]>
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> |
ae847f40 | 147 | #include <net/udp_tunnel.h> |
6621dd29 | 148 | #include <linux/net_namespace.h> |
1da177e4 | 149 | |
342709ef PE |
150 | #include "net-sysfs.h" |
151 | ||
d565b0a1 HX |
152 | /* Instead of increasing this, you should create a hash table. */ |
153 | #define MAX_GRO_SKBS 8 | |
154 | ||
5d38a079 HX |
155 | /* This should be increased if a protocol with a bigger head is added. */ |
156 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
157 | ||
1da177e4 | 158 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 159 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
160 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
161 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 162 | static struct list_head offload_base __read_mostly; |
1da177e4 | 163 | |
ae78dbfa | 164 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 | 165 | static int call_netdevice_notifiers_info(unsigned long val, |
54951194 | 166 | struct netdev_notifier_info *info); |
90b602f8 | 167 | static struct napi_struct *napi_by_id(unsigned int napi_id); |
ae78dbfa | 168 | |
1da177e4 | 169 | /* |
7562f876 | 170 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
171 | * semaphore. |
172 | * | |
c6d14c84 | 173 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
174 | * |
175 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 176 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
177 | * actual updates. This allows pure readers to access the list even |
178 | * while a writer is preparing to update it. | |
179 | * | |
180 | * To put it another way, dev_base_lock is held for writing only to | |
181 | * protect against pure readers; the rtnl semaphore provides the | |
182 | * protection against other writers. | |
183 | * | |
184 | * See, for example usages, register_netdevice() and | |
185 | * unregister_netdevice(), which must be called with the rtnl | |
186 | * semaphore held. | |
187 | */ | |
1da177e4 | 188 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
189 | EXPORT_SYMBOL(dev_base_lock); |
190 | ||
6c557001 FW |
191 | static DEFINE_MUTEX(ifalias_mutex); |
192 | ||
af12fa6e ET |
193 | /* protects napi_hash addition/deletion and napi_gen_id */ |
194 | static DEFINE_SPINLOCK(napi_hash_lock); | |
195 | ||
52bd2d62 | 196 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 197 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 198 | |
18afa4b0 | 199 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 200 | |
4e985ada TG |
201 | static inline void dev_base_seq_inc(struct net *net) |
202 | { | |
643aa9cb | 203 | while (++net->dev_base_seq == 0) |
204 | ; | |
4e985ada TG |
205 | } |
206 | ||
881d966b | 207 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 208 | { |
8387ff25 | 209 | unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); |
95c96174 | 210 | |
08e9897d | 211 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
212 | } |
213 | ||
881d966b | 214 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 215 | { |
7c28bd0b | 216 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
217 | } |
218 | ||
e36fa2f7 | 219 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
220 | { |
221 | #ifdef CONFIG_RPS | |
e36fa2f7 | 222 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
223 | #endif |
224 | } | |
225 | ||
e36fa2f7 | 226 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
227 | { |
228 | #ifdef CONFIG_RPS | |
e36fa2f7 | 229 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
230 | #endif |
231 | } | |
232 | ||
ce286d32 | 233 | /* Device list insertion */ |
53759be9 | 234 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 235 | { |
c346dca1 | 236 | struct net *net = dev_net(dev); |
ce286d32 EB |
237 | |
238 | ASSERT_RTNL(); | |
239 | ||
240 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 241 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 242 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
243 | hlist_add_head_rcu(&dev->index_hlist, |
244 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 245 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
246 | |
247 | dev_base_seq_inc(net); | |
ce286d32 EB |
248 | } |
249 | ||
fb699dfd ED |
250 | /* Device list removal |
251 | * caller must respect a RCU grace period before freeing/reusing dev | |
252 | */ | |
ce286d32 EB |
253 | static void unlist_netdevice(struct net_device *dev) |
254 | { | |
255 | ASSERT_RTNL(); | |
256 | ||
257 | /* Unlink dev from the device chain */ | |
258 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 259 | list_del_rcu(&dev->dev_list); |
72c9528b | 260 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 261 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 262 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
263 | |
264 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
265 | } |
266 | ||
1da177e4 LT |
267 | /* |
268 | * Our notifier list | |
269 | */ | |
270 | ||
f07d5b94 | 271 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
272 | |
273 | /* | |
274 | * Device drivers call our routines to queue packets here. We empty the | |
275 | * queue in the local softnet handler. | |
276 | */ | |
bea3348e | 277 | |
9958da05 | 278 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 279 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 280 | |
cf508b12 | 281 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 282 | /* |
c773e847 | 283 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
284 | * according to dev->type |
285 | */ | |
643aa9cb | 286 | static const unsigned short netdev_lock_type[] = { |
287 | ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
723e98b7 JP |
288 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, |
289 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
290 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
291 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
292 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
293 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
294 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
295 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
296 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
297 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
298 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
299 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
300 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
301 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 302 | |
643aa9cb | 303 | static const char *const netdev_lock_name[] = { |
304 | "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
305 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
306 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
307 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
308 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
309 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
310 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
311 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
312 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
313 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
314 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
315 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
316 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", | |
317 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
318 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
319 | |
320 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 321 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
322 | |
323 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
324 | { | |
325 | int i; | |
326 | ||
327 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
328 | if (netdev_lock_type[i] == dev_type) | |
329 | return i; | |
330 | /* the last key is used by default */ | |
331 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
332 | } | |
333 | ||
cf508b12 DM |
334 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
335 | unsigned short dev_type) | |
723e98b7 JP |
336 | { |
337 | int i; | |
338 | ||
339 | i = netdev_lock_pos(dev_type); | |
340 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
341 | netdev_lock_name[i]); | |
342 | } | |
cf508b12 DM |
343 | |
344 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
345 | { | |
346 | int i; | |
347 | ||
348 | i = netdev_lock_pos(dev->type); | |
349 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
350 | &netdev_addr_lock_key[i], | |
351 | netdev_lock_name[i]); | |
352 | } | |
723e98b7 | 353 | #else |
cf508b12 DM |
354 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
355 | unsigned short dev_type) | |
356 | { | |
357 | } | |
358 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
359 | { |
360 | } | |
361 | #endif | |
1da177e4 LT |
362 | |
363 | /******************************************************************************* | |
eb13da1a | 364 | * |
365 | * Protocol management and registration routines | |
366 | * | |
367 | *******************************************************************************/ | |
1da177e4 | 368 | |
1da177e4 | 369 | |
1da177e4 LT |
370 | /* |
371 | * Add a protocol ID to the list. Now that the input handler is | |
372 | * smarter we can dispense with all the messy stuff that used to be | |
373 | * here. | |
374 | * | |
375 | * BEWARE!!! Protocol handlers, mangling input packets, | |
376 | * MUST BE last in hash buckets and checking protocol handlers | |
377 | * MUST start from promiscuous ptype_all chain in net_bh. | |
378 | * It is true now, do not change it. | |
379 | * Explanation follows: if protocol handler, mangling packet, will | |
380 | * be the first on list, it is not able to sense, that packet | |
381 | * is cloned and should be copied-on-write, so that it will | |
382 | * change it and subsequent readers will get broken packet. | |
383 | * --ANK (980803) | |
384 | */ | |
385 | ||
c07b68e8 ED |
386 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
387 | { | |
388 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 389 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 390 | else |
7866a621 SN |
391 | return pt->dev ? &pt->dev->ptype_specific : |
392 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
393 | } |
394 | ||
1da177e4 LT |
395 | /** |
396 | * dev_add_pack - add packet handler | |
397 | * @pt: packet type declaration | |
398 | * | |
399 | * Add a protocol handler to the networking stack. The passed &packet_type | |
400 | * is linked into kernel lists and may not be freed until it has been | |
401 | * removed from the kernel lists. | |
402 | * | |
4ec93edb | 403 | * This call does not sleep therefore it can not |
1da177e4 LT |
404 | * guarantee all CPU's that are in middle of receiving packets |
405 | * will see the new packet type (until the next received packet). | |
406 | */ | |
407 | ||
408 | void dev_add_pack(struct packet_type *pt) | |
409 | { | |
c07b68e8 | 410 | struct list_head *head = ptype_head(pt); |
1da177e4 | 411 | |
c07b68e8 ED |
412 | spin_lock(&ptype_lock); |
413 | list_add_rcu(&pt->list, head); | |
414 | spin_unlock(&ptype_lock); | |
1da177e4 | 415 | } |
d1b19dff | 416 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 417 | |
1da177e4 LT |
418 | /** |
419 | * __dev_remove_pack - remove packet handler | |
420 | * @pt: packet type declaration | |
421 | * | |
422 | * Remove a protocol handler that was previously added to the kernel | |
423 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
424 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 425 | * returns. |
1da177e4 LT |
426 | * |
427 | * The packet type might still be in use by receivers | |
428 | * and must not be freed until after all the CPU's have gone | |
429 | * through a quiescent state. | |
430 | */ | |
431 | void __dev_remove_pack(struct packet_type *pt) | |
432 | { | |
c07b68e8 | 433 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
434 | struct packet_type *pt1; |
435 | ||
c07b68e8 | 436 | spin_lock(&ptype_lock); |
1da177e4 LT |
437 | |
438 | list_for_each_entry(pt1, head, list) { | |
439 | if (pt == pt1) { | |
440 | list_del_rcu(&pt->list); | |
441 | goto out; | |
442 | } | |
443 | } | |
444 | ||
7b6cd1ce | 445 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 446 | out: |
c07b68e8 | 447 | spin_unlock(&ptype_lock); |
1da177e4 | 448 | } |
d1b19dff ED |
449 | EXPORT_SYMBOL(__dev_remove_pack); |
450 | ||
1da177e4 LT |
451 | /** |
452 | * dev_remove_pack - remove packet handler | |
453 | * @pt: packet type declaration | |
454 | * | |
455 | * Remove a protocol handler that was previously added to the kernel | |
456 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
457 | * from the kernel lists and can be freed or reused once this function | |
458 | * returns. | |
459 | * | |
460 | * This call sleeps to guarantee that no CPU is looking at the packet | |
461 | * type after return. | |
462 | */ | |
463 | void dev_remove_pack(struct packet_type *pt) | |
464 | { | |
465 | __dev_remove_pack(pt); | |
4ec93edb | 466 | |
1da177e4 LT |
467 | synchronize_net(); |
468 | } | |
d1b19dff | 469 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 470 | |
62532da9 VY |
471 | |
472 | /** | |
473 | * dev_add_offload - register offload handlers | |
474 | * @po: protocol offload declaration | |
475 | * | |
476 | * Add protocol offload handlers to the networking stack. The passed | |
477 | * &proto_offload is linked into kernel lists and may not be freed until | |
478 | * it has been removed from the kernel lists. | |
479 | * | |
480 | * This call does not sleep therefore it can not | |
481 | * guarantee all CPU's that are in middle of receiving packets | |
482 | * will see the new offload handlers (until the next received packet). | |
483 | */ | |
484 | void dev_add_offload(struct packet_offload *po) | |
485 | { | |
bdef7de4 | 486 | struct packet_offload *elem; |
62532da9 VY |
487 | |
488 | spin_lock(&offload_lock); | |
bdef7de4 DM |
489 | list_for_each_entry(elem, &offload_base, list) { |
490 | if (po->priority < elem->priority) | |
491 | break; | |
492 | } | |
493 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
494 | spin_unlock(&offload_lock); |
495 | } | |
496 | EXPORT_SYMBOL(dev_add_offload); | |
497 | ||
498 | /** | |
499 | * __dev_remove_offload - remove offload handler | |
500 | * @po: packet offload declaration | |
501 | * | |
502 | * Remove a protocol offload handler that was previously added to the | |
503 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
504 | * is removed from the kernel lists and can be freed or reused once this | |
505 | * function returns. | |
506 | * | |
507 | * The packet type might still be in use by receivers | |
508 | * and must not be freed until after all the CPU's have gone | |
509 | * through a quiescent state. | |
510 | */ | |
1d143d9f | 511 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
512 | { |
513 | struct list_head *head = &offload_base; | |
514 | struct packet_offload *po1; | |
515 | ||
c53aa505 | 516 | spin_lock(&offload_lock); |
62532da9 VY |
517 | |
518 | list_for_each_entry(po1, head, list) { | |
519 | if (po == po1) { | |
520 | list_del_rcu(&po->list); | |
521 | goto out; | |
522 | } | |
523 | } | |
524 | ||
525 | pr_warn("dev_remove_offload: %p not found\n", po); | |
526 | out: | |
c53aa505 | 527 | spin_unlock(&offload_lock); |
62532da9 | 528 | } |
62532da9 VY |
529 | |
530 | /** | |
531 | * dev_remove_offload - remove packet offload handler | |
532 | * @po: packet offload declaration | |
533 | * | |
534 | * Remove a packet offload handler that was previously added to the kernel | |
535 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
536 | * removed from the kernel lists and can be freed or reused once this | |
537 | * function returns. | |
538 | * | |
539 | * This call sleeps to guarantee that no CPU is looking at the packet | |
540 | * type after return. | |
541 | */ | |
542 | void dev_remove_offload(struct packet_offload *po) | |
543 | { | |
544 | __dev_remove_offload(po); | |
545 | ||
546 | synchronize_net(); | |
547 | } | |
548 | EXPORT_SYMBOL(dev_remove_offload); | |
549 | ||
1da177e4 | 550 | /****************************************************************************** |
eb13da1a | 551 | * |
552 | * Device Boot-time Settings Routines | |
553 | * | |
554 | ******************************************************************************/ | |
1da177e4 LT |
555 | |
556 | /* Boot time configuration table */ | |
557 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
558 | ||
559 | /** | |
560 | * netdev_boot_setup_add - add new setup entry | |
561 | * @name: name of the device | |
562 | * @map: configured settings for the device | |
563 | * | |
564 | * Adds new setup entry to the dev_boot_setup list. The function | |
565 | * returns 0 on error and 1 on success. This is a generic routine to | |
566 | * all netdevices. | |
567 | */ | |
568 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
569 | { | |
570 | struct netdev_boot_setup *s; | |
571 | int i; | |
572 | ||
573 | s = dev_boot_setup; | |
574 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
575 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
576 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 577 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
578 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
579 | break; | |
580 | } | |
581 | } | |
582 | ||
583 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
584 | } | |
585 | ||
586 | /** | |
722c9a0c | 587 | * netdev_boot_setup_check - check boot time settings |
588 | * @dev: the netdevice | |
1da177e4 | 589 | * |
722c9a0c | 590 | * Check boot time settings for the device. |
591 | * The found settings are set for the device to be used | |
592 | * later in the device probing. | |
593 | * Returns 0 if no settings found, 1 if they are. | |
1da177e4 LT |
594 | */ |
595 | int netdev_boot_setup_check(struct net_device *dev) | |
596 | { | |
597 | struct netdev_boot_setup *s = dev_boot_setup; | |
598 | int i; | |
599 | ||
600 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
601 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 602 | !strcmp(dev->name, s[i].name)) { |
722c9a0c | 603 | dev->irq = s[i].map.irq; |
604 | dev->base_addr = s[i].map.base_addr; | |
605 | dev->mem_start = s[i].map.mem_start; | |
606 | dev->mem_end = s[i].map.mem_end; | |
1da177e4 LT |
607 | return 1; |
608 | } | |
609 | } | |
610 | return 0; | |
611 | } | |
d1b19dff | 612 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
613 | |
614 | ||
615 | /** | |
722c9a0c | 616 | * netdev_boot_base - get address from boot time settings |
617 | * @prefix: prefix for network device | |
618 | * @unit: id for network device | |
619 | * | |
620 | * Check boot time settings for the base address of device. | |
621 | * The found settings are set for the device to be used | |
622 | * later in the device probing. | |
623 | * Returns 0 if no settings found. | |
1da177e4 LT |
624 | */ |
625 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
626 | { | |
627 | const struct netdev_boot_setup *s = dev_boot_setup; | |
628 | char name[IFNAMSIZ]; | |
629 | int i; | |
630 | ||
631 | sprintf(name, "%s%d", prefix, unit); | |
632 | ||
633 | /* | |
634 | * If device already registered then return base of 1 | |
635 | * to indicate not to probe for this interface | |
636 | */ | |
881d966b | 637 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
638 | return 1; |
639 | ||
640 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
641 | if (!strcmp(name, s[i].name)) | |
642 | return s[i].map.base_addr; | |
643 | return 0; | |
644 | } | |
645 | ||
646 | /* | |
647 | * Saves at boot time configured settings for any netdevice. | |
648 | */ | |
649 | int __init netdev_boot_setup(char *str) | |
650 | { | |
651 | int ints[5]; | |
652 | struct ifmap map; | |
653 | ||
654 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
655 | if (!str || !*str) | |
656 | return 0; | |
657 | ||
658 | /* Save settings */ | |
659 | memset(&map, 0, sizeof(map)); | |
660 | if (ints[0] > 0) | |
661 | map.irq = ints[1]; | |
662 | if (ints[0] > 1) | |
663 | map.base_addr = ints[2]; | |
664 | if (ints[0] > 2) | |
665 | map.mem_start = ints[3]; | |
666 | if (ints[0] > 3) | |
667 | map.mem_end = ints[4]; | |
668 | ||
669 | /* Add new entry to the list */ | |
670 | return netdev_boot_setup_add(str, &map); | |
671 | } | |
672 | ||
673 | __setup("netdev=", netdev_boot_setup); | |
674 | ||
675 | /******************************************************************************* | |
eb13da1a | 676 | * |
677 | * Device Interface Subroutines | |
678 | * | |
679 | *******************************************************************************/ | |
1da177e4 | 680 | |
a54acb3a ND |
681 | /** |
682 | * dev_get_iflink - get 'iflink' value of a interface | |
683 | * @dev: targeted interface | |
684 | * | |
685 | * Indicates the ifindex the interface is linked to. | |
686 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
687 | */ | |
688 | ||
689 | int dev_get_iflink(const struct net_device *dev) | |
690 | { | |
691 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
692 | return dev->netdev_ops->ndo_get_iflink(dev); | |
693 | ||
7a66bbc9 | 694 | return dev->ifindex; |
a54acb3a ND |
695 | } |
696 | EXPORT_SYMBOL(dev_get_iflink); | |
697 | ||
fc4099f1 PS |
698 | /** |
699 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
700 | * @dev: targeted interface | |
701 | * @skb: The packet. | |
702 | * | |
703 | * For better visibility of tunnel traffic OVS needs to retrieve | |
704 | * egress tunnel information for a packet. Following API allows | |
705 | * user to get this info. | |
706 | */ | |
707 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
708 | { | |
709 | struct ip_tunnel_info *info; | |
710 | ||
711 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
712 | return -EINVAL; | |
713 | ||
714 | info = skb_tunnel_info_unclone(skb); | |
715 | if (!info) | |
716 | return -ENOMEM; | |
717 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
718 | return -EINVAL; | |
719 | ||
720 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
721 | } | |
722 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
723 | ||
1da177e4 LT |
724 | /** |
725 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 726 | * @net: the applicable net namespace |
1da177e4 LT |
727 | * @name: name to find |
728 | * | |
729 | * Find an interface by name. Must be called under RTNL semaphore | |
730 | * or @dev_base_lock. If the name is found a pointer to the device | |
731 | * is returned. If the name is not found then %NULL is returned. The | |
732 | * reference counters are not incremented so the caller must be | |
733 | * careful with locks. | |
734 | */ | |
735 | ||
881d966b | 736 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 737 | { |
0bd8d536 ED |
738 | struct net_device *dev; |
739 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 740 | |
b67bfe0d | 741 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
742 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
743 | return dev; | |
0bd8d536 | 744 | |
1da177e4 LT |
745 | return NULL; |
746 | } | |
d1b19dff | 747 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 748 | |
72c9528b | 749 | /** |
722c9a0c | 750 | * dev_get_by_name_rcu - find a device by its name |
751 | * @net: the applicable net namespace | |
752 | * @name: name to find | |
753 | * | |
754 | * Find an interface by name. | |
755 | * If the name is found a pointer to the device is returned. | |
756 | * If the name is not found then %NULL is returned. | |
757 | * The reference counters are not incremented so the caller must be | |
758 | * careful with locks. The caller must hold RCU lock. | |
72c9528b ED |
759 | */ |
760 | ||
761 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
762 | { | |
72c9528b ED |
763 | struct net_device *dev; |
764 | struct hlist_head *head = dev_name_hash(net, name); | |
765 | ||
b67bfe0d | 766 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
767 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
768 | return dev; | |
769 | ||
770 | return NULL; | |
771 | } | |
772 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
773 | ||
1da177e4 LT |
774 | /** |
775 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 776 | * @net: the applicable net namespace |
1da177e4 LT |
777 | * @name: name to find |
778 | * | |
779 | * Find an interface by name. This can be called from any | |
780 | * context and does its own locking. The returned handle has | |
781 | * the usage count incremented and the caller must use dev_put() to | |
782 | * release it when it is no longer needed. %NULL is returned if no | |
783 | * matching device is found. | |
784 | */ | |
785 | ||
881d966b | 786 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
787 | { |
788 | struct net_device *dev; | |
789 | ||
72c9528b ED |
790 | rcu_read_lock(); |
791 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
792 | if (dev) |
793 | dev_hold(dev); | |
72c9528b | 794 | rcu_read_unlock(); |
1da177e4 LT |
795 | return dev; |
796 | } | |
d1b19dff | 797 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
798 | |
799 | /** | |
800 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 801 | * @net: the applicable net namespace |
1da177e4 LT |
802 | * @ifindex: index of device |
803 | * | |
804 | * Search for an interface by index. Returns %NULL if the device | |
805 | * is not found or a pointer to the device. The device has not | |
806 | * had its reference counter increased so the caller must be careful | |
807 | * about locking. The caller must hold either the RTNL semaphore | |
808 | * or @dev_base_lock. | |
809 | */ | |
810 | ||
881d966b | 811 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 812 | { |
0bd8d536 ED |
813 | struct net_device *dev; |
814 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 815 | |
b67bfe0d | 816 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
817 | if (dev->ifindex == ifindex) |
818 | return dev; | |
0bd8d536 | 819 | |
1da177e4 LT |
820 | return NULL; |
821 | } | |
d1b19dff | 822 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 823 | |
fb699dfd ED |
824 | /** |
825 | * dev_get_by_index_rcu - find a device by its ifindex | |
826 | * @net: the applicable net namespace | |
827 | * @ifindex: index of device | |
828 | * | |
829 | * Search for an interface by index. Returns %NULL if the device | |
830 | * is not found or a pointer to the device. The device has not | |
831 | * had its reference counter increased so the caller must be careful | |
832 | * about locking. The caller must hold RCU lock. | |
833 | */ | |
834 | ||
835 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
836 | { | |
fb699dfd ED |
837 | struct net_device *dev; |
838 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
839 | ||
b67bfe0d | 840 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
841 | if (dev->ifindex == ifindex) |
842 | return dev; | |
843 | ||
844 | return NULL; | |
845 | } | |
846 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
847 | ||
1da177e4 LT |
848 | |
849 | /** | |
850 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 851 | * @net: the applicable net namespace |
1da177e4 LT |
852 | * @ifindex: index of device |
853 | * | |
854 | * Search for an interface by index. Returns NULL if the device | |
855 | * is not found or a pointer to the device. The device returned has | |
856 | * had a reference added and the pointer is safe until the user calls | |
857 | * dev_put to indicate they have finished with it. | |
858 | */ | |
859 | ||
881d966b | 860 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
861 | { |
862 | struct net_device *dev; | |
863 | ||
fb699dfd ED |
864 | rcu_read_lock(); |
865 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
866 | if (dev) |
867 | dev_hold(dev); | |
fb699dfd | 868 | rcu_read_unlock(); |
1da177e4 LT |
869 | return dev; |
870 | } | |
d1b19dff | 871 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 872 | |
90b602f8 ML |
873 | /** |
874 | * dev_get_by_napi_id - find a device by napi_id | |
875 | * @napi_id: ID of the NAPI struct | |
876 | * | |
877 | * Search for an interface by NAPI ID. Returns %NULL if the device | |
878 | * is not found or a pointer to the device. The device has not had | |
879 | * its reference counter increased so the caller must be careful | |
880 | * about locking. The caller must hold RCU lock. | |
881 | */ | |
882 | ||
883 | struct net_device *dev_get_by_napi_id(unsigned int napi_id) | |
884 | { | |
885 | struct napi_struct *napi; | |
886 | ||
887 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
888 | ||
889 | if (napi_id < MIN_NAPI_ID) | |
890 | return NULL; | |
891 | ||
892 | napi = napi_by_id(napi_id); | |
893 | ||
894 | return napi ? napi->dev : NULL; | |
895 | } | |
896 | EXPORT_SYMBOL(dev_get_by_napi_id); | |
897 | ||
5dbe7c17 NS |
898 | /** |
899 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
900 | * @net: network namespace | |
901 | * @name: a pointer to the buffer where the name will be stored. | |
902 | * @ifindex: the ifindex of the interface to get the name from. | |
903 | * | |
904 | * The use of raw_seqcount_begin() and cond_resched() before | |
905 | * retrying is required as we want to give the writers a chance | |
906 | * to complete when CONFIG_PREEMPT is not set. | |
907 | */ | |
908 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
909 | { | |
910 | struct net_device *dev; | |
911 | unsigned int seq; | |
912 | ||
913 | retry: | |
914 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
915 | rcu_read_lock(); | |
916 | dev = dev_get_by_index_rcu(net, ifindex); | |
917 | if (!dev) { | |
918 | rcu_read_unlock(); | |
919 | return -ENODEV; | |
920 | } | |
921 | ||
922 | strcpy(name, dev->name); | |
923 | rcu_read_unlock(); | |
924 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
925 | cond_resched(); | |
926 | goto retry; | |
927 | } | |
928 | ||
929 | return 0; | |
930 | } | |
931 | ||
1da177e4 | 932 | /** |
941666c2 | 933 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 934 | * @net: the applicable net namespace |
1da177e4 LT |
935 | * @type: media type of device |
936 | * @ha: hardware address | |
937 | * | |
938 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
939 | * is not found or a pointer to the device. |
940 | * The caller must hold RCU or RTNL. | |
941666c2 | 941 | * The returned device has not had its ref count increased |
1da177e4 LT |
942 | * and the caller must therefore be careful about locking |
943 | * | |
1da177e4 LT |
944 | */ |
945 | ||
941666c2 ED |
946 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
947 | const char *ha) | |
1da177e4 LT |
948 | { |
949 | struct net_device *dev; | |
950 | ||
941666c2 | 951 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
952 | if (dev->type == type && |
953 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
954 | return dev; |
955 | ||
956 | return NULL; | |
1da177e4 | 957 | } |
941666c2 | 958 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 959 | |
881d966b | 960 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
961 | { |
962 | struct net_device *dev; | |
963 | ||
4e9cac2b | 964 | ASSERT_RTNL(); |
881d966b | 965 | for_each_netdev(net, dev) |
4e9cac2b | 966 | if (dev->type == type) |
7562f876 PE |
967 | return dev; |
968 | ||
969 | return NULL; | |
4e9cac2b | 970 | } |
4e9cac2b PM |
971 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
972 | ||
881d966b | 973 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 974 | { |
99fe3c39 | 975 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 976 | |
99fe3c39 ED |
977 | rcu_read_lock(); |
978 | for_each_netdev_rcu(net, dev) | |
979 | if (dev->type == type) { | |
980 | dev_hold(dev); | |
981 | ret = dev; | |
982 | break; | |
983 | } | |
984 | rcu_read_unlock(); | |
985 | return ret; | |
1da177e4 | 986 | } |
1da177e4 LT |
987 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
988 | ||
989 | /** | |
6c555490 | 990 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 991 | * @net: the applicable net namespace |
1da177e4 LT |
992 | * @if_flags: IFF_* values |
993 | * @mask: bitmask of bits in if_flags to check | |
994 | * | |
995 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 996 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 997 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
998 | */ |
999 | ||
6c555490 WC |
1000 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
1001 | unsigned short mask) | |
1da177e4 | 1002 | { |
7562f876 | 1003 | struct net_device *dev, *ret; |
1da177e4 | 1004 | |
6c555490 WC |
1005 | ASSERT_RTNL(); |
1006 | ||
7562f876 | 1007 | ret = NULL; |
6c555490 | 1008 | for_each_netdev(net, dev) { |
1da177e4 | 1009 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 1010 | ret = dev; |
1da177e4 LT |
1011 | break; |
1012 | } | |
1013 | } | |
7562f876 | 1014 | return ret; |
1da177e4 | 1015 | } |
6c555490 | 1016 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
1017 | |
1018 | /** | |
1019 | * dev_valid_name - check if name is okay for network device | |
1020 | * @name: name string | |
1021 | * | |
1022 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
1023 | * to allow sysfs to work. We also disallow any kind of |
1024 | * whitespace. | |
1da177e4 | 1025 | */ |
95f050bf | 1026 | bool dev_valid_name(const char *name) |
1da177e4 | 1027 | { |
c7fa9d18 | 1028 | if (*name == '\0') |
95f050bf | 1029 | return false; |
b6fe17d6 | 1030 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 1031 | return false; |
c7fa9d18 | 1032 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1033 | return false; |
c7fa9d18 DM |
1034 | |
1035 | while (*name) { | |
a4176a93 | 1036 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1037 | return false; |
c7fa9d18 DM |
1038 | name++; |
1039 | } | |
95f050bf | 1040 | return true; |
1da177e4 | 1041 | } |
d1b19dff | 1042 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1043 | |
1044 | /** | |
b267b179 EB |
1045 | * __dev_alloc_name - allocate a name for a device |
1046 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1047 | * @name: name format string |
b267b179 | 1048 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1049 | * |
1050 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1051 | * id. It scans list of devices to build up a free map, then chooses |
1052 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1053 | * while allocating the name and adding the device in order to avoid | |
1054 | * duplicates. | |
1055 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1056 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1057 | */ |
1058 | ||
b267b179 | 1059 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1060 | { |
1061 | int i = 0; | |
1da177e4 LT |
1062 | const char *p; |
1063 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1064 | unsigned long *inuse; |
1da177e4 LT |
1065 | struct net_device *d; |
1066 | ||
1067 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
1068 | if (p) { | |
1069 | /* | |
1070 | * Verify the string as this thing may have come from | |
1071 | * the user. There must be either one "%d" and no other "%" | |
1072 | * characters. | |
1073 | */ | |
1074 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1075 | return -EINVAL; | |
1076 | ||
1077 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1078 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1079 | if (!inuse) |
1080 | return -ENOMEM; | |
1081 | ||
881d966b | 1082 | for_each_netdev(net, d) { |
1da177e4 LT |
1083 | if (!sscanf(d->name, name, &i)) |
1084 | continue; | |
1085 | if (i < 0 || i >= max_netdevices) | |
1086 | continue; | |
1087 | ||
1088 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1089 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1090 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1091 | set_bit(i, inuse); | |
1092 | } | |
1093 | ||
1094 | i = find_first_zero_bit(inuse, max_netdevices); | |
1095 | free_page((unsigned long) inuse); | |
1096 | } | |
1097 | ||
d9031024 OP |
1098 | if (buf != name) |
1099 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1100 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1101 | return i; |
1da177e4 LT |
1102 | |
1103 | /* It is possible to run out of possible slots | |
1104 | * when the name is long and there isn't enough space left | |
1105 | * for the digits, or if all bits are used. | |
1106 | */ | |
1107 | return -ENFILE; | |
1108 | } | |
1109 | ||
b267b179 EB |
1110 | /** |
1111 | * dev_alloc_name - allocate a name for a device | |
1112 | * @dev: device | |
1113 | * @name: name format string | |
1114 | * | |
1115 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1116 | * id. It scans list of devices to build up a free map, then chooses | |
1117 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1118 | * while allocating the name and adding the device in order to avoid | |
1119 | * duplicates. | |
1120 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1121 | * Returns the number of the unit assigned or a negative errno code. | |
1122 | */ | |
1123 | ||
1124 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1125 | { | |
1126 | char buf[IFNAMSIZ]; | |
1127 | struct net *net; | |
1128 | int ret; | |
1129 | ||
c346dca1 YH |
1130 | BUG_ON(!dev_net(dev)); |
1131 | net = dev_net(dev); | |
b267b179 EB |
1132 | ret = __dev_alloc_name(net, name, buf); |
1133 | if (ret >= 0) | |
1134 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1135 | return ret; | |
1136 | } | |
d1b19dff | 1137 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1138 | |
828de4f6 G |
1139 | static int dev_alloc_name_ns(struct net *net, |
1140 | struct net_device *dev, | |
1141 | const char *name) | |
d9031024 | 1142 | { |
828de4f6 G |
1143 | char buf[IFNAMSIZ]; |
1144 | int ret; | |
8ce6cebc | 1145 | |
828de4f6 G |
1146 | ret = __dev_alloc_name(net, name, buf); |
1147 | if (ret >= 0) | |
1148 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1149 | return ret; | |
1150 | } | |
1151 | ||
0ad646c8 CW |
1152 | int dev_get_valid_name(struct net *net, struct net_device *dev, |
1153 | const char *name) | |
828de4f6 G |
1154 | { |
1155 | BUG_ON(!net); | |
8ce6cebc | 1156 | |
d9031024 OP |
1157 | if (!dev_valid_name(name)) |
1158 | return -EINVAL; | |
1159 | ||
1c5cae81 | 1160 | if (strchr(name, '%')) |
828de4f6 | 1161 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1162 | else if (__dev_get_by_name(net, name)) |
1163 | return -EEXIST; | |
8ce6cebc DL |
1164 | else if (dev->name != name) |
1165 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1166 | |
1167 | return 0; | |
1168 | } | |
0ad646c8 | 1169 | EXPORT_SYMBOL(dev_get_valid_name); |
1da177e4 LT |
1170 | |
1171 | /** | |
1172 | * dev_change_name - change name of a device | |
1173 | * @dev: device | |
1174 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1175 | * | |
1176 | * Change name of a device, can pass format strings "eth%d". | |
1177 | * for wildcarding. | |
1178 | */ | |
cf04a4c7 | 1179 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1180 | { |
238fa362 | 1181 | unsigned char old_assign_type; |
fcc5a03a | 1182 | char oldname[IFNAMSIZ]; |
1da177e4 | 1183 | int err = 0; |
fcc5a03a | 1184 | int ret; |
881d966b | 1185 | struct net *net; |
1da177e4 LT |
1186 | |
1187 | ASSERT_RTNL(); | |
c346dca1 | 1188 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1189 | |
c346dca1 | 1190 | net = dev_net(dev); |
1da177e4 LT |
1191 | if (dev->flags & IFF_UP) |
1192 | return -EBUSY; | |
1193 | ||
30e6c9fa | 1194 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1195 | |
1196 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1197 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1198 | return 0; |
c91f6df2 | 1199 | } |
c8d90dca | 1200 | |
fcc5a03a HX |
1201 | memcpy(oldname, dev->name, IFNAMSIZ); |
1202 | ||
828de4f6 | 1203 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1204 | if (err < 0) { |
30e6c9fa | 1205 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1206 | return err; |
c91f6df2 | 1207 | } |
1da177e4 | 1208 | |
6fe82a39 VF |
1209 | if (oldname[0] && !strchr(oldname, '%')) |
1210 | netdev_info(dev, "renamed from %s\n", oldname); | |
1211 | ||
238fa362 TG |
1212 | old_assign_type = dev->name_assign_type; |
1213 | dev->name_assign_type = NET_NAME_RENAMED; | |
1214 | ||
fcc5a03a | 1215 | rollback: |
a1b3f594 EB |
1216 | ret = device_rename(&dev->dev, dev->name); |
1217 | if (ret) { | |
1218 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1219 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1220 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1221 | return ret; |
dcc99773 | 1222 | } |
7f988eab | 1223 | |
30e6c9fa | 1224 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1225 | |
5bb025fa VF |
1226 | netdev_adjacent_rename_links(dev, oldname); |
1227 | ||
7f988eab | 1228 | write_lock_bh(&dev_base_lock); |
372b2312 | 1229 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1230 | write_unlock_bh(&dev_base_lock); |
1231 | ||
1232 | synchronize_rcu(); | |
1233 | ||
1234 | write_lock_bh(&dev_base_lock); | |
1235 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1236 | write_unlock_bh(&dev_base_lock); |
1237 | ||
056925ab | 1238 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1239 | ret = notifier_to_errno(ret); |
1240 | ||
1241 | if (ret) { | |
91e9c07b ED |
1242 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1243 | if (err >= 0) { | |
fcc5a03a | 1244 | err = ret; |
30e6c9fa | 1245 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1246 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1247 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1248 | dev->name_assign_type = old_assign_type; |
1249 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1250 | goto rollback; |
91e9c07b | 1251 | } else { |
7b6cd1ce | 1252 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1253 | dev->name, ret); |
fcc5a03a HX |
1254 | } |
1255 | } | |
1da177e4 LT |
1256 | |
1257 | return err; | |
1258 | } | |
1259 | ||
0b815a1a SH |
1260 | /** |
1261 | * dev_set_alias - change ifalias of a device | |
1262 | * @dev: device | |
1263 | * @alias: name up to IFALIASZ | |
f0db275a | 1264 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1265 | * |
1266 | * Set ifalias for a device, | |
1267 | */ | |
1268 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1269 | { | |
6c557001 | 1270 | struct dev_ifalias *new_alias = NULL; |
0b815a1a SH |
1271 | |
1272 | if (len >= IFALIASZ) | |
1273 | return -EINVAL; | |
1274 | ||
6c557001 FW |
1275 | if (len) { |
1276 | new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); | |
1277 | if (!new_alias) | |
1278 | return -ENOMEM; | |
1279 | ||
1280 | memcpy(new_alias->ifalias, alias, len); | |
1281 | new_alias->ifalias[len] = 0; | |
96ca4a2c OH |
1282 | } |
1283 | ||
6c557001 FW |
1284 | mutex_lock(&ifalias_mutex); |
1285 | rcu_swap_protected(dev->ifalias, new_alias, | |
1286 | mutex_is_locked(&ifalias_mutex)); | |
1287 | mutex_unlock(&ifalias_mutex); | |
1288 | ||
1289 | if (new_alias) | |
1290 | kfree_rcu(new_alias, rcuhead); | |
0b815a1a | 1291 | |
0b815a1a SH |
1292 | return len; |
1293 | } | |
1294 | ||
6c557001 FW |
1295 | /** |
1296 | * dev_get_alias - get ifalias of a device | |
1297 | * @dev: device | |
20e88320 | 1298 | * @name: buffer to store name of ifalias |
6c557001 FW |
1299 | * @len: size of buffer |
1300 | * | |
1301 | * get ifalias for a device. Caller must make sure dev cannot go | |
1302 | * away, e.g. rcu read lock or own a reference count to device. | |
1303 | */ | |
1304 | int dev_get_alias(const struct net_device *dev, char *name, size_t len) | |
1305 | { | |
1306 | const struct dev_ifalias *alias; | |
1307 | int ret = 0; | |
1308 | ||
1309 | rcu_read_lock(); | |
1310 | alias = rcu_dereference(dev->ifalias); | |
1311 | if (alias) | |
1312 | ret = snprintf(name, len, "%s", alias->ifalias); | |
1313 | rcu_read_unlock(); | |
1314 | ||
1315 | return ret; | |
1316 | } | |
0b815a1a | 1317 | |
d8a33ac4 | 1318 | /** |
3041a069 | 1319 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1320 | * @dev: device to cause notification |
1321 | * | |
1322 | * Called to indicate a device has changed features. | |
1323 | */ | |
1324 | void netdev_features_change(struct net_device *dev) | |
1325 | { | |
056925ab | 1326 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1327 | } |
1328 | EXPORT_SYMBOL(netdev_features_change); | |
1329 | ||
1da177e4 LT |
1330 | /** |
1331 | * netdev_state_change - device changes state | |
1332 | * @dev: device to cause notification | |
1333 | * | |
1334 | * Called to indicate a device has changed state. This function calls | |
1335 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1336 | * to the routing socket. | |
1337 | */ | |
1338 | void netdev_state_change(struct net_device *dev) | |
1339 | { | |
1340 | if (dev->flags & IFF_UP) { | |
51d0c047 DA |
1341 | struct netdev_notifier_change_info change_info = { |
1342 | .info.dev = dev, | |
1343 | }; | |
54951194 | 1344 | |
51d0c047 | 1345 | call_netdevice_notifiers_info(NETDEV_CHANGE, |
54951194 | 1346 | &change_info.info); |
7f294054 | 1347 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1348 | } |
1349 | } | |
d1b19dff | 1350 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1351 | |
ee89bab1 | 1352 | /** |
722c9a0c | 1353 | * netdev_notify_peers - notify network peers about existence of @dev |
1354 | * @dev: network device | |
ee89bab1 AW |
1355 | * |
1356 | * Generate traffic such that interested network peers are aware of | |
1357 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1358 | * a device wants to inform the rest of the network about some sort of | |
1359 | * reconfiguration such as a failover event or virtual machine | |
1360 | * migration. | |
1361 | */ | |
1362 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1363 | { |
ee89bab1 AW |
1364 | rtnl_lock(); |
1365 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
37c343b4 | 1366 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); |
ee89bab1 | 1367 | rtnl_unlock(); |
c1da4ac7 | 1368 | } |
ee89bab1 | 1369 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1370 | |
bd380811 | 1371 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1372 | { |
d314774c | 1373 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1374 | int ret; |
1da177e4 | 1375 | |
e46b66bc BH |
1376 | ASSERT_RTNL(); |
1377 | ||
1da177e4 LT |
1378 | if (!netif_device_present(dev)) |
1379 | return -ENODEV; | |
1380 | ||
ca99ca14 NH |
1381 | /* Block netpoll from trying to do any rx path servicing. |
1382 | * If we don't do this there is a chance ndo_poll_controller | |
1383 | * or ndo_poll may be running while we open the device | |
1384 | */ | |
66b5552f | 1385 | netpoll_poll_disable(dev); |
ca99ca14 | 1386 | |
3b8bcfd5 JB |
1387 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1388 | ret = notifier_to_errno(ret); | |
1389 | if (ret) | |
1390 | return ret; | |
1391 | ||
1da177e4 | 1392 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1393 | |
d314774c SH |
1394 | if (ops->ndo_validate_addr) |
1395 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1396 | |
d314774c SH |
1397 | if (!ret && ops->ndo_open) |
1398 | ret = ops->ndo_open(dev); | |
1da177e4 | 1399 | |
66b5552f | 1400 | netpoll_poll_enable(dev); |
ca99ca14 | 1401 | |
bada339b JG |
1402 | if (ret) |
1403 | clear_bit(__LINK_STATE_START, &dev->state); | |
1404 | else { | |
1da177e4 | 1405 | dev->flags |= IFF_UP; |
4417da66 | 1406 | dev_set_rx_mode(dev); |
1da177e4 | 1407 | dev_activate(dev); |
7bf23575 | 1408 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1409 | } |
bada339b | 1410 | |
1da177e4 LT |
1411 | return ret; |
1412 | } | |
1413 | ||
1414 | /** | |
bd380811 PM |
1415 | * dev_open - prepare an interface for use. |
1416 | * @dev: device to open | |
1da177e4 | 1417 | * |
bd380811 PM |
1418 | * Takes a device from down to up state. The device's private open |
1419 | * function is invoked and then the multicast lists are loaded. Finally | |
1420 | * the device is moved into the up state and a %NETDEV_UP message is | |
1421 | * sent to the netdev notifier chain. | |
1422 | * | |
1423 | * Calling this function on an active interface is a nop. On a failure | |
1424 | * a negative errno code is returned. | |
1da177e4 | 1425 | */ |
bd380811 PM |
1426 | int dev_open(struct net_device *dev) |
1427 | { | |
1428 | int ret; | |
1429 | ||
bd380811 PM |
1430 | if (dev->flags & IFF_UP) |
1431 | return 0; | |
1432 | ||
bd380811 PM |
1433 | ret = __dev_open(dev); |
1434 | if (ret < 0) | |
1435 | return ret; | |
1436 | ||
7f294054 | 1437 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1438 | call_netdevice_notifiers(NETDEV_UP, dev); |
1439 | ||
1440 | return ret; | |
1441 | } | |
1442 | EXPORT_SYMBOL(dev_open); | |
1443 | ||
7051b88a | 1444 | static void __dev_close_many(struct list_head *head) |
1da177e4 | 1445 | { |
44345724 | 1446 | struct net_device *dev; |
e46b66bc | 1447 | |
bd380811 | 1448 | ASSERT_RTNL(); |
9d5010db DM |
1449 | might_sleep(); |
1450 | ||
5cde2829 | 1451 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1452 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1453 | netpoll_poll_disable(dev); |
3f4df206 | 1454 | |
44345724 | 1455 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1456 | |
44345724 | 1457 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1458 | |
44345724 OP |
1459 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1460 | * can be even on different cpu. So just clear netif_running(). | |
1461 | * | |
1462 | * dev->stop() will invoke napi_disable() on all of it's | |
1463 | * napi_struct instances on this device. | |
1464 | */ | |
4e857c58 | 1465 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1466 | } |
1da177e4 | 1467 | |
44345724 | 1468 | dev_deactivate_many(head); |
d8b2a4d2 | 1469 | |
5cde2829 | 1470 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1471 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1472 | |
44345724 OP |
1473 | /* |
1474 | * Call the device specific close. This cannot fail. | |
1475 | * Only if device is UP | |
1476 | * | |
1477 | * We allow it to be called even after a DETACH hot-plug | |
1478 | * event. | |
1479 | */ | |
1480 | if (ops->ndo_stop) | |
1481 | ops->ndo_stop(dev); | |
1482 | ||
44345724 | 1483 | dev->flags &= ~IFF_UP; |
66b5552f | 1484 | netpoll_poll_enable(dev); |
44345724 | 1485 | } |
44345724 OP |
1486 | } |
1487 | ||
7051b88a | 1488 | static void __dev_close(struct net_device *dev) |
44345724 OP |
1489 | { |
1490 | LIST_HEAD(single); | |
1491 | ||
5cde2829 | 1492 | list_add(&dev->close_list, &single); |
7051b88a | 1493 | __dev_close_many(&single); |
f87e6f47 | 1494 | list_del(&single); |
44345724 OP |
1495 | } |
1496 | ||
7051b88a | 1497 | void dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1498 | { |
1499 | struct net_device *dev, *tmp; | |
1da177e4 | 1500 | |
5cde2829 EB |
1501 | /* Remove the devices that don't need to be closed */ |
1502 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1503 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1504 | list_del_init(&dev->close_list); |
44345724 OP |
1505 | |
1506 | __dev_close_many(head); | |
1da177e4 | 1507 | |
5cde2829 | 1508 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1509 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1510 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1511 | if (unlink) |
1512 | list_del_init(&dev->close_list); | |
44345724 | 1513 | } |
bd380811 | 1514 | } |
99c4a26a | 1515 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1516 | |
1517 | /** | |
1518 | * dev_close - shutdown an interface. | |
1519 | * @dev: device to shutdown | |
1520 | * | |
1521 | * This function moves an active device into down state. A | |
1522 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1523 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1524 | * chain. | |
1525 | */ | |
7051b88a | 1526 | void dev_close(struct net_device *dev) |
bd380811 | 1527 | { |
e14a5993 ED |
1528 | if (dev->flags & IFF_UP) { |
1529 | LIST_HEAD(single); | |
1da177e4 | 1530 | |
5cde2829 | 1531 | list_add(&dev->close_list, &single); |
99c4a26a | 1532 | dev_close_many(&single, true); |
e14a5993 ED |
1533 | list_del(&single); |
1534 | } | |
1da177e4 | 1535 | } |
d1b19dff | 1536 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1537 | |
1538 | ||
0187bdfb BH |
1539 | /** |
1540 | * dev_disable_lro - disable Large Receive Offload on a device | |
1541 | * @dev: device | |
1542 | * | |
1543 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1544 | * called under RTNL. This is needed if received packets may be | |
1545 | * forwarded to another interface. | |
1546 | */ | |
1547 | void dev_disable_lro(struct net_device *dev) | |
1548 | { | |
fbe168ba MK |
1549 | struct net_device *lower_dev; |
1550 | struct list_head *iter; | |
529d0489 | 1551 | |
bc5787c6 MM |
1552 | dev->wanted_features &= ~NETIF_F_LRO; |
1553 | netdev_update_features(dev); | |
27660515 | 1554 | |
22d5969f MM |
1555 | if (unlikely(dev->features & NETIF_F_LRO)) |
1556 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1557 | |
1558 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1559 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1560 | } |
1561 | EXPORT_SYMBOL(dev_disable_lro); | |
1562 | ||
351638e7 JP |
1563 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1564 | struct net_device *dev) | |
1565 | { | |
51d0c047 DA |
1566 | struct netdev_notifier_info info = { |
1567 | .dev = dev, | |
1568 | }; | |
351638e7 | 1569 | |
351638e7 JP |
1570 | return nb->notifier_call(nb, val, &info); |
1571 | } | |
0187bdfb | 1572 | |
881d966b EB |
1573 | static int dev_boot_phase = 1; |
1574 | ||
1da177e4 | 1575 | /** |
722c9a0c | 1576 | * register_netdevice_notifier - register a network notifier block |
1577 | * @nb: notifier | |
1da177e4 | 1578 | * |
722c9a0c | 1579 | * Register a notifier to be called when network device events occur. |
1580 | * The notifier passed is linked into the kernel structures and must | |
1581 | * not be reused until it has been unregistered. A negative errno code | |
1582 | * is returned on a failure. | |
1da177e4 | 1583 | * |
722c9a0c | 1584 | * When registered all registration and up events are replayed |
1585 | * to the new notifier to allow device to have a race free | |
1586 | * view of the network device list. | |
1da177e4 LT |
1587 | */ |
1588 | ||
1589 | int register_netdevice_notifier(struct notifier_block *nb) | |
1590 | { | |
1591 | struct net_device *dev; | |
fcc5a03a | 1592 | struct net_device *last; |
881d966b | 1593 | struct net *net; |
1da177e4 LT |
1594 | int err; |
1595 | ||
1596 | rtnl_lock(); | |
f07d5b94 | 1597 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1598 | if (err) |
1599 | goto unlock; | |
881d966b EB |
1600 | if (dev_boot_phase) |
1601 | goto unlock; | |
1602 | for_each_net(net) { | |
1603 | for_each_netdev(net, dev) { | |
351638e7 | 1604 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1605 | err = notifier_to_errno(err); |
1606 | if (err) | |
1607 | goto rollback; | |
1608 | ||
1609 | if (!(dev->flags & IFF_UP)) | |
1610 | continue; | |
1da177e4 | 1611 | |
351638e7 | 1612 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1613 | } |
1da177e4 | 1614 | } |
fcc5a03a HX |
1615 | |
1616 | unlock: | |
1da177e4 LT |
1617 | rtnl_unlock(); |
1618 | return err; | |
fcc5a03a HX |
1619 | |
1620 | rollback: | |
1621 | last = dev; | |
881d966b EB |
1622 | for_each_net(net) { |
1623 | for_each_netdev(net, dev) { | |
1624 | if (dev == last) | |
8f891489 | 1625 | goto outroll; |
fcc5a03a | 1626 | |
881d966b | 1627 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1628 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1629 | dev); | |
1630 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1631 | } |
351638e7 | 1632 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1633 | } |
fcc5a03a | 1634 | } |
c67625a1 | 1635 | |
8f891489 | 1636 | outroll: |
c67625a1 | 1637 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1638 | goto unlock; |
1da177e4 | 1639 | } |
d1b19dff | 1640 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1641 | |
1642 | /** | |
722c9a0c | 1643 | * unregister_netdevice_notifier - unregister a network notifier block |
1644 | * @nb: notifier | |
1da177e4 | 1645 | * |
722c9a0c | 1646 | * Unregister a notifier previously registered by |
1647 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1648 | * kernel structures and may then be reused. A negative errno code | |
1649 | * is returned on a failure. | |
7d3d43da | 1650 | * |
722c9a0c | 1651 | * After unregistering unregister and down device events are synthesized |
1652 | * for all devices on the device list to the removed notifier to remove | |
1653 | * the need for special case cleanup code. | |
1da177e4 LT |
1654 | */ |
1655 | ||
1656 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1657 | { | |
7d3d43da EB |
1658 | struct net_device *dev; |
1659 | struct net *net; | |
9f514950 HX |
1660 | int err; |
1661 | ||
1662 | rtnl_lock(); | |
f07d5b94 | 1663 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1664 | if (err) |
1665 | goto unlock; | |
1666 | ||
1667 | for_each_net(net) { | |
1668 | for_each_netdev(net, dev) { | |
1669 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1670 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1671 | dev); | |
1672 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1673 | } |
351638e7 | 1674 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1675 | } |
1676 | } | |
1677 | unlock: | |
9f514950 HX |
1678 | rtnl_unlock(); |
1679 | return err; | |
1da177e4 | 1680 | } |
d1b19dff | 1681 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1682 | |
351638e7 JP |
1683 | /** |
1684 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1685 | * @val: value passed unmodified to notifier function | |
1686 | * @dev: net_device pointer passed unmodified to notifier function | |
1687 | * @info: notifier information data | |
1688 | * | |
1689 | * Call all network notifier blocks. Parameters and return value | |
1690 | * are as for raw_notifier_call_chain(). | |
1691 | */ | |
1692 | ||
1d143d9f | 1693 | static int call_netdevice_notifiers_info(unsigned long val, |
1d143d9f | 1694 | struct netdev_notifier_info *info) |
351638e7 JP |
1695 | { |
1696 | ASSERT_RTNL(); | |
351638e7 JP |
1697 | return raw_notifier_call_chain(&netdev_chain, val, info); |
1698 | } | |
351638e7 | 1699 | |
1da177e4 LT |
1700 | /** |
1701 | * call_netdevice_notifiers - call all network notifier blocks | |
1702 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1703 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1704 | * |
1705 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1706 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1707 | */ |
1708 | ||
ad7379d4 | 1709 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1710 | { |
51d0c047 DA |
1711 | struct netdev_notifier_info info = { |
1712 | .dev = dev, | |
1713 | }; | |
351638e7 | 1714 | |
51d0c047 | 1715 | return call_netdevice_notifiers_info(val, &info); |
1da177e4 | 1716 | } |
edf947f1 | 1717 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1718 | |
1cf51900 | 1719 | #ifdef CONFIG_NET_INGRESS |
4577139b DB |
1720 | static struct static_key ingress_needed __read_mostly; |
1721 | ||
1722 | void net_inc_ingress_queue(void) | |
1723 | { | |
1724 | static_key_slow_inc(&ingress_needed); | |
1725 | } | |
1726 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
1727 | ||
1728 | void net_dec_ingress_queue(void) | |
1729 | { | |
1730 | static_key_slow_dec(&ingress_needed); | |
1731 | } | |
1732 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
1733 | #endif | |
1734 | ||
1f211a1b DB |
1735 | #ifdef CONFIG_NET_EGRESS |
1736 | static struct static_key egress_needed __read_mostly; | |
1737 | ||
1738 | void net_inc_egress_queue(void) | |
1739 | { | |
1740 | static_key_slow_inc(&egress_needed); | |
1741 | } | |
1742 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
1743 | ||
1744 | void net_dec_egress_queue(void) | |
1745 | { | |
1746 | static_key_slow_dec(&egress_needed); | |
1747 | } | |
1748 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
1749 | #endif | |
1750 | ||
c5905afb | 1751 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1752 | #ifdef HAVE_JUMP_LABEL |
b90e5794 | 1753 | static atomic_t netstamp_needed_deferred; |
13baa00a | 1754 | static atomic_t netstamp_wanted; |
5fa8bbda | 1755 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 1756 | { |
b90e5794 | 1757 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 1758 | int wanted; |
b90e5794 | 1759 | |
13baa00a ED |
1760 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
1761 | if (wanted > 0) | |
1762 | static_key_enable(&netstamp_needed); | |
1763 | else | |
1764 | static_key_disable(&netstamp_needed); | |
5fa8bbda ED |
1765 | } |
1766 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 1767 | #endif |
5fa8bbda ED |
1768 | |
1769 | void net_enable_timestamp(void) | |
1770 | { | |
13baa00a ED |
1771 | #ifdef HAVE_JUMP_LABEL |
1772 | int wanted; | |
1773 | ||
1774 | while (1) { | |
1775 | wanted = atomic_read(&netstamp_wanted); | |
1776 | if (wanted <= 0) | |
1777 | break; | |
1778 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) | |
1779 | return; | |
1780 | } | |
1781 | atomic_inc(&netstamp_needed_deferred); | |
1782 | schedule_work(&netstamp_work); | |
1783 | #else | |
c5905afb | 1784 | static_key_slow_inc(&netstamp_needed); |
13baa00a | 1785 | #endif |
1da177e4 | 1786 | } |
d1b19dff | 1787 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1788 | |
1789 | void net_disable_timestamp(void) | |
1790 | { | |
b90e5794 | 1791 | #ifdef HAVE_JUMP_LABEL |
13baa00a ED |
1792 | int wanted; |
1793 | ||
1794 | while (1) { | |
1795 | wanted = atomic_read(&netstamp_wanted); | |
1796 | if (wanted <= 1) | |
1797 | break; | |
1798 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) | |
1799 | return; | |
1800 | } | |
1801 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
1802 | schedule_work(&netstamp_work); |
1803 | #else | |
c5905afb | 1804 | static_key_slow_dec(&netstamp_needed); |
5fa8bbda | 1805 | #endif |
1da177e4 | 1806 | } |
d1b19dff | 1807 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1808 | |
3b098e2d | 1809 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1810 | { |
2456e855 | 1811 | skb->tstamp = 0; |
c5905afb | 1812 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1813 | __net_timestamp(skb); |
1da177e4 LT |
1814 | } |
1815 | ||
588f0330 | 1816 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1817 | if (static_key_false(&netstamp_needed)) { \ |
2456e855 | 1818 | if ((COND) && !(SKB)->tstamp) \ |
588f0330 ED |
1819 | __net_timestamp(SKB); \ |
1820 | } \ | |
3b098e2d | 1821 | |
f4b05d27 | 1822 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 DL |
1823 | { |
1824 | unsigned int len; | |
1825 | ||
1826 | if (!(dev->flags & IFF_UP)) | |
1827 | return false; | |
1828 | ||
1829 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1830 | if (skb->len <= len) | |
1831 | return true; | |
1832 | ||
1833 | /* if TSO is enabled, we don't care about the length as the packet | |
1834 | * could be forwarded without being segmented before | |
1835 | */ | |
1836 | if (skb_is_gso(skb)) | |
1837 | return true; | |
1838 | ||
1839 | return false; | |
1840 | } | |
1ee481fb | 1841 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1842 | |
a0265d28 HX |
1843 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1844 | { | |
4e3264d2 | 1845 | int ret = ____dev_forward_skb(dev, skb); |
a0265d28 | 1846 | |
4e3264d2 MKL |
1847 | if (likely(!ret)) { |
1848 | skb->protocol = eth_type_trans(skb, dev); | |
1849 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
1850 | } | |
a0265d28 | 1851 | |
4e3264d2 | 1852 | return ret; |
a0265d28 HX |
1853 | } |
1854 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1855 | ||
44540960 AB |
1856 | /** |
1857 | * dev_forward_skb - loopback an skb to another netif | |
1858 | * | |
1859 | * @dev: destination network device | |
1860 | * @skb: buffer to forward | |
1861 | * | |
1862 | * return values: | |
1863 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1864 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1865 | * |
1866 | * dev_forward_skb can be used for injecting an skb from the | |
1867 | * start_xmit function of one device into the receive queue | |
1868 | * of another device. | |
1869 | * | |
1870 | * The receiving device may be in another namespace, so | |
1871 | * we have to clear all information in the skb that could | |
1872 | * impact namespace isolation. | |
1873 | */ | |
1874 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1875 | { | |
a0265d28 | 1876 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1877 | } |
1878 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1879 | ||
71d9dec2 CG |
1880 | static inline int deliver_skb(struct sk_buff *skb, |
1881 | struct packet_type *pt_prev, | |
1882 | struct net_device *orig_dev) | |
1883 | { | |
1f8b977a | 1884 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
1080e512 | 1885 | return -ENOMEM; |
63354797 | 1886 | refcount_inc(&skb->users); |
71d9dec2 CG |
1887 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1888 | } | |
1889 | ||
7866a621 SN |
1890 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
1891 | struct packet_type **pt, | |
fbcb2170 JP |
1892 | struct net_device *orig_dev, |
1893 | __be16 type, | |
7866a621 SN |
1894 | struct list_head *ptype_list) |
1895 | { | |
1896 | struct packet_type *ptype, *pt_prev = *pt; | |
1897 | ||
1898 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1899 | if (ptype->type != type) | |
1900 | continue; | |
1901 | if (pt_prev) | |
fbcb2170 | 1902 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
1903 | pt_prev = ptype; |
1904 | } | |
1905 | *pt = pt_prev; | |
1906 | } | |
1907 | ||
c0de08d0 EL |
1908 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1909 | { | |
a3d744e9 | 1910 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1911 | return false; |
1912 | ||
1913 | if (ptype->id_match) | |
1914 | return ptype->id_match(ptype, skb->sk); | |
1915 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1916 | return true; | |
1917 | ||
1918 | return false; | |
1919 | } | |
1920 | ||
1da177e4 LT |
1921 | /* |
1922 | * Support routine. Sends outgoing frames to any network | |
1923 | * taps currently in use. | |
1924 | */ | |
1925 | ||
74b20582 | 1926 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1927 | { |
1928 | struct packet_type *ptype; | |
71d9dec2 CG |
1929 | struct sk_buff *skb2 = NULL; |
1930 | struct packet_type *pt_prev = NULL; | |
7866a621 | 1931 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 1932 | |
1da177e4 | 1933 | rcu_read_lock(); |
7866a621 SN |
1934 | again: |
1935 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1da177e4 LT |
1936 | /* Never send packets back to the socket |
1937 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1938 | */ | |
7866a621 SN |
1939 | if (skb_loop_sk(ptype, skb)) |
1940 | continue; | |
71d9dec2 | 1941 | |
7866a621 SN |
1942 | if (pt_prev) { |
1943 | deliver_skb(skb2, pt_prev, skb->dev); | |
1944 | pt_prev = ptype; | |
1945 | continue; | |
1946 | } | |
1da177e4 | 1947 | |
7866a621 SN |
1948 | /* need to clone skb, done only once */ |
1949 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1950 | if (!skb2) | |
1951 | goto out_unlock; | |
70978182 | 1952 | |
7866a621 | 1953 | net_timestamp_set(skb2); |
1da177e4 | 1954 | |
7866a621 SN |
1955 | /* skb->nh should be correctly |
1956 | * set by sender, so that the second statement is | |
1957 | * just protection against buggy protocols. | |
1958 | */ | |
1959 | skb_reset_mac_header(skb2); | |
1960 | ||
1961 | if (skb_network_header(skb2) < skb2->data || | |
1962 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
1963 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
1964 | ntohs(skb2->protocol), | |
1965 | dev->name); | |
1966 | skb_reset_network_header(skb2); | |
1da177e4 | 1967 | } |
7866a621 SN |
1968 | |
1969 | skb2->transport_header = skb2->network_header; | |
1970 | skb2->pkt_type = PACKET_OUTGOING; | |
1971 | pt_prev = ptype; | |
1972 | } | |
1973 | ||
1974 | if (ptype_list == &ptype_all) { | |
1975 | ptype_list = &dev->ptype_all; | |
1976 | goto again; | |
1da177e4 | 1977 | } |
7866a621 | 1978 | out_unlock: |
581fe0ea WB |
1979 | if (pt_prev) { |
1980 | if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) | |
1981 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1982 | else | |
1983 | kfree_skb(skb2); | |
1984 | } | |
1da177e4 LT |
1985 | rcu_read_unlock(); |
1986 | } | |
74b20582 | 1987 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 1988 | |
2c53040f BH |
1989 | /** |
1990 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1991 | * @dev: Network device |
1992 | * @txq: number of queues available | |
1993 | * | |
1994 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1995 | * valid. To resolve this verify the tc mapping remains valid and if | |
1996 | * not NULL the mapping. With no priorities mapping to this | |
1997 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1998 | * is invalid nothing can be done so disable priority mappings. If is | |
1999 | * expected that drivers will fix this mapping if they can before | |
2000 | * calling netif_set_real_num_tx_queues. | |
2001 | */ | |
bb134d22 | 2002 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
2003 | { |
2004 | int i; | |
2005 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2006 | ||
2007 | /* If TC0 is invalidated disable TC mapping */ | |
2008 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 2009 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
2010 | dev->num_tc = 0; |
2011 | return; | |
2012 | } | |
2013 | ||
2014 | /* Invalidated prio to tc mappings set to TC0 */ | |
2015 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
2016 | int q = netdev_get_prio_tc_map(dev, i); | |
2017 | ||
2018 | tc = &dev->tc_to_txq[q]; | |
2019 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
2020 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
2021 | i, q); | |
4f57c087 JF |
2022 | netdev_set_prio_tc_map(dev, i, 0); |
2023 | } | |
2024 | } | |
2025 | } | |
2026 | ||
8d059b0f AD |
2027 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
2028 | { | |
2029 | if (dev->num_tc) { | |
2030 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2031 | int i; | |
2032 | ||
2033 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { | |
2034 | if ((txq - tc->offset) < tc->count) | |
2035 | return i; | |
2036 | } | |
2037 | ||
2038 | return -1; | |
2039 | } | |
2040 | ||
2041 | return 0; | |
2042 | } | |
8a5f2166 | 2043 | EXPORT_SYMBOL(netdev_txq_to_tc); |
8d059b0f | 2044 | |
537c00de AD |
2045 | #ifdef CONFIG_XPS |
2046 | static DEFINE_MUTEX(xps_map_mutex); | |
2047 | #define xmap_dereference(P) \ | |
2048 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
2049 | ||
6234f874 AD |
2050 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
2051 | int tci, u16 index) | |
537c00de | 2052 | { |
10cdc3f3 AD |
2053 | struct xps_map *map = NULL; |
2054 | int pos; | |
537c00de | 2055 | |
10cdc3f3 | 2056 | if (dev_maps) |
6234f874 AD |
2057 | map = xmap_dereference(dev_maps->cpu_map[tci]); |
2058 | if (!map) | |
2059 | return false; | |
537c00de | 2060 | |
6234f874 AD |
2061 | for (pos = map->len; pos--;) { |
2062 | if (map->queues[pos] != index) | |
2063 | continue; | |
2064 | ||
2065 | if (map->len > 1) { | |
2066 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2067 | break; |
537c00de | 2068 | } |
6234f874 AD |
2069 | |
2070 | RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL); | |
2071 | kfree_rcu(map, rcu); | |
2072 | return false; | |
537c00de AD |
2073 | } |
2074 | ||
6234f874 | 2075 | return true; |
10cdc3f3 AD |
2076 | } |
2077 | ||
6234f874 AD |
2078 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2079 | struct xps_dev_maps *dev_maps, | |
2080 | int cpu, u16 offset, u16 count) | |
2081 | { | |
184c449f AD |
2082 | int num_tc = dev->num_tc ? : 1; |
2083 | bool active = false; | |
2084 | int tci; | |
6234f874 | 2085 | |
184c449f AD |
2086 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2087 | int i, j; | |
2088 | ||
2089 | for (i = count, j = offset; i--; j++) { | |
2090 | if (!remove_xps_queue(dev_maps, cpu, j)) | |
2091 | break; | |
2092 | } | |
2093 | ||
2094 | active |= i < 0; | |
6234f874 AD |
2095 | } |
2096 | ||
184c449f | 2097 | return active; |
6234f874 AD |
2098 | } |
2099 | ||
2100 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, | |
2101 | u16 count) | |
10cdc3f3 AD |
2102 | { |
2103 | struct xps_dev_maps *dev_maps; | |
024e9679 | 2104 | int cpu, i; |
10cdc3f3 AD |
2105 | bool active = false; |
2106 | ||
2107 | mutex_lock(&xps_map_mutex); | |
2108 | dev_maps = xmap_dereference(dev->xps_maps); | |
2109 | ||
2110 | if (!dev_maps) | |
2111 | goto out_no_maps; | |
2112 | ||
6234f874 AD |
2113 | for_each_possible_cpu(cpu) |
2114 | active |= remove_xps_queue_cpu(dev, dev_maps, cpu, | |
2115 | offset, count); | |
10cdc3f3 AD |
2116 | |
2117 | if (!active) { | |
537c00de AD |
2118 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
2119 | kfree_rcu(dev_maps, rcu); | |
2120 | } | |
2121 | ||
6234f874 | 2122 | for (i = offset + (count - 1); count--; i--) |
024e9679 AD |
2123 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), |
2124 | NUMA_NO_NODE); | |
2125 | ||
537c00de AD |
2126 | out_no_maps: |
2127 | mutex_unlock(&xps_map_mutex); | |
2128 | } | |
2129 | ||
6234f874 AD |
2130 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2131 | { | |
2132 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2133 | } | |
2134 | ||
01c5f864 AD |
2135 | static struct xps_map *expand_xps_map(struct xps_map *map, |
2136 | int cpu, u16 index) | |
2137 | { | |
2138 | struct xps_map *new_map; | |
2139 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2140 | int i, pos; | |
2141 | ||
2142 | for (pos = 0; map && pos < map->len; pos++) { | |
2143 | if (map->queues[pos] != index) | |
2144 | continue; | |
2145 | return map; | |
2146 | } | |
2147 | ||
2148 | /* Need to add queue to this CPU's existing map */ | |
2149 | if (map) { | |
2150 | if (pos < map->alloc_len) | |
2151 | return map; | |
2152 | ||
2153 | alloc_len = map->alloc_len * 2; | |
2154 | } | |
2155 | ||
2156 | /* Need to allocate new map to store queue on this CPU's map */ | |
2157 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2158 | cpu_to_node(cpu)); | |
2159 | if (!new_map) | |
2160 | return NULL; | |
2161 | ||
2162 | for (i = 0; i < pos; i++) | |
2163 | new_map->queues[i] = map->queues[i]; | |
2164 | new_map->alloc_len = alloc_len; | |
2165 | new_map->len = pos; | |
2166 | ||
2167 | return new_map; | |
2168 | } | |
2169 | ||
3573540c MT |
2170 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
2171 | u16 index) | |
537c00de | 2172 | { |
01c5f864 | 2173 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
184c449f AD |
2174 | int i, cpu, tci, numa_node_id = -2; |
2175 | int maps_sz, num_tc = 1, tc = 0; | |
537c00de | 2176 | struct xps_map *map, *new_map; |
01c5f864 | 2177 | bool active = false; |
537c00de | 2178 | |
184c449f AD |
2179 | if (dev->num_tc) { |
2180 | num_tc = dev->num_tc; | |
2181 | tc = netdev_txq_to_tc(dev, index); | |
2182 | if (tc < 0) | |
2183 | return -EINVAL; | |
2184 | } | |
2185 | ||
2186 | maps_sz = XPS_DEV_MAPS_SIZE(num_tc); | |
2187 | if (maps_sz < L1_CACHE_BYTES) | |
2188 | maps_sz = L1_CACHE_BYTES; | |
2189 | ||
537c00de AD |
2190 | mutex_lock(&xps_map_mutex); |
2191 | ||
2192 | dev_maps = xmap_dereference(dev->xps_maps); | |
2193 | ||
01c5f864 | 2194 | /* allocate memory for queue storage */ |
184c449f | 2195 | for_each_cpu_and(cpu, cpu_online_mask, mask) { |
01c5f864 AD |
2196 | if (!new_dev_maps) |
2197 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2198 | if (!new_dev_maps) { |
2199 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2200 | return -ENOMEM; |
2bb60cb9 | 2201 | } |
01c5f864 | 2202 | |
184c449f AD |
2203 | tci = cpu * num_tc + tc; |
2204 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) : | |
01c5f864 AD |
2205 | NULL; |
2206 | ||
2207 | map = expand_xps_map(map, cpu, index); | |
2208 | if (!map) | |
2209 | goto error; | |
2210 | ||
184c449f | 2211 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); |
01c5f864 AD |
2212 | } |
2213 | ||
2214 | if (!new_dev_maps) | |
2215 | goto out_no_new_maps; | |
2216 | ||
537c00de | 2217 | for_each_possible_cpu(cpu) { |
184c449f AD |
2218 | /* copy maps belonging to foreign traffic classes */ |
2219 | for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) { | |
2220 | /* fill in the new device map from the old device map */ | |
2221 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
2222 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
2223 | } | |
2224 | ||
2225 | /* We need to explicitly update tci as prevous loop | |
2226 | * could break out early if dev_maps is NULL. | |
2227 | */ | |
2228 | tci = cpu * num_tc + tc; | |
2229 | ||
01c5f864 AD |
2230 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
2231 | /* add queue to CPU maps */ | |
2232 | int pos = 0; | |
2233 | ||
184c449f | 2234 | map = xmap_dereference(new_dev_maps->cpu_map[tci]); |
01c5f864 AD |
2235 | while ((pos < map->len) && (map->queues[pos] != index)) |
2236 | pos++; | |
2237 | ||
2238 | if (pos == map->len) | |
2239 | map->queues[map->len++] = index; | |
537c00de | 2240 | #ifdef CONFIG_NUMA |
537c00de AD |
2241 | if (numa_node_id == -2) |
2242 | numa_node_id = cpu_to_node(cpu); | |
2243 | else if (numa_node_id != cpu_to_node(cpu)) | |
2244 | numa_node_id = -1; | |
537c00de | 2245 | #endif |
01c5f864 AD |
2246 | } else if (dev_maps) { |
2247 | /* fill in the new device map from the old device map */ | |
184c449f AD |
2248 | map = xmap_dereference(dev_maps->cpu_map[tci]); |
2249 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
537c00de | 2250 | } |
01c5f864 | 2251 | |
184c449f AD |
2252 | /* copy maps belonging to foreign traffic classes */ |
2253 | for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { | |
2254 | /* fill in the new device map from the old device map */ | |
2255 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
2256 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
2257 | } | |
537c00de AD |
2258 | } |
2259 | ||
01c5f864 AD |
2260 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2261 | ||
537c00de | 2262 | /* Cleanup old maps */ |
184c449f AD |
2263 | if (!dev_maps) |
2264 | goto out_no_old_maps; | |
2265 | ||
2266 | for_each_possible_cpu(cpu) { | |
2267 | for (i = num_tc, tci = cpu * num_tc; i--; tci++) { | |
2268 | new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); | |
2269 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
01c5f864 AD |
2270 | if (map && map != new_map) |
2271 | kfree_rcu(map, rcu); | |
2272 | } | |
537c00de AD |
2273 | } |
2274 | ||
184c449f AD |
2275 | kfree_rcu(dev_maps, rcu); |
2276 | ||
2277 | out_no_old_maps: | |
01c5f864 AD |
2278 | dev_maps = new_dev_maps; |
2279 | active = true; | |
537c00de | 2280 | |
01c5f864 AD |
2281 | out_no_new_maps: |
2282 | /* update Tx queue numa node */ | |
537c00de AD |
2283 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2284 | (numa_node_id >= 0) ? numa_node_id : | |
2285 | NUMA_NO_NODE); | |
2286 | ||
01c5f864 AD |
2287 | if (!dev_maps) |
2288 | goto out_no_maps; | |
2289 | ||
2290 | /* removes queue from unused CPUs */ | |
2291 | for_each_possible_cpu(cpu) { | |
184c449f AD |
2292 | for (i = tc, tci = cpu * num_tc; i--; tci++) |
2293 | active |= remove_xps_queue(dev_maps, tci, index); | |
2294 | if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu)) | |
2295 | active |= remove_xps_queue(dev_maps, tci, index); | |
2296 | for (i = num_tc - tc, tci++; --i; tci++) | |
2297 | active |= remove_xps_queue(dev_maps, tci, index); | |
01c5f864 AD |
2298 | } |
2299 | ||
2300 | /* free map if not active */ | |
2301 | if (!active) { | |
2302 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2303 | kfree_rcu(dev_maps, rcu); | |
2304 | } | |
2305 | ||
2306 | out_no_maps: | |
537c00de AD |
2307 | mutex_unlock(&xps_map_mutex); |
2308 | ||
2309 | return 0; | |
2310 | error: | |
01c5f864 AD |
2311 | /* remove any maps that we added */ |
2312 | for_each_possible_cpu(cpu) { | |
184c449f AD |
2313 | for (i = num_tc, tci = cpu * num_tc; i--; tci++) { |
2314 | new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); | |
2315 | map = dev_maps ? | |
2316 | xmap_dereference(dev_maps->cpu_map[tci]) : | |
2317 | NULL; | |
2318 | if (new_map && new_map != map) | |
2319 | kfree(new_map); | |
2320 | } | |
01c5f864 AD |
2321 | } |
2322 | ||
537c00de AD |
2323 | mutex_unlock(&xps_map_mutex); |
2324 | ||
537c00de AD |
2325 | kfree(new_dev_maps); |
2326 | return -ENOMEM; | |
2327 | } | |
2328 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2329 | ||
2330 | #endif | |
9cf1f6a8 AD |
2331 | void netdev_reset_tc(struct net_device *dev) |
2332 | { | |
6234f874 AD |
2333 | #ifdef CONFIG_XPS |
2334 | netif_reset_xps_queues_gt(dev, 0); | |
2335 | #endif | |
9cf1f6a8 AD |
2336 | dev->num_tc = 0; |
2337 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2338 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2339 | } | |
2340 | EXPORT_SYMBOL(netdev_reset_tc); | |
2341 | ||
2342 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2343 | { | |
2344 | if (tc >= dev->num_tc) | |
2345 | return -EINVAL; | |
2346 | ||
6234f874 AD |
2347 | #ifdef CONFIG_XPS |
2348 | netif_reset_xps_queues(dev, offset, count); | |
2349 | #endif | |
9cf1f6a8 AD |
2350 | dev->tc_to_txq[tc].count = count; |
2351 | dev->tc_to_txq[tc].offset = offset; | |
2352 | return 0; | |
2353 | } | |
2354 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2355 | ||
2356 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2357 | { | |
2358 | if (num_tc > TC_MAX_QUEUE) | |
2359 | return -EINVAL; | |
2360 | ||
6234f874 AD |
2361 | #ifdef CONFIG_XPS |
2362 | netif_reset_xps_queues_gt(dev, 0); | |
2363 | #endif | |
9cf1f6a8 AD |
2364 | dev->num_tc = num_tc; |
2365 | return 0; | |
2366 | } | |
2367 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2368 | ||
f0796d5c JF |
2369 | /* |
2370 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2371 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2372 | */ | |
e6484930 | 2373 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2374 | { |
1d24eb48 TH |
2375 | int rc; |
2376 | ||
e6484930 TH |
2377 | if (txq < 1 || txq > dev->num_tx_queues) |
2378 | return -EINVAL; | |
f0796d5c | 2379 | |
5c56580b BH |
2380 | if (dev->reg_state == NETREG_REGISTERED || |
2381 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2382 | ASSERT_RTNL(); |
2383 | ||
1d24eb48 TH |
2384 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2385 | txq); | |
bf264145 TH |
2386 | if (rc) |
2387 | return rc; | |
2388 | ||
4f57c087 JF |
2389 | if (dev->num_tc) |
2390 | netif_setup_tc(dev, txq); | |
2391 | ||
024e9679 | 2392 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2393 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2394 | #ifdef CONFIG_XPS |
2395 | netif_reset_xps_queues_gt(dev, txq); | |
2396 | #endif | |
2397 | } | |
f0796d5c | 2398 | } |
e6484930 TH |
2399 | |
2400 | dev->real_num_tx_queues = txq; | |
2401 | return 0; | |
f0796d5c JF |
2402 | } |
2403 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2404 | |
a953be53 | 2405 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2406 | /** |
2407 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2408 | * @dev: Network device | |
2409 | * @rxq: Actual number of RX queues | |
2410 | * | |
2411 | * This must be called either with the rtnl_lock held or before | |
2412 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2413 | * negative error code. If called before registration, it always |
2414 | * succeeds. | |
62fe0b40 BH |
2415 | */ |
2416 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2417 | { | |
2418 | int rc; | |
2419 | ||
bd25fa7b TH |
2420 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2421 | return -EINVAL; | |
2422 | ||
62fe0b40 BH |
2423 | if (dev->reg_state == NETREG_REGISTERED) { |
2424 | ASSERT_RTNL(); | |
2425 | ||
62fe0b40 BH |
2426 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2427 | rxq); | |
2428 | if (rc) | |
2429 | return rc; | |
62fe0b40 BH |
2430 | } |
2431 | ||
2432 | dev->real_num_rx_queues = rxq; | |
2433 | return 0; | |
2434 | } | |
2435 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2436 | #endif | |
2437 | ||
2c53040f BH |
2438 | /** |
2439 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2440 | * |
2441 | * This routine should set an upper limit on the number of RSS queues | |
2442 | * used by default by multiqueue devices. | |
2443 | */ | |
a55b138b | 2444 | int netif_get_num_default_rss_queues(void) |
16917b87 | 2445 | { |
40e4e713 HS |
2446 | return is_kdump_kernel() ? |
2447 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
2448 | } |
2449 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2450 | ||
3bcb846c | 2451 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 2452 | { |
def82a1d JP |
2453 | struct softnet_data *sd; |
2454 | unsigned long flags; | |
56079431 | 2455 | |
def82a1d | 2456 | local_irq_save(flags); |
903ceff7 | 2457 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2458 | q->next_sched = NULL; |
2459 | *sd->output_queue_tailp = q; | |
2460 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2461 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2462 | local_irq_restore(flags); | |
2463 | } | |
2464 | ||
2465 | void __netif_schedule(struct Qdisc *q) | |
2466 | { | |
2467 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2468 | __netif_reschedule(q); | |
56079431 DV |
2469 | } |
2470 | EXPORT_SYMBOL(__netif_schedule); | |
2471 | ||
e6247027 ED |
2472 | struct dev_kfree_skb_cb { |
2473 | enum skb_free_reason reason; | |
2474 | }; | |
2475 | ||
2476 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2477 | { |
e6247027 ED |
2478 | return (struct dev_kfree_skb_cb *)skb->cb; |
2479 | } | |
2480 | ||
46e5da40 JF |
2481 | void netif_schedule_queue(struct netdev_queue *txq) |
2482 | { | |
2483 | rcu_read_lock(); | |
2484 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2485 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2486 | ||
2487 | __netif_schedule(q); | |
2488 | } | |
2489 | rcu_read_unlock(); | |
2490 | } | |
2491 | EXPORT_SYMBOL(netif_schedule_queue); | |
2492 | ||
46e5da40 JF |
2493 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
2494 | { | |
2495 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2496 | struct Qdisc *q; | |
2497 | ||
2498 | rcu_read_lock(); | |
2499 | q = rcu_dereference(dev_queue->qdisc); | |
2500 | __netif_schedule(q); | |
2501 | rcu_read_unlock(); | |
2502 | } | |
2503 | } | |
2504 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2505 | ||
e6247027 | 2506 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2507 | { |
e6247027 | 2508 | unsigned long flags; |
56079431 | 2509 | |
9899886d MJ |
2510 | if (unlikely(!skb)) |
2511 | return; | |
2512 | ||
63354797 | 2513 | if (likely(refcount_read(&skb->users) == 1)) { |
e6247027 | 2514 | smp_rmb(); |
63354797 RE |
2515 | refcount_set(&skb->users, 0); |
2516 | } else if (likely(!refcount_dec_and_test(&skb->users))) { | |
e6247027 | 2517 | return; |
bea3348e | 2518 | } |
e6247027 ED |
2519 | get_kfree_skb_cb(skb)->reason = reason; |
2520 | local_irq_save(flags); | |
2521 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2522 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2523 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2524 | local_irq_restore(flags); | |
56079431 | 2525 | } |
e6247027 | 2526 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2527 | |
e6247027 | 2528 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2529 | { |
2530 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2531 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2532 | else |
2533 | dev_kfree_skb(skb); | |
2534 | } | |
e6247027 | 2535 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2536 | |
2537 | ||
bea3348e SH |
2538 | /** |
2539 | * netif_device_detach - mark device as removed | |
2540 | * @dev: network device | |
2541 | * | |
2542 | * Mark device as removed from system and therefore no longer available. | |
2543 | */ | |
56079431 DV |
2544 | void netif_device_detach(struct net_device *dev) |
2545 | { | |
2546 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2547 | netif_running(dev)) { | |
d543103a | 2548 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2549 | } |
2550 | } | |
2551 | EXPORT_SYMBOL(netif_device_detach); | |
2552 | ||
bea3348e SH |
2553 | /** |
2554 | * netif_device_attach - mark device as attached | |
2555 | * @dev: network device | |
2556 | * | |
2557 | * Mark device as attached from system and restart if needed. | |
2558 | */ | |
56079431 DV |
2559 | void netif_device_attach(struct net_device *dev) |
2560 | { | |
2561 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2562 | netif_running(dev)) { | |
d543103a | 2563 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2564 | __netdev_watchdog_up(dev); |
56079431 DV |
2565 | } |
2566 | } | |
2567 | EXPORT_SYMBOL(netif_device_attach); | |
2568 | ||
5605c762 JP |
2569 | /* |
2570 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2571 | * to be used as a distribution range. | |
2572 | */ | |
2573 | u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, | |
2574 | unsigned int num_tx_queues) | |
2575 | { | |
2576 | u32 hash; | |
2577 | u16 qoffset = 0; | |
2578 | u16 qcount = num_tx_queues; | |
2579 | ||
2580 | if (skb_rx_queue_recorded(skb)) { | |
2581 | hash = skb_get_rx_queue(skb); | |
2582 | while (unlikely(hash >= num_tx_queues)) | |
2583 | hash -= num_tx_queues; | |
2584 | return hash; | |
2585 | } | |
2586 | ||
2587 | if (dev->num_tc) { | |
2588 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
f4563a75 | 2589 | |
5605c762 JP |
2590 | qoffset = dev->tc_to_txq[tc].offset; |
2591 | qcount = dev->tc_to_txq[tc].count; | |
2592 | } | |
2593 | ||
2594 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
2595 | } | |
2596 | EXPORT_SYMBOL(__skb_tx_hash); | |
2597 | ||
36c92474 BH |
2598 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2599 | { | |
84d15ae5 | 2600 | static const netdev_features_t null_features; |
36c92474 | 2601 | struct net_device *dev = skb->dev; |
88ad4175 | 2602 | const char *name = ""; |
36c92474 | 2603 | |
c846ad9b BG |
2604 | if (!net_ratelimit()) |
2605 | return; | |
2606 | ||
88ad4175 BM |
2607 | if (dev) { |
2608 | if (dev->dev.parent) | |
2609 | name = dev_driver_string(dev->dev.parent); | |
2610 | else | |
2611 | name = netdev_name(dev); | |
2612 | } | |
36c92474 BH |
2613 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " |
2614 | "gso_type=%d ip_summed=%d\n", | |
88ad4175 | 2615 | name, dev ? &dev->features : &null_features, |
65e9d2fa | 2616 | skb->sk ? &skb->sk->sk_route_caps : &null_features, |
36c92474 BH |
2617 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2618 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2619 | } | |
2620 | ||
1da177e4 LT |
2621 | /* |
2622 | * Invalidate hardware checksum when packet is to be mangled, and | |
2623 | * complete checksum manually on outgoing path. | |
2624 | */ | |
84fa7933 | 2625 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2626 | { |
d3bc23e7 | 2627 | __wsum csum; |
663ead3b | 2628 | int ret = 0, offset; |
1da177e4 | 2629 | |
84fa7933 | 2630 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2631 | goto out_set_summed; |
2632 | ||
2633 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2634 | skb_warn_bad_offload(skb); |
2635 | return -EINVAL; | |
1da177e4 LT |
2636 | } |
2637 | ||
cef401de ED |
2638 | /* Before computing a checksum, we should make sure no frag could |
2639 | * be modified by an external entity : checksum could be wrong. | |
2640 | */ | |
2641 | if (skb_has_shared_frag(skb)) { | |
2642 | ret = __skb_linearize(skb); | |
2643 | if (ret) | |
2644 | goto out; | |
2645 | } | |
2646 | ||
55508d60 | 2647 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2648 | BUG_ON(offset >= skb_headlen(skb)); |
2649 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2650 | ||
2651 | offset += skb->csum_offset; | |
2652 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2653 | ||
2654 | if (skb_cloned(skb) && | |
2655 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2656 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2657 | if (ret) | |
2658 | goto out; | |
2659 | } | |
2660 | ||
4f2e4ad5 | 2661 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 2662 | out_set_summed: |
1da177e4 | 2663 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2664 | out: |
1da177e4 LT |
2665 | return ret; |
2666 | } | |
d1b19dff | 2667 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2668 | |
b72b5bf6 DC |
2669 | int skb_crc32c_csum_help(struct sk_buff *skb) |
2670 | { | |
2671 | __le32 crc32c_csum; | |
2672 | int ret = 0, offset, start; | |
2673 | ||
2674 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
2675 | goto out; | |
2676 | ||
2677 | if (unlikely(skb_is_gso(skb))) | |
2678 | goto out; | |
2679 | ||
2680 | /* Before computing a checksum, we should make sure no frag could | |
2681 | * be modified by an external entity : checksum could be wrong. | |
2682 | */ | |
2683 | if (unlikely(skb_has_shared_frag(skb))) { | |
2684 | ret = __skb_linearize(skb); | |
2685 | if (ret) | |
2686 | goto out; | |
2687 | } | |
2688 | start = skb_checksum_start_offset(skb); | |
2689 | offset = start + offsetof(struct sctphdr, checksum); | |
2690 | if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { | |
2691 | ret = -EINVAL; | |
2692 | goto out; | |
2693 | } | |
2694 | if (skb_cloned(skb) && | |
2695 | !skb_clone_writable(skb, offset + sizeof(__le32))) { | |
2696 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); | |
2697 | if (ret) | |
2698 | goto out; | |
2699 | } | |
2700 | crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, | |
2701 | skb->len - start, ~(__u32)0, | |
2702 | crc32c_csum_stub)); | |
2703 | *(__le32 *)(skb->data + offset) = crc32c_csum; | |
2704 | skb->ip_summed = CHECKSUM_NONE; | |
dba00306 | 2705 | skb->csum_not_inet = 0; |
b72b5bf6 DC |
2706 | out: |
2707 | return ret; | |
2708 | } | |
2709 | ||
53d6471c | 2710 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2711 | { |
252e3346 | 2712 | __be16 type = skb->protocol; |
f6a78bfc | 2713 | |
19acc327 PS |
2714 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2715 | if (type == htons(ETH_P_TEB)) { | |
2716 | struct ethhdr *eth; | |
2717 | ||
2718 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2719 | return 0; | |
2720 | ||
2721 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2722 | type = eth->h_proto; | |
2723 | } | |
2724 | ||
d4bcef3f | 2725 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
2726 | } |
2727 | ||
2728 | /** | |
2729 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2730 | * @skb: buffer to segment | |
2731 | * @features: features for the output path (see dev->features) | |
2732 | */ | |
2733 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2734 | netdev_features_t features) | |
2735 | { | |
2736 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2737 | struct packet_offload *ptype; | |
53d6471c VY |
2738 | int vlan_depth = skb->mac_len; |
2739 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2740 | |
2741 | if (unlikely(!type)) | |
2742 | return ERR_PTR(-EINVAL); | |
2743 | ||
53d6471c | 2744 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2745 | |
2746 | rcu_read_lock(); | |
22061d80 | 2747 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2748 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2749 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2750 | break; |
2751 | } | |
2752 | } | |
2753 | rcu_read_unlock(); | |
2754 | ||
98e399f8 | 2755 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2756 | |
f6a78bfc HX |
2757 | return segs; |
2758 | } | |
05e8ef4a PS |
2759 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2760 | ||
2761 | ||
2762 | /* openvswitch calls this on rx path, so we need a different check. | |
2763 | */ | |
2764 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2765 | { | |
2766 | if (tx_path) | |
93991221 | 2767 | return skb->ip_summed != CHECKSUM_PARTIAL; |
6e7bc478 ED |
2768 | |
2769 | return skb->ip_summed == CHECKSUM_NONE; | |
05e8ef4a PS |
2770 | } |
2771 | ||
2772 | /** | |
2773 | * __skb_gso_segment - Perform segmentation on skb. | |
2774 | * @skb: buffer to segment | |
2775 | * @features: features for the output path (see dev->features) | |
2776 | * @tx_path: whether it is called in TX path | |
2777 | * | |
2778 | * This function segments the given skb and returns a list of segments. | |
2779 | * | |
2780 | * It may return NULL if the skb requires no segmentation. This is | |
2781 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 KK |
2782 | * |
2783 | * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. | |
05e8ef4a PS |
2784 | */ |
2785 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2786 | netdev_features_t features, bool tx_path) | |
2787 | { | |
b2504a5d ED |
2788 | struct sk_buff *segs; |
2789 | ||
05e8ef4a PS |
2790 | if (unlikely(skb_needs_check(skb, tx_path))) { |
2791 | int err; | |
2792 | ||
b2504a5d | 2793 | /* We're going to init ->check field in TCP or UDP header */ |
a40e0a66 | 2794 | err = skb_cow_head(skb, 0); |
2795 | if (err < 0) | |
05e8ef4a PS |
2796 | return ERR_PTR(err); |
2797 | } | |
2798 | ||
802ab55a AD |
2799 | /* Only report GSO partial support if it will enable us to |
2800 | * support segmentation on this frame without needing additional | |
2801 | * work. | |
2802 | */ | |
2803 | if (features & NETIF_F_GSO_PARTIAL) { | |
2804 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
2805 | struct net_device *dev = skb->dev; | |
2806 | ||
2807 | partial_features |= dev->features & dev->gso_partial_features; | |
2808 | if (!skb_gso_ok(skb, features | partial_features)) | |
2809 | features &= ~NETIF_F_GSO_PARTIAL; | |
2810 | } | |
2811 | ||
9207f9d4 KK |
2812 | BUILD_BUG_ON(SKB_SGO_CB_OFFSET + |
2813 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); | |
2814 | ||
68c33163 | 2815 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2816 | SKB_GSO_CB(skb)->encap_level = 0; |
2817 | ||
05e8ef4a PS |
2818 | skb_reset_mac_header(skb); |
2819 | skb_reset_mac_len(skb); | |
2820 | ||
b2504a5d ED |
2821 | segs = skb_mac_gso_segment(skb, features); |
2822 | ||
2823 | if (unlikely(skb_needs_check(skb, tx_path))) | |
2824 | skb_warn_bad_offload(skb); | |
2825 | ||
2826 | return segs; | |
05e8ef4a | 2827 | } |
12b0004d | 2828 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2829 | |
fb286bb2 HX |
2830 | /* Take action when hardware reception checksum errors are detected. */ |
2831 | #ifdef CONFIG_BUG | |
2832 | void netdev_rx_csum_fault(struct net_device *dev) | |
2833 | { | |
2834 | if (net_ratelimit()) { | |
7b6cd1ce | 2835 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2836 | dump_stack(); |
2837 | } | |
2838 | } | |
2839 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2840 | #endif | |
2841 | ||
1da177e4 LT |
2842 | /* Actually, we should eliminate this check as soon as we know, that: |
2843 | * 1. IOMMU is present and allows to map all the memory. | |
2844 | * 2. No high memory really exists on this machine. | |
2845 | */ | |
2846 | ||
c1e756bf | 2847 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2848 | { |
3d3a8533 | 2849 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2850 | int i; |
f4563a75 | 2851 | |
5acbbd42 | 2852 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2853 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2854 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 2855 | |
ea2ab693 | 2856 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 2857 | return 1; |
ea2ab693 | 2858 | } |
5acbbd42 | 2859 | } |
1da177e4 | 2860 | |
5acbbd42 FT |
2861 | if (PCI_DMA_BUS_IS_PHYS) { |
2862 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2863 | |
9092c658 ED |
2864 | if (!pdev) |
2865 | return 0; | |
5acbbd42 | 2866 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2867 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2868 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
f4563a75 | 2869 | |
5acbbd42 FT |
2870 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2871 | return 1; | |
2872 | } | |
2873 | } | |
3d3a8533 | 2874 | #endif |
1da177e4 LT |
2875 | return 0; |
2876 | } | |
1da177e4 | 2877 | |
3b392ddb SH |
2878 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2879 | * instead of standard features for the netdev. | |
2880 | */ | |
d0edc7bf | 2881 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
2882 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
2883 | netdev_features_t features, | |
2884 | __be16 type) | |
2885 | { | |
25cd9ba0 | 2886 | if (eth_p_mpls(type)) |
3b392ddb SH |
2887 | features &= skb->dev->mpls_features; |
2888 | ||
2889 | return features; | |
2890 | } | |
2891 | #else | |
2892 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2893 | netdev_features_t features, | |
2894 | __be16 type) | |
2895 | { | |
2896 | return features; | |
2897 | } | |
2898 | #endif | |
2899 | ||
c8f44aff | 2900 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2901 | netdev_features_t features) |
f01a5236 | 2902 | { |
53d6471c | 2903 | int tmp; |
3b392ddb SH |
2904 | __be16 type; |
2905 | ||
2906 | type = skb_network_protocol(skb, &tmp); | |
2907 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2908 | |
c0d680e5 | 2909 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2910 | !can_checksum_protocol(features, type)) { |
996e8021 | 2911 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 2912 | } |
7be2c82c ED |
2913 | if (illegal_highdma(skb->dev, skb)) |
2914 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
2915 | |
2916 | return features; | |
2917 | } | |
2918 | ||
e38f3025 TM |
2919 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
2920 | struct net_device *dev, | |
2921 | netdev_features_t features) | |
2922 | { | |
2923 | return features; | |
2924 | } | |
2925 | EXPORT_SYMBOL(passthru_features_check); | |
2926 | ||
8cb65d00 TM |
2927 | static netdev_features_t dflt_features_check(const struct sk_buff *skb, |
2928 | struct net_device *dev, | |
2929 | netdev_features_t features) | |
2930 | { | |
2931 | return vlan_features_check(skb, features); | |
2932 | } | |
2933 | ||
cbc53e08 AD |
2934 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
2935 | struct net_device *dev, | |
2936 | netdev_features_t features) | |
2937 | { | |
2938 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
2939 | ||
2940 | if (gso_segs > dev->gso_max_segs) | |
2941 | return features & ~NETIF_F_GSO_MASK; | |
2942 | ||
802ab55a AD |
2943 | /* Support for GSO partial features requires software |
2944 | * intervention before we can actually process the packets | |
2945 | * so we need to strip support for any partial features now | |
2946 | * and we can pull them back in after we have partially | |
2947 | * segmented the frame. | |
2948 | */ | |
2949 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
2950 | features &= ~dev->gso_partial_features; | |
2951 | ||
2952 | /* Make sure to clear the IPv4 ID mangling feature if the | |
2953 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
2954 | */ |
2955 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
2956 | struct iphdr *iph = skb->encapsulation ? | |
2957 | inner_ip_hdr(skb) : ip_hdr(skb); | |
2958 | ||
2959 | if (!(iph->frag_off & htons(IP_DF))) | |
2960 | features &= ~NETIF_F_TSO_MANGLEID; | |
2961 | } | |
2962 | ||
2963 | return features; | |
2964 | } | |
2965 | ||
c1e756bf | 2966 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 2967 | { |
5f35227e | 2968 | struct net_device *dev = skb->dev; |
fcbeb976 | 2969 | netdev_features_t features = dev->features; |
58e998c6 | 2970 | |
cbc53e08 AD |
2971 | if (skb_is_gso(skb)) |
2972 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 2973 | |
5f35227e JG |
2974 | /* If encapsulation offload request, verify we are testing |
2975 | * hardware encapsulation features instead of standard | |
2976 | * features for the netdev | |
2977 | */ | |
2978 | if (skb->encapsulation) | |
2979 | features &= dev->hw_enc_features; | |
2980 | ||
f5a7fb88 TM |
2981 | if (skb_vlan_tagged(skb)) |
2982 | features = netdev_intersect_features(features, | |
2983 | dev->vlan_features | | |
2984 | NETIF_F_HW_VLAN_CTAG_TX | | |
2985 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2986 | |
5f35227e JG |
2987 | if (dev->netdev_ops->ndo_features_check) |
2988 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
2989 | features); | |
8cb65d00 TM |
2990 | else |
2991 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 2992 | |
c1e756bf | 2993 | return harmonize_features(skb, features); |
58e998c6 | 2994 | } |
c1e756bf | 2995 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2996 | |
2ea25513 | 2997 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2998 | struct netdev_queue *txq, bool more) |
f6a78bfc | 2999 | { |
2ea25513 DM |
3000 | unsigned int len; |
3001 | int rc; | |
00829823 | 3002 | |
7866a621 | 3003 | if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) |
2ea25513 | 3004 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 3005 | |
2ea25513 DM |
3006 | len = skb->len; |
3007 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 3008 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 3009 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 3010 | |
2ea25513 DM |
3011 | return rc; |
3012 | } | |
7b9c6090 | 3013 | |
8dcda22a DM |
3014 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
3015 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
3016 | { |
3017 | struct sk_buff *skb = first; | |
3018 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 3019 | |
7f2e870f DM |
3020 | while (skb) { |
3021 | struct sk_buff *next = skb->next; | |
fc70fb64 | 3022 | |
7f2e870f | 3023 | skb->next = NULL; |
95f6b3dd | 3024 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
3025 | if (unlikely(!dev_xmit_complete(rc))) { |
3026 | skb->next = next; | |
3027 | goto out; | |
3028 | } | |
6afff0ca | 3029 | |
7f2e870f DM |
3030 | skb = next; |
3031 | if (netif_xmit_stopped(txq) && skb) { | |
3032 | rc = NETDEV_TX_BUSY; | |
3033 | break; | |
9ccb8975 | 3034 | } |
7f2e870f | 3035 | } |
9ccb8975 | 3036 | |
7f2e870f DM |
3037 | out: |
3038 | *ret = rc; | |
3039 | return skb; | |
3040 | } | |
b40863c6 | 3041 | |
1ff0dc94 ED |
3042 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
3043 | netdev_features_t features) | |
f6a78bfc | 3044 | { |
df8a39de | 3045 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
3046 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
3047 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
3048 | return skb; |
3049 | } | |
f6a78bfc | 3050 | |
43c26a1a DC |
3051 | int skb_csum_hwoffload_help(struct sk_buff *skb, |
3052 | const netdev_features_t features) | |
3053 | { | |
3054 | if (unlikely(skb->csum_not_inet)) | |
3055 | return !!(features & NETIF_F_SCTP_CRC) ? 0 : | |
3056 | skb_crc32c_csum_help(skb); | |
3057 | ||
3058 | return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb); | |
3059 | } | |
3060 | EXPORT_SYMBOL(skb_csum_hwoffload_help); | |
3061 | ||
55a93b3e | 3062 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
3063 | { |
3064 | netdev_features_t features; | |
f6a78bfc | 3065 | |
eae3f88e DM |
3066 | features = netif_skb_features(skb); |
3067 | skb = validate_xmit_vlan(skb, features); | |
3068 | if (unlikely(!skb)) | |
3069 | goto out_null; | |
7b9c6090 | 3070 | |
8b86a61d | 3071 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
3072 | struct sk_buff *segs; |
3073 | ||
3074 | segs = skb_gso_segment(skb, features); | |
cecda693 | 3075 | if (IS_ERR(segs)) { |
af6dabc9 | 3076 | goto out_kfree_skb; |
cecda693 JW |
3077 | } else if (segs) { |
3078 | consume_skb(skb); | |
3079 | skb = segs; | |
f6a78bfc | 3080 | } |
eae3f88e DM |
3081 | } else { |
3082 | if (skb_needs_linearize(skb, features) && | |
3083 | __skb_linearize(skb)) | |
3084 | goto out_kfree_skb; | |
4ec93edb | 3085 | |
f6e27114 SK |
3086 | if (validate_xmit_xfrm(skb, features)) |
3087 | goto out_kfree_skb; | |
3088 | ||
eae3f88e DM |
3089 | /* If packet is not checksummed and device does not |
3090 | * support checksumming for this protocol, complete | |
3091 | * checksumming here. | |
3092 | */ | |
3093 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
3094 | if (skb->encapsulation) | |
3095 | skb_set_inner_transport_header(skb, | |
3096 | skb_checksum_start_offset(skb)); | |
3097 | else | |
3098 | skb_set_transport_header(skb, | |
3099 | skb_checksum_start_offset(skb)); | |
43c26a1a | 3100 | if (skb_csum_hwoffload_help(skb, features)) |
eae3f88e | 3101 | goto out_kfree_skb; |
7b9c6090 | 3102 | } |
0c772159 | 3103 | } |
7b9c6090 | 3104 | |
eae3f88e | 3105 | return skb; |
fc70fb64 | 3106 | |
f6a78bfc HX |
3107 | out_kfree_skb: |
3108 | kfree_skb(skb); | |
eae3f88e | 3109 | out_null: |
d21fd63e | 3110 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3111 | return NULL; |
3112 | } | |
6afff0ca | 3113 | |
55a93b3e ED |
3114 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
3115 | { | |
3116 | struct sk_buff *next, *head = NULL, *tail; | |
3117 | ||
bec3cfdc | 3118 | for (; skb != NULL; skb = next) { |
55a93b3e ED |
3119 | next = skb->next; |
3120 | skb->next = NULL; | |
bec3cfdc ED |
3121 | |
3122 | /* in case skb wont be segmented, point to itself */ | |
3123 | skb->prev = skb; | |
3124 | ||
55a93b3e | 3125 | skb = validate_xmit_skb(skb, dev); |
bec3cfdc ED |
3126 | if (!skb) |
3127 | continue; | |
55a93b3e | 3128 | |
bec3cfdc ED |
3129 | if (!head) |
3130 | head = skb; | |
3131 | else | |
3132 | tail->next = skb; | |
3133 | /* If skb was segmented, skb->prev points to | |
3134 | * the last segment. If not, it still contains skb. | |
3135 | */ | |
3136 | tail = skb->prev; | |
55a93b3e ED |
3137 | } |
3138 | return head; | |
f6a78bfc | 3139 | } |
104ba78c | 3140 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3141 | |
1def9238 ED |
3142 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3143 | { | |
3144 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3145 | ||
3146 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3147 | ||
3148 | /* To get more precise estimation of bytes sent on wire, | |
3149 | * we add to pkt_len the headers size of all segments | |
3150 | */ | |
3151 | if (shinfo->gso_size) { | |
757b8b1d | 3152 | unsigned int hdr_len; |
15e5a030 | 3153 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3154 | |
757b8b1d ED |
3155 | /* mac layer + network layer */ |
3156 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3157 | ||
3158 | /* + transport layer */ | |
1def9238 ED |
3159 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
3160 | hdr_len += tcp_hdrlen(skb); | |
3161 | else | |
3162 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
3163 | |
3164 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3165 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3166 | shinfo->gso_size); | |
3167 | ||
3168 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3169 | } |
3170 | } | |
3171 | ||
bbd8a0d3 KK |
3172 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3173 | struct net_device *dev, | |
3174 | struct netdev_queue *txq) | |
3175 | { | |
3176 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3177 | struct sk_buff *to_free = NULL; |
a2da570d | 3178 | bool contended; |
bbd8a0d3 KK |
3179 | int rc; |
3180 | ||
a2da570d | 3181 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
3182 | /* |
3183 | * Heuristic to force contended enqueues to serialize on a | |
3184 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3185 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3186 | * often and dequeue packets faster. |
79640a4c | 3187 | */ |
a2da570d | 3188 | contended = qdisc_is_running(q); |
79640a4c ED |
3189 | if (unlikely(contended)) |
3190 | spin_lock(&q->busylock); | |
3191 | ||
bbd8a0d3 KK |
3192 | spin_lock(root_lock); |
3193 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3194 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3195 | rc = NET_XMIT_DROP; |
3196 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3197 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3198 | /* |
3199 | * This is a work-conserving queue; there are no old skbs | |
3200 | * waiting to be sent out; and the qdisc is not running - | |
3201 | * xmit the skb directly. | |
3202 | */ | |
bfe0d029 | 3203 | |
bfe0d029 ED |
3204 | qdisc_bstats_update(q, skb); |
3205 | ||
55a93b3e | 3206 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3207 | if (unlikely(contended)) { |
3208 | spin_unlock(&q->busylock); | |
3209 | contended = false; | |
3210 | } | |
bbd8a0d3 | 3211 | __qdisc_run(q); |
79640a4c | 3212 | } else |
bc135b23 | 3213 | qdisc_run_end(q); |
bbd8a0d3 KK |
3214 | |
3215 | rc = NET_XMIT_SUCCESS; | |
3216 | } else { | |
520ac30f | 3217 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3218 | if (qdisc_run_begin(q)) { |
3219 | if (unlikely(contended)) { | |
3220 | spin_unlock(&q->busylock); | |
3221 | contended = false; | |
3222 | } | |
3223 | __qdisc_run(q); | |
3224 | } | |
bbd8a0d3 KK |
3225 | } |
3226 | spin_unlock(root_lock); | |
520ac30f ED |
3227 | if (unlikely(to_free)) |
3228 | kfree_skb_list(to_free); | |
79640a4c ED |
3229 | if (unlikely(contended)) |
3230 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3231 | return rc; |
3232 | } | |
3233 | ||
86f8515f | 3234 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3235 | static void skb_update_prio(struct sk_buff *skb) |
3236 | { | |
6977a79d | 3237 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 3238 | |
91c68ce2 | 3239 | if (!skb->priority && skb->sk && map) { |
2a56a1fe TH |
3240 | unsigned int prioidx = |
3241 | sock_cgroup_prioidx(&skb->sk->sk_cgrp_data); | |
91c68ce2 ED |
3242 | |
3243 | if (prioidx < map->priomap_len) | |
3244 | skb->priority = map->priomap[prioidx]; | |
3245 | } | |
5bc1421e NH |
3246 | } |
3247 | #else | |
3248 | #define skb_update_prio(skb) | |
3249 | #endif | |
3250 | ||
f60e5990 | 3251 | DEFINE_PER_CPU(int, xmit_recursion); |
3252 | EXPORT_SYMBOL(xmit_recursion); | |
3253 | ||
95603e22 MM |
3254 | /** |
3255 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3256 | * @net: network namespace this loopback is happening in |
3257 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3258 | * @skb: buffer to transmit |
3259 | */ | |
0c4b51f0 | 3260 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3261 | { |
3262 | skb_reset_mac_header(skb); | |
3263 | __skb_pull(skb, skb_network_offset(skb)); | |
3264 | skb->pkt_type = PACKET_LOOPBACK; | |
3265 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3266 | WARN_ON(!skb_dst(skb)); | |
3267 | skb_dst_force(skb); | |
3268 | netif_rx_ni(skb); | |
3269 | return 0; | |
3270 | } | |
3271 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3272 | ||
1f211a1b DB |
3273 | #ifdef CONFIG_NET_EGRESS |
3274 | static struct sk_buff * | |
3275 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3276 | { | |
46209401 | 3277 | struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); |
1f211a1b DB |
3278 | struct tcf_result cl_res; |
3279 | ||
46209401 | 3280 | if (!miniq) |
1f211a1b DB |
3281 | return skb; |
3282 | ||
8dc07fdb | 3283 | /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ |
46209401 | 3284 | mini_qdisc_bstats_cpu_update(miniq, skb); |
1f211a1b | 3285 | |
46209401 | 3286 | switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { |
1f211a1b DB |
3287 | case TC_ACT_OK: |
3288 | case TC_ACT_RECLASSIFY: | |
3289 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3290 | break; | |
3291 | case TC_ACT_SHOT: | |
46209401 | 3292 | mini_qdisc_qstats_cpu_drop(miniq); |
1f211a1b | 3293 | *ret = NET_XMIT_DROP; |
7e2c3aea DB |
3294 | kfree_skb(skb); |
3295 | return NULL; | |
1f211a1b DB |
3296 | case TC_ACT_STOLEN: |
3297 | case TC_ACT_QUEUED: | |
e25ea21f | 3298 | case TC_ACT_TRAP: |
1f211a1b | 3299 | *ret = NET_XMIT_SUCCESS; |
7e2c3aea | 3300 | consume_skb(skb); |
1f211a1b DB |
3301 | return NULL; |
3302 | case TC_ACT_REDIRECT: | |
3303 | /* No need to push/pop skb's mac_header here on egress! */ | |
3304 | skb_do_redirect(skb); | |
3305 | *ret = NET_XMIT_SUCCESS; | |
3306 | return NULL; | |
3307 | default: | |
3308 | break; | |
3309 | } | |
3310 | ||
3311 | return skb; | |
3312 | } | |
3313 | #endif /* CONFIG_NET_EGRESS */ | |
3314 | ||
638b2a69 JP |
3315 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
3316 | { | |
3317 | #ifdef CONFIG_XPS | |
3318 | struct xps_dev_maps *dev_maps; | |
3319 | struct xps_map *map; | |
3320 | int queue_index = -1; | |
3321 | ||
3322 | rcu_read_lock(); | |
3323 | dev_maps = rcu_dereference(dev->xps_maps); | |
3324 | if (dev_maps) { | |
184c449f AD |
3325 | unsigned int tci = skb->sender_cpu - 1; |
3326 | ||
3327 | if (dev->num_tc) { | |
3328 | tci *= dev->num_tc; | |
3329 | tci += netdev_get_prio_tc_map(dev, skb->priority); | |
3330 | } | |
3331 | ||
3332 | map = rcu_dereference(dev_maps->cpu_map[tci]); | |
638b2a69 JP |
3333 | if (map) { |
3334 | if (map->len == 1) | |
3335 | queue_index = map->queues[0]; | |
3336 | else | |
3337 | queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), | |
3338 | map->len)]; | |
3339 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3340 | queue_index = -1; | |
3341 | } | |
3342 | } | |
3343 | rcu_read_unlock(); | |
3344 | ||
3345 | return queue_index; | |
3346 | #else | |
3347 | return -1; | |
3348 | #endif | |
3349 | } | |
3350 | ||
3351 | static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) | |
3352 | { | |
3353 | struct sock *sk = skb->sk; | |
3354 | int queue_index = sk_tx_queue_get(sk); | |
3355 | ||
3356 | if (queue_index < 0 || skb->ooo_okay || | |
3357 | queue_index >= dev->real_num_tx_queues) { | |
3358 | int new_index = get_xps_queue(dev, skb); | |
f4563a75 | 3359 | |
638b2a69 JP |
3360 | if (new_index < 0) |
3361 | new_index = skb_tx_hash(dev, skb); | |
3362 | ||
3363 | if (queue_index != new_index && sk && | |
004a5d01 | 3364 | sk_fullsock(sk) && |
638b2a69 JP |
3365 | rcu_access_pointer(sk->sk_dst_cache)) |
3366 | sk_tx_queue_set(sk, new_index); | |
3367 | ||
3368 | queue_index = new_index; | |
3369 | } | |
3370 | ||
3371 | return queue_index; | |
3372 | } | |
3373 | ||
3374 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, | |
3375 | struct sk_buff *skb, | |
3376 | void *accel_priv) | |
3377 | { | |
3378 | int queue_index = 0; | |
3379 | ||
3380 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
3381 | u32 sender_cpu = skb->sender_cpu - 1; |
3382 | ||
3383 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
3384 | skb->sender_cpu = raw_smp_processor_id() + 1; |
3385 | #endif | |
3386 | ||
3387 | if (dev->real_num_tx_queues != 1) { | |
3388 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 3389 | |
638b2a69 JP |
3390 | if (ops->ndo_select_queue) |
3391 | queue_index = ops->ndo_select_queue(dev, skb, accel_priv, | |
3392 | __netdev_pick_tx); | |
3393 | else | |
3394 | queue_index = __netdev_pick_tx(dev, skb); | |
3395 | ||
3396 | if (!accel_priv) | |
3397 | queue_index = netdev_cap_txqueue(dev, queue_index); | |
3398 | } | |
3399 | ||
3400 | skb_set_queue_mapping(skb, queue_index); | |
3401 | return netdev_get_tx_queue(dev, queue_index); | |
3402 | } | |
3403 | ||
d29f749e | 3404 | /** |
9d08dd3d | 3405 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3406 | * @skb: buffer to transmit |
9d08dd3d | 3407 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
3408 | * |
3409 | * Queue a buffer for transmission to a network device. The caller must | |
3410 | * have set the device and priority and built the buffer before calling | |
3411 | * this function. The function can be called from an interrupt. | |
3412 | * | |
3413 | * A negative errno code is returned on a failure. A success does not | |
3414 | * guarantee the frame will be transmitted as it may be dropped due | |
3415 | * to congestion or traffic shaping. | |
3416 | * | |
3417 | * ----------------------------------------------------------------------------------- | |
3418 | * I notice this method can also return errors from the queue disciplines, | |
3419 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3420 | * be positive. | |
3421 | * | |
3422 | * Regardless of the return value, the skb is consumed, so it is currently | |
3423 | * difficult to retry a send to this method. (You can bump the ref count | |
3424 | * before sending to hold a reference for retry if you are careful.) | |
3425 | * | |
3426 | * When calling this method, interrupts MUST be enabled. This is because | |
3427 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3428 | * --BLG | |
3429 | */ | |
0a59f3a9 | 3430 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
3431 | { |
3432 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3433 | struct netdev_queue *txq; |
1da177e4 LT |
3434 | struct Qdisc *q; |
3435 | int rc = -ENOMEM; | |
3436 | ||
6d1ccff6 ED |
3437 | skb_reset_mac_header(skb); |
3438 | ||
e7fd2885 WB |
3439 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3440 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3441 | ||
4ec93edb YH |
3442 | /* Disable soft irqs for various locks below. Also |
3443 | * stops preemption for RCU. | |
1da177e4 | 3444 | */ |
4ec93edb | 3445 | rcu_read_lock_bh(); |
1da177e4 | 3446 | |
5bc1421e NH |
3447 | skb_update_prio(skb); |
3448 | ||
1f211a1b DB |
3449 | qdisc_pkt_len_init(skb); |
3450 | #ifdef CONFIG_NET_CLS_ACT | |
8dc07fdb | 3451 | skb->tc_at_ingress = 0; |
1f211a1b DB |
3452 | # ifdef CONFIG_NET_EGRESS |
3453 | if (static_key_false(&egress_needed)) { | |
3454 | skb = sch_handle_egress(skb, &rc, dev); | |
3455 | if (!skb) | |
3456 | goto out; | |
3457 | } | |
3458 | # endif | |
3459 | #endif | |
02875878 ED |
3460 | /* If device/qdisc don't need skb->dst, release it right now while |
3461 | * its hot in this cpu cache. | |
3462 | */ | |
3463 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
3464 | skb_dst_drop(skb); | |
3465 | else | |
3466 | skb_dst_force(skb); | |
3467 | ||
f663dd9a | 3468 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 3469 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 3470 | |
cf66ba58 | 3471 | trace_net_dev_queue(skb); |
1da177e4 | 3472 | if (q->enqueue) { |
bbd8a0d3 | 3473 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 3474 | goto out; |
1da177e4 LT |
3475 | } |
3476 | ||
3477 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 3478 | * loopback, all the sorts of tunnels... |
1da177e4 | 3479 | |
eb13da1a | 3480 | * Really, it is unlikely that netif_tx_lock protection is necessary |
3481 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
3482 | * counters.) | |
3483 | * However, it is possible, that they rely on protection | |
3484 | * made by us here. | |
1da177e4 | 3485 | |
eb13da1a | 3486 | * Check this and shot the lock. It is not prone from deadlocks. |
3487 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
3488 | */ |
3489 | if (dev->flags & IFF_UP) { | |
3490 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
3491 | ||
c773e847 | 3492 | if (txq->xmit_lock_owner != cpu) { |
a70b506e DB |
3493 | if (unlikely(__this_cpu_read(xmit_recursion) > |
3494 | XMIT_RECURSION_LIMIT)) | |
745e20f1 ED |
3495 | goto recursion_alert; |
3496 | ||
1f59533f JDB |
3497 | skb = validate_xmit_skb(skb, dev); |
3498 | if (!skb) | |
d21fd63e | 3499 | goto out; |
1f59533f | 3500 | |
c773e847 | 3501 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 3502 | |
73466498 | 3503 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 3504 | __this_cpu_inc(xmit_recursion); |
ce93718f | 3505 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 3506 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 3507 | if (dev_xmit_complete(rc)) { |
c773e847 | 3508 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
3509 | goto out; |
3510 | } | |
3511 | } | |
c773e847 | 3512 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
3513 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
3514 | dev->name); | |
1da177e4 LT |
3515 | } else { |
3516 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
3517 | * unfortunately |
3518 | */ | |
3519 | recursion_alert: | |
e87cc472 JP |
3520 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
3521 | dev->name); | |
1da177e4 LT |
3522 | } |
3523 | } | |
3524 | ||
3525 | rc = -ENETDOWN; | |
d4828d85 | 3526 | rcu_read_unlock_bh(); |
1da177e4 | 3527 | |
015f0688 | 3528 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 3529 | kfree_skb_list(skb); |
1da177e4 LT |
3530 | return rc; |
3531 | out: | |
d4828d85 | 3532 | rcu_read_unlock_bh(); |
1da177e4 LT |
3533 | return rc; |
3534 | } | |
f663dd9a | 3535 | |
2b4aa3ce | 3536 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
3537 | { |
3538 | return __dev_queue_xmit(skb, NULL); | |
3539 | } | |
2b4aa3ce | 3540 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3541 | |
f663dd9a JW |
3542 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3543 | { | |
3544 | return __dev_queue_xmit(skb, accel_priv); | |
3545 | } | |
3546 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3547 | ||
1da177e4 | 3548 | |
eb13da1a | 3549 | /************************************************************************* |
3550 | * Receiver routines | |
3551 | *************************************************************************/ | |
1da177e4 | 3552 | |
6b2bedc3 | 3553 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3554 | EXPORT_SYMBOL(netdev_max_backlog); |
3555 | ||
3b098e2d | 3556 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 | 3557 | int netdev_budget __read_mostly = 300; |
7acf8a1e | 3558 | unsigned int __read_mostly netdev_budget_usecs = 2000; |
3d48b53f MT |
3559 | int weight_p __read_mostly = 64; /* old backlog weight */ |
3560 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
3561 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
3562 | int dev_rx_weight __read_mostly = 64; | |
3563 | int dev_tx_weight __read_mostly = 64; | |
1da177e4 | 3564 | |
eecfd7c4 ED |
3565 | /* Called with irq disabled */ |
3566 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3567 | struct napi_struct *napi) | |
3568 | { | |
3569 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3570 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3571 | } | |
3572 | ||
bfb564e7 KK |
3573 | #ifdef CONFIG_RPS |
3574 | ||
3575 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3576 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 3577 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
3578 | u32 rps_cpu_mask __read_mostly; |
3579 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 3580 | |
c5905afb | 3581 | struct static_key rps_needed __read_mostly; |
3df97ba8 | 3582 | EXPORT_SYMBOL(rps_needed); |
13bfff25 ED |
3583 | struct static_key rfs_needed __read_mostly; |
3584 | EXPORT_SYMBOL(rfs_needed); | |
adc9300e | 3585 | |
c445477d BH |
3586 | static struct rps_dev_flow * |
3587 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3588 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3589 | { | |
a31196b0 | 3590 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
3591 | #ifdef CONFIG_RFS_ACCEL |
3592 | struct netdev_rx_queue *rxqueue; | |
3593 | struct rps_dev_flow_table *flow_table; | |
3594 | struct rps_dev_flow *old_rflow; | |
3595 | u32 flow_id; | |
3596 | u16 rxq_index; | |
3597 | int rc; | |
3598 | ||
3599 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3600 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3601 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3602 | goto out; |
3603 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3604 | if (rxq_index == skb_get_rx_queue(skb)) | |
3605 | goto out; | |
3606 | ||
3607 | rxqueue = dev->_rx + rxq_index; | |
3608 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3609 | if (!flow_table) | |
3610 | goto out; | |
61b905da | 3611 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3612 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3613 | rxq_index, flow_id); | |
3614 | if (rc < 0) | |
3615 | goto out; | |
3616 | old_rflow = rflow; | |
3617 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3618 | rflow->filter = rc; |
3619 | if (old_rflow->filter == rflow->filter) | |
3620 | old_rflow->filter = RPS_NO_FILTER; | |
3621 | out: | |
3622 | #endif | |
3623 | rflow->last_qtail = | |
09994d1b | 3624 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3625 | } |
3626 | ||
09994d1b | 3627 | rflow->cpu = next_cpu; |
c445477d BH |
3628 | return rflow; |
3629 | } | |
3630 | ||
bfb564e7 KK |
3631 | /* |
3632 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3633 | * CPU from the RPS map of the receiving queue for a given skb. | |
3634 | * rcu_read_lock must be held on entry. | |
3635 | */ | |
3636 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3637 | struct rps_dev_flow **rflowp) | |
3638 | { | |
567e4b79 ED |
3639 | const struct rps_sock_flow_table *sock_flow_table; |
3640 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 3641 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 3642 | struct rps_map *map; |
bfb564e7 | 3643 | int cpu = -1; |
567e4b79 | 3644 | u32 tcpu; |
61b905da | 3645 | u32 hash; |
bfb564e7 KK |
3646 | |
3647 | if (skb_rx_queue_recorded(skb)) { | |
3648 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 3649 | |
62fe0b40 BH |
3650 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3651 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3652 | "%s received packet on queue %u, but number " | |
3653 | "of RX queues is %u\n", | |
3654 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3655 | goto done; |
3656 | } | |
567e4b79 ED |
3657 | rxqueue += index; |
3658 | } | |
bfb564e7 | 3659 | |
567e4b79 ED |
3660 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
3661 | ||
3662 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 3663 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 3664 | if (!flow_table && !map) |
bfb564e7 KK |
3665 | goto done; |
3666 | ||
2d47b459 | 3667 | skb_reset_network_header(skb); |
61b905da TH |
3668 | hash = skb_get_hash(skb); |
3669 | if (!hash) | |
bfb564e7 KK |
3670 | goto done; |
3671 | ||
fec5e652 TH |
3672 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
3673 | if (flow_table && sock_flow_table) { | |
fec5e652 | 3674 | struct rps_dev_flow *rflow; |
567e4b79 ED |
3675 | u32 next_cpu; |
3676 | u32 ident; | |
3677 | ||
3678 | /* First check into global flow table if there is a match */ | |
3679 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
3680 | if ((ident ^ hash) & ~rps_cpu_mask) | |
3681 | goto try_rps; | |
fec5e652 | 3682 | |
567e4b79 ED |
3683 | next_cpu = ident & rps_cpu_mask; |
3684 | ||
3685 | /* OK, now we know there is a match, | |
3686 | * we can look at the local (per receive queue) flow table | |
3687 | */ | |
61b905da | 3688 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3689 | tcpu = rflow->cpu; |
3690 | ||
fec5e652 TH |
3691 | /* |
3692 | * If the desired CPU (where last recvmsg was done) is | |
3693 | * different from current CPU (one in the rx-queue flow | |
3694 | * table entry), switch if one of the following holds: | |
a31196b0 | 3695 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
3696 | * - Current CPU is offline. |
3697 | * - The current CPU's queue tail has advanced beyond the | |
3698 | * last packet that was enqueued using this table entry. | |
3699 | * This guarantees that all previous packets for the flow | |
3700 | * have been dequeued, thus preserving in order delivery. | |
3701 | */ | |
3702 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 3703 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 3704 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
3705 | rflow->last_qtail)) >= 0)) { |
3706 | tcpu = next_cpu; | |
c445477d | 3707 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3708 | } |
c445477d | 3709 | |
a31196b0 | 3710 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
3711 | *rflowp = rflow; |
3712 | cpu = tcpu; | |
3713 | goto done; | |
3714 | } | |
3715 | } | |
3716 | ||
567e4b79 ED |
3717 | try_rps: |
3718 | ||
0a9627f2 | 3719 | if (map) { |
8fc54f68 | 3720 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3721 | if (cpu_online(tcpu)) { |
3722 | cpu = tcpu; | |
3723 | goto done; | |
3724 | } | |
3725 | } | |
3726 | ||
3727 | done: | |
0a9627f2 TH |
3728 | return cpu; |
3729 | } | |
3730 | ||
c445477d BH |
3731 | #ifdef CONFIG_RFS_ACCEL |
3732 | ||
3733 | /** | |
3734 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3735 | * @dev: Device on which the filter was set | |
3736 | * @rxq_index: RX queue index | |
3737 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3738 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3739 | * | |
3740 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3741 | * this function for each installed filter and remove the filters for | |
3742 | * which it returns %true. | |
3743 | */ | |
3744 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3745 | u32 flow_id, u16 filter_id) | |
3746 | { | |
3747 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3748 | struct rps_dev_flow_table *flow_table; | |
3749 | struct rps_dev_flow *rflow; | |
3750 | bool expire = true; | |
a31196b0 | 3751 | unsigned int cpu; |
c445477d BH |
3752 | |
3753 | rcu_read_lock(); | |
3754 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3755 | if (flow_table && flow_id <= flow_table->mask) { | |
3756 | rflow = &flow_table->flows[flow_id]; | |
3757 | cpu = ACCESS_ONCE(rflow->cpu); | |
a31196b0 | 3758 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
3759 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
3760 | rflow->last_qtail) < | |
3761 | (int)(10 * flow_table->mask))) | |
3762 | expire = false; | |
3763 | } | |
3764 | rcu_read_unlock(); | |
3765 | return expire; | |
3766 | } | |
3767 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3768 | ||
3769 | #endif /* CONFIG_RFS_ACCEL */ | |
3770 | ||
0a9627f2 | 3771 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3772 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3773 | { |
e36fa2f7 ED |
3774 | struct softnet_data *sd = data; |
3775 | ||
eecfd7c4 | 3776 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3777 | sd->received_rps++; |
0a9627f2 | 3778 | } |
e36fa2f7 | 3779 | |
fec5e652 | 3780 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3781 | |
e36fa2f7 ED |
3782 | /* |
3783 | * Check if this softnet_data structure is another cpu one | |
3784 | * If yes, queue it to our IPI list and return 1 | |
3785 | * If no, return 0 | |
3786 | */ | |
3787 | static int rps_ipi_queued(struct softnet_data *sd) | |
3788 | { | |
3789 | #ifdef CONFIG_RPS | |
903ceff7 | 3790 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
3791 | |
3792 | if (sd != mysd) { | |
3793 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3794 | mysd->rps_ipi_list = sd; | |
3795 | ||
3796 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3797 | return 1; | |
3798 | } | |
3799 | #endif /* CONFIG_RPS */ | |
3800 | return 0; | |
3801 | } | |
3802 | ||
99bbc707 WB |
3803 | #ifdef CONFIG_NET_FLOW_LIMIT |
3804 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3805 | #endif | |
3806 | ||
3807 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3808 | { | |
3809 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3810 | struct sd_flow_limit *fl; | |
3811 | struct softnet_data *sd; | |
3812 | unsigned int old_flow, new_flow; | |
3813 | ||
3814 | if (qlen < (netdev_max_backlog >> 1)) | |
3815 | return false; | |
3816 | ||
903ceff7 | 3817 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
3818 | |
3819 | rcu_read_lock(); | |
3820 | fl = rcu_dereference(sd->flow_limit); | |
3821 | if (fl) { | |
3958afa1 | 3822 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3823 | old_flow = fl->history[fl->history_head]; |
3824 | fl->history[fl->history_head] = new_flow; | |
3825 | ||
3826 | fl->history_head++; | |
3827 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3828 | ||
3829 | if (likely(fl->buckets[old_flow])) | |
3830 | fl->buckets[old_flow]--; | |
3831 | ||
3832 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3833 | fl->count++; | |
3834 | rcu_read_unlock(); | |
3835 | return true; | |
3836 | } | |
3837 | } | |
3838 | rcu_read_unlock(); | |
3839 | #endif | |
3840 | return false; | |
3841 | } | |
3842 | ||
0a9627f2 TH |
3843 | /* |
3844 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3845 | * queue (may be a remote CPU queue). | |
3846 | */ | |
fec5e652 TH |
3847 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3848 | unsigned int *qtail) | |
0a9627f2 | 3849 | { |
e36fa2f7 | 3850 | struct softnet_data *sd; |
0a9627f2 | 3851 | unsigned long flags; |
99bbc707 | 3852 | unsigned int qlen; |
0a9627f2 | 3853 | |
e36fa2f7 | 3854 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3855 | |
3856 | local_irq_save(flags); | |
0a9627f2 | 3857 | |
e36fa2f7 | 3858 | rps_lock(sd); |
e9e4dd32 JA |
3859 | if (!netif_running(skb->dev)) |
3860 | goto drop; | |
99bbc707 WB |
3861 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3862 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 3863 | if (qlen) { |
0a9627f2 | 3864 | enqueue: |
e36fa2f7 | 3865 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3866 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3867 | rps_unlock(sd); |
152102c7 | 3868 | local_irq_restore(flags); |
0a9627f2 TH |
3869 | return NET_RX_SUCCESS; |
3870 | } | |
3871 | ||
ebda37c2 ED |
3872 | /* Schedule NAPI for backlog device |
3873 | * We can use non atomic operation since we own the queue lock | |
3874 | */ | |
3875 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3876 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3877 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3878 | } |
3879 | goto enqueue; | |
3880 | } | |
3881 | ||
e9e4dd32 | 3882 | drop: |
dee42870 | 3883 | sd->dropped++; |
e36fa2f7 | 3884 | rps_unlock(sd); |
0a9627f2 | 3885 | |
0a9627f2 TH |
3886 | local_irq_restore(flags); |
3887 | ||
caf586e5 | 3888 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3889 | kfree_skb(skb); |
3890 | return NET_RX_DROP; | |
3891 | } | |
1da177e4 | 3892 | |
d4455169 JF |
3893 | static u32 netif_receive_generic_xdp(struct sk_buff *skb, |
3894 | struct bpf_prog *xdp_prog) | |
3895 | { | |
de8f3a83 | 3896 | u32 metalen, act = XDP_DROP; |
d4455169 | 3897 | struct xdp_buff xdp; |
d4455169 JF |
3898 | void *orig_data; |
3899 | int hlen, off; | |
3900 | u32 mac_len; | |
3901 | ||
3902 | /* Reinjected packets coming from act_mirred or similar should | |
3903 | * not get XDP generic processing. | |
3904 | */ | |
3905 | if (skb_cloned(skb)) | |
3906 | return XDP_PASS; | |
3907 | ||
de8f3a83 DB |
3908 | /* XDP packets must be linear and must have sufficient headroom |
3909 | * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also | |
3910 | * native XDP provides, thus we need to do it here as well. | |
3911 | */ | |
3912 | if (skb_is_nonlinear(skb) || | |
3913 | skb_headroom(skb) < XDP_PACKET_HEADROOM) { | |
3914 | int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); | |
3915 | int troom = skb->tail + skb->data_len - skb->end; | |
3916 | ||
3917 | /* In case we have to go down the path and also linearize, | |
3918 | * then lets do the pskb_expand_head() work just once here. | |
3919 | */ | |
3920 | if (pskb_expand_head(skb, | |
3921 | hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, | |
3922 | troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) | |
3923 | goto do_drop; | |
3924 | if (troom > 0 && __skb_linearize(skb)) | |
3925 | goto do_drop; | |
3926 | } | |
d4455169 JF |
3927 | |
3928 | /* The XDP program wants to see the packet starting at the MAC | |
3929 | * header. | |
3930 | */ | |
3931 | mac_len = skb->data - skb_mac_header(skb); | |
3932 | hlen = skb_headlen(skb) + mac_len; | |
3933 | xdp.data = skb->data - mac_len; | |
de8f3a83 | 3934 | xdp.data_meta = xdp.data; |
d4455169 JF |
3935 | xdp.data_end = xdp.data + hlen; |
3936 | xdp.data_hard_start = skb->data - skb_headroom(skb); | |
3937 | orig_data = xdp.data; | |
3938 | ||
3939 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
3940 | ||
3941 | off = xdp.data - orig_data; | |
3942 | if (off > 0) | |
3943 | __skb_pull(skb, off); | |
3944 | else if (off < 0) | |
3945 | __skb_push(skb, -off); | |
92dd5452 | 3946 | skb->mac_header += off; |
d4455169 JF |
3947 | |
3948 | switch (act) { | |
6103aa96 | 3949 | case XDP_REDIRECT: |
d4455169 JF |
3950 | case XDP_TX: |
3951 | __skb_push(skb, mac_len); | |
de8f3a83 | 3952 | break; |
d4455169 | 3953 | case XDP_PASS: |
de8f3a83 DB |
3954 | metalen = xdp.data - xdp.data_meta; |
3955 | if (metalen) | |
3956 | skb_metadata_set(skb, metalen); | |
d4455169 | 3957 | break; |
d4455169 JF |
3958 | default: |
3959 | bpf_warn_invalid_xdp_action(act); | |
3960 | /* fall through */ | |
3961 | case XDP_ABORTED: | |
3962 | trace_xdp_exception(skb->dev, xdp_prog, act); | |
3963 | /* fall through */ | |
3964 | case XDP_DROP: | |
3965 | do_drop: | |
3966 | kfree_skb(skb); | |
3967 | break; | |
3968 | } | |
3969 | ||
3970 | return act; | |
3971 | } | |
3972 | ||
3973 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
3974 | * network taps in order to match in-driver-XDP behavior. | |
3975 | */ | |
7c497478 | 3976 | void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) |
d4455169 JF |
3977 | { |
3978 | struct net_device *dev = skb->dev; | |
3979 | struct netdev_queue *txq; | |
3980 | bool free_skb = true; | |
3981 | int cpu, rc; | |
3982 | ||
3983 | txq = netdev_pick_tx(dev, skb, NULL); | |
3984 | cpu = smp_processor_id(); | |
3985 | HARD_TX_LOCK(dev, txq, cpu); | |
3986 | if (!netif_xmit_stopped(txq)) { | |
3987 | rc = netdev_start_xmit(skb, dev, txq, 0); | |
3988 | if (dev_xmit_complete(rc)) | |
3989 | free_skb = false; | |
3990 | } | |
3991 | HARD_TX_UNLOCK(dev, txq); | |
3992 | if (free_skb) { | |
3993 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
3994 | kfree_skb(skb); | |
3995 | } | |
3996 | } | |
7c497478 | 3997 | EXPORT_SYMBOL_GPL(generic_xdp_tx); |
d4455169 JF |
3998 | |
3999 | static struct static_key generic_xdp_needed __read_mostly; | |
4000 | ||
7c497478 | 4001 | int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) |
d4455169 | 4002 | { |
d4455169 JF |
4003 | if (xdp_prog) { |
4004 | u32 act = netif_receive_generic_xdp(skb, xdp_prog); | |
6103aa96 | 4005 | int err; |
d4455169 JF |
4006 | |
4007 | if (act != XDP_PASS) { | |
6103aa96 JF |
4008 | switch (act) { |
4009 | case XDP_REDIRECT: | |
2facaad6 JDB |
4010 | err = xdp_do_generic_redirect(skb->dev, skb, |
4011 | xdp_prog); | |
6103aa96 JF |
4012 | if (err) |
4013 | goto out_redir; | |
4014 | /* fallthru to submit skb */ | |
4015 | case XDP_TX: | |
d4455169 | 4016 | generic_xdp_tx(skb, xdp_prog); |
6103aa96 JF |
4017 | break; |
4018 | } | |
d4455169 JF |
4019 | return XDP_DROP; |
4020 | } | |
4021 | } | |
4022 | return XDP_PASS; | |
6103aa96 | 4023 | out_redir: |
6103aa96 JF |
4024 | kfree_skb(skb); |
4025 | return XDP_DROP; | |
d4455169 | 4026 | } |
7c497478 | 4027 | EXPORT_SYMBOL_GPL(do_xdp_generic); |
d4455169 | 4028 | |
ae78dbfa | 4029 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 4030 | { |
b0e28f1e | 4031 | int ret; |
1da177e4 | 4032 | |
588f0330 | 4033 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 4034 | |
cf66ba58 | 4035 | trace_netif_rx(skb); |
d4455169 JF |
4036 | |
4037 | if (static_key_false(&generic_xdp_needed)) { | |
bbbe211c JF |
4038 | int ret; |
4039 | ||
4040 | preempt_disable(); | |
4041 | rcu_read_lock(); | |
4042 | ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
4043 | rcu_read_unlock(); | |
4044 | preempt_enable(); | |
d4455169 | 4045 | |
6103aa96 JF |
4046 | /* Consider XDP consuming the packet a success from |
4047 | * the netdev point of view we do not want to count | |
4048 | * this as an error. | |
4049 | */ | |
d4455169 | 4050 | if (ret != XDP_PASS) |
6103aa96 | 4051 | return NET_RX_SUCCESS; |
d4455169 JF |
4052 | } |
4053 | ||
df334545 | 4054 | #ifdef CONFIG_RPS |
c5905afb | 4055 | if (static_key_false(&rps_needed)) { |
fec5e652 | 4056 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
4057 | int cpu; |
4058 | ||
cece1945 | 4059 | preempt_disable(); |
b0e28f1e | 4060 | rcu_read_lock(); |
fec5e652 TH |
4061 | |
4062 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
4063 | if (cpu < 0) |
4064 | cpu = smp_processor_id(); | |
fec5e652 TH |
4065 | |
4066 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4067 | ||
b0e28f1e | 4068 | rcu_read_unlock(); |
cece1945 | 4069 | preempt_enable(); |
adc9300e ED |
4070 | } else |
4071 | #endif | |
fec5e652 TH |
4072 | { |
4073 | unsigned int qtail; | |
f4563a75 | 4074 | |
fec5e652 TH |
4075 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); |
4076 | put_cpu(); | |
4077 | } | |
b0e28f1e | 4078 | return ret; |
1da177e4 | 4079 | } |
ae78dbfa BH |
4080 | |
4081 | /** | |
4082 | * netif_rx - post buffer to the network code | |
4083 | * @skb: buffer to post | |
4084 | * | |
4085 | * This function receives a packet from a device driver and queues it for | |
4086 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
4087 | * may be dropped during processing for congestion control or by the | |
4088 | * protocol layers. | |
4089 | * | |
4090 | * return values: | |
4091 | * NET_RX_SUCCESS (no congestion) | |
4092 | * NET_RX_DROP (packet was dropped) | |
4093 | * | |
4094 | */ | |
4095 | ||
4096 | int netif_rx(struct sk_buff *skb) | |
4097 | { | |
4098 | trace_netif_rx_entry(skb); | |
4099 | ||
4100 | return netif_rx_internal(skb); | |
4101 | } | |
d1b19dff | 4102 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
4103 | |
4104 | int netif_rx_ni(struct sk_buff *skb) | |
4105 | { | |
4106 | int err; | |
4107 | ||
ae78dbfa BH |
4108 | trace_netif_rx_ni_entry(skb); |
4109 | ||
1da177e4 | 4110 | preempt_disable(); |
ae78dbfa | 4111 | err = netif_rx_internal(skb); |
1da177e4 LT |
4112 | if (local_softirq_pending()) |
4113 | do_softirq(); | |
4114 | preempt_enable(); | |
4115 | ||
4116 | return err; | |
4117 | } | |
1da177e4 LT |
4118 | EXPORT_SYMBOL(netif_rx_ni); |
4119 | ||
0766f788 | 4120 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 4121 | { |
903ceff7 | 4122 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
4123 | |
4124 | if (sd->completion_queue) { | |
4125 | struct sk_buff *clist; | |
4126 | ||
4127 | local_irq_disable(); | |
4128 | clist = sd->completion_queue; | |
4129 | sd->completion_queue = NULL; | |
4130 | local_irq_enable(); | |
4131 | ||
4132 | while (clist) { | |
4133 | struct sk_buff *skb = clist; | |
f4563a75 | 4134 | |
1da177e4 LT |
4135 | clist = clist->next; |
4136 | ||
63354797 | 4137 | WARN_ON(refcount_read(&skb->users)); |
e6247027 ED |
4138 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
4139 | trace_consume_skb(skb); | |
4140 | else | |
4141 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
4142 | |
4143 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
4144 | __kfree_skb(skb); | |
4145 | else | |
4146 | __kfree_skb_defer(skb); | |
1da177e4 | 4147 | } |
15fad714 JDB |
4148 | |
4149 | __kfree_skb_flush(); | |
1da177e4 LT |
4150 | } |
4151 | ||
4152 | if (sd->output_queue) { | |
37437bb2 | 4153 | struct Qdisc *head; |
1da177e4 LT |
4154 | |
4155 | local_irq_disable(); | |
4156 | head = sd->output_queue; | |
4157 | sd->output_queue = NULL; | |
a9cbd588 | 4158 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
4159 | local_irq_enable(); |
4160 | ||
4161 | while (head) { | |
37437bb2 DM |
4162 | struct Qdisc *q = head; |
4163 | spinlock_t *root_lock; | |
4164 | ||
1da177e4 LT |
4165 | head = head->next_sched; |
4166 | ||
5fb66229 | 4167 | root_lock = qdisc_lock(q); |
3bcb846c ED |
4168 | spin_lock(root_lock); |
4169 | /* We need to make sure head->next_sched is read | |
4170 | * before clearing __QDISC_STATE_SCHED | |
4171 | */ | |
4172 | smp_mb__before_atomic(); | |
4173 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
4174 | qdisc_run(q); | |
4175 | spin_unlock(root_lock); | |
1da177e4 LT |
4176 | } |
4177 | } | |
4178 | } | |
4179 | ||
181402a5 | 4180 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
4181 | /* This hook is defined here for ATM LANE */ |
4182 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
4183 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 4184 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 4185 | #endif |
1da177e4 | 4186 | |
1f211a1b DB |
4187 | static inline struct sk_buff * |
4188 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
4189 | struct net_device *orig_dev) | |
f697c3e8 | 4190 | { |
e7582bab | 4191 | #ifdef CONFIG_NET_CLS_ACT |
46209401 | 4192 | struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); |
d2788d34 | 4193 | struct tcf_result cl_res; |
24824a09 | 4194 | |
c9e99fd0 DB |
4195 | /* If there's at least one ingress present somewhere (so |
4196 | * we get here via enabled static key), remaining devices | |
4197 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 4198 | * out here. |
c9e99fd0 | 4199 | */ |
46209401 | 4200 | if (!miniq) |
4577139b | 4201 | return skb; |
46209401 | 4202 | |
f697c3e8 HX |
4203 | if (*pt_prev) { |
4204 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4205 | *pt_prev = NULL; | |
1da177e4 LT |
4206 | } |
4207 | ||
3365495c | 4208 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
8dc07fdb | 4209 | skb->tc_at_ingress = 1; |
46209401 | 4210 | mini_qdisc_bstats_cpu_update(miniq, skb); |
c9e99fd0 | 4211 | |
46209401 | 4212 | switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { |
d2788d34 DB |
4213 | case TC_ACT_OK: |
4214 | case TC_ACT_RECLASSIFY: | |
4215 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
4216 | break; | |
4217 | case TC_ACT_SHOT: | |
46209401 | 4218 | mini_qdisc_qstats_cpu_drop(miniq); |
8a3a4c6e ED |
4219 | kfree_skb(skb); |
4220 | return NULL; | |
d2788d34 DB |
4221 | case TC_ACT_STOLEN: |
4222 | case TC_ACT_QUEUED: | |
e25ea21f | 4223 | case TC_ACT_TRAP: |
8a3a4c6e | 4224 | consume_skb(skb); |
d2788d34 | 4225 | return NULL; |
27b29f63 AS |
4226 | case TC_ACT_REDIRECT: |
4227 | /* skb_mac_header check was done by cls/act_bpf, so | |
4228 | * we can safely push the L2 header back before | |
4229 | * redirecting to another netdev | |
4230 | */ | |
4231 | __skb_push(skb, skb->mac_len); | |
4232 | skb_do_redirect(skb); | |
4233 | return NULL; | |
d2788d34 DB |
4234 | default: |
4235 | break; | |
f697c3e8 | 4236 | } |
e7582bab | 4237 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
4238 | return skb; |
4239 | } | |
1da177e4 | 4240 | |
24b27fc4 MB |
4241 | /** |
4242 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
4243 | * @dev: device to check | |
4244 | * | |
4245 | * Check if a receive handler is already registered for a given device. | |
4246 | * Return true if there one. | |
4247 | * | |
4248 | * The caller must hold the rtnl_mutex. | |
4249 | */ | |
4250 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
4251 | { | |
4252 | ASSERT_RTNL(); | |
4253 | return dev && rtnl_dereference(dev->rx_handler); | |
4254 | } | |
4255 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
4256 | ||
ab95bfe0 JP |
4257 | /** |
4258 | * netdev_rx_handler_register - register receive handler | |
4259 | * @dev: device to register a handler for | |
4260 | * @rx_handler: receive handler to register | |
93e2c32b | 4261 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 4262 | * |
e227867f | 4263 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
4264 | * called from __netif_receive_skb. A negative errno code is returned |
4265 | * on a failure. | |
4266 | * | |
4267 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
4268 | * |
4269 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
4270 | */ |
4271 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
4272 | rx_handler_func_t *rx_handler, |
4273 | void *rx_handler_data) | |
ab95bfe0 | 4274 | { |
1b7cd004 | 4275 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
4276 | return -EBUSY; |
4277 | ||
00cfec37 | 4278 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 4279 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
4280 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
4281 | ||
4282 | return 0; | |
4283 | } | |
4284 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
4285 | ||
4286 | /** | |
4287 | * netdev_rx_handler_unregister - unregister receive handler | |
4288 | * @dev: device to unregister a handler from | |
4289 | * | |
166ec369 | 4290 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
4291 | * |
4292 | * The caller must hold the rtnl_mutex. | |
4293 | */ | |
4294 | void netdev_rx_handler_unregister(struct net_device *dev) | |
4295 | { | |
4296 | ||
4297 | ASSERT_RTNL(); | |
a9b3cd7f | 4298 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
4299 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
4300 | * section has a guarantee to see a non NULL rx_handler_data | |
4301 | * as well. | |
4302 | */ | |
4303 | synchronize_net(); | |
a9b3cd7f | 4304 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
4305 | } |
4306 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
4307 | ||
b4b9e355 MG |
4308 | /* |
4309 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
4310 | * the special handling of PFMEMALLOC skbs. | |
4311 | */ | |
4312 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
4313 | { | |
4314 | switch (skb->protocol) { | |
2b8837ae JP |
4315 | case htons(ETH_P_ARP): |
4316 | case htons(ETH_P_IP): | |
4317 | case htons(ETH_P_IPV6): | |
4318 | case htons(ETH_P_8021Q): | |
4319 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
4320 | return true; |
4321 | default: | |
4322 | return false; | |
4323 | } | |
4324 | } | |
4325 | ||
e687ad60 PN |
4326 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
4327 | int *ret, struct net_device *orig_dev) | |
4328 | { | |
e7582bab | 4329 | #ifdef CONFIG_NETFILTER_INGRESS |
e687ad60 | 4330 | if (nf_hook_ingress_active(skb)) { |
2c1e2703 AC |
4331 | int ingress_retval; |
4332 | ||
e687ad60 PN |
4333 | if (*pt_prev) { |
4334 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4335 | *pt_prev = NULL; | |
4336 | } | |
4337 | ||
2c1e2703 AC |
4338 | rcu_read_lock(); |
4339 | ingress_retval = nf_hook_ingress(skb); | |
4340 | rcu_read_unlock(); | |
4341 | return ingress_retval; | |
e687ad60 | 4342 | } |
e7582bab | 4343 | #endif /* CONFIG_NETFILTER_INGRESS */ |
e687ad60 PN |
4344 | return 0; |
4345 | } | |
e687ad60 | 4346 | |
9754e293 | 4347 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
4348 | { |
4349 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 4350 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 4351 | struct net_device *orig_dev; |
8a4eb573 | 4352 | bool deliver_exact = false; |
1da177e4 | 4353 | int ret = NET_RX_DROP; |
252e3346 | 4354 | __be16 type; |
1da177e4 | 4355 | |
588f0330 | 4356 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 4357 | |
cf66ba58 | 4358 | trace_netif_receive_skb(skb); |
9b22ea56 | 4359 | |
cc9bd5ce | 4360 | orig_dev = skb->dev; |
8f903c70 | 4361 | |
c1d2bbe1 | 4362 | skb_reset_network_header(skb); |
fda55eca ED |
4363 | if (!skb_transport_header_was_set(skb)) |
4364 | skb_reset_transport_header(skb); | |
0b5c9db1 | 4365 | skb_reset_mac_len(skb); |
1da177e4 LT |
4366 | |
4367 | pt_prev = NULL; | |
4368 | ||
63d8ea7f | 4369 | another_round: |
b6858177 | 4370 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
4371 | |
4372 | __this_cpu_inc(softnet_data.processed); | |
4373 | ||
8ad227ff PM |
4374 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
4375 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 4376 | skb = skb_vlan_untag(skb); |
bcc6d479 | 4377 | if (unlikely(!skb)) |
2c17d27c | 4378 | goto out; |
bcc6d479 JP |
4379 | } |
4380 | ||
e7246e12 WB |
4381 | if (skb_skip_tc_classify(skb)) |
4382 | goto skip_classify; | |
1da177e4 | 4383 | |
9754e293 | 4384 | if (pfmemalloc) |
b4b9e355 MG |
4385 | goto skip_taps; |
4386 | ||
1da177e4 | 4387 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
4388 | if (pt_prev) |
4389 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4390 | pt_prev = ptype; | |
4391 | } | |
4392 | ||
4393 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
4394 | if (pt_prev) | |
4395 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4396 | pt_prev = ptype; | |
1da177e4 LT |
4397 | } |
4398 | ||
b4b9e355 | 4399 | skip_taps: |
1cf51900 | 4400 | #ifdef CONFIG_NET_INGRESS |
4577139b | 4401 | if (static_key_false(&ingress_needed)) { |
1f211a1b | 4402 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); |
4577139b | 4403 | if (!skb) |
2c17d27c | 4404 | goto out; |
e687ad60 PN |
4405 | |
4406 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 4407 | goto out; |
4577139b | 4408 | } |
1cf51900 | 4409 | #endif |
a5135bcf | 4410 | skb_reset_tc(skb); |
e7246e12 | 4411 | skip_classify: |
9754e293 | 4412 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
4413 | goto drop; |
4414 | ||
df8a39de | 4415 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
4416 | if (pt_prev) { |
4417 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4418 | pt_prev = NULL; | |
4419 | } | |
48cc32d3 | 4420 | if (vlan_do_receive(&skb)) |
2425717b JF |
4421 | goto another_round; |
4422 | else if (unlikely(!skb)) | |
2c17d27c | 4423 | goto out; |
2425717b JF |
4424 | } |
4425 | ||
48cc32d3 | 4426 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
4427 | if (rx_handler) { |
4428 | if (pt_prev) { | |
4429 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4430 | pt_prev = NULL; | |
4431 | } | |
8a4eb573 JP |
4432 | switch (rx_handler(&skb)) { |
4433 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 4434 | ret = NET_RX_SUCCESS; |
2c17d27c | 4435 | goto out; |
8a4eb573 | 4436 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 4437 | goto another_round; |
8a4eb573 JP |
4438 | case RX_HANDLER_EXACT: |
4439 | deliver_exact = true; | |
4440 | case RX_HANDLER_PASS: | |
4441 | break; | |
4442 | default: | |
4443 | BUG(); | |
4444 | } | |
ab95bfe0 | 4445 | } |
1da177e4 | 4446 | |
df8a39de JP |
4447 | if (unlikely(skb_vlan_tag_present(skb))) { |
4448 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
4449 | skb->pkt_type = PACKET_OTHERHOST; |
4450 | /* Note: we might in the future use prio bits | |
4451 | * and set skb->priority like in vlan_do_receive() | |
4452 | * For the time being, just ignore Priority Code Point | |
4453 | */ | |
4454 | skb->vlan_tci = 0; | |
4455 | } | |
48cc32d3 | 4456 | |
7866a621 SN |
4457 | type = skb->protocol; |
4458 | ||
63d8ea7f | 4459 | /* deliver only exact match when indicated */ |
7866a621 SN |
4460 | if (likely(!deliver_exact)) { |
4461 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4462 | &ptype_base[ntohs(type) & | |
4463 | PTYPE_HASH_MASK]); | |
4464 | } | |
1f3c8804 | 4465 | |
7866a621 SN |
4466 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
4467 | &orig_dev->ptype_specific); | |
4468 | ||
4469 | if (unlikely(skb->dev != orig_dev)) { | |
4470 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4471 | &skb->dev->ptype_specific); | |
1da177e4 LT |
4472 | } |
4473 | ||
4474 | if (pt_prev) { | |
1f8b977a | 4475 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
0e698bf6 | 4476 | goto drop; |
1080e512 MT |
4477 | else |
4478 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 4479 | } else { |
b4b9e355 | 4480 | drop: |
6e7333d3 JW |
4481 | if (!deliver_exact) |
4482 | atomic_long_inc(&skb->dev->rx_dropped); | |
4483 | else | |
4484 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
4485 | kfree_skb(skb); |
4486 | /* Jamal, now you will not able to escape explaining | |
4487 | * me how you were going to use this. :-) | |
4488 | */ | |
4489 | ret = NET_RX_DROP; | |
4490 | } | |
4491 | ||
2c17d27c | 4492 | out: |
9754e293 DM |
4493 | return ret; |
4494 | } | |
4495 | ||
1c601d82 JDB |
4496 | /** |
4497 | * netif_receive_skb_core - special purpose version of netif_receive_skb | |
4498 | * @skb: buffer to process | |
4499 | * | |
4500 | * More direct receive version of netif_receive_skb(). It should | |
4501 | * only be used by callers that have a need to skip RPS and Generic XDP. | |
4502 | * Caller must also take care of handling if (page_is_)pfmemalloc. | |
4503 | * | |
4504 | * This function may only be called from softirq context and interrupts | |
4505 | * should be enabled. | |
4506 | * | |
4507 | * Return values (usually ignored): | |
4508 | * NET_RX_SUCCESS: no congestion | |
4509 | * NET_RX_DROP: packet was dropped | |
4510 | */ | |
4511 | int netif_receive_skb_core(struct sk_buff *skb) | |
4512 | { | |
4513 | int ret; | |
4514 | ||
4515 | rcu_read_lock(); | |
4516 | ret = __netif_receive_skb_core(skb, false); | |
4517 | rcu_read_unlock(); | |
4518 | ||
4519 | return ret; | |
4520 | } | |
4521 | EXPORT_SYMBOL(netif_receive_skb_core); | |
4522 | ||
9754e293 DM |
4523 | static int __netif_receive_skb(struct sk_buff *skb) |
4524 | { | |
4525 | int ret; | |
4526 | ||
4527 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 4528 | unsigned int noreclaim_flag; |
9754e293 DM |
4529 | |
4530 | /* | |
4531 | * PFMEMALLOC skbs are special, they should | |
4532 | * - be delivered to SOCK_MEMALLOC sockets only | |
4533 | * - stay away from userspace | |
4534 | * - have bounded memory usage | |
4535 | * | |
4536 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
4537 | * context down to all allocation sites. | |
4538 | */ | |
f1083048 | 4539 | noreclaim_flag = memalloc_noreclaim_save(); |
9754e293 | 4540 | ret = __netif_receive_skb_core(skb, true); |
f1083048 | 4541 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 DM |
4542 | } else |
4543 | ret = __netif_receive_skb_core(skb, false); | |
4544 | ||
1da177e4 LT |
4545 | return ret; |
4546 | } | |
0a9627f2 | 4547 | |
f4e63525 | 4548 | static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) |
b5cdae32 | 4549 | { |
58038695 | 4550 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); |
b5cdae32 DM |
4551 | struct bpf_prog *new = xdp->prog; |
4552 | int ret = 0; | |
4553 | ||
4554 | switch (xdp->command) { | |
58038695 | 4555 | case XDP_SETUP_PROG: |
b5cdae32 DM |
4556 | rcu_assign_pointer(dev->xdp_prog, new); |
4557 | if (old) | |
4558 | bpf_prog_put(old); | |
4559 | ||
4560 | if (old && !new) { | |
4561 | static_key_slow_dec(&generic_xdp_needed); | |
4562 | } else if (new && !old) { | |
4563 | static_key_slow_inc(&generic_xdp_needed); | |
4564 | dev_disable_lro(dev); | |
4565 | } | |
4566 | break; | |
b5cdae32 DM |
4567 | |
4568 | case XDP_QUERY_PROG: | |
58038695 MKL |
4569 | xdp->prog_attached = !!old; |
4570 | xdp->prog_id = old ? old->aux->id : 0; | |
b5cdae32 DM |
4571 | break; |
4572 | ||
4573 | default: | |
4574 | ret = -EINVAL; | |
4575 | break; | |
4576 | } | |
4577 | ||
4578 | return ret; | |
4579 | } | |
4580 | ||
ae78dbfa | 4581 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 4582 | { |
2c17d27c JA |
4583 | int ret; |
4584 | ||
588f0330 | 4585 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 4586 | |
c1f19b51 RC |
4587 | if (skb_defer_rx_timestamp(skb)) |
4588 | return NET_RX_SUCCESS; | |
4589 | ||
b5cdae32 | 4590 | if (static_key_false(&generic_xdp_needed)) { |
bbbe211c | 4591 | int ret; |
b5cdae32 | 4592 | |
bbbe211c JF |
4593 | preempt_disable(); |
4594 | rcu_read_lock(); | |
4595 | ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
4596 | rcu_read_unlock(); | |
4597 | preempt_enable(); | |
4598 | ||
4599 | if (ret != XDP_PASS) | |
d4455169 | 4600 | return NET_RX_DROP; |
b5cdae32 DM |
4601 | } |
4602 | ||
bbbe211c | 4603 | rcu_read_lock(); |
df334545 | 4604 | #ifdef CONFIG_RPS |
c5905afb | 4605 | if (static_key_false(&rps_needed)) { |
3b098e2d | 4606 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 4607 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 4608 | |
3b098e2d ED |
4609 | if (cpu >= 0) { |
4610 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4611 | rcu_read_unlock(); | |
adc9300e | 4612 | return ret; |
3b098e2d | 4613 | } |
fec5e652 | 4614 | } |
1e94d72f | 4615 | #endif |
2c17d27c JA |
4616 | ret = __netif_receive_skb(skb); |
4617 | rcu_read_unlock(); | |
4618 | return ret; | |
0a9627f2 | 4619 | } |
ae78dbfa BH |
4620 | |
4621 | /** | |
4622 | * netif_receive_skb - process receive buffer from network | |
4623 | * @skb: buffer to process | |
4624 | * | |
4625 | * netif_receive_skb() is the main receive data processing function. | |
4626 | * It always succeeds. The buffer may be dropped during processing | |
4627 | * for congestion control or by the protocol layers. | |
4628 | * | |
4629 | * This function may only be called from softirq context and interrupts | |
4630 | * should be enabled. | |
4631 | * | |
4632 | * Return values (usually ignored): | |
4633 | * NET_RX_SUCCESS: no congestion | |
4634 | * NET_RX_DROP: packet was dropped | |
4635 | */ | |
04eb4489 | 4636 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa BH |
4637 | { |
4638 | trace_netif_receive_skb_entry(skb); | |
4639 | ||
4640 | return netif_receive_skb_internal(skb); | |
4641 | } | |
04eb4489 | 4642 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 4643 | |
41852497 | 4644 | DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
4645 | |
4646 | /* Network device is going away, flush any packets still pending */ | |
4647 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 4648 | { |
6e583ce5 | 4649 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
4650 | struct softnet_data *sd; |
4651 | ||
4652 | local_bh_disable(); | |
4653 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 4654 | |
145dd5f9 | 4655 | local_irq_disable(); |
e36fa2f7 | 4656 | rps_lock(sd); |
6e7676c1 | 4657 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 4658 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 4659 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 4660 | kfree_skb(skb); |
76cc8b13 | 4661 | input_queue_head_incr(sd); |
6e583ce5 | 4662 | } |
6e7676c1 | 4663 | } |
e36fa2f7 | 4664 | rps_unlock(sd); |
145dd5f9 | 4665 | local_irq_enable(); |
6e7676c1 CG |
4666 | |
4667 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 4668 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
4669 | __skb_unlink(skb, &sd->process_queue); |
4670 | kfree_skb(skb); | |
76cc8b13 | 4671 | input_queue_head_incr(sd); |
6e7676c1 CG |
4672 | } |
4673 | } | |
145dd5f9 PA |
4674 | local_bh_enable(); |
4675 | } | |
4676 | ||
41852497 | 4677 | static void flush_all_backlogs(void) |
145dd5f9 PA |
4678 | { |
4679 | unsigned int cpu; | |
4680 | ||
4681 | get_online_cpus(); | |
4682 | ||
41852497 ED |
4683 | for_each_online_cpu(cpu) |
4684 | queue_work_on(cpu, system_highpri_wq, | |
4685 | per_cpu_ptr(&flush_works, cpu)); | |
145dd5f9 PA |
4686 | |
4687 | for_each_online_cpu(cpu) | |
41852497 | 4688 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 PA |
4689 | |
4690 | put_online_cpus(); | |
6e583ce5 SH |
4691 | } |
4692 | ||
d565b0a1 HX |
4693 | static int napi_gro_complete(struct sk_buff *skb) |
4694 | { | |
22061d80 | 4695 | struct packet_offload *ptype; |
d565b0a1 | 4696 | __be16 type = skb->protocol; |
22061d80 | 4697 | struct list_head *head = &offload_base; |
d565b0a1 HX |
4698 | int err = -ENOENT; |
4699 | ||
c3c7c254 ED |
4700 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
4701 | ||
fc59f9a3 HX |
4702 | if (NAPI_GRO_CB(skb)->count == 1) { |
4703 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 4704 | goto out; |
fc59f9a3 | 4705 | } |
d565b0a1 HX |
4706 | |
4707 | rcu_read_lock(); | |
4708 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4709 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
4710 | continue; |
4711 | ||
299603e8 | 4712 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
4713 | break; |
4714 | } | |
4715 | rcu_read_unlock(); | |
4716 | ||
4717 | if (err) { | |
4718 | WARN_ON(&ptype->list == head); | |
4719 | kfree_skb(skb); | |
4720 | return NET_RX_SUCCESS; | |
4721 | } | |
4722 | ||
4723 | out: | |
ae78dbfa | 4724 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
4725 | } |
4726 | ||
2e71a6f8 ED |
4727 | /* napi->gro_list contains packets ordered by age. |
4728 | * youngest packets at the head of it. | |
4729 | * Complete skbs in reverse order to reduce latencies. | |
4730 | */ | |
4731 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 4732 | { |
2e71a6f8 | 4733 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 4734 | |
2e71a6f8 ED |
4735 | /* scan list and build reverse chain */ |
4736 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
4737 | skb->prev = prev; | |
4738 | prev = skb; | |
4739 | } | |
4740 | ||
4741 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 4742 | skb->next = NULL; |
2e71a6f8 ED |
4743 | |
4744 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
4745 | return; | |
4746 | ||
4747 | prev = skb->prev; | |
d565b0a1 | 4748 | napi_gro_complete(skb); |
2e71a6f8 | 4749 | napi->gro_count--; |
d565b0a1 HX |
4750 | } |
4751 | ||
4752 | napi->gro_list = NULL; | |
4753 | } | |
86cac58b | 4754 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 4755 | |
89c5fa33 ED |
4756 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
4757 | { | |
4758 | struct sk_buff *p; | |
4759 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 4760 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
4761 | |
4762 | for (p = napi->gro_list; p; p = p->next) { | |
4763 | unsigned long diffs; | |
4764 | ||
0b4cec8c TH |
4765 | NAPI_GRO_CB(p)->flush = 0; |
4766 | ||
4767 | if (hash != skb_get_hash_raw(p)) { | |
4768 | NAPI_GRO_CB(p)->same_flow = 0; | |
4769 | continue; | |
4770 | } | |
4771 | ||
89c5fa33 ED |
4772 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
4773 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
ce87fc6c | 4774 | diffs |= skb_metadata_dst_cmp(p, skb); |
de8f3a83 | 4775 | diffs |= skb_metadata_differs(p, skb); |
89c5fa33 ED |
4776 | if (maclen == ETH_HLEN) |
4777 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 4778 | skb_mac_header(skb)); |
89c5fa33 ED |
4779 | else if (!diffs) |
4780 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 4781 | skb_mac_header(skb), |
89c5fa33 ED |
4782 | maclen); |
4783 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
4784 | } |
4785 | } | |
4786 | ||
299603e8 JC |
4787 | static void skb_gro_reset_offset(struct sk_buff *skb) |
4788 | { | |
4789 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4790 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
4791 | ||
4792 | NAPI_GRO_CB(skb)->data_offset = 0; | |
4793 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
4794 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
4795 | ||
4796 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
4797 | pinfo->nr_frags && | |
4798 | !PageHighMem(skb_frag_page(frag0))) { | |
4799 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
7cfd5fd5 ED |
4800 | NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, |
4801 | skb_frag_size(frag0), | |
4802 | skb->end - skb->tail); | |
89c5fa33 ED |
4803 | } |
4804 | } | |
4805 | ||
a50e233c ED |
4806 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
4807 | { | |
4808 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4809 | ||
4810 | BUG_ON(skb->end - skb->tail < grow); | |
4811 | ||
4812 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
4813 | ||
4814 | skb->data_len -= grow; | |
4815 | skb->tail += grow; | |
4816 | ||
4817 | pinfo->frags[0].page_offset += grow; | |
4818 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
4819 | ||
4820 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
4821 | skb_frag_unref(skb, 0); | |
4822 | memmove(pinfo->frags, pinfo->frags + 1, | |
4823 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
4824 | } | |
4825 | } | |
4826 | ||
bb728820 | 4827 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
4828 | { |
4829 | struct sk_buff **pp = NULL; | |
22061d80 | 4830 | struct packet_offload *ptype; |
d565b0a1 | 4831 | __be16 type = skb->protocol; |
22061d80 | 4832 | struct list_head *head = &offload_base; |
0da2afd5 | 4833 | int same_flow; |
5b252f0c | 4834 | enum gro_result ret; |
a50e233c | 4835 | int grow; |
d565b0a1 | 4836 | |
b5cdae32 | 4837 | if (netif_elide_gro(skb->dev)) |
d565b0a1 HX |
4838 | goto normal; |
4839 | ||
89c5fa33 ED |
4840 | gro_list_prepare(napi, skb); |
4841 | ||
d565b0a1 HX |
4842 | rcu_read_lock(); |
4843 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4844 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4845 | continue; |
4846 | ||
86911732 | 4847 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4848 | skb_reset_mac_len(skb); |
d565b0a1 | 4849 | NAPI_GRO_CB(skb)->same_flow = 0; |
d61d072e | 4850 | NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); |
5d38a079 | 4851 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 4852 | NAPI_GRO_CB(skb)->encap_mark = 0; |
fcd91dd4 | 4853 | NAPI_GRO_CB(skb)->recursion_counter = 0; |
a0ca153f | 4854 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 4855 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 4856 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 4857 | |
662880f4 TH |
4858 | /* Setup for GRO checksum validation */ |
4859 | switch (skb->ip_summed) { | |
4860 | case CHECKSUM_COMPLETE: | |
4861 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4862 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4863 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4864 | break; | |
4865 | case CHECKSUM_UNNECESSARY: | |
4866 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4867 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4868 | break; | |
4869 | default: | |
4870 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4871 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4872 | } | |
d565b0a1 | 4873 | |
f191a1d1 | 4874 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4875 | break; |
4876 | } | |
4877 | rcu_read_unlock(); | |
4878 | ||
4879 | if (&ptype->list == head) | |
4880 | goto normal; | |
4881 | ||
25393d3f SK |
4882 | if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) { |
4883 | ret = GRO_CONSUMED; | |
4884 | goto ok; | |
4885 | } | |
4886 | ||
0da2afd5 | 4887 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4888 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4889 | |
d565b0a1 HX |
4890 | if (pp) { |
4891 | struct sk_buff *nskb = *pp; | |
4892 | ||
4893 | *pp = nskb->next; | |
4894 | nskb->next = NULL; | |
4895 | napi_gro_complete(nskb); | |
4ae5544f | 4896 | napi->gro_count--; |
d565b0a1 HX |
4897 | } |
4898 | ||
0da2afd5 | 4899 | if (same_flow) |
d565b0a1 HX |
4900 | goto ok; |
4901 | ||
600adc18 | 4902 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4903 | goto normal; |
d565b0a1 | 4904 | |
600adc18 ED |
4905 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4906 | struct sk_buff *nskb = napi->gro_list; | |
4907 | ||
4908 | /* locate the end of the list to select the 'oldest' flow */ | |
4909 | while (nskb->next) { | |
4910 | pp = &nskb->next; | |
4911 | nskb = *pp; | |
4912 | } | |
4913 | *pp = NULL; | |
4914 | nskb->next = NULL; | |
4915 | napi_gro_complete(nskb); | |
4916 | } else { | |
4917 | napi->gro_count++; | |
4918 | } | |
d565b0a1 | 4919 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4920 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4921 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4922 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4923 | skb->next = napi->gro_list; |
4924 | napi->gro_list = skb; | |
5d0d9be8 | 4925 | ret = GRO_HELD; |
d565b0a1 | 4926 | |
ad0f9904 | 4927 | pull: |
a50e233c ED |
4928 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4929 | if (grow > 0) | |
4930 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4931 | ok: |
5d0d9be8 | 4932 | return ret; |
d565b0a1 HX |
4933 | |
4934 | normal: | |
ad0f9904 HX |
4935 | ret = GRO_NORMAL; |
4936 | goto pull; | |
5d38a079 | 4937 | } |
96e93eab | 4938 | |
bf5a755f JC |
4939 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4940 | { | |
4941 | struct list_head *offload_head = &offload_base; | |
4942 | struct packet_offload *ptype; | |
4943 | ||
4944 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4945 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4946 | continue; | |
4947 | return ptype; | |
4948 | } | |
4949 | return NULL; | |
4950 | } | |
e27a2f83 | 4951 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4952 | |
4953 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4954 | { | |
4955 | struct list_head *offload_head = &offload_base; | |
4956 | struct packet_offload *ptype; | |
4957 | ||
4958 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4959 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4960 | continue; | |
4961 | return ptype; | |
4962 | } | |
4963 | return NULL; | |
4964 | } | |
e27a2f83 | 4965 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4966 | |
e44699d2 MK |
4967 | static void napi_skb_free_stolen_head(struct sk_buff *skb) |
4968 | { | |
4969 | skb_dst_drop(skb); | |
4970 | secpath_reset(skb); | |
4971 | kmem_cache_free(skbuff_head_cache, skb); | |
4972 | } | |
4973 | ||
bb728820 | 4974 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4975 | { |
5d0d9be8 HX |
4976 | switch (ret) { |
4977 | case GRO_NORMAL: | |
ae78dbfa | 4978 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4979 | ret = GRO_DROP; |
4980 | break; | |
5d38a079 | 4981 | |
5d0d9be8 | 4982 | case GRO_DROP: |
5d38a079 HX |
4983 | kfree_skb(skb); |
4984 | break; | |
5b252f0c | 4985 | |
daa86548 | 4986 | case GRO_MERGED_FREE: |
e44699d2 MK |
4987 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
4988 | napi_skb_free_stolen_head(skb); | |
4989 | else | |
d7e8883c | 4990 | __kfree_skb(skb); |
daa86548 ED |
4991 | break; |
4992 | ||
5b252f0c BH |
4993 | case GRO_HELD: |
4994 | case GRO_MERGED: | |
25393d3f | 4995 | case GRO_CONSUMED: |
5b252f0c | 4996 | break; |
5d38a079 HX |
4997 | } |
4998 | ||
c7c4b3b6 | 4999 | return ret; |
5d0d9be8 | 5000 | } |
5d0d9be8 | 5001 | |
c7c4b3b6 | 5002 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 5003 | { |
93f93a44 | 5004 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 5005 | trace_napi_gro_receive_entry(skb); |
86911732 | 5006 | |
a50e233c ED |
5007 | skb_gro_reset_offset(skb); |
5008 | ||
89c5fa33 | 5009 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
5010 | } |
5011 | EXPORT_SYMBOL(napi_gro_receive); | |
5012 | ||
d0c2b0d2 | 5013 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 5014 | { |
93a35f59 ED |
5015 | if (unlikely(skb->pfmemalloc)) { |
5016 | consume_skb(skb); | |
5017 | return; | |
5018 | } | |
96e93eab | 5019 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
5020 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
5021 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 5022 | skb->vlan_tci = 0; |
66c46d74 | 5023 | skb->dev = napi->dev; |
6d152e23 | 5024 | skb->skb_iif = 0; |
c3caf119 JC |
5025 | skb->encapsulation = 0; |
5026 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 5027 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
f991bb9d | 5028 | secpath_reset(skb); |
96e93eab HX |
5029 | |
5030 | napi->skb = skb; | |
5031 | } | |
96e93eab | 5032 | |
76620aaf | 5033 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 5034 | { |
5d38a079 | 5035 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
5036 | |
5037 | if (!skb) { | |
fd11a83d | 5038 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
5039 | if (skb) { |
5040 | napi->skb = skb; | |
5041 | skb_mark_napi_id(skb, napi); | |
5042 | } | |
80595d59 | 5043 | } |
96e93eab HX |
5044 | return skb; |
5045 | } | |
76620aaf | 5046 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 5047 | |
a50e233c ED |
5048 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
5049 | struct sk_buff *skb, | |
5050 | gro_result_t ret) | |
96e93eab | 5051 | { |
5d0d9be8 HX |
5052 | switch (ret) { |
5053 | case GRO_NORMAL: | |
a50e233c ED |
5054 | case GRO_HELD: |
5055 | __skb_push(skb, ETH_HLEN); | |
5056 | skb->protocol = eth_type_trans(skb, skb->dev); | |
5057 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 5058 | ret = GRO_DROP; |
86911732 | 5059 | break; |
5d38a079 | 5060 | |
5d0d9be8 | 5061 | case GRO_DROP: |
5d0d9be8 HX |
5062 | napi_reuse_skb(napi, skb); |
5063 | break; | |
5b252f0c | 5064 | |
e44699d2 MK |
5065 | case GRO_MERGED_FREE: |
5066 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) | |
5067 | napi_skb_free_stolen_head(skb); | |
5068 | else | |
5069 | napi_reuse_skb(napi, skb); | |
5070 | break; | |
5071 | ||
5b252f0c | 5072 | case GRO_MERGED: |
25393d3f | 5073 | case GRO_CONSUMED: |
5b252f0c | 5074 | break; |
5d0d9be8 | 5075 | } |
5d38a079 | 5076 | |
c7c4b3b6 | 5077 | return ret; |
5d38a079 | 5078 | } |
5d0d9be8 | 5079 | |
a50e233c ED |
5080 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
5081 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
5082 | * We copy ethernet header into skb->data to have a common layout. | |
5083 | */ | |
4adb9c4a | 5084 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
5085 | { |
5086 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
5087 | const struct ethhdr *eth; |
5088 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
5089 | |
5090 | napi->skb = NULL; | |
5091 | ||
a50e233c ED |
5092 | skb_reset_mac_header(skb); |
5093 | skb_gro_reset_offset(skb); | |
5094 | ||
5095 | eth = skb_gro_header_fast(skb, 0); | |
5096 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
5097 | eth = skb_gro_header_slow(skb, hlen, 0); | |
5098 | if (unlikely(!eth)) { | |
4da46ceb AC |
5099 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
5100 | __func__, napi->dev->name); | |
a50e233c ED |
5101 | napi_reuse_skb(napi, skb); |
5102 | return NULL; | |
5103 | } | |
5104 | } else { | |
5105 | gro_pull_from_frag0(skb, hlen); | |
5106 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
5107 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 5108 | } |
a50e233c ED |
5109 | __skb_pull(skb, hlen); |
5110 | ||
5111 | /* | |
5112 | * This works because the only protocols we care about don't require | |
5113 | * special handling. | |
5114 | * We'll fix it up properly in napi_frags_finish() | |
5115 | */ | |
5116 | skb->protocol = eth->h_proto; | |
76620aaf | 5117 | |
76620aaf HX |
5118 | return skb; |
5119 | } | |
76620aaf | 5120 | |
c7c4b3b6 | 5121 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 5122 | { |
76620aaf | 5123 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
5124 | |
5125 | if (!skb) | |
c7c4b3b6 | 5126 | return GRO_DROP; |
5d0d9be8 | 5127 | |
ae78dbfa BH |
5128 | trace_napi_gro_frags_entry(skb); |
5129 | ||
89c5fa33 | 5130 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 5131 | } |
5d38a079 HX |
5132 | EXPORT_SYMBOL(napi_gro_frags); |
5133 | ||
573e8fca TH |
5134 | /* Compute the checksum from gro_offset and return the folded value |
5135 | * after adding in any pseudo checksum. | |
5136 | */ | |
5137 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
5138 | { | |
5139 | __wsum wsum; | |
5140 | __sum16 sum; | |
5141 | ||
5142 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
5143 | ||
5144 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
5145 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
5146 | if (likely(!sum)) { | |
5147 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
5148 | !skb->csum_complete_sw) | |
5149 | netdev_rx_csum_fault(skb->dev); | |
5150 | } | |
5151 | ||
5152 | NAPI_GRO_CB(skb)->csum = wsum; | |
5153 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
5154 | ||
5155 | return sum; | |
5156 | } | |
5157 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
5158 | ||
773fc8f6 | 5159 | static void net_rps_send_ipi(struct softnet_data *remsd) |
5160 | { | |
5161 | #ifdef CONFIG_RPS | |
5162 | while (remsd) { | |
5163 | struct softnet_data *next = remsd->rps_ipi_next; | |
5164 | ||
5165 | if (cpu_online(remsd->cpu)) | |
5166 | smp_call_function_single_async(remsd->cpu, &remsd->csd); | |
5167 | remsd = next; | |
5168 | } | |
5169 | #endif | |
5170 | } | |
5171 | ||
e326bed2 | 5172 | /* |
855abcf0 | 5173 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
5174 | * Note: called with local irq disabled, but exits with local irq enabled. |
5175 | */ | |
5176 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
5177 | { | |
5178 | #ifdef CONFIG_RPS | |
5179 | struct softnet_data *remsd = sd->rps_ipi_list; | |
5180 | ||
5181 | if (remsd) { | |
5182 | sd->rps_ipi_list = NULL; | |
5183 | ||
5184 | local_irq_enable(); | |
5185 | ||
5186 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
773fc8f6 | 5187 | net_rps_send_ipi(remsd); |
e326bed2 ED |
5188 | } else |
5189 | #endif | |
5190 | local_irq_enable(); | |
5191 | } | |
5192 | ||
d75b1ade ED |
5193 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
5194 | { | |
5195 | #ifdef CONFIG_RPS | |
5196 | return sd->rps_ipi_list != NULL; | |
5197 | #else | |
5198 | return false; | |
5199 | #endif | |
5200 | } | |
5201 | ||
bea3348e | 5202 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 5203 | { |
eecfd7c4 | 5204 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
5205 | bool again = true; |
5206 | int work = 0; | |
1da177e4 | 5207 | |
e326bed2 ED |
5208 | /* Check if we have pending ipi, its better to send them now, |
5209 | * not waiting net_rx_action() end. | |
5210 | */ | |
d75b1ade | 5211 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
5212 | local_irq_disable(); |
5213 | net_rps_action_and_irq_enable(sd); | |
5214 | } | |
d75b1ade | 5215 | |
3d48b53f | 5216 | napi->weight = dev_rx_weight; |
145dd5f9 | 5217 | while (again) { |
1da177e4 | 5218 | struct sk_buff *skb; |
6e7676c1 CG |
5219 | |
5220 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 5221 | rcu_read_lock(); |
6e7676c1 | 5222 | __netif_receive_skb(skb); |
2c17d27c | 5223 | rcu_read_unlock(); |
76cc8b13 | 5224 | input_queue_head_incr(sd); |
145dd5f9 | 5225 | if (++work >= quota) |
76cc8b13 | 5226 | return work; |
145dd5f9 | 5227 | |
6e7676c1 | 5228 | } |
1da177e4 | 5229 | |
145dd5f9 | 5230 | local_irq_disable(); |
e36fa2f7 | 5231 | rps_lock(sd); |
11ef7a89 | 5232 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
5233 | /* |
5234 | * Inline a custom version of __napi_complete(). | |
5235 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
5236 | * and NAPI_STATE_SCHED is the only possible flag set |
5237 | * on backlog. | |
5238 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
5239 | * and we dont need an smp_mb() memory barrier. |
5240 | */ | |
eecfd7c4 | 5241 | napi->state = 0; |
145dd5f9 PA |
5242 | again = false; |
5243 | } else { | |
5244 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
5245 | &sd->process_queue); | |
bea3348e | 5246 | } |
e36fa2f7 | 5247 | rps_unlock(sd); |
145dd5f9 | 5248 | local_irq_enable(); |
6e7676c1 | 5249 | } |
1da177e4 | 5250 | |
bea3348e SH |
5251 | return work; |
5252 | } | |
1da177e4 | 5253 | |
bea3348e SH |
5254 | /** |
5255 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 5256 | * @n: entry to schedule |
bea3348e | 5257 | * |
bc9ad166 ED |
5258 | * The entry's receive function will be scheduled to run. |
5259 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 5260 | */ |
b5606c2d | 5261 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
5262 | { |
5263 | unsigned long flags; | |
1da177e4 | 5264 | |
bea3348e | 5265 | local_irq_save(flags); |
903ceff7 | 5266 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 5267 | local_irq_restore(flags); |
1da177e4 | 5268 | } |
bea3348e SH |
5269 | EXPORT_SYMBOL(__napi_schedule); |
5270 | ||
39e6c820 ED |
5271 | /** |
5272 | * napi_schedule_prep - check if napi can be scheduled | |
5273 | * @n: napi context | |
5274 | * | |
5275 | * Test if NAPI routine is already running, and if not mark | |
5276 | * it as running. This is used as a condition variable | |
5277 | * insure only one NAPI poll instance runs. We also make | |
5278 | * sure there is no pending NAPI disable. | |
5279 | */ | |
5280 | bool napi_schedule_prep(struct napi_struct *n) | |
5281 | { | |
5282 | unsigned long val, new; | |
5283 | ||
5284 | do { | |
5285 | val = READ_ONCE(n->state); | |
5286 | if (unlikely(val & NAPIF_STATE_DISABLE)) | |
5287 | return false; | |
5288 | new = val | NAPIF_STATE_SCHED; | |
5289 | ||
5290 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
5291 | * This was suggested by Alexander Duyck, as compiler | |
5292 | * emits better code than : | |
5293 | * if (val & NAPIF_STATE_SCHED) | |
5294 | * new |= NAPIF_STATE_MISSED; | |
5295 | */ | |
5296 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
5297 | NAPIF_STATE_MISSED; | |
5298 | } while (cmpxchg(&n->state, val, new) != val); | |
5299 | ||
5300 | return !(val & NAPIF_STATE_SCHED); | |
5301 | } | |
5302 | EXPORT_SYMBOL(napi_schedule_prep); | |
5303 | ||
bc9ad166 ED |
5304 | /** |
5305 | * __napi_schedule_irqoff - schedule for receive | |
5306 | * @n: entry to schedule | |
5307 | * | |
5308 | * Variant of __napi_schedule() assuming hard irqs are masked | |
5309 | */ | |
5310 | void __napi_schedule_irqoff(struct napi_struct *n) | |
5311 | { | |
5312 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
5313 | } | |
5314 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
5315 | ||
364b6055 | 5316 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 5317 | { |
39e6c820 | 5318 | unsigned long flags, val, new; |
d565b0a1 HX |
5319 | |
5320 | /* | |
217f6974 ED |
5321 | * 1) Don't let napi dequeue from the cpu poll list |
5322 | * just in case its running on a different cpu. | |
5323 | * 2) If we are busy polling, do nothing here, we have | |
5324 | * the guarantee we will be called later. | |
d565b0a1 | 5325 | */ |
217f6974 ED |
5326 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
5327 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 5328 | return false; |
d565b0a1 | 5329 | |
3b47d303 ED |
5330 | if (n->gro_list) { |
5331 | unsigned long timeout = 0; | |
d75b1ade | 5332 | |
3b47d303 ED |
5333 | if (work_done) |
5334 | timeout = n->dev->gro_flush_timeout; | |
5335 | ||
5336 | if (timeout) | |
5337 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
5338 | HRTIMER_MODE_REL_PINNED); | |
5339 | else | |
5340 | napi_gro_flush(n, false); | |
5341 | } | |
02c1602e | 5342 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
5343 | /* If n->poll_list is not empty, we need to mask irqs */ |
5344 | local_irq_save(flags); | |
02c1602e | 5345 | list_del_init(&n->poll_list); |
d75b1ade ED |
5346 | local_irq_restore(flags); |
5347 | } | |
39e6c820 ED |
5348 | |
5349 | do { | |
5350 | val = READ_ONCE(n->state); | |
5351 | ||
5352 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); | |
5353 | ||
5354 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); | |
5355 | ||
5356 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
5357 | * because we will call napi->poll() one more time. | |
5358 | * This C code was suggested by Alexander Duyck to help gcc. | |
5359 | */ | |
5360 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
5361 | NAPIF_STATE_SCHED; | |
5362 | } while (cmpxchg(&n->state, val, new) != val); | |
5363 | ||
5364 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
5365 | __napi_schedule(n); | |
5366 | return false; | |
5367 | } | |
5368 | ||
364b6055 | 5369 | return true; |
d565b0a1 | 5370 | } |
3b47d303 | 5371 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 5372 | |
af12fa6e | 5373 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 5374 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
5375 | { |
5376 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
5377 | struct napi_struct *napi; | |
5378 | ||
5379 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
5380 | if (napi->napi_id == napi_id) | |
5381 | return napi; | |
5382 | ||
5383 | return NULL; | |
5384 | } | |
02d62e86 ED |
5385 | |
5386 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
217f6974 | 5387 | |
ce6aea93 | 5388 | #define BUSY_POLL_BUDGET 8 |
217f6974 ED |
5389 | |
5390 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) | |
5391 | { | |
5392 | int rc; | |
5393 | ||
39e6c820 ED |
5394 | /* Busy polling means there is a high chance device driver hard irq |
5395 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
5396 | * set in napi_schedule_prep(). | |
5397 | * Since we are about to call napi->poll() once more, we can safely | |
5398 | * clear NAPI_STATE_MISSED. | |
5399 | * | |
5400 | * Note: x86 could use a single "lock and ..." instruction | |
5401 | * to perform these two clear_bit() | |
5402 | */ | |
5403 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
5404 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
5405 | ||
5406 | local_bh_disable(); | |
5407 | ||
5408 | /* All we really want here is to re-enable device interrupts. | |
5409 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
5410 | */ | |
5411 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
1e22391e | 5412 | trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); |
217f6974 ED |
5413 | netpoll_poll_unlock(have_poll_lock); |
5414 | if (rc == BUSY_POLL_BUDGET) | |
5415 | __napi_schedule(napi); | |
5416 | local_bh_enable(); | |
217f6974 ED |
5417 | } |
5418 | ||
7db6b048 SS |
5419 | void napi_busy_loop(unsigned int napi_id, |
5420 | bool (*loop_end)(void *, unsigned long), | |
5421 | void *loop_end_arg) | |
02d62e86 | 5422 | { |
7db6b048 | 5423 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 5424 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 5425 | void *have_poll_lock = NULL; |
02d62e86 | 5426 | struct napi_struct *napi; |
217f6974 ED |
5427 | |
5428 | restart: | |
217f6974 | 5429 | napi_poll = NULL; |
02d62e86 | 5430 | |
2a028ecb | 5431 | rcu_read_lock(); |
02d62e86 | 5432 | |
545cd5e5 | 5433 | napi = napi_by_id(napi_id); |
02d62e86 ED |
5434 | if (!napi) |
5435 | goto out; | |
5436 | ||
217f6974 ED |
5437 | preempt_disable(); |
5438 | for (;;) { | |
2b5cd0df AD |
5439 | int work = 0; |
5440 | ||
2a028ecb | 5441 | local_bh_disable(); |
217f6974 ED |
5442 | if (!napi_poll) { |
5443 | unsigned long val = READ_ONCE(napi->state); | |
5444 | ||
5445 | /* If multiple threads are competing for this napi, | |
5446 | * we avoid dirtying napi->state as much as we can. | |
5447 | */ | |
5448 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
5449 | NAPIF_STATE_IN_BUSY_POLL)) | |
5450 | goto count; | |
5451 | if (cmpxchg(&napi->state, val, | |
5452 | val | NAPIF_STATE_IN_BUSY_POLL | | |
5453 | NAPIF_STATE_SCHED) != val) | |
5454 | goto count; | |
5455 | have_poll_lock = netpoll_poll_lock(napi); | |
5456 | napi_poll = napi->poll; | |
5457 | } | |
2b5cd0df AD |
5458 | work = napi_poll(napi, BUSY_POLL_BUDGET); |
5459 | trace_napi_poll(napi, work, BUSY_POLL_BUDGET); | |
217f6974 | 5460 | count: |
2b5cd0df | 5461 | if (work > 0) |
7db6b048 | 5462 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 5463 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
2a028ecb | 5464 | local_bh_enable(); |
02d62e86 | 5465 | |
7db6b048 | 5466 | if (!loop_end || loop_end(loop_end_arg, start_time)) |
217f6974 | 5467 | break; |
02d62e86 | 5468 | |
217f6974 ED |
5469 | if (unlikely(need_resched())) { |
5470 | if (napi_poll) | |
5471 | busy_poll_stop(napi, have_poll_lock); | |
5472 | preempt_enable(); | |
5473 | rcu_read_unlock(); | |
5474 | cond_resched(); | |
7db6b048 | 5475 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 5476 | return; |
217f6974 ED |
5477 | goto restart; |
5478 | } | |
6cdf89b1 | 5479 | cpu_relax(); |
217f6974 ED |
5480 | } |
5481 | if (napi_poll) | |
5482 | busy_poll_stop(napi, have_poll_lock); | |
5483 | preempt_enable(); | |
02d62e86 | 5484 | out: |
2a028ecb | 5485 | rcu_read_unlock(); |
02d62e86 | 5486 | } |
7db6b048 | 5487 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
5488 | |
5489 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 5490 | |
149d6ad8 | 5491 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 5492 | { |
d64b5e85 ED |
5493 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
5494 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 5495 | return; |
af12fa6e | 5496 | |
52bd2d62 | 5497 | spin_lock(&napi_hash_lock); |
af12fa6e | 5498 | |
545cd5e5 | 5499 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 5500 | do { |
545cd5e5 AD |
5501 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
5502 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
5503 | } while (napi_by_id(napi_gen_id)); |
5504 | napi->napi_id = napi_gen_id; | |
af12fa6e | 5505 | |
52bd2d62 ED |
5506 | hlist_add_head_rcu(&napi->napi_hash_node, |
5507 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 5508 | |
52bd2d62 | 5509 | spin_unlock(&napi_hash_lock); |
af12fa6e | 5510 | } |
af12fa6e ET |
5511 | |
5512 | /* Warning : caller is responsible to make sure rcu grace period | |
5513 | * is respected before freeing memory containing @napi | |
5514 | */ | |
34cbe27e | 5515 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 5516 | { |
34cbe27e ED |
5517 | bool rcu_sync_needed = false; |
5518 | ||
af12fa6e ET |
5519 | spin_lock(&napi_hash_lock); |
5520 | ||
34cbe27e ED |
5521 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
5522 | rcu_sync_needed = true; | |
af12fa6e | 5523 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 5524 | } |
af12fa6e | 5525 | spin_unlock(&napi_hash_lock); |
34cbe27e | 5526 | return rcu_sync_needed; |
af12fa6e ET |
5527 | } |
5528 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
5529 | ||
3b47d303 ED |
5530 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
5531 | { | |
5532 | struct napi_struct *napi; | |
5533 | ||
5534 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
5535 | |
5536 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
5537 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
5538 | */ | |
5539 | if (napi->gro_list && !napi_disable_pending(napi) && | |
5540 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) | |
5541 | __napi_schedule_irqoff(napi); | |
3b47d303 ED |
5542 | |
5543 | return HRTIMER_NORESTART; | |
5544 | } | |
5545 | ||
d565b0a1 HX |
5546 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
5547 | int (*poll)(struct napi_struct *, int), int weight) | |
5548 | { | |
5549 | INIT_LIST_HEAD(&napi->poll_list); | |
3b47d303 ED |
5550 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
5551 | napi->timer.function = napi_watchdog; | |
4ae5544f | 5552 | napi->gro_count = 0; |
d565b0a1 | 5553 | napi->gro_list = NULL; |
5d38a079 | 5554 | napi->skb = NULL; |
d565b0a1 | 5555 | napi->poll = poll; |
82dc3c63 ED |
5556 | if (weight > NAPI_POLL_WEIGHT) |
5557 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
5558 | weight, dev->name); | |
d565b0a1 HX |
5559 | napi->weight = weight; |
5560 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 5561 | napi->dev = dev; |
5d38a079 | 5562 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
5563 | napi->poll_owner = -1; |
5564 | #endif | |
5565 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 5566 | napi_hash_add(napi); |
d565b0a1 HX |
5567 | } |
5568 | EXPORT_SYMBOL(netif_napi_add); | |
5569 | ||
3b47d303 ED |
5570 | void napi_disable(struct napi_struct *n) |
5571 | { | |
5572 | might_sleep(); | |
5573 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
5574 | ||
5575 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
5576 | msleep(1); | |
2d8bff12 NH |
5577 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
5578 | msleep(1); | |
3b47d303 ED |
5579 | |
5580 | hrtimer_cancel(&n->timer); | |
5581 | ||
5582 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
5583 | } | |
5584 | EXPORT_SYMBOL(napi_disable); | |
5585 | ||
93d05d4a | 5586 | /* Must be called in process context */ |
d565b0a1 HX |
5587 | void netif_napi_del(struct napi_struct *napi) |
5588 | { | |
93d05d4a ED |
5589 | might_sleep(); |
5590 | if (napi_hash_del(napi)) | |
5591 | synchronize_net(); | |
d7b06636 | 5592 | list_del_init(&napi->dev_list); |
76620aaf | 5593 | napi_free_frags(napi); |
d565b0a1 | 5594 | |
289dccbe | 5595 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 5596 | napi->gro_list = NULL; |
4ae5544f | 5597 | napi->gro_count = 0; |
d565b0a1 HX |
5598 | } |
5599 | EXPORT_SYMBOL(netif_napi_del); | |
5600 | ||
726ce70e HX |
5601 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
5602 | { | |
5603 | void *have; | |
5604 | int work, weight; | |
5605 | ||
5606 | list_del_init(&n->poll_list); | |
5607 | ||
5608 | have = netpoll_poll_lock(n); | |
5609 | ||
5610 | weight = n->weight; | |
5611 | ||
5612 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
5613 | * with netpoll's poll_napi(). Only the entity which | |
5614 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
5615 | * actually make the ->poll() call. Therefore we avoid | |
5616 | * accidentally calling ->poll() when NAPI is not scheduled. | |
5617 | */ | |
5618 | work = 0; | |
5619 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
5620 | work = n->poll(n, weight); | |
1db19db7 | 5621 | trace_napi_poll(n, work, weight); |
726ce70e HX |
5622 | } |
5623 | ||
5624 | WARN_ON_ONCE(work > weight); | |
5625 | ||
5626 | if (likely(work < weight)) | |
5627 | goto out_unlock; | |
5628 | ||
5629 | /* Drivers must not modify the NAPI state if they | |
5630 | * consume the entire weight. In such cases this code | |
5631 | * still "owns" the NAPI instance and therefore can | |
5632 | * move the instance around on the list at-will. | |
5633 | */ | |
5634 | if (unlikely(napi_disable_pending(n))) { | |
5635 | napi_complete(n); | |
5636 | goto out_unlock; | |
5637 | } | |
5638 | ||
5639 | if (n->gro_list) { | |
5640 | /* flush too old packets | |
5641 | * If HZ < 1000, flush all packets. | |
5642 | */ | |
5643 | napi_gro_flush(n, HZ >= 1000); | |
5644 | } | |
5645 | ||
001ce546 HX |
5646 | /* Some drivers may have called napi_schedule |
5647 | * prior to exhausting their budget. | |
5648 | */ | |
5649 | if (unlikely(!list_empty(&n->poll_list))) { | |
5650 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
5651 | n->dev ? n->dev->name : "backlog"); | |
5652 | goto out_unlock; | |
5653 | } | |
5654 | ||
726ce70e HX |
5655 | list_add_tail(&n->poll_list, repoll); |
5656 | ||
5657 | out_unlock: | |
5658 | netpoll_poll_unlock(have); | |
5659 | ||
5660 | return work; | |
5661 | } | |
5662 | ||
0766f788 | 5663 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 5664 | { |
903ceff7 | 5665 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e MW |
5666 | unsigned long time_limit = jiffies + |
5667 | usecs_to_jiffies(netdev_budget_usecs); | |
51b0bded | 5668 | int budget = netdev_budget; |
d75b1ade ED |
5669 | LIST_HEAD(list); |
5670 | LIST_HEAD(repoll); | |
53fb95d3 | 5671 | |
1da177e4 | 5672 | local_irq_disable(); |
d75b1ade ED |
5673 | list_splice_init(&sd->poll_list, &list); |
5674 | local_irq_enable(); | |
1da177e4 | 5675 | |
ceb8d5bf | 5676 | for (;;) { |
bea3348e | 5677 | struct napi_struct *n; |
1da177e4 | 5678 | |
ceb8d5bf HX |
5679 | if (list_empty(&list)) { |
5680 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
f52dffe0 | 5681 | goto out; |
ceb8d5bf HX |
5682 | break; |
5683 | } | |
5684 | ||
6bd373eb HX |
5685 | n = list_first_entry(&list, struct napi_struct, poll_list); |
5686 | budget -= napi_poll(n, &repoll); | |
5687 | ||
d75b1ade | 5688 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
5689 | * Allow this to run for 2 jiffies since which will allow |
5690 | * an average latency of 1.5/HZ. | |
bea3348e | 5691 | */ |
ceb8d5bf HX |
5692 | if (unlikely(budget <= 0 || |
5693 | time_after_eq(jiffies, time_limit))) { | |
5694 | sd->time_squeeze++; | |
5695 | break; | |
5696 | } | |
1da177e4 | 5697 | } |
d75b1ade | 5698 | |
d75b1ade ED |
5699 | local_irq_disable(); |
5700 | ||
5701 | list_splice_tail_init(&sd->poll_list, &list); | |
5702 | list_splice_tail(&repoll, &list); | |
5703 | list_splice(&list, &sd->poll_list); | |
5704 | if (!list_empty(&sd->poll_list)) | |
5705 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
5706 | ||
e326bed2 | 5707 | net_rps_action_and_irq_enable(sd); |
f52dffe0 ED |
5708 | out: |
5709 | __kfree_skb_flush(); | |
1da177e4 LT |
5710 | } |
5711 | ||
aa9d8560 | 5712 | struct netdev_adjacent { |
9ff162a8 | 5713 | struct net_device *dev; |
5d261913 VF |
5714 | |
5715 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 5716 | bool master; |
5d261913 | 5717 | |
5d261913 VF |
5718 | /* counter for the number of times this device was added to us */ |
5719 | u16 ref_nr; | |
5720 | ||
402dae96 VF |
5721 | /* private field for the users */ |
5722 | void *private; | |
5723 | ||
9ff162a8 JP |
5724 | struct list_head list; |
5725 | struct rcu_head rcu; | |
9ff162a8 JP |
5726 | }; |
5727 | ||
6ea29da1 | 5728 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 5729 | struct list_head *adj_list) |
9ff162a8 | 5730 | { |
5d261913 | 5731 | struct netdev_adjacent *adj; |
5d261913 | 5732 | |
2f268f12 | 5733 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
5734 | if (adj->dev == adj_dev) |
5735 | return adj; | |
9ff162a8 JP |
5736 | } |
5737 | return NULL; | |
5738 | } | |
5739 | ||
f1170fd4 DA |
5740 | static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data) |
5741 | { | |
5742 | struct net_device *dev = data; | |
5743 | ||
5744 | return upper_dev == dev; | |
5745 | } | |
5746 | ||
9ff162a8 JP |
5747 | /** |
5748 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
5749 | * @dev: device | |
5750 | * @upper_dev: upper device to check | |
5751 | * | |
5752 | * Find out if a device is linked to specified upper device and return true | |
5753 | * in case it is. Note that this checks only immediate upper device, | |
5754 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
5755 | */ | |
5756 | bool netdev_has_upper_dev(struct net_device *dev, | |
5757 | struct net_device *upper_dev) | |
5758 | { | |
5759 | ASSERT_RTNL(); | |
5760 | ||
f1170fd4 DA |
5761 | return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, |
5762 | upper_dev); | |
9ff162a8 JP |
5763 | } |
5764 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
5765 | ||
1a3f060c DA |
5766 | /** |
5767 | * netdev_has_upper_dev_all - Check if device is linked to an upper device | |
5768 | * @dev: device | |
5769 | * @upper_dev: upper device to check | |
5770 | * | |
5771 | * Find out if a device is linked to specified upper device and return true | |
5772 | * in case it is. Note that this checks the entire upper device chain. | |
5773 | * The caller must hold rcu lock. | |
5774 | */ | |
5775 | ||
1a3f060c DA |
5776 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
5777 | struct net_device *upper_dev) | |
5778 | { | |
5779 | return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, | |
5780 | upper_dev); | |
5781 | } | |
5782 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
5783 | ||
9ff162a8 JP |
5784 | /** |
5785 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
5786 | * @dev: device | |
5787 | * | |
5788 | * Find out if a device is linked to an upper device and return true in case | |
5789 | * it is. The caller must hold the RTNL lock. | |
5790 | */ | |
25cc72a3 | 5791 | bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
5792 | { |
5793 | ASSERT_RTNL(); | |
5794 | ||
f1170fd4 | 5795 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 5796 | } |
25cc72a3 | 5797 | EXPORT_SYMBOL(netdev_has_any_upper_dev); |
9ff162a8 JP |
5798 | |
5799 | /** | |
5800 | * netdev_master_upper_dev_get - Get master upper device | |
5801 | * @dev: device | |
5802 | * | |
5803 | * Find a master upper device and return pointer to it or NULL in case | |
5804 | * it's not there. The caller must hold the RTNL lock. | |
5805 | */ | |
5806 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
5807 | { | |
aa9d8560 | 5808 | struct netdev_adjacent *upper; |
9ff162a8 JP |
5809 | |
5810 | ASSERT_RTNL(); | |
5811 | ||
2f268f12 | 5812 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
5813 | return NULL; |
5814 | ||
2f268f12 | 5815 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 5816 | struct netdev_adjacent, list); |
9ff162a8 JP |
5817 | if (likely(upper->master)) |
5818 | return upper->dev; | |
5819 | return NULL; | |
5820 | } | |
5821 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
5822 | ||
0f524a80 DA |
5823 | /** |
5824 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
5825 | * @dev: device | |
5826 | * | |
5827 | * Find out if a device is linked to a lower device and return true in case | |
5828 | * it is. The caller must hold the RTNL lock. | |
5829 | */ | |
5830 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
5831 | { | |
5832 | ASSERT_RTNL(); | |
5833 | ||
5834 | return !list_empty(&dev->adj_list.lower); | |
5835 | } | |
5836 | ||
b6ccba4c VF |
5837 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
5838 | { | |
5839 | struct netdev_adjacent *adj; | |
5840 | ||
5841 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
5842 | ||
5843 | return adj->private; | |
5844 | } | |
5845 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
5846 | ||
44a40855 VY |
5847 | /** |
5848 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
5849 | * @dev: device | |
5850 | * @iter: list_head ** of the current position | |
5851 | * | |
5852 | * Gets the next device from the dev's upper list, starting from iter | |
5853 | * position. The caller must hold RCU read lock. | |
5854 | */ | |
5855 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
5856 | struct list_head **iter) | |
5857 | { | |
5858 | struct netdev_adjacent *upper; | |
5859 | ||
5860 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5861 | ||
5862 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5863 | ||
5864 | if (&upper->list == &dev->adj_list.upper) | |
5865 | return NULL; | |
5866 | ||
5867 | *iter = &upper->list; | |
5868 | ||
5869 | return upper->dev; | |
5870 | } | |
5871 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
5872 | ||
1a3f060c DA |
5873 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
5874 | struct list_head **iter) | |
5875 | { | |
5876 | struct netdev_adjacent *upper; | |
5877 | ||
5878 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5879 | ||
5880 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5881 | ||
5882 | if (&upper->list == &dev->adj_list.upper) | |
5883 | return NULL; | |
5884 | ||
5885 | *iter = &upper->list; | |
5886 | ||
5887 | return upper->dev; | |
5888 | } | |
5889 | ||
5890 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, | |
5891 | int (*fn)(struct net_device *dev, | |
5892 | void *data), | |
5893 | void *data) | |
5894 | { | |
5895 | struct net_device *udev; | |
5896 | struct list_head *iter; | |
5897 | int ret; | |
5898 | ||
5899 | for (iter = &dev->adj_list.upper, | |
5900 | udev = netdev_next_upper_dev_rcu(dev, &iter); | |
5901 | udev; | |
5902 | udev = netdev_next_upper_dev_rcu(dev, &iter)) { | |
5903 | /* first is the upper device itself */ | |
5904 | ret = fn(udev, data); | |
5905 | if (ret) | |
5906 | return ret; | |
5907 | ||
5908 | /* then look at all of its upper devices */ | |
5909 | ret = netdev_walk_all_upper_dev_rcu(udev, fn, data); | |
5910 | if (ret) | |
5911 | return ret; | |
5912 | } | |
5913 | ||
5914 | return 0; | |
5915 | } | |
5916 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
5917 | ||
31088a11 VF |
5918 | /** |
5919 | * netdev_lower_get_next_private - Get the next ->private from the | |
5920 | * lower neighbour list | |
5921 | * @dev: device | |
5922 | * @iter: list_head ** of the current position | |
5923 | * | |
5924 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5925 | * list, starting from iter position. The caller must hold either hold the | |
5926 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 5927 | * list will remain unchanged. |
31088a11 VF |
5928 | */ |
5929 | void *netdev_lower_get_next_private(struct net_device *dev, | |
5930 | struct list_head **iter) | |
5931 | { | |
5932 | struct netdev_adjacent *lower; | |
5933 | ||
5934 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
5935 | ||
5936 | if (&lower->list == &dev->adj_list.lower) | |
5937 | return NULL; | |
5938 | ||
6859e7df | 5939 | *iter = lower->list.next; |
31088a11 VF |
5940 | |
5941 | return lower->private; | |
5942 | } | |
5943 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
5944 | ||
5945 | /** | |
5946 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
5947 | * lower neighbour list, RCU | |
5948 | * variant | |
5949 | * @dev: device | |
5950 | * @iter: list_head ** of the current position | |
5951 | * | |
5952 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5953 | * list, starting from iter position. The caller must hold RCU read lock. | |
5954 | */ | |
5955 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
5956 | struct list_head **iter) | |
5957 | { | |
5958 | struct netdev_adjacent *lower; | |
5959 | ||
5960 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5961 | ||
5962 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5963 | ||
5964 | if (&lower->list == &dev->adj_list.lower) | |
5965 | return NULL; | |
5966 | ||
6859e7df | 5967 | *iter = &lower->list; |
31088a11 VF |
5968 | |
5969 | return lower->private; | |
5970 | } | |
5971 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
5972 | ||
4085ebe8 VY |
5973 | /** |
5974 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
5975 | * list | |
5976 | * @dev: device | |
5977 | * @iter: list_head ** of the current position | |
5978 | * | |
5979 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
5980 | * list, starting from iter position. The caller must hold RTNL lock or | |
5981 | * its own locking that guarantees that the neighbour lower | |
b469139e | 5982 | * list will remain unchanged. |
4085ebe8 VY |
5983 | */ |
5984 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
5985 | { | |
5986 | struct netdev_adjacent *lower; | |
5987 | ||
cfdd28be | 5988 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
5989 | |
5990 | if (&lower->list == &dev->adj_list.lower) | |
5991 | return NULL; | |
5992 | ||
cfdd28be | 5993 | *iter = lower->list.next; |
4085ebe8 VY |
5994 | |
5995 | return lower->dev; | |
5996 | } | |
5997 | EXPORT_SYMBOL(netdev_lower_get_next); | |
5998 | ||
1a3f060c DA |
5999 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
6000 | struct list_head **iter) | |
6001 | { | |
6002 | struct netdev_adjacent *lower; | |
6003 | ||
46b5ab1a | 6004 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
6005 | |
6006 | if (&lower->list == &dev->adj_list.lower) | |
6007 | return NULL; | |
6008 | ||
46b5ab1a | 6009 | *iter = &lower->list; |
1a3f060c DA |
6010 | |
6011 | return lower->dev; | |
6012 | } | |
6013 | ||
6014 | int netdev_walk_all_lower_dev(struct net_device *dev, | |
6015 | int (*fn)(struct net_device *dev, | |
6016 | void *data), | |
6017 | void *data) | |
6018 | { | |
6019 | struct net_device *ldev; | |
6020 | struct list_head *iter; | |
6021 | int ret; | |
6022 | ||
6023 | for (iter = &dev->adj_list.lower, | |
6024 | ldev = netdev_next_lower_dev(dev, &iter); | |
6025 | ldev; | |
6026 | ldev = netdev_next_lower_dev(dev, &iter)) { | |
6027 | /* first is the lower device itself */ | |
6028 | ret = fn(ldev, data); | |
6029 | if (ret) | |
6030 | return ret; | |
6031 | ||
6032 | /* then look at all of its lower devices */ | |
6033 | ret = netdev_walk_all_lower_dev(ldev, fn, data); | |
6034 | if (ret) | |
6035 | return ret; | |
6036 | } | |
6037 | ||
6038 | return 0; | |
6039 | } | |
6040 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
6041 | ||
1a3f060c DA |
6042 | static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
6043 | struct list_head **iter) | |
6044 | { | |
6045 | struct netdev_adjacent *lower; | |
6046 | ||
6047 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6048 | if (&lower->list == &dev->adj_list.lower) | |
6049 | return NULL; | |
6050 | ||
6051 | *iter = &lower->list; | |
6052 | ||
6053 | return lower->dev; | |
6054 | } | |
6055 | ||
6056 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
6057 | int (*fn)(struct net_device *dev, | |
6058 | void *data), | |
6059 | void *data) | |
6060 | { | |
6061 | struct net_device *ldev; | |
6062 | struct list_head *iter; | |
6063 | int ret; | |
6064 | ||
6065 | for (iter = &dev->adj_list.lower, | |
6066 | ldev = netdev_next_lower_dev_rcu(dev, &iter); | |
6067 | ldev; | |
6068 | ldev = netdev_next_lower_dev_rcu(dev, &iter)) { | |
6069 | /* first is the lower device itself */ | |
6070 | ret = fn(ldev, data); | |
6071 | if (ret) | |
6072 | return ret; | |
6073 | ||
6074 | /* then look at all of its lower devices */ | |
6075 | ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data); | |
6076 | if (ret) | |
6077 | return ret; | |
6078 | } | |
6079 | ||
6080 | return 0; | |
6081 | } | |
6082 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
6083 | ||
e001bfad | 6084 | /** |
6085 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
6086 | * lower neighbour list, RCU | |
6087 | * variant | |
6088 | * @dev: device | |
6089 | * | |
6090 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
6091 | * list. The caller must hold RCU read lock. | |
6092 | */ | |
6093 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
6094 | { | |
6095 | struct netdev_adjacent *lower; | |
6096 | ||
6097 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
6098 | struct netdev_adjacent, list); | |
6099 | if (lower) | |
6100 | return lower->private; | |
6101 | return NULL; | |
6102 | } | |
6103 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
6104 | ||
9ff162a8 JP |
6105 | /** |
6106 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
6107 | * @dev: device | |
6108 | * | |
6109 | * Find a master upper device and return pointer to it or NULL in case | |
6110 | * it's not there. The caller must hold the RCU read lock. | |
6111 | */ | |
6112 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
6113 | { | |
aa9d8560 | 6114 | struct netdev_adjacent *upper; |
9ff162a8 | 6115 | |
2f268f12 | 6116 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 6117 | struct netdev_adjacent, list); |
9ff162a8 JP |
6118 | if (upper && likely(upper->master)) |
6119 | return upper->dev; | |
6120 | return NULL; | |
6121 | } | |
6122 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
6123 | ||
0a59f3a9 | 6124 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
6125 | struct net_device *adj_dev, |
6126 | struct list_head *dev_list) | |
6127 | { | |
6128 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 6129 | |
3ee32707 VF |
6130 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
6131 | "upper_%s" : "lower_%s", adj_dev->name); | |
6132 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
6133 | linkname); | |
6134 | } | |
0a59f3a9 | 6135 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
6136 | char *name, |
6137 | struct list_head *dev_list) | |
6138 | { | |
6139 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 6140 | |
3ee32707 VF |
6141 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
6142 | "upper_%s" : "lower_%s", name); | |
6143 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
6144 | } | |
6145 | ||
7ce64c79 AF |
6146 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
6147 | struct net_device *adj_dev, | |
6148 | struct list_head *dev_list) | |
6149 | { | |
6150 | return (dev_list == &dev->adj_list.upper || | |
6151 | dev_list == &dev->adj_list.lower) && | |
6152 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
6153 | } | |
3ee32707 | 6154 | |
5d261913 VF |
6155 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
6156 | struct net_device *adj_dev, | |
7863c054 | 6157 | struct list_head *dev_list, |
402dae96 | 6158 | void *private, bool master) |
5d261913 VF |
6159 | { |
6160 | struct netdev_adjacent *adj; | |
842d67a7 | 6161 | int ret; |
5d261913 | 6162 | |
6ea29da1 | 6163 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
6164 | |
6165 | if (adj) { | |
790510d9 | 6166 | adj->ref_nr += 1; |
67b62f98 DA |
6167 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
6168 | dev->name, adj_dev->name, adj->ref_nr); | |
6169 | ||
5d261913 VF |
6170 | return 0; |
6171 | } | |
6172 | ||
6173 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
6174 | if (!adj) | |
6175 | return -ENOMEM; | |
6176 | ||
6177 | adj->dev = adj_dev; | |
6178 | adj->master = master; | |
790510d9 | 6179 | adj->ref_nr = 1; |
402dae96 | 6180 | adj->private = private; |
5d261913 | 6181 | dev_hold(adj_dev); |
2f268f12 | 6182 | |
67b62f98 DA |
6183 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
6184 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 6185 | |
7ce64c79 | 6186 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 6187 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
6188 | if (ret) |
6189 | goto free_adj; | |
6190 | } | |
6191 | ||
7863c054 | 6192 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
6193 | if (master) { |
6194 | ret = sysfs_create_link(&(dev->dev.kobj), | |
6195 | &(adj_dev->dev.kobj), "master"); | |
6196 | if (ret) | |
5831d66e | 6197 | goto remove_symlinks; |
842d67a7 | 6198 | |
7863c054 | 6199 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 6200 | } else { |
7863c054 | 6201 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 6202 | } |
5d261913 VF |
6203 | |
6204 | return 0; | |
842d67a7 | 6205 | |
5831d66e | 6206 | remove_symlinks: |
7ce64c79 | 6207 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6208 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
6209 | free_adj: |
6210 | kfree(adj); | |
974daef7 | 6211 | dev_put(adj_dev); |
842d67a7 VF |
6212 | |
6213 | return ret; | |
5d261913 VF |
6214 | } |
6215 | ||
1d143d9f | 6216 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
6217 | struct net_device *adj_dev, | |
93409033 | 6218 | u16 ref_nr, |
1d143d9f | 6219 | struct list_head *dev_list) |
5d261913 VF |
6220 | { |
6221 | struct netdev_adjacent *adj; | |
6222 | ||
67b62f98 DA |
6223 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
6224 | dev->name, adj_dev->name, ref_nr); | |
6225 | ||
6ea29da1 | 6226 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 6227 | |
2f268f12 | 6228 | if (!adj) { |
67b62f98 | 6229 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 6230 | dev->name, adj_dev->name); |
67b62f98 DA |
6231 | WARN_ON(1); |
6232 | return; | |
2f268f12 | 6233 | } |
5d261913 | 6234 | |
93409033 | 6235 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
6236 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
6237 | dev->name, adj_dev->name, ref_nr, | |
6238 | adj->ref_nr - ref_nr); | |
93409033 | 6239 | adj->ref_nr -= ref_nr; |
5d261913 VF |
6240 | return; |
6241 | } | |
6242 | ||
842d67a7 VF |
6243 | if (adj->master) |
6244 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
6245 | ||
7ce64c79 | 6246 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6247 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 6248 | |
5d261913 | 6249 | list_del_rcu(&adj->list); |
67b62f98 | 6250 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 6251 | adj_dev->name, dev->name, adj_dev->name); |
5d261913 VF |
6252 | dev_put(adj_dev); |
6253 | kfree_rcu(adj, rcu); | |
6254 | } | |
6255 | ||
1d143d9f | 6256 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
6257 | struct net_device *upper_dev, | |
6258 | struct list_head *up_list, | |
6259 | struct list_head *down_list, | |
6260 | void *private, bool master) | |
5d261913 VF |
6261 | { |
6262 | int ret; | |
6263 | ||
790510d9 | 6264 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 6265 | private, master); |
5d261913 VF |
6266 | if (ret) |
6267 | return ret; | |
6268 | ||
790510d9 | 6269 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 6270 | private, false); |
5d261913 | 6271 | if (ret) { |
790510d9 | 6272 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
6273 | return ret; |
6274 | } | |
6275 | ||
6276 | return 0; | |
6277 | } | |
6278 | ||
1d143d9f | 6279 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
6280 | struct net_device *upper_dev, | |
93409033 | 6281 | u16 ref_nr, |
1d143d9f | 6282 | struct list_head *up_list, |
6283 | struct list_head *down_list) | |
5d261913 | 6284 | { |
93409033 AC |
6285 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
6286 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
6287 | } |
6288 | ||
1d143d9f | 6289 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
6290 | struct net_device *upper_dev, | |
6291 | void *private, bool master) | |
2f268f12 | 6292 | { |
f1170fd4 DA |
6293 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
6294 | &dev->adj_list.upper, | |
6295 | &upper_dev->adj_list.lower, | |
6296 | private, master); | |
5d261913 VF |
6297 | } |
6298 | ||
1d143d9f | 6299 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
6300 | struct net_device *upper_dev) | |
2f268f12 | 6301 | { |
93409033 | 6302 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
6303 | &dev->adj_list.upper, |
6304 | &upper_dev->adj_list.lower); | |
6305 | } | |
5d261913 | 6306 | |
9ff162a8 | 6307 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 6308 | struct net_device *upper_dev, bool master, |
42ab19ee DA |
6309 | void *upper_priv, void *upper_info, |
6310 | struct netlink_ext_ack *extack) | |
9ff162a8 | 6311 | { |
51d0c047 DA |
6312 | struct netdev_notifier_changeupper_info changeupper_info = { |
6313 | .info = { | |
6314 | .dev = dev, | |
42ab19ee | 6315 | .extack = extack, |
51d0c047 DA |
6316 | }, |
6317 | .upper_dev = upper_dev, | |
6318 | .master = master, | |
6319 | .linking = true, | |
6320 | .upper_info = upper_info, | |
6321 | }; | |
5d261913 | 6322 | int ret = 0; |
9ff162a8 JP |
6323 | |
6324 | ASSERT_RTNL(); | |
6325 | ||
6326 | if (dev == upper_dev) | |
6327 | return -EBUSY; | |
6328 | ||
6329 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
f1170fd4 | 6330 | if (netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
6331 | return -EBUSY; |
6332 | ||
f1170fd4 | 6333 | if (netdev_has_upper_dev(dev, upper_dev)) |
9ff162a8 JP |
6334 | return -EEXIST; |
6335 | ||
6336 | if (master && netdev_master_upper_dev_get(dev)) | |
6337 | return -EBUSY; | |
6338 | ||
51d0c047 | 6339 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
6340 | &changeupper_info.info); |
6341 | ret = notifier_to_errno(ret); | |
6342 | if (ret) | |
6343 | return ret; | |
6344 | ||
6dffb044 | 6345 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 6346 | master); |
5d261913 VF |
6347 | if (ret) |
6348 | return ret; | |
9ff162a8 | 6349 | |
51d0c047 | 6350 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
b03804e7 IS |
6351 | &changeupper_info.info); |
6352 | ret = notifier_to_errno(ret); | |
6353 | if (ret) | |
f1170fd4 | 6354 | goto rollback; |
b03804e7 | 6355 | |
9ff162a8 | 6356 | return 0; |
5d261913 | 6357 | |
f1170fd4 | 6358 | rollback: |
2f268f12 | 6359 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
6360 | |
6361 | return ret; | |
9ff162a8 JP |
6362 | } |
6363 | ||
6364 | /** | |
6365 | * netdev_upper_dev_link - Add a link to the upper device | |
6366 | * @dev: device | |
6367 | * @upper_dev: new upper device | |
6368 | * | |
6369 | * Adds a link to device which is upper to this one. The caller must hold | |
6370 | * the RTNL lock. On a failure a negative errno code is returned. | |
6371 | * On success the reference counts are adjusted and the function | |
6372 | * returns zero. | |
6373 | */ | |
6374 | int netdev_upper_dev_link(struct net_device *dev, | |
42ab19ee DA |
6375 | struct net_device *upper_dev, |
6376 | struct netlink_ext_ack *extack) | |
9ff162a8 | 6377 | { |
42ab19ee DA |
6378 | return __netdev_upper_dev_link(dev, upper_dev, false, |
6379 | NULL, NULL, extack); | |
9ff162a8 JP |
6380 | } |
6381 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
6382 | ||
6383 | /** | |
6384 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
6385 | * @dev: device | |
6386 | * @upper_dev: new upper device | |
6dffb044 | 6387 | * @upper_priv: upper device private |
29bf24af | 6388 | * @upper_info: upper info to be passed down via notifier |
9ff162a8 JP |
6389 | * |
6390 | * Adds a link to device which is upper to this one. In this case, only | |
6391 | * one master upper device can be linked, although other non-master devices | |
6392 | * might be linked as well. The caller must hold the RTNL lock. | |
6393 | * On a failure a negative errno code is returned. On success the reference | |
6394 | * counts are adjusted and the function returns zero. | |
6395 | */ | |
6396 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 6397 | struct net_device *upper_dev, |
42ab19ee DA |
6398 | void *upper_priv, void *upper_info, |
6399 | struct netlink_ext_ack *extack) | |
9ff162a8 | 6400 | { |
29bf24af | 6401 | return __netdev_upper_dev_link(dev, upper_dev, true, |
42ab19ee | 6402 | upper_priv, upper_info, extack); |
9ff162a8 JP |
6403 | } |
6404 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
6405 | ||
6406 | /** | |
6407 | * netdev_upper_dev_unlink - Removes a link to upper device | |
6408 | * @dev: device | |
6409 | * @upper_dev: new upper device | |
6410 | * | |
6411 | * Removes a link to device which is upper to this one. The caller must hold | |
6412 | * the RTNL lock. | |
6413 | */ | |
6414 | void netdev_upper_dev_unlink(struct net_device *dev, | |
6415 | struct net_device *upper_dev) | |
6416 | { | |
51d0c047 DA |
6417 | struct netdev_notifier_changeupper_info changeupper_info = { |
6418 | .info = { | |
6419 | .dev = dev, | |
6420 | }, | |
6421 | .upper_dev = upper_dev, | |
6422 | .linking = false, | |
6423 | }; | |
f4563a75 | 6424 | |
9ff162a8 JP |
6425 | ASSERT_RTNL(); |
6426 | ||
0e4ead9d | 6427 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; |
0e4ead9d | 6428 | |
51d0c047 | 6429 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
6430 | &changeupper_info.info); |
6431 | ||
2f268f12 | 6432 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 6433 | |
51d0c047 | 6434 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
0e4ead9d | 6435 | &changeupper_info.info); |
9ff162a8 JP |
6436 | } |
6437 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
6438 | ||
61bd3857 MS |
6439 | /** |
6440 | * netdev_bonding_info_change - Dispatch event about slave change | |
6441 | * @dev: device | |
4a26e453 | 6442 | * @bonding_info: info to dispatch |
61bd3857 MS |
6443 | * |
6444 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
6445 | * The caller must hold the RTNL lock. | |
6446 | */ | |
6447 | void netdev_bonding_info_change(struct net_device *dev, | |
6448 | struct netdev_bonding_info *bonding_info) | |
6449 | { | |
51d0c047 DA |
6450 | struct netdev_notifier_bonding_info info = { |
6451 | .info.dev = dev, | |
6452 | }; | |
61bd3857 MS |
6453 | |
6454 | memcpy(&info.bonding_info, bonding_info, | |
6455 | sizeof(struct netdev_bonding_info)); | |
51d0c047 | 6456 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, |
61bd3857 MS |
6457 | &info.info); |
6458 | } | |
6459 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
6460 | ||
2ce1ee17 | 6461 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
6462 | { |
6463 | struct netdev_adjacent *iter; | |
6464 | ||
6465 | struct net *net = dev_net(dev); | |
6466 | ||
6467 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 6468 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6469 | continue; |
6470 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6471 | &iter->dev->adj_list.lower); | |
6472 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
6473 | &dev->adj_list.upper); | |
6474 | } | |
6475 | ||
6476 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 6477 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6478 | continue; |
6479 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6480 | &iter->dev->adj_list.upper); | |
6481 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
6482 | &dev->adj_list.lower); | |
6483 | } | |
6484 | } | |
6485 | ||
2ce1ee17 | 6486 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
6487 | { |
6488 | struct netdev_adjacent *iter; | |
6489 | ||
6490 | struct net *net = dev_net(dev); | |
6491 | ||
6492 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 6493 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6494 | continue; |
6495 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6496 | &iter->dev->adj_list.lower); | |
6497 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6498 | &dev->adj_list.upper); | |
6499 | } | |
6500 | ||
6501 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 6502 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6503 | continue; |
6504 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6505 | &iter->dev->adj_list.upper); | |
6506 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6507 | &dev->adj_list.lower); | |
6508 | } | |
6509 | } | |
6510 | ||
5bb025fa | 6511 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 6512 | { |
5bb025fa | 6513 | struct netdev_adjacent *iter; |
402dae96 | 6514 | |
4c75431a AF |
6515 | struct net *net = dev_net(dev); |
6516 | ||
5bb025fa | 6517 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 6518 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6519 | continue; |
5bb025fa VF |
6520 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6521 | &iter->dev->adj_list.lower); | |
6522 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6523 | &iter->dev->adj_list.lower); | |
6524 | } | |
402dae96 | 6525 | |
5bb025fa | 6526 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 6527 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6528 | continue; |
5bb025fa VF |
6529 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6530 | &iter->dev->adj_list.upper); | |
6531 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6532 | &iter->dev->adj_list.upper); | |
6533 | } | |
402dae96 | 6534 | } |
402dae96 VF |
6535 | |
6536 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
6537 | struct net_device *lower_dev) | |
6538 | { | |
6539 | struct netdev_adjacent *lower; | |
6540 | ||
6541 | if (!lower_dev) | |
6542 | return NULL; | |
6ea29da1 | 6543 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
6544 | if (!lower) |
6545 | return NULL; | |
6546 | ||
6547 | return lower->private; | |
6548 | } | |
6549 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
6550 | ||
4085ebe8 | 6551 | |
952fcfd0 | 6552 | int dev_get_nest_level(struct net_device *dev) |
4085ebe8 VY |
6553 | { |
6554 | struct net_device *lower = NULL; | |
6555 | struct list_head *iter; | |
6556 | int max_nest = -1; | |
6557 | int nest; | |
6558 | ||
6559 | ASSERT_RTNL(); | |
6560 | ||
6561 | netdev_for_each_lower_dev(dev, lower, iter) { | |
952fcfd0 | 6562 | nest = dev_get_nest_level(lower); |
4085ebe8 VY |
6563 | if (max_nest < nest) |
6564 | max_nest = nest; | |
6565 | } | |
6566 | ||
952fcfd0 | 6567 | return max_nest + 1; |
4085ebe8 VY |
6568 | } |
6569 | EXPORT_SYMBOL(dev_get_nest_level); | |
6570 | ||
04d48266 JP |
6571 | /** |
6572 | * netdev_lower_change - Dispatch event about lower device state change | |
6573 | * @lower_dev: device | |
6574 | * @lower_state_info: state to dispatch | |
6575 | * | |
6576 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
6577 | * The caller must hold the RTNL lock. | |
6578 | */ | |
6579 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
6580 | void *lower_state_info) | |
6581 | { | |
51d0c047 DA |
6582 | struct netdev_notifier_changelowerstate_info changelowerstate_info = { |
6583 | .info.dev = lower_dev, | |
6584 | }; | |
04d48266 JP |
6585 | |
6586 | ASSERT_RTNL(); | |
6587 | changelowerstate_info.lower_state_info = lower_state_info; | |
51d0c047 | 6588 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, |
04d48266 JP |
6589 | &changelowerstate_info.info); |
6590 | } | |
6591 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
6592 | ||
b6c40d68 PM |
6593 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
6594 | { | |
d314774c SH |
6595 | const struct net_device_ops *ops = dev->netdev_ops; |
6596 | ||
d2615bf4 | 6597 | if (ops->ndo_change_rx_flags) |
d314774c | 6598 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
6599 | } |
6600 | ||
991fb3f7 | 6601 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6602 | { |
b536db93 | 6603 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
6604 | kuid_t uid; |
6605 | kgid_t gid; | |
1da177e4 | 6606 | |
24023451 PM |
6607 | ASSERT_RTNL(); |
6608 | ||
dad9b335 WC |
6609 | dev->flags |= IFF_PROMISC; |
6610 | dev->promiscuity += inc; | |
6611 | if (dev->promiscuity == 0) { | |
6612 | /* | |
6613 | * Avoid overflow. | |
6614 | * If inc causes overflow, untouch promisc and return error. | |
6615 | */ | |
6616 | if (inc < 0) | |
6617 | dev->flags &= ~IFF_PROMISC; | |
6618 | else { | |
6619 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
6620 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
6621 | dev->name); | |
dad9b335 WC |
6622 | return -EOVERFLOW; |
6623 | } | |
6624 | } | |
52609c0b | 6625 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
6626 | pr_info("device %s %s promiscuous mode\n", |
6627 | dev->name, | |
6628 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
6629 | if (audit_enabled) { |
6630 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
6631 | audit_log(current->audit_context, GFP_ATOMIC, |
6632 | AUDIT_ANOM_PROMISCUOUS, | |
6633 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
6634 | dev->name, (dev->flags & IFF_PROMISC), | |
6635 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 6636 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
6637 | from_kuid(&init_user_ns, uid), |
6638 | from_kgid(&init_user_ns, gid), | |
7759db82 | 6639 | audit_get_sessionid(current)); |
8192b0c4 | 6640 | } |
24023451 | 6641 | |
b6c40d68 | 6642 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 6643 | } |
991fb3f7 ND |
6644 | if (notify) |
6645 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 6646 | return 0; |
1da177e4 LT |
6647 | } |
6648 | ||
4417da66 PM |
6649 | /** |
6650 | * dev_set_promiscuity - update promiscuity count on a device | |
6651 | * @dev: device | |
6652 | * @inc: modifier | |
6653 | * | |
6654 | * Add or remove promiscuity from a device. While the count in the device | |
6655 | * remains above zero the interface remains promiscuous. Once it hits zero | |
6656 | * the device reverts back to normal filtering operation. A negative inc | |
6657 | * value is used to drop promiscuity on the device. | |
dad9b335 | 6658 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 6659 | */ |
dad9b335 | 6660 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 6661 | { |
b536db93 | 6662 | unsigned int old_flags = dev->flags; |
dad9b335 | 6663 | int err; |
4417da66 | 6664 | |
991fb3f7 | 6665 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 6666 | if (err < 0) |
dad9b335 | 6667 | return err; |
4417da66 PM |
6668 | if (dev->flags != old_flags) |
6669 | dev_set_rx_mode(dev); | |
dad9b335 | 6670 | return err; |
4417da66 | 6671 | } |
d1b19dff | 6672 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 6673 | |
991fb3f7 | 6674 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6675 | { |
991fb3f7 | 6676 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 6677 | |
24023451 PM |
6678 | ASSERT_RTNL(); |
6679 | ||
1da177e4 | 6680 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
6681 | dev->allmulti += inc; |
6682 | if (dev->allmulti == 0) { | |
6683 | /* | |
6684 | * Avoid overflow. | |
6685 | * If inc causes overflow, untouch allmulti and return error. | |
6686 | */ | |
6687 | if (inc < 0) | |
6688 | dev->flags &= ~IFF_ALLMULTI; | |
6689 | else { | |
6690 | dev->allmulti -= inc; | |
7b6cd1ce JP |
6691 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
6692 | dev->name); | |
dad9b335 WC |
6693 | return -EOVERFLOW; |
6694 | } | |
6695 | } | |
24023451 | 6696 | if (dev->flags ^ old_flags) { |
b6c40d68 | 6697 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 6698 | dev_set_rx_mode(dev); |
991fb3f7 ND |
6699 | if (notify) |
6700 | __dev_notify_flags(dev, old_flags, | |
6701 | dev->gflags ^ old_gflags); | |
24023451 | 6702 | } |
dad9b335 | 6703 | return 0; |
4417da66 | 6704 | } |
991fb3f7 ND |
6705 | |
6706 | /** | |
6707 | * dev_set_allmulti - update allmulti count on a device | |
6708 | * @dev: device | |
6709 | * @inc: modifier | |
6710 | * | |
6711 | * Add or remove reception of all multicast frames to a device. While the | |
6712 | * count in the device remains above zero the interface remains listening | |
6713 | * to all interfaces. Once it hits zero the device reverts back to normal | |
6714 | * filtering operation. A negative @inc value is used to drop the counter | |
6715 | * when releasing a resource needing all multicasts. | |
6716 | * Return 0 if successful or a negative errno code on error. | |
6717 | */ | |
6718 | ||
6719 | int dev_set_allmulti(struct net_device *dev, int inc) | |
6720 | { | |
6721 | return __dev_set_allmulti(dev, inc, true); | |
6722 | } | |
d1b19dff | 6723 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
6724 | |
6725 | /* | |
6726 | * Upload unicast and multicast address lists to device and | |
6727 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 6728 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
6729 | * are present. |
6730 | */ | |
6731 | void __dev_set_rx_mode(struct net_device *dev) | |
6732 | { | |
d314774c SH |
6733 | const struct net_device_ops *ops = dev->netdev_ops; |
6734 | ||
4417da66 PM |
6735 | /* dev_open will call this function so the list will stay sane. */ |
6736 | if (!(dev->flags&IFF_UP)) | |
6737 | return; | |
6738 | ||
6739 | if (!netif_device_present(dev)) | |
40b77c94 | 6740 | return; |
4417da66 | 6741 | |
01789349 | 6742 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
6743 | /* Unicast addresses changes may only happen under the rtnl, |
6744 | * therefore calling __dev_set_promiscuity here is safe. | |
6745 | */ | |
32e7bfc4 | 6746 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 6747 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 6748 | dev->uc_promisc = true; |
32e7bfc4 | 6749 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 6750 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 6751 | dev->uc_promisc = false; |
4417da66 | 6752 | } |
4417da66 | 6753 | } |
01789349 JP |
6754 | |
6755 | if (ops->ndo_set_rx_mode) | |
6756 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
6757 | } |
6758 | ||
6759 | void dev_set_rx_mode(struct net_device *dev) | |
6760 | { | |
b9e40857 | 6761 | netif_addr_lock_bh(dev); |
4417da66 | 6762 | __dev_set_rx_mode(dev); |
b9e40857 | 6763 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
6764 | } |
6765 | ||
f0db275a SH |
6766 | /** |
6767 | * dev_get_flags - get flags reported to userspace | |
6768 | * @dev: device | |
6769 | * | |
6770 | * Get the combination of flag bits exported through APIs to userspace. | |
6771 | */ | |
95c96174 | 6772 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 6773 | { |
95c96174 | 6774 | unsigned int flags; |
1da177e4 LT |
6775 | |
6776 | flags = (dev->flags & ~(IFF_PROMISC | | |
6777 | IFF_ALLMULTI | | |
b00055aa SR |
6778 | IFF_RUNNING | |
6779 | IFF_LOWER_UP | | |
6780 | IFF_DORMANT)) | | |
1da177e4 LT |
6781 | (dev->gflags & (IFF_PROMISC | |
6782 | IFF_ALLMULTI)); | |
6783 | ||
b00055aa SR |
6784 | if (netif_running(dev)) { |
6785 | if (netif_oper_up(dev)) | |
6786 | flags |= IFF_RUNNING; | |
6787 | if (netif_carrier_ok(dev)) | |
6788 | flags |= IFF_LOWER_UP; | |
6789 | if (netif_dormant(dev)) | |
6790 | flags |= IFF_DORMANT; | |
6791 | } | |
1da177e4 LT |
6792 | |
6793 | return flags; | |
6794 | } | |
d1b19dff | 6795 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 6796 | |
bd380811 | 6797 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 6798 | { |
b536db93 | 6799 | unsigned int old_flags = dev->flags; |
bd380811 | 6800 | int ret; |
1da177e4 | 6801 | |
24023451 PM |
6802 | ASSERT_RTNL(); |
6803 | ||
1da177e4 LT |
6804 | /* |
6805 | * Set the flags on our device. | |
6806 | */ | |
6807 | ||
6808 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
6809 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
6810 | IFF_AUTOMEDIA)) | | |
6811 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
6812 | IFF_ALLMULTI)); | |
6813 | ||
6814 | /* | |
6815 | * Load in the correct multicast list now the flags have changed. | |
6816 | */ | |
6817 | ||
b6c40d68 PM |
6818 | if ((old_flags ^ flags) & IFF_MULTICAST) |
6819 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 6820 | |
4417da66 | 6821 | dev_set_rx_mode(dev); |
1da177e4 LT |
6822 | |
6823 | /* | |
6824 | * Have we downed the interface. We handle IFF_UP ourselves | |
6825 | * according to user attempts to set it, rather than blindly | |
6826 | * setting it. | |
6827 | */ | |
6828 | ||
6829 | ret = 0; | |
7051b88a | 6830 | if ((old_flags ^ flags) & IFF_UP) { |
6831 | if (old_flags & IFF_UP) | |
6832 | __dev_close(dev); | |
6833 | else | |
6834 | ret = __dev_open(dev); | |
6835 | } | |
1da177e4 | 6836 | |
1da177e4 | 6837 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 6838 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 6839 | unsigned int old_flags = dev->flags; |
d1b19dff | 6840 | |
1da177e4 | 6841 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
6842 | |
6843 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
6844 | if (dev->flags != old_flags) | |
6845 | dev_set_rx_mode(dev); | |
1da177e4 LT |
6846 | } |
6847 | ||
6848 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 6849 | * is important. Some (broken) drivers set IFF_PROMISC, when |
6850 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
6851 | */ |
6852 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
6853 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
6854 | ||
1da177e4 | 6855 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 6856 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
6857 | } |
6858 | ||
bd380811 PM |
6859 | return ret; |
6860 | } | |
6861 | ||
a528c219 ND |
6862 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
6863 | unsigned int gchanges) | |
bd380811 PM |
6864 | { |
6865 | unsigned int changes = dev->flags ^ old_flags; | |
6866 | ||
a528c219 | 6867 | if (gchanges) |
7f294054 | 6868 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 6869 | |
bd380811 PM |
6870 | if (changes & IFF_UP) { |
6871 | if (dev->flags & IFF_UP) | |
6872 | call_netdevice_notifiers(NETDEV_UP, dev); | |
6873 | else | |
6874 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
6875 | } | |
6876 | ||
6877 | if (dev->flags & IFF_UP && | |
be9efd36 | 6878 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
51d0c047 DA |
6879 | struct netdev_notifier_change_info change_info = { |
6880 | .info = { | |
6881 | .dev = dev, | |
6882 | }, | |
6883 | .flags_changed = changes, | |
6884 | }; | |
6885 | ||
6886 | call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); | |
be9efd36 | 6887 | } |
bd380811 PM |
6888 | } |
6889 | ||
6890 | /** | |
6891 | * dev_change_flags - change device settings | |
6892 | * @dev: device | |
6893 | * @flags: device state flags | |
6894 | * | |
6895 | * Change settings on device based state flags. The flags are | |
6896 | * in the userspace exported format. | |
6897 | */ | |
b536db93 | 6898 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 6899 | { |
b536db93 | 6900 | int ret; |
991fb3f7 | 6901 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
6902 | |
6903 | ret = __dev_change_flags(dev, flags); | |
6904 | if (ret < 0) | |
6905 | return ret; | |
6906 | ||
991fb3f7 | 6907 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 6908 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
6909 | return ret; |
6910 | } | |
d1b19dff | 6911 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 6912 | |
f51048c3 | 6913 | int __dev_set_mtu(struct net_device *dev, int new_mtu) |
2315dc91 VF |
6914 | { |
6915 | const struct net_device_ops *ops = dev->netdev_ops; | |
6916 | ||
6917 | if (ops->ndo_change_mtu) | |
6918 | return ops->ndo_change_mtu(dev, new_mtu); | |
6919 | ||
6920 | dev->mtu = new_mtu; | |
6921 | return 0; | |
6922 | } | |
f51048c3 | 6923 | EXPORT_SYMBOL(__dev_set_mtu); |
2315dc91 | 6924 | |
f0db275a SH |
6925 | /** |
6926 | * dev_set_mtu - Change maximum transfer unit | |
6927 | * @dev: device | |
6928 | * @new_mtu: new transfer unit | |
6929 | * | |
6930 | * Change the maximum transfer size of the network device. | |
6931 | */ | |
1da177e4 LT |
6932 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
6933 | { | |
2315dc91 | 6934 | int err, orig_mtu; |
1da177e4 LT |
6935 | |
6936 | if (new_mtu == dev->mtu) | |
6937 | return 0; | |
6938 | ||
61e84623 JW |
6939 | /* MTU must be positive, and in range */ |
6940 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
6941 | net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n", | |
6942 | dev->name, new_mtu, dev->min_mtu); | |
1da177e4 | 6943 | return -EINVAL; |
61e84623 JW |
6944 | } |
6945 | ||
6946 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
6947 | net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n", | |
a0e65de7 | 6948 | dev->name, new_mtu, dev->max_mtu); |
61e84623 JW |
6949 | return -EINVAL; |
6950 | } | |
1da177e4 LT |
6951 | |
6952 | if (!netif_device_present(dev)) | |
6953 | return -ENODEV; | |
6954 | ||
1d486bfb VF |
6955 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
6956 | err = notifier_to_errno(err); | |
6957 | if (err) | |
6958 | return err; | |
d314774c | 6959 | |
2315dc91 VF |
6960 | orig_mtu = dev->mtu; |
6961 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 6962 | |
2315dc91 VF |
6963 | if (!err) { |
6964 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6965 | err = notifier_to_errno(err); | |
6966 | if (err) { | |
6967 | /* setting mtu back and notifying everyone again, | |
6968 | * so that they have a chance to revert changes. | |
6969 | */ | |
6970 | __dev_set_mtu(dev, orig_mtu); | |
6971 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6972 | } | |
6973 | } | |
1da177e4 LT |
6974 | return err; |
6975 | } | |
d1b19dff | 6976 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 6977 | |
cbda10fa VD |
6978 | /** |
6979 | * dev_set_group - Change group this device belongs to | |
6980 | * @dev: device | |
6981 | * @new_group: group this device should belong to | |
6982 | */ | |
6983 | void dev_set_group(struct net_device *dev, int new_group) | |
6984 | { | |
6985 | dev->group = new_group; | |
6986 | } | |
6987 | EXPORT_SYMBOL(dev_set_group); | |
6988 | ||
f0db275a SH |
6989 | /** |
6990 | * dev_set_mac_address - Change Media Access Control Address | |
6991 | * @dev: device | |
6992 | * @sa: new address | |
6993 | * | |
6994 | * Change the hardware (MAC) address of the device | |
6995 | */ | |
1da177e4 LT |
6996 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
6997 | { | |
d314774c | 6998 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
6999 | int err; |
7000 | ||
d314774c | 7001 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
7002 | return -EOPNOTSUPP; |
7003 | if (sa->sa_family != dev->type) | |
7004 | return -EINVAL; | |
7005 | if (!netif_device_present(dev)) | |
7006 | return -ENODEV; | |
d314774c | 7007 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
7008 | if (err) |
7009 | return err; | |
fbdeca2d | 7010 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 7011 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 7012 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 7013 | return 0; |
1da177e4 | 7014 | } |
d1b19dff | 7015 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 7016 | |
4bf84c35 JP |
7017 | /** |
7018 | * dev_change_carrier - Change device carrier | |
7019 | * @dev: device | |
691b3b7e | 7020 | * @new_carrier: new value |
4bf84c35 JP |
7021 | * |
7022 | * Change device carrier | |
7023 | */ | |
7024 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
7025 | { | |
7026 | const struct net_device_ops *ops = dev->netdev_ops; | |
7027 | ||
7028 | if (!ops->ndo_change_carrier) | |
7029 | return -EOPNOTSUPP; | |
7030 | if (!netif_device_present(dev)) | |
7031 | return -ENODEV; | |
7032 | return ops->ndo_change_carrier(dev, new_carrier); | |
7033 | } | |
7034 | EXPORT_SYMBOL(dev_change_carrier); | |
7035 | ||
66b52b0d JP |
7036 | /** |
7037 | * dev_get_phys_port_id - Get device physical port ID | |
7038 | * @dev: device | |
7039 | * @ppid: port ID | |
7040 | * | |
7041 | * Get device physical port ID | |
7042 | */ | |
7043 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 7044 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
7045 | { |
7046 | const struct net_device_ops *ops = dev->netdev_ops; | |
7047 | ||
7048 | if (!ops->ndo_get_phys_port_id) | |
7049 | return -EOPNOTSUPP; | |
7050 | return ops->ndo_get_phys_port_id(dev, ppid); | |
7051 | } | |
7052 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
7053 | ||
db24a904 DA |
7054 | /** |
7055 | * dev_get_phys_port_name - Get device physical port name | |
7056 | * @dev: device | |
7057 | * @name: port name | |
ed49e650 | 7058 | * @len: limit of bytes to copy to name |
db24a904 DA |
7059 | * |
7060 | * Get device physical port name | |
7061 | */ | |
7062 | int dev_get_phys_port_name(struct net_device *dev, | |
7063 | char *name, size_t len) | |
7064 | { | |
7065 | const struct net_device_ops *ops = dev->netdev_ops; | |
7066 | ||
7067 | if (!ops->ndo_get_phys_port_name) | |
7068 | return -EOPNOTSUPP; | |
7069 | return ops->ndo_get_phys_port_name(dev, name, len); | |
7070 | } | |
7071 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
7072 | ||
d746d707 AK |
7073 | /** |
7074 | * dev_change_proto_down - update protocol port state information | |
7075 | * @dev: device | |
7076 | * @proto_down: new value | |
7077 | * | |
7078 | * This info can be used by switch drivers to set the phys state of the | |
7079 | * port. | |
7080 | */ | |
7081 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
7082 | { | |
7083 | const struct net_device_ops *ops = dev->netdev_ops; | |
7084 | ||
7085 | if (!ops->ndo_change_proto_down) | |
7086 | return -EOPNOTSUPP; | |
7087 | if (!netif_device_present(dev)) | |
7088 | return -ENODEV; | |
7089 | return ops->ndo_change_proto_down(dev, proto_down); | |
7090 | } | |
7091 | EXPORT_SYMBOL(dev_change_proto_down); | |
7092 | ||
f4e63525 | 7093 | u8 __dev_xdp_attached(struct net_device *dev, bpf_op_t bpf_op, u32 *prog_id) |
d67b9cd2 | 7094 | { |
f4e63525 | 7095 | struct netdev_bpf xdp; |
d67b9cd2 DB |
7096 | |
7097 | memset(&xdp, 0, sizeof(xdp)); | |
7098 | xdp.command = XDP_QUERY_PROG; | |
7099 | ||
7100 | /* Query must always succeed. */ | |
f4e63525 | 7101 | WARN_ON(bpf_op(dev, &xdp) < 0); |
58038695 MKL |
7102 | if (prog_id) |
7103 | *prog_id = xdp.prog_id; | |
7104 | ||
d67b9cd2 DB |
7105 | return xdp.prog_attached; |
7106 | } | |
7107 | ||
f4e63525 | 7108 | static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op, |
32d60277 | 7109 | struct netlink_ext_ack *extack, u32 flags, |
d67b9cd2 DB |
7110 | struct bpf_prog *prog) |
7111 | { | |
f4e63525 | 7112 | struct netdev_bpf xdp; |
d67b9cd2 DB |
7113 | |
7114 | memset(&xdp, 0, sizeof(xdp)); | |
ee5d032f JK |
7115 | if (flags & XDP_FLAGS_HW_MODE) |
7116 | xdp.command = XDP_SETUP_PROG_HW; | |
7117 | else | |
7118 | xdp.command = XDP_SETUP_PROG; | |
d67b9cd2 | 7119 | xdp.extack = extack; |
32d60277 | 7120 | xdp.flags = flags; |
d67b9cd2 DB |
7121 | xdp.prog = prog; |
7122 | ||
f4e63525 | 7123 | return bpf_op(dev, &xdp); |
d67b9cd2 DB |
7124 | } |
7125 | ||
a7862b45 BB |
7126 | /** |
7127 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
7128 | * @dev: device | |
b5d60989 | 7129 | * @extack: netlink extended ack |
a7862b45 | 7130 | * @fd: new program fd or negative value to clear |
85de8576 | 7131 | * @flags: xdp-related flags |
a7862b45 BB |
7132 | * |
7133 | * Set or clear a bpf program for a device | |
7134 | */ | |
ddf9f970 JK |
7135 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, |
7136 | int fd, u32 flags) | |
a7862b45 BB |
7137 | { |
7138 | const struct net_device_ops *ops = dev->netdev_ops; | |
7139 | struct bpf_prog *prog = NULL; | |
f4e63525 | 7140 | bpf_op_t bpf_op, bpf_chk; |
a7862b45 BB |
7141 | int err; |
7142 | ||
85de8576 DB |
7143 | ASSERT_RTNL(); |
7144 | ||
f4e63525 JK |
7145 | bpf_op = bpf_chk = ops->ndo_bpf; |
7146 | if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) | |
0489df9a | 7147 | return -EOPNOTSUPP; |
f4e63525 JK |
7148 | if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE)) |
7149 | bpf_op = generic_xdp_install; | |
7150 | if (bpf_op == bpf_chk) | |
7151 | bpf_chk = generic_xdp_install; | |
b5cdae32 | 7152 | |
a7862b45 | 7153 | if (fd >= 0) { |
f4e63525 | 7154 | if (bpf_chk && __dev_xdp_attached(dev, bpf_chk, NULL)) |
d67b9cd2 DB |
7155 | return -EEXIST; |
7156 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && | |
f4e63525 | 7157 | __dev_xdp_attached(dev, bpf_op, NULL)) |
d67b9cd2 | 7158 | return -EBUSY; |
85de8576 | 7159 | |
248f346f JK |
7160 | if (bpf_op == ops->ndo_bpf) |
7161 | prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, | |
7162 | dev); | |
7163 | else | |
7164 | prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP); | |
a7862b45 BB |
7165 | if (IS_ERR(prog)) |
7166 | return PTR_ERR(prog); | |
7167 | } | |
7168 | ||
f4e63525 | 7169 | err = dev_xdp_install(dev, bpf_op, extack, flags, prog); |
a7862b45 BB |
7170 | if (err < 0 && prog) |
7171 | bpf_prog_put(prog); | |
7172 | ||
7173 | return err; | |
7174 | } | |
a7862b45 | 7175 | |
1da177e4 LT |
7176 | /** |
7177 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 7178 | * @net: the applicable net namespace |
1da177e4 LT |
7179 | * |
7180 | * Returns a suitable unique value for a new device interface | |
7181 | * number. The caller must hold the rtnl semaphore or the | |
7182 | * dev_base_lock to be sure it remains unique. | |
7183 | */ | |
881d966b | 7184 | static int dev_new_index(struct net *net) |
1da177e4 | 7185 | { |
aa79e66e | 7186 | int ifindex = net->ifindex; |
f4563a75 | 7187 | |
1da177e4 LT |
7188 | for (;;) { |
7189 | if (++ifindex <= 0) | |
7190 | ifindex = 1; | |
881d966b | 7191 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 7192 | return net->ifindex = ifindex; |
1da177e4 LT |
7193 | } |
7194 | } | |
7195 | ||
1da177e4 | 7196 | /* Delayed registration/unregisteration */ |
3b5b34fd | 7197 | static LIST_HEAD(net_todo_list); |
200b916f | 7198 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 7199 | |
6f05f629 | 7200 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 7201 | { |
1da177e4 | 7202 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 7203 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
7204 | } |
7205 | ||
9b5e383c | 7206 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 7207 | { |
e93737b0 | 7208 | struct net_device *dev, *tmp; |
5cde2829 | 7209 | LIST_HEAD(close_head); |
9b5e383c | 7210 | |
93ee31f1 DL |
7211 | BUG_ON(dev_boot_phase); |
7212 | ASSERT_RTNL(); | |
7213 | ||
e93737b0 | 7214 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 7215 | /* Some devices call without registering |
e93737b0 KK |
7216 | * for initialization unwind. Remove those |
7217 | * devices and proceed with the remaining. | |
9b5e383c ED |
7218 | */ |
7219 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
7220 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
7221 | dev->name, dev); | |
93ee31f1 | 7222 | |
9b5e383c | 7223 | WARN_ON(1); |
e93737b0 KK |
7224 | list_del(&dev->unreg_list); |
7225 | continue; | |
9b5e383c | 7226 | } |
449f4544 | 7227 | dev->dismantle = true; |
9b5e383c | 7228 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 7229 | } |
93ee31f1 | 7230 | |
44345724 | 7231 | /* If device is running, close it first. */ |
5cde2829 EB |
7232 | list_for_each_entry(dev, head, unreg_list) |
7233 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 7234 | dev_close_many(&close_head, true); |
93ee31f1 | 7235 | |
44345724 | 7236 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
7237 | /* And unlink it from device chain. */ |
7238 | unlist_netdevice(dev); | |
93ee31f1 | 7239 | |
9b5e383c ED |
7240 | dev->reg_state = NETREG_UNREGISTERING; |
7241 | } | |
41852497 | 7242 | flush_all_backlogs(); |
93ee31f1 DL |
7243 | |
7244 | synchronize_net(); | |
7245 | ||
9b5e383c | 7246 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
7247 | struct sk_buff *skb = NULL; |
7248 | ||
9b5e383c ED |
7249 | /* Shutdown queueing discipline. */ |
7250 | dev_shutdown(dev); | |
93ee31f1 DL |
7251 | |
7252 | ||
9b5e383c | 7253 | /* Notify protocols, that we are about to destroy |
eb13da1a | 7254 | * this device. They should clean all the things. |
7255 | */ | |
9b5e383c | 7256 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
93ee31f1 | 7257 | |
395eea6c MB |
7258 | if (!dev->rtnl_link_ops || |
7259 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
3d3ea5af | 7260 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, |
6621dd29 | 7261 | GFP_KERNEL, NULL); |
395eea6c | 7262 | |
9b5e383c ED |
7263 | /* |
7264 | * Flush the unicast and multicast chains | |
7265 | */ | |
a748ee24 | 7266 | dev_uc_flush(dev); |
22bedad3 | 7267 | dev_mc_flush(dev); |
93ee31f1 | 7268 | |
9b5e383c ED |
7269 | if (dev->netdev_ops->ndo_uninit) |
7270 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 7271 | |
395eea6c MB |
7272 | if (skb) |
7273 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 7274 | |
9ff162a8 JP |
7275 | /* Notifier chain MUST detach us all upper devices. */ |
7276 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
0f524a80 | 7277 | WARN_ON(netdev_has_any_lower_dev(dev)); |
93ee31f1 | 7278 | |
9b5e383c ED |
7279 | /* Remove entries from kobject tree */ |
7280 | netdev_unregister_kobject(dev); | |
024e9679 AD |
7281 | #ifdef CONFIG_XPS |
7282 | /* Remove XPS queueing entries */ | |
7283 | netif_reset_xps_queues_gt(dev, 0); | |
7284 | #endif | |
9b5e383c | 7285 | } |
93ee31f1 | 7286 | |
850a545b | 7287 | synchronize_net(); |
395264d5 | 7288 | |
a5ee1551 | 7289 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
7290 | dev_put(dev); |
7291 | } | |
7292 | ||
7293 | static void rollback_registered(struct net_device *dev) | |
7294 | { | |
7295 | LIST_HEAD(single); | |
7296 | ||
7297 | list_add(&dev->unreg_list, &single); | |
7298 | rollback_registered_many(&single); | |
ceaaec98 | 7299 | list_del(&single); |
93ee31f1 DL |
7300 | } |
7301 | ||
fd867d51 JW |
7302 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
7303 | struct net_device *upper, netdev_features_t features) | |
7304 | { | |
7305 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
7306 | netdev_features_t feature; | |
5ba3f7d6 | 7307 | int feature_bit; |
fd867d51 | 7308 | |
5ba3f7d6 JW |
7309 | for_each_netdev_feature(&upper_disables, feature_bit) { |
7310 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
7311 | if (!(upper->wanted_features & feature) |
7312 | && (features & feature)) { | |
7313 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
7314 | &feature, upper->name); | |
7315 | features &= ~feature; | |
7316 | } | |
7317 | } | |
7318 | ||
7319 | return features; | |
7320 | } | |
7321 | ||
7322 | static void netdev_sync_lower_features(struct net_device *upper, | |
7323 | struct net_device *lower, netdev_features_t features) | |
7324 | { | |
7325 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
7326 | netdev_features_t feature; | |
5ba3f7d6 | 7327 | int feature_bit; |
fd867d51 | 7328 | |
5ba3f7d6 JW |
7329 | for_each_netdev_feature(&upper_disables, feature_bit) { |
7330 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
7331 | if (!(features & feature) && (lower->features & feature)) { |
7332 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
7333 | &feature, lower->name); | |
7334 | lower->wanted_features &= ~feature; | |
7335 | netdev_update_features(lower); | |
7336 | ||
7337 | if (unlikely(lower->features & feature)) | |
7338 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
7339 | &feature, lower->name); | |
7340 | } | |
7341 | } | |
7342 | } | |
7343 | ||
c8f44aff MM |
7344 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
7345 | netdev_features_t features) | |
b63365a2 | 7346 | { |
57422dc5 MM |
7347 | /* Fix illegal checksum combinations */ |
7348 | if ((features & NETIF_F_HW_CSUM) && | |
7349 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 7350 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
7351 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
7352 | } | |
7353 | ||
b63365a2 | 7354 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 7355 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 7356 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 7357 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
7358 | } |
7359 | ||
ec5f0615 PS |
7360 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
7361 | !(features & NETIF_F_IP_CSUM)) { | |
7362 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
7363 | features &= ~NETIF_F_TSO; | |
7364 | features &= ~NETIF_F_TSO_ECN; | |
7365 | } | |
7366 | ||
7367 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
7368 | !(features & NETIF_F_IPV6_CSUM)) { | |
7369 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
7370 | features &= ~NETIF_F_TSO6; | |
7371 | } | |
7372 | ||
b1dc497b AD |
7373 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
7374 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
7375 | features &= ~NETIF_F_TSO_MANGLEID; | |
7376 | ||
31d8b9e0 BH |
7377 | /* TSO ECN requires that TSO is present as well. */ |
7378 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
7379 | features &= ~NETIF_F_TSO_ECN; | |
7380 | ||
212b573f MM |
7381 | /* Software GSO depends on SG. */ |
7382 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 7383 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
7384 | features &= ~NETIF_F_GSO; |
7385 | } | |
7386 | ||
802ab55a AD |
7387 | /* GSO partial features require GSO partial be set */ |
7388 | if ((features & dev->gso_partial_features) && | |
7389 | !(features & NETIF_F_GSO_PARTIAL)) { | |
7390 | netdev_dbg(dev, | |
7391 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
7392 | features &= ~dev->gso_partial_features; | |
7393 | } | |
7394 | ||
b63365a2 HX |
7395 | return features; |
7396 | } | |
b63365a2 | 7397 | |
6cb6a27c | 7398 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 7399 | { |
fd867d51 | 7400 | struct net_device *upper, *lower; |
c8f44aff | 7401 | netdev_features_t features; |
fd867d51 | 7402 | struct list_head *iter; |
e7868a85 | 7403 | int err = -1; |
5455c699 | 7404 | |
87267485 MM |
7405 | ASSERT_RTNL(); |
7406 | ||
5455c699 MM |
7407 | features = netdev_get_wanted_features(dev); |
7408 | ||
7409 | if (dev->netdev_ops->ndo_fix_features) | |
7410 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
7411 | ||
7412 | /* driver might be less strict about feature dependencies */ | |
7413 | features = netdev_fix_features(dev, features); | |
7414 | ||
fd867d51 JW |
7415 | /* some features can't be enabled if they're off an an upper device */ |
7416 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
7417 | features = netdev_sync_upper_features(dev, upper, features); | |
7418 | ||
5455c699 | 7419 | if (dev->features == features) |
e7868a85 | 7420 | goto sync_lower; |
5455c699 | 7421 | |
c8f44aff MM |
7422 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
7423 | &dev->features, &features); | |
5455c699 MM |
7424 | |
7425 | if (dev->netdev_ops->ndo_set_features) | |
7426 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
7427 | else |
7428 | err = 0; | |
5455c699 | 7429 | |
6cb6a27c | 7430 | if (unlikely(err < 0)) { |
5455c699 | 7431 | netdev_err(dev, |
c8f44aff MM |
7432 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
7433 | err, &features, &dev->features); | |
17b85d29 NA |
7434 | /* return non-0 since some features might have changed and |
7435 | * it's better to fire a spurious notification than miss it | |
7436 | */ | |
7437 | return -1; | |
6cb6a27c MM |
7438 | } |
7439 | ||
e7868a85 | 7440 | sync_lower: |
fd867d51 JW |
7441 | /* some features must be disabled on lower devices when disabled |
7442 | * on an upper device (think: bonding master or bridge) | |
7443 | */ | |
7444 | netdev_for_each_lower_dev(dev, lower, iter) | |
7445 | netdev_sync_lower_features(dev, lower, features); | |
7446 | ||
ae847f40 SD |
7447 | if (!err) { |
7448 | netdev_features_t diff = features ^ dev->features; | |
7449 | ||
7450 | if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
7451 | /* udp_tunnel_{get,drop}_rx_info both need | |
7452 | * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the | |
7453 | * device, or they won't do anything. | |
7454 | * Thus we need to update dev->features | |
7455 | * *before* calling udp_tunnel_get_rx_info, | |
7456 | * but *after* calling udp_tunnel_drop_rx_info. | |
7457 | */ | |
7458 | if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
7459 | dev->features = features; | |
7460 | udp_tunnel_get_rx_info(dev); | |
7461 | } else { | |
7462 | udp_tunnel_drop_rx_info(dev); | |
7463 | } | |
7464 | } | |
7465 | ||
6cb6a27c | 7466 | dev->features = features; |
ae847f40 | 7467 | } |
6cb6a27c | 7468 | |
e7868a85 | 7469 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
7470 | } |
7471 | ||
afe12cc8 MM |
7472 | /** |
7473 | * netdev_update_features - recalculate device features | |
7474 | * @dev: the device to check | |
7475 | * | |
7476 | * Recalculate dev->features set and send notifications if it | |
7477 | * has changed. Should be called after driver or hardware dependent | |
7478 | * conditions might have changed that influence the features. | |
7479 | */ | |
6cb6a27c MM |
7480 | void netdev_update_features(struct net_device *dev) |
7481 | { | |
7482 | if (__netdev_update_features(dev)) | |
7483 | netdev_features_change(dev); | |
5455c699 MM |
7484 | } |
7485 | EXPORT_SYMBOL(netdev_update_features); | |
7486 | ||
afe12cc8 MM |
7487 | /** |
7488 | * netdev_change_features - recalculate device features | |
7489 | * @dev: the device to check | |
7490 | * | |
7491 | * Recalculate dev->features set and send notifications even | |
7492 | * if they have not changed. Should be called instead of | |
7493 | * netdev_update_features() if also dev->vlan_features might | |
7494 | * have changed to allow the changes to be propagated to stacked | |
7495 | * VLAN devices. | |
7496 | */ | |
7497 | void netdev_change_features(struct net_device *dev) | |
7498 | { | |
7499 | __netdev_update_features(dev); | |
7500 | netdev_features_change(dev); | |
7501 | } | |
7502 | EXPORT_SYMBOL(netdev_change_features); | |
7503 | ||
fc4a7489 PM |
7504 | /** |
7505 | * netif_stacked_transfer_operstate - transfer operstate | |
7506 | * @rootdev: the root or lower level device to transfer state from | |
7507 | * @dev: the device to transfer operstate to | |
7508 | * | |
7509 | * Transfer operational state from root to device. This is normally | |
7510 | * called when a stacking relationship exists between the root | |
7511 | * device and the device(a leaf device). | |
7512 | */ | |
7513 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
7514 | struct net_device *dev) | |
7515 | { | |
7516 | if (rootdev->operstate == IF_OPER_DORMANT) | |
7517 | netif_dormant_on(dev); | |
7518 | else | |
7519 | netif_dormant_off(dev); | |
7520 | ||
0575c86b ZS |
7521 | if (netif_carrier_ok(rootdev)) |
7522 | netif_carrier_on(dev); | |
7523 | else | |
7524 | netif_carrier_off(dev); | |
fc4a7489 PM |
7525 | } |
7526 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
7527 | ||
a953be53 | 7528 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
7529 | static int netif_alloc_rx_queues(struct net_device *dev) |
7530 | { | |
1b4bf461 | 7531 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 7532 | struct netdev_rx_queue *rx; |
10595902 | 7533 | size_t sz = count * sizeof(*rx); |
1b4bf461 | 7534 | |
bd25fa7b | 7535 | BUG_ON(count < 1); |
1b4bf461 | 7536 | |
dcda9b04 | 7537 | rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
7538 | if (!rx) |
7539 | return -ENOMEM; | |
7540 | ||
bd25fa7b TH |
7541 | dev->_rx = rx; |
7542 | ||
bd25fa7b | 7543 | for (i = 0; i < count; i++) |
fe822240 | 7544 | rx[i].dev = dev; |
1b4bf461 ED |
7545 | return 0; |
7546 | } | |
bf264145 | 7547 | #endif |
1b4bf461 | 7548 | |
aa942104 CG |
7549 | static void netdev_init_one_queue(struct net_device *dev, |
7550 | struct netdev_queue *queue, void *_unused) | |
7551 | { | |
7552 | /* Initialize queue lock */ | |
7553 | spin_lock_init(&queue->_xmit_lock); | |
7554 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
7555 | queue->xmit_lock_owner = -1; | |
b236da69 | 7556 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 7557 | queue->dev = dev; |
114cf580 TH |
7558 | #ifdef CONFIG_BQL |
7559 | dql_init(&queue->dql, HZ); | |
7560 | #endif | |
aa942104 CG |
7561 | } |
7562 | ||
60877a32 ED |
7563 | static void netif_free_tx_queues(struct net_device *dev) |
7564 | { | |
4cb28970 | 7565 | kvfree(dev->_tx); |
60877a32 ED |
7566 | } |
7567 | ||
e6484930 TH |
7568 | static int netif_alloc_netdev_queues(struct net_device *dev) |
7569 | { | |
7570 | unsigned int count = dev->num_tx_queues; | |
7571 | struct netdev_queue *tx; | |
60877a32 | 7572 | size_t sz = count * sizeof(*tx); |
e6484930 | 7573 | |
d339727c ED |
7574 | if (count < 1 || count > 0xffff) |
7575 | return -EINVAL; | |
62b5942a | 7576 | |
dcda9b04 | 7577 | tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
7578 | if (!tx) |
7579 | return -ENOMEM; | |
7580 | ||
e6484930 | 7581 | dev->_tx = tx; |
1d24eb48 | 7582 | |
e6484930 TH |
7583 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
7584 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
7585 | |
7586 | return 0; | |
e6484930 TH |
7587 | } |
7588 | ||
a2029240 DV |
7589 | void netif_tx_stop_all_queues(struct net_device *dev) |
7590 | { | |
7591 | unsigned int i; | |
7592 | ||
7593 | for (i = 0; i < dev->num_tx_queues; i++) { | |
7594 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 7595 | |
a2029240 DV |
7596 | netif_tx_stop_queue(txq); |
7597 | } | |
7598 | } | |
7599 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
7600 | ||
1da177e4 LT |
7601 | /** |
7602 | * register_netdevice - register a network device | |
7603 | * @dev: device to register | |
7604 | * | |
7605 | * Take a completed network device structure and add it to the kernel | |
7606 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7607 | * chain. 0 is returned on success. A negative errno code is returned | |
7608 | * on a failure to set up the device, or if the name is a duplicate. | |
7609 | * | |
7610 | * Callers must hold the rtnl semaphore. You may want | |
7611 | * register_netdev() instead of this. | |
7612 | * | |
7613 | * BUGS: | |
7614 | * The locking appears insufficient to guarantee two parallel registers | |
7615 | * will not get the same name. | |
7616 | */ | |
7617 | ||
7618 | int register_netdevice(struct net_device *dev) | |
7619 | { | |
1da177e4 | 7620 | int ret; |
d314774c | 7621 | struct net *net = dev_net(dev); |
1da177e4 LT |
7622 | |
7623 | BUG_ON(dev_boot_phase); | |
7624 | ASSERT_RTNL(); | |
7625 | ||
b17a7c17 SH |
7626 | might_sleep(); |
7627 | ||
1da177e4 LT |
7628 | /* When net_device's are persistent, this will be fatal. */ |
7629 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 7630 | BUG_ON(!net); |
1da177e4 | 7631 | |
f1f28aa3 | 7632 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 7633 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 7634 | |
828de4f6 | 7635 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
7636 | if (ret < 0) |
7637 | goto out; | |
7638 | ||
1da177e4 | 7639 | /* Init, if this function is available */ |
d314774c SH |
7640 | if (dev->netdev_ops->ndo_init) { |
7641 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
7642 | if (ret) { |
7643 | if (ret > 0) | |
7644 | ret = -EIO; | |
90833aa4 | 7645 | goto out; |
1da177e4 LT |
7646 | } |
7647 | } | |
4ec93edb | 7648 | |
f646968f PM |
7649 | if (((dev->hw_features | dev->features) & |
7650 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
7651 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
7652 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
7653 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
7654 | ret = -EINVAL; | |
7655 | goto err_uninit; | |
7656 | } | |
7657 | ||
9c7dafbf PE |
7658 | ret = -EBUSY; |
7659 | if (!dev->ifindex) | |
7660 | dev->ifindex = dev_new_index(net); | |
7661 | else if (__dev_get_by_index(net, dev->ifindex)) | |
7662 | goto err_uninit; | |
7663 | ||
5455c699 MM |
7664 | /* Transfer changeable features to wanted_features and enable |
7665 | * software offloads (GSO and GRO). | |
7666 | */ | |
7667 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f | 7668 | dev->features |= NETIF_F_SOFT_FEATURES; |
d764a122 SD |
7669 | |
7670 | if (dev->netdev_ops->ndo_udp_tunnel_add) { | |
7671 | dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
7672 | dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
7673 | } | |
7674 | ||
14d1232f | 7675 | dev->wanted_features = dev->features & dev->hw_features; |
1da177e4 | 7676 | |
cbc53e08 | 7677 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 7678 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 7679 | |
7f348a60 AD |
7680 | /* If IPv4 TCP segmentation offload is supported we should also |
7681 | * allow the device to enable segmenting the frame with the option | |
7682 | * of ignoring a static IP ID value. This doesn't enable the | |
7683 | * feature itself but allows the user to enable it later. | |
7684 | */ | |
cbc53e08 AD |
7685 | if (dev->hw_features & NETIF_F_TSO) |
7686 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
7687 | if (dev->vlan_features & NETIF_F_TSO) |
7688 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
7689 | if (dev->mpls_features & NETIF_F_TSO) | |
7690 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
7691 | if (dev->hw_enc_features & NETIF_F_TSO) | |
7692 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 7693 | |
1180e7d6 | 7694 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 7695 | */ |
1180e7d6 | 7696 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 7697 | |
ee579677 PS |
7698 | /* Make NETIF_F_SG inheritable to tunnel devices. |
7699 | */ | |
802ab55a | 7700 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 7701 | |
0d89d203 SH |
7702 | /* Make NETIF_F_SG inheritable to MPLS. |
7703 | */ | |
7704 | dev->mpls_features |= NETIF_F_SG; | |
7705 | ||
7ffbe3fd JB |
7706 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
7707 | ret = notifier_to_errno(ret); | |
7708 | if (ret) | |
7709 | goto err_uninit; | |
7710 | ||
8b41d188 | 7711 | ret = netdev_register_kobject(dev); |
b17a7c17 | 7712 | if (ret) |
7ce1b0ed | 7713 | goto err_uninit; |
b17a7c17 SH |
7714 | dev->reg_state = NETREG_REGISTERED; |
7715 | ||
6cb6a27c | 7716 | __netdev_update_features(dev); |
8e9b59b2 | 7717 | |
1da177e4 LT |
7718 | /* |
7719 | * Default initial state at registry is that the | |
7720 | * device is present. | |
7721 | */ | |
7722 | ||
7723 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7724 | ||
8f4cccbb BH |
7725 | linkwatch_init_dev(dev); |
7726 | ||
1da177e4 | 7727 | dev_init_scheduler(dev); |
1da177e4 | 7728 | dev_hold(dev); |
ce286d32 | 7729 | list_netdevice(dev); |
7bf23575 | 7730 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 7731 | |
948b337e JP |
7732 | /* If the device has permanent device address, driver should |
7733 | * set dev_addr and also addr_assign_type should be set to | |
7734 | * NET_ADDR_PERM (default value). | |
7735 | */ | |
7736 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
7737 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
7738 | ||
1da177e4 | 7739 | /* Notify protocols, that a new device appeared. */ |
056925ab | 7740 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 7741 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
7742 | if (ret) { |
7743 | rollback_registered(dev); | |
7744 | dev->reg_state = NETREG_UNREGISTERED; | |
7745 | } | |
d90a909e EB |
7746 | /* |
7747 | * Prevent userspace races by waiting until the network | |
7748 | * device is fully setup before sending notifications. | |
7749 | */ | |
a2835763 PM |
7750 | if (!dev->rtnl_link_ops || |
7751 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 7752 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
7753 | |
7754 | out: | |
7755 | return ret; | |
7ce1b0ed HX |
7756 | |
7757 | err_uninit: | |
d314774c SH |
7758 | if (dev->netdev_ops->ndo_uninit) |
7759 | dev->netdev_ops->ndo_uninit(dev); | |
cf124db5 DM |
7760 | if (dev->priv_destructor) |
7761 | dev->priv_destructor(dev); | |
7ce1b0ed | 7762 | goto out; |
1da177e4 | 7763 | } |
d1b19dff | 7764 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 7765 | |
937f1ba5 BH |
7766 | /** |
7767 | * init_dummy_netdev - init a dummy network device for NAPI | |
7768 | * @dev: device to init | |
7769 | * | |
7770 | * This takes a network device structure and initialize the minimum | |
7771 | * amount of fields so it can be used to schedule NAPI polls without | |
7772 | * registering a full blown interface. This is to be used by drivers | |
7773 | * that need to tie several hardware interfaces to a single NAPI | |
7774 | * poll scheduler due to HW limitations. | |
7775 | */ | |
7776 | int init_dummy_netdev(struct net_device *dev) | |
7777 | { | |
7778 | /* Clear everything. Note we don't initialize spinlocks | |
7779 | * are they aren't supposed to be taken by any of the | |
7780 | * NAPI code and this dummy netdev is supposed to be | |
7781 | * only ever used for NAPI polls | |
7782 | */ | |
7783 | memset(dev, 0, sizeof(struct net_device)); | |
7784 | ||
7785 | /* make sure we BUG if trying to hit standard | |
7786 | * register/unregister code path | |
7787 | */ | |
7788 | dev->reg_state = NETREG_DUMMY; | |
7789 | ||
937f1ba5 BH |
7790 | /* NAPI wants this */ |
7791 | INIT_LIST_HEAD(&dev->napi_list); | |
7792 | ||
7793 | /* a dummy interface is started by default */ | |
7794 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7795 | set_bit(__LINK_STATE_START, &dev->state); | |
7796 | ||
29b4433d ED |
7797 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
7798 | * because users of this 'device' dont need to change | |
7799 | * its refcount. | |
7800 | */ | |
7801 | ||
937f1ba5 BH |
7802 | return 0; |
7803 | } | |
7804 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
7805 | ||
7806 | ||
1da177e4 LT |
7807 | /** |
7808 | * register_netdev - register a network device | |
7809 | * @dev: device to register | |
7810 | * | |
7811 | * Take a completed network device structure and add it to the kernel | |
7812 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7813 | * chain. 0 is returned on success. A negative errno code is returned | |
7814 | * on a failure to set up the device, or if the name is a duplicate. | |
7815 | * | |
38b4da38 | 7816 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
7817 | * and expands the device name if you passed a format string to |
7818 | * alloc_netdev. | |
7819 | */ | |
7820 | int register_netdev(struct net_device *dev) | |
7821 | { | |
7822 | int err; | |
7823 | ||
7824 | rtnl_lock(); | |
1da177e4 | 7825 | err = register_netdevice(dev); |
1da177e4 LT |
7826 | rtnl_unlock(); |
7827 | return err; | |
7828 | } | |
7829 | EXPORT_SYMBOL(register_netdev); | |
7830 | ||
29b4433d ED |
7831 | int netdev_refcnt_read(const struct net_device *dev) |
7832 | { | |
7833 | int i, refcnt = 0; | |
7834 | ||
7835 | for_each_possible_cpu(i) | |
7836 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
7837 | return refcnt; | |
7838 | } | |
7839 | EXPORT_SYMBOL(netdev_refcnt_read); | |
7840 | ||
2c53040f | 7841 | /** |
1da177e4 | 7842 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 7843 | * @dev: target net_device |
1da177e4 LT |
7844 | * |
7845 | * This is called when unregistering network devices. | |
7846 | * | |
7847 | * Any protocol or device that holds a reference should register | |
7848 | * for netdevice notification, and cleanup and put back the | |
7849 | * reference if they receive an UNREGISTER event. | |
7850 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 7851 | * call dev_put. |
1da177e4 LT |
7852 | */ |
7853 | static void netdev_wait_allrefs(struct net_device *dev) | |
7854 | { | |
7855 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 7856 | int refcnt; |
1da177e4 | 7857 | |
e014debe ED |
7858 | linkwatch_forget_dev(dev); |
7859 | ||
1da177e4 | 7860 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
7861 | refcnt = netdev_refcnt_read(dev); |
7862 | ||
7863 | while (refcnt != 0) { | |
1da177e4 | 7864 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 7865 | rtnl_lock(); |
1da177e4 LT |
7866 | |
7867 | /* Rebroadcast unregister notification */ | |
056925ab | 7868 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 7869 | |
748e2d93 | 7870 | __rtnl_unlock(); |
0115e8e3 | 7871 | rcu_barrier(); |
748e2d93 ED |
7872 | rtnl_lock(); |
7873 | ||
0115e8e3 | 7874 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
7875 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
7876 | &dev->state)) { | |
7877 | /* We must not have linkwatch events | |
7878 | * pending on unregister. If this | |
7879 | * happens, we simply run the queue | |
7880 | * unscheduled, resulting in a noop | |
7881 | * for this device. | |
7882 | */ | |
7883 | linkwatch_run_queue(); | |
7884 | } | |
7885 | ||
6756ae4b | 7886 | __rtnl_unlock(); |
1da177e4 LT |
7887 | |
7888 | rebroadcast_time = jiffies; | |
7889 | } | |
7890 | ||
7891 | msleep(250); | |
7892 | ||
29b4433d ED |
7893 | refcnt = netdev_refcnt_read(dev); |
7894 | ||
1da177e4 | 7895 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
7896 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
7897 | dev->name, refcnt); | |
1da177e4 LT |
7898 | warning_time = jiffies; |
7899 | } | |
7900 | } | |
7901 | } | |
7902 | ||
7903 | /* The sequence is: | |
7904 | * | |
7905 | * rtnl_lock(); | |
7906 | * ... | |
7907 | * register_netdevice(x1); | |
7908 | * register_netdevice(x2); | |
7909 | * ... | |
7910 | * unregister_netdevice(y1); | |
7911 | * unregister_netdevice(y2); | |
7912 | * ... | |
7913 | * rtnl_unlock(); | |
7914 | * free_netdev(y1); | |
7915 | * free_netdev(y2); | |
7916 | * | |
58ec3b4d | 7917 | * We are invoked by rtnl_unlock(). |
1da177e4 | 7918 | * This allows us to deal with problems: |
b17a7c17 | 7919 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
7920 | * without deadlocking with linkwatch via keventd. |
7921 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
7922 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
7923 | * |
7924 | * We must not return until all unregister events added during | |
7925 | * the interval the lock was held have been completed. | |
1da177e4 | 7926 | */ |
1da177e4 LT |
7927 | void netdev_run_todo(void) |
7928 | { | |
626ab0e6 | 7929 | struct list_head list; |
1da177e4 | 7930 | |
1da177e4 | 7931 | /* Snapshot list, allow later requests */ |
626ab0e6 | 7932 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
7933 | |
7934 | __rtnl_unlock(); | |
626ab0e6 | 7935 | |
0115e8e3 ED |
7936 | |
7937 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
7938 | if (!list_empty(&list)) |
7939 | rcu_barrier(); | |
7940 | ||
1da177e4 LT |
7941 | while (!list_empty(&list)) { |
7942 | struct net_device *dev | |
e5e26d75 | 7943 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
7944 | list_del(&dev->todo_list); |
7945 | ||
748e2d93 | 7946 | rtnl_lock(); |
0115e8e3 | 7947 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 7948 | __rtnl_unlock(); |
0115e8e3 | 7949 | |
b17a7c17 | 7950 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 7951 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
7952 | dev->name, dev->reg_state); |
7953 | dump_stack(); | |
7954 | continue; | |
7955 | } | |
1da177e4 | 7956 | |
b17a7c17 | 7957 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 7958 | |
b17a7c17 | 7959 | netdev_wait_allrefs(dev); |
1da177e4 | 7960 | |
b17a7c17 | 7961 | /* paranoia */ |
29b4433d | 7962 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
7963 | BUG_ON(!list_empty(&dev->ptype_all)); |
7964 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
7965 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
7966 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 7967 | WARN_ON(dev->dn_ptr); |
1da177e4 | 7968 | |
cf124db5 DM |
7969 | if (dev->priv_destructor) |
7970 | dev->priv_destructor(dev); | |
7971 | if (dev->needs_free_netdev) | |
7972 | free_netdev(dev); | |
9093bbb2 | 7973 | |
50624c93 EB |
7974 | /* Report a network device has been unregistered */ |
7975 | rtnl_lock(); | |
7976 | dev_net(dev)->dev_unreg_count--; | |
7977 | __rtnl_unlock(); | |
7978 | wake_up(&netdev_unregistering_wq); | |
7979 | ||
9093bbb2 SH |
7980 | /* Free network device */ |
7981 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 7982 | } |
1da177e4 LT |
7983 | } |
7984 | ||
9256645a JW |
7985 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
7986 | * all the same fields in the same order as net_device_stats, with only | |
7987 | * the type differing, but rtnl_link_stats64 may have additional fields | |
7988 | * at the end for newer counters. | |
3cfde79c | 7989 | */ |
77a1abf5 ED |
7990 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
7991 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
7992 | { |
7993 | #if BITS_PER_LONG == 64 | |
9256645a | 7994 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
9af9959e | 7995 | memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); |
9256645a JW |
7996 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7997 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
7998 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 7999 | #else |
9256645a | 8000 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
8001 | const unsigned long *src = (const unsigned long *)netdev_stats; |
8002 | u64 *dst = (u64 *)stats64; | |
8003 | ||
9256645a | 8004 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
8005 | for (i = 0; i < n; i++) |
8006 | dst[i] = src[i]; | |
9256645a JW |
8007 | /* zero out counters that only exist in rtnl_link_stats64 */ |
8008 | memset((char *)stats64 + n * sizeof(u64), 0, | |
8009 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
8010 | #endif |
8011 | } | |
77a1abf5 | 8012 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 8013 | |
eeda3fd6 SH |
8014 | /** |
8015 | * dev_get_stats - get network device statistics | |
8016 | * @dev: device to get statistics from | |
28172739 | 8017 | * @storage: place to store stats |
eeda3fd6 | 8018 | * |
d7753516 BH |
8019 | * Get network statistics from device. Return @storage. |
8020 | * The device driver may provide its own method by setting | |
8021 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
8022 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 8023 | */ |
d7753516 BH |
8024 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
8025 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 8026 | { |
eeda3fd6 SH |
8027 | const struct net_device_ops *ops = dev->netdev_ops; |
8028 | ||
28172739 ED |
8029 | if (ops->ndo_get_stats64) { |
8030 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
8031 | ops->ndo_get_stats64(dev, storage); |
8032 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 8033 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
8034 | } else { |
8035 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 8036 | } |
6f64ec74 ED |
8037 | storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); |
8038 | storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); | |
8039 | storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); | |
28172739 | 8040 | return storage; |
c45d286e | 8041 | } |
eeda3fd6 | 8042 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 8043 | |
24824a09 | 8044 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 8045 | { |
24824a09 | 8046 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 8047 | |
24824a09 ED |
8048 | #ifdef CONFIG_NET_CLS_ACT |
8049 | if (queue) | |
8050 | return queue; | |
8051 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
8052 | if (!queue) | |
8053 | return NULL; | |
8054 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 8055 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
8056 | queue->qdisc_sleeping = &noop_qdisc; |
8057 | rcu_assign_pointer(dev->ingress_queue, queue); | |
8058 | #endif | |
8059 | return queue; | |
bb949fbd DM |
8060 | } |
8061 | ||
2c60db03 ED |
8062 | static const struct ethtool_ops default_ethtool_ops; |
8063 | ||
d07d7507 SG |
8064 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
8065 | const struct ethtool_ops *ops) | |
8066 | { | |
8067 | if (dev->ethtool_ops == &default_ethtool_ops) | |
8068 | dev->ethtool_ops = ops; | |
8069 | } | |
8070 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
8071 | ||
74d332c1 ED |
8072 | void netdev_freemem(struct net_device *dev) |
8073 | { | |
8074 | char *addr = (char *)dev - dev->padded; | |
8075 | ||
4cb28970 | 8076 | kvfree(addr); |
74d332c1 ED |
8077 | } |
8078 | ||
1da177e4 | 8079 | /** |
722c9a0c | 8080 | * alloc_netdev_mqs - allocate network device |
8081 | * @sizeof_priv: size of private data to allocate space for | |
8082 | * @name: device name format string | |
8083 | * @name_assign_type: origin of device name | |
8084 | * @setup: callback to initialize device | |
8085 | * @txqs: the number of TX subqueues to allocate | |
8086 | * @rxqs: the number of RX subqueues to allocate | |
8087 | * | |
8088 | * Allocates a struct net_device with private data area for driver use | |
8089 | * and performs basic initialization. Also allocates subqueue structs | |
8090 | * for each queue on the device. | |
1da177e4 | 8091 | */ |
36909ea4 | 8092 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 8093 | unsigned char name_assign_type, |
36909ea4 TH |
8094 | void (*setup)(struct net_device *), |
8095 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 8096 | { |
1da177e4 | 8097 | struct net_device *dev; |
52a59bd5 | 8098 | unsigned int alloc_size; |
1ce8e7b5 | 8099 | struct net_device *p; |
1da177e4 | 8100 | |
b6fe17d6 SH |
8101 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
8102 | ||
36909ea4 | 8103 | if (txqs < 1) { |
7b6cd1ce | 8104 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
8105 | return NULL; |
8106 | } | |
8107 | ||
a953be53 | 8108 | #ifdef CONFIG_SYSFS |
36909ea4 | 8109 | if (rxqs < 1) { |
7b6cd1ce | 8110 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
8111 | return NULL; |
8112 | } | |
8113 | #endif | |
8114 | ||
fd2ea0a7 | 8115 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
8116 | if (sizeof_priv) { |
8117 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 8118 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
8119 | alloc_size += sizeof_priv; |
8120 | } | |
8121 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 8122 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 8123 | |
dcda9b04 | 8124 | p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
62b5942a | 8125 | if (!p) |
1da177e4 | 8126 | return NULL; |
1da177e4 | 8127 | |
1ce8e7b5 | 8128 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 8129 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 8130 | |
29b4433d ED |
8131 | dev->pcpu_refcnt = alloc_percpu(int); |
8132 | if (!dev->pcpu_refcnt) | |
74d332c1 | 8133 | goto free_dev; |
ab9c73cc | 8134 | |
ab9c73cc | 8135 | if (dev_addr_init(dev)) |
29b4433d | 8136 | goto free_pcpu; |
ab9c73cc | 8137 | |
22bedad3 | 8138 | dev_mc_init(dev); |
a748ee24 | 8139 | dev_uc_init(dev); |
ccffad25 | 8140 | |
c346dca1 | 8141 | dev_net_set(dev, &init_net); |
1da177e4 | 8142 | |
8d3bdbd5 | 8143 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 8144 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 8145 | |
8d3bdbd5 DM |
8146 | INIT_LIST_HEAD(&dev->napi_list); |
8147 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 8148 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 8149 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
8150 | INIT_LIST_HEAD(&dev->adj_list.upper); |
8151 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
8152 | INIT_LIST_HEAD(&dev->ptype_all); |
8153 | INIT_LIST_HEAD(&dev->ptype_specific); | |
59cc1f61 JK |
8154 | #ifdef CONFIG_NET_SCHED |
8155 | hash_init(dev->qdisc_hash); | |
8156 | #endif | |
02875878 | 8157 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
8158 | setup(dev); |
8159 | ||
a813104d | 8160 | if (!dev->tx_queue_len) { |
f84bb1ea | 8161 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 8162 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 8163 | } |
906470c1 | 8164 | |
36909ea4 TH |
8165 | dev->num_tx_queues = txqs; |
8166 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 8167 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 8168 | goto free_all; |
e8a0464c | 8169 | |
a953be53 | 8170 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
8171 | dev->num_rx_queues = rxqs; |
8172 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 8173 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 8174 | goto free_all; |
df334545 | 8175 | #endif |
0a9627f2 | 8176 | |
1da177e4 | 8177 | strcpy(dev->name, name); |
c835a677 | 8178 | dev->name_assign_type = name_assign_type; |
cbda10fa | 8179 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
8180 | if (!dev->ethtool_ops) |
8181 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 PN |
8182 | |
8183 | nf_hook_ingress_init(dev); | |
8184 | ||
1da177e4 | 8185 | return dev; |
ab9c73cc | 8186 | |
8d3bdbd5 DM |
8187 | free_all: |
8188 | free_netdev(dev); | |
8189 | return NULL; | |
8190 | ||
29b4433d ED |
8191 | free_pcpu: |
8192 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
8193 | free_dev: |
8194 | netdev_freemem(dev); | |
ab9c73cc | 8195 | return NULL; |
1da177e4 | 8196 | } |
36909ea4 | 8197 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
8198 | |
8199 | /** | |
722c9a0c | 8200 | * free_netdev - free network device |
8201 | * @dev: device | |
1da177e4 | 8202 | * |
722c9a0c | 8203 | * This function does the last stage of destroying an allocated device |
8204 | * interface. The reference to the device object is released. If this | |
8205 | * is the last reference then it will be freed.Must be called in process | |
8206 | * context. | |
1da177e4 LT |
8207 | */ |
8208 | void free_netdev(struct net_device *dev) | |
8209 | { | |
d565b0a1 | 8210 | struct napi_struct *p, *n; |
b5cdae32 | 8211 | struct bpf_prog *prog; |
d565b0a1 | 8212 | |
93d05d4a | 8213 | might_sleep(); |
60877a32 | 8214 | netif_free_tx_queues(dev); |
a953be53 | 8215 | #ifdef CONFIG_SYSFS |
10595902 | 8216 | kvfree(dev->_rx); |
fe822240 | 8217 | #endif |
e8a0464c | 8218 | |
33d480ce | 8219 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 8220 | |
f001fde5 JP |
8221 | /* Flush device addresses */ |
8222 | dev_addr_flush(dev); | |
8223 | ||
d565b0a1 HX |
8224 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
8225 | netif_napi_del(p); | |
8226 | ||
29b4433d ED |
8227 | free_percpu(dev->pcpu_refcnt); |
8228 | dev->pcpu_refcnt = NULL; | |
8229 | ||
b5cdae32 DM |
8230 | prog = rcu_dereference_protected(dev->xdp_prog, 1); |
8231 | if (prog) { | |
8232 | bpf_prog_put(prog); | |
8233 | static_key_slow_dec(&generic_xdp_needed); | |
8234 | } | |
8235 | ||
3041a069 | 8236 | /* Compatibility with error handling in drivers */ |
1da177e4 | 8237 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 8238 | netdev_freemem(dev); |
1da177e4 LT |
8239 | return; |
8240 | } | |
8241 | ||
8242 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
8243 | dev->reg_state = NETREG_RELEASED; | |
8244 | ||
43cb76d9 GKH |
8245 | /* will free via device release */ |
8246 | put_device(&dev->dev); | |
1da177e4 | 8247 | } |
d1b19dff | 8248 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 8249 | |
f0db275a SH |
8250 | /** |
8251 | * synchronize_net - Synchronize with packet receive processing | |
8252 | * | |
8253 | * Wait for packets currently being received to be done. | |
8254 | * Does not block later packets from starting. | |
8255 | */ | |
4ec93edb | 8256 | void synchronize_net(void) |
1da177e4 LT |
8257 | { |
8258 | might_sleep(); | |
be3fc413 ED |
8259 | if (rtnl_is_locked()) |
8260 | synchronize_rcu_expedited(); | |
8261 | else | |
8262 | synchronize_rcu(); | |
1da177e4 | 8263 | } |
d1b19dff | 8264 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
8265 | |
8266 | /** | |
44a0873d | 8267 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 8268 | * @dev: device |
44a0873d | 8269 | * @head: list |
6ebfbc06 | 8270 | * |
1da177e4 | 8271 | * This function shuts down a device interface and removes it |
d59b54b1 | 8272 | * from the kernel tables. |
44a0873d | 8273 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
8274 | * |
8275 | * Callers must hold the rtnl semaphore. You may want | |
8276 | * unregister_netdev() instead of this. | |
8277 | */ | |
8278 | ||
44a0873d | 8279 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 8280 | { |
a6620712 HX |
8281 | ASSERT_RTNL(); |
8282 | ||
44a0873d | 8283 | if (head) { |
9fdce099 | 8284 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
8285 | } else { |
8286 | rollback_registered(dev); | |
8287 | /* Finish processing unregister after unlock */ | |
8288 | net_set_todo(dev); | |
8289 | } | |
1da177e4 | 8290 | } |
44a0873d | 8291 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 8292 | |
9b5e383c ED |
8293 | /** |
8294 | * unregister_netdevice_many - unregister many devices | |
8295 | * @head: list of devices | |
87757a91 ED |
8296 | * |
8297 | * Note: As most callers use a stack allocated list_head, | |
8298 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
8299 | */ |
8300 | void unregister_netdevice_many(struct list_head *head) | |
8301 | { | |
8302 | struct net_device *dev; | |
8303 | ||
8304 | if (!list_empty(head)) { | |
8305 | rollback_registered_many(head); | |
8306 | list_for_each_entry(dev, head, unreg_list) | |
8307 | net_set_todo(dev); | |
87757a91 | 8308 | list_del(head); |
9b5e383c ED |
8309 | } |
8310 | } | |
63c8099d | 8311 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 8312 | |
1da177e4 LT |
8313 | /** |
8314 | * unregister_netdev - remove device from the kernel | |
8315 | * @dev: device | |
8316 | * | |
8317 | * This function shuts down a device interface and removes it | |
d59b54b1 | 8318 | * from the kernel tables. |
1da177e4 LT |
8319 | * |
8320 | * This is just a wrapper for unregister_netdevice that takes | |
8321 | * the rtnl semaphore. In general you want to use this and not | |
8322 | * unregister_netdevice. | |
8323 | */ | |
8324 | void unregister_netdev(struct net_device *dev) | |
8325 | { | |
8326 | rtnl_lock(); | |
8327 | unregister_netdevice(dev); | |
8328 | rtnl_unlock(); | |
8329 | } | |
1da177e4 LT |
8330 | EXPORT_SYMBOL(unregister_netdev); |
8331 | ||
ce286d32 EB |
8332 | /** |
8333 | * dev_change_net_namespace - move device to different nethost namespace | |
8334 | * @dev: device | |
8335 | * @net: network namespace | |
8336 | * @pat: If not NULL name pattern to try if the current device name | |
8337 | * is already taken in the destination network namespace. | |
8338 | * | |
8339 | * This function shuts down a device interface and moves it | |
8340 | * to a new network namespace. On success 0 is returned, on | |
8341 | * a failure a netagive errno code is returned. | |
8342 | * | |
8343 | * Callers must hold the rtnl semaphore. | |
8344 | */ | |
8345 | ||
8346 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
8347 | { | |
6621dd29 | 8348 | int err, new_nsid; |
ce286d32 EB |
8349 | |
8350 | ASSERT_RTNL(); | |
8351 | ||
8352 | /* Don't allow namespace local devices to be moved. */ | |
8353 | err = -EINVAL; | |
8354 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8355 | goto out; | |
8356 | ||
8357 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
8358 | if (dev->reg_state != NETREG_REGISTERED) |
8359 | goto out; | |
8360 | ||
8361 | /* Get out if there is nothing todo */ | |
8362 | err = 0; | |
878628fb | 8363 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
8364 | goto out; |
8365 | ||
8366 | /* Pick the destination device name, and ensure | |
8367 | * we can use it in the destination network namespace. | |
8368 | */ | |
8369 | err = -EEXIST; | |
d9031024 | 8370 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
8371 | /* We get here if we can't use the current device name */ |
8372 | if (!pat) | |
8373 | goto out; | |
828de4f6 | 8374 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
8375 | goto out; |
8376 | } | |
8377 | ||
8378 | /* | |
8379 | * And now a mini version of register_netdevice unregister_netdevice. | |
8380 | */ | |
8381 | ||
8382 | /* If device is running close it first. */ | |
9b772652 | 8383 | dev_close(dev); |
ce286d32 EB |
8384 | |
8385 | /* And unlink it from device chain */ | |
8386 | err = -ENODEV; | |
8387 | unlist_netdevice(dev); | |
8388 | ||
8389 | synchronize_net(); | |
8390 | ||
8391 | /* Shutdown queueing discipline. */ | |
8392 | dev_shutdown(dev); | |
8393 | ||
8394 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 8395 | * this device. They should clean all the things. |
8396 | * | |
8397 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
8398 | * This is wanted because this way 8021q and macvlan know | |
8399 | * the device is just moving and can keep their slaves up. | |
8400 | */ | |
ce286d32 | 8401 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 G |
8402 | rcu_barrier(); |
8403 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
6621dd29 ND |
8404 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->get_link_net) |
8405 | new_nsid = peernet2id_alloc(dev_net(dev), net); | |
8406 | else | |
8407 | new_nsid = peernet2id(dev_net(dev), net); | |
8408 | rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid); | |
ce286d32 EB |
8409 | |
8410 | /* | |
8411 | * Flush the unicast and multicast chains | |
8412 | */ | |
a748ee24 | 8413 | dev_uc_flush(dev); |
22bedad3 | 8414 | dev_mc_flush(dev); |
ce286d32 | 8415 | |
4e66ae2e SH |
8416 | /* Send a netdev-removed uevent to the old namespace */ |
8417 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 8418 | netdev_adjacent_del_links(dev); |
4e66ae2e | 8419 | |
ce286d32 | 8420 | /* Actually switch the network namespace */ |
c346dca1 | 8421 | dev_net_set(dev, net); |
ce286d32 | 8422 | |
ce286d32 | 8423 | /* If there is an ifindex conflict assign a new one */ |
7a66bbc9 | 8424 | if (__dev_get_by_index(net, dev->ifindex)) |
ce286d32 | 8425 | dev->ifindex = dev_new_index(net); |
ce286d32 | 8426 | |
4e66ae2e SH |
8427 | /* Send a netdev-add uevent to the new namespace */ |
8428 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 8429 | netdev_adjacent_add_links(dev); |
4e66ae2e | 8430 | |
8b41d188 | 8431 | /* Fixup kobjects */ |
a1b3f594 | 8432 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 8433 | WARN_ON(err); |
ce286d32 EB |
8434 | |
8435 | /* Add the device back in the hashes */ | |
8436 | list_netdevice(dev); | |
8437 | ||
8438 | /* Notify protocols, that a new device appeared. */ | |
8439 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
8440 | ||
d90a909e EB |
8441 | /* |
8442 | * Prevent userspace races by waiting until the network | |
8443 | * device is fully setup before sending notifications. | |
8444 | */ | |
7f294054 | 8445 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 8446 | |
ce286d32 EB |
8447 | synchronize_net(); |
8448 | err = 0; | |
8449 | out: | |
8450 | return err; | |
8451 | } | |
463d0183 | 8452 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 8453 | |
f0bf90de | 8454 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
8455 | { |
8456 | struct sk_buff **list_skb; | |
1da177e4 | 8457 | struct sk_buff *skb; |
f0bf90de | 8458 | unsigned int cpu; |
97d8b6e3 | 8459 | struct softnet_data *sd, *oldsd, *remsd = NULL; |
1da177e4 | 8460 | |
1da177e4 LT |
8461 | local_irq_disable(); |
8462 | cpu = smp_processor_id(); | |
8463 | sd = &per_cpu(softnet_data, cpu); | |
8464 | oldsd = &per_cpu(softnet_data, oldcpu); | |
8465 | ||
8466 | /* Find end of our completion_queue. */ | |
8467 | list_skb = &sd->completion_queue; | |
8468 | while (*list_skb) | |
8469 | list_skb = &(*list_skb)->next; | |
8470 | /* Append completion queue from offline CPU. */ | |
8471 | *list_skb = oldsd->completion_queue; | |
8472 | oldsd->completion_queue = NULL; | |
8473 | ||
1da177e4 | 8474 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
8475 | if (oldsd->output_queue) { |
8476 | *sd->output_queue_tailp = oldsd->output_queue; | |
8477 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
8478 | oldsd->output_queue = NULL; | |
8479 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
8480 | } | |
ac64da0b ED |
8481 | /* Append NAPI poll list from offline CPU, with one exception : |
8482 | * process_backlog() must be called by cpu owning percpu backlog. | |
8483 | * We properly handle process_queue & input_pkt_queue later. | |
8484 | */ | |
8485 | while (!list_empty(&oldsd->poll_list)) { | |
8486 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
8487 | struct napi_struct, | |
8488 | poll_list); | |
8489 | ||
8490 | list_del_init(&napi->poll_list); | |
8491 | if (napi->poll == process_backlog) | |
8492 | napi->state = 0; | |
8493 | else | |
8494 | ____napi_schedule(sd, napi); | |
264524d5 | 8495 | } |
1da177e4 LT |
8496 | |
8497 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
8498 | local_irq_enable(); | |
8499 | ||
773fc8f6 | 8500 | #ifdef CONFIG_RPS |
8501 | remsd = oldsd->rps_ipi_list; | |
8502 | oldsd->rps_ipi_list = NULL; | |
8503 | #endif | |
8504 | /* send out pending IPI's on offline CPU */ | |
8505 | net_rps_send_ipi(remsd); | |
8506 | ||
1da177e4 | 8507 | /* Process offline CPU's input_pkt_queue */ |
76cc8b13 | 8508 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 8509 | netif_rx_ni(skb); |
76cc8b13 | 8510 | input_queue_head_incr(oldsd); |
fec5e652 | 8511 | } |
ac64da0b | 8512 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 8513 | netif_rx_ni(skb); |
76cc8b13 TH |
8514 | input_queue_head_incr(oldsd); |
8515 | } | |
1da177e4 | 8516 | |
f0bf90de | 8517 | return 0; |
1da177e4 | 8518 | } |
1da177e4 | 8519 | |
7f353bf2 | 8520 | /** |
b63365a2 HX |
8521 | * netdev_increment_features - increment feature set by one |
8522 | * @all: current feature set | |
8523 | * @one: new feature set | |
8524 | * @mask: mask feature set | |
7f353bf2 HX |
8525 | * |
8526 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
8527 | * @one to the master device with current feature set @all. Will not |
8528 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 8529 | */ |
c8f44aff MM |
8530 | netdev_features_t netdev_increment_features(netdev_features_t all, |
8531 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 8532 | { |
c8cd0989 | 8533 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 8534 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 8535 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 8536 | |
a188222b | 8537 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 8538 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 8539 | |
1742f183 | 8540 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
8541 | if (all & NETIF_F_HW_CSUM) |
8542 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
8543 | |
8544 | return all; | |
8545 | } | |
b63365a2 | 8546 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 8547 | |
430f03cd | 8548 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
8549 | { |
8550 | int i; | |
8551 | struct hlist_head *hash; | |
8552 | ||
8553 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
8554 | if (hash != NULL) | |
8555 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
8556 | INIT_HLIST_HEAD(&hash[i]); | |
8557 | ||
8558 | return hash; | |
8559 | } | |
8560 | ||
881d966b | 8561 | /* Initialize per network namespace state */ |
4665079c | 8562 | static int __net_init netdev_init(struct net *net) |
881d966b | 8563 | { |
734b6541 RM |
8564 | if (net != &init_net) |
8565 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 8566 | |
30d97d35 PE |
8567 | net->dev_name_head = netdev_create_hash(); |
8568 | if (net->dev_name_head == NULL) | |
8569 | goto err_name; | |
881d966b | 8570 | |
30d97d35 PE |
8571 | net->dev_index_head = netdev_create_hash(); |
8572 | if (net->dev_index_head == NULL) | |
8573 | goto err_idx; | |
881d966b EB |
8574 | |
8575 | return 0; | |
30d97d35 PE |
8576 | |
8577 | err_idx: | |
8578 | kfree(net->dev_name_head); | |
8579 | err_name: | |
8580 | return -ENOMEM; | |
881d966b EB |
8581 | } |
8582 | ||
f0db275a SH |
8583 | /** |
8584 | * netdev_drivername - network driver for the device | |
8585 | * @dev: network device | |
f0db275a SH |
8586 | * |
8587 | * Determine network driver for device. | |
8588 | */ | |
3019de12 | 8589 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 8590 | { |
cf04a4c7 SH |
8591 | const struct device_driver *driver; |
8592 | const struct device *parent; | |
3019de12 | 8593 | const char *empty = ""; |
6579e57b AV |
8594 | |
8595 | parent = dev->dev.parent; | |
6579e57b | 8596 | if (!parent) |
3019de12 | 8597 | return empty; |
6579e57b AV |
8598 | |
8599 | driver = parent->driver; | |
8600 | if (driver && driver->name) | |
3019de12 DM |
8601 | return driver->name; |
8602 | return empty; | |
6579e57b AV |
8603 | } |
8604 | ||
6ea754eb JP |
8605 | static void __netdev_printk(const char *level, const struct net_device *dev, |
8606 | struct va_format *vaf) | |
256df2f3 | 8607 | { |
b004ff49 | 8608 | if (dev && dev->dev.parent) { |
6ea754eb JP |
8609 | dev_printk_emit(level[1] - '0', |
8610 | dev->dev.parent, | |
8611 | "%s %s %s%s: %pV", | |
8612 | dev_driver_string(dev->dev.parent), | |
8613 | dev_name(dev->dev.parent), | |
8614 | netdev_name(dev), netdev_reg_state(dev), | |
8615 | vaf); | |
b004ff49 | 8616 | } else if (dev) { |
6ea754eb JP |
8617 | printk("%s%s%s: %pV", |
8618 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 8619 | } else { |
6ea754eb | 8620 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 8621 | } |
256df2f3 JP |
8622 | } |
8623 | ||
6ea754eb JP |
8624 | void netdev_printk(const char *level, const struct net_device *dev, |
8625 | const char *format, ...) | |
256df2f3 JP |
8626 | { |
8627 | struct va_format vaf; | |
8628 | va_list args; | |
256df2f3 JP |
8629 | |
8630 | va_start(args, format); | |
8631 | ||
8632 | vaf.fmt = format; | |
8633 | vaf.va = &args; | |
8634 | ||
6ea754eb | 8635 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 8636 | |
256df2f3 | 8637 | va_end(args); |
256df2f3 JP |
8638 | } |
8639 | EXPORT_SYMBOL(netdev_printk); | |
8640 | ||
8641 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 8642 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 8643 | { \ |
256df2f3 JP |
8644 | struct va_format vaf; \ |
8645 | va_list args; \ | |
8646 | \ | |
8647 | va_start(args, fmt); \ | |
8648 | \ | |
8649 | vaf.fmt = fmt; \ | |
8650 | vaf.va = &args; \ | |
8651 | \ | |
6ea754eb | 8652 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 8653 | \ |
256df2f3 | 8654 | va_end(args); \ |
256df2f3 JP |
8655 | } \ |
8656 | EXPORT_SYMBOL(func); | |
8657 | ||
8658 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
8659 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
8660 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
8661 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
8662 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
8663 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
8664 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
8665 | ||
4665079c | 8666 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
8667 | { |
8668 | kfree(net->dev_name_head); | |
8669 | kfree(net->dev_index_head); | |
8670 | } | |
8671 | ||
022cbae6 | 8672 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
8673 | .init = netdev_init, |
8674 | .exit = netdev_exit, | |
8675 | }; | |
8676 | ||
4665079c | 8677 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 8678 | { |
e008b5fc | 8679 | struct net_device *dev, *aux; |
ce286d32 | 8680 | /* |
e008b5fc | 8681 | * Push all migratable network devices back to the |
ce286d32 EB |
8682 | * initial network namespace |
8683 | */ | |
8684 | rtnl_lock(); | |
e008b5fc | 8685 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 8686 | int err; |
aca51397 | 8687 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
8688 | |
8689 | /* Ignore unmoveable devices (i.e. loopback) */ | |
8690 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8691 | continue; | |
8692 | ||
e008b5fc EB |
8693 | /* Leave virtual devices for the generic cleanup */ |
8694 | if (dev->rtnl_link_ops) | |
8695 | continue; | |
d0c082ce | 8696 | |
25985edc | 8697 | /* Push remaining network devices to init_net */ |
aca51397 PE |
8698 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
8699 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 8700 | if (err) { |
7b6cd1ce JP |
8701 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
8702 | __func__, dev->name, err); | |
aca51397 | 8703 | BUG(); |
ce286d32 EB |
8704 | } |
8705 | } | |
8706 | rtnl_unlock(); | |
8707 | } | |
8708 | ||
50624c93 EB |
8709 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
8710 | { | |
8711 | /* Return with the rtnl_lock held when there are no network | |
8712 | * devices unregistering in any network namespace in net_list. | |
8713 | */ | |
8714 | struct net *net; | |
8715 | bool unregistering; | |
ff960a73 | 8716 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 8717 | |
ff960a73 | 8718 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 8719 | for (;;) { |
50624c93 EB |
8720 | unregistering = false; |
8721 | rtnl_lock(); | |
8722 | list_for_each_entry(net, net_list, exit_list) { | |
8723 | if (net->dev_unreg_count > 0) { | |
8724 | unregistering = true; | |
8725 | break; | |
8726 | } | |
8727 | } | |
8728 | if (!unregistering) | |
8729 | break; | |
8730 | __rtnl_unlock(); | |
ff960a73 PZ |
8731 | |
8732 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 8733 | } |
ff960a73 | 8734 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
8735 | } |
8736 | ||
04dc7f6b EB |
8737 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
8738 | { | |
8739 | /* At exit all network devices most be removed from a network | |
b595076a | 8740 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
8741 | * Do this across as many network namespaces as possible to |
8742 | * improve batching efficiency. | |
8743 | */ | |
8744 | struct net_device *dev; | |
8745 | struct net *net; | |
8746 | LIST_HEAD(dev_kill_list); | |
8747 | ||
50624c93 EB |
8748 | /* To prevent network device cleanup code from dereferencing |
8749 | * loopback devices or network devices that have been freed | |
8750 | * wait here for all pending unregistrations to complete, | |
8751 | * before unregistring the loopback device and allowing the | |
8752 | * network namespace be freed. | |
8753 | * | |
8754 | * The netdev todo list containing all network devices | |
8755 | * unregistrations that happen in default_device_exit_batch | |
8756 | * will run in the rtnl_unlock() at the end of | |
8757 | * default_device_exit_batch. | |
8758 | */ | |
8759 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
8760 | list_for_each_entry(net, net_list, exit_list) { |
8761 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 8762 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
8763 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
8764 | else | |
8765 | unregister_netdevice_queue(dev, &dev_kill_list); | |
8766 | } | |
8767 | } | |
8768 | unregister_netdevice_many(&dev_kill_list); | |
8769 | rtnl_unlock(); | |
8770 | } | |
8771 | ||
022cbae6 | 8772 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 8773 | .exit = default_device_exit, |
04dc7f6b | 8774 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
8775 | }; |
8776 | ||
1da177e4 LT |
8777 | /* |
8778 | * Initialize the DEV module. At boot time this walks the device list and | |
8779 | * unhooks any devices that fail to initialise (normally hardware not | |
8780 | * present) and leaves us with a valid list of present and active devices. | |
8781 | * | |
8782 | */ | |
8783 | ||
8784 | /* | |
8785 | * This is called single threaded during boot, so no need | |
8786 | * to take the rtnl semaphore. | |
8787 | */ | |
8788 | static int __init net_dev_init(void) | |
8789 | { | |
8790 | int i, rc = -ENOMEM; | |
8791 | ||
8792 | BUG_ON(!dev_boot_phase); | |
8793 | ||
1da177e4 LT |
8794 | if (dev_proc_init()) |
8795 | goto out; | |
8796 | ||
8b41d188 | 8797 | if (netdev_kobject_init()) |
1da177e4 LT |
8798 | goto out; |
8799 | ||
8800 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 8801 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
8802 | INIT_LIST_HEAD(&ptype_base[i]); |
8803 | ||
62532da9 VY |
8804 | INIT_LIST_HEAD(&offload_base); |
8805 | ||
881d966b EB |
8806 | if (register_pernet_subsys(&netdev_net_ops)) |
8807 | goto out; | |
1da177e4 LT |
8808 | |
8809 | /* | |
8810 | * Initialise the packet receive queues. | |
8811 | */ | |
8812 | ||
6f912042 | 8813 | for_each_possible_cpu(i) { |
41852497 | 8814 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 8815 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 8816 | |
41852497 ED |
8817 | INIT_WORK(flush, flush_backlog); |
8818 | ||
e36fa2f7 | 8819 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 8820 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 8821 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 8822 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 8823 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
8824 | sd->csd.func = rps_trigger_softirq; |
8825 | sd->csd.info = sd; | |
e36fa2f7 | 8826 | sd->cpu = i; |
1e94d72f | 8827 | #endif |
0a9627f2 | 8828 | |
e36fa2f7 ED |
8829 | sd->backlog.poll = process_backlog; |
8830 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
8831 | } |
8832 | ||
1da177e4 LT |
8833 | dev_boot_phase = 0; |
8834 | ||
505d4f73 EB |
8835 | /* The loopback device is special if any other network devices |
8836 | * is present in a network namespace the loopback device must | |
8837 | * be present. Since we now dynamically allocate and free the | |
8838 | * loopback device ensure this invariant is maintained by | |
8839 | * keeping the loopback device as the first device on the | |
8840 | * list of network devices. Ensuring the loopback devices | |
8841 | * is the first device that appears and the last network device | |
8842 | * that disappears. | |
8843 | */ | |
8844 | if (register_pernet_device(&loopback_net_ops)) | |
8845 | goto out; | |
8846 | ||
8847 | if (register_pernet_device(&default_device_ops)) | |
8848 | goto out; | |
8849 | ||
962cf36c CM |
8850 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
8851 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 8852 | |
f0bf90de SAS |
8853 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
8854 | NULL, dev_cpu_dead); | |
8855 | WARN_ON(rc < 0); | |
1da177e4 LT |
8856 | rc = 0; |
8857 | out: | |
8858 | return rc; | |
8859 | } | |
8860 | ||
8861 | subsys_initcall(net_dev_init); |