]>
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> |
1da177e4 | 148 | |
342709ef PE |
149 | #include "net-sysfs.h" |
150 | ||
d565b0a1 HX |
151 | /* Instead of increasing this, you should create a hash table. */ |
152 | #define MAX_GRO_SKBS 8 | |
153 | ||
5d38a079 HX |
154 | /* This should be increased if a protocol with a bigger head is added. */ |
155 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
156 | ||
1da177e4 | 157 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 158 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
159 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
160 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 161 | static struct list_head offload_base __read_mostly; |
1da177e4 | 162 | |
ae78dbfa | 163 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
164 | static int call_netdevice_notifiers_info(unsigned long val, |
165 | struct net_device *dev, | |
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 | ||
af12fa6e ET |
191 | /* protects napi_hash addition/deletion and napi_gen_id */ |
192 | static DEFINE_SPINLOCK(napi_hash_lock); | |
193 | ||
52bd2d62 | 194 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 195 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 196 | |
18afa4b0 | 197 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 198 | |
4e985ada TG |
199 | static inline void dev_base_seq_inc(struct net *net) |
200 | { | |
643aa9cb | 201 | while (++net->dev_base_seq == 0) |
202 | ; | |
4e985ada TG |
203 | } |
204 | ||
881d966b | 205 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 206 | { |
8387ff25 | 207 | unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); |
95c96174 | 208 | |
08e9897d | 209 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
210 | } |
211 | ||
881d966b | 212 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 213 | { |
7c28bd0b | 214 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
215 | } |
216 | ||
e36fa2f7 | 217 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
218 | { |
219 | #ifdef CONFIG_RPS | |
e36fa2f7 | 220 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
221 | #endif |
222 | } | |
223 | ||
e36fa2f7 | 224 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
225 | { |
226 | #ifdef CONFIG_RPS | |
e36fa2f7 | 227 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
228 | #endif |
229 | } | |
230 | ||
ce286d32 | 231 | /* Device list insertion */ |
53759be9 | 232 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 233 | { |
c346dca1 | 234 | struct net *net = dev_net(dev); |
ce286d32 EB |
235 | |
236 | ASSERT_RTNL(); | |
237 | ||
238 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 239 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 240 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
241 | hlist_add_head_rcu(&dev->index_hlist, |
242 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 243 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
244 | |
245 | dev_base_seq_inc(net); | |
ce286d32 EB |
246 | } |
247 | ||
fb699dfd ED |
248 | /* Device list removal |
249 | * caller must respect a RCU grace period before freeing/reusing dev | |
250 | */ | |
ce286d32 EB |
251 | static void unlist_netdevice(struct net_device *dev) |
252 | { | |
253 | ASSERT_RTNL(); | |
254 | ||
255 | /* Unlink dev from the device chain */ | |
256 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 257 | list_del_rcu(&dev->dev_list); |
72c9528b | 258 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 259 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 260 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
261 | |
262 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
263 | } |
264 | ||
1da177e4 LT |
265 | /* |
266 | * Our notifier list | |
267 | */ | |
268 | ||
f07d5b94 | 269 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
270 | |
271 | /* | |
272 | * Device drivers call our routines to queue packets here. We empty the | |
273 | * queue in the local softnet handler. | |
274 | */ | |
bea3348e | 275 | |
9958da05 | 276 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 277 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 278 | |
cf508b12 | 279 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 280 | /* |
c773e847 | 281 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
282 | * according to dev->type |
283 | */ | |
643aa9cb | 284 | static const unsigned short netdev_lock_type[] = { |
285 | ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
723e98b7 JP |
286 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, |
287 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
288 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
289 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
290 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
291 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
292 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
293 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
294 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
295 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
296 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
297 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
298 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
299 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 300 | |
643aa9cb | 301 | static const char *const netdev_lock_name[] = { |
302 | "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
303 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
304 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
305 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
306 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
307 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
308 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
309 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
310 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
311 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
312 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
313 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
314 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", | |
315 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
316 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
317 | |
318 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 319 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
320 | |
321 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
322 | { | |
323 | int i; | |
324 | ||
325 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
326 | if (netdev_lock_type[i] == dev_type) | |
327 | return i; | |
328 | /* the last key is used by default */ | |
329 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
330 | } | |
331 | ||
cf508b12 DM |
332 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
333 | unsigned short dev_type) | |
723e98b7 JP |
334 | { |
335 | int i; | |
336 | ||
337 | i = netdev_lock_pos(dev_type); | |
338 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
339 | netdev_lock_name[i]); | |
340 | } | |
cf508b12 DM |
341 | |
342 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
343 | { | |
344 | int i; | |
345 | ||
346 | i = netdev_lock_pos(dev->type); | |
347 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
348 | &netdev_addr_lock_key[i], | |
349 | netdev_lock_name[i]); | |
350 | } | |
723e98b7 | 351 | #else |
cf508b12 DM |
352 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
353 | unsigned short dev_type) | |
354 | { | |
355 | } | |
356 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
357 | { |
358 | } | |
359 | #endif | |
1da177e4 LT |
360 | |
361 | /******************************************************************************* | |
eb13da1a | 362 | * |
363 | * Protocol management and registration routines | |
364 | * | |
365 | *******************************************************************************/ | |
1da177e4 | 366 | |
1da177e4 | 367 | |
1da177e4 LT |
368 | /* |
369 | * Add a protocol ID to the list. Now that the input handler is | |
370 | * smarter we can dispense with all the messy stuff that used to be | |
371 | * here. | |
372 | * | |
373 | * BEWARE!!! Protocol handlers, mangling input packets, | |
374 | * MUST BE last in hash buckets and checking protocol handlers | |
375 | * MUST start from promiscuous ptype_all chain in net_bh. | |
376 | * It is true now, do not change it. | |
377 | * Explanation follows: if protocol handler, mangling packet, will | |
378 | * be the first on list, it is not able to sense, that packet | |
379 | * is cloned and should be copied-on-write, so that it will | |
380 | * change it and subsequent readers will get broken packet. | |
381 | * --ANK (980803) | |
382 | */ | |
383 | ||
c07b68e8 ED |
384 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
385 | { | |
386 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 387 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 388 | else |
7866a621 SN |
389 | return pt->dev ? &pt->dev->ptype_specific : |
390 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
391 | } |
392 | ||
1da177e4 LT |
393 | /** |
394 | * dev_add_pack - add packet handler | |
395 | * @pt: packet type declaration | |
396 | * | |
397 | * Add a protocol handler to the networking stack. The passed &packet_type | |
398 | * is linked into kernel lists and may not be freed until it has been | |
399 | * removed from the kernel lists. | |
400 | * | |
4ec93edb | 401 | * This call does not sleep therefore it can not |
1da177e4 LT |
402 | * guarantee all CPU's that are in middle of receiving packets |
403 | * will see the new packet type (until the next received packet). | |
404 | */ | |
405 | ||
406 | void dev_add_pack(struct packet_type *pt) | |
407 | { | |
c07b68e8 | 408 | struct list_head *head = ptype_head(pt); |
1da177e4 | 409 | |
c07b68e8 ED |
410 | spin_lock(&ptype_lock); |
411 | list_add_rcu(&pt->list, head); | |
412 | spin_unlock(&ptype_lock); | |
1da177e4 | 413 | } |
d1b19dff | 414 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 415 | |
1da177e4 LT |
416 | /** |
417 | * __dev_remove_pack - remove packet handler | |
418 | * @pt: packet type declaration | |
419 | * | |
420 | * Remove a protocol handler that was previously added to the kernel | |
421 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
422 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 423 | * returns. |
1da177e4 LT |
424 | * |
425 | * The packet type might still be in use by receivers | |
426 | * and must not be freed until after all the CPU's have gone | |
427 | * through a quiescent state. | |
428 | */ | |
429 | void __dev_remove_pack(struct packet_type *pt) | |
430 | { | |
c07b68e8 | 431 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
432 | struct packet_type *pt1; |
433 | ||
c07b68e8 | 434 | spin_lock(&ptype_lock); |
1da177e4 LT |
435 | |
436 | list_for_each_entry(pt1, head, list) { | |
437 | if (pt == pt1) { | |
438 | list_del_rcu(&pt->list); | |
439 | goto out; | |
440 | } | |
441 | } | |
442 | ||
7b6cd1ce | 443 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 444 | out: |
c07b68e8 | 445 | spin_unlock(&ptype_lock); |
1da177e4 | 446 | } |
d1b19dff ED |
447 | EXPORT_SYMBOL(__dev_remove_pack); |
448 | ||
1da177e4 LT |
449 | /** |
450 | * dev_remove_pack - remove packet handler | |
451 | * @pt: packet type declaration | |
452 | * | |
453 | * Remove a protocol handler that was previously added to the kernel | |
454 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
455 | * from the kernel lists and can be freed or reused once this function | |
456 | * returns. | |
457 | * | |
458 | * This call sleeps to guarantee that no CPU is looking at the packet | |
459 | * type after return. | |
460 | */ | |
461 | void dev_remove_pack(struct packet_type *pt) | |
462 | { | |
463 | __dev_remove_pack(pt); | |
4ec93edb | 464 | |
1da177e4 LT |
465 | synchronize_net(); |
466 | } | |
d1b19dff | 467 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 468 | |
62532da9 VY |
469 | |
470 | /** | |
471 | * dev_add_offload - register offload handlers | |
472 | * @po: protocol offload declaration | |
473 | * | |
474 | * Add protocol offload handlers to the networking stack. The passed | |
475 | * &proto_offload is linked into kernel lists and may not be freed until | |
476 | * it has been removed from the kernel lists. | |
477 | * | |
478 | * This call does not sleep therefore it can not | |
479 | * guarantee all CPU's that are in middle of receiving packets | |
480 | * will see the new offload handlers (until the next received packet). | |
481 | */ | |
482 | void dev_add_offload(struct packet_offload *po) | |
483 | { | |
bdef7de4 | 484 | struct packet_offload *elem; |
62532da9 VY |
485 | |
486 | spin_lock(&offload_lock); | |
bdef7de4 DM |
487 | list_for_each_entry(elem, &offload_base, list) { |
488 | if (po->priority < elem->priority) | |
489 | break; | |
490 | } | |
491 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
492 | spin_unlock(&offload_lock); |
493 | } | |
494 | EXPORT_SYMBOL(dev_add_offload); | |
495 | ||
496 | /** | |
497 | * __dev_remove_offload - remove offload handler | |
498 | * @po: packet offload declaration | |
499 | * | |
500 | * Remove a protocol offload handler that was previously added to the | |
501 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
502 | * is removed from the kernel lists and can be freed or reused once this | |
503 | * function returns. | |
504 | * | |
505 | * The packet type might still be in use by receivers | |
506 | * and must not be freed until after all the CPU's have gone | |
507 | * through a quiescent state. | |
508 | */ | |
1d143d9f | 509 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
510 | { |
511 | struct list_head *head = &offload_base; | |
512 | struct packet_offload *po1; | |
513 | ||
c53aa505 | 514 | spin_lock(&offload_lock); |
62532da9 VY |
515 | |
516 | list_for_each_entry(po1, head, list) { | |
517 | if (po == po1) { | |
518 | list_del_rcu(&po->list); | |
519 | goto out; | |
520 | } | |
521 | } | |
522 | ||
523 | pr_warn("dev_remove_offload: %p not found\n", po); | |
524 | out: | |
c53aa505 | 525 | spin_unlock(&offload_lock); |
62532da9 | 526 | } |
62532da9 VY |
527 | |
528 | /** | |
529 | * dev_remove_offload - remove packet offload handler | |
530 | * @po: packet offload declaration | |
531 | * | |
532 | * Remove a packet offload handler that was previously added to the kernel | |
533 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
534 | * removed from the kernel lists and can be freed or reused once this | |
535 | * function returns. | |
536 | * | |
537 | * This call sleeps to guarantee that no CPU is looking at the packet | |
538 | * type after return. | |
539 | */ | |
540 | void dev_remove_offload(struct packet_offload *po) | |
541 | { | |
542 | __dev_remove_offload(po); | |
543 | ||
544 | synchronize_net(); | |
545 | } | |
546 | EXPORT_SYMBOL(dev_remove_offload); | |
547 | ||
1da177e4 | 548 | /****************************************************************************** |
eb13da1a | 549 | * |
550 | * Device Boot-time Settings Routines | |
551 | * | |
552 | ******************************************************************************/ | |
1da177e4 LT |
553 | |
554 | /* Boot time configuration table */ | |
555 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
556 | ||
557 | /** | |
558 | * netdev_boot_setup_add - add new setup entry | |
559 | * @name: name of the device | |
560 | * @map: configured settings for the device | |
561 | * | |
562 | * Adds new setup entry to the dev_boot_setup list. The function | |
563 | * returns 0 on error and 1 on success. This is a generic routine to | |
564 | * all netdevices. | |
565 | */ | |
566 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
567 | { | |
568 | struct netdev_boot_setup *s; | |
569 | int i; | |
570 | ||
571 | s = dev_boot_setup; | |
572 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
573 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
574 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 575 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
576 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
577 | break; | |
578 | } | |
579 | } | |
580 | ||
581 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
582 | } | |
583 | ||
584 | /** | |
722c9a0c | 585 | * netdev_boot_setup_check - check boot time settings |
586 | * @dev: the netdevice | |
1da177e4 | 587 | * |
722c9a0c | 588 | * Check boot time settings for the device. |
589 | * The found settings are set for the device to be used | |
590 | * later in the device probing. | |
591 | * Returns 0 if no settings found, 1 if they are. | |
1da177e4 LT |
592 | */ |
593 | int netdev_boot_setup_check(struct net_device *dev) | |
594 | { | |
595 | struct netdev_boot_setup *s = dev_boot_setup; | |
596 | int i; | |
597 | ||
598 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
599 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 600 | !strcmp(dev->name, s[i].name)) { |
722c9a0c | 601 | dev->irq = s[i].map.irq; |
602 | dev->base_addr = s[i].map.base_addr; | |
603 | dev->mem_start = s[i].map.mem_start; | |
604 | dev->mem_end = s[i].map.mem_end; | |
1da177e4 LT |
605 | return 1; |
606 | } | |
607 | } | |
608 | return 0; | |
609 | } | |
d1b19dff | 610 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
611 | |
612 | ||
613 | /** | |
722c9a0c | 614 | * netdev_boot_base - get address from boot time settings |
615 | * @prefix: prefix for network device | |
616 | * @unit: id for network device | |
617 | * | |
618 | * Check boot time settings for the base address of device. | |
619 | * The found settings are set for the device to be used | |
620 | * later in the device probing. | |
621 | * Returns 0 if no settings found. | |
1da177e4 LT |
622 | */ |
623 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
624 | { | |
625 | const struct netdev_boot_setup *s = dev_boot_setup; | |
626 | char name[IFNAMSIZ]; | |
627 | int i; | |
628 | ||
629 | sprintf(name, "%s%d", prefix, unit); | |
630 | ||
631 | /* | |
632 | * If device already registered then return base of 1 | |
633 | * to indicate not to probe for this interface | |
634 | */ | |
881d966b | 635 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
636 | return 1; |
637 | ||
638 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
639 | if (!strcmp(name, s[i].name)) | |
640 | return s[i].map.base_addr; | |
641 | return 0; | |
642 | } | |
643 | ||
644 | /* | |
645 | * Saves at boot time configured settings for any netdevice. | |
646 | */ | |
647 | int __init netdev_boot_setup(char *str) | |
648 | { | |
649 | int ints[5]; | |
650 | struct ifmap map; | |
651 | ||
652 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
653 | if (!str || !*str) | |
654 | return 0; | |
655 | ||
656 | /* Save settings */ | |
657 | memset(&map, 0, sizeof(map)); | |
658 | if (ints[0] > 0) | |
659 | map.irq = ints[1]; | |
660 | if (ints[0] > 1) | |
661 | map.base_addr = ints[2]; | |
662 | if (ints[0] > 2) | |
663 | map.mem_start = ints[3]; | |
664 | if (ints[0] > 3) | |
665 | map.mem_end = ints[4]; | |
666 | ||
667 | /* Add new entry to the list */ | |
668 | return netdev_boot_setup_add(str, &map); | |
669 | } | |
670 | ||
671 | __setup("netdev=", netdev_boot_setup); | |
672 | ||
673 | /******************************************************************************* | |
eb13da1a | 674 | * |
675 | * Device Interface Subroutines | |
676 | * | |
677 | *******************************************************************************/ | |
1da177e4 | 678 | |
a54acb3a ND |
679 | /** |
680 | * dev_get_iflink - get 'iflink' value of a interface | |
681 | * @dev: targeted interface | |
682 | * | |
683 | * Indicates the ifindex the interface is linked to. | |
684 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
685 | */ | |
686 | ||
687 | int dev_get_iflink(const struct net_device *dev) | |
688 | { | |
689 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
690 | return dev->netdev_ops->ndo_get_iflink(dev); | |
691 | ||
7a66bbc9 | 692 | return dev->ifindex; |
a54acb3a ND |
693 | } |
694 | EXPORT_SYMBOL(dev_get_iflink); | |
695 | ||
fc4099f1 PS |
696 | /** |
697 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
698 | * @dev: targeted interface | |
699 | * @skb: The packet. | |
700 | * | |
701 | * For better visibility of tunnel traffic OVS needs to retrieve | |
702 | * egress tunnel information for a packet. Following API allows | |
703 | * user to get this info. | |
704 | */ | |
705 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
706 | { | |
707 | struct ip_tunnel_info *info; | |
708 | ||
709 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
710 | return -EINVAL; | |
711 | ||
712 | info = skb_tunnel_info_unclone(skb); | |
713 | if (!info) | |
714 | return -ENOMEM; | |
715 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
716 | return -EINVAL; | |
717 | ||
718 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
719 | } | |
720 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
721 | ||
1da177e4 LT |
722 | /** |
723 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 724 | * @net: the applicable net namespace |
1da177e4 LT |
725 | * @name: name to find |
726 | * | |
727 | * Find an interface by name. Must be called under RTNL semaphore | |
728 | * or @dev_base_lock. If the name is found a pointer to the device | |
729 | * is returned. If the name is not found then %NULL is returned. The | |
730 | * reference counters are not incremented so the caller must be | |
731 | * careful with locks. | |
732 | */ | |
733 | ||
881d966b | 734 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 735 | { |
0bd8d536 ED |
736 | struct net_device *dev; |
737 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 738 | |
b67bfe0d | 739 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
740 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
741 | return dev; | |
0bd8d536 | 742 | |
1da177e4 LT |
743 | return NULL; |
744 | } | |
d1b19dff | 745 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 746 | |
72c9528b | 747 | /** |
722c9a0c | 748 | * dev_get_by_name_rcu - find a device by its name |
749 | * @net: the applicable net namespace | |
750 | * @name: name to find | |
751 | * | |
752 | * Find an interface by name. | |
753 | * If the name is found a pointer to the device is returned. | |
754 | * If the name is not found then %NULL is returned. | |
755 | * The reference counters are not incremented so the caller must be | |
756 | * careful with locks. The caller must hold RCU lock. | |
72c9528b ED |
757 | */ |
758 | ||
759 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
760 | { | |
72c9528b ED |
761 | struct net_device *dev; |
762 | struct hlist_head *head = dev_name_hash(net, name); | |
763 | ||
b67bfe0d | 764 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
765 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
766 | return dev; | |
767 | ||
768 | return NULL; | |
769 | } | |
770 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
771 | ||
1da177e4 LT |
772 | /** |
773 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 774 | * @net: the applicable net namespace |
1da177e4 LT |
775 | * @name: name to find |
776 | * | |
777 | * Find an interface by name. This can be called from any | |
778 | * context and does its own locking. The returned handle has | |
779 | * the usage count incremented and the caller must use dev_put() to | |
780 | * release it when it is no longer needed. %NULL is returned if no | |
781 | * matching device is found. | |
782 | */ | |
783 | ||
881d966b | 784 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
785 | { |
786 | struct net_device *dev; | |
787 | ||
72c9528b ED |
788 | rcu_read_lock(); |
789 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
790 | if (dev) |
791 | dev_hold(dev); | |
72c9528b | 792 | rcu_read_unlock(); |
1da177e4 LT |
793 | return dev; |
794 | } | |
d1b19dff | 795 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
796 | |
797 | /** | |
798 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 799 | * @net: the applicable net namespace |
1da177e4 LT |
800 | * @ifindex: index of device |
801 | * | |
802 | * Search for an interface by index. Returns %NULL if the device | |
803 | * is not found or a pointer to the device. The device has not | |
804 | * had its reference counter increased so the caller must be careful | |
805 | * about locking. The caller must hold either the RTNL semaphore | |
806 | * or @dev_base_lock. | |
807 | */ | |
808 | ||
881d966b | 809 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 810 | { |
0bd8d536 ED |
811 | struct net_device *dev; |
812 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 813 | |
b67bfe0d | 814 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
815 | if (dev->ifindex == ifindex) |
816 | return dev; | |
0bd8d536 | 817 | |
1da177e4 LT |
818 | return NULL; |
819 | } | |
d1b19dff | 820 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 821 | |
fb699dfd ED |
822 | /** |
823 | * dev_get_by_index_rcu - find a device by its ifindex | |
824 | * @net: the applicable net namespace | |
825 | * @ifindex: index of device | |
826 | * | |
827 | * Search for an interface by index. Returns %NULL if the device | |
828 | * is not found or a pointer to the device. The device has not | |
829 | * had its reference counter increased so the caller must be careful | |
830 | * about locking. The caller must hold RCU lock. | |
831 | */ | |
832 | ||
833 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
834 | { | |
fb699dfd ED |
835 | struct net_device *dev; |
836 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
837 | ||
b67bfe0d | 838 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
839 | if (dev->ifindex == ifindex) |
840 | return dev; | |
841 | ||
842 | return NULL; | |
843 | } | |
844 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
845 | ||
1da177e4 LT |
846 | |
847 | /** | |
848 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 849 | * @net: the applicable net namespace |
1da177e4 LT |
850 | * @ifindex: index of device |
851 | * | |
852 | * Search for an interface by index. Returns NULL if the device | |
853 | * is not found or a pointer to the device. The device returned has | |
854 | * had a reference added and the pointer is safe until the user calls | |
855 | * dev_put to indicate they have finished with it. | |
856 | */ | |
857 | ||
881d966b | 858 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
859 | { |
860 | struct net_device *dev; | |
861 | ||
fb699dfd ED |
862 | rcu_read_lock(); |
863 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
864 | if (dev) |
865 | dev_hold(dev); | |
fb699dfd | 866 | rcu_read_unlock(); |
1da177e4 LT |
867 | return dev; |
868 | } | |
d1b19dff | 869 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 870 | |
90b602f8 ML |
871 | /** |
872 | * dev_get_by_napi_id - find a device by napi_id | |
873 | * @napi_id: ID of the NAPI struct | |
874 | * | |
875 | * Search for an interface by NAPI ID. Returns %NULL if the device | |
876 | * is not found or a pointer to the device. The device has not had | |
877 | * its reference counter increased so the caller must be careful | |
878 | * about locking. The caller must hold RCU lock. | |
879 | */ | |
880 | ||
881 | struct net_device *dev_get_by_napi_id(unsigned int napi_id) | |
882 | { | |
883 | struct napi_struct *napi; | |
884 | ||
885 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
886 | ||
887 | if (napi_id < MIN_NAPI_ID) | |
888 | return NULL; | |
889 | ||
890 | napi = napi_by_id(napi_id); | |
891 | ||
892 | return napi ? napi->dev : NULL; | |
893 | } | |
894 | EXPORT_SYMBOL(dev_get_by_napi_id); | |
895 | ||
5dbe7c17 NS |
896 | /** |
897 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
898 | * @net: network namespace | |
899 | * @name: a pointer to the buffer where the name will be stored. | |
900 | * @ifindex: the ifindex of the interface to get the name from. | |
901 | * | |
902 | * The use of raw_seqcount_begin() and cond_resched() before | |
903 | * retrying is required as we want to give the writers a chance | |
904 | * to complete when CONFIG_PREEMPT is not set. | |
905 | */ | |
906 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
907 | { | |
908 | struct net_device *dev; | |
909 | unsigned int seq; | |
910 | ||
911 | retry: | |
912 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
913 | rcu_read_lock(); | |
914 | dev = dev_get_by_index_rcu(net, ifindex); | |
915 | if (!dev) { | |
916 | rcu_read_unlock(); | |
917 | return -ENODEV; | |
918 | } | |
919 | ||
920 | strcpy(name, dev->name); | |
921 | rcu_read_unlock(); | |
922 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
923 | cond_resched(); | |
924 | goto retry; | |
925 | } | |
926 | ||
927 | return 0; | |
928 | } | |
929 | ||
1da177e4 | 930 | /** |
941666c2 | 931 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 932 | * @net: the applicable net namespace |
1da177e4 LT |
933 | * @type: media type of device |
934 | * @ha: hardware address | |
935 | * | |
936 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
937 | * is not found or a pointer to the device. |
938 | * The caller must hold RCU or RTNL. | |
941666c2 | 939 | * The returned device has not had its ref count increased |
1da177e4 LT |
940 | * and the caller must therefore be careful about locking |
941 | * | |
1da177e4 LT |
942 | */ |
943 | ||
941666c2 ED |
944 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
945 | const char *ha) | |
1da177e4 LT |
946 | { |
947 | struct net_device *dev; | |
948 | ||
941666c2 | 949 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
950 | if (dev->type == type && |
951 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
952 | return dev; |
953 | ||
954 | return NULL; | |
1da177e4 | 955 | } |
941666c2 | 956 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 957 | |
881d966b | 958 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
959 | { |
960 | struct net_device *dev; | |
961 | ||
4e9cac2b | 962 | ASSERT_RTNL(); |
881d966b | 963 | for_each_netdev(net, dev) |
4e9cac2b | 964 | if (dev->type == type) |
7562f876 PE |
965 | return dev; |
966 | ||
967 | return NULL; | |
4e9cac2b | 968 | } |
4e9cac2b PM |
969 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
970 | ||
881d966b | 971 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 972 | { |
99fe3c39 | 973 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 974 | |
99fe3c39 ED |
975 | rcu_read_lock(); |
976 | for_each_netdev_rcu(net, dev) | |
977 | if (dev->type == type) { | |
978 | dev_hold(dev); | |
979 | ret = dev; | |
980 | break; | |
981 | } | |
982 | rcu_read_unlock(); | |
983 | return ret; | |
1da177e4 | 984 | } |
1da177e4 LT |
985 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
986 | ||
987 | /** | |
6c555490 | 988 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 989 | * @net: the applicable net namespace |
1da177e4 LT |
990 | * @if_flags: IFF_* values |
991 | * @mask: bitmask of bits in if_flags to check | |
992 | * | |
993 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 994 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 995 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
996 | */ |
997 | ||
6c555490 WC |
998 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
999 | unsigned short mask) | |
1da177e4 | 1000 | { |
7562f876 | 1001 | struct net_device *dev, *ret; |
1da177e4 | 1002 | |
6c555490 WC |
1003 | ASSERT_RTNL(); |
1004 | ||
7562f876 | 1005 | ret = NULL; |
6c555490 | 1006 | for_each_netdev(net, dev) { |
1da177e4 | 1007 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 1008 | ret = dev; |
1da177e4 LT |
1009 | break; |
1010 | } | |
1011 | } | |
7562f876 | 1012 | return ret; |
1da177e4 | 1013 | } |
6c555490 | 1014 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
1015 | |
1016 | /** | |
1017 | * dev_valid_name - check if name is okay for network device | |
1018 | * @name: name string | |
1019 | * | |
1020 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
1021 | * to allow sysfs to work. We also disallow any kind of |
1022 | * whitespace. | |
1da177e4 | 1023 | */ |
95f050bf | 1024 | bool dev_valid_name(const char *name) |
1da177e4 | 1025 | { |
c7fa9d18 | 1026 | if (*name == '\0') |
95f050bf | 1027 | return false; |
b6fe17d6 | 1028 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 1029 | return false; |
c7fa9d18 | 1030 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1031 | return false; |
c7fa9d18 DM |
1032 | |
1033 | while (*name) { | |
a4176a93 | 1034 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1035 | return false; |
c7fa9d18 DM |
1036 | name++; |
1037 | } | |
95f050bf | 1038 | return true; |
1da177e4 | 1039 | } |
d1b19dff | 1040 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1041 | |
1042 | /** | |
b267b179 EB |
1043 | * __dev_alloc_name - allocate a name for a device |
1044 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1045 | * @name: name format string |
b267b179 | 1046 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1047 | * |
1048 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1049 | * id. It scans list of devices to build up a free map, then chooses |
1050 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1051 | * while allocating the name and adding the device in order to avoid | |
1052 | * duplicates. | |
1053 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1054 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1055 | */ |
1056 | ||
b267b179 | 1057 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1058 | { |
1059 | int i = 0; | |
1da177e4 LT |
1060 | const char *p; |
1061 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1062 | unsigned long *inuse; |
1da177e4 LT |
1063 | struct net_device *d; |
1064 | ||
1065 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
1066 | if (p) { | |
1067 | /* | |
1068 | * Verify the string as this thing may have come from | |
1069 | * the user. There must be either one "%d" and no other "%" | |
1070 | * characters. | |
1071 | */ | |
1072 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1073 | return -EINVAL; | |
1074 | ||
1075 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1076 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1077 | if (!inuse) |
1078 | return -ENOMEM; | |
1079 | ||
881d966b | 1080 | for_each_netdev(net, d) { |
1da177e4 LT |
1081 | if (!sscanf(d->name, name, &i)) |
1082 | continue; | |
1083 | if (i < 0 || i >= max_netdevices) | |
1084 | continue; | |
1085 | ||
1086 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1087 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1088 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1089 | set_bit(i, inuse); | |
1090 | } | |
1091 | ||
1092 | i = find_first_zero_bit(inuse, max_netdevices); | |
1093 | free_page((unsigned long) inuse); | |
1094 | } | |
1095 | ||
d9031024 OP |
1096 | if (buf != name) |
1097 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1098 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1099 | return i; |
1da177e4 LT |
1100 | |
1101 | /* It is possible to run out of possible slots | |
1102 | * when the name is long and there isn't enough space left | |
1103 | * for the digits, or if all bits are used. | |
1104 | */ | |
1105 | return -ENFILE; | |
1106 | } | |
1107 | ||
b267b179 EB |
1108 | /** |
1109 | * dev_alloc_name - allocate a name for a device | |
1110 | * @dev: device | |
1111 | * @name: name format string | |
1112 | * | |
1113 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1114 | * id. It scans list of devices to build up a free map, then chooses | |
1115 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1116 | * while allocating the name and adding the device in order to avoid | |
1117 | * duplicates. | |
1118 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1119 | * Returns the number of the unit assigned or a negative errno code. | |
1120 | */ | |
1121 | ||
1122 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1123 | { | |
1124 | char buf[IFNAMSIZ]; | |
1125 | struct net *net; | |
1126 | int ret; | |
1127 | ||
c346dca1 YH |
1128 | BUG_ON(!dev_net(dev)); |
1129 | net = dev_net(dev); | |
b267b179 EB |
1130 | ret = __dev_alloc_name(net, name, buf); |
1131 | if (ret >= 0) | |
1132 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1133 | return ret; | |
1134 | } | |
d1b19dff | 1135 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1136 | |
828de4f6 G |
1137 | static int dev_alloc_name_ns(struct net *net, |
1138 | struct net_device *dev, | |
1139 | const char *name) | |
d9031024 | 1140 | { |
828de4f6 G |
1141 | char buf[IFNAMSIZ]; |
1142 | int ret; | |
8ce6cebc | 1143 | |
828de4f6 G |
1144 | ret = __dev_alloc_name(net, name, buf); |
1145 | if (ret >= 0) | |
1146 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1147 | return ret; | |
1148 | } | |
1149 | ||
0ad646c8 CW |
1150 | int dev_get_valid_name(struct net *net, struct net_device *dev, |
1151 | const char *name) | |
828de4f6 G |
1152 | { |
1153 | BUG_ON(!net); | |
8ce6cebc | 1154 | |
d9031024 OP |
1155 | if (!dev_valid_name(name)) |
1156 | return -EINVAL; | |
1157 | ||
1c5cae81 | 1158 | if (strchr(name, '%')) |
828de4f6 | 1159 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1160 | else if (__dev_get_by_name(net, name)) |
1161 | return -EEXIST; | |
8ce6cebc DL |
1162 | else if (dev->name != name) |
1163 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1164 | |
1165 | return 0; | |
1166 | } | |
0ad646c8 | 1167 | EXPORT_SYMBOL(dev_get_valid_name); |
1da177e4 LT |
1168 | |
1169 | /** | |
1170 | * dev_change_name - change name of a device | |
1171 | * @dev: device | |
1172 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1173 | * | |
1174 | * Change name of a device, can pass format strings "eth%d". | |
1175 | * for wildcarding. | |
1176 | */ | |
cf04a4c7 | 1177 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1178 | { |
238fa362 | 1179 | unsigned char old_assign_type; |
fcc5a03a | 1180 | char oldname[IFNAMSIZ]; |
1da177e4 | 1181 | int err = 0; |
fcc5a03a | 1182 | int ret; |
881d966b | 1183 | struct net *net; |
1da177e4 LT |
1184 | |
1185 | ASSERT_RTNL(); | |
c346dca1 | 1186 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1187 | |
c346dca1 | 1188 | net = dev_net(dev); |
1da177e4 LT |
1189 | if (dev->flags & IFF_UP) |
1190 | return -EBUSY; | |
1191 | ||
30e6c9fa | 1192 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1193 | |
1194 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1195 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1196 | return 0; |
c91f6df2 | 1197 | } |
c8d90dca | 1198 | |
fcc5a03a HX |
1199 | memcpy(oldname, dev->name, IFNAMSIZ); |
1200 | ||
828de4f6 | 1201 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1202 | if (err < 0) { |
30e6c9fa | 1203 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1204 | return err; |
c91f6df2 | 1205 | } |
1da177e4 | 1206 | |
6fe82a39 VF |
1207 | if (oldname[0] && !strchr(oldname, '%')) |
1208 | netdev_info(dev, "renamed from %s\n", oldname); | |
1209 | ||
238fa362 TG |
1210 | old_assign_type = dev->name_assign_type; |
1211 | dev->name_assign_type = NET_NAME_RENAMED; | |
1212 | ||
fcc5a03a | 1213 | rollback: |
a1b3f594 EB |
1214 | ret = device_rename(&dev->dev, dev->name); |
1215 | if (ret) { | |
1216 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1217 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1218 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1219 | return ret; |
dcc99773 | 1220 | } |
7f988eab | 1221 | |
30e6c9fa | 1222 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1223 | |
5bb025fa VF |
1224 | netdev_adjacent_rename_links(dev, oldname); |
1225 | ||
7f988eab | 1226 | write_lock_bh(&dev_base_lock); |
372b2312 | 1227 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1228 | write_unlock_bh(&dev_base_lock); |
1229 | ||
1230 | synchronize_rcu(); | |
1231 | ||
1232 | write_lock_bh(&dev_base_lock); | |
1233 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1234 | write_unlock_bh(&dev_base_lock); |
1235 | ||
056925ab | 1236 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1237 | ret = notifier_to_errno(ret); |
1238 | ||
1239 | if (ret) { | |
91e9c07b ED |
1240 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1241 | if (err >= 0) { | |
fcc5a03a | 1242 | err = ret; |
30e6c9fa | 1243 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1244 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1245 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1246 | dev->name_assign_type = old_assign_type; |
1247 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1248 | goto rollback; |
91e9c07b | 1249 | } else { |
7b6cd1ce | 1250 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1251 | dev->name, ret); |
fcc5a03a HX |
1252 | } |
1253 | } | |
1da177e4 LT |
1254 | |
1255 | return err; | |
1256 | } | |
1257 | ||
0b815a1a SH |
1258 | /** |
1259 | * dev_set_alias - change ifalias of a device | |
1260 | * @dev: device | |
1261 | * @alias: name up to IFALIASZ | |
f0db275a | 1262 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1263 | * |
1264 | * Set ifalias for a device, | |
1265 | */ | |
1266 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1267 | { | |
7364e445 AK |
1268 | char *new_ifalias; |
1269 | ||
0b815a1a SH |
1270 | ASSERT_RTNL(); |
1271 | ||
1272 | if (len >= IFALIASZ) | |
1273 | return -EINVAL; | |
1274 | ||
96ca4a2c | 1275 | if (!len) { |
388dfc2d SK |
1276 | kfree(dev->ifalias); |
1277 | dev->ifalias = NULL; | |
96ca4a2c OH |
1278 | return 0; |
1279 | } | |
1280 | ||
7364e445 AK |
1281 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1282 | if (!new_ifalias) | |
0b815a1a | 1283 | return -ENOMEM; |
7364e445 | 1284 | dev->ifalias = new_ifalias; |
c28294b9 AP |
1285 | memcpy(dev->ifalias, alias, len); |
1286 | dev->ifalias[len] = 0; | |
0b815a1a | 1287 | |
0b815a1a SH |
1288 | return len; |
1289 | } | |
1290 | ||
1291 | ||
d8a33ac4 | 1292 | /** |
3041a069 | 1293 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1294 | * @dev: device to cause notification |
1295 | * | |
1296 | * Called to indicate a device has changed features. | |
1297 | */ | |
1298 | void netdev_features_change(struct net_device *dev) | |
1299 | { | |
056925ab | 1300 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1301 | } |
1302 | EXPORT_SYMBOL(netdev_features_change); | |
1303 | ||
1da177e4 LT |
1304 | /** |
1305 | * netdev_state_change - device changes state | |
1306 | * @dev: device to cause notification | |
1307 | * | |
1308 | * Called to indicate a device has changed state. This function calls | |
1309 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1310 | * to the routing socket. | |
1311 | */ | |
1312 | void netdev_state_change(struct net_device *dev) | |
1313 | { | |
1314 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1315 | struct netdev_notifier_change_info change_info; |
1316 | ||
1317 | change_info.flags_changed = 0; | |
1318 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1319 | &change_info.info); | |
7f294054 | 1320 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1321 | } |
1322 | } | |
d1b19dff | 1323 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1324 | |
ee89bab1 | 1325 | /** |
722c9a0c | 1326 | * netdev_notify_peers - notify network peers about existence of @dev |
1327 | * @dev: network device | |
ee89bab1 AW |
1328 | * |
1329 | * Generate traffic such that interested network peers are aware of | |
1330 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1331 | * a device wants to inform the rest of the network about some sort of | |
1332 | * reconfiguration such as a failover event or virtual machine | |
1333 | * migration. | |
1334 | */ | |
1335 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1336 | { |
ee89bab1 AW |
1337 | rtnl_lock(); |
1338 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
37c343b4 | 1339 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); |
ee89bab1 | 1340 | rtnl_unlock(); |
c1da4ac7 | 1341 | } |
ee89bab1 | 1342 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1343 | |
bd380811 | 1344 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1345 | { |
d314774c | 1346 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1347 | int ret; |
1da177e4 | 1348 | |
e46b66bc BH |
1349 | ASSERT_RTNL(); |
1350 | ||
1da177e4 LT |
1351 | if (!netif_device_present(dev)) |
1352 | return -ENODEV; | |
1353 | ||
ca99ca14 NH |
1354 | /* Block netpoll from trying to do any rx path servicing. |
1355 | * If we don't do this there is a chance ndo_poll_controller | |
1356 | * or ndo_poll may be running while we open the device | |
1357 | */ | |
66b5552f | 1358 | netpoll_poll_disable(dev); |
ca99ca14 | 1359 | |
3b8bcfd5 JB |
1360 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1361 | ret = notifier_to_errno(ret); | |
1362 | if (ret) | |
1363 | return ret; | |
1364 | ||
1da177e4 | 1365 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1366 | |
d314774c SH |
1367 | if (ops->ndo_validate_addr) |
1368 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1369 | |
d314774c SH |
1370 | if (!ret && ops->ndo_open) |
1371 | ret = ops->ndo_open(dev); | |
1da177e4 | 1372 | |
66b5552f | 1373 | netpoll_poll_enable(dev); |
ca99ca14 | 1374 | |
bada339b JG |
1375 | if (ret) |
1376 | clear_bit(__LINK_STATE_START, &dev->state); | |
1377 | else { | |
1da177e4 | 1378 | dev->flags |= IFF_UP; |
4417da66 | 1379 | dev_set_rx_mode(dev); |
1da177e4 | 1380 | dev_activate(dev); |
7bf23575 | 1381 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1382 | } |
bada339b | 1383 | |
1da177e4 LT |
1384 | return ret; |
1385 | } | |
1386 | ||
1387 | /** | |
bd380811 PM |
1388 | * dev_open - prepare an interface for use. |
1389 | * @dev: device to open | |
1da177e4 | 1390 | * |
bd380811 PM |
1391 | * Takes a device from down to up state. The device's private open |
1392 | * function is invoked and then the multicast lists are loaded. Finally | |
1393 | * the device is moved into the up state and a %NETDEV_UP message is | |
1394 | * sent to the netdev notifier chain. | |
1395 | * | |
1396 | * Calling this function on an active interface is a nop. On a failure | |
1397 | * a negative errno code is returned. | |
1da177e4 | 1398 | */ |
bd380811 PM |
1399 | int dev_open(struct net_device *dev) |
1400 | { | |
1401 | int ret; | |
1402 | ||
bd380811 PM |
1403 | if (dev->flags & IFF_UP) |
1404 | return 0; | |
1405 | ||
bd380811 PM |
1406 | ret = __dev_open(dev); |
1407 | if (ret < 0) | |
1408 | return ret; | |
1409 | ||
7f294054 | 1410 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1411 | call_netdevice_notifiers(NETDEV_UP, dev); |
1412 | ||
1413 | return ret; | |
1414 | } | |
1415 | EXPORT_SYMBOL(dev_open); | |
1416 | ||
7051b88a | 1417 | static void __dev_close_many(struct list_head *head) |
1da177e4 | 1418 | { |
44345724 | 1419 | struct net_device *dev; |
e46b66bc | 1420 | |
bd380811 | 1421 | ASSERT_RTNL(); |
9d5010db DM |
1422 | might_sleep(); |
1423 | ||
5cde2829 | 1424 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1425 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1426 | netpoll_poll_disable(dev); |
3f4df206 | 1427 | |
44345724 | 1428 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1429 | |
44345724 | 1430 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1431 | |
44345724 OP |
1432 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1433 | * can be even on different cpu. So just clear netif_running(). | |
1434 | * | |
1435 | * dev->stop() will invoke napi_disable() on all of it's | |
1436 | * napi_struct instances on this device. | |
1437 | */ | |
4e857c58 | 1438 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1439 | } |
1da177e4 | 1440 | |
44345724 | 1441 | dev_deactivate_many(head); |
d8b2a4d2 | 1442 | |
5cde2829 | 1443 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1444 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1445 | |
44345724 OP |
1446 | /* |
1447 | * Call the device specific close. This cannot fail. | |
1448 | * Only if device is UP | |
1449 | * | |
1450 | * We allow it to be called even after a DETACH hot-plug | |
1451 | * event. | |
1452 | */ | |
1453 | if (ops->ndo_stop) | |
1454 | ops->ndo_stop(dev); | |
1455 | ||
44345724 | 1456 | dev->flags &= ~IFF_UP; |
66b5552f | 1457 | netpoll_poll_enable(dev); |
44345724 | 1458 | } |
44345724 OP |
1459 | } |
1460 | ||
7051b88a | 1461 | static void __dev_close(struct net_device *dev) |
44345724 OP |
1462 | { |
1463 | LIST_HEAD(single); | |
1464 | ||
5cde2829 | 1465 | list_add(&dev->close_list, &single); |
7051b88a | 1466 | __dev_close_many(&single); |
f87e6f47 | 1467 | list_del(&single); |
44345724 OP |
1468 | } |
1469 | ||
7051b88a | 1470 | void dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1471 | { |
1472 | struct net_device *dev, *tmp; | |
1da177e4 | 1473 | |
5cde2829 EB |
1474 | /* Remove the devices that don't need to be closed */ |
1475 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1476 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1477 | list_del_init(&dev->close_list); |
44345724 OP |
1478 | |
1479 | __dev_close_many(head); | |
1da177e4 | 1480 | |
5cde2829 | 1481 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1482 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1483 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1484 | if (unlink) |
1485 | list_del_init(&dev->close_list); | |
44345724 | 1486 | } |
bd380811 | 1487 | } |
99c4a26a | 1488 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1489 | |
1490 | /** | |
1491 | * dev_close - shutdown an interface. | |
1492 | * @dev: device to shutdown | |
1493 | * | |
1494 | * This function moves an active device into down state. A | |
1495 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1496 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1497 | * chain. | |
1498 | */ | |
7051b88a | 1499 | void dev_close(struct net_device *dev) |
bd380811 | 1500 | { |
e14a5993 ED |
1501 | if (dev->flags & IFF_UP) { |
1502 | LIST_HEAD(single); | |
1da177e4 | 1503 | |
5cde2829 | 1504 | list_add(&dev->close_list, &single); |
99c4a26a | 1505 | dev_close_many(&single, true); |
e14a5993 ED |
1506 | list_del(&single); |
1507 | } | |
1da177e4 | 1508 | } |
d1b19dff | 1509 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1510 | |
1511 | ||
0187bdfb BH |
1512 | /** |
1513 | * dev_disable_lro - disable Large Receive Offload on a device | |
1514 | * @dev: device | |
1515 | * | |
1516 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1517 | * called under RTNL. This is needed if received packets may be | |
1518 | * forwarded to another interface. | |
1519 | */ | |
1520 | void dev_disable_lro(struct net_device *dev) | |
1521 | { | |
fbe168ba MK |
1522 | struct net_device *lower_dev; |
1523 | struct list_head *iter; | |
529d0489 | 1524 | |
bc5787c6 MM |
1525 | dev->wanted_features &= ~NETIF_F_LRO; |
1526 | netdev_update_features(dev); | |
27660515 | 1527 | |
22d5969f MM |
1528 | if (unlikely(dev->features & NETIF_F_LRO)) |
1529 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1530 | |
1531 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1532 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1533 | } |
1534 | EXPORT_SYMBOL(dev_disable_lro); | |
1535 | ||
351638e7 JP |
1536 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1537 | struct net_device *dev) | |
1538 | { | |
1539 | struct netdev_notifier_info info; | |
1540 | ||
1541 | netdev_notifier_info_init(&info, dev); | |
1542 | return nb->notifier_call(nb, val, &info); | |
1543 | } | |
0187bdfb | 1544 | |
881d966b EB |
1545 | static int dev_boot_phase = 1; |
1546 | ||
1da177e4 | 1547 | /** |
722c9a0c | 1548 | * register_netdevice_notifier - register a network notifier block |
1549 | * @nb: notifier | |
1da177e4 | 1550 | * |
722c9a0c | 1551 | * Register a notifier to be called when network device events occur. |
1552 | * The notifier passed is linked into the kernel structures and must | |
1553 | * not be reused until it has been unregistered. A negative errno code | |
1554 | * is returned on a failure. | |
1da177e4 | 1555 | * |
722c9a0c | 1556 | * When registered all registration and up events are replayed |
1557 | * to the new notifier to allow device to have a race free | |
1558 | * view of the network device list. | |
1da177e4 LT |
1559 | */ |
1560 | ||
1561 | int register_netdevice_notifier(struct notifier_block *nb) | |
1562 | { | |
1563 | struct net_device *dev; | |
fcc5a03a | 1564 | struct net_device *last; |
881d966b | 1565 | struct net *net; |
1da177e4 LT |
1566 | int err; |
1567 | ||
1568 | rtnl_lock(); | |
f07d5b94 | 1569 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1570 | if (err) |
1571 | goto unlock; | |
881d966b EB |
1572 | if (dev_boot_phase) |
1573 | goto unlock; | |
1574 | for_each_net(net) { | |
1575 | for_each_netdev(net, dev) { | |
351638e7 | 1576 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1577 | err = notifier_to_errno(err); |
1578 | if (err) | |
1579 | goto rollback; | |
1580 | ||
1581 | if (!(dev->flags & IFF_UP)) | |
1582 | continue; | |
1da177e4 | 1583 | |
351638e7 | 1584 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1585 | } |
1da177e4 | 1586 | } |
fcc5a03a HX |
1587 | |
1588 | unlock: | |
1da177e4 LT |
1589 | rtnl_unlock(); |
1590 | return err; | |
fcc5a03a HX |
1591 | |
1592 | rollback: | |
1593 | last = dev; | |
881d966b EB |
1594 | for_each_net(net) { |
1595 | for_each_netdev(net, dev) { | |
1596 | if (dev == last) | |
8f891489 | 1597 | goto outroll; |
fcc5a03a | 1598 | |
881d966b | 1599 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1600 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1601 | dev); | |
1602 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1603 | } |
351638e7 | 1604 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1605 | } |
fcc5a03a | 1606 | } |
c67625a1 | 1607 | |
8f891489 | 1608 | outroll: |
c67625a1 | 1609 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1610 | goto unlock; |
1da177e4 | 1611 | } |
d1b19dff | 1612 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1613 | |
1614 | /** | |
722c9a0c | 1615 | * unregister_netdevice_notifier - unregister a network notifier block |
1616 | * @nb: notifier | |
1da177e4 | 1617 | * |
722c9a0c | 1618 | * Unregister a notifier previously registered by |
1619 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1620 | * kernel structures and may then be reused. A negative errno code | |
1621 | * is returned on a failure. | |
7d3d43da | 1622 | * |
722c9a0c | 1623 | * After unregistering unregister and down device events are synthesized |
1624 | * for all devices on the device list to the removed notifier to remove | |
1625 | * the need for special case cleanup code. | |
1da177e4 LT |
1626 | */ |
1627 | ||
1628 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1629 | { | |
7d3d43da EB |
1630 | struct net_device *dev; |
1631 | struct net *net; | |
9f514950 HX |
1632 | int err; |
1633 | ||
1634 | rtnl_lock(); | |
f07d5b94 | 1635 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1636 | if (err) |
1637 | goto unlock; | |
1638 | ||
1639 | for_each_net(net) { | |
1640 | for_each_netdev(net, dev) { | |
1641 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1642 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1643 | dev); | |
1644 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1645 | } |
351638e7 | 1646 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1647 | } |
1648 | } | |
1649 | unlock: | |
9f514950 HX |
1650 | rtnl_unlock(); |
1651 | return err; | |
1da177e4 | 1652 | } |
d1b19dff | 1653 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1654 | |
351638e7 JP |
1655 | /** |
1656 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1657 | * @val: value passed unmodified to notifier function | |
1658 | * @dev: net_device pointer passed unmodified to notifier function | |
1659 | * @info: notifier information data | |
1660 | * | |
1661 | * Call all network notifier blocks. Parameters and return value | |
1662 | * are as for raw_notifier_call_chain(). | |
1663 | */ | |
1664 | ||
1d143d9f | 1665 | static int call_netdevice_notifiers_info(unsigned long val, |
1666 | struct net_device *dev, | |
1667 | struct netdev_notifier_info *info) | |
351638e7 JP |
1668 | { |
1669 | ASSERT_RTNL(); | |
1670 | netdev_notifier_info_init(info, dev); | |
1671 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1672 | } | |
351638e7 | 1673 | |
1da177e4 LT |
1674 | /** |
1675 | * call_netdevice_notifiers - call all network notifier blocks | |
1676 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1677 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1678 | * |
1679 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1680 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1681 | */ |
1682 | ||
ad7379d4 | 1683 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1684 | { |
351638e7 JP |
1685 | struct netdev_notifier_info info; |
1686 | ||
1687 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1688 | } |
edf947f1 | 1689 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1690 | |
1cf51900 | 1691 | #ifdef CONFIG_NET_INGRESS |
4577139b DB |
1692 | static struct static_key ingress_needed __read_mostly; |
1693 | ||
1694 | void net_inc_ingress_queue(void) | |
1695 | { | |
1696 | static_key_slow_inc(&ingress_needed); | |
1697 | } | |
1698 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
1699 | ||
1700 | void net_dec_ingress_queue(void) | |
1701 | { | |
1702 | static_key_slow_dec(&ingress_needed); | |
1703 | } | |
1704 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
1705 | #endif | |
1706 | ||
1f211a1b DB |
1707 | #ifdef CONFIG_NET_EGRESS |
1708 | static struct static_key egress_needed __read_mostly; | |
1709 | ||
1710 | void net_inc_egress_queue(void) | |
1711 | { | |
1712 | static_key_slow_inc(&egress_needed); | |
1713 | } | |
1714 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
1715 | ||
1716 | void net_dec_egress_queue(void) | |
1717 | { | |
1718 | static_key_slow_dec(&egress_needed); | |
1719 | } | |
1720 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
1721 | #endif | |
1722 | ||
c5905afb | 1723 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1724 | #ifdef HAVE_JUMP_LABEL |
b90e5794 | 1725 | static atomic_t netstamp_needed_deferred; |
13baa00a | 1726 | static atomic_t netstamp_wanted; |
5fa8bbda | 1727 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 1728 | { |
b90e5794 | 1729 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 1730 | int wanted; |
b90e5794 | 1731 | |
13baa00a ED |
1732 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
1733 | if (wanted > 0) | |
1734 | static_key_enable(&netstamp_needed); | |
1735 | else | |
1736 | static_key_disable(&netstamp_needed); | |
5fa8bbda ED |
1737 | } |
1738 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 1739 | #endif |
5fa8bbda ED |
1740 | |
1741 | void net_enable_timestamp(void) | |
1742 | { | |
13baa00a ED |
1743 | #ifdef HAVE_JUMP_LABEL |
1744 | int wanted; | |
1745 | ||
1746 | while (1) { | |
1747 | wanted = atomic_read(&netstamp_wanted); | |
1748 | if (wanted <= 0) | |
1749 | break; | |
1750 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) | |
1751 | return; | |
1752 | } | |
1753 | atomic_inc(&netstamp_needed_deferred); | |
1754 | schedule_work(&netstamp_work); | |
1755 | #else | |
c5905afb | 1756 | static_key_slow_inc(&netstamp_needed); |
13baa00a | 1757 | #endif |
1da177e4 | 1758 | } |
d1b19dff | 1759 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1760 | |
1761 | void net_disable_timestamp(void) | |
1762 | { | |
b90e5794 | 1763 | #ifdef HAVE_JUMP_LABEL |
13baa00a ED |
1764 | int wanted; |
1765 | ||
1766 | while (1) { | |
1767 | wanted = atomic_read(&netstamp_wanted); | |
1768 | if (wanted <= 1) | |
1769 | break; | |
1770 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) | |
1771 | return; | |
1772 | } | |
1773 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
1774 | schedule_work(&netstamp_work); |
1775 | #else | |
c5905afb | 1776 | static_key_slow_dec(&netstamp_needed); |
5fa8bbda | 1777 | #endif |
1da177e4 | 1778 | } |
d1b19dff | 1779 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1780 | |
3b098e2d | 1781 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1782 | { |
2456e855 | 1783 | skb->tstamp = 0; |
c5905afb | 1784 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1785 | __net_timestamp(skb); |
1da177e4 LT |
1786 | } |
1787 | ||
588f0330 | 1788 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1789 | if (static_key_false(&netstamp_needed)) { \ |
2456e855 | 1790 | if ((COND) && !(SKB)->tstamp) \ |
588f0330 ED |
1791 | __net_timestamp(SKB); \ |
1792 | } \ | |
3b098e2d | 1793 | |
f4b05d27 | 1794 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 DL |
1795 | { |
1796 | unsigned int len; | |
1797 | ||
1798 | if (!(dev->flags & IFF_UP)) | |
1799 | return false; | |
1800 | ||
1801 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1802 | if (skb->len <= len) | |
1803 | return true; | |
1804 | ||
1805 | /* if TSO is enabled, we don't care about the length as the packet | |
1806 | * could be forwarded without being segmented before | |
1807 | */ | |
1808 | if (skb_is_gso(skb)) | |
1809 | return true; | |
1810 | ||
1811 | return false; | |
1812 | } | |
1ee481fb | 1813 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1814 | |
a0265d28 HX |
1815 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1816 | { | |
4e3264d2 | 1817 | int ret = ____dev_forward_skb(dev, skb); |
a0265d28 | 1818 | |
4e3264d2 MKL |
1819 | if (likely(!ret)) { |
1820 | skb->protocol = eth_type_trans(skb, dev); | |
1821 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
1822 | } | |
a0265d28 | 1823 | |
4e3264d2 | 1824 | return ret; |
a0265d28 HX |
1825 | } |
1826 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1827 | ||
44540960 AB |
1828 | /** |
1829 | * dev_forward_skb - loopback an skb to another netif | |
1830 | * | |
1831 | * @dev: destination network device | |
1832 | * @skb: buffer to forward | |
1833 | * | |
1834 | * return values: | |
1835 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1836 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1837 | * |
1838 | * dev_forward_skb can be used for injecting an skb from the | |
1839 | * start_xmit function of one device into the receive queue | |
1840 | * of another device. | |
1841 | * | |
1842 | * The receiving device may be in another namespace, so | |
1843 | * we have to clear all information in the skb that could | |
1844 | * impact namespace isolation. | |
1845 | */ | |
1846 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1847 | { | |
a0265d28 | 1848 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1849 | } |
1850 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1851 | ||
71d9dec2 CG |
1852 | static inline int deliver_skb(struct sk_buff *skb, |
1853 | struct packet_type *pt_prev, | |
1854 | struct net_device *orig_dev) | |
1855 | { | |
1f8b977a | 1856 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
1080e512 | 1857 | return -ENOMEM; |
63354797 | 1858 | refcount_inc(&skb->users); |
71d9dec2 CG |
1859 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1860 | } | |
1861 | ||
7866a621 SN |
1862 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
1863 | struct packet_type **pt, | |
fbcb2170 JP |
1864 | struct net_device *orig_dev, |
1865 | __be16 type, | |
7866a621 SN |
1866 | struct list_head *ptype_list) |
1867 | { | |
1868 | struct packet_type *ptype, *pt_prev = *pt; | |
1869 | ||
1870 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1871 | if (ptype->type != type) | |
1872 | continue; | |
1873 | if (pt_prev) | |
fbcb2170 | 1874 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
1875 | pt_prev = ptype; |
1876 | } | |
1877 | *pt = pt_prev; | |
1878 | } | |
1879 | ||
c0de08d0 EL |
1880 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1881 | { | |
a3d744e9 | 1882 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1883 | return false; |
1884 | ||
1885 | if (ptype->id_match) | |
1886 | return ptype->id_match(ptype, skb->sk); | |
1887 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1888 | return true; | |
1889 | ||
1890 | return false; | |
1891 | } | |
1892 | ||
1da177e4 LT |
1893 | /* |
1894 | * Support routine. Sends outgoing frames to any network | |
1895 | * taps currently in use. | |
1896 | */ | |
1897 | ||
74b20582 | 1898 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1899 | { |
1900 | struct packet_type *ptype; | |
71d9dec2 CG |
1901 | struct sk_buff *skb2 = NULL; |
1902 | struct packet_type *pt_prev = NULL; | |
7866a621 | 1903 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 1904 | |
1da177e4 | 1905 | rcu_read_lock(); |
7866a621 SN |
1906 | again: |
1907 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1da177e4 LT |
1908 | /* Never send packets back to the socket |
1909 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1910 | */ | |
7866a621 SN |
1911 | if (skb_loop_sk(ptype, skb)) |
1912 | continue; | |
71d9dec2 | 1913 | |
7866a621 SN |
1914 | if (pt_prev) { |
1915 | deliver_skb(skb2, pt_prev, skb->dev); | |
1916 | pt_prev = ptype; | |
1917 | continue; | |
1918 | } | |
1da177e4 | 1919 | |
7866a621 SN |
1920 | /* need to clone skb, done only once */ |
1921 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1922 | if (!skb2) | |
1923 | goto out_unlock; | |
70978182 | 1924 | |
7866a621 | 1925 | net_timestamp_set(skb2); |
1da177e4 | 1926 | |
7866a621 SN |
1927 | /* skb->nh should be correctly |
1928 | * set by sender, so that the second statement is | |
1929 | * just protection against buggy protocols. | |
1930 | */ | |
1931 | skb_reset_mac_header(skb2); | |
1932 | ||
1933 | if (skb_network_header(skb2) < skb2->data || | |
1934 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
1935 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
1936 | ntohs(skb2->protocol), | |
1937 | dev->name); | |
1938 | skb_reset_network_header(skb2); | |
1da177e4 | 1939 | } |
7866a621 SN |
1940 | |
1941 | skb2->transport_header = skb2->network_header; | |
1942 | skb2->pkt_type = PACKET_OUTGOING; | |
1943 | pt_prev = ptype; | |
1944 | } | |
1945 | ||
1946 | if (ptype_list == &ptype_all) { | |
1947 | ptype_list = &dev->ptype_all; | |
1948 | goto again; | |
1da177e4 | 1949 | } |
7866a621 | 1950 | out_unlock: |
581fe0ea WB |
1951 | if (pt_prev) { |
1952 | if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) | |
1953 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1954 | else | |
1955 | kfree_skb(skb2); | |
1956 | } | |
1da177e4 LT |
1957 | rcu_read_unlock(); |
1958 | } | |
74b20582 | 1959 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 1960 | |
2c53040f BH |
1961 | /** |
1962 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1963 | * @dev: Network device |
1964 | * @txq: number of queues available | |
1965 | * | |
1966 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1967 | * valid. To resolve this verify the tc mapping remains valid and if | |
1968 | * not NULL the mapping. With no priorities mapping to this | |
1969 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1970 | * is invalid nothing can be done so disable priority mappings. If is | |
1971 | * expected that drivers will fix this mapping if they can before | |
1972 | * calling netif_set_real_num_tx_queues. | |
1973 | */ | |
bb134d22 | 1974 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1975 | { |
1976 | int i; | |
1977 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1978 | ||
1979 | /* If TC0 is invalidated disable TC mapping */ | |
1980 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1981 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1982 | dev->num_tc = 0; |
1983 | return; | |
1984 | } | |
1985 | ||
1986 | /* Invalidated prio to tc mappings set to TC0 */ | |
1987 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1988 | int q = netdev_get_prio_tc_map(dev, i); | |
1989 | ||
1990 | tc = &dev->tc_to_txq[q]; | |
1991 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1992 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1993 | i, q); | |
4f57c087 JF |
1994 | netdev_set_prio_tc_map(dev, i, 0); |
1995 | } | |
1996 | } | |
1997 | } | |
1998 | ||
8d059b0f AD |
1999 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
2000 | { | |
2001 | if (dev->num_tc) { | |
2002 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2003 | int i; | |
2004 | ||
2005 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { | |
2006 | if ((txq - tc->offset) < tc->count) | |
2007 | return i; | |
2008 | } | |
2009 | ||
2010 | return -1; | |
2011 | } | |
2012 | ||
2013 | return 0; | |
2014 | } | |
2015 | ||
537c00de AD |
2016 | #ifdef CONFIG_XPS |
2017 | static DEFINE_MUTEX(xps_map_mutex); | |
2018 | #define xmap_dereference(P) \ | |
2019 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
2020 | ||
6234f874 AD |
2021 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
2022 | int tci, u16 index) | |
537c00de | 2023 | { |
10cdc3f3 AD |
2024 | struct xps_map *map = NULL; |
2025 | int pos; | |
537c00de | 2026 | |
10cdc3f3 | 2027 | if (dev_maps) |
6234f874 AD |
2028 | map = xmap_dereference(dev_maps->cpu_map[tci]); |
2029 | if (!map) | |
2030 | return false; | |
537c00de | 2031 | |
6234f874 AD |
2032 | for (pos = map->len; pos--;) { |
2033 | if (map->queues[pos] != index) | |
2034 | continue; | |
2035 | ||
2036 | if (map->len > 1) { | |
2037 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2038 | break; |
537c00de | 2039 | } |
6234f874 AD |
2040 | |
2041 | RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL); | |
2042 | kfree_rcu(map, rcu); | |
2043 | return false; | |
537c00de AD |
2044 | } |
2045 | ||
6234f874 | 2046 | return true; |
10cdc3f3 AD |
2047 | } |
2048 | ||
6234f874 AD |
2049 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2050 | struct xps_dev_maps *dev_maps, | |
2051 | int cpu, u16 offset, u16 count) | |
2052 | { | |
184c449f AD |
2053 | int num_tc = dev->num_tc ? : 1; |
2054 | bool active = false; | |
2055 | int tci; | |
6234f874 | 2056 | |
184c449f AD |
2057 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2058 | int i, j; | |
2059 | ||
2060 | for (i = count, j = offset; i--; j++) { | |
2061 | if (!remove_xps_queue(dev_maps, cpu, j)) | |
2062 | break; | |
2063 | } | |
2064 | ||
2065 | active |= i < 0; | |
6234f874 AD |
2066 | } |
2067 | ||
184c449f | 2068 | return active; |
6234f874 AD |
2069 | } |
2070 | ||
2071 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, | |
2072 | u16 count) | |
10cdc3f3 AD |
2073 | { |
2074 | struct xps_dev_maps *dev_maps; | |
024e9679 | 2075 | int cpu, i; |
10cdc3f3 AD |
2076 | bool active = false; |
2077 | ||
2078 | mutex_lock(&xps_map_mutex); | |
2079 | dev_maps = xmap_dereference(dev->xps_maps); | |
2080 | ||
2081 | if (!dev_maps) | |
2082 | goto out_no_maps; | |
2083 | ||
6234f874 AD |
2084 | for_each_possible_cpu(cpu) |
2085 | active |= remove_xps_queue_cpu(dev, dev_maps, cpu, | |
2086 | offset, count); | |
10cdc3f3 AD |
2087 | |
2088 | if (!active) { | |
537c00de AD |
2089 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
2090 | kfree_rcu(dev_maps, rcu); | |
2091 | } | |
2092 | ||
6234f874 | 2093 | for (i = offset + (count - 1); count--; i--) |
024e9679 AD |
2094 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), |
2095 | NUMA_NO_NODE); | |
2096 | ||
537c00de AD |
2097 | out_no_maps: |
2098 | mutex_unlock(&xps_map_mutex); | |
2099 | } | |
2100 | ||
6234f874 AD |
2101 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2102 | { | |
2103 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2104 | } | |
2105 | ||
01c5f864 AD |
2106 | static struct xps_map *expand_xps_map(struct xps_map *map, |
2107 | int cpu, u16 index) | |
2108 | { | |
2109 | struct xps_map *new_map; | |
2110 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2111 | int i, pos; | |
2112 | ||
2113 | for (pos = 0; map && pos < map->len; pos++) { | |
2114 | if (map->queues[pos] != index) | |
2115 | continue; | |
2116 | return map; | |
2117 | } | |
2118 | ||
2119 | /* Need to add queue to this CPU's existing map */ | |
2120 | if (map) { | |
2121 | if (pos < map->alloc_len) | |
2122 | return map; | |
2123 | ||
2124 | alloc_len = map->alloc_len * 2; | |
2125 | } | |
2126 | ||
2127 | /* Need to allocate new map to store queue on this CPU's map */ | |
2128 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2129 | cpu_to_node(cpu)); | |
2130 | if (!new_map) | |
2131 | return NULL; | |
2132 | ||
2133 | for (i = 0; i < pos; i++) | |
2134 | new_map->queues[i] = map->queues[i]; | |
2135 | new_map->alloc_len = alloc_len; | |
2136 | new_map->len = pos; | |
2137 | ||
2138 | return new_map; | |
2139 | } | |
2140 | ||
3573540c MT |
2141 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
2142 | u16 index) | |
537c00de | 2143 | { |
01c5f864 | 2144 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
184c449f AD |
2145 | int i, cpu, tci, numa_node_id = -2; |
2146 | int maps_sz, num_tc = 1, tc = 0; | |
537c00de | 2147 | struct xps_map *map, *new_map; |
01c5f864 | 2148 | bool active = false; |
537c00de | 2149 | |
184c449f AD |
2150 | if (dev->num_tc) { |
2151 | num_tc = dev->num_tc; | |
2152 | tc = netdev_txq_to_tc(dev, index); | |
2153 | if (tc < 0) | |
2154 | return -EINVAL; | |
2155 | } | |
2156 | ||
2157 | maps_sz = XPS_DEV_MAPS_SIZE(num_tc); | |
2158 | if (maps_sz < L1_CACHE_BYTES) | |
2159 | maps_sz = L1_CACHE_BYTES; | |
2160 | ||
537c00de AD |
2161 | mutex_lock(&xps_map_mutex); |
2162 | ||
2163 | dev_maps = xmap_dereference(dev->xps_maps); | |
2164 | ||
01c5f864 | 2165 | /* allocate memory for queue storage */ |
184c449f | 2166 | for_each_cpu_and(cpu, cpu_online_mask, mask) { |
01c5f864 AD |
2167 | if (!new_dev_maps) |
2168 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2169 | if (!new_dev_maps) { |
2170 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2171 | return -ENOMEM; |
2bb60cb9 | 2172 | } |
01c5f864 | 2173 | |
184c449f AD |
2174 | tci = cpu * num_tc + tc; |
2175 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) : | |
01c5f864 AD |
2176 | NULL; |
2177 | ||
2178 | map = expand_xps_map(map, cpu, index); | |
2179 | if (!map) | |
2180 | goto error; | |
2181 | ||
184c449f | 2182 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); |
01c5f864 AD |
2183 | } |
2184 | ||
2185 | if (!new_dev_maps) | |
2186 | goto out_no_new_maps; | |
2187 | ||
537c00de | 2188 | for_each_possible_cpu(cpu) { |
184c449f AD |
2189 | /* copy maps belonging to foreign traffic classes */ |
2190 | for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) { | |
2191 | /* fill in the new device map from the old device map */ | |
2192 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
2193 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
2194 | } | |
2195 | ||
2196 | /* We need to explicitly update tci as prevous loop | |
2197 | * could break out early if dev_maps is NULL. | |
2198 | */ | |
2199 | tci = cpu * num_tc + tc; | |
2200 | ||
01c5f864 AD |
2201 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
2202 | /* add queue to CPU maps */ | |
2203 | int pos = 0; | |
2204 | ||
184c449f | 2205 | map = xmap_dereference(new_dev_maps->cpu_map[tci]); |
01c5f864 AD |
2206 | while ((pos < map->len) && (map->queues[pos] != index)) |
2207 | pos++; | |
2208 | ||
2209 | if (pos == map->len) | |
2210 | map->queues[map->len++] = index; | |
537c00de | 2211 | #ifdef CONFIG_NUMA |
537c00de AD |
2212 | if (numa_node_id == -2) |
2213 | numa_node_id = cpu_to_node(cpu); | |
2214 | else if (numa_node_id != cpu_to_node(cpu)) | |
2215 | numa_node_id = -1; | |
537c00de | 2216 | #endif |
01c5f864 AD |
2217 | } else if (dev_maps) { |
2218 | /* fill in the new device map from the old device map */ | |
184c449f AD |
2219 | map = xmap_dereference(dev_maps->cpu_map[tci]); |
2220 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
537c00de | 2221 | } |
01c5f864 | 2222 | |
184c449f AD |
2223 | /* copy maps belonging to foreign traffic classes */ |
2224 | for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { | |
2225 | /* fill in the new device map from the old device map */ | |
2226 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
2227 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
2228 | } | |
537c00de AD |
2229 | } |
2230 | ||
01c5f864 AD |
2231 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2232 | ||
537c00de | 2233 | /* Cleanup old maps */ |
184c449f AD |
2234 | if (!dev_maps) |
2235 | goto out_no_old_maps; | |
2236 | ||
2237 | for_each_possible_cpu(cpu) { | |
2238 | for (i = num_tc, tci = cpu * num_tc; i--; tci++) { | |
2239 | new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); | |
2240 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
01c5f864 AD |
2241 | if (map && map != new_map) |
2242 | kfree_rcu(map, rcu); | |
2243 | } | |
537c00de AD |
2244 | } |
2245 | ||
184c449f AD |
2246 | kfree_rcu(dev_maps, rcu); |
2247 | ||
2248 | out_no_old_maps: | |
01c5f864 AD |
2249 | dev_maps = new_dev_maps; |
2250 | active = true; | |
537c00de | 2251 | |
01c5f864 AD |
2252 | out_no_new_maps: |
2253 | /* update Tx queue numa node */ | |
537c00de AD |
2254 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2255 | (numa_node_id >= 0) ? numa_node_id : | |
2256 | NUMA_NO_NODE); | |
2257 | ||
01c5f864 AD |
2258 | if (!dev_maps) |
2259 | goto out_no_maps; | |
2260 | ||
2261 | /* removes queue from unused CPUs */ | |
2262 | for_each_possible_cpu(cpu) { | |
184c449f AD |
2263 | for (i = tc, tci = cpu * num_tc; i--; tci++) |
2264 | active |= remove_xps_queue(dev_maps, tci, index); | |
2265 | if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu)) | |
2266 | active |= remove_xps_queue(dev_maps, tci, index); | |
2267 | for (i = num_tc - tc, tci++; --i; tci++) | |
2268 | active |= remove_xps_queue(dev_maps, tci, index); | |
01c5f864 AD |
2269 | } |
2270 | ||
2271 | /* free map if not active */ | |
2272 | if (!active) { | |
2273 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2274 | kfree_rcu(dev_maps, rcu); | |
2275 | } | |
2276 | ||
2277 | out_no_maps: | |
537c00de AD |
2278 | mutex_unlock(&xps_map_mutex); |
2279 | ||
2280 | return 0; | |
2281 | error: | |
01c5f864 AD |
2282 | /* remove any maps that we added */ |
2283 | for_each_possible_cpu(cpu) { | |
184c449f AD |
2284 | for (i = num_tc, tci = cpu * num_tc; i--; tci++) { |
2285 | new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); | |
2286 | map = dev_maps ? | |
2287 | xmap_dereference(dev_maps->cpu_map[tci]) : | |
2288 | NULL; | |
2289 | if (new_map && new_map != map) | |
2290 | kfree(new_map); | |
2291 | } | |
01c5f864 AD |
2292 | } |
2293 | ||
537c00de AD |
2294 | mutex_unlock(&xps_map_mutex); |
2295 | ||
537c00de AD |
2296 | kfree(new_dev_maps); |
2297 | return -ENOMEM; | |
2298 | } | |
2299 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2300 | ||
2301 | #endif | |
9cf1f6a8 AD |
2302 | void netdev_reset_tc(struct net_device *dev) |
2303 | { | |
6234f874 AD |
2304 | #ifdef CONFIG_XPS |
2305 | netif_reset_xps_queues_gt(dev, 0); | |
2306 | #endif | |
9cf1f6a8 AD |
2307 | dev->num_tc = 0; |
2308 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2309 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2310 | } | |
2311 | EXPORT_SYMBOL(netdev_reset_tc); | |
2312 | ||
2313 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2314 | { | |
2315 | if (tc >= dev->num_tc) | |
2316 | return -EINVAL; | |
2317 | ||
6234f874 AD |
2318 | #ifdef CONFIG_XPS |
2319 | netif_reset_xps_queues(dev, offset, count); | |
2320 | #endif | |
9cf1f6a8 AD |
2321 | dev->tc_to_txq[tc].count = count; |
2322 | dev->tc_to_txq[tc].offset = offset; | |
2323 | return 0; | |
2324 | } | |
2325 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2326 | ||
2327 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2328 | { | |
2329 | if (num_tc > TC_MAX_QUEUE) | |
2330 | return -EINVAL; | |
2331 | ||
6234f874 AD |
2332 | #ifdef CONFIG_XPS |
2333 | netif_reset_xps_queues_gt(dev, 0); | |
2334 | #endif | |
9cf1f6a8 AD |
2335 | dev->num_tc = num_tc; |
2336 | return 0; | |
2337 | } | |
2338 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2339 | ||
f0796d5c JF |
2340 | /* |
2341 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2342 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2343 | */ | |
e6484930 | 2344 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2345 | { |
1d24eb48 TH |
2346 | int rc; |
2347 | ||
e6484930 TH |
2348 | if (txq < 1 || txq > dev->num_tx_queues) |
2349 | return -EINVAL; | |
f0796d5c | 2350 | |
5c56580b BH |
2351 | if (dev->reg_state == NETREG_REGISTERED || |
2352 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2353 | ASSERT_RTNL(); |
2354 | ||
1d24eb48 TH |
2355 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2356 | txq); | |
bf264145 TH |
2357 | if (rc) |
2358 | return rc; | |
2359 | ||
4f57c087 JF |
2360 | if (dev->num_tc) |
2361 | netif_setup_tc(dev, txq); | |
2362 | ||
024e9679 | 2363 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2364 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2365 | #ifdef CONFIG_XPS |
2366 | netif_reset_xps_queues_gt(dev, txq); | |
2367 | #endif | |
2368 | } | |
f0796d5c | 2369 | } |
e6484930 TH |
2370 | |
2371 | dev->real_num_tx_queues = txq; | |
2372 | return 0; | |
f0796d5c JF |
2373 | } |
2374 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2375 | |
a953be53 | 2376 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2377 | /** |
2378 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2379 | * @dev: Network device | |
2380 | * @rxq: Actual number of RX queues | |
2381 | * | |
2382 | * This must be called either with the rtnl_lock held or before | |
2383 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2384 | * negative error code. If called before registration, it always |
2385 | * succeeds. | |
62fe0b40 BH |
2386 | */ |
2387 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2388 | { | |
2389 | int rc; | |
2390 | ||
bd25fa7b TH |
2391 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2392 | return -EINVAL; | |
2393 | ||
62fe0b40 BH |
2394 | if (dev->reg_state == NETREG_REGISTERED) { |
2395 | ASSERT_RTNL(); | |
2396 | ||
62fe0b40 BH |
2397 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2398 | rxq); | |
2399 | if (rc) | |
2400 | return rc; | |
62fe0b40 BH |
2401 | } |
2402 | ||
2403 | dev->real_num_rx_queues = rxq; | |
2404 | return 0; | |
2405 | } | |
2406 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2407 | #endif | |
2408 | ||
2c53040f BH |
2409 | /** |
2410 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2411 | * |
2412 | * This routine should set an upper limit on the number of RSS queues | |
2413 | * used by default by multiqueue devices. | |
2414 | */ | |
a55b138b | 2415 | int netif_get_num_default_rss_queues(void) |
16917b87 | 2416 | { |
40e4e713 HS |
2417 | return is_kdump_kernel() ? |
2418 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
2419 | } |
2420 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2421 | ||
3bcb846c | 2422 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 2423 | { |
def82a1d JP |
2424 | struct softnet_data *sd; |
2425 | unsigned long flags; | |
56079431 | 2426 | |
def82a1d | 2427 | local_irq_save(flags); |
903ceff7 | 2428 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2429 | q->next_sched = NULL; |
2430 | *sd->output_queue_tailp = q; | |
2431 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2432 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2433 | local_irq_restore(flags); | |
2434 | } | |
2435 | ||
2436 | void __netif_schedule(struct Qdisc *q) | |
2437 | { | |
2438 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2439 | __netif_reschedule(q); | |
56079431 DV |
2440 | } |
2441 | EXPORT_SYMBOL(__netif_schedule); | |
2442 | ||
e6247027 ED |
2443 | struct dev_kfree_skb_cb { |
2444 | enum skb_free_reason reason; | |
2445 | }; | |
2446 | ||
2447 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2448 | { |
e6247027 ED |
2449 | return (struct dev_kfree_skb_cb *)skb->cb; |
2450 | } | |
2451 | ||
46e5da40 JF |
2452 | void netif_schedule_queue(struct netdev_queue *txq) |
2453 | { | |
2454 | rcu_read_lock(); | |
2455 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2456 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2457 | ||
2458 | __netif_schedule(q); | |
2459 | } | |
2460 | rcu_read_unlock(); | |
2461 | } | |
2462 | EXPORT_SYMBOL(netif_schedule_queue); | |
2463 | ||
46e5da40 JF |
2464 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
2465 | { | |
2466 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2467 | struct Qdisc *q; | |
2468 | ||
2469 | rcu_read_lock(); | |
2470 | q = rcu_dereference(dev_queue->qdisc); | |
2471 | __netif_schedule(q); | |
2472 | rcu_read_unlock(); | |
2473 | } | |
2474 | } | |
2475 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2476 | ||
e6247027 | 2477 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2478 | { |
e6247027 | 2479 | unsigned long flags; |
56079431 | 2480 | |
9899886d MJ |
2481 | if (unlikely(!skb)) |
2482 | return; | |
2483 | ||
63354797 | 2484 | if (likely(refcount_read(&skb->users) == 1)) { |
e6247027 | 2485 | smp_rmb(); |
63354797 RE |
2486 | refcount_set(&skb->users, 0); |
2487 | } else if (likely(!refcount_dec_and_test(&skb->users))) { | |
e6247027 | 2488 | return; |
bea3348e | 2489 | } |
e6247027 ED |
2490 | get_kfree_skb_cb(skb)->reason = reason; |
2491 | local_irq_save(flags); | |
2492 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2493 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2494 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2495 | local_irq_restore(flags); | |
56079431 | 2496 | } |
e6247027 | 2497 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2498 | |
e6247027 | 2499 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2500 | { |
2501 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2502 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2503 | else |
2504 | dev_kfree_skb(skb); | |
2505 | } | |
e6247027 | 2506 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2507 | |
2508 | ||
bea3348e SH |
2509 | /** |
2510 | * netif_device_detach - mark device as removed | |
2511 | * @dev: network device | |
2512 | * | |
2513 | * Mark device as removed from system and therefore no longer available. | |
2514 | */ | |
56079431 DV |
2515 | void netif_device_detach(struct net_device *dev) |
2516 | { | |
2517 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2518 | netif_running(dev)) { | |
d543103a | 2519 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2520 | } |
2521 | } | |
2522 | EXPORT_SYMBOL(netif_device_detach); | |
2523 | ||
bea3348e SH |
2524 | /** |
2525 | * netif_device_attach - mark device as attached | |
2526 | * @dev: network device | |
2527 | * | |
2528 | * Mark device as attached from system and restart if needed. | |
2529 | */ | |
56079431 DV |
2530 | void netif_device_attach(struct net_device *dev) |
2531 | { | |
2532 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2533 | netif_running(dev)) { | |
d543103a | 2534 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2535 | __netdev_watchdog_up(dev); |
56079431 DV |
2536 | } |
2537 | } | |
2538 | EXPORT_SYMBOL(netif_device_attach); | |
2539 | ||
5605c762 JP |
2540 | /* |
2541 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2542 | * to be used as a distribution range. | |
2543 | */ | |
2544 | u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, | |
2545 | unsigned int num_tx_queues) | |
2546 | { | |
2547 | u32 hash; | |
2548 | u16 qoffset = 0; | |
2549 | u16 qcount = num_tx_queues; | |
2550 | ||
2551 | if (skb_rx_queue_recorded(skb)) { | |
2552 | hash = skb_get_rx_queue(skb); | |
2553 | while (unlikely(hash >= num_tx_queues)) | |
2554 | hash -= num_tx_queues; | |
2555 | return hash; | |
2556 | } | |
2557 | ||
2558 | if (dev->num_tc) { | |
2559 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
f4563a75 | 2560 | |
5605c762 JP |
2561 | qoffset = dev->tc_to_txq[tc].offset; |
2562 | qcount = dev->tc_to_txq[tc].count; | |
2563 | } | |
2564 | ||
2565 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
2566 | } | |
2567 | EXPORT_SYMBOL(__skb_tx_hash); | |
2568 | ||
36c92474 BH |
2569 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2570 | { | |
84d15ae5 | 2571 | static const netdev_features_t null_features; |
36c92474 | 2572 | struct net_device *dev = skb->dev; |
88ad4175 | 2573 | const char *name = ""; |
36c92474 | 2574 | |
c846ad9b BG |
2575 | if (!net_ratelimit()) |
2576 | return; | |
2577 | ||
88ad4175 BM |
2578 | if (dev) { |
2579 | if (dev->dev.parent) | |
2580 | name = dev_driver_string(dev->dev.parent); | |
2581 | else | |
2582 | name = netdev_name(dev); | |
2583 | } | |
36c92474 BH |
2584 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " |
2585 | "gso_type=%d ip_summed=%d\n", | |
88ad4175 | 2586 | name, dev ? &dev->features : &null_features, |
65e9d2fa | 2587 | skb->sk ? &skb->sk->sk_route_caps : &null_features, |
36c92474 BH |
2588 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2589 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2590 | } | |
2591 | ||
1da177e4 LT |
2592 | /* |
2593 | * Invalidate hardware checksum when packet is to be mangled, and | |
2594 | * complete checksum manually on outgoing path. | |
2595 | */ | |
84fa7933 | 2596 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2597 | { |
d3bc23e7 | 2598 | __wsum csum; |
663ead3b | 2599 | int ret = 0, offset; |
1da177e4 | 2600 | |
84fa7933 | 2601 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2602 | goto out_set_summed; |
2603 | ||
2604 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2605 | skb_warn_bad_offload(skb); |
2606 | return -EINVAL; | |
1da177e4 LT |
2607 | } |
2608 | ||
cef401de ED |
2609 | /* Before computing a checksum, we should make sure no frag could |
2610 | * be modified by an external entity : checksum could be wrong. | |
2611 | */ | |
2612 | if (skb_has_shared_frag(skb)) { | |
2613 | ret = __skb_linearize(skb); | |
2614 | if (ret) | |
2615 | goto out; | |
2616 | } | |
2617 | ||
55508d60 | 2618 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2619 | BUG_ON(offset >= skb_headlen(skb)); |
2620 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2621 | ||
2622 | offset += skb->csum_offset; | |
2623 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2624 | ||
2625 | if (skb_cloned(skb) && | |
2626 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2627 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2628 | if (ret) | |
2629 | goto out; | |
2630 | } | |
2631 | ||
4f2e4ad5 | 2632 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 2633 | out_set_summed: |
1da177e4 | 2634 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2635 | out: |
1da177e4 LT |
2636 | return ret; |
2637 | } | |
d1b19dff | 2638 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2639 | |
b72b5bf6 DC |
2640 | int skb_crc32c_csum_help(struct sk_buff *skb) |
2641 | { | |
2642 | __le32 crc32c_csum; | |
2643 | int ret = 0, offset, start; | |
2644 | ||
2645 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
2646 | goto out; | |
2647 | ||
2648 | if (unlikely(skb_is_gso(skb))) | |
2649 | goto out; | |
2650 | ||
2651 | /* Before computing a checksum, we should make sure no frag could | |
2652 | * be modified by an external entity : checksum could be wrong. | |
2653 | */ | |
2654 | if (unlikely(skb_has_shared_frag(skb))) { | |
2655 | ret = __skb_linearize(skb); | |
2656 | if (ret) | |
2657 | goto out; | |
2658 | } | |
2659 | start = skb_checksum_start_offset(skb); | |
2660 | offset = start + offsetof(struct sctphdr, checksum); | |
2661 | if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { | |
2662 | ret = -EINVAL; | |
2663 | goto out; | |
2664 | } | |
2665 | if (skb_cloned(skb) && | |
2666 | !skb_clone_writable(skb, offset + sizeof(__le32))) { | |
2667 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); | |
2668 | if (ret) | |
2669 | goto out; | |
2670 | } | |
2671 | crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, | |
2672 | skb->len - start, ~(__u32)0, | |
2673 | crc32c_csum_stub)); | |
2674 | *(__le32 *)(skb->data + offset) = crc32c_csum; | |
2675 | skb->ip_summed = CHECKSUM_NONE; | |
dba00306 | 2676 | skb->csum_not_inet = 0; |
b72b5bf6 DC |
2677 | out: |
2678 | return ret; | |
2679 | } | |
2680 | ||
53d6471c | 2681 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2682 | { |
252e3346 | 2683 | __be16 type = skb->protocol; |
f6a78bfc | 2684 | |
19acc327 PS |
2685 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2686 | if (type == htons(ETH_P_TEB)) { | |
2687 | struct ethhdr *eth; | |
2688 | ||
2689 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2690 | return 0; | |
2691 | ||
2692 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2693 | type = eth->h_proto; | |
2694 | } | |
2695 | ||
d4bcef3f | 2696 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
2697 | } |
2698 | ||
2699 | /** | |
2700 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2701 | * @skb: buffer to segment | |
2702 | * @features: features for the output path (see dev->features) | |
2703 | */ | |
2704 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2705 | netdev_features_t features) | |
2706 | { | |
2707 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2708 | struct packet_offload *ptype; | |
53d6471c VY |
2709 | int vlan_depth = skb->mac_len; |
2710 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2711 | |
2712 | if (unlikely(!type)) | |
2713 | return ERR_PTR(-EINVAL); | |
2714 | ||
53d6471c | 2715 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2716 | |
2717 | rcu_read_lock(); | |
22061d80 | 2718 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2719 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2720 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2721 | break; |
2722 | } | |
2723 | } | |
2724 | rcu_read_unlock(); | |
2725 | ||
98e399f8 | 2726 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2727 | |
f6a78bfc HX |
2728 | return segs; |
2729 | } | |
05e8ef4a PS |
2730 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2731 | ||
2732 | ||
2733 | /* openvswitch calls this on rx path, so we need a different check. | |
2734 | */ | |
2735 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2736 | { | |
2737 | if (tx_path) | |
93991221 | 2738 | return skb->ip_summed != CHECKSUM_PARTIAL; |
6e7bc478 ED |
2739 | |
2740 | return skb->ip_summed == CHECKSUM_NONE; | |
05e8ef4a PS |
2741 | } |
2742 | ||
2743 | /** | |
2744 | * __skb_gso_segment - Perform segmentation on skb. | |
2745 | * @skb: buffer to segment | |
2746 | * @features: features for the output path (see dev->features) | |
2747 | * @tx_path: whether it is called in TX path | |
2748 | * | |
2749 | * This function segments the given skb and returns a list of segments. | |
2750 | * | |
2751 | * It may return NULL if the skb requires no segmentation. This is | |
2752 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 KK |
2753 | * |
2754 | * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. | |
05e8ef4a PS |
2755 | */ |
2756 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2757 | netdev_features_t features, bool tx_path) | |
2758 | { | |
b2504a5d ED |
2759 | struct sk_buff *segs; |
2760 | ||
05e8ef4a PS |
2761 | if (unlikely(skb_needs_check(skb, tx_path))) { |
2762 | int err; | |
2763 | ||
b2504a5d | 2764 | /* We're going to init ->check field in TCP or UDP header */ |
a40e0a66 | 2765 | err = skb_cow_head(skb, 0); |
2766 | if (err < 0) | |
05e8ef4a PS |
2767 | return ERR_PTR(err); |
2768 | } | |
2769 | ||
802ab55a AD |
2770 | /* Only report GSO partial support if it will enable us to |
2771 | * support segmentation on this frame without needing additional | |
2772 | * work. | |
2773 | */ | |
2774 | if (features & NETIF_F_GSO_PARTIAL) { | |
2775 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
2776 | struct net_device *dev = skb->dev; | |
2777 | ||
2778 | partial_features |= dev->features & dev->gso_partial_features; | |
2779 | if (!skb_gso_ok(skb, features | partial_features)) | |
2780 | features &= ~NETIF_F_GSO_PARTIAL; | |
2781 | } | |
2782 | ||
9207f9d4 KK |
2783 | BUILD_BUG_ON(SKB_SGO_CB_OFFSET + |
2784 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); | |
2785 | ||
68c33163 | 2786 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2787 | SKB_GSO_CB(skb)->encap_level = 0; |
2788 | ||
05e8ef4a PS |
2789 | skb_reset_mac_header(skb); |
2790 | skb_reset_mac_len(skb); | |
2791 | ||
b2504a5d ED |
2792 | segs = skb_mac_gso_segment(skb, features); |
2793 | ||
2794 | if (unlikely(skb_needs_check(skb, tx_path))) | |
2795 | skb_warn_bad_offload(skb); | |
2796 | ||
2797 | return segs; | |
05e8ef4a | 2798 | } |
12b0004d | 2799 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2800 | |
fb286bb2 HX |
2801 | /* Take action when hardware reception checksum errors are detected. */ |
2802 | #ifdef CONFIG_BUG | |
2803 | void netdev_rx_csum_fault(struct net_device *dev) | |
2804 | { | |
2805 | if (net_ratelimit()) { | |
7b6cd1ce | 2806 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2807 | dump_stack(); |
2808 | } | |
2809 | } | |
2810 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2811 | #endif | |
2812 | ||
1da177e4 LT |
2813 | /* Actually, we should eliminate this check as soon as we know, that: |
2814 | * 1. IOMMU is present and allows to map all the memory. | |
2815 | * 2. No high memory really exists on this machine. | |
2816 | */ | |
2817 | ||
c1e756bf | 2818 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2819 | { |
3d3a8533 | 2820 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2821 | int i; |
f4563a75 | 2822 | |
5acbbd42 | 2823 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2824 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2825 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 2826 | |
ea2ab693 | 2827 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 2828 | return 1; |
ea2ab693 | 2829 | } |
5acbbd42 | 2830 | } |
1da177e4 | 2831 | |
5acbbd42 FT |
2832 | if (PCI_DMA_BUS_IS_PHYS) { |
2833 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2834 | |
9092c658 ED |
2835 | if (!pdev) |
2836 | return 0; | |
5acbbd42 | 2837 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2838 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2839 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
f4563a75 | 2840 | |
5acbbd42 FT |
2841 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2842 | return 1; | |
2843 | } | |
2844 | } | |
3d3a8533 | 2845 | #endif |
1da177e4 LT |
2846 | return 0; |
2847 | } | |
1da177e4 | 2848 | |
3b392ddb SH |
2849 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2850 | * instead of standard features for the netdev. | |
2851 | */ | |
d0edc7bf | 2852 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
2853 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
2854 | netdev_features_t features, | |
2855 | __be16 type) | |
2856 | { | |
25cd9ba0 | 2857 | if (eth_p_mpls(type)) |
3b392ddb SH |
2858 | features &= skb->dev->mpls_features; |
2859 | ||
2860 | return features; | |
2861 | } | |
2862 | #else | |
2863 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2864 | netdev_features_t features, | |
2865 | __be16 type) | |
2866 | { | |
2867 | return features; | |
2868 | } | |
2869 | #endif | |
2870 | ||
c8f44aff | 2871 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2872 | netdev_features_t features) |
f01a5236 | 2873 | { |
53d6471c | 2874 | int tmp; |
3b392ddb SH |
2875 | __be16 type; |
2876 | ||
2877 | type = skb_network_protocol(skb, &tmp); | |
2878 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2879 | |
c0d680e5 | 2880 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2881 | !can_checksum_protocol(features, type)) { |
996e8021 | 2882 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 2883 | } |
7be2c82c ED |
2884 | if (illegal_highdma(skb->dev, skb)) |
2885 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
2886 | |
2887 | return features; | |
2888 | } | |
2889 | ||
e38f3025 TM |
2890 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
2891 | struct net_device *dev, | |
2892 | netdev_features_t features) | |
2893 | { | |
2894 | return features; | |
2895 | } | |
2896 | EXPORT_SYMBOL(passthru_features_check); | |
2897 | ||
8cb65d00 TM |
2898 | static netdev_features_t dflt_features_check(const struct sk_buff *skb, |
2899 | struct net_device *dev, | |
2900 | netdev_features_t features) | |
2901 | { | |
2902 | return vlan_features_check(skb, features); | |
2903 | } | |
2904 | ||
cbc53e08 AD |
2905 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
2906 | struct net_device *dev, | |
2907 | netdev_features_t features) | |
2908 | { | |
2909 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
2910 | ||
2911 | if (gso_segs > dev->gso_max_segs) | |
2912 | return features & ~NETIF_F_GSO_MASK; | |
2913 | ||
802ab55a AD |
2914 | /* Support for GSO partial features requires software |
2915 | * intervention before we can actually process the packets | |
2916 | * so we need to strip support for any partial features now | |
2917 | * and we can pull them back in after we have partially | |
2918 | * segmented the frame. | |
2919 | */ | |
2920 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
2921 | features &= ~dev->gso_partial_features; | |
2922 | ||
2923 | /* Make sure to clear the IPv4 ID mangling feature if the | |
2924 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
2925 | */ |
2926 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
2927 | struct iphdr *iph = skb->encapsulation ? | |
2928 | inner_ip_hdr(skb) : ip_hdr(skb); | |
2929 | ||
2930 | if (!(iph->frag_off & htons(IP_DF))) | |
2931 | features &= ~NETIF_F_TSO_MANGLEID; | |
2932 | } | |
2933 | ||
2934 | return features; | |
2935 | } | |
2936 | ||
c1e756bf | 2937 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 2938 | { |
5f35227e | 2939 | struct net_device *dev = skb->dev; |
fcbeb976 | 2940 | netdev_features_t features = dev->features; |
58e998c6 | 2941 | |
cbc53e08 AD |
2942 | if (skb_is_gso(skb)) |
2943 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 2944 | |
5f35227e JG |
2945 | /* If encapsulation offload request, verify we are testing |
2946 | * hardware encapsulation features instead of standard | |
2947 | * features for the netdev | |
2948 | */ | |
2949 | if (skb->encapsulation) | |
2950 | features &= dev->hw_enc_features; | |
2951 | ||
f5a7fb88 TM |
2952 | if (skb_vlan_tagged(skb)) |
2953 | features = netdev_intersect_features(features, | |
2954 | dev->vlan_features | | |
2955 | NETIF_F_HW_VLAN_CTAG_TX | | |
2956 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2957 | |
5f35227e JG |
2958 | if (dev->netdev_ops->ndo_features_check) |
2959 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
2960 | features); | |
8cb65d00 TM |
2961 | else |
2962 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 2963 | |
c1e756bf | 2964 | return harmonize_features(skb, features); |
58e998c6 | 2965 | } |
c1e756bf | 2966 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2967 | |
2ea25513 | 2968 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2969 | struct netdev_queue *txq, bool more) |
f6a78bfc | 2970 | { |
2ea25513 DM |
2971 | unsigned int len; |
2972 | int rc; | |
00829823 | 2973 | |
7866a621 | 2974 | if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) |
2ea25513 | 2975 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 2976 | |
2ea25513 DM |
2977 | len = skb->len; |
2978 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 2979 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 2980 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 2981 | |
2ea25513 DM |
2982 | return rc; |
2983 | } | |
7b9c6090 | 2984 | |
8dcda22a DM |
2985 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
2986 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
2987 | { |
2988 | struct sk_buff *skb = first; | |
2989 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 2990 | |
7f2e870f DM |
2991 | while (skb) { |
2992 | struct sk_buff *next = skb->next; | |
fc70fb64 | 2993 | |
7f2e870f | 2994 | skb->next = NULL; |
95f6b3dd | 2995 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
2996 | if (unlikely(!dev_xmit_complete(rc))) { |
2997 | skb->next = next; | |
2998 | goto out; | |
2999 | } | |
6afff0ca | 3000 | |
7f2e870f DM |
3001 | skb = next; |
3002 | if (netif_xmit_stopped(txq) && skb) { | |
3003 | rc = NETDEV_TX_BUSY; | |
3004 | break; | |
9ccb8975 | 3005 | } |
7f2e870f | 3006 | } |
9ccb8975 | 3007 | |
7f2e870f DM |
3008 | out: |
3009 | *ret = rc; | |
3010 | return skb; | |
3011 | } | |
b40863c6 | 3012 | |
1ff0dc94 ED |
3013 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
3014 | netdev_features_t features) | |
f6a78bfc | 3015 | { |
df8a39de | 3016 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
3017 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
3018 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
3019 | return skb; |
3020 | } | |
f6a78bfc | 3021 | |
43c26a1a DC |
3022 | int skb_csum_hwoffload_help(struct sk_buff *skb, |
3023 | const netdev_features_t features) | |
3024 | { | |
3025 | if (unlikely(skb->csum_not_inet)) | |
3026 | return !!(features & NETIF_F_SCTP_CRC) ? 0 : | |
3027 | skb_crc32c_csum_help(skb); | |
3028 | ||
3029 | return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb); | |
3030 | } | |
3031 | EXPORT_SYMBOL(skb_csum_hwoffload_help); | |
3032 | ||
55a93b3e | 3033 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
3034 | { |
3035 | netdev_features_t features; | |
f6a78bfc | 3036 | |
eae3f88e DM |
3037 | features = netif_skb_features(skb); |
3038 | skb = validate_xmit_vlan(skb, features); | |
3039 | if (unlikely(!skb)) | |
3040 | goto out_null; | |
7b9c6090 | 3041 | |
8b86a61d | 3042 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
3043 | struct sk_buff *segs; |
3044 | ||
3045 | segs = skb_gso_segment(skb, features); | |
cecda693 | 3046 | if (IS_ERR(segs)) { |
af6dabc9 | 3047 | goto out_kfree_skb; |
cecda693 JW |
3048 | } else if (segs) { |
3049 | consume_skb(skb); | |
3050 | skb = segs; | |
f6a78bfc | 3051 | } |
eae3f88e DM |
3052 | } else { |
3053 | if (skb_needs_linearize(skb, features) && | |
3054 | __skb_linearize(skb)) | |
3055 | goto out_kfree_skb; | |
4ec93edb | 3056 | |
f6e27114 SK |
3057 | if (validate_xmit_xfrm(skb, features)) |
3058 | goto out_kfree_skb; | |
3059 | ||
eae3f88e DM |
3060 | /* If packet is not checksummed and device does not |
3061 | * support checksumming for this protocol, complete | |
3062 | * checksumming here. | |
3063 | */ | |
3064 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
3065 | if (skb->encapsulation) | |
3066 | skb_set_inner_transport_header(skb, | |
3067 | skb_checksum_start_offset(skb)); | |
3068 | else | |
3069 | skb_set_transport_header(skb, | |
3070 | skb_checksum_start_offset(skb)); | |
43c26a1a | 3071 | if (skb_csum_hwoffload_help(skb, features)) |
eae3f88e | 3072 | goto out_kfree_skb; |
7b9c6090 | 3073 | } |
0c772159 | 3074 | } |
7b9c6090 | 3075 | |
eae3f88e | 3076 | return skb; |
fc70fb64 | 3077 | |
f6a78bfc HX |
3078 | out_kfree_skb: |
3079 | kfree_skb(skb); | |
eae3f88e | 3080 | out_null: |
d21fd63e | 3081 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3082 | return NULL; |
3083 | } | |
6afff0ca | 3084 | |
55a93b3e ED |
3085 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
3086 | { | |
3087 | struct sk_buff *next, *head = NULL, *tail; | |
3088 | ||
bec3cfdc | 3089 | for (; skb != NULL; skb = next) { |
55a93b3e ED |
3090 | next = skb->next; |
3091 | skb->next = NULL; | |
bec3cfdc ED |
3092 | |
3093 | /* in case skb wont be segmented, point to itself */ | |
3094 | skb->prev = skb; | |
3095 | ||
55a93b3e | 3096 | skb = validate_xmit_skb(skb, dev); |
bec3cfdc ED |
3097 | if (!skb) |
3098 | continue; | |
55a93b3e | 3099 | |
bec3cfdc ED |
3100 | if (!head) |
3101 | head = skb; | |
3102 | else | |
3103 | tail->next = skb; | |
3104 | /* If skb was segmented, skb->prev points to | |
3105 | * the last segment. If not, it still contains skb. | |
3106 | */ | |
3107 | tail = skb->prev; | |
55a93b3e ED |
3108 | } |
3109 | return head; | |
f6a78bfc | 3110 | } |
104ba78c | 3111 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3112 | |
1def9238 ED |
3113 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3114 | { | |
3115 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3116 | ||
3117 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3118 | ||
3119 | /* To get more precise estimation of bytes sent on wire, | |
3120 | * we add to pkt_len the headers size of all segments | |
3121 | */ | |
3122 | if (shinfo->gso_size) { | |
757b8b1d | 3123 | unsigned int hdr_len; |
15e5a030 | 3124 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3125 | |
757b8b1d ED |
3126 | /* mac layer + network layer */ |
3127 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3128 | ||
3129 | /* + transport layer */ | |
1def9238 ED |
3130 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
3131 | hdr_len += tcp_hdrlen(skb); | |
3132 | else | |
3133 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
3134 | |
3135 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3136 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3137 | shinfo->gso_size); | |
3138 | ||
3139 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3140 | } |
3141 | } | |
3142 | ||
bbd8a0d3 KK |
3143 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3144 | struct net_device *dev, | |
3145 | struct netdev_queue *txq) | |
3146 | { | |
3147 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3148 | struct sk_buff *to_free = NULL; |
a2da570d | 3149 | bool contended; |
bbd8a0d3 KK |
3150 | int rc; |
3151 | ||
a2da570d | 3152 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
3153 | /* |
3154 | * Heuristic to force contended enqueues to serialize on a | |
3155 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3156 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3157 | * often and dequeue packets faster. |
79640a4c | 3158 | */ |
a2da570d | 3159 | contended = qdisc_is_running(q); |
79640a4c ED |
3160 | if (unlikely(contended)) |
3161 | spin_lock(&q->busylock); | |
3162 | ||
bbd8a0d3 KK |
3163 | spin_lock(root_lock); |
3164 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3165 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3166 | rc = NET_XMIT_DROP; |
3167 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3168 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3169 | /* |
3170 | * This is a work-conserving queue; there are no old skbs | |
3171 | * waiting to be sent out; and the qdisc is not running - | |
3172 | * xmit the skb directly. | |
3173 | */ | |
bfe0d029 | 3174 | |
bfe0d029 ED |
3175 | qdisc_bstats_update(q, skb); |
3176 | ||
55a93b3e | 3177 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3178 | if (unlikely(contended)) { |
3179 | spin_unlock(&q->busylock); | |
3180 | contended = false; | |
3181 | } | |
bbd8a0d3 | 3182 | __qdisc_run(q); |
79640a4c | 3183 | } else |
bc135b23 | 3184 | qdisc_run_end(q); |
bbd8a0d3 KK |
3185 | |
3186 | rc = NET_XMIT_SUCCESS; | |
3187 | } else { | |
520ac30f | 3188 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3189 | if (qdisc_run_begin(q)) { |
3190 | if (unlikely(contended)) { | |
3191 | spin_unlock(&q->busylock); | |
3192 | contended = false; | |
3193 | } | |
3194 | __qdisc_run(q); | |
3195 | } | |
bbd8a0d3 KK |
3196 | } |
3197 | spin_unlock(root_lock); | |
520ac30f ED |
3198 | if (unlikely(to_free)) |
3199 | kfree_skb_list(to_free); | |
79640a4c ED |
3200 | if (unlikely(contended)) |
3201 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3202 | return rc; |
3203 | } | |
3204 | ||
86f8515f | 3205 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3206 | static void skb_update_prio(struct sk_buff *skb) |
3207 | { | |
6977a79d | 3208 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 3209 | |
91c68ce2 | 3210 | if (!skb->priority && skb->sk && map) { |
2a56a1fe TH |
3211 | unsigned int prioidx = |
3212 | sock_cgroup_prioidx(&skb->sk->sk_cgrp_data); | |
91c68ce2 ED |
3213 | |
3214 | if (prioidx < map->priomap_len) | |
3215 | skb->priority = map->priomap[prioidx]; | |
3216 | } | |
5bc1421e NH |
3217 | } |
3218 | #else | |
3219 | #define skb_update_prio(skb) | |
3220 | #endif | |
3221 | ||
f60e5990 | 3222 | DEFINE_PER_CPU(int, xmit_recursion); |
3223 | EXPORT_SYMBOL(xmit_recursion); | |
3224 | ||
95603e22 MM |
3225 | /** |
3226 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3227 | * @net: network namespace this loopback is happening in |
3228 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3229 | * @skb: buffer to transmit |
3230 | */ | |
0c4b51f0 | 3231 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3232 | { |
3233 | skb_reset_mac_header(skb); | |
3234 | __skb_pull(skb, skb_network_offset(skb)); | |
3235 | skb->pkt_type = PACKET_LOOPBACK; | |
3236 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3237 | WARN_ON(!skb_dst(skb)); | |
3238 | skb_dst_force(skb); | |
3239 | netif_rx_ni(skb); | |
3240 | return 0; | |
3241 | } | |
3242 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3243 | ||
1f211a1b DB |
3244 | #ifdef CONFIG_NET_EGRESS |
3245 | static struct sk_buff * | |
3246 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3247 | { | |
3248 | struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list); | |
3249 | struct tcf_result cl_res; | |
3250 | ||
3251 | if (!cl) | |
3252 | return skb; | |
3253 | ||
8dc07fdb | 3254 | /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ |
1f211a1b DB |
3255 | qdisc_bstats_cpu_update(cl->q, skb); |
3256 | ||
87d83093 | 3257 | switch (tcf_classify(skb, cl, &cl_res, false)) { |
1f211a1b DB |
3258 | case TC_ACT_OK: |
3259 | case TC_ACT_RECLASSIFY: | |
3260 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3261 | break; | |
3262 | case TC_ACT_SHOT: | |
3263 | qdisc_qstats_cpu_drop(cl->q); | |
3264 | *ret = NET_XMIT_DROP; | |
7e2c3aea DB |
3265 | kfree_skb(skb); |
3266 | return NULL; | |
1f211a1b DB |
3267 | case TC_ACT_STOLEN: |
3268 | case TC_ACT_QUEUED: | |
e25ea21f | 3269 | case TC_ACT_TRAP: |
1f211a1b | 3270 | *ret = NET_XMIT_SUCCESS; |
7e2c3aea | 3271 | consume_skb(skb); |
1f211a1b DB |
3272 | return NULL; |
3273 | case TC_ACT_REDIRECT: | |
3274 | /* No need to push/pop skb's mac_header here on egress! */ | |
3275 | skb_do_redirect(skb); | |
3276 | *ret = NET_XMIT_SUCCESS; | |
3277 | return NULL; | |
3278 | default: | |
3279 | break; | |
3280 | } | |
3281 | ||
3282 | return skb; | |
3283 | } | |
3284 | #endif /* CONFIG_NET_EGRESS */ | |
3285 | ||
638b2a69 JP |
3286 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
3287 | { | |
3288 | #ifdef CONFIG_XPS | |
3289 | struct xps_dev_maps *dev_maps; | |
3290 | struct xps_map *map; | |
3291 | int queue_index = -1; | |
3292 | ||
3293 | rcu_read_lock(); | |
3294 | dev_maps = rcu_dereference(dev->xps_maps); | |
3295 | if (dev_maps) { | |
184c449f AD |
3296 | unsigned int tci = skb->sender_cpu - 1; |
3297 | ||
3298 | if (dev->num_tc) { | |
3299 | tci *= dev->num_tc; | |
3300 | tci += netdev_get_prio_tc_map(dev, skb->priority); | |
3301 | } | |
3302 | ||
3303 | map = rcu_dereference(dev_maps->cpu_map[tci]); | |
638b2a69 JP |
3304 | if (map) { |
3305 | if (map->len == 1) | |
3306 | queue_index = map->queues[0]; | |
3307 | else | |
3308 | queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), | |
3309 | map->len)]; | |
3310 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3311 | queue_index = -1; | |
3312 | } | |
3313 | } | |
3314 | rcu_read_unlock(); | |
3315 | ||
3316 | return queue_index; | |
3317 | #else | |
3318 | return -1; | |
3319 | #endif | |
3320 | } | |
3321 | ||
3322 | static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) | |
3323 | { | |
3324 | struct sock *sk = skb->sk; | |
3325 | int queue_index = sk_tx_queue_get(sk); | |
3326 | ||
3327 | if (queue_index < 0 || skb->ooo_okay || | |
3328 | queue_index >= dev->real_num_tx_queues) { | |
3329 | int new_index = get_xps_queue(dev, skb); | |
f4563a75 | 3330 | |
638b2a69 JP |
3331 | if (new_index < 0) |
3332 | new_index = skb_tx_hash(dev, skb); | |
3333 | ||
3334 | if (queue_index != new_index && sk && | |
004a5d01 | 3335 | sk_fullsock(sk) && |
638b2a69 JP |
3336 | rcu_access_pointer(sk->sk_dst_cache)) |
3337 | sk_tx_queue_set(sk, new_index); | |
3338 | ||
3339 | queue_index = new_index; | |
3340 | } | |
3341 | ||
3342 | return queue_index; | |
3343 | } | |
3344 | ||
3345 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, | |
3346 | struct sk_buff *skb, | |
3347 | void *accel_priv) | |
3348 | { | |
3349 | int queue_index = 0; | |
3350 | ||
3351 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
3352 | u32 sender_cpu = skb->sender_cpu - 1; |
3353 | ||
3354 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
3355 | skb->sender_cpu = raw_smp_processor_id() + 1; |
3356 | #endif | |
3357 | ||
3358 | if (dev->real_num_tx_queues != 1) { | |
3359 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 3360 | |
638b2a69 JP |
3361 | if (ops->ndo_select_queue) |
3362 | queue_index = ops->ndo_select_queue(dev, skb, accel_priv, | |
3363 | __netdev_pick_tx); | |
3364 | else | |
3365 | queue_index = __netdev_pick_tx(dev, skb); | |
3366 | ||
3367 | if (!accel_priv) | |
3368 | queue_index = netdev_cap_txqueue(dev, queue_index); | |
3369 | } | |
3370 | ||
3371 | skb_set_queue_mapping(skb, queue_index); | |
3372 | return netdev_get_tx_queue(dev, queue_index); | |
3373 | } | |
3374 | ||
d29f749e | 3375 | /** |
9d08dd3d | 3376 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3377 | * @skb: buffer to transmit |
9d08dd3d | 3378 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
3379 | * |
3380 | * Queue a buffer for transmission to a network device. The caller must | |
3381 | * have set the device and priority and built the buffer before calling | |
3382 | * this function. The function can be called from an interrupt. | |
3383 | * | |
3384 | * A negative errno code is returned on a failure. A success does not | |
3385 | * guarantee the frame will be transmitted as it may be dropped due | |
3386 | * to congestion or traffic shaping. | |
3387 | * | |
3388 | * ----------------------------------------------------------------------------------- | |
3389 | * I notice this method can also return errors from the queue disciplines, | |
3390 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3391 | * be positive. | |
3392 | * | |
3393 | * Regardless of the return value, the skb is consumed, so it is currently | |
3394 | * difficult to retry a send to this method. (You can bump the ref count | |
3395 | * before sending to hold a reference for retry if you are careful.) | |
3396 | * | |
3397 | * When calling this method, interrupts MUST be enabled. This is because | |
3398 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3399 | * --BLG | |
3400 | */ | |
0a59f3a9 | 3401 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
3402 | { |
3403 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3404 | struct netdev_queue *txq; |
1da177e4 LT |
3405 | struct Qdisc *q; |
3406 | int rc = -ENOMEM; | |
3407 | ||
6d1ccff6 ED |
3408 | skb_reset_mac_header(skb); |
3409 | ||
e7fd2885 WB |
3410 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3411 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3412 | ||
4ec93edb YH |
3413 | /* Disable soft irqs for various locks below. Also |
3414 | * stops preemption for RCU. | |
1da177e4 | 3415 | */ |
4ec93edb | 3416 | rcu_read_lock_bh(); |
1da177e4 | 3417 | |
5bc1421e NH |
3418 | skb_update_prio(skb); |
3419 | ||
1f211a1b DB |
3420 | qdisc_pkt_len_init(skb); |
3421 | #ifdef CONFIG_NET_CLS_ACT | |
8dc07fdb | 3422 | skb->tc_at_ingress = 0; |
1f211a1b DB |
3423 | # ifdef CONFIG_NET_EGRESS |
3424 | if (static_key_false(&egress_needed)) { | |
3425 | skb = sch_handle_egress(skb, &rc, dev); | |
3426 | if (!skb) | |
3427 | goto out; | |
3428 | } | |
3429 | # endif | |
3430 | #endif | |
02875878 ED |
3431 | /* If device/qdisc don't need skb->dst, release it right now while |
3432 | * its hot in this cpu cache. | |
3433 | */ | |
3434 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
3435 | skb_dst_drop(skb); | |
3436 | else | |
3437 | skb_dst_force(skb); | |
3438 | ||
f663dd9a | 3439 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 3440 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 3441 | |
cf66ba58 | 3442 | trace_net_dev_queue(skb); |
1da177e4 | 3443 | if (q->enqueue) { |
bbd8a0d3 | 3444 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 3445 | goto out; |
1da177e4 LT |
3446 | } |
3447 | ||
3448 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 3449 | * loopback, all the sorts of tunnels... |
1da177e4 | 3450 | |
eb13da1a | 3451 | * Really, it is unlikely that netif_tx_lock protection is necessary |
3452 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
3453 | * counters.) | |
3454 | * However, it is possible, that they rely on protection | |
3455 | * made by us here. | |
1da177e4 | 3456 | |
eb13da1a | 3457 | * Check this and shot the lock. It is not prone from deadlocks. |
3458 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
3459 | */ |
3460 | if (dev->flags & IFF_UP) { | |
3461 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
3462 | ||
c773e847 | 3463 | if (txq->xmit_lock_owner != cpu) { |
a70b506e DB |
3464 | if (unlikely(__this_cpu_read(xmit_recursion) > |
3465 | XMIT_RECURSION_LIMIT)) | |
745e20f1 ED |
3466 | goto recursion_alert; |
3467 | ||
1f59533f JDB |
3468 | skb = validate_xmit_skb(skb, dev); |
3469 | if (!skb) | |
d21fd63e | 3470 | goto out; |
1f59533f | 3471 | |
c773e847 | 3472 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 3473 | |
73466498 | 3474 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 3475 | __this_cpu_inc(xmit_recursion); |
ce93718f | 3476 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 3477 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 3478 | if (dev_xmit_complete(rc)) { |
c773e847 | 3479 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
3480 | goto out; |
3481 | } | |
3482 | } | |
c773e847 | 3483 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
3484 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
3485 | dev->name); | |
1da177e4 LT |
3486 | } else { |
3487 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
3488 | * unfortunately |
3489 | */ | |
3490 | recursion_alert: | |
e87cc472 JP |
3491 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
3492 | dev->name); | |
1da177e4 LT |
3493 | } |
3494 | } | |
3495 | ||
3496 | rc = -ENETDOWN; | |
d4828d85 | 3497 | rcu_read_unlock_bh(); |
1da177e4 | 3498 | |
015f0688 | 3499 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 3500 | kfree_skb_list(skb); |
1da177e4 LT |
3501 | return rc; |
3502 | out: | |
d4828d85 | 3503 | rcu_read_unlock_bh(); |
1da177e4 LT |
3504 | return rc; |
3505 | } | |
f663dd9a | 3506 | |
2b4aa3ce | 3507 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
3508 | { |
3509 | return __dev_queue_xmit(skb, NULL); | |
3510 | } | |
2b4aa3ce | 3511 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3512 | |
f663dd9a JW |
3513 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3514 | { | |
3515 | return __dev_queue_xmit(skb, accel_priv); | |
3516 | } | |
3517 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3518 | ||
1da177e4 | 3519 | |
eb13da1a | 3520 | /************************************************************************* |
3521 | * Receiver routines | |
3522 | *************************************************************************/ | |
1da177e4 | 3523 | |
6b2bedc3 | 3524 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3525 | EXPORT_SYMBOL(netdev_max_backlog); |
3526 | ||
3b098e2d | 3527 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 | 3528 | int netdev_budget __read_mostly = 300; |
7acf8a1e | 3529 | unsigned int __read_mostly netdev_budget_usecs = 2000; |
3d48b53f MT |
3530 | int weight_p __read_mostly = 64; /* old backlog weight */ |
3531 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
3532 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
3533 | int dev_rx_weight __read_mostly = 64; | |
3534 | int dev_tx_weight __read_mostly = 64; | |
1da177e4 | 3535 | |
eecfd7c4 ED |
3536 | /* Called with irq disabled */ |
3537 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3538 | struct napi_struct *napi) | |
3539 | { | |
3540 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3541 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3542 | } | |
3543 | ||
bfb564e7 KK |
3544 | #ifdef CONFIG_RPS |
3545 | ||
3546 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3547 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 3548 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
3549 | u32 rps_cpu_mask __read_mostly; |
3550 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 3551 | |
c5905afb | 3552 | struct static_key rps_needed __read_mostly; |
3df97ba8 | 3553 | EXPORT_SYMBOL(rps_needed); |
13bfff25 ED |
3554 | struct static_key rfs_needed __read_mostly; |
3555 | EXPORT_SYMBOL(rfs_needed); | |
adc9300e | 3556 | |
c445477d BH |
3557 | static struct rps_dev_flow * |
3558 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3559 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3560 | { | |
a31196b0 | 3561 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
3562 | #ifdef CONFIG_RFS_ACCEL |
3563 | struct netdev_rx_queue *rxqueue; | |
3564 | struct rps_dev_flow_table *flow_table; | |
3565 | struct rps_dev_flow *old_rflow; | |
3566 | u32 flow_id; | |
3567 | u16 rxq_index; | |
3568 | int rc; | |
3569 | ||
3570 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3571 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3572 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3573 | goto out; |
3574 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3575 | if (rxq_index == skb_get_rx_queue(skb)) | |
3576 | goto out; | |
3577 | ||
3578 | rxqueue = dev->_rx + rxq_index; | |
3579 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3580 | if (!flow_table) | |
3581 | goto out; | |
61b905da | 3582 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3583 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3584 | rxq_index, flow_id); | |
3585 | if (rc < 0) | |
3586 | goto out; | |
3587 | old_rflow = rflow; | |
3588 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3589 | rflow->filter = rc; |
3590 | if (old_rflow->filter == rflow->filter) | |
3591 | old_rflow->filter = RPS_NO_FILTER; | |
3592 | out: | |
3593 | #endif | |
3594 | rflow->last_qtail = | |
09994d1b | 3595 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3596 | } |
3597 | ||
09994d1b | 3598 | rflow->cpu = next_cpu; |
c445477d BH |
3599 | return rflow; |
3600 | } | |
3601 | ||
bfb564e7 KK |
3602 | /* |
3603 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3604 | * CPU from the RPS map of the receiving queue for a given skb. | |
3605 | * rcu_read_lock must be held on entry. | |
3606 | */ | |
3607 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3608 | struct rps_dev_flow **rflowp) | |
3609 | { | |
567e4b79 ED |
3610 | const struct rps_sock_flow_table *sock_flow_table; |
3611 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 3612 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 3613 | struct rps_map *map; |
bfb564e7 | 3614 | int cpu = -1; |
567e4b79 | 3615 | u32 tcpu; |
61b905da | 3616 | u32 hash; |
bfb564e7 KK |
3617 | |
3618 | if (skb_rx_queue_recorded(skb)) { | |
3619 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 3620 | |
62fe0b40 BH |
3621 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3622 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3623 | "%s received packet on queue %u, but number " | |
3624 | "of RX queues is %u\n", | |
3625 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3626 | goto done; |
3627 | } | |
567e4b79 ED |
3628 | rxqueue += index; |
3629 | } | |
bfb564e7 | 3630 | |
567e4b79 ED |
3631 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
3632 | ||
3633 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 3634 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 3635 | if (!flow_table && !map) |
bfb564e7 KK |
3636 | goto done; |
3637 | ||
2d47b459 | 3638 | skb_reset_network_header(skb); |
61b905da TH |
3639 | hash = skb_get_hash(skb); |
3640 | if (!hash) | |
bfb564e7 KK |
3641 | goto done; |
3642 | ||
fec5e652 TH |
3643 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
3644 | if (flow_table && sock_flow_table) { | |
fec5e652 | 3645 | struct rps_dev_flow *rflow; |
567e4b79 ED |
3646 | u32 next_cpu; |
3647 | u32 ident; | |
3648 | ||
3649 | /* First check into global flow table if there is a match */ | |
3650 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
3651 | if ((ident ^ hash) & ~rps_cpu_mask) | |
3652 | goto try_rps; | |
fec5e652 | 3653 | |
567e4b79 ED |
3654 | next_cpu = ident & rps_cpu_mask; |
3655 | ||
3656 | /* OK, now we know there is a match, | |
3657 | * we can look at the local (per receive queue) flow table | |
3658 | */ | |
61b905da | 3659 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3660 | tcpu = rflow->cpu; |
3661 | ||
fec5e652 TH |
3662 | /* |
3663 | * If the desired CPU (where last recvmsg was done) is | |
3664 | * different from current CPU (one in the rx-queue flow | |
3665 | * table entry), switch if one of the following holds: | |
a31196b0 | 3666 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
3667 | * - Current CPU is offline. |
3668 | * - The current CPU's queue tail has advanced beyond the | |
3669 | * last packet that was enqueued using this table entry. | |
3670 | * This guarantees that all previous packets for the flow | |
3671 | * have been dequeued, thus preserving in order delivery. | |
3672 | */ | |
3673 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 3674 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 3675 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
3676 | rflow->last_qtail)) >= 0)) { |
3677 | tcpu = next_cpu; | |
c445477d | 3678 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3679 | } |
c445477d | 3680 | |
a31196b0 | 3681 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
3682 | *rflowp = rflow; |
3683 | cpu = tcpu; | |
3684 | goto done; | |
3685 | } | |
3686 | } | |
3687 | ||
567e4b79 ED |
3688 | try_rps: |
3689 | ||
0a9627f2 | 3690 | if (map) { |
8fc54f68 | 3691 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3692 | if (cpu_online(tcpu)) { |
3693 | cpu = tcpu; | |
3694 | goto done; | |
3695 | } | |
3696 | } | |
3697 | ||
3698 | done: | |
0a9627f2 TH |
3699 | return cpu; |
3700 | } | |
3701 | ||
c445477d BH |
3702 | #ifdef CONFIG_RFS_ACCEL |
3703 | ||
3704 | /** | |
3705 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3706 | * @dev: Device on which the filter was set | |
3707 | * @rxq_index: RX queue index | |
3708 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3709 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3710 | * | |
3711 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3712 | * this function for each installed filter and remove the filters for | |
3713 | * which it returns %true. | |
3714 | */ | |
3715 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3716 | u32 flow_id, u16 filter_id) | |
3717 | { | |
3718 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3719 | struct rps_dev_flow_table *flow_table; | |
3720 | struct rps_dev_flow *rflow; | |
3721 | bool expire = true; | |
a31196b0 | 3722 | unsigned int cpu; |
c445477d BH |
3723 | |
3724 | rcu_read_lock(); | |
3725 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3726 | if (flow_table && flow_id <= flow_table->mask) { | |
3727 | rflow = &flow_table->flows[flow_id]; | |
6aa7de05 | 3728 | cpu = READ_ONCE(rflow->cpu); |
a31196b0 | 3729 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
3730 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
3731 | rflow->last_qtail) < | |
3732 | (int)(10 * flow_table->mask))) | |
3733 | expire = false; | |
3734 | } | |
3735 | rcu_read_unlock(); | |
3736 | return expire; | |
3737 | } | |
3738 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3739 | ||
3740 | #endif /* CONFIG_RFS_ACCEL */ | |
3741 | ||
0a9627f2 | 3742 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3743 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3744 | { |
e36fa2f7 ED |
3745 | struct softnet_data *sd = data; |
3746 | ||
eecfd7c4 | 3747 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3748 | sd->received_rps++; |
0a9627f2 | 3749 | } |
e36fa2f7 | 3750 | |
fec5e652 | 3751 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3752 | |
e36fa2f7 ED |
3753 | /* |
3754 | * Check if this softnet_data structure is another cpu one | |
3755 | * If yes, queue it to our IPI list and return 1 | |
3756 | * If no, return 0 | |
3757 | */ | |
3758 | static int rps_ipi_queued(struct softnet_data *sd) | |
3759 | { | |
3760 | #ifdef CONFIG_RPS | |
903ceff7 | 3761 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
3762 | |
3763 | if (sd != mysd) { | |
3764 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3765 | mysd->rps_ipi_list = sd; | |
3766 | ||
3767 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3768 | return 1; | |
3769 | } | |
3770 | #endif /* CONFIG_RPS */ | |
3771 | return 0; | |
3772 | } | |
3773 | ||
99bbc707 WB |
3774 | #ifdef CONFIG_NET_FLOW_LIMIT |
3775 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3776 | #endif | |
3777 | ||
3778 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3779 | { | |
3780 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3781 | struct sd_flow_limit *fl; | |
3782 | struct softnet_data *sd; | |
3783 | unsigned int old_flow, new_flow; | |
3784 | ||
3785 | if (qlen < (netdev_max_backlog >> 1)) | |
3786 | return false; | |
3787 | ||
903ceff7 | 3788 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
3789 | |
3790 | rcu_read_lock(); | |
3791 | fl = rcu_dereference(sd->flow_limit); | |
3792 | if (fl) { | |
3958afa1 | 3793 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3794 | old_flow = fl->history[fl->history_head]; |
3795 | fl->history[fl->history_head] = new_flow; | |
3796 | ||
3797 | fl->history_head++; | |
3798 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3799 | ||
3800 | if (likely(fl->buckets[old_flow])) | |
3801 | fl->buckets[old_flow]--; | |
3802 | ||
3803 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3804 | fl->count++; | |
3805 | rcu_read_unlock(); | |
3806 | return true; | |
3807 | } | |
3808 | } | |
3809 | rcu_read_unlock(); | |
3810 | #endif | |
3811 | return false; | |
3812 | } | |
3813 | ||
0a9627f2 TH |
3814 | /* |
3815 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3816 | * queue (may be a remote CPU queue). | |
3817 | */ | |
fec5e652 TH |
3818 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3819 | unsigned int *qtail) | |
0a9627f2 | 3820 | { |
e36fa2f7 | 3821 | struct softnet_data *sd; |
0a9627f2 | 3822 | unsigned long flags; |
99bbc707 | 3823 | unsigned int qlen; |
0a9627f2 | 3824 | |
e36fa2f7 | 3825 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3826 | |
3827 | local_irq_save(flags); | |
0a9627f2 | 3828 | |
e36fa2f7 | 3829 | rps_lock(sd); |
e9e4dd32 JA |
3830 | if (!netif_running(skb->dev)) |
3831 | goto drop; | |
99bbc707 WB |
3832 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3833 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 3834 | if (qlen) { |
0a9627f2 | 3835 | enqueue: |
e36fa2f7 | 3836 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3837 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3838 | rps_unlock(sd); |
152102c7 | 3839 | local_irq_restore(flags); |
0a9627f2 TH |
3840 | return NET_RX_SUCCESS; |
3841 | } | |
3842 | ||
ebda37c2 ED |
3843 | /* Schedule NAPI for backlog device |
3844 | * We can use non atomic operation since we own the queue lock | |
3845 | */ | |
3846 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3847 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3848 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3849 | } |
3850 | goto enqueue; | |
3851 | } | |
3852 | ||
e9e4dd32 | 3853 | drop: |
dee42870 | 3854 | sd->dropped++; |
e36fa2f7 | 3855 | rps_unlock(sd); |
0a9627f2 | 3856 | |
0a9627f2 TH |
3857 | local_irq_restore(flags); |
3858 | ||
caf586e5 | 3859 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3860 | kfree_skb(skb); |
3861 | return NET_RX_DROP; | |
3862 | } | |
1da177e4 | 3863 | |
d4455169 JF |
3864 | static u32 netif_receive_generic_xdp(struct sk_buff *skb, |
3865 | struct bpf_prog *xdp_prog) | |
3866 | { | |
3867 | struct xdp_buff xdp; | |
3868 | u32 act = XDP_DROP; | |
3869 | void *orig_data; | |
3870 | int hlen, off; | |
3871 | u32 mac_len; | |
3872 | ||
3873 | /* Reinjected packets coming from act_mirred or similar should | |
3874 | * not get XDP generic processing. | |
3875 | */ | |
3876 | if (skb_cloned(skb)) | |
3877 | return XDP_PASS; | |
3878 | ||
3879 | if (skb_linearize(skb)) | |
3880 | goto do_drop; | |
3881 | ||
3882 | /* The XDP program wants to see the packet starting at the MAC | |
3883 | * header. | |
3884 | */ | |
3885 | mac_len = skb->data - skb_mac_header(skb); | |
3886 | hlen = skb_headlen(skb) + mac_len; | |
3887 | xdp.data = skb->data - mac_len; | |
3888 | xdp.data_end = xdp.data + hlen; | |
3889 | xdp.data_hard_start = skb->data - skb_headroom(skb); | |
3890 | orig_data = xdp.data; | |
3891 | ||
3892 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
3893 | ||
3894 | off = xdp.data - orig_data; | |
3895 | if (off > 0) | |
3896 | __skb_pull(skb, off); | |
3897 | else if (off < 0) | |
3898 | __skb_push(skb, -off); | |
92dd5452 | 3899 | skb->mac_header += off; |
d4455169 JF |
3900 | |
3901 | switch (act) { | |
6103aa96 | 3902 | case XDP_REDIRECT: |
d4455169 JF |
3903 | case XDP_TX: |
3904 | __skb_push(skb, mac_len); | |
3905 | /* fall through */ | |
3906 | case XDP_PASS: | |
3907 | break; | |
3908 | ||
3909 | default: | |
3910 | bpf_warn_invalid_xdp_action(act); | |
3911 | /* fall through */ | |
3912 | case XDP_ABORTED: | |
3913 | trace_xdp_exception(skb->dev, xdp_prog, act); | |
3914 | /* fall through */ | |
3915 | case XDP_DROP: | |
3916 | do_drop: | |
3917 | kfree_skb(skb); | |
3918 | break; | |
3919 | } | |
3920 | ||
3921 | return act; | |
3922 | } | |
3923 | ||
3924 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
3925 | * network taps in order to match in-driver-XDP behavior. | |
3926 | */ | |
7c497478 | 3927 | void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) |
d4455169 JF |
3928 | { |
3929 | struct net_device *dev = skb->dev; | |
3930 | struct netdev_queue *txq; | |
3931 | bool free_skb = true; | |
3932 | int cpu, rc; | |
3933 | ||
3934 | txq = netdev_pick_tx(dev, skb, NULL); | |
3935 | cpu = smp_processor_id(); | |
3936 | HARD_TX_LOCK(dev, txq, cpu); | |
3937 | if (!netif_xmit_stopped(txq)) { | |
3938 | rc = netdev_start_xmit(skb, dev, txq, 0); | |
3939 | if (dev_xmit_complete(rc)) | |
3940 | free_skb = false; | |
3941 | } | |
3942 | HARD_TX_UNLOCK(dev, txq); | |
3943 | if (free_skb) { | |
3944 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
3945 | kfree_skb(skb); | |
3946 | } | |
3947 | } | |
7c497478 | 3948 | EXPORT_SYMBOL_GPL(generic_xdp_tx); |
d4455169 JF |
3949 | |
3950 | static struct static_key generic_xdp_needed __read_mostly; | |
3951 | ||
7c497478 | 3952 | int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) |
d4455169 | 3953 | { |
d4455169 JF |
3954 | if (xdp_prog) { |
3955 | u32 act = netif_receive_generic_xdp(skb, xdp_prog); | |
6103aa96 | 3956 | int err; |
d4455169 JF |
3957 | |
3958 | if (act != XDP_PASS) { | |
6103aa96 JF |
3959 | switch (act) { |
3960 | case XDP_REDIRECT: | |
2facaad6 JDB |
3961 | err = xdp_do_generic_redirect(skb->dev, skb, |
3962 | xdp_prog); | |
6103aa96 JF |
3963 | if (err) |
3964 | goto out_redir; | |
3965 | /* fallthru to submit skb */ | |
3966 | case XDP_TX: | |
d4455169 | 3967 | generic_xdp_tx(skb, xdp_prog); |
6103aa96 JF |
3968 | break; |
3969 | } | |
d4455169 JF |
3970 | return XDP_DROP; |
3971 | } | |
3972 | } | |
3973 | return XDP_PASS; | |
6103aa96 | 3974 | out_redir: |
6103aa96 JF |
3975 | kfree_skb(skb); |
3976 | return XDP_DROP; | |
d4455169 | 3977 | } |
7c497478 | 3978 | EXPORT_SYMBOL_GPL(do_xdp_generic); |
d4455169 | 3979 | |
ae78dbfa | 3980 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3981 | { |
b0e28f1e | 3982 | int ret; |
1da177e4 | 3983 | |
588f0330 | 3984 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3985 | |
cf66ba58 | 3986 | trace_netif_rx(skb); |
d4455169 JF |
3987 | |
3988 | if (static_key_false(&generic_xdp_needed)) { | |
bbbe211c JF |
3989 | int ret; |
3990 | ||
3991 | preempt_disable(); | |
3992 | rcu_read_lock(); | |
3993 | ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
3994 | rcu_read_unlock(); | |
3995 | preempt_enable(); | |
d4455169 | 3996 | |
6103aa96 JF |
3997 | /* Consider XDP consuming the packet a success from |
3998 | * the netdev point of view we do not want to count | |
3999 | * this as an error. | |
4000 | */ | |
d4455169 | 4001 | if (ret != XDP_PASS) |
6103aa96 | 4002 | return NET_RX_SUCCESS; |
d4455169 JF |
4003 | } |
4004 | ||
df334545 | 4005 | #ifdef CONFIG_RPS |
c5905afb | 4006 | if (static_key_false(&rps_needed)) { |
fec5e652 | 4007 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
4008 | int cpu; |
4009 | ||
cece1945 | 4010 | preempt_disable(); |
b0e28f1e | 4011 | rcu_read_lock(); |
fec5e652 TH |
4012 | |
4013 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
4014 | if (cpu < 0) |
4015 | cpu = smp_processor_id(); | |
fec5e652 TH |
4016 | |
4017 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4018 | ||
b0e28f1e | 4019 | rcu_read_unlock(); |
cece1945 | 4020 | preempt_enable(); |
adc9300e ED |
4021 | } else |
4022 | #endif | |
fec5e652 TH |
4023 | { |
4024 | unsigned int qtail; | |
f4563a75 | 4025 | |
fec5e652 TH |
4026 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); |
4027 | put_cpu(); | |
4028 | } | |
b0e28f1e | 4029 | return ret; |
1da177e4 | 4030 | } |
ae78dbfa BH |
4031 | |
4032 | /** | |
4033 | * netif_rx - post buffer to the network code | |
4034 | * @skb: buffer to post | |
4035 | * | |
4036 | * This function receives a packet from a device driver and queues it for | |
4037 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
4038 | * may be dropped during processing for congestion control or by the | |
4039 | * protocol layers. | |
4040 | * | |
4041 | * return values: | |
4042 | * NET_RX_SUCCESS (no congestion) | |
4043 | * NET_RX_DROP (packet was dropped) | |
4044 | * | |
4045 | */ | |
4046 | ||
4047 | int netif_rx(struct sk_buff *skb) | |
4048 | { | |
4049 | trace_netif_rx_entry(skb); | |
4050 | ||
4051 | return netif_rx_internal(skb); | |
4052 | } | |
d1b19dff | 4053 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
4054 | |
4055 | int netif_rx_ni(struct sk_buff *skb) | |
4056 | { | |
4057 | int err; | |
4058 | ||
ae78dbfa BH |
4059 | trace_netif_rx_ni_entry(skb); |
4060 | ||
1da177e4 | 4061 | preempt_disable(); |
ae78dbfa | 4062 | err = netif_rx_internal(skb); |
1da177e4 LT |
4063 | if (local_softirq_pending()) |
4064 | do_softirq(); | |
4065 | preempt_enable(); | |
4066 | ||
4067 | return err; | |
4068 | } | |
1da177e4 LT |
4069 | EXPORT_SYMBOL(netif_rx_ni); |
4070 | ||
0766f788 | 4071 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 4072 | { |
903ceff7 | 4073 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
4074 | |
4075 | if (sd->completion_queue) { | |
4076 | struct sk_buff *clist; | |
4077 | ||
4078 | local_irq_disable(); | |
4079 | clist = sd->completion_queue; | |
4080 | sd->completion_queue = NULL; | |
4081 | local_irq_enable(); | |
4082 | ||
4083 | while (clist) { | |
4084 | struct sk_buff *skb = clist; | |
f4563a75 | 4085 | |
1da177e4 LT |
4086 | clist = clist->next; |
4087 | ||
63354797 | 4088 | WARN_ON(refcount_read(&skb->users)); |
e6247027 ED |
4089 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
4090 | trace_consume_skb(skb); | |
4091 | else | |
4092 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
4093 | |
4094 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
4095 | __kfree_skb(skb); | |
4096 | else | |
4097 | __kfree_skb_defer(skb); | |
1da177e4 | 4098 | } |
15fad714 JDB |
4099 | |
4100 | __kfree_skb_flush(); | |
1da177e4 LT |
4101 | } |
4102 | ||
4103 | if (sd->output_queue) { | |
37437bb2 | 4104 | struct Qdisc *head; |
1da177e4 LT |
4105 | |
4106 | local_irq_disable(); | |
4107 | head = sd->output_queue; | |
4108 | sd->output_queue = NULL; | |
a9cbd588 | 4109 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
4110 | local_irq_enable(); |
4111 | ||
4112 | while (head) { | |
37437bb2 DM |
4113 | struct Qdisc *q = head; |
4114 | spinlock_t *root_lock; | |
4115 | ||
1da177e4 LT |
4116 | head = head->next_sched; |
4117 | ||
5fb66229 | 4118 | root_lock = qdisc_lock(q); |
3bcb846c ED |
4119 | spin_lock(root_lock); |
4120 | /* We need to make sure head->next_sched is read | |
4121 | * before clearing __QDISC_STATE_SCHED | |
4122 | */ | |
4123 | smp_mb__before_atomic(); | |
4124 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
4125 | qdisc_run(q); | |
4126 | spin_unlock(root_lock); | |
1da177e4 LT |
4127 | } |
4128 | } | |
4129 | } | |
4130 | ||
181402a5 | 4131 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
4132 | /* This hook is defined here for ATM LANE */ |
4133 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
4134 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 4135 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 4136 | #endif |
1da177e4 | 4137 | |
1f211a1b DB |
4138 | static inline struct sk_buff * |
4139 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
4140 | struct net_device *orig_dev) | |
f697c3e8 | 4141 | { |
e7582bab | 4142 | #ifdef CONFIG_NET_CLS_ACT |
d2788d34 DB |
4143 | struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list); |
4144 | struct tcf_result cl_res; | |
24824a09 | 4145 | |
c9e99fd0 DB |
4146 | /* If there's at least one ingress present somewhere (so |
4147 | * we get here via enabled static key), remaining devices | |
4148 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 4149 | * out here. |
c9e99fd0 | 4150 | */ |
d2788d34 | 4151 | if (!cl) |
4577139b | 4152 | return skb; |
f697c3e8 HX |
4153 | if (*pt_prev) { |
4154 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4155 | *pt_prev = NULL; | |
1da177e4 LT |
4156 | } |
4157 | ||
3365495c | 4158 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
8dc07fdb | 4159 | skb->tc_at_ingress = 1; |
24ea591d | 4160 | qdisc_bstats_cpu_update(cl->q, skb); |
c9e99fd0 | 4161 | |
87d83093 | 4162 | switch (tcf_classify(skb, cl, &cl_res, false)) { |
d2788d34 DB |
4163 | case TC_ACT_OK: |
4164 | case TC_ACT_RECLASSIFY: | |
4165 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
4166 | break; | |
4167 | case TC_ACT_SHOT: | |
24ea591d | 4168 | qdisc_qstats_cpu_drop(cl->q); |
8a3a4c6e ED |
4169 | kfree_skb(skb); |
4170 | return NULL; | |
d2788d34 DB |
4171 | case TC_ACT_STOLEN: |
4172 | case TC_ACT_QUEUED: | |
e25ea21f | 4173 | case TC_ACT_TRAP: |
8a3a4c6e | 4174 | consume_skb(skb); |
d2788d34 | 4175 | return NULL; |
27b29f63 AS |
4176 | case TC_ACT_REDIRECT: |
4177 | /* skb_mac_header check was done by cls/act_bpf, so | |
4178 | * we can safely push the L2 header back before | |
4179 | * redirecting to another netdev | |
4180 | */ | |
4181 | __skb_push(skb, skb->mac_len); | |
4182 | skb_do_redirect(skb); | |
4183 | return NULL; | |
d2788d34 DB |
4184 | default: |
4185 | break; | |
f697c3e8 | 4186 | } |
e7582bab | 4187 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
4188 | return skb; |
4189 | } | |
1da177e4 | 4190 | |
24b27fc4 MB |
4191 | /** |
4192 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
4193 | * @dev: device to check | |
4194 | * | |
4195 | * Check if a receive handler is already registered for a given device. | |
4196 | * Return true if there one. | |
4197 | * | |
4198 | * The caller must hold the rtnl_mutex. | |
4199 | */ | |
4200 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
4201 | { | |
4202 | ASSERT_RTNL(); | |
4203 | return dev && rtnl_dereference(dev->rx_handler); | |
4204 | } | |
4205 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
4206 | ||
ab95bfe0 JP |
4207 | /** |
4208 | * netdev_rx_handler_register - register receive handler | |
4209 | * @dev: device to register a handler for | |
4210 | * @rx_handler: receive handler to register | |
93e2c32b | 4211 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 4212 | * |
e227867f | 4213 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
4214 | * called from __netif_receive_skb. A negative errno code is returned |
4215 | * on a failure. | |
4216 | * | |
4217 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
4218 | * |
4219 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
4220 | */ |
4221 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
4222 | rx_handler_func_t *rx_handler, |
4223 | void *rx_handler_data) | |
ab95bfe0 | 4224 | { |
1b7cd004 | 4225 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
4226 | return -EBUSY; |
4227 | ||
00cfec37 | 4228 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 4229 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
4230 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
4231 | ||
4232 | return 0; | |
4233 | } | |
4234 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
4235 | ||
4236 | /** | |
4237 | * netdev_rx_handler_unregister - unregister receive handler | |
4238 | * @dev: device to unregister a handler from | |
4239 | * | |
166ec369 | 4240 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
4241 | * |
4242 | * The caller must hold the rtnl_mutex. | |
4243 | */ | |
4244 | void netdev_rx_handler_unregister(struct net_device *dev) | |
4245 | { | |
4246 | ||
4247 | ASSERT_RTNL(); | |
a9b3cd7f | 4248 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
4249 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
4250 | * section has a guarantee to see a non NULL rx_handler_data | |
4251 | * as well. | |
4252 | */ | |
4253 | synchronize_net(); | |
a9b3cd7f | 4254 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
4255 | } |
4256 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
4257 | ||
b4b9e355 MG |
4258 | /* |
4259 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
4260 | * the special handling of PFMEMALLOC skbs. | |
4261 | */ | |
4262 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
4263 | { | |
4264 | switch (skb->protocol) { | |
2b8837ae JP |
4265 | case htons(ETH_P_ARP): |
4266 | case htons(ETH_P_IP): | |
4267 | case htons(ETH_P_IPV6): | |
4268 | case htons(ETH_P_8021Q): | |
4269 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
4270 | return true; |
4271 | default: | |
4272 | return false; | |
4273 | } | |
4274 | } | |
4275 | ||
e687ad60 PN |
4276 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
4277 | int *ret, struct net_device *orig_dev) | |
4278 | { | |
e7582bab | 4279 | #ifdef CONFIG_NETFILTER_INGRESS |
e687ad60 | 4280 | if (nf_hook_ingress_active(skb)) { |
2c1e2703 AC |
4281 | int ingress_retval; |
4282 | ||
e687ad60 PN |
4283 | if (*pt_prev) { |
4284 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4285 | *pt_prev = NULL; | |
4286 | } | |
4287 | ||
2c1e2703 AC |
4288 | rcu_read_lock(); |
4289 | ingress_retval = nf_hook_ingress(skb); | |
4290 | rcu_read_unlock(); | |
4291 | return ingress_retval; | |
e687ad60 | 4292 | } |
e7582bab | 4293 | #endif /* CONFIG_NETFILTER_INGRESS */ |
e687ad60 PN |
4294 | return 0; |
4295 | } | |
e687ad60 | 4296 | |
9754e293 | 4297 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
4298 | { |
4299 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 4300 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 4301 | struct net_device *orig_dev; |
8a4eb573 | 4302 | bool deliver_exact = false; |
1da177e4 | 4303 | int ret = NET_RX_DROP; |
252e3346 | 4304 | __be16 type; |
1da177e4 | 4305 | |
588f0330 | 4306 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 4307 | |
cf66ba58 | 4308 | trace_netif_receive_skb(skb); |
9b22ea56 | 4309 | |
cc9bd5ce | 4310 | orig_dev = skb->dev; |
8f903c70 | 4311 | |
c1d2bbe1 | 4312 | skb_reset_network_header(skb); |
fda55eca ED |
4313 | if (!skb_transport_header_was_set(skb)) |
4314 | skb_reset_transport_header(skb); | |
0b5c9db1 | 4315 | skb_reset_mac_len(skb); |
1da177e4 LT |
4316 | |
4317 | pt_prev = NULL; | |
4318 | ||
63d8ea7f | 4319 | another_round: |
b6858177 | 4320 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
4321 | |
4322 | __this_cpu_inc(softnet_data.processed); | |
4323 | ||
8ad227ff PM |
4324 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
4325 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 4326 | skb = skb_vlan_untag(skb); |
bcc6d479 | 4327 | if (unlikely(!skb)) |
2c17d27c | 4328 | goto out; |
bcc6d479 JP |
4329 | } |
4330 | ||
e7246e12 WB |
4331 | if (skb_skip_tc_classify(skb)) |
4332 | goto skip_classify; | |
1da177e4 | 4333 | |
9754e293 | 4334 | if (pfmemalloc) |
b4b9e355 MG |
4335 | goto skip_taps; |
4336 | ||
1da177e4 | 4337 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
4338 | if (pt_prev) |
4339 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4340 | pt_prev = ptype; | |
4341 | } | |
4342 | ||
4343 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
4344 | if (pt_prev) | |
4345 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4346 | pt_prev = ptype; | |
1da177e4 LT |
4347 | } |
4348 | ||
b4b9e355 | 4349 | skip_taps: |
1cf51900 | 4350 | #ifdef CONFIG_NET_INGRESS |
4577139b | 4351 | if (static_key_false(&ingress_needed)) { |
1f211a1b | 4352 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); |
4577139b | 4353 | if (!skb) |
2c17d27c | 4354 | goto out; |
e687ad60 PN |
4355 | |
4356 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 4357 | goto out; |
4577139b | 4358 | } |
1cf51900 | 4359 | #endif |
a5135bcf | 4360 | skb_reset_tc(skb); |
e7246e12 | 4361 | skip_classify: |
9754e293 | 4362 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
4363 | goto drop; |
4364 | ||
df8a39de | 4365 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
4366 | if (pt_prev) { |
4367 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4368 | pt_prev = NULL; | |
4369 | } | |
48cc32d3 | 4370 | if (vlan_do_receive(&skb)) |
2425717b JF |
4371 | goto another_round; |
4372 | else if (unlikely(!skb)) | |
2c17d27c | 4373 | goto out; |
2425717b JF |
4374 | } |
4375 | ||
48cc32d3 | 4376 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
4377 | if (rx_handler) { |
4378 | if (pt_prev) { | |
4379 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4380 | pt_prev = NULL; | |
4381 | } | |
8a4eb573 JP |
4382 | switch (rx_handler(&skb)) { |
4383 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 4384 | ret = NET_RX_SUCCESS; |
2c17d27c | 4385 | goto out; |
8a4eb573 | 4386 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 4387 | goto another_round; |
8a4eb573 JP |
4388 | case RX_HANDLER_EXACT: |
4389 | deliver_exact = true; | |
4390 | case RX_HANDLER_PASS: | |
4391 | break; | |
4392 | default: | |
4393 | BUG(); | |
4394 | } | |
ab95bfe0 | 4395 | } |
1da177e4 | 4396 | |
df8a39de JP |
4397 | if (unlikely(skb_vlan_tag_present(skb))) { |
4398 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
4399 | skb->pkt_type = PACKET_OTHERHOST; |
4400 | /* Note: we might in the future use prio bits | |
4401 | * and set skb->priority like in vlan_do_receive() | |
4402 | * For the time being, just ignore Priority Code Point | |
4403 | */ | |
4404 | skb->vlan_tci = 0; | |
4405 | } | |
48cc32d3 | 4406 | |
7866a621 SN |
4407 | type = skb->protocol; |
4408 | ||
63d8ea7f | 4409 | /* deliver only exact match when indicated */ |
7866a621 SN |
4410 | if (likely(!deliver_exact)) { |
4411 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4412 | &ptype_base[ntohs(type) & | |
4413 | PTYPE_HASH_MASK]); | |
4414 | } | |
1f3c8804 | 4415 | |
7866a621 SN |
4416 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
4417 | &orig_dev->ptype_specific); | |
4418 | ||
4419 | if (unlikely(skb->dev != orig_dev)) { | |
4420 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4421 | &skb->dev->ptype_specific); | |
1da177e4 LT |
4422 | } |
4423 | ||
4424 | if (pt_prev) { | |
1f8b977a | 4425 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
0e698bf6 | 4426 | goto drop; |
1080e512 MT |
4427 | else |
4428 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 4429 | } else { |
b4b9e355 | 4430 | drop: |
6e7333d3 JW |
4431 | if (!deliver_exact) |
4432 | atomic_long_inc(&skb->dev->rx_dropped); | |
4433 | else | |
4434 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
4435 | kfree_skb(skb); |
4436 | /* Jamal, now you will not able to escape explaining | |
4437 | * me how you were going to use this. :-) | |
4438 | */ | |
4439 | ret = NET_RX_DROP; | |
4440 | } | |
4441 | ||
2c17d27c | 4442 | out: |
9754e293 DM |
4443 | return ret; |
4444 | } | |
4445 | ||
4446 | static int __netif_receive_skb(struct sk_buff *skb) | |
4447 | { | |
4448 | int ret; | |
4449 | ||
4450 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 4451 | unsigned int noreclaim_flag; |
9754e293 DM |
4452 | |
4453 | /* | |
4454 | * PFMEMALLOC skbs are special, they should | |
4455 | * - be delivered to SOCK_MEMALLOC sockets only | |
4456 | * - stay away from userspace | |
4457 | * - have bounded memory usage | |
4458 | * | |
4459 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
4460 | * context down to all allocation sites. | |
4461 | */ | |
f1083048 | 4462 | noreclaim_flag = memalloc_noreclaim_save(); |
9754e293 | 4463 | ret = __netif_receive_skb_core(skb, true); |
f1083048 | 4464 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 DM |
4465 | } else |
4466 | ret = __netif_receive_skb_core(skb, false); | |
4467 | ||
1da177e4 LT |
4468 | return ret; |
4469 | } | |
0a9627f2 | 4470 | |
b5cdae32 DM |
4471 | static int generic_xdp_install(struct net_device *dev, struct netdev_xdp *xdp) |
4472 | { | |
58038695 | 4473 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); |
b5cdae32 DM |
4474 | struct bpf_prog *new = xdp->prog; |
4475 | int ret = 0; | |
4476 | ||
4477 | switch (xdp->command) { | |
58038695 | 4478 | case XDP_SETUP_PROG: |
b5cdae32 DM |
4479 | rcu_assign_pointer(dev->xdp_prog, new); |
4480 | if (old) | |
4481 | bpf_prog_put(old); | |
4482 | ||
4483 | if (old && !new) { | |
4484 | static_key_slow_dec(&generic_xdp_needed); | |
4485 | } else if (new && !old) { | |
4486 | static_key_slow_inc(&generic_xdp_needed); | |
4487 | dev_disable_lro(dev); | |
4488 | } | |
4489 | break; | |
b5cdae32 DM |
4490 | |
4491 | case XDP_QUERY_PROG: | |
58038695 MKL |
4492 | xdp->prog_attached = !!old; |
4493 | xdp->prog_id = old ? old->aux->id : 0; | |
b5cdae32 DM |
4494 | break; |
4495 | ||
4496 | default: | |
4497 | ret = -EINVAL; | |
4498 | break; | |
4499 | } | |
4500 | ||
4501 | return ret; | |
4502 | } | |
4503 | ||
ae78dbfa | 4504 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 4505 | { |
2c17d27c JA |
4506 | int ret; |
4507 | ||
588f0330 | 4508 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 4509 | |
c1f19b51 RC |
4510 | if (skb_defer_rx_timestamp(skb)) |
4511 | return NET_RX_SUCCESS; | |
4512 | ||
b5cdae32 | 4513 | if (static_key_false(&generic_xdp_needed)) { |
bbbe211c | 4514 | int ret; |
b5cdae32 | 4515 | |
bbbe211c JF |
4516 | preempt_disable(); |
4517 | rcu_read_lock(); | |
4518 | ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
4519 | rcu_read_unlock(); | |
4520 | preempt_enable(); | |
4521 | ||
4522 | if (ret != XDP_PASS) | |
d4455169 | 4523 | return NET_RX_DROP; |
b5cdae32 DM |
4524 | } |
4525 | ||
bbbe211c | 4526 | rcu_read_lock(); |
df334545 | 4527 | #ifdef CONFIG_RPS |
c5905afb | 4528 | if (static_key_false(&rps_needed)) { |
3b098e2d | 4529 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 4530 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 4531 | |
3b098e2d ED |
4532 | if (cpu >= 0) { |
4533 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4534 | rcu_read_unlock(); | |
adc9300e | 4535 | return ret; |
3b098e2d | 4536 | } |
fec5e652 | 4537 | } |
1e94d72f | 4538 | #endif |
2c17d27c JA |
4539 | ret = __netif_receive_skb(skb); |
4540 | rcu_read_unlock(); | |
4541 | return ret; | |
0a9627f2 | 4542 | } |
ae78dbfa BH |
4543 | |
4544 | /** | |
4545 | * netif_receive_skb - process receive buffer from network | |
4546 | * @skb: buffer to process | |
4547 | * | |
4548 | * netif_receive_skb() is the main receive data processing function. | |
4549 | * It always succeeds. The buffer may be dropped during processing | |
4550 | * for congestion control or by the protocol layers. | |
4551 | * | |
4552 | * This function may only be called from softirq context and interrupts | |
4553 | * should be enabled. | |
4554 | * | |
4555 | * Return values (usually ignored): | |
4556 | * NET_RX_SUCCESS: no congestion | |
4557 | * NET_RX_DROP: packet was dropped | |
4558 | */ | |
04eb4489 | 4559 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa BH |
4560 | { |
4561 | trace_netif_receive_skb_entry(skb); | |
4562 | ||
4563 | return netif_receive_skb_internal(skb); | |
4564 | } | |
04eb4489 | 4565 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 4566 | |
41852497 | 4567 | DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
4568 | |
4569 | /* Network device is going away, flush any packets still pending */ | |
4570 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 4571 | { |
6e583ce5 | 4572 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
4573 | struct softnet_data *sd; |
4574 | ||
4575 | local_bh_disable(); | |
4576 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 4577 | |
145dd5f9 | 4578 | local_irq_disable(); |
e36fa2f7 | 4579 | rps_lock(sd); |
6e7676c1 | 4580 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 4581 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 4582 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 4583 | kfree_skb(skb); |
76cc8b13 | 4584 | input_queue_head_incr(sd); |
6e583ce5 | 4585 | } |
6e7676c1 | 4586 | } |
e36fa2f7 | 4587 | rps_unlock(sd); |
145dd5f9 | 4588 | local_irq_enable(); |
6e7676c1 CG |
4589 | |
4590 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 4591 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
4592 | __skb_unlink(skb, &sd->process_queue); |
4593 | kfree_skb(skb); | |
76cc8b13 | 4594 | input_queue_head_incr(sd); |
6e7676c1 CG |
4595 | } |
4596 | } | |
145dd5f9 PA |
4597 | local_bh_enable(); |
4598 | } | |
4599 | ||
41852497 | 4600 | static void flush_all_backlogs(void) |
145dd5f9 PA |
4601 | { |
4602 | unsigned int cpu; | |
4603 | ||
4604 | get_online_cpus(); | |
4605 | ||
41852497 ED |
4606 | for_each_online_cpu(cpu) |
4607 | queue_work_on(cpu, system_highpri_wq, | |
4608 | per_cpu_ptr(&flush_works, cpu)); | |
145dd5f9 PA |
4609 | |
4610 | for_each_online_cpu(cpu) | |
41852497 | 4611 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 PA |
4612 | |
4613 | put_online_cpus(); | |
6e583ce5 SH |
4614 | } |
4615 | ||
d565b0a1 HX |
4616 | static int napi_gro_complete(struct sk_buff *skb) |
4617 | { | |
22061d80 | 4618 | struct packet_offload *ptype; |
d565b0a1 | 4619 | __be16 type = skb->protocol; |
22061d80 | 4620 | struct list_head *head = &offload_base; |
d565b0a1 HX |
4621 | int err = -ENOENT; |
4622 | ||
c3c7c254 ED |
4623 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
4624 | ||
fc59f9a3 HX |
4625 | if (NAPI_GRO_CB(skb)->count == 1) { |
4626 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 4627 | goto out; |
fc59f9a3 | 4628 | } |
d565b0a1 HX |
4629 | |
4630 | rcu_read_lock(); | |
4631 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4632 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
4633 | continue; |
4634 | ||
299603e8 | 4635 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
4636 | break; |
4637 | } | |
4638 | rcu_read_unlock(); | |
4639 | ||
4640 | if (err) { | |
4641 | WARN_ON(&ptype->list == head); | |
4642 | kfree_skb(skb); | |
4643 | return NET_RX_SUCCESS; | |
4644 | } | |
4645 | ||
4646 | out: | |
ae78dbfa | 4647 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
4648 | } |
4649 | ||
2e71a6f8 ED |
4650 | /* napi->gro_list contains packets ordered by age. |
4651 | * youngest packets at the head of it. | |
4652 | * Complete skbs in reverse order to reduce latencies. | |
4653 | */ | |
4654 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 4655 | { |
2e71a6f8 | 4656 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 4657 | |
2e71a6f8 ED |
4658 | /* scan list and build reverse chain */ |
4659 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
4660 | skb->prev = prev; | |
4661 | prev = skb; | |
4662 | } | |
4663 | ||
4664 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 4665 | skb->next = NULL; |
2e71a6f8 ED |
4666 | |
4667 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
4668 | return; | |
4669 | ||
4670 | prev = skb->prev; | |
d565b0a1 | 4671 | napi_gro_complete(skb); |
2e71a6f8 | 4672 | napi->gro_count--; |
d565b0a1 HX |
4673 | } |
4674 | ||
4675 | napi->gro_list = NULL; | |
4676 | } | |
86cac58b | 4677 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 4678 | |
89c5fa33 ED |
4679 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
4680 | { | |
4681 | struct sk_buff *p; | |
4682 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 4683 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
4684 | |
4685 | for (p = napi->gro_list; p; p = p->next) { | |
4686 | unsigned long diffs; | |
4687 | ||
0b4cec8c TH |
4688 | NAPI_GRO_CB(p)->flush = 0; |
4689 | ||
4690 | if (hash != skb_get_hash_raw(p)) { | |
4691 | NAPI_GRO_CB(p)->same_flow = 0; | |
4692 | continue; | |
4693 | } | |
4694 | ||
89c5fa33 ED |
4695 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
4696 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
ce87fc6c | 4697 | diffs |= skb_metadata_dst_cmp(p, skb); |
89c5fa33 ED |
4698 | if (maclen == ETH_HLEN) |
4699 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 4700 | skb_mac_header(skb)); |
89c5fa33 ED |
4701 | else if (!diffs) |
4702 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 4703 | skb_mac_header(skb), |
89c5fa33 ED |
4704 | maclen); |
4705 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
4706 | } |
4707 | } | |
4708 | ||
299603e8 JC |
4709 | static void skb_gro_reset_offset(struct sk_buff *skb) |
4710 | { | |
4711 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4712 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
4713 | ||
4714 | NAPI_GRO_CB(skb)->data_offset = 0; | |
4715 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
4716 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
4717 | ||
4718 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
4719 | pinfo->nr_frags && | |
4720 | !PageHighMem(skb_frag_page(frag0))) { | |
4721 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
7cfd5fd5 ED |
4722 | NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, |
4723 | skb_frag_size(frag0), | |
4724 | skb->end - skb->tail); | |
89c5fa33 ED |
4725 | } |
4726 | } | |
4727 | ||
a50e233c ED |
4728 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
4729 | { | |
4730 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4731 | ||
4732 | BUG_ON(skb->end - skb->tail < grow); | |
4733 | ||
4734 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
4735 | ||
4736 | skb->data_len -= grow; | |
4737 | skb->tail += grow; | |
4738 | ||
4739 | pinfo->frags[0].page_offset += grow; | |
4740 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
4741 | ||
4742 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
4743 | skb_frag_unref(skb, 0); | |
4744 | memmove(pinfo->frags, pinfo->frags + 1, | |
4745 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
4746 | } | |
4747 | } | |
4748 | ||
bb728820 | 4749 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
4750 | { |
4751 | struct sk_buff **pp = NULL; | |
22061d80 | 4752 | struct packet_offload *ptype; |
d565b0a1 | 4753 | __be16 type = skb->protocol; |
22061d80 | 4754 | struct list_head *head = &offload_base; |
0da2afd5 | 4755 | int same_flow; |
5b252f0c | 4756 | enum gro_result ret; |
a50e233c | 4757 | int grow; |
d565b0a1 | 4758 | |
b5cdae32 | 4759 | if (netif_elide_gro(skb->dev)) |
d565b0a1 HX |
4760 | goto normal; |
4761 | ||
89c5fa33 ED |
4762 | gro_list_prepare(napi, skb); |
4763 | ||
d565b0a1 HX |
4764 | rcu_read_lock(); |
4765 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4766 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4767 | continue; |
4768 | ||
86911732 | 4769 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4770 | skb_reset_mac_len(skb); |
d565b0a1 | 4771 | NAPI_GRO_CB(skb)->same_flow = 0; |
d61d072e | 4772 | NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); |
5d38a079 | 4773 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 4774 | NAPI_GRO_CB(skb)->encap_mark = 0; |
fcd91dd4 | 4775 | NAPI_GRO_CB(skb)->recursion_counter = 0; |
a0ca153f | 4776 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 4777 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 4778 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 4779 | |
662880f4 TH |
4780 | /* Setup for GRO checksum validation */ |
4781 | switch (skb->ip_summed) { | |
4782 | case CHECKSUM_COMPLETE: | |
4783 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4784 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4785 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4786 | break; | |
4787 | case CHECKSUM_UNNECESSARY: | |
4788 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4789 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4790 | break; | |
4791 | default: | |
4792 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4793 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4794 | } | |
d565b0a1 | 4795 | |
f191a1d1 | 4796 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4797 | break; |
4798 | } | |
4799 | rcu_read_unlock(); | |
4800 | ||
4801 | if (&ptype->list == head) | |
4802 | goto normal; | |
4803 | ||
25393d3f SK |
4804 | if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) { |
4805 | ret = GRO_CONSUMED; | |
4806 | goto ok; | |
4807 | } | |
4808 | ||
0da2afd5 | 4809 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4810 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4811 | |
d565b0a1 HX |
4812 | if (pp) { |
4813 | struct sk_buff *nskb = *pp; | |
4814 | ||
4815 | *pp = nskb->next; | |
4816 | nskb->next = NULL; | |
4817 | napi_gro_complete(nskb); | |
4ae5544f | 4818 | napi->gro_count--; |
d565b0a1 HX |
4819 | } |
4820 | ||
0da2afd5 | 4821 | if (same_flow) |
d565b0a1 HX |
4822 | goto ok; |
4823 | ||
600adc18 | 4824 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4825 | goto normal; |
d565b0a1 | 4826 | |
600adc18 ED |
4827 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4828 | struct sk_buff *nskb = napi->gro_list; | |
4829 | ||
4830 | /* locate the end of the list to select the 'oldest' flow */ | |
4831 | while (nskb->next) { | |
4832 | pp = &nskb->next; | |
4833 | nskb = *pp; | |
4834 | } | |
4835 | *pp = NULL; | |
4836 | nskb->next = NULL; | |
4837 | napi_gro_complete(nskb); | |
4838 | } else { | |
4839 | napi->gro_count++; | |
4840 | } | |
d565b0a1 | 4841 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4842 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4843 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4844 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4845 | skb->next = napi->gro_list; |
4846 | napi->gro_list = skb; | |
5d0d9be8 | 4847 | ret = GRO_HELD; |
d565b0a1 | 4848 | |
ad0f9904 | 4849 | pull: |
a50e233c ED |
4850 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4851 | if (grow > 0) | |
4852 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4853 | ok: |
5d0d9be8 | 4854 | return ret; |
d565b0a1 HX |
4855 | |
4856 | normal: | |
ad0f9904 HX |
4857 | ret = GRO_NORMAL; |
4858 | goto pull; | |
5d38a079 | 4859 | } |
96e93eab | 4860 | |
bf5a755f JC |
4861 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4862 | { | |
4863 | struct list_head *offload_head = &offload_base; | |
4864 | struct packet_offload *ptype; | |
4865 | ||
4866 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4867 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4868 | continue; | |
4869 | return ptype; | |
4870 | } | |
4871 | return NULL; | |
4872 | } | |
e27a2f83 | 4873 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4874 | |
4875 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4876 | { | |
4877 | struct list_head *offload_head = &offload_base; | |
4878 | struct packet_offload *ptype; | |
4879 | ||
4880 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4881 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4882 | continue; | |
4883 | return ptype; | |
4884 | } | |
4885 | return NULL; | |
4886 | } | |
e27a2f83 | 4887 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4888 | |
e44699d2 MK |
4889 | static void napi_skb_free_stolen_head(struct sk_buff *skb) |
4890 | { | |
4891 | skb_dst_drop(skb); | |
4892 | secpath_reset(skb); | |
4893 | kmem_cache_free(skbuff_head_cache, skb); | |
4894 | } | |
4895 | ||
bb728820 | 4896 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4897 | { |
5d0d9be8 HX |
4898 | switch (ret) { |
4899 | case GRO_NORMAL: | |
ae78dbfa | 4900 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4901 | ret = GRO_DROP; |
4902 | break; | |
5d38a079 | 4903 | |
5d0d9be8 | 4904 | case GRO_DROP: |
5d38a079 HX |
4905 | kfree_skb(skb); |
4906 | break; | |
5b252f0c | 4907 | |
daa86548 | 4908 | case GRO_MERGED_FREE: |
e44699d2 MK |
4909 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
4910 | napi_skb_free_stolen_head(skb); | |
4911 | else | |
d7e8883c | 4912 | __kfree_skb(skb); |
daa86548 ED |
4913 | break; |
4914 | ||
5b252f0c BH |
4915 | case GRO_HELD: |
4916 | case GRO_MERGED: | |
25393d3f | 4917 | case GRO_CONSUMED: |
5b252f0c | 4918 | break; |
5d38a079 HX |
4919 | } |
4920 | ||
c7c4b3b6 | 4921 | return ret; |
5d0d9be8 | 4922 | } |
5d0d9be8 | 4923 | |
c7c4b3b6 | 4924 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4925 | { |
93f93a44 | 4926 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 4927 | trace_napi_gro_receive_entry(skb); |
86911732 | 4928 | |
a50e233c ED |
4929 | skb_gro_reset_offset(skb); |
4930 | ||
89c5fa33 | 4931 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4932 | } |
4933 | EXPORT_SYMBOL(napi_gro_receive); | |
4934 | ||
d0c2b0d2 | 4935 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4936 | { |
93a35f59 ED |
4937 | if (unlikely(skb->pfmemalloc)) { |
4938 | consume_skb(skb); | |
4939 | return; | |
4940 | } | |
96e93eab | 4941 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4942 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4943 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4944 | skb->vlan_tci = 0; |
66c46d74 | 4945 | skb->dev = napi->dev; |
6d152e23 | 4946 | skb->skb_iif = 0; |
c3caf119 JC |
4947 | skb->encapsulation = 0; |
4948 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4949 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
f991bb9d | 4950 | secpath_reset(skb); |
96e93eab HX |
4951 | |
4952 | napi->skb = skb; | |
4953 | } | |
96e93eab | 4954 | |
76620aaf | 4955 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4956 | { |
5d38a079 | 4957 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4958 | |
4959 | if (!skb) { | |
fd11a83d | 4960 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
4961 | if (skb) { |
4962 | napi->skb = skb; | |
4963 | skb_mark_napi_id(skb, napi); | |
4964 | } | |
80595d59 | 4965 | } |
96e93eab HX |
4966 | return skb; |
4967 | } | |
76620aaf | 4968 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4969 | |
a50e233c ED |
4970 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4971 | struct sk_buff *skb, | |
4972 | gro_result_t ret) | |
96e93eab | 4973 | { |
5d0d9be8 HX |
4974 | switch (ret) { |
4975 | case GRO_NORMAL: | |
a50e233c ED |
4976 | case GRO_HELD: |
4977 | __skb_push(skb, ETH_HLEN); | |
4978 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4979 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4980 | ret = GRO_DROP; |
86911732 | 4981 | break; |
5d38a079 | 4982 | |
5d0d9be8 | 4983 | case GRO_DROP: |
5d0d9be8 HX |
4984 | napi_reuse_skb(napi, skb); |
4985 | break; | |
5b252f0c | 4986 | |
e44699d2 MK |
4987 | case GRO_MERGED_FREE: |
4988 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) | |
4989 | napi_skb_free_stolen_head(skb); | |
4990 | else | |
4991 | napi_reuse_skb(napi, skb); | |
4992 | break; | |
4993 | ||
5b252f0c | 4994 | case GRO_MERGED: |
25393d3f | 4995 | case GRO_CONSUMED: |
5b252f0c | 4996 | break; |
5d0d9be8 | 4997 | } |
5d38a079 | 4998 | |
c7c4b3b6 | 4999 | return ret; |
5d38a079 | 5000 | } |
5d0d9be8 | 5001 | |
a50e233c ED |
5002 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
5003 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
5004 | * We copy ethernet header into skb->data to have a common layout. | |
5005 | */ | |
4adb9c4a | 5006 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
5007 | { |
5008 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
5009 | const struct ethhdr *eth; |
5010 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
5011 | |
5012 | napi->skb = NULL; | |
5013 | ||
a50e233c ED |
5014 | skb_reset_mac_header(skb); |
5015 | skb_gro_reset_offset(skb); | |
5016 | ||
5017 | eth = skb_gro_header_fast(skb, 0); | |
5018 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
5019 | eth = skb_gro_header_slow(skb, hlen, 0); | |
5020 | if (unlikely(!eth)) { | |
4da46ceb AC |
5021 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
5022 | __func__, napi->dev->name); | |
a50e233c ED |
5023 | napi_reuse_skb(napi, skb); |
5024 | return NULL; | |
5025 | } | |
5026 | } else { | |
5027 | gro_pull_from_frag0(skb, hlen); | |
5028 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
5029 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 5030 | } |
a50e233c ED |
5031 | __skb_pull(skb, hlen); |
5032 | ||
5033 | /* | |
5034 | * This works because the only protocols we care about don't require | |
5035 | * special handling. | |
5036 | * We'll fix it up properly in napi_frags_finish() | |
5037 | */ | |
5038 | skb->protocol = eth->h_proto; | |
76620aaf | 5039 | |
76620aaf HX |
5040 | return skb; |
5041 | } | |
76620aaf | 5042 | |
c7c4b3b6 | 5043 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 5044 | { |
76620aaf | 5045 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
5046 | |
5047 | if (!skb) | |
c7c4b3b6 | 5048 | return GRO_DROP; |
5d0d9be8 | 5049 | |
ae78dbfa BH |
5050 | trace_napi_gro_frags_entry(skb); |
5051 | ||
89c5fa33 | 5052 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 5053 | } |
5d38a079 HX |
5054 | EXPORT_SYMBOL(napi_gro_frags); |
5055 | ||
573e8fca TH |
5056 | /* Compute the checksum from gro_offset and return the folded value |
5057 | * after adding in any pseudo checksum. | |
5058 | */ | |
5059 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
5060 | { | |
5061 | __wsum wsum; | |
5062 | __sum16 sum; | |
5063 | ||
5064 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
5065 | ||
5066 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
5067 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
5068 | if (likely(!sum)) { | |
5069 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
5070 | !skb->csum_complete_sw) | |
5071 | netdev_rx_csum_fault(skb->dev); | |
5072 | } | |
5073 | ||
5074 | NAPI_GRO_CB(skb)->csum = wsum; | |
5075 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
5076 | ||
5077 | return sum; | |
5078 | } | |
5079 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
5080 | ||
773fc8f6 | 5081 | static void net_rps_send_ipi(struct softnet_data *remsd) |
5082 | { | |
5083 | #ifdef CONFIG_RPS | |
5084 | while (remsd) { | |
5085 | struct softnet_data *next = remsd->rps_ipi_next; | |
5086 | ||
5087 | if (cpu_online(remsd->cpu)) | |
5088 | smp_call_function_single_async(remsd->cpu, &remsd->csd); | |
5089 | remsd = next; | |
5090 | } | |
5091 | #endif | |
5092 | } | |
5093 | ||
e326bed2 | 5094 | /* |
855abcf0 | 5095 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
5096 | * Note: called with local irq disabled, but exits with local irq enabled. |
5097 | */ | |
5098 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
5099 | { | |
5100 | #ifdef CONFIG_RPS | |
5101 | struct softnet_data *remsd = sd->rps_ipi_list; | |
5102 | ||
5103 | if (remsd) { | |
5104 | sd->rps_ipi_list = NULL; | |
5105 | ||
5106 | local_irq_enable(); | |
5107 | ||
5108 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
773fc8f6 | 5109 | net_rps_send_ipi(remsd); |
e326bed2 ED |
5110 | } else |
5111 | #endif | |
5112 | local_irq_enable(); | |
5113 | } | |
5114 | ||
d75b1ade ED |
5115 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
5116 | { | |
5117 | #ifdef CONFIG_RPS | |
5118 | return sd->rps_ipi_list != NULL; | |
5119 | #else | |
5120 | return false; | |
5121 | #endif | |
5122 | } | |
5123 | ||
bea3348e | 5124 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 5125 | { |
eecfd7c4 | 5126 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
5127 | bool again = true; |
5128 | int work = 0; | |
1da177e4 | 5129 | |
e326bed2 ED |
5130 | /* Check if we have pending ipi, its better to send them now, |
5131 | * not waiting net_rx_action() end. | |
5132 | */ | |
d75b1ade | 5133 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
5134 | local_irq_disable(); |
5135 | net_rps_action_and_irq_enable(sd); | |
5136 | } | |
d75b1ade | 5137 | |
3d48b53f | 5138 | napi->weight = dev_rx_weight; |
145dd5f9 | 5139 | while (again) { |
1da177e4 | 5140 | struct sk_buff *skb; |
6e7676c1 CG |
5141 | |
5142 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 5143 | rcu_read_lock(); |
6e7676c1 | 5144 | __netif_receive_skb(skb); |
2c17d27c | 5145 | rcu_read_unlock(); |
76cc8b13 | 5146 | input_queue_head_incr(sd); |
145dd5f9 | 5147 | if (++work >= quota) |
76cc8b13 | 5148 | return work; |
145dd5f9 | 5149 | |
6e7676c1 | 5150 | } |
1da177e4 | 5151 | |
145dd5f9 | 5152 | local_irq_disable(); |
e36fa2f7 | 5153 | rps_lock(sd); |
11ef7a89 | 5154 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
5155 | /* |
5156 | * Inline a custom version of __napi_complete(). | |
5157 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
5158 | * and NAPI_STATE_SCHED is the only possible flag set |
5159 | * on backlog. | |
5160 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
5161 | * and we dont need an smp_mb() memory barrier. |
5162 | */ | |
eecfd7c4 | 5163 | napi->state = 0; |
145dd5f9 PA |
5164 | again = false; |
5165 | } else { | |
5166 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
5167 | &sd->process_queue); | |
bea3348e | 5168 | } |
e36fa2f7 | 5169 | rps_unlock(sd); |
145dd5f9 | 5170 | local_irq_enable(); |
6e7676c1 | 5171 | } |
1da177e4 | 5172 | |
bea3348e SH |
5173 | return work; |
5174 | } | |
1da177e4 | 5175 | |
bea3348e SH |
5176 | /** |
5177 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 5178 | * @n: entry to schedule |
bea3348e | 5179 | * |
bc9ad166 ED |
5180 | * The entry's receive function will be scheduled to run. |
5181 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 5182 | */ |
b5606c2d | 5183 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
5184 | { |
5185 | unsigned long flags; | |
1da177e4 | 5186 | |
bea3348e | 5187 | local_irq_save(flags); |
903ceff7 | 5188 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 5189 | local_irq_restore(flags); |
1da177e4 | 5190 | } |
bea3348e SH |
5191 | EXPORT_SYMBOL(__napi_schedule); |
5192 | ||
39e6c820 ED |
5193 | /** |
5194 | * napi_schedule_prep - check if napi can be scheduled | |
5195 | * @n: napi context | |
5196 | * | |
5197 | * Test if NAPI routine is already running, and if not mark | |
5198 | * it as running. This is used as a condition variable | |
5199 | * insure only one NAPI poll instance runs. We also make | |
5200 | * sure there is no pending NAPI disable. | |
5201 | */ | |
5202 | bool napi_schedule_prep(struct napi_struct *n) | |
5203 | { | |
5204 | unsigned long val, new; | |
5205 | ||
5206 | do { | |
5207 | val = READ_ONCE(n->state); | |
5208 | if (unlikely(val & NAPIF_STATE_DISABLE)) | |
5209 | return false; | |
5210 | new = val | NAPIF_STATE_SCHED; | |
5211 | ||
5212 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
5213 | * This was suggested by Alexander Duyck, as compiler | |
5214 | * emits better code than : | |
5215 | * if (val & NAPIF_STATE_SCHED) | |
5216 | * new |= NAPIF_STATE_MISSED; | |
5217 | */ | |
5218 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
5219 | NAPIF_STATE_MISSED; | |
5220 | } while (cmpxchg(&n->state, val, new) != val); | |
5221 | ||
5222 | return !(val & NAPIF_STATE_SCHED); | |
5223 | } | |
5224 | EXPORT_SYMBOL(napi_schedule_prep); | |
5225 | ||
bc9ad166 ED |
5226 | /** |
5227 | * __napi_schedule_irqoff - schedule for receive | |
5228 | * @n: entry to schedule | |
5229 | * | |
5230 | * Variant of __napi_schedule() assuming hard irqs are masked | |
5231 | */ | |
5232 | void __napi_schedule_irqoff(struct napi_struct *n) | |
5233 | { | |
5234 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
5235 | } | |
5236 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
5237 | ||
364b6055 | 5238 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 5239 | { |
39e6c820 | 5240 | unsigned long flags, val, new; |
d565b0a1 HX |
5241 | |
5242 | /* | |
217f6974 ED |
5243 | * 1) Don't let napi dequeue from the cpu poll list |
5244 | * just in case its running on a different cpu. | |
5245 | * 2) If we are busy polling, do nothing here, we have | |
5246 | * the guarantee we will be called later. | |
d565b0a1 | 5247 | */ |
217f6974 ED |
5248 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
5249 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 5250 | return false; |
d565b0a1 | 5251 | |
3b47d303 ED |
5252 | if (n->gro_list) { |
5253 | unsigned long timeout = 0; | |
d75b1ade | 5254 | |
3b47d303 ED |
5255 | if (work_done) |
5256 | timeout = n->dev->gro_flush_timeout; | |
5257 | ||
5258 | if (timeout) | |
5259 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
5260 | HRTIMER_MODE_REL_PINNED); | |
5261 | else | |
5262 | napi_gro_flush(n, false); | |
5263 | } | |
02c1602e | 5264 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
5265 | /* If n->poll_list is not empty, we need to mask irqs */ |
5266 | local_irq_save(flags); | |
02c1602e | 5267 | list_del_init(&n->poll_list); |
d75b1ade ED |
5268 | local_irq_restore(flags); |
5269 | } | |
39e6c820 ED |
5270 | |
5271 | do { | |
5272 | val = READ_ONCE(n->state); | |
5273 | ||
5274 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); | |
5275 | ||
5276 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); | |
5277 | ||
5278 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
5279 | * because we will call napi->poll() one more time. | |
5280 | * This C code was suggested by Alexander Duyck to help gcc. | |
5281 | */ | |
5282 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
5283 | NAPIF_STATE_SCHED; | |
5284 | } while (cmpxchg(&n->state, val, new) != val); | |
5285 | ||
5286 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
5287 | __napi_schedule(n); | |
5288 | return false; | |
5289 | } | |
5290 | ||
364b6055 | 5291 | return true; |
d565b0a1 | 5292 | } |
3b47d303 | 5293 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 5294 | |
af12fa6e | 5295 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 5296 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
5297 | { |
5298 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
5299 | struct napi_struct *napi; | |
5300 | ||
5301 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
5302 | if (napi->napi_id == napi_id) | |
5303 | return napi; | |
5304 | ||
5305 | return NULL; | |
5306 | } | |
02d62e86 ED |
5307 | |
5308 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
217f6974 | 5309 | |
ce6aea93 | 5310 | #define BUSY_POLL_BUDGET 8 |
217f6974 ED |
5311 | |
5312 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) | |
5313 | { | |
5314 | int rc; | |
5315 | ||
39e6c820 ED |
5316 | /* Busy polling means there is a high chance device driver hard irq |
5317 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
5318 | * set in napi_schedule_prep(). | |
5319 | * Since we are about to call napi->poll() once more, we can safely | |
5320 | * clear NAPI_STATE_MISSED. | |
5321 | * | |
5322 | * Note: x86 could use a single "lock and ..." instruction | |
5323 | * to perform these two clear_bit() | |
5324 | */ | |
5325 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
5326 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
5327 | ||
5328 | local_bh_disable(); | |
5329 | ||
5330 | /* All we really want here is to re-enable device interrupts. | |
5331 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
5332 | */ | |
5333 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
1e22391e | 5334 | trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); |
217f6974 ED |
5335 | netpoll_poll_unlock(have_poll_lock); |
5336 | if (rc == BUSY_POLL_BUDGET) | |
5337 | __napi_schedule(napi); | |
5338 | local_bh_enable(); | |
217f6974 ED |
5339 | } |
5340 | ||
7db6b048 SS |
5341 | void napi_busy_loop(unsigned int napi_id, |
5342 | bool (*loop_end)(void *, unsigned long), | |
5343 | void *loop_end_arg) | |
02d62e86 | 5344 | { |
7db6b048 | 5345 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 5346 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 5347 | void *have_poll_lock = NULL; |
02d62e86 | 5348 | struct napi_struct *napi; |
217f6974 ED |
5349 | |
5350 | restart: | |
217f6974 | 5351 | napi_poll = NULL; |
02d62e86 | 5352 | |
2a028ecb | 5353 | rcu_read_lock(); |
02d62e86 | 5354 | |
545cd5e5 | 5355 | napi = napi_by_id(napi_id); |
02d62e86 ED |
5356 | if (!napi) |
5357 | goto out; | |
5358 | ||
217f6974 ED |
5359 | preempt_disable(); |
5360 | for (;;) { | |
2b5cd0df AD |
5361 | int work = 0; |
5362 | ||
2a028ecb | 5363 | local_bh_disable(); |
217f6974 ED |
5364 | if (!napi_poll) { |
5365 | unsigned long val = READ_ONCE(napi->state); | |
5366 | ||
5367 | /* If multiple threads are competing for this napi, | |
5368 | * we avoid dirtying napi->state as much as we can. | |
5369 | */ | |
5370 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
5371 | NAPIF_STATE_IN_BUSY_POLL)) | |
5372 | goto count; | |
5373 | if (cmpxchg(&napi->state, val, | |
5374 | val | NAPIF_STATE_IN_BUSY_POLL | | |
5375 | NAPIF_STATE_SCHED) != val) | |
5376 | goto count; | |
5377 | have_poll_lock = netpoll_poll_lock(napi); | |
5378 | napi_poll = napi->poll; | |
5379 | } | |
2b5cd0df AD |
5380 | work = napi_poll(napi, BUSY_POLL_BUDGET); |
5381 | trace_napi_poll(napi, work, BUSY_POLL_BUDGET); | |
217f6974 | 5382 | count: |
2b5cd0df | 5383 | if (work > 0) |
7db6b048 | 5384 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 5385 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
2a028ecb | 5386 | local_bh_enable(); |
02d62e86 | 5387 | |
7db6b048 | 5388 | if (!loop_end || loop_end(loop_end_arg, start_time)) |
217f6974 | 5389 | break; |
02d62e86 | 5390 | |
217f6974 ED |
5391 | if (unlikely(need_resched())) { |
5392 | if (napi_poll) | |
5393 | busy_poll_stop(napi, have_poll_lock); | |
5394 | preempt_enable(); | |
5395 | rcu_read_unlock(); | |
5396 | cond_resched(); | |
7db6b048 | 5397 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 5398 | return; |
217f6974 ED |
5399 | goto restart; |
5400 | } | |
6cdf89b1 | 5401 | cpu_relax(); |
217f6974 ED |
5402 | } |
5403 | if (napi_poll) | |
5404 | busy_poll_stop(napi, have_poll_lock); | |
5405 | preempt_enable(); | |
02d62e86 | 5406 | out: |
2a028ecb | 5407 | rcu_read_unlock(); |
02d62e86 | 5408 | } |
7db6b048 | 5409 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
5410 | |
5411 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 5412 | |
149d6ad8 | 5413 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 5414 | { |
d64b5e85 ED |
5415 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
5416 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 5417 | return; |
af12fa6e | 5418 | |
52bd2d62 | 5419 | spin_lock(&napi_hash_lock); |
af12fa6e | 5420 | |
545cd5e5 | 5421 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 5422 | do { |
545cd5e5 AD |
5423 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
5424 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
5425 | } while (napi_by_id(napi_gen_id)); |
5426 | napi->napi_id = napi_gen_id; | |
af12fa6e | 5427 | |
52bd2d62 ED |
5428 | hlist_add_head_rcu(&napi->napi_hash_node, |
5429 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 5430 | |
52bd2d62 | 5431 | spin_unlock(&napi_hash_lock); |
af12fa6e | 5432 | } |
af12fa6e ET |
5433 | |
5434 | /* Warning : caller is responsible to make sure rcu grace period | |
5435 | * is respected before freeing memory containing @napi | |
5436 | */ | |
34cbe27e | 5437 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 5438 | { |
34cbe27e ED |
5439 | bool rcu_sync_needed = false; |
5440 | ||
af12fa6e ET |
5441 | spin_lock(&napi_hash_lock); |
5442 | ||
34cbe27e ED |
5443 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
5444 | rcu_sync_needed = true; | |
af12fa6e | 5445 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 5446 | } |
af12fa6e | 5447 | spin_unlock(&napi_hash_lock); |
34cbe27e | 5448 | return rcu_sync_needed; |
af12fa6e ET |
5449 | } |
5450 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
5451 | ||
3b47d303 ED |
5452 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
5453 | { | |
5454 | struct napi_struct *napi; | |
5455 | ||
5456 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
5457 | |
5458 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
5459 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
5460 | */ | |
5461 | if (napi->gro_list && !napi_disable_pending(napi) && | |
5462 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) | |
5463 | __napi_schedule_irqoff(napi); | |
3b47d303 ED |
5464 | |
5465 | return HRTIMER_NORESTART; | |
5466 | } | |
5467 | ||
d565b0a1 HX |
5468 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
5469 | int (*poll)(struct napi_struct *, int), int weight) | |
5470 | { | |
5471 | INIT_LIST_HEAD(&napi->poll_list); | |
3b47d303 ED |
5472 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
5473 | napi->timer.function = napi_watchdog; | |
4ae5544f | 5474 | napi->gro_count = 0; |
d565b0a1 | 5475 | napi->gro_list = NULL; |
5d38a079 | 5476 | napi->skb = NULL; |
d565b0a1 | 5477 | napi->poll = poll; |
82dc3c63 ED |
5478 | if (weight > NAPI_POLL_WEIGHT) |
5479 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
5480 | weight, dev->name); | |
d565b0a1 HX |
5481 | napi->weight = weight; |
5482 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 5483 | napi->dev = dev; |
5d38a079 | 5484 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
5485 | napi->poll_owner = -1; |
5486 | #endif | |
5487 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 5488 | napi_hash_add(napi); |
d565b0a1 HX |
5489 | } |
5490 | EXPORT_SYMBOL(netif_napi_add); | |
5491 | ||
3b47d303 ED |
5492 | void napi_disable(struct napi_struct *n) |
5493 | { | |
5494 | might_sleep(); | |
5495 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
5496 | ||
5497 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
5498 | msleep(1); | |
2d8bff12 NH |
5499 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
5500 | msleep(1); | |
3b47d303 ED |
5501 | |
5502 | hrtimer_cancel(&n->timer); | |
5503 | ||
5504 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
5505 | } | |
5506 | EXPORT_SYMBOL(napi_disable); | |
5507 | ||
93d05d4a | 5508 | /* Must be called in process context */ |
d565b0a1 HX |
5509 | void netif_napi_del(struct napi_struct *napi) |
5510 | { | |
93d05d4a ED |
5511 | might_sleep(); |
5512 | if (napi_hash_del(napi)) | |
5513 | synchronize_net(); | |
d7b06636 | 5514 | list_del_init(&napi->dev_list); |
76620aaf | 5515 | napi_free_frags(napi); |
d565b0a1 | 5516 | |
289dccbe | 5517 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 5518 | napi->gro_list = NULL; |
4ae5544f | 5519 | napi->gro_count = 0; |
d565b0a1 HX |
5520 | } |
5521 | EXPORT_SYMBOL(netif_napi_del); | |
5522 | ||
726ce70e HX |
5523 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
5524 | { | |
5525 | void *have; | |
5526 | int work, weight; | |
5527 | ||
5528 | list_del_init(&n->poll_list); | |
5529 | ||
5530 | have = netpoll_poll_lock(n); | |
5531 | ||
5532 | weight = n->weight; | |
5533 | ||
5534 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
5535 | * with netpoll's poll_napi(). Only the entity which | |
5536 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
5537 | * actually make the ->poll() call. Therefore we avoid | |
5538 | * accidentally calling ->poll() when NAPI is not scheduled. | |
5539 | */ | |
5540 | work = 0; | |
5541 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
5542 | work = n->poll(n, weight); | |
1db19db7 | 5543 | trace_napi_poll(n, work, weight); |
726ce70e HX |
5544 | } |
5545 | ||
5546 | WARN_ON_ONCE(work > weight); | |
5547 | ||
5548 | if (likely(work < weight)) | |
5549 | goto out_unlock; | |
5550 | ||
5551 | /* Drivers must not modify the NAPI state if they | |
5552 | * consume the entire weight. In such cases this code | |
5553 | * still "owns" the NAPI instance and therefore can | |
5554 | * move the instance around on the list at-will. | |
5555 | */ | |
5556 | if (unlikely(napi_disable_pending(n))) { | |
5557 | napi_complete(n); | |
5558 | goto out_unlock; | |
5559 | } | |
5560 | ||
5561 | if (n->gro_list) { | |
5562 | /* flush too old packets | |
5563 | * If HZ < 1000, flush all packets. | |
5564 | */ | |
5565 | napi_gro_flush(n, HZ >= 1000); | |
5566 | } | |
5567 | ||
001ce546 HX |
5568 | /* Some drivers may have called napi_schedule |
5569 | * prior to exhausting their budget. | |
5570 | */ | |
5571 | if (unlikely(!list_empty(&n->poll_list))) { | |
5572 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
5573 | n->dev ? n->dev->name : "backlog"); | |
5574 | goto out_unlock; | |
5575 | } | |
5576 | ||
726ce70e HX |
5577 | list_add_tail(&n->poll_list, repoll); |
5578 | ||
5579 | out_unlock: | |
5580 | netpoll_poll_unlock(have); | |
5581 | ||
5582 | return work; | |
5583 | } | |
5584 | ||
0766f788 | 5585 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 5586 | { |
903ceff7 | 5587 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e MW |
5588 | unsigned long time_limit = jiffies + |
5589 | usecs_to_jiffies(netdev_budget_usecs); | |
51b0bded | 5590 | int budget = netdev_budget; |
d75b1ade ED |
5591 | LIST_HEAD(list); |
5592 | LIST_HEAD(repoll); | |
53fb95d3 | 5593 | |
1da177e4 | 5594 | local_irq_disable(); |
d75b1ade ED |
5595 | list_splice_init(&sd->poll_list, &list); |
5596 | local_irq_enable(); | |
1da177e4 | 5597 | |
ceb8d5bf | 5598 | for (;;) { |
bea3348e | 5599 | struct napi_struct *n; |
1da177e4 | 5600 | |
ceb8d5bf HX |
5601 | if (list_empty(&list)) { |
5602 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
f52dffe0 | 5603 | goto out; |
ceb8d5bf HX |
5604 | break; |
5605 | } | |
5606 | ||
6bd373eb HX |
5607 | n = list_first_entry(&list, struct napi_struct, poll_list); |
5608 | budget -= napi_poll(n, &repoll); | |
5609 | ||
d75b1ade | 5610 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
5611 | * Allow this to run for 2 jiffies since which will allow |
5612 | * an average latency of 1.5/HZ. | |
bea3348e | 5613 | */ |
ceb8d5bf HX |
5614 | if (unlikely(budget <= 0 || |
5615 | time_after_eq(jiffies, time_limit))) { | |
5616 | sd->time_squeeze++; | |
5617 | break; | |
5618 | } | |
1da177e4 | 5619 | } |
d75b1ade | 5620 | |
d75b1ade ED |
5621 | local_irq_disable(); |
5622 | ||
5623 | list_splice_tail_init(&sd->poll_list, &list); | |
5624 | list_splice_tail(&repoll, &list); | |
5625 | list_splice(&list, &sd->poll_list); | |
5626 | if (!list_empty(&sd->poll_list)) | |
5627 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
5628 | ||
e326bed2 | 5629 | net_rps_action_and_irq_enable(sd); |
f52dffe0 ED |
5630 | out: |
5631 | __kfree_skb_flush(); | |
1da177e4 LT |
5632 | } |
5633 | ||
aa9d8560 | 5634 | struct netdev_adjacent { |
9ff162a8 | 5635 | struct net_device *dev; |
5d261913 VF |
5636 | |
5637 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 5638 | bool master; |
5d261913 | 5639 | |
5d261913 VF |
5640 | /* counter for the number of times this device was added to us */ |
5641 | u16 ref_nr; | |
5642 | ||
402dae96 VF |
5643 | /* private field for the users */ |
5644 | void *private; | |
5645 | ||
9ff162a8 JP |
5646 | struct list_head list; |
5647 | struct rcu_head rcu; | |
9ff162a8 JP |
5648 | }; |
5649 | ||
6ea29da1 | 5650 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 5651 | struct list_head *adj_list) |
9ff162a8 | 5652 | { |
5d261913 | 5653 | struct netdev_adjacent *adj; |
5d261913 | 5654 | |
2f268f12 | 5655 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
5656 | if (adj->dev == adj_dev) |
5657 | return adj; | |
9ff162a8 JP |
5658 | } |
5659 | return NULL; | |
5660 | } | |
5661 | ||
f1170fd4 DA |
5662 | static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data) |
5663 | { | |
5664 | struct net_device *dev = data; | |
5665 | ||
5666 | return upper_dev == dev; | |
5667 | } | |
5668 | ||
9ff162a8 JP |
5669 | /** |
5670 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
5671 | * @dev: device | |
5672 | * @upper_dev: upper device to check | |
5673 | * | |
5674 | * Find out if a device is linked to specified upper device and return true | |
5675 | * in case it is. Note that this checks only immediate upper device, | |
5676 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
5677 | */ | |
5678 | bool netdev_has_upper_dev(struct net_device *dev, | |
5679 | struct net_device *upper_dev) | |
5680 | { | |
5681 | ASSERT_RTNL(); | |
5682 | ||
f1170fd4 DA |
5683 | return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, |
5684 | upper_dev); | |
9ff162a8 JP |
5685 | } |
5686 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
5687 | ||
1a3f060c DA |
5688 | /** |
5689 | * netdev_has_upper_dev_all - Check if device is linked to an upper device | |
5690 | * @dev: device | |
5691 | * @upper_dev: upper device to check | |
5692 | * | |
5693 | * Find out if a device is linked to specified upper device and return true | |
5694 | * in case it is. Note that this checks the entire upper device chain. | |
5695 | * The caller must hold rcu lock. | |
5696 | */ | |
5697 | ||
1a3f060c DA |
5698 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
5699 | struct net_device *upper_dev) | |
5700 | { | |
5701 | return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, | |
5702 | upper_dev); | |
5703 | } | |
5704 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
5705 | ||
9ff162a8 JP |
5706 | /** |
5707 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
5708 | * @dev: device | |
5709 | * | |
5710 | * Find out if a device is linked to an upper device and return true in case | |
5711 | * it is. The caller must hold the RTNL lock. | |
5712 | */ | |
25cc72a3 | 5713 | bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
5714 | { |
5715 | ASSERT_RTNL(); | |
5716 | ||
f1170fd4 | 5717 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 5718 | } |
25cc72a3 | 5719 | EXPORT_SYMBOL(netdev_has_any_upper_dev); |
9ff162a8 JP |
5720 | |
5721 | /** | |
5722 | * netdev_master_upper_dev_get - Get master upper device | |
5723 | * @dev: device | |
5724 | * | |
5725 | * Find a master upper device and return pointer to it or NULL in case | |
5726 | * it's not there. The caller must hold the RTNL lock. | |
5727 | */ | |
5728 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
5729 | { | |
aa9d8560 | 5730 | struct netdev_adjacent *upper; |
9ff162a8 JP |
5731 | |
5732 | ASSERT_RTNL(); | |
5733 | ||
2f268f12 | 5734 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
5735 | return NULL; |
5736 | ||
2f268f12 | 5737 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 5738 | struct netdev_adjacent, list); |
9ff162a8 JP |
5739 | if (likely(upper->master)) |
5740 | return upper->dev; | |
5741 | return NULL; | |
5742 | } | |
5743 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
5744 | ||
0f524a80 DA |
5745 | /** |
5746 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
5747 | * @dev: device | |
5748 | * | |
5749 | * Find out if a device is linked to a lower device and return true in case | |
5750 | * it is. The caller must hold the RTNL lock. | |
5751 | */ | |
5752 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
5753 | { | |
5754 | ASSERT_RTNL(); | |
5755 | ||
5756 | return !list_empty(&dev->adj_list.lower); | |
5757 | } | |
5758 | ||
b6ccba4c VF |
5759 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
5760 | { | |
5761 | struct netdev_adjacent *adj; | |
5762 | ||
5763 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
5764 | ||
5765 | return adj->private; | |
5766 | } | |
5767 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
5768 | ||
44a40855 VY |
5769 | /** |
5770 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
5771 | * @dev: device | |
5772 | * @iter: list_head ** of the current position | |
5773 | * | |
5774 | * Gets the next device from the dev's upper list, starting from iter | |
5775 | * position. The caller must hold RCU read lock. | |
5776 | */ | |
5777 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
5778 | struct list_head **iter) | |
5779 | { | |
5780 | struct netdev_adjacent *upper; | |
5781 | ||
5782 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5783 | ||
5784 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5785 | ||
5786 | if (&upper->list == &dev->adj_list.upper) | |
5787 | return NULL; | |
5788 | ||
5789 | *iter = &upper->list; | |
5790 | ||
5791 | return upper->dev; | |
5792 | } | |
5793 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
5794 | ||
1a3f060c DA |
5795 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
5796 | struct list_head **iter) | |
5797 | { | |
5798 | struct netdev_adjacent *upper; | |
5799 | ||
5800 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5801 | ||
5802 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5803 | ||
5804 | if (&upper->list == &dev->adj_list.upper) | |
5805 | return NULL; | |
5806 | ||
5807 | *iter = &upper->list; | |
5808 | ||
5809 | return upper->dev; | |
5810 | } | |
5811 | ||
5812 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, | |
5813 | int (*fn)(struct net_device *dev, | |
5814 | void *data), | |
5815 | void *data) | |
5816 | { | |
5817 | struct net_device *udev; | |
5818 | struct list_head *iter; | |
5819 | int ret; | |
5820 | ||
5821 | for (iter = &dev->adj_list.upper, | |
5822 | udev = netdev_next_upper_dev_rcu(dev, &iter); | |
5823 | udev; | |
5824 | udev = netdev_next_upper_dev_rcu(dev, &iter)) { | |
5825 | /* first is the upper device itself */ | |
5826 | ret = fn(udev, data); | |
5827 | if (ret) | |
5828 | return ret; | |
5829 | ||
5830 | /* then look at all of its upper devices */ | |
5831 | ret = netdev_walk_all_upper_dev_rcu(udev, fn, data); | |
5832 | if (ret) | |
5833 | return ret; | |
5834 | } | |
5835 | ||
5836 | return 0; | |
5837 | } | |
5838 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
5839 | ||
31088a11 VF |
5840 | /** |
5841 | * netdev_lower_get_next_private - Get the next ->private from the | |
5842 | * lower neighbour list | |
5843 | * @dev: device | |
5844 | * @iter: list_head ** of the current position | |
5845 | * | |
5846 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5847 | * list, starting from iter position. The caller must hold either hold the | |
5848 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 5849 | * list will remain unchanged. |
31088a11 VF |
5850 | */ |
5851 | void *netdev_lower_get_next_private(struct net_device *dev, | |
5852 | struct list_head **iter) | |
5853 | { | |
5854 | struct netdev_adjacent *lower; | |
5855 | ||
5856 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
5857 | ||
5858 | if (&lower->list == &dev->adj_list.lower) | |
5859 | return NULL; | |
5860 | ||
6859e7df | 5861 | *iter = lower->list.next; |
31088a11 VF |
5862 | |
5863 | return lower->private; | |
5864 | } | |
5865 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
5866 | ||
5867 | /** | |
5868 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
5869 | * lower neighbour list, RCU | |
5870 | * variant | |
5871 | * @dev: device | |
5872 | * @iter: list_head ** of the current position | |
5873 | * | |
5874 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5875 | * list, starting from iter position. The caller must hold RCU read lock. | |
5876 | */ | |
5877 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
5878 | struct list_head **iter) | |
5879 | { | |
5880 | struct netdev_adjacent *lower; | |
5881 | ||
5882 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5883 | ||
5884 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5885 | ||
5886 | if (&lower->list == &dev->adj_list.lower) | |
5887 | return NULL; | |
5888 | ||
6859e7df | 5889 | *iter = &lower->list; |
31088a11 VF |
5890 | |
5891 | return lower->private; | |
5892 | } | |
5893 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
5894 | ||
4085ebe8 VY |
5895 | /** |
5896 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
5897 | * list | |
5898 | * @dev: device | |
5899 | * @iter: list_head ** of the current position | |
5900 | * | |
5901 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
5902 | * list, starting from iter position. The caller must hold RTNL lock or | |
5903 | * its own locking that guarantees that the neighbour lower | |
b469139e | 5904 | * list will remain unchanged. |
4085ebe8 VY |
5905 | */ |
5906 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
5907 | { | |
5908 | struct netdev_adjacent *lower; | |
5909 | ||
cfdd28be | 5910 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
5911 | |
5912 | if (&lower->list == &dev->adj_list.lower) | |
5913 | return NULL; | |
5914 | ||
cfdd28be | 5915 | *iter = lower->list.next; |
4085ebe8 VY |
5916 | |
5917 | return lower->dev; | |
5918 | } | |
5919 | EXPORT_SYMBOL(netdev_lower_get_next); | |
5920 | ||
1a3f060c DA |
5921 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
5922 | struct list_head **iter) | |
5923 | { | |
5924 | struct netdev_adjacent *lower; | |
5925 | ||
46b5ab1a | 5926 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
5927 | |
5928 | if (&lower->list == &dev->adj_list.lower) | |
5929 | return NULL; | |
5930 | ||
46b5ab1a | 5931 | *iter = &lower->list; |
1a3f060c DA |
5932 | |
5933 | return lower->dev; | |
5934 | } | |
5935 | ||
5936 | int netdev_walk_all_lower_dev(struct net_device *dev, | |
5937 | int (*fn)(struct net_device *dev, | |
5938 | void *data), | |
5939 | void *data) | |
5940 | { | |
5941 | struct net_device *ldev; | |
5942 | struct list_head *iter; | |
5943 | int ret; | |
5944 | ||
5945 | for (iter = &dev->adj_list.lower, | |
5946 | ldev = netdev_next_lower_dev(dev, &iter); | |
5947 | ldev; | |
5948 | ldev = netdev_next_lower_dev(dev, &iter)) { | |
5949 | /* first is the lower device itself */ | |
5950 | ret = fn(ldev, data); | |
5951 | if (ret) | |
5952 | return ret; | |
5953 | ||
5954 | /* then look at all of its lower devices */ | |
5955 | ret = netdev_walk_all_lower_dev(ldev, fn, data); | |
5956 | if (ret) | |
5957 | return ret; | |
5958 | } | |
5959 | ||
5960 | return 0; | |
5961 | } | |
5962 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
5963 | ||
1a3f060c DA |
5964 | static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
5965 | struct list_head **iter) | |
5966 | { | |
5967 | struct netdev_adjacent *lower; | |
5968 | ||
5969 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5970 | if (&lower->list == &dev->adj_list.lower) | |
5971 | return NULL; | |
5972 | ||
5973 | *iter = &lower->list; | |
5974 | ||
5975 | return lower->dev; | |
5976 | } | |
5977 | ||
5978 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
5979 | int (*fn)(struct net_device *dev, | |
5980 | void *data), | |
5981 | void *data) | |
5982 | { | |
5983 | struct net_device *ldev; | |
5984 | struct list_head *iter; | |
5985 | int ret; | |
5986 | ||
5987 | for (iter = &dev->adj_list.lower, | |
5988 | ldev = netdev_next_lower_dev_rcu(dev, &iter); | |
5989 | ldev; | |
5990 | ldev = netdev_next_lower_dev_rcu(dev, &iter)) { | |
5991 | /* first is the lower device itself */ | |
5992 | ret = fn(ldev, data); | |
5993 | if (ret) | |
5994 | return ret; | |
5995 | ||
5996 | /* then look at all of its lower devices */ | |
5997 | ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data); | |
5998 | if (ret) | |
5999 | return ret; | |
6000 | } | |
6001 | ||
6002 | return 0; | |
6003 | } | |
6004 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
6005 | ||
e001bfad | 6006 | /** |
6007 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
6008 | * lower neighbour list, RCU | |
6009 | * variant | |
6010 | * @dev: device | |
6011 | * | |
6012 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
6013 | * list. The caller must hold RCU read lock. | |
6014 | */ | |
6015 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
6016 | { | |
6017 | struct netdev_adjacent *lower; | |
6018 | ||
6019 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
6020 | struct netdev_adjacent, list); | |
6021 | if (lower) | |
6022 | return lower->private; | |
6023 | return NULL; | |
6024 | } | |
6025 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
6026 | ||
9ff162a8 JP |
6027 | /** |
6028 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
6029 | * @dev: device | |
6030 | * | |
6031 | * Find a master upper device and return pointer to it or NULL in case | |
6032 | * it's not there. The caller must hold the RCU read lock. | |
6033 | */ | |
6034 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
6035 | { | |
aa9d8560 | 6036 | struct netdev_adjacent *upper; |
9ff162a8 | 6037 | |
2f268f12 | 6038 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 6039 | struct netdev_adjacent, list); |
9ff162a8 JP |
6040 | if (upper && likely(upper->master)) |
6041 | return upper->dev; | |
6042 | return NULL; | |
6043 | } | |
6044 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
6045 | ||
0a59f3a9 | 6046 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
6047 | struct net_device *adj_dev, |
6048 | struct list_head *dev_list) | |
6049 | { | |
6050 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 6051 | |
3ee32707 VF |
6052 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
6053 | "upper_%s" : "lower_%s", adj_dev->name); | |
6054 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
6055 | linkname); | |
6056 | } | |
0a59f3a9 | 6057 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
6058 | char *name, |
6059 | struct list_head *dev_list) | |
6060 | { | |
6061 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 6062 | |
3ee32707 VF |
6063 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
6064 | "upper_%s" : "lower_%s", name); | |
6065 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
6066 | } | |
6067 | ||
7ce64c79 AF |
6068 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
6069 | struct net_device *adj_dev, | |
6070 | struct list_head *dev_list) | |
6071 | { | |
6072 | return (dev_list == &dev->adj_list.upper || | |
6073 | dev_list == &dev->adj_list.lower) && | |
6074 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
6075 | } | |
3ee32707 | 6076 | |
5d261913 VF |
6077 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
6078 | struct net_device *adj_dev, | |
7863c054 | 6079 | struct list_head *dev_list, |
402dae96 | 6080 | void *private, bool master) |
5d261913 VF |
6081 | { |
6082 | struct netdev_adjacent *adj; | |
842d67a7 | 6083 | int ret; |
5d261913 | 6084 | |
6ea29da1 | 6085 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
6086 | |
6087 | if (adj) { | |
790510d9 | 6088 | adj->ref_nr += 1; |
67b62f98 DA |
6089 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
6090 | dev->name, adj_dev->name, adj->ref_nr); | |
6091 | ||
5d261913 VF |
6092 | return 0; |
6093 | } | |
6094 | ||
6095 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
6096 | if (!adj) | |
6097 | return -ENOMEM; | |
6098 | ||
6099 | adj->dev = adj_dev; | |
6100 | adj->master = master; | |
790510d9 | 6101 | adj->ref_nr = 1; |
402dae96 | 6102 | adj->private = private; |
5d261913 | 6103 | dev_hold(adj_dev); |
2f268f12 | 6104 | |
67b62f98 DA |
6105 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
6106 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 6107 | |
7ce64c79 | 6108 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 6109 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
6110 | if (ret) |
6111 | goto free_adj; | |
6112 | } | |
6113 | ||
7863c054 | 6114 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
6115 | if (master) { |
6116 | ret = sysfs_create_link(&(dev->dev.kobj), | |
6117 | &(adj_dev->dev.kobj), "master"); | |
6118 | if (ret) | |
5831d66e | 6119 | goto remove_symlinks; |
842d67a7 | 6120 | |
7863c054 | 6121 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 6122 | } else { |
7863c054 | 6123 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 6124 | } |
5d261913 VF |
6125 | |
6126 | return 0; | |
842d67a7 | 6127 | |
5831d66e | 6128 | remove_symlinks: |
7ce64c79 | 6129 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6130 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
6131 | free_adj: |
6132 | kfree(adj); | |
974daef7 | 6133 | dev_put(adj_dev); |
842d67a7 VF |
6134 | |
6135 | return ret; | |
5d261913 VF |
6136 | } |
6137 | ||
1d143d9f | 6138 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
6139 | struct net_device *adj_dev, | |
93409033 | 6140 | u16 ref_nr, |
1d143d9f | 6141 | struct list_head *dev_list) |
5d261913 VF |
6142 | { |
6143 | struct netdev_adjacent *adj; | |
6144 | ||
67b62f98 DA |
6145 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
6146 | dev->name, adj_dev->name, ref_nr); | |
6147 | ||
6ea29da1 | 6148 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 6149 | |
2f268f12 | 6150 | if (!adj) { |
67b62f98 | 6151 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 6152 | dev->name, adj_dev->name); |
67b62f98 DA |
6153 | WARN_ON(1); |
6154 | return; | |
2f268f12 | 6155 | } |
5d261913 | 6156 | |
93409033 | 6157 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
6158 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
6159 | dev->name, adj_dev->name, ref_nr, | |
6160 | adj->ref_nr - ref_nr); | |
93409033 | 6161 | adj->ref_nr -= ref_nr; |
5d261913 VF |
6162 | return; |
6163 | } | |
6164 | ||
842d67a7 VF |
6165 | if (adj->master) |
6166 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
6167 | ||
7ce64c79 | 6168 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6169 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 6170 | |
5d261913 | 6171 | list_del_rcu(&adj->list); |
67b62f98 | 6172 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 6173 | adj_dev->name, dev->name, adj_dev->name); |
5d261913 VF |
6174 | dev_put(adj_dev); |
6175 | kfree_rcu(adj, rcu); | |
6176 | } | |
6177 | ||
1d143d9f | 6178 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
6179 | struct net_device *upper_dev, | |
6180 | struct list_head *up_list, | |
6181 | struct list_head *down_list, | |
6182 | void *private, bool master) | |
5d261913 VF |
6183 | { |
6184 | int ret; | |
6185 | ||
790510d9 | 6186 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 6187 | private, master); |
5d261913 VF |
6188 | if (ret) |
6189 | return ret; | |
6190 | ||
790510d9 | 6191 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 6192 | private, false); |
5d261913 | 6193 | if (ret) { |
790510d9 | 6194 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
6195 | return ret; |
6196 | } | |
6197 | ||
6198 | return 0; | |
6199 | } | |
6200 | ||
1d143d9f | 6201 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
6202 | struct net_device *upper_dev, | |
93409033 | 6203 | u16 ref_nr, |
1d143d9f | 6204 | struct list_head *up_list, |
6205 | struct list_head *down_list) | |
5d261913 | 6206 | { |
93409033 AC |
6207 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
6208 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
6209 | } |
6210 | ||
1d143d9f | 6211 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
6212 | struct net_device *upper_dev, | |
6213 | void *private, bool master) | |
2f268f12 | 6214 | { |
f1170fd4 DA |
6215 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
6216 | &dev->adj_list.upper, | |
6217 | &upper_dev->adj_list.lower, | |
6218 | private, master); | |
5d261913 VF |
6219 | } |
6220 | ||
1d143d9f | 6221 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
6222 | struct net_device *upper_dev) | |
2f268f12 | 6223 | { |
93409033 | 6224 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
6225 | &dev->adj_list.upper, |
6226 | &upper_dev->adj_list.lower); | |
6227 | } | |
5d261913 | 6228 | |
9ff162a8 | 6229 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 6230 | struct net_device *upper_dev, bool master, |
29bf24af | 6231 | void *upper_priv, void *upper_info) |
9ff162a8 | 6232 | { |
0e4ead9d | 6233 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 | 6234 | int ret = 0; |
9ff162a8 JP |
6235 | |
6236 | ASSERT_RTNL(); | |
6237 | ||
6238 | if (dev == upper_dev) | |
6239 | return -EBUSY; | |
6240 | ||
6241 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
f1170fd4 | 6242 | if (netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
6243 | return -EBUSY; |
6244 | ||
f1170fd4 | 6245 | if (netdev_has_upper_dev(dev, upper_dev)) |
9ff162a8 JP |
6246 | return -EEXIST; |
6247 | ||
6248 | if (master && netdev_master_upper_dev_get(dev)) | |
6249 | return -EBUSY; | |
6250 | ||
0e4ead9d JP |
6251 | changeupper_info.upper_dev = upper_dev; |
6252 | changeupper_info.master = master; | |
6253 | changeupper_info.linking = true; | |
29bf24af | 6254 | changeupper_info.upper_info = upper_info; |
0e4ead9d | 6255 | |
573c7ba0 JP |
6256 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
6257 | &changeupper_info.info); | |
6258 | ret = notifier_to_errno(ret); | |
6259 | if (ret) | |
6260 | return ret; | |
6261 | ||
6dffb044 | 6262 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 6263 | master); |
5d261913 VF |
6264 | if (ret) |
6265 | return ret; | |
9ff162a8 | 6266 | |
b03804e7 IS |
6267 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
6268 | &changeupper_info.info); | |
6269 | ret = notifier_to_errno(ret); | |
6270 | if (ret) | |
f1170fd4 | 6271 | goto rollback; |
b03804e7 | 6272 | |
9ff162a8 | 6273 | return 0; |
5d261913 | 6274 | |
f1170fd4 | 6275 | rollback: |
2f268f12 | 6276 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
6277 | |
6278 | return ret; | |
9ff162a8 JP |
6279 | } |
6280 | ||
6281 | /** | |
6282 | * netdev_upper_dev_link - Add a link to the upper device | |
6283 | * @dev: device | |
6284 | * @upper_dev: new upper device | |
6285 | * | |
6286 | * Adds a link to device which is upper to this one. The caller must hold | |
6287 | * the RTNL lock. On a failure a negative errno code is returned. | |
6288 | * On success the reference counts are adjusted and the function | |
6289 | * returns zero. | |
6290 | */ | |
6291 | int netdev_upper_dev_link(struct net_device *dev, | |
6292 | struct net_device *upper_dev) | |
6293 | { | |
29bf24af | 6294 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL); |
9ff162a8 JP |
6295 | } |
6296 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
6297 | ||
6298 | /** | |
6299 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
6300 | * @dev: device | |
6301 | * @upper_dev: new upper device | |
6dffb044 | 6302 | * @upper_priv: upper device private |
29bf24af | 6303 | * @upper_info: upper info to be passed down via notifier |
9ff162a8 JP |
6304 | * |
6305 | * Adds a link to device which is upper to this one. In this case, only | |
6306 | * one master upper device can be linked, although other non-master devices | |
6307 | * might be linked as well. The caller must hold the RTNL lock. | |
6308 | * On a failure a negative errno code is returned. On success the reference | |
6309 | * counts are adjusted and the function returns zero. | |
6310 | */ | |
6311 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 6312 | struct net_device *upper_dev, |
29bf24af | 6313 | void *upper_priv, void *upper_info) |
9ff162a8 | 6314 | { |
29bf24af JP |
6315 | return __netdev_upper_dev_link(dev, upper_dev, true, |
6316 | upper_priv, upper_info); | |
9ff162a8 JP |
6317 | } |
6318 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
6319 | ||
6320 | /** | |
6321 | * netdev_upper_dev_unlink - Removes a link to upper device | |
6322 | * @dev: device | |
6323 | * @upper_dev: new upper device | |
6324 | * | |
6325 | * Removes a link to device which is upper to this one. The caller must hold | |
6326 | * the RTNL lock. | |
6327 | */ | |
6328 | void netdev_upper_dev_unlink(struct net_device *dev, | |
6329 | struct net_device *upper_dev) | |
6330 | { | |
0e4ead9d | 6331 | struct netdev_notifier_changeupper_info changeupper_info; |
f4563a75 | 6332 | |
9ff162a8 JP |
6333 | ASSERT_RTNL(); |
6334 | ||
0e4ead9d JP |
6335 | changeupper_info.upper_dev = upper_dev; |
6336 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; | |
6337 | changeupper_info.linking = false; | |
6338 | ||
573c7ba0 JP |
6339 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
6340 | &changeupper_info.info); | |
6341 | ||
2f268f12 | 6342 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 6343 | |
0e4ead9d JP |
6344 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
6345 | &changeupper_info.info); | |
9ff162a8 JP |
6346 | } |
6347 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
6348 | ||
61bd3857 MS |
6349 | /** |
6350 | * netdev_bonding_info_change - Dispatch event about slave change | |
6351 | * @dev: device | |
4a26e453 | 6352 | * @bonding_info: info to dispatch |
61bd3857 MS |
6353 | * |
6354 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
6355 | * The caller must hold the RTNL lock. | |
6356 | */ | |
6357 | void netdev_bonding_info_change(struct net_device *dev, | |
6358 | struct netdev_bonding_info *bonding_info) | |
6359 | { | |
6360 | struct netdev_notifier_bonding_info info; | |
6361 | ||
6362 | memcpy(&info.bonding_info, bonding_info, | |
6363 | sizeof(struct netdev_bonding_info)); | |
6364 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, | |
6365 | &info.info); | |
6366 | } | |
6367 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
6368 | ||
2ce1ee17 | 6369 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
6370 | { |
6371 | struct netdev_adjacent *iter; | |
6372 | ||
6373 | struct net *net = dev_net(dev); | |
6374 | ||
6375 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 6376 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6377 | continue; |
6378 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6379 | &iter->dev->adj_list.lower); | |
6380 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
6381 | &dev->adj_list.upper); | |
6382 | } | |
6383 | ||
6384 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 6385 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6386 | continue; |
6387 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6388 | &iter->dev->adj_list.upper); | |
6389 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
6390 | &dev->adj_list.lower); | |
6391 | } | |
6392 | } | |
6393 | ||
2ce1ee17 | 6394 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
6395 | { |
6396 | struct netdev_adjacent *iter; | |
6397 | ||
6398 | struct net *net = dev_net(dev); | |
6399 | ||
6400 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 6401 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6402 | continue; |
6403 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6404 | &iter->dev->adj_list.lower); | |
6405 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6406 | &dev->adj_list.upper); | |
6407 | } | |
6408 | ||
6409 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 6410 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6411 | continue; |
6412 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6413 | &iter->dev->adj_list.upper); | |
6414 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6415 | &dev->adj_list.lower); | |
6416 | } | |
6417 | } | |
6418 | ||
5bb025fa | 6419 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 6420 | { |
5bb025fa | 6421 | struct netdev_adjacent *iter; |
402dae96 | 6422 | |
4c75431a AF |
6423 | struct net *net = dev_net(dev); |
6424 | ||
5bb025fa | 6425 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 6426 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6427 | continue; |
5bb025fa VF |
6428 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6429 | &iter->dev->adj_list.lower); | |
6430 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6431 | &iter->dev->adj_list.lower); | |
6432 | } | |
402dae96 | 6433 | |
5bb025fa | 6434 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 6435 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6436 | continue; |
5bb025fa VF |
6437 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6438 | &iter->dev->adj_list.upper); | |
6439 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6440 | &iter->dev->adj_list.upper); | |
6441 | } | |
402dae96 | 6442 | } |
402dae96 VF |
6443 | |
6444 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
6445 | struct net_device *lower_dev) | |
6446 | { | |
6447 | struct netdev_adjacent *lower; | |
6448 | ||
6449 | if (!lower_dev) | |
6450 | return NULL; | |
6ea29da1 | 6451 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
6452 | if (!lower) |
6453 | return NULL; | |
6454 | ||
6455 | return lower->private; | |
6456 | } | |
6457 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
6458 | ||
4085ebe8 | 6459 | |
952fcfd0 | 6460 | int dev_get_nest_level(struct net_device *dev) |
4085ebe8 VY |
6461 | { |
6462 | struct net_device *lower = NULL; | |
6463 | struct list_head *iter; | |
6464 | int max_nest = -1; | |
6465 | int nest; | |
6466 | ||
6467 | ASSERT_RTNL(); | |
6468 | ||
6469 | netdev_for_each_lower_dev(dev, lower, iter) { | |
952fcfd0 | 6470 | nest = dev_get_nest_level(lower); |
4085ebe8 VY |
6471 | if (max_nest < nest) |
6472 | max_nest = nest; | |
6473 | } | |
6474 | ||
952fcfd0 | 6475 | return max_nest + 1; |
4085ebe8 VY |
6476 | } |
6477 | EXPORT_SYMBOL(dev_get_nest_level); | |
6478 | ||
04d48266 JP |
6479 | /** |
6480 | * netdev_lower_change - Dispatch event about lower device state change | |
6481 | * @lower_dev: device | |
6482 | * @lower_state_info: state to dispatch | |
6483 | * | |
6484 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
6485 | * The caller must hold the RTNL lock. | |
6486 | */ | |
6487 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
6488 | void *lower_state_info) | |
6489 | { | |
6490 | struct netdev_notifier_changelowerstate_info changelowerstate_info; | |
6491 | ||
6492 | ASSERT_RTNL(); | |
6493 | changelowerstate_info.lower_state_info = lower_state_info; | |
6494 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev, | |
6495 | &changelowerstate_info.info); | |
6496 | } | |
6497 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
6498 | ||
b6c40d68 PM |
6499 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
6500 | { | |
d314774c SH |
6501 | const struct net_device_ops *ops = dev->netdev_ops; |
6502 | ||
d2615bf4 | 6503 | if (ops->ndo_change_rx_flags) |
d314774c | 6504 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
6505 | } |
6506 | ||
991fb3f7 | 6507 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6508 | { |
b536db93 | 6509 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
6510 | kuid_t uid; |
6511 | kgid_t gid; | |
1da177e4 | 6512 | |
24023451 PM |
6513 | ASSERT_RTNL(); |
6514 | ||
dad9b335 WC |
6515 | dev->flags |= IFF_PROMISC; |
6516 | dev->promiscuity += inc; | |
6517 | if (dev->promiscuity == 0) { | |
6518 | /* | |
6519 | * Avoid overflow. | |
6520 | * If inc causes overflow, untouch promisc and return error. | |
6521 | */ | |
6522 | if (inc < 0) | |
6523 | dev->flags &= ~IFF_PROMISC; | |
6524 | else { | |
6525 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
6526 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
6527 | dev->name); | |
dad9b335 WC |
6528 | return -EOVERFLOW; |
6529 | } | |
6530 | } | |
52609c0b | 6531 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
6532 | pr_info("device %s %s promiscuous mode\n", |
6533 | dev->name, | |
6534 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
6535 | if (audit_enabled) { |
6536 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
6537 | audit_log(current->audit_context, GFP_ATOMIC, |
6538 | AUDIT_ANOM_PROMISCUOUS, | |
6539 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
6540 | dev->name, (dev->flags & IFF_PROMISC), | |
6541 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 6542 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
6543 | from_kuid(&init_user_ns, uid), |
6544 | from_kgid(&init_user_ns, gid), | |
7759db82 | 6545 | audit_get_sessionid(current)); |
8192b0c4 | 6546 | } |
24023451 | 6547 | |
b6c40d68 | 6548 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 6549 | } |
991fb3f7 ND |
6550 | if (notify) |
6551 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 6552 | return 0; |
1da177e4 LT |
6553 | } |
6554 | ||
4417da66 PM |
6555 | /** |
6556 | * dev_set_promiscuity - update promiscuity count on a device | |
6557 | * @dev: device | |
6558 | * @inc: modifier | |
6559 | * | |
6560 | * Add or remove promiscuity from a device. While the count in the device | |
6561 | * remains above zero the interface remains promiscuous. Once it hits zero | |
6562 | * the device reverts back to normal filtering operation. A negative inc | |
6563 | * value is used to drop promiscuity on the device. | |
dad9b335 | 6564 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 6565 | */ |
dad9b335 | 6566 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 6567 | { |
b536db93 | 6568 | unsigned int old_flags = dev->flags; |
dad9b335 | 6569 | int err; |
4417da66 | 6570 | |
991fb3f7 | 6571 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 6572 | if (err < 0) |
dad9b335 | 6573 | return err; |
4417da66 PM |
6574 | if (dev->flags != old_flags) |
6575 | dev_set_rx_mode(dev); | |
dad9b335 | 6576 | return err; |
4417da66 | 6577 | } |
d1b19dff | 6578 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 6579 | |
991fb3f7 | 6580 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6581 | { |
991fb3f7 | 6582 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 6583 | |
24023451 PM |
6584 | ASSERT_RTNL(); |
6585 | ||
1da177e4 | 6586 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
6587 | dev->allmulti += inc; |
6588 | if (dev->allmulti == 0) { | |
6589 | /* | |
6590 | * Avoid overflow. | |
6591 | * If inc causes overflow, untouch allmulti and return error. | |
6592 | */ | |
6593 | if (inc < 0) | |
6594 | dev->flags &= ~IFF_ALLMULTI; | |
6595 | else { | |
6596 | dev->allmulti -= inc; | |
7b6cd1ce JP |
6597 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
6598 | dev->name); | |
dad9b335 WC |
6599 | return -EOVERFLOW; |
6600 | } | |
6601 | } | |
24023451 | 6602 | if (dev->flags ^ old_flags) { |
b6c40d68 | 6603 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 6604 | dev_set_rx_mode(dev); |
991fb3f7 ND |
6605 | if (notify) |
6606 | __dev_notify_flags(dev, old_flags, | |
6607 | dev->gflags ^ old_gflags); | |
24023451 | 6608 | } |
dad9b335 | 6609 | return 0; |
4417da66 | 6610 | } |
991fb3f7 ND |
6611 | |
6612 | /** | |
6613 | * dev_set_allmulti - update allmulti count on a device | |
6614 | * @dev: device | |
6615 | * @inc: modifier | |
6616 | * | |
6617 | * Add or remove reception of all multicast frames to a device. While the | |
6618 | * count in the device remains above zero the interface remains listening | |
6619 | * to all interfaces. Once it hits zero the device reverts back to normal | |
6620 | * filtering operation. A negative @inc value is used to drop the counter | |
6621 | * when releasing a resource needing all multicasts. | |
6622 | * Return 0 if successful or a negative errno code on error. | |
6623 | */ | |
6624 | ||
6625 | int dev_set_allmulti(struct net_device *dev, int inc) | |
6626 | { | |
6627 | return __dev_set_allmulti(dev, inc, true); | |
6628 | } | |
d1b19dff | 6629 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
6630 | |
6631 | /* | |
6632 | * Upload unicast and multicast address lists to device and | |
6633 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 6634 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
6635 | * are present. |
6636 | */ | |
6637 | void __dev_set_rx_mode(struct net_device *dev) | |
6638 | { | |
d314774c SH |
6639 | const struct net_device_ops *ops = dev->netdev_ops; |
6640 | ||
4417da66 PM |
6641 | /* dev_open will call this function so the list will stay sane. */ |
6642 | if (!(dev->flags&IFF_UP)) | |
6643 | return; | |
6644 | ||
6645 | if (!netif_device_present(dev)) | |
40b77c94 | 6646 | return; |
4417da66 | 6647 | |
01789349 | 6648 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
6649 | /* Unicast addresses changes may only happen under the rtnl, |
6650 | * therefore calling __dev_set_promiscuity here is safe. | |
6651 | */ | |
32e7bfc4 | 6652 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 6653 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 6654 | dev->uc_promisc = true; |
32e7bfc4 | 6655 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 6656 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 6657 | dev->uc_promisc = false; |
4417da66 | 6658 | } |
4417da66 | 6659 | } |
01789349 JP |
6660 | |
6661 | if (ops->ndo_set_rx_mode) | |
6662 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
6663 | } |
6664 | ||
6665 | void dev_set_rx_mode(struct net_device *dev) | |
6666 | { | |
b9e40857 | 6667 | netif_addr_lock_bh(dev); |
4417da66 | 6668 | __dev_set_rx_mode(dev); |
b9e40857 | 6669 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
6670 | } |
6671 | ||
f0db275a SH |
6672 | /** |
6673 | * dev_get_flags - get flags reported to userspace | |
6674 | * @dev: device | |
6675 | * | |
6676 | * Get the combination of flag bits exported through APIs to userspace. | |
6677 | */ | |
95c96174 | 6678 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 6679 | { |
95c96174 | 6680 | unsigned int flags; |
1da177e4 LT |
6681 | |
6682 | flags = (dev->flags & ~(IFF_PROMISC | | |
6683 | IFF_ALLMULTI | | |
b00055aa SR |
6684 | IFF_RUNNING | |
6685 | IFF_LOWER_UP | | |
6686 | IFF_DORMANT)) | | |
1da177e4 LT |
6687 | (dev->gflags & (IFF_PROMISC | |
6688 | IFF_ALLMULTI)); | |
6689 | ||
b00055aa SR |
6690 | if (netif_running(dev)) { |
6691 | if (netif_oper_up(dev)) | |
6692 | flags |= IFF_RUNNING; | |
6693 | if (netif_carrier_ok(dev)) | |
6694 | flags |= IFF_LOWER_UP; | |
6695 | if (netif_dormant(dev)) | |
6696 | flags |= IFF_DORMANT; | |
6697 | } | |
1da177e4 LT |
6698 | |
6699 | return flags; | |
6700 | } | |
d1b19dff | 6701 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 6702 | |
bd380811 | 6703 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 6704 | { |
b536db93 | 6705 | unsigned int old_flags = dev->flags; |
bd380811 | 6706 | int ret; |
1da177e4 | 6707 | |
24023451 PM |
6708 | ASSERT_RTNL(); |
6709 | ||
1da177e4 LT |
6710 | /* |
6711 | * Set the flags on our device. | |
6712 | */ | |
6713 | ||
6714 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
6715 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
6716 | IFF_AUTOMEDIA)) | | |
6717 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
6718 | IFF_ALLMULTI)); | |
6719 | ||
6720 | /* | |
6721 | * Load in the correct multicast list now the flags have changed. | |
6722 | */ | |
6723 | ||
b6c40d68 PM |
6724 | if ((old_flags ^ flags) & IFF_MULTICAST) |
6725 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 6726 | |
4417da66 | 6727 | dev_set_rx_mode(dev); |
1da177e4 LT |
6728 | |
6729 | /* | |
6730 | * Have we downed the interface. We handle IFF_UP ourselves | |
6731 | * according to user attempts to set it, rather than blindly | |
6732 | * setting it. | |
6733 | */ | |
6734 | ||
6735 | ret = 0; | |
7051b88a | 6736 | if ((old_flags ^ flags) & IFF_UP) { |
6737 | if (old_flags & IFF_UP) | |
6738 | __dev_close(dev); | |
6739 | else | |
6740 | ret = __dev_open(dev); | |
6741 | } | |
1da177e4 | 6742 | |
1da177e4 | 6743 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 6744 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 6745 | unsigned int old_flags = dev->flags; |
d1b19dff | 6746 | |
1da177e4 | 6747 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
6748 | |
6749 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
6750 | if (dev->flags != old_flags) | |
6751 | dev_set_rx_mode(dev); | |
1da177e4 LT |
6752 | } |
6753 | ||
6754 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 6755 | * is important. Some (broken) drivers set IFF_PROMISC, when |
6756 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
6757 | */ |
6758 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
6759 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
6760 | ||
1da177e4 | 6761 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 6762 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
6763 | } |
6764 | ||
bd380811 PM |
6765 | return ret; |
6766 | } | |
6767 | ||
a528c219 ND |
6768 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
6769 | unsigned int gchanges) | |
bd380811 PM |
6770 | { |
6771 | unsigned int changes = dev->flags ^ old_flags; | |
6772 | ||
a528c219 | 6773 | if (gchanges) |
7f294054 | 6774 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 6775 | |
bd380811 PM |
6776 | if (changes & IFF_UP) { |
6777 | if (dev->flags & IFF_UP) | |
6778 | call_netdevice_notifiers(NETDEV_UP, dev); | |
6779 | else | |
6780 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
6781 | } | |
6782 | ||
6783 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
6784 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
6785 | struct netdev_notifier_change_info change_info; | |
6786 | ||
6787 | change_info.flags_changed = changes; | |
6788 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
6789 | &change_info.info); | |
6790 | } | |
bd380811 PM |
6791 | } |
6792 | ||
6793 | /** | |
6794 | * dev_change_flags - change device settings | |
6795 | * @dev: device | |
6796 | * @flags: device state flags | |
6797 | * | |
6798 | * Change settings on device based state flags. The flags are | |
6799 | * in the userspace exported format. | |
6800 | */ | |
b536db93 | 6801 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 6802 | { |
b536db93 | 6803 | int ret; |
991fb3f7 | 6804 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
6805 | |
6806 | ret = __dev_change_flags(dev, flags); | |
6807 | if (ret < 0) | |
6808 | return ret; | |
6809 | ||
991fb3f7 | 6810 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 6811 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
6812 | return ret; |
6813 | } | |
d1b19dff | 6814 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 6815 | |
f51048c3 | 6816 | int __dev_set_mtu(struct net_device *dev, int new_mtu) |
2315dc91 VF |
6817 | { |
6818 | const struct net_device_ops *ops = dev->netdev_ops; | |
6819 | ||
6820 | if (ops->ndo_change_mtu) | |
6821 | return ops->ndo_change_mtu(dev, new_mtu); | |
6822 | ||
6823 | dev->mtu = new_mtu; | |
6824 | return 0; | |
6825 | } | |
f51048c3 | 6826 | EXPORT_SYMBOL(__dev_set_mtu); |
2315dc91 | 6827 | |
f0db275a SH |
6828 | /** |
6829 | * dev_set_mtu - Change maximum transfer unit | |
6830 | * @dev: device | |
6831 | * @new_mtu: new transfer unit | |
6832 | * | |
6833 | * Change the maximum transfer size of the network device. | |
6834 | */ | |
1da177e4 LT |
6835 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
6836 | { | |
2315dc91 | 6837 | int err, orig_mtu; |
1da177e4 LT |
6838 | |
6839 | if (new_mtu == dev->mtu) | |
6840 | return 0; | |
6841 | ||
61e84623 JW |
6842 | /* MTU must be positive, and in range */ |
6843 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
6844 | net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n", | |
6845 | dev->name, new_mtu, dev->min_mtu); | |
1da177e4 | 6846 | return -EINVAL; |
61e84623 JW |
6847 | } |
6848 | ||
6849 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
6850 | net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n", | |
a0e65de7 | 6851 | dev->name, new_mtu, dev->max_mtu); |
61e84623 JW |
6852 | return -EINVAL; |
6853 | } | |
1da177e4 LT |
6854 | |
6855 | if (!netif_device_present(dev)) | |
6856 | return -ENODEV; | |
6857 | ||
1d486bfb VF |
6858 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
6859 | err = notifier_to_errno(err); | |
6860 | if (err) | |
6861 | return err; | |
d314774c | 6862 | |
2315dc91 VF |
6863 | orig_mtu = dev->mtu; |
6864 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 6865 | |
2315dc91 VF |
6866 | if (!err) { |
6867 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6868 | err = notifier_to_errno(err); | |
6869 | if (err) { | |
6870 | /* setting mtu back and notifying everyone again, | |
6871 | * so that they have a chance to revert changes. | |
6872 | */ | |
6873 | __dev_set_mtu(dev, orig_mtu); | |
6874 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6875 | } | |
6876 | } | |
1da177e4 LT |
6877 | return err; |
6878 | } | |
d1b19dff | 6879 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 6880 | |
cbda10fa VD |
6881 | /** |
6882 | * dev_set_group - Change group this device belongs to | |
6883 | * @dev: device | |
6884 | * @new_group: group this device should belong to | |
6885 | */ | |
6886 | void dev_set_group(struct net_device *dev, int new_group) | |
6887 | { | |
6888 | dev->group = new_group; | |
6889 | } | |
6890 | EXPORT_SYMBOL(dev_set_group); | |
6891 | ||
f0db275a SH |
6892 | /** |
6893 | * dev_set_mac_address - Change Media Access Control Address | |
6894 | * @dev: device | |
6895 | * @sa: new address | |
6896 | * | |
6897 | * Change the hardware (MAC) address of the device | |
6898 | */ | |
1da177e4 LT |
6899 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
6900 | { | |
d314774c | 6901 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
6902 | int err; |
6903 | ||
d314774c | 6904 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
6905 | return -EOPNOTSUPP; |
6906 | if (sa->sa_family != dev->type) | |
6907 | return -EINVAL; | |
6908 | if (!netif_device_present(dev)) | |
6909 | return -ENODEV; | |
d314774c | 6910 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
6911 | if (err) |
6912 | return err; | |
fbdeca2d | 6913 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 6914 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 6915 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 6916 | return 0; |
1da177e4 | 6917 | } |
d1b19dff | 6918 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 6919 | |
4bf84c35 JP |
6920 | /** |
6921 | * dev_change_carrier - Change device carrier | |
6922 | * @dev: device | |
691b3b7e | 6923 | * @new_carrier: new value |
4bf84c35 JP |
6924 | * |
6925 | * Change device carrier | |
6926 | */ | |
6927 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
6928 | { | |
6929 | const struct net_device_ops *ops = dev->netdev_ops; | |
6930 | ||
6931 | if (!ops->ndo_change_carrier) | |
6932 | return -EOPNOTSUPP; | |
6933 | if (!netif_device_present(dev)) | |
6934 | return -ENODEV; | |
6935 | return ops->ndo_change_carrier(dev, new_carrier); | |
6936 | } | |
6937 | EXPORT_SYMBOL(dev_change_carrier); | |
6938 | ||
66b52b0d JP |
6939 | /** |
6940 | * dev_get_phys_port_id - Get device physical port ID | |
6941 | * @dev: device | |
6942 | * @ppid: port ID | |
6943 | * | |
6944 | * Get device physical port ID | |
6945 | */ | |
6946 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 6947 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
6948 | { |
6949 | const struct net_device_ops *ops = dev->netdev_ops; | |
6950 | ||
6951 | if (!ops->ndo_get_phys_port_id) | |
6952 | return -EOPNOTSUPP; | |
6953 | return ops->ndo_get_phys_port_id(dev, ppid); | |
6954 | } | |
6955 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
6956 | ||
db24a904 DA |
6957 | /** |
6958 | * dev_get_phys_port_name - Get device physical port name | |
6959 | * @dev: device | |
6960 | * @name: port name | |
ed49e650 | 6961 | * @len: limit of bytes to copy to name |
db24a904 DA |
6962 | * |
6963 | * Get device physical port name | |
6964 | */ | |
6965 | int dev_get_phys_port_name(struct net_device *dev, | |
6966 | char *name, size_t len) | |
6967 | { | |
6968 | const struct net_device_ops *ops = dev->netdev_ops; | |
6969 | ||
6970 | if (!ops->ndo_get_phys_port_name) | |
6971 | return -EOPNOTSUPP; | |
6972 | return ops->ndo_get_phys_port_name(dev, name, len); | |
6973 | } | |
6974 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
6975 | ||
d746d707 AK |
6976 | /** |
6977 | * dev_change_proto_down - update protocol port state information | |
6978 | * @dev: device | |
6979 | * @proto_down: new value | |
6980 | * | |
6981 | * This info can be used by switch drivers to set the phys state of the | |
6982 | * port. | |
6983 | */ | |
6984 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
6985 | { | |
6986 | const struct net_device_ops *ops = dev->netdev_ops; | |
6987 | ||
6988 | if (!ops->ndo_change_proto_down) | |
6989 | return -EOPNOTSUPP; | |
6990 | if (!netif_device_present(dev)) | |
6991 | return -ENODEV; | |
6992 | return ops->ndo_change_proto_down(dev, proto_down); | |
6993 | } | |
6994 | EXPORT_SYMBOL(dev_change_proto_down); | |
6995 | ||
ce158e58 | 6996 | u8 __dev_xdp_attached(struct net_device *dev, xdp_op_t xdp_op, u32 *prog_id) |
d67b9cd2 DB |
6997 | { |
6998 | struct netdev_xdp xdp; | |
6999 | ||
7000 | memset(&xdp, 0, sizeof(xdp)); | |
7001 | xdp.command = XDP_QUERY_PROG; | |
7002 | ||
7003 | /* Query must always succeed. */ | |
7004 | WARN_ON(xdp_op(dev, &xdp) < 0); | |
58038695 MKL |
7005 | if (prog_id) |
7006 | *prog_id = xdp.prog_id; | |
7007 | ||
d67b9cd2 DB |
7008 | return xdp.prog_attached; |
7009 | } | |
7010 | ||
7011 | static int dev_xdp_install(struct net_device *dev, xdp_op_t xdp_op, | |
32d60277 | 7012 | struct netlink_ext_ack *extack, u32 flags, |
d67b9cd2 DB |
7013 | struct bpf_prog *prog) |
7014 | { | |
7015 | struct netdev_xdp xdp; | |
7016 | ||
7017 | memset(&xdp, 0, sizeof(xdp)); | |
ee5d032f JK |
7018 | if (flags & XDP_FLAGS_HW_MODE) |
7019 | xdp.command = XDP_SETUP_PROG_HW; | |
7020 | else | |
7021 | xdp.command = XDP_SETUP_PROG; | |
d67b9cd2 | 7022 | xdp.extack = extack; |
32d60277 | 7023 | xdp.flags = flags; |
d67b9cd2 DB |
7024 | xdp.prog = prog; |
7025 | ||
7026 | return xdp_op(dev, &xdp); | |
7027 | } | |
7028 | ||
a7862b45 BB |
7029 | /** |
7030 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
7031 | * @dev: device | |
b5d60989 | 7032 | * @extack: netlink extended ack |
a7862b45 | 7033 | * @fd: new program fd or negative value to clear |
85de8576 | 7034 | * @flags: xdp-related flags |
a7862b45 BB |
7035 | * |
7036 | * Set or clear a bpf program for a device | |
7037 | */ | |
ddf9f970 JK |
7038 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, |
7039 | int fd, u32 flags) | |
a7862b45 BB |
7040 | { |
7041 | const struct net_device_ops *ops = dev->netdev_ops; | |
7042 | struct bpf_prog *prog = NULL; | |
d67b9cd2 | 7043 | xdp_op_t xdp_op, xdp_chk; |
a7862b45 BB |
7044 | int err; |
7045 | ||
85de8576 DB |
7046 | ASSERT_RTNL(); |
7047 | ||
d67b9cd2 | 7048 | xdp_op = xdp_chk = ops->ndo_xdp; |
ee5d032f | 7049 | if (!xdp_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) |
0489df9a | 7050 | return -EOPNOTSUPP; |
b5cdae32 DM |
7051 | if (!xdp_op || (flags & XDP_FLAGS_SKB_MODE)) |
7052 | xdp_op = generic_xdp_install; | |
d67b9cd2 DB |
7053 | if (xdp_op == xdp_chk) |
7054 | xdp_chk = generic_xdp_install; | |
b5cdae32 | 7055 | |
a7862b45 | 7056 | if (fd >= 0) { |
58038695 | 7057 | if (xdp_chk && __dev_xdp_attached(dev, xdp_chk, NULL)) |
d67b9cd2 DB |
7058 | return -EEXIST; |
7059 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && | |
58038695 | 7060 | __dev_xdp_attached(dev, xdp_op, NULL)) |
d67b9cd2 | 7061 | return -EBUSY; |
85de8576 | 7062 | |
a7862b45 BB |
7063 | prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP); |
7064 | if (IS_ERR(prog)) | |
7065 | return PTR_ERR(prog); | |
7066 | } | |
7067 | ||
32d60277 | 7068 | err = dev_xdp_install(dev, xdp_op, extack, flags, prog); |
a7862b45 BB |
7069 | if (err < 0 && prog) |
7070 | bpf_prog_put(prog); | |
7071 | ||
7072 | return err; | |
7073 | } | |
a7862b45 | 7074 | |
1da177e4 LT |
7075 | /** |
7076 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 7077 | * @net: the applicable net namespace |
1da177e4 LT |
7078 | * |
7079 | * Returns a suitable unique value for a new device interface | |
7080 | * number. The caller must hold the rtnl semaphore or the | |
7081 | * dev_base_lock to be sure it remains unique. | |
7082 | */ | |
881d966b | 7083 | static int dev_new_index(struct net *net) |
1da177e4 | 7084 | { |
aa79e66e | 7085 | int ifindex = net->ifindex; |
f4563a75 | 7086 | |
1da177e4 LT |
7087 | for (;;) { |
7088 | if (++ifindex <= 0) | |
7089 | ifindex = 1; | |
881d966b | 7090 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 7091 | return net->ifindex = ifindex; |
1da177e4 LT |
7092 | } |
7093 | } | |
7094 | ||
1da177e4 | 7095 | /* Delayed registration/unregisteration */ |
3b5b34fd | 7096 | static LIST_HEAD(net_todo_list); |
200b916f | 7097 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 7098 | |
6f05f629 | 7099 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 7100 | { |
1da177e4 | 7101 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 7102 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
7103 | } |
7104 | ||
9b5e383c | 7105 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 7106 | { |
e93737b0 | 7107 | struct net_device *dev, *tmp; |
5cde2829 | 7108 | LIST_HEAD(close_head); |
9b5e383c | 7109 | |
93ee31f1 DL |
7110 | BUG_ON(dev_boot_phase); |
7111 | ASSERT_RTNL(); | |
7112 | ||
e93737b0 | 7113 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 7114 | /* Some devices call without registering |
e93737b0 KK |
7115 | * for initialization unwind. Remove those |
7116 | * devices and proceed with the remaining. | |
9b5e383c ED |
7117 | */ |
7118 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
7119 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
7120 | dev->name, dev); | |
93ee31f1 | 7121 | |
9b5e383c | 7122 | WARN_ON(1); |
e93737b0 KK |
7123 | list_del(&dev->unreg_list); |
7124 | continue; | |
9b5e383c | 7125 | } |
449f4544 | 7126 | dev->dismantle = true; |
9b5e383c | 7127 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 7128 | } |
93ee31f1 | 7129 | |
44345724 | 7130 | /* If device is running, close it first. */ |
5cde2829 EB |
7131 | list_for_each_entry(dev, head, unreg_list) |
7132 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 7133 | dev_close_many(&close_head, true); |
93ee31f1 | 7134 | |
44345724 | 7135 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
7136 | /* And unlink it from device chain. */ |
7137 | unlist_netdevice(dev); | |
93ee31f1 | 7138 | |
9b5e383c ED |
7139 | dev->reg_state = NETREG_UNREGISTERING; |
7140 | } | |
41852497 | 7141 | flush_all_backlogs(); |
93ee31f1 DL |
7142 | |
7143 | synchronize_net(); | |
7144 | ||
9b5e383c | 7145 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
7146 | struct sk_buff *skb = NULL; |
7147 | ||
9b5e383c ED |
7148 | /* Shutdown queueing discipline. */ |
7149 | dev_shutdown(dev); | |
93ee31f1 DL |
7150 | |
7151 | ||
9b5e383c | 7152 | /* Notify protocols, that we are about to destroy |
eb13da1a | 7153 | * this device. They should clean all the things. |
7154 | */ | |
9b5e383c | 7155 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
93ee31f1 | 7156 | |
395eea6c MB |
7157 | if (!dev->rtnl_link_ops || |
7158 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
3d3ea5af | 7159 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, |
395eea6c MB |
7160 | GFP_KERNEL); |
7161 | ||
9b5e383c ED |
7162 | /* |
7163 | * Flush the unicast and multicast chains | |
7164 | */ | |
a748ee24 | 7165 | dev_uc_flush(dev); |
22bedad3 | 7166 | dev_mc_flush(dev); |
93ee31f1 | 7167 | |
9b5e383c ED |
7168 | if (dev->netdev_ops->ndo_uninit) |
7169 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 7170 | |
395eea6c MB |
7171 | if (skb) |
7172 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 7173 | |
9ff162a8 JP |
7174 | /* Notifier chain MUST detach us all upper devices. */ |
7175 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
0f524a80 | 7176 | WARN_ON(netdev_has_any_lower_dev(dev)); |
93ee31f1 | 7177 | |
9b5e383c ED |
7178 | /* Remove entries from kobject tree */ |
7179 | netdev_unregister_kobject(dev); | |
024e9679 AD |
7180 | #ifdef CONFIG_XPS |
7181 | /* Remove XPS queueing entries */ | |
7182 | netif_reset_xps_queues_gt(dev, 0); | |
7183 | #endif | |
9b5e383c | 7184 | } |
93ee31f1 | 7185 | |
850a545b | 7186 | synchronize_net(); |
395264d5 | 7187 | |
a5ee1551 | 7188 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
7189 | dev_put(dev); |
7190 | } | |
7191 | ||
7192 | static void rollback_registered(struct net_device *dev) | |
7193 | { | |
7194 | LIST_HEAD(single); | |
7195 | ||
7196 | list_add(&dev->unreg_list, &single); | |
7197 | rollback_registered_many(&single); | |
ceaaec98 | 7198 | list_del(&single); |
93ee31f1 DL |
7199 | } |
7200 | ||
fd867d51 JW |
7201 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
7202 | struct net_device *upper, netdev_features_t features) | |
7203 | { | |
7204 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
7205 | netdev_features_t feature; | |
5ba3f7d6 | 7206 | int feature_bit; |
fd867d51 | 7207 | |
5ba3f7d6 JW |
7208 | for_each_netdev_feature(&upper_disables, feature_bit) { |
7209 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
7210 | if (!(upper->wanted_features & feature) |
7211 | && (features & feature)) { | |
7212 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
7213 | &feature, upper->name); | |
7214 | features &= ~feature; | |
7215 | } | |
7216 | } | |
7217 | ||
7218 | return features; | |
7219 | } | |
7220 | ||
7221 | static void netdev_sync_lower_features(struct net_device *upper, | |
7222 | struct net_device *lower, netdev_features_t features) | |
7223 | { | |
7224 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
7225 | netdev_features_t feature; | |
5ba3f7d6 | 7226 | int feature_bit; |
fd867d51 | 7227 | |
5ba3f7d6 JW |
7228 | for_each_netdev_feature(&upper_disables, feature_bit) { |
7229 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
7230 | if (!(features & feature) && (lower->features & feature)) { |
7231 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
7232 | &feature, lower->name); | |
7233 | lower->wanted_features &= ~feature; | |
7234 | netdev_update_features(lower); | |
7235 | ||
7236 | if (unlikely(lower->features & feature)) | |
7237 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
7238 | &feature, lower->name); | |
7239 | } | |
7240 | } | |
7241 | } | |
7242 | ||
c8f44aff MM |
7243 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
7244 | netdev_features_t features) | |
b63365a2 | 7245 | { |
57422dc5 MM |
7246 | /* Fix illegal checksum combinations */ |
7247 | if ((features & NETIF_F_HW_CSUM) && | |
7248 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 7249 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
7250 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
7251 | } | |
7252 | ||
b63365a2 | 7253 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 7254 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 7255 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 7256 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
7257 | } |
7258 | ||
ec5f0615 PS |
7259 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
7260 | !(features & NETIF_F_IP_CSUM)) { | |
7261 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
7262 | features &= ~NETIF_F_TSO; | |
7263 | features &= ~NETIF_F_TSO_ECN; | |
7264 | } | |
7265 | ||
7266 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
7267 | !(features & NETIF_F_IPV6_CSUM)) { | |
7268 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
7269 | features &= ~NETIF_F_TSO6; | |
7270 | } | |
7271 | ||
b1dc497b AD |
7272 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
7273 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
7274 | features &= ~NETIF_F_TSO_MANGLEID; | |
7275 | ||
31d8b9e0 BH |
7276 | /* TSO ECN requires that TSO is present as well. */ |
7277 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
7278 | features &= ~NETIF_F_TSO_ECN; | |
7279 | ||
212b573f MM |
7280 | /* Software GSO depends on SG. */ |
7281 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 7282 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
7283 | features &= ~NETIF_F_GSO; |
7284 | } | |
7285 | ||
802ab55a AD |
7286 | /* GSO partial features require GSO partial be set */ |
7287 | if ((features & dev->gso_partial_features) && | |
7288 | !(features & NETIF_F_GSO_PARTIAL)) { | |
7289 | netdev_dbg(dev, | |
7290 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
7291 | features &= ~dev->gso_partial_features; | |
7292 | } | |
7293 | ||
b63365a2 HX |
7294 | return features; |
7295 | } | |
b63365a2 | 7296 | |
6cb6a27c | 7297 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 7298 | { |
fd867d51 | 7299 | struct net_device *upper, *lower; |
c8f44aff | 7300 | netdev_features_t features; |
fd867d51 | 7301 | struct list_head *iter; |
e7868a85 | 7302 | int err = -1; |
5455c699 | 7303 | |
87267485 MM |
7304 | ASSERT_RTNL(); |
7305 | ||
5455c699 MM |
7306 | features = netdev_get_wanted_features(dev); |
7307 | ||
7308 | if (dev->netdev_ops->ndo_fix_features) | |
7309 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
7310 | ||
7311 | /* driver might be less strict about feature dependencies */ | |
7312 | features = netdev_fix_features(dev, features); | |
7313 | ||
fd867d51 JW |
7314 | /* some features can't be enabled if they're off an an upper device */ |
7315 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
7316 | features = netdev_sync_upper_features(dev, upper, features); | |
7317 | ||
5455c699 | 7318 | if (dev->features == features) |
e7868a85 | 7319 | goto sync_lower; |
5455c699 | 7320 | |
c8f44aff MM |
7321 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
7322 | &dev->features, &features); | |
5455c699 MM |
7323 | |
7324 | if (dev->netdev_ops->ndo_set_features) | |
7325 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
7326 | else |
7327 | err = 0; | |
5455c699 | 7328 | |
6cb6a27c | 7329 | if (unlikely(err < 0)) { |
5455c699 | 7330 | netdev_err(dev, |
c8f44aff MM |
7331 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
7332 | err, &features, &dev->features); | |
17b85d29 NA |
7333 | /* return non-0 since some features might have changed and |
7334 | * it's better to fire a spurious notification than miss it | |
7335 | */ | |
7336 | return -1; | |
6cb6a27c MM |
7337 | } |
7338 | ||
e7868a85 | 7339 | sync_lower: |
fd867d51 JW |
7340 | /* some features must be disabled on lower devices when disabled |
7341 | * on an upper device (think: bonding master or bridge) | |
7342 | */ | |
7343 | netdev_for_each_lower_dev(dev, lower, iter) | |
7344 | netdev_sync_lower_features(dev, lower, features); | |
7345 | ||
ae847f40 SD |
7346 | if (!err) { |
7347 | netdev_features_t diff = features ^ dev->features; | |
7348 | ||
7349 | if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
7350 | /* udp_tunnel_{get,drop}_rx_info both need | |
7351 | * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the | |
7352 | * device, or they won't do anything. | |
7353 | * Thus we need to update dev->features | |
7354 | * *before* calling udp_tunnel_get_rx_info, | |
7355 | * but *after* calling udp_tunnel_drop_rx_info. | |
7356 | */ | |
7357 | if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
7358 | dev->features = features; | |
7359 | udp_tunnel_get_rx_info(dev); | |
7360 | } else { | |
7361 | udp_tunnel_drop_rx_info(dev); | |
7362 | } | |
7363 | } | |
7364 | ||
6cb6a27c | 7365 | dev->features = features; |
ae847f40 | 7366 | } |
6cb6a27c | 7367 | |
e7868a85 | 7368 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
7369 | } |
7370 | ||
afe12cc8 MM |
7371 | /** |
7372 | * netdev_update_features - recalculate device features | |
7373 | * @dev: the device to check | |
7374 | * | |
7375 | * Recalculate dev->features set and send notifications if it | |
7376 | * has changed. Should be called after driver or hardware dependent | |
7377 | * conditions might have changed that influence the features. | |
7378 | */ | |
6cb6a27c MM |
7379 | void netdev_update_features(struct net_device *dev) |
7380 | { | |
7381 | if (__netdev_update_features(dev)) | |
7382 | netdev_features_change(dev); | |
5455c699 MM |
7383 | } |
7384 | EXPORT_SYMBOL(netdev_update_features); | |
7385 | ||
afe12cc8 MM |
7386 | /** |
7387 | * netdev_change_features - recalculate device features | |
7388 | * @dev: the device to check | |
7389 | * | |
7390 | * Recalculate dev->features set and send notifications even | |
7391 | * if they have not changed. Should be called instead of | |
7392 | * netdev_update_features() if also dev->vlan_features might | |
7393 | * have changed to allow the changes to be propagated to stacked | |
7394 | * VLAN devices. | |
7395 | */ | |
7396 | void netdev_change_features(struct net_device *dev) | |
7397 | { | |
7398 | __netdev_update_features(dev); | |
7399 | netdev_features_change(dev); | |
7400 | } | |
7401 | EXPORT_SYMBOL(netdev_change_features); | |
7402 | ||
fc4a7489 PM |
7403 | /** |
7404 | * netif_stacked_transfer_operstate - transfer operstate | |
7405 | * @rootdev: the root or lower level device to transfer state from | |
7406 | * @dev: the device to transfer operstate to | |
7407 | * | |
7408 | * Transfer operational state from root to device. This is normally | |
7409 | * called when a stacking relationship exists between the root | |
7410 | * device and the device(a leaf device). | |
7411 | */ | |
7412 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
7413 | struct net_device *dev) | |
7414 | { | |
7415 | if (rootdev->operstate == IF_OPER_DORMANT) | |
7416 | netif_dormant_on(dev); | |
7417 | else | |
7418 | netif_dormant_off(dev); | |
7419 | ||
0575c86b ZS |
7420 | if (netif_carrier_ok(rootdev)) |
7421 | netif_carrier_on(dev); | |
7422 | else | |
7423 | netif_carrier_off(dev); | |
fc4a7489 PM |
7424 | } |
7425 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
7426 | ||
a953be53 | 7427 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
7428 | static int netif_alloc_rx_queues(struct net_device *dev) |
7429 | { | |
1b4bf461 | 7430 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 7431 | struct netdev_rx_queue *rx; |
10595902 | 7432 | size_t sz = count * sizeof(*rx); |
1b4bf461 | 7433 | |
bd25fa7b | 7434 | BUG_ON(count < 1); |
1b4bf461 | 7435 | |
dcda9b04 | 7436 | rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
7437 | if (!rx) |
7438 | return -ENOMEM; | |
7439 | ||
bd25fa7b TH |
7440 | dev->_rx = rx; |
7441 | ||
bd25fa7b | 7442 | for (i = 0; i < count; i++) |
fe822240 | 7443 | rx[i].dev = dev; |
1b4bf461 ED |
7444 | return 0; |
7445 | } | |
bf264145 | 7446 | #endif |
1b4bf461 | 7447 | |
aa942104 CG |
7448 | static void netdev_init_one_queue(struct net_device *dev, |
7449 | struct netdev_queue *queue, void *_unused) | |
7450 | { | |
7451 | /* Initialize queue lock */ | |
7452 | spin_lock_init(&queue->_xmit_lock); | |
7453 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
7454 | queue->xmit_lock_owner = -1; | |
b236da69 | 7455 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 7456 | queue->dev = dev; |
114cf580 TH |
7457 | #ifdef CONFIG_BQL |
7458 | dql_init(&queue->dql, HZ); | |
7459 | #endif | |
aa942104 CG |
7460 | } |
7461 | ||
60877a32 ED |
7462 | static void netif_free_tx_queues(struct net_device *dev) |
7463 | { | |
4cb28970 | 7464 | kvfree(dev->_tx); |
60877a32 ED |
7465 | } |
7466 | ||
e6484930 TH |
7467 | static int netif_alloc_netdev_queues(struct net_device *dev) |
7468 | { | |
7469 | unsigned int count = dev->num_tx_queues; | |
7470 | struct netdev_queue *tx; | |
60877a32 | 7471 | size_t sz = count * sizeof(*tx); |
e6484930 | 7472 | |
d339727c ED |
7473 | if (count < 1 || count > 0xffff) |
7474 | return -EINVAL; | |
62b5942a | 7475 | |
dcda9b04 | 7476 | tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
7477 | if (!tx) |
7478 | return -ENOMEM; | |
7479 | ||
e6484930 | 7480 | dev->_tx = tx; |
1d24eb48 | 7481 | |
e6484930 TH |
7482 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
7483 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
7484 | |
7485 | return 0; | |
e6484930 TH |
7486 | } |
7487 | ||
a2029240 DV |
7488 | void netif_tx_stop_all_queues(struct net_device *dev) |
7489 | { | |
7490 | unsigned int i; | |
7491 | ||
7492 | for (i = 0; i < dev->num_tx_queues; i++) { | |
7493 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 7494 | |
a2029240 DV |
7495 | netif_tx_stop_queue(txq); |
7496 | } | |
7497 | } | |
7498 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
7499 | ||
1da177e4 LT |
7500 | /** |
7501 | * register_netdevice - register a network device | |
7502 | * @dev: device to register | |
7503 | * | |
7504 | * Take a completed network device structure and add it to the kernel | |
7505 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7506 | * chain. 0 is returned on success. A negative errno code is returned | |
7507 | * on a failure to set up the device, or if the name is a duplicate. | |
7508 | * | |
7509 | * Callers must hold the rtnl semaphore. You may want | |
7510 | * register_netdev() instead of this. | |
7511 | * | |
7512 | * BUGS: | |
7513 | * The locking appears insufficient to guarantee two parallel registers | |
7514 | * will not get the same name. | |
7515 | */ | |
7516 | ||
7517 | int register_netdevice(struct net_device *dev) | |
7518 | { | |
1da177e4 | 7519 | int ret; |
d314774c | 7520 | struct net *net = dev_net(dev); |
1da177e4 LT |
7521 | |
7522 | BUG_ON(dev_boot_phase); | |
7523 | ASSERT_RTNL(); | |
7524 | ||
b17a7c17 SH |
7525 | might_sleep(); |
7526 | ||
1da177e4 LT |
7527 | /* When net_device's are persistent, this will be fatal. */ |
7528 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 7529 | BUG_ON(!net); |
1da177e4 | 7530 | |
f1f28aa3 | 7531 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 7532 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 7533 | |
828de4f6 | 7534 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
7535 | if (ret < 0) |
7536 | goto out; | |
7537 | ||
1da177e4 | 7538 | /* Init, if this function is available */ |
d314774c SH |
7539 | if (dev->netdev_ops->ndo_init) { |
7540 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
7541 | if (ret) { |
7542 | if (ret > 0) | |
7543 | ret = -EIO; | |
90833aa4 | 7544 | goto out; |
1da177e4 LT |
7545 | } |
7546 | } | |
4ec93edb | 7547 | |
f646968f PM |
7548 | if (((dev->hw_features | dev->features) & |
7549 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
7550 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
7551 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
7552 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
7553 | ret = -EINVAL; | |
7554 | goto err_uninit; | |
7555 | } | |
7556 | ||
9c7dafbf PE |
7557 | ret = -EBUSY; |
7558 | if (!dev->ifindex) | |
7559 | dev->ifindex = dev_new_index(net); | |
7560 | else if (__dev_get_by_index(net, dev->ifindex)) | |
7561 | goto err_uninit; | |
7562 | ||
5455c699 MM |
7563 | /* Transfer changeable features to wanted_features and enable |
7564 | * software offloads (GSO and GRO). | |
7565 | */ | |
7566 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f | 7567 | dev->features |= NETIF_F_SOFT_FEATURES; |
d764a122 SD |
7568 | |
7569 | if (dev->netdev_ops->ndo_udp_tunnel_add) { | |
7570 | dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
7571 | dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
7572 | } | |
7573 | ||
14d1232f | 7574 | dev->wanted_features = dev->features & dev->hw_features; |
1da177e4 | 7575 | |
cbc53e08 | 7576 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 7577 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 7578 | |
7f348a60 AD |
7579 | /* If IPv4 TCP segmentation offload is supported we should also |
7580 | * allow the device to enable segmenting the frame with the option | |
7581 | * of ignoring a static IP ID value. This doesn't enable the | |
7582 | * feature itself but allows the user to enable it later. | |
7583 | */ | |
cbc53e08 AD |
7584 | if (dev->hw_features & NETIF_F_TSO) |
7585 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
7586 | if (dev->vlan_features & NETIF_F_TSO) |
7587 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
7588 | if (dev->mpls_features & NETIF_F_TSO) | |
7589 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
7590 | if (dev->hw_enc_features & NETIF_F_TSO) | |
7591 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 7592 | |
1180e7d6 | 7593 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 7594 | */ |
1180e7d6 | 7595 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 7596 | |
ee579677 PS |
7597 | /* Make NETIF_F_SG inheritable to tunnel devices. |
7598 | */ | |
802ab55a | 7599 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 7600 | |
0d89d203 SH |
7601 | /* Make NETIF_F_SG inheritable to MPLS. |
7602 | */ | |
7603 | dev->mpls_features |= NETIF_F_SG; | |
7604 | ||
7ffbe3fd JB |
7605 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
7606 | ret = notifier_to_errno(ret); | |
7607 | if (ret) | |
7608 | goto err_uninit; | |
7609 | ||
8b41d188 | 7610 | ret = netdev_register_kobject(dev); |
b17a7c17 | 7611 | if (ret) |
7ce1b0ed | 7612 | goto err_uninit; |
b17a7c17 SH |
7613 | dev->reg_state = NETREG_REGISTERED; |
7614 | ||
6cb6a27c | 7615 | __netdev_update_features(dev); |
8e9b59b2 | 7616 | |
1da177e4 LT |
7617 | /* |
7618 | * Default initial state at registry is that the | |
7619 | * device is present. | |
7620 | */ | |
7621 | ||
7622 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7623 | ||
8f4cccbb BH |
7624 | linkwatch_init_dev(dev); |
7625 | ||
1da177e4 | 7626 | dev_init_scheduler(dev); |
1da177e4 | 7627 | dev_hold(dev); |
ce286d32 | 7628 | list_netdevice(dev); |
7bf23575 | 7629 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 7630 | |
948b337e JP |
7631 | /* If the device has permanent device address, driver should |
7632 | * set dev_addr and also addr_assign_type should be set to | |
7633 | * NET_ADDR_PERM (default value). | |
7634 | */ | |
7635 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
7636 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
7637 | ||
1da177e4 | 7638 | /* Notify protocols, that a new device appeared. */ |
056925ab | 7639 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 7640 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
7641 | if (ret) { |
7642 | rollback_registered(dev); | |
7643 | dev->reg_state = NETREG_UNREGISTERED; | |
7644 | } | |
d90a909e EB |
7645 | /* |
7646 | * Prevent userspace races by waiting until the network | |
7647 | * device is fully setup before sending notifications. | |
7648 | */ | |
a2835763 PM |
7649 | if (!dev->rtnl_link_ops || |
7650 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 7651 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
7652 | |
7653 | out: | |
7654 | return ret; | |
7ce1b0ed HX |
7655 | |
7656 | err_uninit: | |
d314774c SH |
7657 | if (dev->netdev_ops->ndo_uninit) |
7658 | dev->netdev_ops->ndo_uninit(dev); | |
cf124db5 DM |
7659 | if (dev->priv_destructor) |
7660 | dev->priv_destructor(dev); | |
7ce1b0ed | 7661 | goto out; |
1da177e4 | 7662 | } |
d1b19dff | 7663 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 7664 | |
937f1ba5 BH |
7665 | /** |
7666 | * init_dummy_netdev - init a dummy network device for NAPI | |
7667 | * @dev: device to init | |
7668 | * | |
7669 | * This takes a network device structure and initialize the minimum | |
7670 | * amount of fields so it can be used to schedule NAPI polls without | |
7671 | * registering a full blown interface. This is to be used by drivers | |
7672 | * that need to tie several hardware interfaces to a single NAPI | |
7673 | * poll scheduler due to HW limitations. | |
7674 | */ | |
7675 | int init_dummy_netdev(struct net_device *dev) | |
7676 | { | |
7677 | /* Clear everything. Note we don't initialize spinlocks | |
7678 | * are they aren't supposed to be taken by any of the | |
7679 | * NAPI code and this dummy netdev is supposed to be | |
7680 | * only ever used for NAPI polls | |
7681 | */ | |
7682 | memset(dev, 0, sizeof(struct net_device)); | |
7683 | ||
7684 | /* make sure we BUG if trying to hit standard | |
7685 | * register/unregister code path | |
7686 | */ | |
7687 | dev->reg_state = NETREG_DUMMY; | |
7688 | ||
937f1ba5 BH |
7689 | /* NAPI wants this */ |
7690 | INIT_LIST_HEAD(&dev->napi_list); | |
7691 | ||
7692 | /* a dummy interface is started by default */ | |
7693 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7694 | set_bit(__LINK_STATE_START, &dev->state); | |
7695 | ||
29b4433d ED |
7696 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
7697 | * because users of this 'device' dont need to change | |
7698 | * its refcount. | |
7699 | */ | |
7700 | ||
937f1ba5 BH |
7701 | return 0; |
7702 | } | |
7703 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
7704 | ||
7705 | ||
1da177e4 LT |
7706 | /** |
7707 | * register_netdev - register a network device | |
7708 | * @dev: device to register | |
7709 | * | |
7710 | * Take a completed network device structure and add it to the kernel | |
7711 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7712 | * chain. 0 is returned on success. A negative errno code is returned | |
7713 | * on a failure to set up the device, or if the name is a duplicate. | |
7714 | * | |
38b4da38 | 7715 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
7716 | * and expands the device name if you passed a format string to |
7717 | * alloc_netdev. | |
7718 | */ | |
7719 | int register_netdev(struct net_device *dev) | |
7720 | { | |
7721 | int err; | |
7722 | ||
7723 | rtnl_lock(); | |
1da177e4 | 7724 | err = register_netdevice(dev); |
1da177e4 LT |
7725 | rtnl_unlock(); |
7726 | return err; | |
7727 | } | |
7728 | EXPORT_SYMBOL(register_netdev); | |
7729 | ||
29b4433d ED |
7730 | int netdev_refcnt_read(const struct net_device *dev) |
7731 | { | |
7732 | int i, refcnt = 0; | |
7733 | ||
7734 | for_each_possible_cpu(i) | |
7735 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
7736 | return refcnt; | |
7737 | } | |
7738 | EXPORT_SYMBOL(netdev_refcnt_read); | |
7739 | ||
2c53040f | 7740 | /** |
1da177e4 | 7741 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 7742 | * @dev: target net_device |
1da177e4 LT |
7743 | * |
7744 | * This is called when unregistering network devices. | |
7745 | * | |
7746 | * Any protocol or device that holds a reference should register | |
7747 | * for netdevice notification, and cleanup and put back the | |
7748 | * reference if they receive an UNREGISTER event. | |
7749 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 7750 | * call dev_put. |
1da177e4 LT |
7751 | */ |
7752 | static void netdev_wait_allrefs(struct net_device *dev) | |
7753 | { | |
7754 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 7755 | int refcnt; |
1da177e4 | 7756 | |
e014debe ED |
7757 | linkwatch_forget_dev(dev); |
7758 | ||
1da177e4 | 7759 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
7760 | refcnt = netdev_refcnt_read(dev); |
7761 | ||
7762 | while (refcnt != 0) { | |
1da177e4 | 7763 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 7764 | rtnl_lock(); |
1da177e4 LT |
7765 | |
7766 | /* Rebroadcast unregister notification */ | |
056925ab | 7767 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 7768 | |
748e2d93 | 7769 | __rtnl_unlock(); |
0115e8e3 | 7770 | rcu_barrier(); |
748e2d93 ED |
7771 | rtnl_lock(); |
7772 | ||
0115e8e3 | 7773 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
7774 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
7775 | &dev->state)) { | |
7776 | /* We must not have linkwatch events | |
7777 | * pending on unregister. If this | |
7778 | * happens, we simply run the queue | |
7779 | * unscheduled, resulting in a noop | |
7780 | * for this device. | |
7781 | */ | |
7782 | linkwatch_run_queue(); | |
7783 | } | |
7784 | ||
6756ae4b | 7785 | __rtnl_unlock(); |
1da177e4 LT |
7786 | |
7787 | rebroadcast_time = jiffies; | |
7788 | } | |
7789 | ||
7790 | msleep(250); | |
7791 | ||
29b4433d ED |
7792 | refcnt = netdev_refcnt_read(dev); |
7793 | ||
1da177e4 | 7794 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
7795 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
7796 | dev->name, refcnt); | |
1da177e4 LT |
7797 | warning_time = jiffies; |
7798 | } | |
7799 | } | |
7800 | } | |
7801 | ||
7802 | /* The sequence is: | |
7803 | * | |
7804 | * rtnl_lock(); | |
7805 | * ... | |
7806 | * register_netdevice(x1); | |
7807 | * register_netdevice(x2); | |
7808 | * ... | |
7809 | * unregister_netdevice(y1); | |
7810 | * unregister_netdevice(y2); | |
7811 | * ... | |
7812 | * rtnl_unlock(); | |
7813 | * free_netdev(y1); | |
7814 | * free_netdev(y2); | |
7815 | * | |
58ec3b4d | 7816 | * We are invoked by rtnl_unlock(). |
1da177e4 | 7817 | * This allows us to deal with problems: |
b17a7c17 | 7818 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
7819 | * without deadlocking with linkwatch via keventd. |
7820 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
7821 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
7822 | * |
7823 | * We must not return until all unregister events added during | |
7824 | * the interval the lock was held have been completed. | |
1da177e4 | 7825 | */ |
1da177e4 LT |
7826 | void netdev_run_todo(void) |
7827 | { | |
626ab0e6 | 7828 | struct list_head list; |
1da177e4 | 7829 | |
1da177e4 | 7830 | /* Snapshot list, allow later requests */ |
626ab0e6 | 7831 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
7832 | |
7833 | __rtnl_unlock(); | |
626ab0e6 | 7834 | |
0115e8e3 ED |
7835 | |
7836 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
7837 | if (!list_empty(&list)) |
7838 | rcu_barrier(); | |
7839 | ||
1da177e4 LT |
7840 | while (!list_empty(&list)) { |
7841 | struct net_device *dev | |
e5e26d75 | 7842 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
7843 | list_del(&dev->todo_list); |
7844 | ||
748e2d93 | 7845 | rtnl_lock(); |
0115e8e3 | 7846 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 7847 | __rtnl_unlock(); |
0115e8e3 | 7848 | |
b17a7c17 | 7849 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 7850 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
7851 | dev->name, dev->reg_state); |
7852 | dump_stack(); | |
7853 | continue; | |
7854 | } | |
1da177e4 | 7855 | |
b17a7c17 | 7856 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 7857 | |
b17a7c17 | 7858 | netdev_wait_allrefs(dev); |
1da177e4 | 7859 | |
b17a7c17 | 7860 | /* paranoia */ |
29b4433d | 7861 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
7862 | BUG_ON(!list_empty(&dev->ptype_all)); |
7863 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
7864 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
7865 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 7866 | WARN_ON(dev->dn_ptr); |
1da177e4 | 7867 | |
cf124db5 DM |
7868 | if (dev->priv_destructor) |
7869 | dev->priv_destructor(dev); | |
7870 | if (dev->needs_free_netdev) | |
7871 | free_netdev(dev); | |
9093bbb2 | 7872 | |
50624c93 EB |
7873 | /* Report a network device has been unregistered */ |
7874 | rtnl_lock(); | |
7875 | dev_net(dev)->dev_unreg_count--; | |
7876 | __rtnl_unlock(); | |
7877 | wake_up(&netdev_unregistering_wq); | |
7878 | ||
9093bbb2 SH |
7879 | /* Free network device */ |
7880 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 7881 | } |
1da177e4 LT |
7882 | } |
7883 | ||
9256645a JW |
7884 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
7885 | * all the same fields in the same order as net_device_stats, with only | |
7886 | * the type differing, but rtnl_link_stats64 may have additional fields | |
7887 | * at the end for newer counters. | |
3cfde79c | 7888 | */ |
77a1abf5 ED |
7889 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
7890 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
7891 | { |
7892 | #if BITS_PER_LONG == 64 | |
9256645a | 7893 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
9af9959e | 7894 | memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); |
9256645a JW |
7895 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7896 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
7897 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 7898 | #else |
9256645a | 7899 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
7900 | const unsigned long *src = (const unsigned long *)netdev_stats; |
7901 | u64 *dst = (u64 *)stats64; | |
7902 | ||
9256645a | 7903 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
7904 | for (i = 0; i < n; i++) |
7905 | dst[i] = src[i]; | |
9256645a JW |
7906 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7907 | memset((char *)stats64 + n * sizeof(u64), 0, | |
7908 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
7909 | #endif |
7910 | } | |
77a1abf5 | 7911 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 7912 | |
eeda3fd6 SH |
7913 | /** |
7914 | * dev_get_stats - get network device statistics | |
7915 | * @dev: device to get statistics from | |
28172739 | 7916 | * @storage: place to store stats |
eeda3fd6 | 7917 | * |
d7753516 BH |
7918 | * Get network statistics from device. Return @storage. |
7919 | * The device driver may provide its own method by setting | |
7920 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
7921 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 7922 | */ |
d7753516 BH |
7923 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
7924 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 7925 | { |
eeda3fd6 SH |
7926 | const struct net_device_ops *ops = dev->netdev_ops; |
7927 | ||
28172739 ED |
7928 | if (ops->ndo_get_stats64) { |
7929 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
7930 | ops->ndo_get_stats64(dev, storage); |
7931 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 7932 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
7933 | } else { |
7934 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 7935 | } |
6f64ec74 ED |
7936 | storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); |
7937 | storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); | |
7938 | storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); | |
28172739 | 7939 | return storage; |
c45d286e | 7940 | } |
eeda3fd6 | 7941 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 7942 | |
24824a09 | 7943 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 7944 | { |
24824a09 | 7945 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 7946 | |
24824a09 ED |
7947 | #ifdef CONFIG_NET_CLS_ACT |
7948 | if (queue) | |
7949 | return queue; | |
7950 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
7951 | if (!queue) | |
7952 | return NULL; | |
7953 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 7954 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
7955 | queue->qdisc_sleeping = &noop_qdisc; |
7956 | rcu_assign_pointer(dev->ingress_queue, queue); | |
7957 | #endif | |
7958 | return queue; | |
bb949fbd DM |
7959 | } |
7960 | ||
2c60db03 ED |
7961 | static const struct ethtool_ops default_ethtool_ops; |
7962 | ||
d07d7507 SG |
7963 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
7964 | const struct ethtool_ops *ops) | |
7965 | { | |
7966 | if (dev->ethtool_ops == &default_ethtool_ops) | |
7967 | dev->ethtool_ops = ops; | |
7968 | } | |
7969 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
7970 | ||
74d332c1 ED |
7971 | void netdev_freemem(struct net_device *dev) |
7972 | { | |
7973 | char *addr = (char *)dev - dev->padded; | |
7974 | ||
4cb28970 | 7975 | kvfree(addr); |
74d332c1 ED |
7976 | } |
7977 | ||
1da177e4 | 7978 | /** |
722c9a0c | 7979 | * alloc_netdev_mqs - allocate network device |
7980 | * @sizeof_priv: size of private data to allocate space for | |
7981 | * @name: device name format string | |
7982 | * @name_assign_type: origin of device name | |
7983 | * @setup: callback to initialize device | |
7984 | * @txqs: the number of TX subqueues to allocate | |
7985 | * @rxqs: the number of RX subqueues to allocate | |
7986 | * | |
7987 | * Allocates a struct net_device with private data area for driver use | |
7988 | * and performs basic initialization. Also allocates subqueue structs | |
7989 | * for each queue on the device. | |
1da177e4 | 7990 | */ |
36909ea4 | 7991 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 7992 | unsigned char name_assign_type, |
36909ea4 TH |
7993 | void (*setup)(struct net_device *), |
7994 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 7995 | { |
1da177e4 | 7996 | struct net_device *dev; |
7943986c | 7997 | size_t alloc_size; |
1ce8e7b5 | 7998 | struct net_device *p; |
1da177e4 | 7999 | |
b6fe17d6 SH |
8000 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
8001 | ||
36909ea4 | 8002 | if (txqs < 1) { |
7b6cd1ce | 8003 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
8004 | return NULL; |
8005 | } | |
8006 | ||
a953be53 | 8007 | #ifdef CONFIG_SYSFS |
36909ea4 | 8008 | if (rxqs < 1) { |
7b6cd1ce | 8009 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
8010 | return NULL; |
8011 | } | |
8012 | #endif | |
8013 | ||
fd2ea0a7 | 8014 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
8015 | if (sizeof_priv) { |
8016 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 8017 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
8018 | alloc_size += sizeof_priv; |
8019 | } | |
8020 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 8021 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 8022 | |
dcda9b04 | 8023 | p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
62b5942a | 8024 | if (!p) |
1da177e4 | 8025 | return NULL; |
1da177e4 | 8026 | |
1ce8e7b5 | 8027 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 8028 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 8029 | |
29b4433d ED |
8030 | dev->pcpu_refcnt = alloc_percpu(int); |
8031 | if (!dev->pcpu_refcnt) | |
74d332c1 | 8032 | goto free_dev; |
ab9c73cc | 8033 | |
ab9c73cc | 8034 | if (dev_addr_init(dev)) |
29b4433d | 8035 | goto free_pcpu; |
ab9c73cc | 8036 | |
22bedad3 | 8037 | dev_mc_init(dev); |
a748ee24 | 8038 | dev_uc_init(dev); |
ccffad25 | 8039 | |
c346dca1 | 8040 | dev_net_set(dev, &init_net); |
1da177e4 | 8041 | |
8d3bdbd5 | 8042 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 8043 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 8044 | |
8d3bdbd5 DM |
8045 | INIT_LIST_HEAD(&dev->napi_list); |
8046 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 8047 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 8048 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
8049 | INIT_LIST_HEAD(&dev->adj_list.upper); |
8050 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
8051 | INIT_LIST_HEAD(&dev->ptype_all); |
8052 | INIT_LIST_HEAD(&dev->ptype_specific); | |
59cc1f61 JK |
8053 | #ifdef CONFIG_NET_SCHED |
8054 | hash_init(dev->qdisc_hash); | |
8055 | #endif | |
02875878 | 8056 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
8057 | setup(dev); |
8058 | ||
a813104d | 8059 | if (!dev->tx_queue_len) { |
f84bb1ea | 8060 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 8061 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 8062 | } |
906470c1 | 8063 | |
36909ea4 TH |
8064 | dev->num_tx_queues = txqs; |
8065 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 8066 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 8067 | goto free_all; |
e8a0464c | 8068 | |
a953be53 | 8069 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
8070 | dev->num_rx_queues = rxqs; |
8071 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 8072 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 8073 | goto free_all; |
df334545 | 8074 | #endif |
0a9627f2 | 8075 | |
1da177e4 | 8076 | strcpy(dev->name, name); |
c835a677 | 8077 | dev->name_assign_type = name_assign_type; |
cbda10fa | 8078 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
8079 | if (!dev->ethtool_ops) |
8080 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 PN |
8081 | |
8082 | nf_hook_ingress_init(dev); | |
8083 | ||
1da177e4 | 8084 | return dev; |
ab9c73cc | 8085 | |
8d3bdbd5 DM |
8086 | free_all: |
8087 | free_netdev(dev); | |
8088 | return NULL; | |
8089 | ||
29b4433d ED |
8090 | free_pcpu: |
8091 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
8092 | free_dev: |
8093 | netdev_freemem(dev); | |
ab9c73cc | 8094 | return NULL; |
1da177e4 | 8095 | } |
36909ea4 | 8096 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
8097 | |
8098 | /** | |
722c9a0c | 8099 | * free_netdev - free network device |
8100 | * @dev: device | |
1da177e4 | 8101 | * |
722c9a0c | 8102 | * This function does the last stage of destroying an allocated device |
8103 | * interface. The reference to the device object is released. If this | |
8104 | * is the last reference then it will be freed.Must be called in process | |
8105 | * context. | |
1da177e4 LT |
8106 | */ |
8107 | void free_netdev(struct net_device *dev) | |
8108 | { | |
d565b0a1 | 8109 | struct napi_struct *p, *n; |
b5cdae32 | 8110 | struct bpf_prog *prog; |
d565b0a1 | 8111 | |
93d05d4a | 8112 | might_sleep(); |
60877a32 | 8113 | netif_free_tx_queues(dev); |
a953be53 | 8114 | #ifdef CONFIG_SYSFS |
10595902 | 8115 | kvfree(dev->_rx); |
fe822240 | 8116 | #endif |
e8a0464c | 8117 | |
33d480ce | 8118 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 8119 | |
f001fde5 JP |
8120 | /* Flush device addresses */ |
8121 | dev_addr_flush(dev); | |
8122 | ||
d565b0a1 HX |
8123 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
8124 | netif_napi_del(p); | |
8125 | ||
29b4433d ED |
8126 | free_percpu(dev->pcpu_refcnt); |
8127 | dev->pcpu_refcnt = NULL; | |
8128 | ||
b5cdae32 DM |
8129 | prog = rcu_dereference_protected(dev->xdp_prog, 1); |
8130 | if (prog) { | |
8131 | bpf_prog_put(prog); | |
8132 | static_key_slow_dec(&generic_xdp_needed); | |
8133 | } | |
8134 | ||
3041a069 | 8135 | /* Compatibility with error handling in drivers */ |
1da177e4 | 8136 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 8137 | netdev_freemem(dev); |
1da177e4 LT |
8138 | return; |
8139 | } | |
8140 | ||
8141 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
8142 | dev->reg_state = NETREG_RELEASED; | |
8143 | ||
43cb76d9 GKH |
8144 | /* will free via device release */ |
8145 | put_device(&dev->dev); | |
1da177e4 | 8146 | } |
d1b19dff | 8147 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 8148 | |
f0db275a SH |
8149 | /** |
8150 | * synchronize_net - Synchronize with packet receive processing | |
8151 | * | |
8152 | * Wait for packets currently being received to be done. | |
8153 | * Does not block later packets from starting. | |
8154 | */ | |
4ec93edb | 8155 | void synchronize_net(void) |
1da177e4 LT |
8156 | { |
8157 | might_sleep(); | |
be3fc413 ED |
8158 | if (rtnl_is_locked()) |
8159 | synchronize_rcu_expedited(); | |
8160 | else | |
8161 | synchronize_rcu(); | |
1da177e4 | 8162 | } |
d1b19dff | 8163 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
8164 | |
8165 | /** | |
44a0873d | 8166 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 8167 | * @dev: device |
44a0873d | 8168 | * @head: list |
6ebfbc06 | 8169 | * |
1da177e4 | 8170 | * This function shuts down a device interface and removes it |
d59b54b1 | 8171 | * from the kernel tables. |
44a0873d | 8172 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
8173 | * |
8174 | * Callers must hold the rtnl semaphore. You may want | |
8175 | * unregister_netdev() instead of this. | |
8176 | */ | |
8177 | ||
44a0873d | 8178 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 8179 | { |
a6620712 HX |
8180 | ASSERT_RTNL(); |
8181 | ||
44a0873d | 8182 | if (head) { |
9fdce099 | 8183 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
8184 | } else { |
8185 | rollback_registered(dev); | |
8186 | /* Finish processing unregister after unlock */ | |
8187 | net_set_todo(dev); | |
8188 | } | |
1da177e4 | 8189 | } |
44a0873d | 8190 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 8191 | |
9b5e383c ED |
8192 | /** |
8193 | * unregister_netdevice_many - unregister many devices | |
8194 | * @head: list of devices | |
87757a91 ED |
8195 | * |
8196 | * Note: As most callers use a stack allocated list_head, | |
8197 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
8198 | */ |
8199 | void unregister_netdevice_many(struct list_head *head) | |
8200 | { | |
8201 | struct net_device *dev; | |
8202 | ||
8203 | if (!list_empty(head)) { | |
8204 | rollback_registered_many(head); | |
8205 | list_for_each_entry(dev, head, unreg_list) | |
8206 | net_set_todo(dev); | |
87757a91 | 8207 | list_del(head); |
9b5e383c ED |
8208 | } |
8209 | } | |
63c8099d | 8210 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 8211 | |
1da177e4 LT |
8212 | /** |
8213 | * unregister_netdev - remove device from the kernel | |
8214 | * @dev: device | |
8215 | * | |
8216 | * This function shuts down a device interface and removes it | |
d59b54b1 | 8217 | * from the kernel tables. |
1da177e4 LT |
8218 | * |
8219 | * This is just a wrapper for unregister_netdevice that takes | |
8220 | * the rtnl semaphore. In general you want to use this and not | |
8221 | * unregister_netdevice. | |
8222 | */ | |
8223 | void unregister_netdev(struct net_device *dev) | |
8224 | { | |
8225 | rtnl_lock(); | |
8226 | unregister_netdevice(dev); | |
8227 | rtnl_unlock(); | |
8228 | } | |
1da177e4 LT |
8229 | EXPORT_SYMBOL(unregister_netdev); |
8230 | ||
ce286d32 EB |
8231 | /** |
8232 | * dev_change_net_namespace - move device to different nethost namespace | |
8233 | * @dev: device | |
8234 | * @net: network namespace | |
8235 | * @pat: If not NULL name pattern to try if the current device name | |
8236 | * is already taken in the destination network namespace. | |
8237 | * | |
8238 | * This function shuts down a device interface and moves it | |
8239 | * to a new network namespace. On success 0 is returned, on | |
8240 | * a failure a netagive errno code is returned. | |
8241 | * | |
8242 | * Callers must hold the rtnl semaphore. | |
8243 | */ | |
8244 | ||
8245 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
8246 | { | |
ce286d32 EB |
8247 | int err; |
8248 | ||
8249 | ASSERT_RTNL(); | |
8250 | ||
8251 | /* Don't allow namespace local devices to be moved. */ | |
8252 | err = -EINVAL; | |
8253 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8254 | goto out; | |
8255 | ||
8256 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
8257 | if (dev->reg_state != NETREG_REGISTERED) |
8258 | goto out; | |
8259 | ||
8260 | /* Get out if there is nothing todo */ | |
8261 | err = 0; | |
878628fb | 8262 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
8263 | goto out; |
8264 | ||
8265 | /* Pick the destination device name, and ensure | |
8266 | * we can use it in the destination network namespace. | |
8267 | */ | |
8268 | err = -EEXIST; | |
d9031024 | 8269 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
8270 | /* We get here if we can't use the current device name */ |
8271 | if (!pat) | |
8272 | goto out; | |
828de4f6 | 8273 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
8274 | goto out; |
8275 | } | |
8276 | ||
8277 | /* | |
8278 | * And now a mini version of register_netdevice unregister_netdevice. | |
8279 | */ | |
8280 | ||
8281 | /* If device is running close it first. */ | |
9b772652 | 8282 | dev_close(dev); |
ce286d32 EB |
8283 | |
8284 | /* And unlink it from device chain */ | |
8285 | err = -ENODEV; | |
8286 | unlist_netdevice(dev); | |
8287 | ||
8288 | synchronize_net(); | |
8289 | ||
8290 | /* Shutdown queueing discipline. */ | |
8291 | dev_shutdown(dev); | |
8292 | ||
8293 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 8294 | * this device. They should clean all the things. |
8295 | * | |
8296 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
8297 | * This is wanted because this way 8021q and macvlan know | |
8298 | * the device is just moving and can keep their slaves up. | |
8299 | */ | |
ce286d32 | 8300 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 G |
8301 | rcu_barrier(); |
8302 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 8303 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
8304 | |
8305 | /* | |
8306 | * Flush the unicast and multicast chains | |
8307 | */ | |
a748ee24 | 8308 | dev_uc_flush(dev); |
22bedad3 | 8309 | dev_mc_flush(dev); |
ce286d32 | 8310 | |
4e66ae2e SH |
8311 | /* Send a netdev-removed uevent to the old namespace */ |
8312 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 8313 | netdev_adjacent_del_links(dev); |
4e66ae2e | 8314 | |
ce286d32 | 8315 | /* Actually switch the network namespace */ |
c346dca1 | 8316 | dev_net_set(dev, net); |
ce286d32 | 8317 | |
ce286d32 | 8318 | /* If there is an ifindex conflict assign a new one */ |
7a66bbc9 | 8319 | if (__dev_get_by_index(net, dev->ifindex)) |
ce286d32 | 8320 | dev->ifindex = dev_new_index(net); |
ce286d32 | 8321 | |
4e66ae2e SH |
8322 | /* Send a netdev-add uevent to the new namespace */ |
8323 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 8324 | netdev_adjacent_add_links(dev); |
4e66ae2e | 8325 | |
8b41d188 | 8326 | /* Fixup kobjects */ |
a1b3f594 | 8327 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 8328 | WARN_ON(err); |
ce286d32 EB |
8329 | |
8330 | /* Add the device back in the hashes */ | |
8331 | list_netdevice(dev); | |
8332 | ||
8333 | /* Notify protocols, that a new device appeared. */ | |
8334 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
8335 | ||
d90a909e EB |
8336 | /* |
8337 | * Prevent userspace races by waiting until the network | |
8338 | * device is fully setup before sending notifications. | |
8339 | */ | |
7f294054 | 8340 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 8341 | |
ce286d32 EB |
8342 | synchronize_net(); |
8343 | err = 0; | |
8344 | out: | |
8345 | return err; | |
8346 | } | |
463d0183 | 8347 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 8348 | |
f0bf90de | 8349 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
8350 | { |
8351 | struct sk_buff **list_skb; | |
1da177e4 | 8352 | struct sk_buff *skb; |
f0bf90de | 8353 | unsigned int cpu; |
97d8b6e3 | 8354 | struct softnet_data *sd, *oldsd, *remsd = NULL; |
1da177e4 | 8355 | |
1da177e4 LT |
8356 | local_irq_disable(); |
8357 | cpu = smp_processor_id(); | |
8358 | sd = &per_cpu(softnet_data, cpu); | |
8359 | oldsd = &per_cpu(softnet_data, oldcpu); | |
8360 | ||
8361 | /* Find end of our completion_queue. */ | |
8362 | list_skb = &sd->completion_queue; | |
8363 | while (*list_skb) | |
8364 | list_skb = &(*list_skb)->next; | |
8365 | /* Append completion queue from offline CPU. */ | |
8366 | *list_skb = oldsd->completion_queue; | |
8367 | oldsd->completion_queue = NULL; | |
8368 | ||
1da177e4 | 8369 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
8370 | if (oldsd->output_queue) { |
8371 | *sd->output_queue_tailp = oldsd->output_queue; | |
8372 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
8373 | oldsd->output_queue = NULL; | |
8374 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
8375 | } | |
ac64da0b ED |
8376 | /* Append NAPI poll list from offline CPU, with one exception : |
8377 | * process_backlog() must be called by cpu owning percpu backlog. | |
8378 | * We properly handle process_queue & input_pkt_queue later. | |
8379 | */ | |
8380 | while (!list_empty(&oldsd->poll_list)) { | |
8381 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
8382 | struct napi_struct, | |
8383 | poll_list); | |
8384 | ||
8385 | list_del_init(&napi->poll_list); | |
8386 | if (napi->poll == process_backlog) | |
8387 | napi->state = 0; | |
8388 | else | |
8389 | ____napi_schedule(sd, napi); | |
264524d5 | 8390 | } |
1da177e4 LT |
8391 | |
8392 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
8393 | local_irq_enable(); | |
8394 | ||
773fc8f6 | 8395 | #ifdef CONFIG_RPS |
8396 | remsd = oldsd->rps_ipi_list; | |
8397 | oldsd->rps_ipi_list = NULL; | |
8398 | #endif | |
8399 | /* send out pending IPI's on offline CPU */ | |
8400 | net_rps_send_ipi(remsd); | |
8401 | ||
1da177e4 | 8402 | /* Process offline CPU's input_pkt_queue */ |
76cc8b13 | 8403 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 8404 | netif_rx_ni(skb); |
76cc8b13 | 8405 | input_queue_head_incr(oldsd); |
fec5e652 | 8406 | } |
ac64da0b | 8407 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 8408 | netif_rx_ni(skb); |
76cc8b13 TH |
8409 | input_queue_head_incr(oldsd); |
8410 | } | |
1da177e4 | 8411 | |
f0bf90de | 8412 | return 0; |
1da177e4 | 8413 | } |
1da177e4 | 8414 | |
7f353bf2 | 8415 | /** |
b63365a2 HX |
8416 | * netdev_increment_features - increment feature set by one |
8417 | * @all: current feature set | |
8418 | * @one: new feature set | |
8419 | * @mask: mask feature set | |
7f353bf2 HX |
8420 | * |
8421 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
8422 | * @one to the master device with current feature set @all. Will not |
8423 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 8424 | */ |
c8f44aff MM |
8425 | netdev_features_t netdev_increment_features(netdev_features_t all, |
8426 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 8427 | { |
c8cd0989 | 8428 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 8429 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 8430 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 8431 | |
a188222b | 8432 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 8433 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 8434 | |
1742f183 | 8435 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
8436 | if (all & NETIF_F_HW_CSUM) |
8437 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
8438 | |
8439 | return all; | |
8440 | } | |
b63365a2 | 8441 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 8442 | |
430f03cd | 8443 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
8444 | { |
8445 | int i; | |
8446 | struct hlist_head *hash; | |
8447 | ||
8448 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
8449 | if (hash != NULL) | |
8450 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
8451 | INIT_HLIST_HEAD(&hash[i]); | |
8452 | ||
8453 | return hash; | |
8454 | } | |
8455 | ||
881d966b | 8456 | /* Initialize per network namespace state */ |
4665079c | 8457 | static int __net_init netdev_init(struct net *net) |
881d966b | 8458 | { |
734b6541 RM |
8459 | if (net != &init_net) |
8460 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 8461 | |
30d97d35 PE |
8462 | net->dev_name_head = netdev_create_hash(); |
8463 | if (net->dev_name_head == NULL) | |
8464 | goto err_name; | |
881d966b | 8465 | |
30d97d35 PE |
8466 | net->dev_index_head = netdev_create_hash(); |
8467 | if (net->dev_index_head == NULL) | |
8468 | goto err_idx; | |
881d966b EB |
8469 | |
8470 | return 0; | |
30d97d35 PE |
8471 | |
8472 | err_idx: | |
8473 | kfree(net->dev_name_head); | |
8474 | err_name: | |
8475 | return -ENOMEM; | |
881d966b EB |
8476 | } |
8477 | ||
f0db275a SH |
8478 | /** |
8479 | * netdev_drivername - network driver for the device | |
8480 | * @dev: network device | |
f0db275a SH |
8481 | * |
8482 | * Determine network driver for device. | |
8483 | */ | |
3019de12 | 8484 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 8485 | { |
cf04a4c7 SH |
8486 | const struct device_driver *driver; |
8487 | const struct device *parent; | |
3019de12 | 8488 | const char *empty = ""; |
6579e57b AV |
8489 | |
8490 | parent = dev->dev.parent; | |
6579e57b | 8491 | if (!parent) |
3019de12 | 8492 | return empty; |
6579e57b AV |
8493 | |
8494 | driver = parent->driver; | |
8495 | if (driver && driver->name) | |
3019de12 DM |
8496 | return driver->name; |
8497 | return empty; | |
6579e57b AV |
8498 | } |
8499 | ||
6ea754eb JP |
8500 | static void __netdev_printk(const char *level, const struct net_device *dev, |
8501 | struct va_format *vaf) | |
256df2f3 | 8502 | { |
b004ff49 | 8503 | if (dev && dev->dev.parent) { |
6ea754eb JP |
8504 | dev_printk_emit(level[1] - '0', |
8505 | dev->dev.parent, | |
8506 | "%s %s %s%s: %pV", | |
8507 | dev_driver_string(dev->dev.parent), | |
8508 | dev_name(dev->dev.parent), | |
8509 | netdev_name(dev), netdev_reg_state(dev), | |
8510 | vaf); | |
b004ff49 | 8511 | } else if (dev) { |
6ea754eb JP |
8512 | printk("%s%s%s: %pV", |
8513 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 8514 | } else { |
6ea754eb | 8515 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 8516 | } |
256df2f3 JP |
8517 | } |
8518 | ||
6ea754eb JP |
8519 | void netdev_printk(const char *level, const struct net_device *dev, |
8520 | const char *format, ...) | |
256df2f3 JP |
8521 | { |
8522 | struct va_format vaf; | |
8523 | va_list args; | |
256df2f3 JP |
8524 | |
8525 | va_start(args, format); | |
8526 | ||
8527 | vaf.fmt = format; | |
8528 | vaf.va = &args; | |
8529 | ||
6ea754eb | 8530 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 8531 | |
256df2f3 | 8532 | va_end(args); |
256df2f3 JP |
8533 | } |
8534 | EXPORT_SYMBOL(netdev_printk); | |
8535 | ||
8536 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 8537 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 8538 | { \ |
256df2f3 JP |
8539 | struct va_format vaf; \ |
8540 | va_list args; \ | |
8541 | \ | |
8542 | va_start(args, fmt); \ | |
8543 | \ | |
8544 | vaf.fmt = fmt; \ | |
8545 | vaf.va = &args; \ | |
8546 | \ | |
6ea754eb | 8547 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 8548 | \ |
256df2f3 | 8549 | va_end(args); \ |
256df2f3 JP |
8550 | } \ |
8551 | EXPORT_SYMBOL(func); | |
8552 | ||
8553 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
8554 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
8555 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
8556 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
8557 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
8558 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
8559 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
8560 | ||
4665079c | 8561 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
8562 | { |
8563 | kfree(net->dev_name_head); | |
8564 | kfree(net->dev_index_head); | |
8565 | } | |
8566 | ||
022cbae6 | 8567 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
8568 | .init = netdev_init, |
8569 | .exit = netdev_exit, | |
8570 | }; | |
8571 | ||
4665079c | 8572 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 8573 | { |
e008b5fc | 8574 | struct net_device *dev, *aux; |
ce286d32 | 8575 | /* |
e008b5fc | 8576 | * Push all migratable network devices back to the |
ce286d32 EB |
8577 | * initial network namespace |
8578 | */ | |
8579 | rtnl_lock(); | |
e008b5fc | 8580 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 8581 | int err; |
aca51397 | 8582 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
8583 | |
8584 | /* Ignore unmoveable devices (i.e. loopback) */ | |
8585 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8586 | continue; | |
8587 | ||
e008b5fc EB |
8588 | /* Leave virtual devices for the generic cleanup */ |
8589 | if (dev->rtnl_link_ops) | |
8590 | continue; | |
d0c082ce | 8591 | |
25985edc | 8592 | /* Push remaining network devices to init_net */ |
aca51397 PE |
8593 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
8594 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 8595 | if (err) { |
7b6cd1ce JP |
8596 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
8597 | __func__, dev->name, err); | |
aca51397 | 8598 | BUG(); |
ce286d32 EB |
8599 | } |
8600 | } | |
8601 | rtnl_unlock(); | |
8602 | } | |
8603 | ||
50624c93 EB |
8604 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
8605 | { | |
8606 | /* Return with the rtnl_lock held when there are no network | |
8607 | * devices unregistering in any network namespace in net_list. | |
8608 | */ | |
8609 | struct net *net; | |
8610 | bool unregistering; | |
ff960a73 | 8611 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 8612 | |
ff960a73 | 8613 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 8614 | for (;;) { |
50624c93 EB |
8615 | unregistering = false; |
8616 | rtnl_lock(); | |
8617 | list_for_each_entry(net, net_list, exit_list) { | |
8618 | if (net->dev_unreg_count > 0) { | |
8619 | unregistering = true; | |
8620 | break; | |
8621 | } | |
8622 | } | |
8623 | if (!unregistering) | |
8624 | break; | |
8625 | __rtnl_unlock(); | |
ff960a73 PZ |
8626 | |
8627 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 8628 | } |
ff960a73 | 8629 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
8630 | } |
8631 | ||
04dc7f6b EB |
8632 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
8633 | { | |
8634 | /* At exit all network devices most be removed from a network | |
b595076a | 8635 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
8636 | * Do this across as many network namespaces as possible to |
8637 | * improve batching efficiency. | |
8638 | */ | |
8639 | struct net_device *dev; | |
8640 | struct net *net; | |
8641 | LIST_HEAD(dev_kill_list); | |
8642 | ||
50624c93 EB |
8643 | /* To prevent network device cleanup code from dereferencing |
8644 | * loopback devices or network devices that have been freed | |
8645 | * wait here for all pending unregistrations to complete, | |
8646 | * before unregistring the loopback device and allowing the | |
8647 | * network namespace be freed. | |
8648 | * | |
8649 | * The netdev todo list containing all network devices | |
8650 | * unregistrations that happen in default_device_exit_batch | |
8651 | * will run in the rtnl_unlock() at the end of | |
8652 | * default_device_exit_batch. | |
8653 | */ | |
8654 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
8655 | list_for_each_entry(net, net_list, exit_list) { |
8656 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 8657 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
8658 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
8659 | else | |
8660 | unregister_netdevice_queue(dev, &dev_kill_list); | |
8661 | } | |
8662 | } | |
8663 | unregister_netdevice_many(&dev_kill_list); | |
8664 | rtnl_unlock(); | |
8665 | } | |
8666 | ||
022cbae6 | 8667 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 8668 | .exit = default_device_exit, |
04dc7f6b | 8669 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
8670 | }; |
8671 | ||
1da177e4 LT |
8672 | /* |
8673 | * Initialize the DEV module. At boot time this walks the device list and | |
8674 | * unhooks any devices that fail to initialise (normally hardware not | |
8675 | * present) and leaves us with a valid list of present and active devices. | |
8676 | * | |
8677 | */ | |
8678 | ||
8679 | /* | |
8680 | * This is called single threaded during boot, so no need | |
8681 | * to take the rtnl semaphore. | |
8682 | */ | |
8683 | static int __init net_dev_init(void) | |
8684 | { | |
8685 | int i, rc = -ENOMEM; | |
8686 | ||
8687 | BUG_ON(!dev_boot_phase); | |
8688 | ||
1da177e4 LT |
8689 | if (dev_proc_init()) |
8690 | goto out; | |
8691 | ||
8b41d188 | 8692 | if (netdev_kobject_init()) |
1da177e4 LT |
8693 | goto out; |
8694 | ||
8695 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 8696 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
8697 | INIT_LIST_HEAD(&ptype_base[i]); |
8698 | ||
62532da9 VY |
8699 | INIT_LIST_HEAD(&offload_base); |
8700 | ||
881d966b EB |
8701 | if (register_pernet_subsys(&netdev_net_ops)) |
8702 | goto out; | |
1da177e4 LT |
8703 | |
8704 | /* | |
8705 | * Initialise the packet receive queues. | |
8706 | */ | |
8707 | ||
6f912042 | 8708 | for_each_possible_cpu(i) { |
41852497 | 8709 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 8710 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 8711 | |
41852497 ED |
8712 | INIT_WORK(flush, flush_backlog); |
8713 | ||
e36fa2f7 | 8714 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 8715 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 8716 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 8717 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 8718 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
8719 | sd->csd.func = rps_trigger_softirq; |
8720 | sd->csd.info = sd; | |
e36fa2f7 | 8721 | sd->cpu = i; |
1e94d72f | 8722 | #endif |
0a9627f2 | 8723 | |
e36fa2f7 ED |
8724 | sd->backlog.poll = process_backlog; |
8725 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
8726 | } |
8727 | ||
1da177e4 LT |
8728 | dev_boot_phase = 0; |
8729 | ||
505d4f73 EB |
8730 | /* The loopback device is special if any other network devices |
8731 | * is present in a network namespace the loopback device must | |
8732 | * be present. Since we now dynamically allocate and free the | |
8733 | * loopback device ensure this invariant is maintained by | |
8734 | * keeping the loopback device as the first device on the | |
8735 | * list of network devices. Ensuring the loopback devices | |
8736 | * is the first device that appears and the last network device | |
8737 | * that disappears. | |
8738 | */ | |
8739 | if (register_pernet_device(&loopback_net_ops)) | |
8740 | goto out; | |
8741 | ||
8742 | if (register_pernet_device(&default_device_ops)) | |
8743 | goto out; | |
8744 | ||
962cf36c CM |
8745 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
8746 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 8747 | |
f0bf90de SAS |
8748 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
8749 | NULL, dev_cpu_dead); | |
8750 | WARN_ON(rc < 0); | |
1da177e4 LT |
8751 | rc = 0; |
8752 | out: | |
8753 | return rc; | |
8754 | } | |
8755 | ||
8756 | subsys_initcall(net_dev_init); |