the packet receive queues. | |
8651 | */ | |
8652 | ||
6f912042 | 8653 | for_each_possible_cpu(i) { |
41852497 | 8654 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 8655 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 8656 | |
41852497 ED |
8657 | INIT_WORK(flush, flush_backlog); |
8658 | ||
e36fa2f7 | 8659 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 8660 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 8661 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 8662 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 8663 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
8664 | sd->csd.func = rps_trigger_softirq; |
8665 | sd->csd.info = sd; | |
e36fa2f7 | 8666 | sd->cpu = i; |
1e94d72f | 8667 | #endif |
0a9627f2 | 8668 | |
e36fa2f7 ED |
8669 | sd->backlog.poll = process_backlog; |
8670 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
8671 | } |
8672 | ||
1da177e4 LT |
8673 | dev_boot_phase = 0; |
8674 | ||
505d4f73 EB |
8675 | /* The loopback device is special if any other network devices |
8676 | * is present in a network namespace the loopback device must | |
8677 | * be present. Since we now dynamically allocate and free the | |
8678 | * loopback device ensure this invariant is maintained by | |
8679 | * keeping the loopback device as the first device on the | |
8680 | * list of network devices. Ensuring the loopback devices | |
8681 | * is the first device that appears and the last network device | |
8682 | * that disappears. | |
8683 | */ | |
8684 | if (register_pernet_device(&loopback_net_ops)) | |
8685 | goto out; | |
8686 | ||
8687 | if (register_pernet_device(&default_device_ops)) | |
8688 | goto out; | |
8689 | ||
962cf36c CM |
8690 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
8691 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 8692 | |
f0bf90de SAS |
8693 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
8694 | NULL, dev_cpu_dead); | |
8695 | WARN_ON(rc < 0); | |
1da177e4 LT |
8696 | rc = 0; |
8697 | out: | |
8698 | return rc; | |
8699 | } | |
8700 | ||
8701 | subsys_initcall(net_dev_init); |