]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
722c9a0c | 2 | * NET3 Protocol independent device support routines. |
1da177e4 LT |
3 | * |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Derived from the non IP parts of dev.c 1.0.19 | |
722c9a0c | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * | |
14 | * Additional Authors: | |
15 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
16 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
17 | * David Hinds <dahinds@users.sourceforge.net> | |
18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
19 | * Adam Sulmicki <adam@cfar.umd.edu> | |
20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
21 | * | |
22 | * Changes: | |
23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
722c9a0c | 24 | * to 2 if register_netdev gets called |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
1da177e4 LT |
27 | * Alan Cox : device private ioctl copies fields back. |
28 | * Alan Cox : Transmit queue code does relevant | |
29 | * stunts to keep the queue safe. | |
30 | * Alan Cox : Fixed double lock. | |
31 | * Alan Cox : Fixed promisc NULL pointer trap | |
32 | * ???????? : Support the full private ioctl range | |
33 | * Alan Cox : Moved ioctl permission check into | |
34 | * drivers | |
35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
36 | * Alan Cox : 100 backlog just doesn't cut it when | |
37 | * you start doing multicast video 8) | |
38 | * Alan Cox : Rewrote net_bh and list manager. | |
722c9a0c | 39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. |
1da177e4 LT |
40 | * Alan Cox : Took out transmit every packet pass |
41 | * Saved a few bytes in the ioctl handler | |
42 | * Alan Cox : Network driver sets packet type before | |
43 | * calling netif_rx. Saves a function | |
44 | * call a packet. | |
45 | * Alan Cox : Hashed net_bh() | |
46 | * Richard Kooijman: Timestamp fixes. | |
47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
48 | * Alan Cox : Device lock protection. | |
722c9a0c | 49 | * Alan Cox : Fixed nasty side effect of device close |
1da177e4 LT |
50 | * changes. |
51 | * Rudi Cilibrasi : Pass the right thing to | |
52 | * set_mac_address() | |
53 | * Dave Miller : 32bit quantity for the device lock to | |
54 | * make it work out on a Sparc. | |
55 | * Bjorn Ekwall : Added KERNELD hack. | |
56 | * Alan Cox : Cleaned up the backlog initialise. | |
57 | * Craig Metz : SIOCGIFCONF fix if space for under | |
58 | * 1 device. | |
59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
60 | * is no device open function. | |
61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
63 | * Cyrus Durgin : Cleaned for KMOD | |
64 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
65 | * A network device unload needs to purge | |
66 | * the backlog queue. | |
67 | * Paul Rusty Russell : SIOCSIFNAME | |
68 | * Pekka Riikonen : Netdev boot-time settings code | |
69 | * Andrew Morton : Make unregister_netdevice wait | |
722c9a0c | 70 | * indefinitely on dev->refcnt |
71 | * J Hadi Salim : - Backlog queue sampling | |
1da177e4 LT |
72 | * - netif_rx() feedback |
73 | */ | |
74 | ||
7c0f6ba6 | 75 | #include <linux/uaccess.h> |
1da177e4 | 76 | #include <linux/bitops.h> |
4fc268d2 | 77 | #include <linux/capability.h> |
1da177e4 LT |
78 | #include <linux/cpu.h> |
79 | #include <linux/types.h> | |
80 | #include <linux/kernel.h> | |
08e9897d | 81 | #include <linux/hash.h> |
5a0e3ad6 | 82 | #include <linux/slab.h> |
1da177e4 | 83 | #include <linux/sched.h> |
f1083048 | 84 | #include <linux/sched/mm.h> |
4a3e2f71 | 85 | #include <linux/mutex.h> |
1da177e4 LT |
86 | #include <linux/string.h> |
87 | #include <linux/mm.h> | |
88 | #include <linux/socket.h> | |
89 | #include <linux/sockios.h> | |
90 | #include <linux/errno.h> | |
91 | #include <linux/interrupt.h> | |
92 | #include <linux/if_ether.h> | |
93 | #include <linux/netdevice.h> | |
94 | #include <linux/etherdevice.h> | |
0187bdfb | 95 | #include <linux/ethtool.h> |
1da177e4 LT |
96 | #include <linux/notifier.h> |
97 | #include <linux/skbuff.h> | |
a7862b45 | 98 | #include <linux/bpf.h> |
b5cdae32 | 99 | #include <linux/bpf_trace.h> |
457c4cbc | 100 | #include <net/net_namespace.h> |
1da177e4 | 101 | #include <net/sock.h> |
02d62e86 | 102 | #include <net/busy_poll.h> |
1da177e4 | 103 | #include <linux/rtnetlink.h> |
1da177e4 | 104 | #include <linux/stat.h> |
1da177e4 | 105 | #include <net/dst.h> |
fc4099f1 | 106 | #include <net/dst_metadata.h> |
1da177e4 | 107 | #include <net/pkt_sched.h> |
87d83093 | 108 | #include <net/pkt_cls.h> |
1da177e4 | 109 | #include <net/checksum.h> |
44540960 | 110 | #include <net/xfrm.h> |
1da177e4 LT |
111 | #include <linux/highmem.h> |
112 | #include <linux/init.h> | |
1da177e4 | 113 | #include <linux/module.h> |
1da177e4 LT |
114 | #include <linux/netpoll.h> |
115 | #include <linux/rcupdate.h> | |
116 | #include <linux/delay.h> | |
1da177e4 | 117 | #include <net/iw_handler.h> |
1da177e4 | 118 | #include <asm/current.h> |
5bdb9886 | 119 | #include <linux/audit.h> |
db217334 | 120 | #include <linux/dmaengine.h> |
f6a78bfc | 121 | #include <linux/err.h> |
c7fa9d18 | 122 | #include <linux/ctype.h> |
723e98b7 | 123 | #include <linux/if_arp.h> |
6de329e2 | 124 | #include <linux/if_vlan.h> |
8f0f2223 | 125 | #include <linux/ip.h> |
ad55dcaf | 126 | #include <net/ip.h> |
25cd9ba0 | 127 | #include <net/mpls.h> |
8f0f2223 DM |
128 | #include <linux/ipv6.h> |
129 | #include <linux/in.h> | |
b6b2fed1 DM |
130 | #include <linux/jhash.h> |
131 | #include <linux/random.h> | |
9cbc1cb8 | 132 | #include <trace/events/napi.h> |
cf66ba58 | 133 | #include <trace/events/net.h> |
07dc22e7 | 134 | #include <trace/events/skb.h> |
5acbbd42 | 135 | #include <linux/pci.h> |
caeda9b9 | 136 | #include <linux/inetdevice.h> |
c445477d | 137 | #include <linux/cpu_rmap.h> |
c5905afb | 138 | #include <linux/static_key.h> |
af12fa6e | 139 | #include <linux/hashtable.h> |
60877a32 | 140 | #include <linux/vmalloc.h> |
529d0489 | 141 | #include <linux/if_macvlan.h> |
e7fd2885 | 142 | #include <linux/errqueue.h> |
3b47d303 | 143 | #include <linux/hrtimer.h> |
e687ad60 | 144 | #include <linux/netfilter_ingress.h> |
40e4e713 | 145 | #include <linux/crash_dump.h> |
b72b5bf6 | 146 | #include <linux/sctp.h> |
1da177e4 | 147 | |
342709ef PE |
148 | #include "net-sysfs.h" |
149 | ||
d565b0a1 HX |
150 | /* Instead of increasing this, you should create a hash table. */ |
151 | #define MAX_GRO_SKBS 8 | |
152 | ||
5d38a079 HX |
153 | /* This should be increased if a protocol with a bigger head is added. */ |
154 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
155 | ||
1da177e4 | 156 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 157 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
158 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
159 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 160 | static struct list_head offload_base __read_mostly; |
1da177e4 | 161 | |
ae78dbfa | 162 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
163 | static int call_netdevice_notifiers_info(unsigned long val, |
164 | struct net_device *dev, | |
165 | struct netdev_notifier_info *info); | |
ae78dbfa | 166 | |
1da177e4 | 167 | /* |
7562f876 | 168 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
169 | * semaphore. |
170 | * | |
c6d14c84 | 171 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
172 | * |
173 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 174 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
175 | * actual updates. This allows pure readers to access the list even |
176 | * while a writer is preparing to update it. | |
177 | * | |
178 | * To put it another way, dev_base_lock is held for writing only to | |
179 | * protect against pure readers; the rtnl semaphore provides the | |
180 | * protection against other writers. | |
181 | * | |
182 | * See, for example usages, register_netdevice() and | |
183 | * unregister_netdevice(), which must be called with the rtnl | |
184 | * semaphore held. | |
185 | */ | |
1da177e4 | 186 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
187 | EXPORT_SYMBOL(dev_base_lock); |
188 | ||
af12fa6e ET |
189 | /* protects napi_hash addition/deletion and napi_gen_id */ |
190 | static DEFINE_SPINLOCK(napi_hash_lock); | |
191 | ||
52bd2d62 | 192 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 193 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 194 | |
18afa4b0 | 195 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 196 | |
4e985ada TG |
197 | static inline void dev_base_seq_inc(struct net *net) |
198 | { | |
643aa9cb | 199 | while (++net->dev_base_seq == 0) |
200 | ; | |
4e985ada TG |
201 | } |
202 | ||
881d966b | 203 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 204 | { |
8387ff25 | 205 | unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); |
95c96174 | 206 | |
08e9897d | 207 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
208 | } |
209 | ||
881d966b | 210 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 211 | { |
7c28bd0b | 212 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
213 | } |
214 | ||
e36fa2f7 | 215 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
216 | { |
217 | #ifdef CONFIG_RPS | |
e36fa2f7 | 218 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
219 | #endif |
220 | } | |
221 | ||
e36fa2f7 | 222 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
223 | { |
224 | #ifdef CONFIG_RPS | |
e36fa2f7 | 225 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
226 | #endif |
227 | } | |
228 | ||
ce286d32 | 229 | /* Device list insertion */ |
53759be9 | 230 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 231 | { |
c346dca1 | 232 | struct net *net = dev_net(dev); |
ce286d32 EB |
233 | |
234 | ASSERT_RTNL(); | |
235 | ||
236 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 237 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 238 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
239 | hlist_add_head_rcu(&dev->index_hlist, |
240 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 241 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
242 | |
243 | dev_base_seq_inc(net); | |
ce286d32 EB |
244 | } |
245 | ||
fb699dfd ED |
246 | /* Device list removal |
247 | * caller must respect a RCU grace period before freeing/reusing dev | |
248 | */ | |
ce286d32 EB |
249 | static void unlist_netdevice(struct net_device *dev) |
250 | { | |
251 | ASSERT_RTNL(); | |
252 | ||
253 | /* Unlink dev from the device chain */ | |
254 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 255 | list_del_rcu(&dev->dev_list); |
72c9528b | 256 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 257 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 258 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
259 | |
260 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
261 | } |
262 | ||
1da177e4 LT |
263 | /* |
264 | * Our notifier list | |
265 | */ | |
266 | ||
f07d5b94 | 267 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
268 | |
269 | /* | |
270 | * Device drivers call our routines to queue packets here. We empty the | |
271 | * queue in the local softnet handler. | |
272 | */ | |
bea3348e | 273 | |
9958da05 | 274 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 275 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 276 | |
cf508b12 | 277 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 278 | /* |
c773e847 | 279 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
280 | * according to dev->type |
281 | */ | |
643aa9cb | 282 | static const unsigned short netdev_lock_type[] = { |
283 | ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
723e98b7 JP |
284 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, |
285 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
286 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
287 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
288 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
289 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
290 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
291 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
292 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
293 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
294 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
295 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
296 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
297 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 298 | |
643aa9cb | 299 | static const char *const netdev_lock_name[] = { |
300 | "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
301 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
302 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
303 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
304 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
305 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
306 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
307 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
308 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
309 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
310 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
311 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
312 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", | |
313 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
314 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
315 | |
316 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 317 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
318 | |
319 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
320 | { | |
321 | int i; | |
322 | ||
323 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
324 | if (netdev_lock_type[i] == dev_type) | |
325 | return i; | |
326 | /* the last key is used by default */ | |
327 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
328 | } | |
329 | ||
cf508b12 DM |
330 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
331 | unsigned short dev_type) | |
723e98b7 JP |
332 | { |
333 | int i; | |
334 | ||
335 | i = netdev_lock_pos(dev_type); | |
336 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
337 | netdev_lock_name[i]); | |
338 | } | |
cf508b12 DM |
339 | |
340 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
341 | { | |
342 | int i; | |
343 | ||
344 | i = netdev_lock_pos(dev->type); | |
345 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
346 | &netdev_addr_lock_key[i], | |
347 | netdev_lock_name[i]); | |
348 | } | |
723e98b7 | 349 | #else |
cf508b12 DM |
350 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
351 | unsigned short dev_type) | |
352 | { | |
353 | } | |
354 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
355 | { |
356 | } | |
357 | #endif | |
1da177e4 LT |
358 | |
359 | /******************************************************************************* | |
eb13da1a | 360 | * |
361 | * Protocol management and registration routines | |
362 | * | |
363 | *******************************************************************************/ | |
1da177e4 | 364 | |
1da177e4 | 365 | |
1da177e4 LT |
366 | /* |
367 | * Add a protocol ID to the list. Now that the input handler is | |
368 | * smarter we can dispense with all the messy stuff that used to be | |
369 | * here. | |
370 | * | |
371 | * BEWARE!!! Protocol handlers, mangling input packets, | |
372 | * MUST BE last in hash buckets and checking protocol handlers | |
373 | * MUST start from promiscuous ptype_all chain in net_bh. | |
374 | * It is true now, do not change it. | |
375 | * Explanation follows: if protocol handler, mangling packet, will | |
376 | * be the first on list, it is not able to sense, that packet | |
377 | * is cloned and should be copied-on-write, so that it will | |
378 | * change it and subsequent readers will get broken packet. | |
379 | * --ANK (980803) | |
380 | */ | |
381 | ||
c07b68e8 ED |
382 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
383 | { | |
384 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 385 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 386 | else |
7866a621 SN |
387 | return pt->dev ? &pt->dev->ptype_specific : |
388 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
389 | } |
390 | ||
1da177e4 LT |
391 | /** |
392 | * dev_add_pack - add packet handler | |
393 | * @pt: packet type declaration | |
394 | * | |
395 | * Add a protocol handler to the networking stack. The passed &packet_type | |
396 | * is linked into kernel lists and may not be freed until it has been | |
397 | * removed from the kernel lists. | |
398 | * | |
4ec93edb | 399 | * This call does not sleep therefore it can not |
1da177e4 LT |
400 | * guarantee all CPU's that are in middle of receiving packets |
401 | * will see the new packet type (until the next received packet). | |
402 | */ | |
403 | ||
404 | void dev_add_pack(struct packet_type *pt) | |
405 | { | |
c07b68e8 | 406 | struct list_head *head = ptype_head(pt); |
1da177e4 | 407 | |
c07b68e8 ED |
408 | spin_lock(&ptype_lock); |
409 | list_add_rcu(&pt->list, head); | |
410 | spin_unlock(&ptype_lock); | |
1da177e4 | 411 | } |
d1b19dff | 412 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 413 | |
1da177e4 LT |
414 | /** |
415 | * __dev_remove_pack - remove packet handler | |
416 | * @pt: packet type declaration | |
417 | * | |
418 | * Remove a protocol handler that was previously added to the kernel | |
419 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
420 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 421 | * returns. |
1da177e4 LT |
422 | * |
423 | * The packet type might still be in use by receivers | |
424 | * and must not be freed until after all the CPU's have gone | |
425 | * through a quiescent state. | |
426 | */ | |
427 | void __dev_remove_pack(struct packet_type *pt) | |
428 | { | |
c07b68e8 | 429 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
430 | struct packet_type *pt1; |
431 | ||
c07b68e8 | 432 | spin_lock(&ptype_lock); |
1da177e4 LT |
433 | |
434 | list_for_each_entry(pt1, head, list) { | |
435 | if (pt == pt1) { | |
436 | list_del_rcu(&pt->list); | |
437 | goto out; | |
438 | } | |
439 | } | |
440 | ||
7b6cd1ce | 441 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 442 | out: |
c07b68e8 | 443 | spin_unlock(&ptype_lock); |
1da177e4 | 444 | } |
d1b19dff ED |
445 | EXPORT_SYMBOL(__dev_remove_pack); |
446 | ||
1da177e4 LT |
447 | /** |
448 | * dev_remove_pack - remove packet handler | |
449 | * @pt: packet type declaration | |
450 | * | |
451 | * Remove a protocol handler that was previously added to the kernel | |
452 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
453 | * from the kernel lists and can be freed or reused once this function | |
454 | * returns. | |
455 | * | |
456 | * This call sleeps to guarantee that no CPU is looking at the packet | |
457 | * type after return. | |
458 | */ | |
459 | void dev_remove_pack(struct packet_type *pt) | |
460 | { | |
461 | __dev_remove_pack(pt); | |
4ec93edb | 462 | |
1da177e4 LT |
463 | synchronize_net(); |
464 | } | |
d1b19dff | 465 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 466 | |
62532da9 VY |
467 | |
468 | /** | |
469 | * dev_add_offload - register offload handlers | |
470 | * @po: protocol offload declaration | |
471 | * | |
472 | * Add protocol offload handlers to the networking stack. The passed | |
473 | * &proto_offload is linked into kernel lists and may not be freed until | |
474 | * it has been removed from the kernel lists. | |
475 | * | |
476 | * This call does not sleep therefore it can not | |
477 | * guarantee all CPU's that are in middle of receiving packets | |
478 | * will see the new offload handlers (until the next received packet). | |
479 | */ | |
480 | void dev_add_offload(struct packet_offload *po) | |
481 | { | |
bdef7de4 | 482 | struct packet_offload *elem; |
62532da9 VY |
483 | |
484 | spin_lock(&offload_lock); | |
bdef7de4 DM |
485 | list_for_each_entry(elem, &offload_base, list) { |
486 | if (po->priority < elem->priority) | |
487 | break; | |
488 | } | |
489 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
490 | spin_unlock(&offload_lock); |
491 | } | |
492 | EXPORT_SYMBOL(dev_add_offload); | |
493 | ||
494 | /** | |
495 | * __dev_remove_offload - remove offload handler | |
496 | * @po: packet offload declaration | |
497 | * | |
498 | * Remove a protocol offload handler that was previously added to the | |
499 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
500 | * is removed from the kernel lists and can be freed or reused once this | |
501 | * function returns. | |
502 | * | |
503 | * The packet type might still be in use by receivers | |
504 | * and must not be freed until after all the CPU's have gone | |
505 | * through a quiescent state. | |
506 | */ | |
1d143d9f | 507 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
508 | { |
509 | struct list_head *head = &offload_base; | |
510 | struct packet_offload *po1; | |
511 | ||
c53aa505 | 512 | spin_lock(&offload_lock); |
62532da9 VY |
513 | |
514 | list_for_each_entry(po1, head, list) { | |
515 | if (po == po1) { | |
516 | list_del_rcu(&po->list); | |
517 | goto out; | |
518 | } | |
519 | } | |
520 | ||
521 | pr_warn("dev_remove_offload: %p not found\n", po); | |
522 | out: | |
c53aa505 | 523 | spin_unlock(&offload_lock); |
62532da9 | 524 | } |
62532da9 VY |
525 | |
526 | /** | |
527 | * dev_remove_offload - remove packet offload handler | |
528 | * @po: packet offload declaration | |
529 | * | |
530 | * Remove a packet offload handler that was previously added to the kernel | |
531 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
532 | * removed from the kernel lists and can be freed or reused once this | |
533 | * function returns. | |
534 | * | |
535 | * This call sleeps to guarantee that no CPU is looking at the packet | |
536 | * type after return. | |
537 | */ | |
538 | void dev_remove_offload(struct packet_offload *po) | |
539 | { | |
540 | __dev_remove_offload(po); | |
541 | ||
542 | synchronize_net(); | |
543 | } | |
544 | EXPORT_SYMBOL(dev_remove_offload); | |
545 | ||
1da177e4 | 546 | /****************************************************************************** |
eb13da1a | 547 | * |
548 | * Device Boot-time Settings Routines | |
549 | * | |
550 | ******************************************************************************/ | |
1da177e4 LT |
551 | |
552 | /* Boot time configuration table */ | |
553 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
554 | ||
555 | /** | |
556 | * netdev_boot_setup_add - add new setup entry | |
557 | * @name: name of the device | |
558 | * @map: configured settings for the device | |
559 | * | |
560 | * Adds new setup entry to the dev_boot_setup list. The function | |
561 | * returns 0 on error and 1 on success. This is a generic routine to | |
562 | * all netdevices. | |
563 | */ | |
564 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
565 | { | |
566 | struct netdev_boot_setup *s; | |
567 | int i; | |
568 | ||
569 | s = dev_boot_setup; | |
570 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
571 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
572 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 573 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
574 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
575 | break; | |
576 | } | |
577 | } | |
578 | ||
579 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
580 | } | |
581 | ||
582 | /** | |
722c9a0c | 583 | * netdev_boot_setup_check - check boot time settings |
584 | * @dev: the netdevice | |
1da177e4 | 585 | * |
722c9a0c | 586 | * Check boot time settings for the device. |
587 | * The found settings are set for the device to be used | |
588 | * later in the device probing. | |
589 | * Returns 0 if no settings found, 1 if they are. | |
1da177e4 LT |
590 | */ |
591 | int netdev_boot_setup_check(struct net_device *dev) | |
592 | { | |
593 | struct netdev_boot_setup *s = dev_boot_setup; | |
594 | int i; | |
595 | ||
596 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
597 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 598 | !strcmp(dev->name, s[i].name)) { |
722c9a0c | 599 | dev->irq = s[i].map.irq; |
600 | dev->base_addr = s[i].map.base_addr; | |
601 | dev->mem_start = s[i].map.mem_start; | |
602 | dev->mem_end = s[i].map.mem_end; | |
1da177e4 LT |
603 | return 1; |
604 | } | |
605 | } | |
606 | return 0; | |
607 | } | |
d1b19dff | 608 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
609 | |
610 | ||
611 | /** | |
722c9a0c | 612 | * netdev_boot_base - get address from boot time settings |
613 | * @prefix: prefix for network device | |
614 | * @unit: id for network device | |
615 | * | |
616 | * Check boot time settings for the base address of device. | |
617 | * The found settings are set for the device to be used | |
618 | * later in the device probing. | |
619 | * Returns 0 if no settings found. | |
1da177e4 LT |
620 | */ |
621 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
622 | { | |
623 | const struct netdev_boot_setup *s = dev_boot_setup; | |
624 | char name[IFNAMSIZ]; | |
625 | int i; | |
626 | ||
627 | sprintf(name, "%s%d", prefix, unit); | |
628 | ||
629 | /* | |
630 | * If device already registered then return base of 1 | |
631 | * to indicate not to probe for this interface | |
632 | */ | |
881d966b | 633 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
634 | return 1; |
635 | ||
636 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
637 | if (!strcmp(name, s[i].name)) | |
638 | return s[i].map.base_addr; | |
639 | return 0; | |
640 | } | |
641 | ||
642 | /* | |
643 | * Saves at boot time configured settings for any netdevice. | |
644 | */ | |
645 | int __init netdev_boot_setup(char *str) | |
646 | { | |
647 | int ints[5]; | |
648 | struct ifmap map; | |
649 | ||
650 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
651 | if (!str || !*str) | |
652 | return 0; | |
653 | ||
654 | /* Save settings */ | |
655 | memset(&map, 0, sizeof(map)); | |
656 | if (ints[0] > 0) | |
657 | map.irq = ints[1]; | |
658 | if (ints[0] > 1) | |
659 | map.base_addr = ints[2]; | |
660 | if (ints[0] > 2) | |
661 | map.mem_start = ints[3]; | |
662 | if (ints[0] > 3) | |
663 | map.mem_end = ints[4]; | |
664 | ||
665 | /* Add new entry to the list */ | |
666 | return netdev_boot_setup_add(str, &map); | |
667 | } | |
668 | ||
669 | __setup("netdev=", netdev_boot_setup); | |
670 | ||
671 | /******************************************************************************* | |
eb13da1a | 672 | * |
673 | * Device Interface Subroutines | |
674 | * | |
675 | *******************************************************************************/ | |
1da177e4 | 676 | |
a54acb3a ND |
677 | /** |
678 | * dev_get_iflink - get 'iflink' value of a interface | |
679 | * @dev: targeted interface | |
680 | * | |
681 | * Indicates the ifindex the interface is linked to. | |
682 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
683 | */ | |
684 | ||
685 | int dev_get_iflink(const struct net_device *dev) | |
686 | { | |
687 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
688 | return dev->netdev_ops->ndo_get_iflink(dev); | |
689 | ||
7a66bbc9 | 690 | return dev->ifindex; |
a54acb3a ND |
691 | } |
692 | EXPORT_SYMBOL(dev_get_iflink); | |
693 | ||
fc4099f1 PS |
694 | /** |
695 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
696 | * @dev: targeted interface | |
697 | * @skb: The packet. | |
698 | * | |
699 | * For better visibility of tunnel traffic OVS needs to retrieve | |
700 | * egress tunnel information for a packet. Following API allows | |
701 | * user to get this info. | |
702 | */ | |
703 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
704 | { | |
705 | struct ip_tunnel_info *info; | |
706 | ||
707 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
708 | return -EINVAL; | |
709 | ||
710 | info = skb_tunnel_info_unclone(skb); | |
711 | if (!info) | |
712 | return -ENOMEM; | |
713 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
714 | return -EINVAL; | |
715 | ||
716 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
717 | } | |
718 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
719 | ||
1da177e4 LT |
720 | /** |
721 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 722 | * @net: the applicable net namespace |
1da177e4 LT |
723 | * @name: name to find |
724 | * | |
725 | * Find an interface by name. Must be called under RTNL semaphore | |
726 | * or @dev_base_lock. If the name is found a pointer to the device | |
727 | * is returned. If the name is not found then %NULL is returned. The | |
728 | * reference counters are not incremented so the caller must be | |
729 | * careful with locks. | |
730 | */ | |
731 | ||
881d966b | 732 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 733 | { |
0bd8d536 ED |
734 | struct net_device *dev; |
735 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 736 | |
b67bfe0d | 737 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
738 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
739 | return dev; | |
0bd8d536 | 740 | |
1da177e4 LT |
741 | return NULL; |
742 | } | |
d1b19dff | 743 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 744 | |
72c9528b | 745 | /** |
722c9a0c | 746 | * dev_get_by_name_rcu - find a device by its name |
747 | * @net: the applicable net namespace | |
748 | * @name: name to find | |
749 | * | |
750 | * Find an interface by name. | |
751 | * If the name is found a pointer to the device is returned. | |
752 | * If the name is not found then %NULL is returned. | |
753 | * The reference counters are not incremented so the caller must be | |
754 | * careful with locks. The caller must hold RCU lock. | |
72c9528b ED |
755 | */ |
756 | ||
757 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
758 | { | |
72c9528b ED |
759 | struct net_device *dev; |
760 | struct hlist_head *head = dev_name_hash(net, name); | |
761 | ||
b67bfe0d | 762 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
763 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
764 | return dev; | |
765 | ||
766 | return NULL; | |
767 | } | |
768 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
769 | ||
1da177e4 LT |
770 | /** |
771 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 772 | * @net: the applicable net namespace |
1da177e4 LT |
773 | * @name: name to find |
774 | * | |
775 | * Find an interface by name. This can be called from any | |
776 | * context and does its own locking. The returned handle has | |
777 | * the usage count incremented and the caller must use dev_put() to | |
778 | * release it when it is no longer needed. %NULL is returned if no | |
779 | * matching device is found. | |
780 | */ | |
781 | ||
881d966b | 782 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
783 | { |
784 | struct net_device *dev; | |
785 | ||
72c9528b ED |
786 | rcu_read_lock(); |
787 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
788 | if (dev) |
789 | dev_hold(dev); | |
72c9528b | 790 | rcu_read_unlock(); |
1da177e4 LT |
791 | return dev; |
792 | } | |
d1b19dff | 793 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
794 | |
795 | /** | |
796 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 797 | * @net: the applicable net namespace |
1da177e4 LT |
798 | * @ifindex: index of device |
799 | * | |
800 | * Search for an interface by index. Returns %NULL if the device | |
801 | * is not found or a pointer to the device. The device has not | |
802 | * had its reference counter increased so the caller must be careful | |
803 | * about locking. The caller must hold either the RTNL semaphore | |
804 | * or @dev_base_lock. | |
805 | */ | |
806 | ||
881d966b | 807 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 808 | { |
0bd8d536 ED |
809 | struct net_device *dev; |
810 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 811 | |
b67bfe0d | 812 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
813 | if (dev->ifindex == ifindex) |
814 | return dev; | |
0bd8d536 | 815 | |
1da177e4 LT |
816 | return NULL; |
817 | } | |
d1b19dff | 818 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 819 | |
fb699dfd ED |
820 | /** |
821 | * dev_get_by_index_rcu - find a device by its ifindex | |
822 | * @net: the applicable net namespace | |
823 | * @ifindex: index of device | |
824 | * | |
825 | * Search for an interface by index. Returns %NULL if the device | |
826 | * is not found or a pointer to the device. The device has not | |
827 | * had its reference counter increased so the caller must be careful | |
828 | * about locking. The caller must hold RCU lock. | |
829 | */ | |
830 | ||
831 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
832 | { | |
fb699dfd ED |
833 | struct net_device *dev; |
834 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
835 | ||
b67bfe0d | 836 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
837 | if (dev->ifindex == ifindex) |
838 | return dev; | |
839 | ||
840 | return NULL; | |
841 | } | |
842 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
843 | ||
1da177e4 LT |
844 | |
845 | /** | |
846 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 847 | * @net: the applicable net namespace |
1da177e4 LT |
848 | * @ifindex: index of device |
849 | * | |
850 | * Search for an interface by index. Returns NULL if the device | |
851 | * is not found or a pointer to the device. The device returned has | |
852 | * had a reference added and the pointer is safe until the user calls | |
853 | * dev_put to indicate they have finished with it. | |
854 | */ | |
855 | ||
881d966b | 856 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
857 | { |
858 | struct net_device *dev; | |
859 | ||
fb699dfd ED |
860 | rcu_read_lock(); |
861 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
862 | if (dev) |
863 | dev_hold(dev); | |
fb699dfd | 864 | rcu_read_unlock(); |
1da177e4 LT |
865 | return dev; |
866 | } | |
d1b19dff | 867 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 868 | |
5dbe7c17 NS |
869 | /** |
870 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
871 | * @net: network namespace | |
872 | * @name: a pointer to the buffer where the name will be stored. | |
873 | * @ifindex: the ifindex of the interface to get the name from. | |
874 | * | |
875 | * The use of raw_seqcount_begin() and cond_resched() before | |
876 | * retrying is required as we want to give the writers a chance | |
877 | * to complete when CONFIG_PREEMPT is not set. | |
878 | */ | |
879 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
880 | { | |
881 | struct net_device *dev; | |
882 | unsigned int seq; | |
883 | ||
884 | retry: | |
885 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
886 | rcu_read_lock(); | |
887 | dev = dev_get_by_index_rcu(net, ifindex); | |
888 | if (!dev) { | |
889 | rcu_read_unlock(); | |
890 | return -ENODEV; | |
891 | } | |
892 | ||
893 | strcpy(name, dev->name); | |
894 | rcu_read_unlock(); | |
895 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
896 | cond_resched(); | |
897 | goto retry; | |
898 | } | |
899 | ||
900 | return 0; | |
901 | } | |
902 | ||
1da177e4 | 903 | /** |
941666c2 | 904 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 905 | * @net: the applicable net namespace |
1da177e4 LT |
906 | * @type: media type of device |
907 | * @ha: hardware address | |
908 | * | |
909 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
910 | * is not found or a pointer to the device. |
911 | * The caller must hold RCU or RTNL. | |
941666c2 | 912 | * The returned device has not had its ref count increased |
1da177e4 LT |
913 | * and the caller must therefore be careful about locking |
914 | * | |
1da177e4 LT |
915 | */ |
916 | ||
941666c2 ED |
917 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
918 | const char *ha) | |
1da177e4 LT |
919 | { |
920 | struct net_device *dev; | |
921 | ||
941666c2 | 922 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
923 | if (dev->type == type && |
924 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
925 | return dev; |
926 | ||
927 | return NULL; | |
1da177e4 | 928 | } |
941666c2 | 929 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 930 | |
881d966b | 931 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
932 | { |
933 | struct net_device *dev; | |
934 | ||
4e9cac2b | 935 | ASSERT_RTNL(); |
881d966b | 936 | for_each_netdev(net, dev) |
4e9cac2b | 937 | if (dev->type == type) |
7562f876 PE |
938 | return dev; |
939 | ||
940 | return NULL; | |
4e9cac2b | 941 | } |
4e9cac2b PM |
942 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
943 | ||
881d966b | 944 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 945 | { |
99fe3c39 | 946 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 947 | |
99fe3c39 ED |
948 | rcu_read_lock(); |
949 | for_each_netdev_rcu(net, dev) | |
950 | if (dev->type == type) { | |
951 | dev_hold(dev); | |
952 | ret = dev; | |
953 | break; | |
954 | } | |
955 | rcu_read_unlock(); | |
956 | return ret; | |
1da177e4 | 957 | } |
1da177e4 LT |
958 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
959 | ||
960 | /** | |
6c555490 | 961 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 962 | * @net: the applicable net namespace |
1da177e4 LT |
963 | * @if_flags: IFF_* values |
964 | * @mask: bitmask of bits in if_flags to check | |
965 | * | |
966 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 967 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 968 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
969 | */ |
970 | ||
6c555490 WC |
971 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
972 | unsigned short mask) | |
1da177e4 | 973 | { |
7562f876 | 974 | struct net_device *dev, *ret; |
1da177e4 | 975 | |
6c555490 WC |
976 | ASSERT_RTNL(); |
977 | ||
7562f876 | 978 | ret = NULL; |
6c555490 | 979 | for_each_netdev(net, dev) { |
1da177e4 | 980 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 981 | ret = dev; |
1da177e4 LT |
982 | break; |
983 | } | |
984 | } | |
7562f876 | 985 | return ret; |
1da177e4 | 986 | } |
6c555490 | 987 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
988 | |
989 | /** | |
990 | * dev_valid_name - check if name is okay for network device | |
991 | * @name: name string | |
992 | * | |
993 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
994 | * to allow sysfs to work. We also disallow any kind of |
995 | * whitespace. | |
1da177e4 | 996 | */ |
95f050bf | 997 | bool dev_valid_name(const char *name) |
1da177e4 | 998 | { |
c7fa9d18 | 999 | if (*name == '\0') |
95f050bf | 1000 | return false; |
b6fe17d6 | 1001 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 1002 | return false; |
c7fa9d18 | 1003 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1004 | return false; |
c7fa9d18 DM |
1005 | |
1006 | while (*name) { | |
a4176a93 | 1007 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1008 | return false; |
c7fa9d18 DM |
1009 | name++; |
1010 | } | |
95f050bf | 1011 | return true; |
1da177e4 | 1012 | } |
d1b19dff | 1013 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1014 | |
1015 | /** | |
b267b179 EB |
1016 | * __dev_alloc_name - allocate a name for a device |
1017 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1018 | * @name: name format string |
b267b179 | 1019 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1020 | * |
1021 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1022 | * id. It scans list of devices to build up a free map, then chooses |
1023 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1024 | * while allocating the name and adding the device in order to avoid | |
1025 | * duplicates. | |
1026 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1027 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1028 | */ |
1029 | ||
b267b179 | 1030 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1031 | { |
1032 | int i = 0; | |
1da177e4 LT |
1033 | const char *p; |
1034 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1035 | unsigned long *inuse; |
1da177e4 LT |
1036 | struct net_device *d; |
1037 | ||
1038 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
1039 | if (p) { | |
1040 | /* | |
1041 | * Verify the string as this thing may have come from | |
1042 | * the user. There must be either one "%d" and no other "%" | |
1043 | * characters. | |
1044 | */ | |
1045 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1046 | return -EINVAL; | |
1047 | ||
1048 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1049 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1050 | if (!inuse) |
1051 | return -ENOMEM; | |
1052 | ||
881d966b | 1053 | for_each_netdev(net, d) { |
1da177e4 LT |
1054 | if (!sscanf(d->name, name, &i)) |
1055 | continue; | |
1056 | if (i < 0 || i >= max_netdevices) | |
1057 | continue; | |
1058 | ||
1059 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1060 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1061 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1062 | set_bit(i, inuse); | |
1063 | } | |
1064 | ||
1065 | i = find_first_zero_bit(inuse, max_netdevices); | |
1066 | free_page((unsigned long) inuse); | |
1067 | } | |
1068 | ||
d9031024 OP |
1069 | if (buf != name) |
1070 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1071 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1072 | return i; |
1da177e4 LT |
1073 | |
1074 | /* It is possible to run out of possible slots | |
1075 | * when the name is long and there isn't enough space left | |
1076 | * for the digits, or if all bits are used. | |
1077 | */ | |
1078 | return -ENFILE; | |
1079 | } | |
1080 | ||
b267b179 EB |
1081 | /** |
1082 | * dev_alloc_name - allocate a name for a device | |
1083 | * @dev: device | |
1084 | * @name: name format string | |
1085 | * | |
1086 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1087 | * id. It scans list of devices to build up a free map, then chooses | |
1088 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1089 | * while allocating the name and adding the device in order to avoid | |
1090 | * duplicates. | |
1091 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1092 | * Returns the number of the unit assigned or a negative errno code. | |
1093 | */ | |
1094 | ||
1095 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1096 | { | |
1097 | char buf[IFNAMSIZ]; | |
1098 | struct net *net; | |
1099 | int ret; | |
1100 | ||
c346dca1 YH |
1101 | BUG_ON(!dev_net(dev)); |
1102 | net = dev_net(dev); | |
b267b179 EB |
1103 | ret = __dev_alloc_name(net, name, buf); |
1104 | if (ret >= 0) | |
1105 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1106 | return ret; | |
1107 | } | |
d1b19dff | 1108 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1109 | |
828de4f6 G |
1110 | static int dev_alloc_name_ns(struct net *net, |
1111 | struct net_device *dev, | |
1112 | const char *name) | |
d9031024 | 1113 | { |
828de4f6 G |
1114 | char buf[IFNAMSIZ]; |
1115 | int ret; | |
8ce6cebc | 1116 | |
828de4f6 G |
1117 | ret = __dev_alloc_name(net, name, buf); |
1118 | if (ret >= 0) | |
1119 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1120 | return ret; | |
1121 | } | |
1122 | ||
1123 | static int dev_get_valid_name(struct net *net, | |
1124 | struct net_device *dev, | |
1125 | const char *name) | |
1126 | { | |
1127 | BUG_ON(!net); | |
8ce6cebc | 1128 | |
d9031024 OP |
1129 | if (!dev_valid_name(name)) |
1130 | return -EINVAL; | |
1131 | ||
1c5cae81 | 1132 | if (strchr(name, '%')) |
828de4f6 | 1133 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1134 | else if (__dev_get_by_name(net, name)) |
1135 | return -EEXIST; | |
8ce6cebc DL |
1136 | else if (dev->name != name) |
1137 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1138 | |
1139 | return 0; | |
1140 | } | |
1da177e4 LT |
1141 | |
1142 | /** | |
1143 | * dev_change_name - change name of a device | |
1144 | * @dev: device | |
1145 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1146 | * | |
1147 | * Change name of a device, can pass format strings "eth%d". | |
1148 | * for wildcarding. | |
1149 | */ | |
cf04a4c7 | 1150 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1151 | { |
238fa362 | 1152 | unsigned char old_assign_type; |
fcc5a03a | 1153 | char oldname[IFNAMSIZ]; |
1da177e4 | 1154 | int err = 0; |
fcc5a03a | 1155 | int ret; |
881d966b | 1156 | struct net *net; |
1da177e4 LT |
1157 | |
1158 | ASSERT_RTNL(); | |
c346dca1 | 1159 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1160 | |
c346dca1 | 1161 | net = dev_net(dev); |
1da177e4 LT |
1162 | if (dev->flags & IFF_UP) |
1163 | return -EBUSY; | |
1164 | ||
30e6c9fa | 1165 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1166 | |
1167 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1168 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1169 | return 0; |
c91f6df2 | 1170 | } |
c8d90dca | 1171 | |
fcc5a03a HX |
1172 | memcpy(oldname, dev->name, IFNAMSIZ); |
1173 | ||
828de4f6 | 1174 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1175 | if (err < 0) { |
30e6c9fa | 1176 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1177 | return err; |
c91f6df2 | 1178 | } |
1da177e4 | 1179 | |
6fe82a39 VF |
1180 | if (oldname[0] && !strchr(oldname, '%')) |
1181 | netdev_info(dev, "renamed from %s\n", oldname); | |
1182 | ||
238fa362 TG |
1183 | old_assign_type = dev->name_assign_type; |
1184 | dev->name_assign_type = NET_NAME_RENAMED; | |
1185 | ||
fcc5a03a | 1186 | rollback: |
a1b3f594 EB |
1187 | ret = device_rename(&dev->dev, dev->name); |
1188 | if (ret) { | |
1189 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1190 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1191 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1192 | return ret; |
dcc99773 | 1193 | } |
7f988eab | 1194 | |
30e6c9fa | 1195 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1196 | |
5bb025fa VF |
1197 | netdev_adjacent_rename_links(dev, oldname); |
1198 | ||
7f988eab | 1199 | write_lock_bh(&dev_base_lock); |
372b2312 | 1200 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1201 | write_unlock_bh(&dev_base_lock); |
1202 | ||
1203 | synchronize_rcu(); | |
1204 | ||
1205 | write_lock_bh(&dev_base_lock); | |
1206 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1207 | write_unlock_bh(&dev_base_lock); |
1208 | ||
056925ab | 1209 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1210 | ret = notifier_to_errno(ret); |
1211 | ||
1212 | if (ret) { | |
91e9c07b ED |
1213 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1214 | if (err >= 0) { | |
fcc5a03a | 1215 | err = ret; |
30e6c9fa | 1216 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1217 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1218 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1219 | dev->name_assign_type = old_assign_type; |
1220 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1221 | goto rollback; |
91e9c07b | 1222 | } else { |
7b6cd1ce | 1223 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1224 | dev->name, ret); |
fcc5a03a HX |
1225 | } |
1226 | } | |
1da177e4 LT |
1227 | |
1228 | return err; | |
1229 | } | |
1230 | ||
0b815a1a SH |
1231 | /** |
1232 | * dev_set_alias - change ifalias of a device | |
1233 | * @dev: device | |
1234 | * @alias: name up to IFALIASZ | |
f0db275a | 1235 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1236 | * |
1237 | * Set ifalias for a device, | |
1238 | */ | |
1239 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1240 | { | |
7364e445 AK |
1241 | char *new_ifalias; |
1242 | ||
0b815a1a SH |
1243 | ASSERT_RTNL(); |
1244 | ||
1245 | if (len >= IFALIASZ) | |
1246 | return -EINVAL; | |
1247 | ||
96ca4a2c | 1248 | if (!len) { |
388dfc2d SK |
1249 | kfree(dev->ifalias); |
1250 | dev->ifalias = NULL; | |
96ca4a2c OH |
1251 | return 0; |
1252 | } | |
1253 | ||
7364e445 AK |
1254 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1255 | if (!new_ifalias) | |
0b815a1a | 1256 | return -ENOMEM; |
7364e445 | 1257 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1258 | |
1259 | strlcpy(dev->ifalias, alias, len+1); | |
1260 | return len; | |
1261 | } | |
1262 | ||
1263 | ||
d8a33ac4 | 1264 | /** |
3041a069 | 1265 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1266 | * @dev: device to cause notification |
1267 | * | |
1268 | * Called to indicate a device has changed features. | |
1269 | */ | |
1270 | void netdev_features_change(struct net_device *dev) | |
1271 | { | |
056925ab | 1272 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1273 | } |
1274 | EXPORT_SYMBOL(netdev_features_change); | |
1275 | ||
1da177e4 LT |
1276 | /** |
1277 | * netdev_state_change - device changes state | |
1278 | * @dev: device to cause notification | |
1279 | * | |
1280 | * Called to indicate a device has changed state. This function calls | |
1281 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1282 | * to the routing socket. | |
1283 | */ | |
1284 | void netdev_state_change(struct net_device *dev) | |
1285 | { | |
1286 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1287 | struct netdev_notifier_change_info change_info; |
1288 | ||
1289 | change_info.flags_changed = 0; | |
1290 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1291 | &change_info.info); | |
7f294054 | 1292 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1293 | } |
1294 | } | |
d1b19dff | 1295 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1296 | |
ee89bab1 | 1297 | /** |
722c9a0c | 1298 | * netdev_notify_peers - notify network peers about existence of @dev |
1299 | * @dev: network device | |
ee89bab1 AW |
1300 | * |
1301 | * Generate traffic such that interested network peers are aware of | |
1302 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1303 | * a device wants to inform the rest of the network about some sort of | |
1304 | * reconfiguration such as a failover event or virtual machine | |
1305 | * migration. | |
1306 | */ | |
1307 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1308 | { |
ee89bab1 AW |
1309 | rtnl_lock(); |
1310 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
37c343b4 | 1311 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); |
ee89bab1 | 1312 | rtnl_unlock(); |
c1da4ac7 | 1313 | } |
ee89bab1 | 1314 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1315 | |
bd380811 | 1316 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1317 | { |
d314774c | 1318 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1319 | int ret; |
1da177e4 | 1320 | |
e46b66bc BH |
1321 | ASSERT_RTNL(); |
1322 | ||
1da177e4 LT |
1323 | if (!netif_device_present(dev)) |
1324 | return -ENODEV; | |
1325 | ||
ca99ca14 NH |
1326 | /* Block netpoll from trying to do any rx path servicing. |
1327 | * If we don't do this there is a chance ndo_poll_controller | |
1328 | * or ndo_poll may be running while we open the device | |
1329 | */ | |
66b5552f | 1330 | netpoll_poll_disable(dev); |
ca99ca14 | 1331 | |
3b8bcfd5 JB |
1332 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1333 | ret = notifier_to_errno(ret); | |
1334 | if (ret) | |
1335 | return ret; | |
1336 | ||
1da177e4 | 1337 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1338 | |
d314774c SH |
1339 | if (ops->ndo_validate_addr) |
1340 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1341 | |
d314774c SH |
1342 | if (!ret && ops->ndo_open) |
1343 | ret = ops->ndo_open(dev); | |
1da177e4 | 1344 | |
66b5552f | 1345 | netpoll_poll_enable(dev); |
ca99ca14 | 1346 | |
bada339b JG |
1347 | if (ret) |
1348 | clear_bit(__LINK_STATE_START, &dev->state); | |
1349 | else { | |
1da177e4 | 1350 | dev->flags |= IFF_UP; |
4417da66 | 1351 | dev_set_rx_mode(dev); |
1da177e4 | 1352 | dev_activate(dev); |
7bf23575 | 1353 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1354 | } |
bada339b | 1355 | |
1da177e4 LT |
1356 | return ret; |
1357 | } | |
1358 | ||
1359 | /** | |
bd380811 PM |
1360 | * dev_open - prepare an interface for use. |
1361 | * @dev: device to open | |
1da177e4 | 1362 | * |
bd380811 PM |
1363 | * Takes a device from down to up state. The device's private open |
1364 | * function is invoked and then the multicast lists are loaded. Finally | |
1365 | * the device is moved into the up state and a %NETDEV_UP message is | |
1366 | * sent to the netdev notifier chain. | |
1367 | * | |
1368 | * Calling this function on an active interface is a nop. On a failure | |
1369 | * a negative errno code is returned. | |
1da177e4 | 1370 | */ |
bd380811 PM |
1371 | int dev_open(struct net_device *dev) |
1372 | { | |
1373 | int ret; | |
1374 | ||
bd380811 PM |
1375 | if (dev->flags & IFF_UP) |
1376 | return 0; | |
1377 | ||
bd380811 PM |
1378 | ret = __dev_open(dev); |
1379 | if (ret < 0) | |
1380 | return ret; | |
1381 | ||
7f294054 | 1382 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1383 | call_netdevice_notifiers(NETDEV_UP, dev); |
1384 | ||
1385 | return ret; | |
1386 | } | |
1387 | EXPORT_SYMBOL(dev_open); | |
1388 | ||
44345724 | 1389 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1390 | { |
44345724 | 1391 | struct net_device *dev; |
e46b66bc | 1392 | |
bd380811 | 1393 | ASSERT_RTNL(); |
9d5010db DM |
1394 | might_sleep(); |
1395 | ||
5cde2829 | 1396 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1397 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1398 | netpoll_poll_disable(dev); |
3f4df206 | 1399 | |
44345724 | 1400 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1401 | |
44345724 | 1402 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1403 | |
44345724 OP |
1404 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1405 | * can be even on different cpu. So just clear netif_running(). | |
1406 | * | |
1407 | * dev->stop() will invoke napi_disable() on all of it's | |
1408 | * napi_struct instances on this device. | |
1409 | */ | |
4e857c58 | 1410 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1411 | } |
1da177e4 | 1412 | |
44345724 | 1413 | dev_deactivate_many(head); |
d8b2a4d2 | 1414 | |
5cde2829 | 1415 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1416 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1417 | |
44345724 OP |
1418 | /* |
1419 | * Call the device specific close. This cannot fail. | |
1420 | * Only if device is UP | |
1421 | * | |
1422 | * We allow it to be called even after a DETACH hot-plug | |
1423 | * event. | |
1424 | */ | |
1425 | if (ops->ndo_stop) | |
1426 | ops->ndo_stop(dev); | |
1427 | ||
44345724 | 1428 | dev->flags &= ~IFF_UP; |
66b5552f | 1429 | netpoll_poll_enable(dev); |
44345724 OP |
1430 | } |
1431 | ||
1432 | return 0; | |
1433 | } | |
1434 | ||
1435 | static int __dev_close(struct net_device *dev) | |
1436 | { | |
f87e6f47 | 1437 | int retval; |
44345724 OP |
1438 | LIST_HEAD(single); |
1439 | ||
5cde2829 | 1440 | list_add(&dev->close_list, &single); |
f87e6f47 LT |
1441 | retval = __dev_close_many(&single); |
1442 | list_del(&single); | |
ca99ca14 | 1443 | |
f87e6f47 | 1444 | return retval; |
44345724 OP |
1445 | } |
1446 | ||
99c4a26a | 1447 | int dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1448 | { |
1449 | struct net_device *dev, *tmp; | |
1da177e4 | 1450 | |
5cde2829 EB |
1451 | /* Remove the devices that don't need to be closed */ |
1452 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1453 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1454 | list_del_init(&dev->close_list); |
44345724 OP |
1455 | |
1456 | __dev_close_many(head); | |
1da177e4 | 1457 | |
5cde2829 | 1458 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1459 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1460 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1461 | if (unlink) |
1462 | list_del_init(&dev->close_list); | |
44345724 | 1463 | } |
bd380811 PM |
1464 | |
1465 | return 0; | |
1466 | } | |
99c4a26a | 1467 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1468 | |
1469 | /** | |
1470 | * dev_close - shutdown an interface. | |
1471 | * @dev: device to shutdown | |
1472 | * | |
1473 | * This function moves an active device into down state. A | |
1474 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1475 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1476 | * chain. | |
1477 | */ | |
1478 | int dev_close(struct net_device *dev) | |
1479 | { | |
e14a5993 ED |
1480 | if (dev->flags & IFF_UP) { |
1481 | LIST_HEAD(single); | |
1da177e4 | 1482 | |
5cde2829 | 1483 | list_add(&dev->close_list, &single); |
99c4a26a | 1484 | dev_close_many(&single, true); |
e14a5993 ED |
1485 | list_del(&single); |
1486 | } | |
da6e378b | 1487 | return 0; |
1da177e4 | 1488 | } |
d1b19dff | 1489 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1490 | |
1491 | ||
0187bdfb BH |
1492 | /** |
1493 | * dev_disable_lro - disable Large Receive Offload on a device | |
1494 | * @dev: device | |
1495 | * | |
1496 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1497 | * called under RTNL. This is needed if received packets may be | |
1498 | * forwarded to another interface. | |
1499 | */ | |
1500 | void dev_disable_lro(struct net_device *dev) | |
1501 | { | |
fbe168ba MK |
1502 | struct net_device *lower_dev; |
1503 | struct list_head *iter; | |
529d0489 | 1504 | |
bc5787c6 MM |
1505 | dev->wanted_features &= ~NETIF_F_LRO; |
1506 | netdev_update_features(dev); | |
27660515 | 1507 | |
22d5969f MM |
1508 | if (unlikely(dev->features & NETIF_F_LRO)) |
1509 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1510 | |
1511 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1512 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1513 | } |
1514 | EXPORT_SYMBOL(dev_disable_lro); | |
1515 | ||
351638e7 JP |
1516 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1517 | struct net_device *dev) | |
1518 | { | |
1519 | struct netdev_notifier_info info; | |
1520 | ||
1521 | netdev_notifier_info_init(&info, dev); | |
1522 | return nb->notifier_call(nb, val, &info); | |
1523 | } | |
0187bdfb | 1524 | |
881d966b EB |
1525 | static int dev_boot_phase = 1; |
1526 | ||
1da177e4 | 1527 | /** |
722c9a0c | 1528 | * register_netdevice_notifier - register a network notifier block |
1529 | * @nb: notifier | |
1da177e4 | 1530 | * |
722c9a0c | 1531 | * Register a notifier to be called when network device events occur. |
1532 | * The notifier passed is linked into the kernel structures and must | |
1533 | * not be reused until it has been unregistered. A negative errno code | |
1534 | * is returned on a failure. | |
1da177e4 | 1535 | * |
722c9a0c | 1536 | * When registered all registration and up events are replayed |
1537 | * to the new notifier to allow device to have a race free | |
1538 | * view of the network device list. | |
1da177e4 LT |
1539 | */ |
1540 | ||
1541 | int register_netdevice_notifier(struct notifier_block *nb) | |
1542 | { | |
1543 | struct net_device *dev; | |
fcc5a03a | 1544 | struct net_device *last; |
881d966b | 1545 | struct net *net; |
1da177e4 LT |
1546 | int err; |
1547 | ||
1548 | rtnl_lock(); | |
f07d5b94 | 1549 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1550 | if (err) |
1551 | goto unlock; | |
881d966b EB |
1552 | if (dev_boot_phase) |
1553 | goto unlock; | |
1554 | for_each_net(net) { | |
1555 | for_each_netdev(net, dev) { | |
351638e7 | 1556 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1557 | err = notifier_to_errno(err); |
1558 | if (err) | |
1559 | goto rollback; | |
1560 | ||
1561 | if (!(dev->flags & IFF_UP)) | |
1562 | continue; | |
1da177e4 | 1563 | |
351638e7 | 1564 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1565 | } |
1da177e4 | 1566 | } |
fcc5a03a HX |
1567 | |
1568 | unlock: | |
1da177e4 LT |
1569 | rtnl_unlock(); |
1570 | return err; | |
fcc5a03a HX |
1571 | |
1572 | rollback: | |
1573 | last = dev; | |
881d966b EB |
1574 | for_each_net(net) { |
1575 | for_each_netdev(net, dev) { | |
1576 | if (dev == last) | |
8f891489 | 1577 | goto outroll; |
fcc5a03a | 1578 | |
881d966b | 1579 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1580 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1581 | dev); | |
1582 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1583 | } |
351638e7 | 1584 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1585 | } |
fcc5a03a | 1586 | } |
c67625a1 | 1587 | |
8f891489 | 1588 | outroll: |
c67625a1 | 1589 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1590 | goto unlock; |
1da177e4 | 1591 | } |
d1b19dff | 1592 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1593 | |
1594 | /** | |
722c9a0c | 1595 | * unregister_netdevice_notifier - unregister a network notifier block |
1596 | * @nb: notifier | |
1da177e4 | 1597 | * |
722c9a0c | 1598 | * Unregister a notifier previously registered by |
1599 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1600 | * kernel structures and may then be reused. A negative errno code | |
1601 | * is returned on a failure. | |
7d3d43da | 1602 | * |
722c9a0c | 1603 | * After unregistering unregister and down device events are synthesized |
1604 | * for all devices on the device list to the removed notifier to remove | |
1605 | * the need for special case cleanup code. | |
1da177e4 LT |
1606 | */ |
1607 | ||
1608 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1609 | { | |
7d3d43da EB |
1610 | struct net_device *dev; |
1611 | struct net *net; | |
9f514950 HX |
1612 | int err; |
1613 | ||
1614 | rtnl_lock(); | |
f07d5b94 | 1615 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1616 | if (err) |
1617 | goto unlock; | |
1618 | ||
1619 | for_each_net(net) { | |
1620 | for_each_netdev(net, dev) { | |
1621 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1622 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1623 | dev); | |
1624 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1625 | } |
351638e7 | 1626 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1627 | } |
1628 | } | |
1629 | unlock: | |
9f514950 HX |
1630 | rtnl_unlock(); |
1631 | return err; | |
1da177e4 | 1632 | } |
d1b19dff | 1633 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1634 | |
351638e7 JP |
1635 | /** |
1636 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1637 | * @val: value passed unmodified to notifier function | |
1638 | * @dev: net_device pointer passed unmodified to notifier function | |
1639 | * @info: notifier information data | |
1640 | * | |
1641 | * Call all network notifier blocks. Parameters and return value | |
1642 | * are as for raw_notifier_call_chain(). | |
1643 | */ | |
1644 | ||
1d143d9f | 1645 | static int call_netdevice_notifiers_info(unsigned long val, |
1646 | struct net_device *dev, | |
1647 | struct netdev_notifier_info *info) | |
351638e7 JP |
1648 | { |
1649 | ASSERT_RTNL(); | |
1650 | netdev_notifier_info_init(info, dev); | |
1651 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1652 | } | |
351638e7 | 1653 | |
1da177e4 LT |
1654 | /** |
1655 | * call_netdevice_notifiers - call all network notifier blocks | |
1656 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1657 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1658 | * |
1659 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1660 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1661 | */ |
1662 | ||
ad7379d4 | 1663 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1664 | { |
351638e7 JP |
1665 | struct netdev_notifier_info info; |
1666 | ||
1667 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1668 | } |
edf947f1 | 1669 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1670 | |
1cf51900 | 1671 | #ifdef CONFIG_NET_INGRESS |
4577139b DB |
1672 | static struct static_key ingress_needed __read_mostly; |
1673 | ||
1674 | void net_inc_ingress_queue(void) | |
1675 | { | |
1676 | static_key_slow_inc(&ingress_needed); | |
1677 | } | |
1678 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
1679 | ||
1680 | void net_dec_ingress_queue(void) | |
1681 | { | |
1682 | static_key_slow_dec(&ingress_needed); | |
1683 | } | |
1684 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
1685 | #endif | |
1686 | ||
1f211a1b DB |
1687 | #ifdef CONFIG_NET_EGRESS |
1688 | static struct static_key egress_needed __read_mostly; | |
1689 | ||
1690 | void net_inc_egress_queue(void) | |
1691 | { | |
1692 | static_key_slow_inc(&egress_needed); | |
1693 | } | |
1694 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
1695 | ||
1696 | void net_dec_egress_queue(void) | |
1697 | { | |
1698 | static_key_slow_dec(&egress_needed); | |
1699 | } | |
1700 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
1701 | #endif | |
1702 | ||
c5905afb | 1703 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1704 | #ifdef HAVE_JUMP_LABEL |
b90e5794 | 1705 | static atomic_t netstamp_needed_deferred; |
13baa00a | 1706 | static atomic_t netstamp_wanted; |
5fa8bbda | 1707 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 1708 | { |
b90e5794 | 1709 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 1710 | int wanted; |
b90e5794 | 1711 | |
13baa00a ED |
1712 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
1713 | if (wanted > 0) | |
1714 | static_key_enable(&netstamp_needed); | |
1715 | else | |
1716 | static_key_disable(&netstamp_needed); | |
5fa8bbda ED |
1717 | } |
1718 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 1719 | #endif |
5fa8bbda ED |
1720 | |
1721 | void net_enable_timestamp(void) | |
1722 | { | |
13baa00a ED |
1723 | #ifdef HAVE_JUMP_LABEL |
1724 | int wanted; | |
1725 | ||
1726 | while (1) { | |
1727 | wanted = atomic_read(&netstamp_wanted); | |
1728 | if (wanted <= 0) | |
1729 | break; | |
1730 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) | |
1731 | return; | |
1732 | } | |
1733 | atomic_inc(&netstamp_needed_deferred); | |
1734 | schedule_work(&netstamp_work); | |
1735 | #else | |
c5905afb | 1736 | static_key_slow_inc(&netstamp_needed); |
13baa00a | 1737 | #endif |
1da177e4 | 1738 | } |
d1b19dff | 1739 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1740 | |
1741 | void net_disable_timestamp(void) | |
1742 | { | |
b90e5794 | 1743 | #ifdef HAVE_JUMP_LABEL |
13baa00a ED |
1744 | int wanted; |
1745 | ||
1746 | while (1) { | |
1747 | wanted = atomic_read(&netstamp_wanted); | |
1748 | if (wanted <= 1) | |
1749 | break; | |
1750 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) | |
1751 | return; | |
1752 | } | |
1753 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
1754 | schedule_work(&netstamp_work); |
1755 | #else | |
c5905afb | 1756 | static_key_slow_dec(&netstamp_needed); |
5fa8bbda | 1757 | #endif |
1da177e4 | 1758 | } |
d1b19dff | 1759 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1760 | |
3b098e2d | 1761 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1762 | { |
2456e855 | 1763 | skb->tstamp = 0; |
c5905afb | 1764 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1765 | __net_timestamp(skb); |
1da177e4 LT |
1766 | } |
1767 | ||
588f0330 | 1768 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1769 | if (static_key_false(&netstamp_needed)) { \ |
2456e855 | 1770 | if ((COND) && !(SKB)->tstamp) \ |
588f0330 ED |
1771 | __net_timestamp(SKB); \ |
1772 | } \ | |
3b098e2d | 1773 | |
f4b05d27 | 1774 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 DL |
1775 | { |
1776 | unsigned int len; | |
1777 | ||
1778 | if (!(dev->flags & IFF_UP)) | |
1779 | return false; | |
1780 | ||
1781 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1782 | if (skb->len <= len) | |
1783 | return true; | |
1784 | ||
1785 | /* if TSO is enabled, we don't care about the length as the packet | |
1786 | * could be forwarded without being segmented before | |
1787 | */ | |
1788 | if (skb_is_gso(skb)) | |
1789 | return true; | |
1790 | ||
1791 | return false; | |
1792 | } | |
1ee481fb | 1793 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1794 | |
a0265d28 HX |
1795 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1796 | { | |
4e3264d2 | 1797 | int ret = ____dev_forward_skb(dev, skb); |
a0265d28 | 1798 | |
4e3264d2 MKL |
1799 | if (likely(!ret)) { |
1800 | skb->protocol = eth_type_trans(skb, dev); | |
1801 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
1802 | } | |
a0265d28 | 1803 | |
4e3264d2 | 1804 | return ret; |
a0265d28 HX |
1805 | } |
1806 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1807 | ||
44540960 AB |
1808 | /** |
1809 | * dev_forward_skb - loopback an skb to another netif | |
1810 | * | |
1811 | * @dev: destination network device | |
1812 | * @skb: buffer to forward | |
1813 | * | |
1814 | * return values: | |
1815 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1816 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1817 | * |
1818 | * dev_forward_skb can be used for injecting an skb from the | |
1819 | * start_xmit function of one device into the receive queue | |
1820 | * of another device. | |
1821 | * | |
1822 | * The receiving device may be in another namespace, so | |
1823 | * we have to clear all information in the skb that could | |
1824 | * impact namespace isolation. | |
1825 | */ | |
1826 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1827 | { | |
a0265d28 | 1828 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1829 | } |
1830 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1831 | ||
71d9dec2 CG |
1832 | static inline int deliver_skb(struct sk_buff *skb, |
1833 | struct packet_type *pt_prev, | |
1834 | struct net_device *orig_dev) | |
1835 | { | |
1080e512 MT |
1836 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1837 | return -ENOMEM; | |
71d9dec2 CG |
1838 | atomic_inc(&skb->users); |
1839 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1840 | } | |
1841 | ||
7866a621 SN |
1842 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
1843 | struct packet_type **pt, | |
fbcb2170 JP |
1844 | struct net_device *orig_dev, |
1845 | __be16 type, | |
7866a621 SN |
1846 | struct list_head *ptype_list) |
1847 | { | |
1848 | struct packet_type *ptype, *pt_prev = *pt; | |
1849 | ||
1850 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1851 | if (ptype->type != type) | |
1852 | continue; | |
1853 | if (pt_prev) | |
fbcb2170 | 1854 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
1855 | pt_prev = ptype; |
1856 | } | |
1857 | *pt = pt_prev; | |
1858 | } | |
1859 | ||
c0de08d0 EL |
1860 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1861 | { | |
a3d744e9 | 1862 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1863 | return false; |
1864 | ||
1865 | if (ptype->id_match) | |
1866 | return ptype->id_match(ptype, skb->sk); | |
1867 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1868 | return true; | |
1869 | ||
1870 | return false; | |
1871 | } | |
1872 | ||
1da177e4 LT |
1873 | /* |
1874 | * Support routine. Sends outgoing frames to any network | |
1875 | * taps currently in use. | |
1876 | */ | |
1877 | ||
74b20582 | 1878 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1879 | { |
1880 | struct packet_type *ptype; | |
71d9dec2 CG |
1881 | struct sk_buff *skb2 = NULL; |
1882 | struct packet_type *pt_prev = NULL; | |
7866a621 | 1883 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 1884 | |
1da177e4 | 1885 | rcu_read_lock(); |
7866a621 SN |
1886 | again: |
1887 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1da177e4 LT |
1888 | /* Never send packets back to the socket |
1889 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1890 | */ | |
7866a621 SN |
1891 | if (skb_loop_sk(ptype, skb)) |
1892 | continue; | |
71d9dec2 | 1893 | |
7866a621 SN |
1894 | if (pt_prev) { |
1895 | deliver_skb(skb2, pt_prev, skb->dev); | |
1896 | pt_prev = ptype; | |
1897 | continue; | |
1898 | } | |
1da177e4 | 1899 | |
7866a621 SN |
1900 | /* need to clone skb, done only once */ |
1901 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1902 | if (!skb2) | |
1903 | goto out_unlock; | |
70978182 | 1904 | |
7866a621 | 1905 | net_timestamp_set(skb2); |
1da177e4 | 1906 | |
7866a621 SN |
1907 | /* skb->nh should be correctly |
1908 | * set by sender, so that the second statement is | |
1909 | * just protection against buggy protocols. | |
1910 | */ | |
1911 | skb_reset_mac_header(skb2); | |
1912 | ||
1913 | if (skb_network_header(skb2) < skb2->data || | |
1914 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
1915 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
1916 | ntohs(skb2->protocol), | |
1917 | dev->name); | |
1918 | skb_reset_network_header(skb2); | |
1da177e4 | 1919 | } |
7866a621 SN |
1920 | |
1921 | skb2->transport_header = skb2->network_header; | |
1922 | skb2->pkt_type = PACKET_OUTGOING; | |
1923 | pt_prev = ptype; | |
1924 | } | |
1925 | ||
1926 | if (ptype_list == &ptype_all) { | |
1927 | ptype_list = &dev->ptype_all; | |
1928 | goto again; | |
1da177e4 | 1929 | } |
7866a621 | 1930 | out_unlock: |
71d9dec2 CG |
1931 | if (pt_prev) |
1932 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1933 | rcu_read_unlock(); |
1934 | } | |
74b20582 | 1935 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 1936 | |
2c53040f BH |
1937 | /** |
1938 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1939 | * @dev: Network device |
1940 | * @txq: number of queues available | |
1941 | * | |
1942 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1943 | * valid. To resolve this verify the tc mapping remains valid and if | |
1944 | * not NULL the mapping. With no priorities mapping to this | |
1945 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1946 | * is invalid nothing can be done so disable priority mappings. If is | |
1947 | * expected that drivers will fix this mapping if they can before | |
1948 | * calling netif_set_real_num_tx_queues. | |
1949 | */ | |
bb134d22 | 1950 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1951 | { |
1952 | int i; | |
1953 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1954 | ||
1955 | /* If TC0 is invalidated disable TC mapping */ | |
1956 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1957 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1958 | dev->num_tc = 0; |
1959 | return; | |
1960 | } | |
1961 | ||
1962 | /* Invalidated prio to tc mappings set to TC0 */ | |
1963 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1964 | int q = netdev_get_prio_tc_map(dev, i); | |
1965 | ||
1966 | tc = &dev->tc_to_txq[q]; | |
1967 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1968 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1969 | i, q); | |
4f57c087 JF |
1970 | netdev_set_prio_tc_map(dev, i, 0); |
1971 | } | |
1972 | } | |
1973 | } | |
1974 | ||
8d059b0f AD |
1975 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
1976 | { | |
1977 | if (dev->num_tc) { | |
1978 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1979 | int i; | |
1980 | ||
1981 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { | |
1982 | if ((txq - tc->offset) < tc->count) | |
1983 | return i; | |
1984 | } | |
1985 | ||
1986 | return -1; | |
1987 | } | |
1988 | ||
1989 | return 0; | |
1990 | } | |
1991 | ||
537c00de AD |
1992 | #ifdef CONFIG_XPS |
1993 | static DEFINE_MUTEX(xps_map_mutex); | |
1994 | #define xmap_dereference(P) \ | |
1995 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1996 | ||
6234f874 AD |
1997 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
1998 | int tci, u16 index) | |
537c00de | 1999 | { |
10cdc3f3 AD |
2000 | struct xps_map *map = NULL; |
2001 | int pos; | |
537c00de | 2002 | |
10cdc3f3 | 2003 | if (dev_maps) |
6234f874 AD |
2004 | map = xmap_dereference(dev_maps->cpu_map[tci]); |
2005 | if (!map) | |
2006 | return false; | |
537c00de | 2007 | |
6234f874 AD |
2008 | for (pos = map->len; pos--;) { |
2009 | if (map->queues[pos] != index) | |
2010 | continue; | |
2011 | ||
2012 | if (map->len > 1) { | |
2013 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2014 | break; |
537c00de | 2015 | } |
6234f874 AD |
2016 | |
2017 | RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL); | |
2018 | kfree_rcu(map, rcu); | |
2019 | return false; | |
537c00de AD |
2020 | } |
2021 | ||
6234f874 | 2022 | return true; |
10cdc3f3 AD |
2023 | } |
2024 | ||
6234f874 AD |
2025 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2026 | struct xps_dev_maps *dev_maps, | |
2027 | int cpu, u16 offset, u16 count) | |
2028 | { | |
184c449f AD |
2029 | int num_tc = dev->num_tc ? : 1; |
2030 | bool active = false; | |
2031 | int tci; | |
6234f874 | 2032 | |
184c449f AD |
2033 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2034 | int i, j; | |
2035 | ||
2036 | for (i = count, j = offset; i--; j++) { | |
2037 | if (!remove_xps_queue(dev_maps, cpu, j)) | |
2038 | break; | |
2039 | } | |
2040 | ||
2041 | active |= i < 0; | |
6234f874 AD |
2042 | } |
2043 | ||
184c449f | 2044 | return active; |
6234f874 AD |
2045 | } |
2046 | ||
2047 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, | |
2048 | u16 count) | |
10cdc3f3 AD |
2049 | { |
2050 | struct xps_dev_maps *dev_maps; | |
024e9679 | 2051 | int cpu, i; |
10cdc3f3 AD |
2052 | bool active = false; |
2053 | ||
2054 | mutex_lock(&xps_map_mutex); | |
2055 | dev_maps = xmap_dereference(dev->xps_maps); | |
2056 | ||
2057 | if (!dev_maps) | |
2058 | goto out_no_maps; | |
2059 | ||
6234f874 AD |
2060 | for_each_possible_cpu(cpu) |
2061 | active |= remove_xps_queue_cpu(dev, dev_maps, cpu, | |
2062 | offset, count); | |
10cdc3f3 AD |
2063 | |
2064 | if (!active) { | |
537c00de AD |
2065 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
2066 | kfree_rcu(dev_maps, rcu); | |
2067 | } | |
2068 | ||
6234f874 | 2069 | for (i = offset + (count - 1); count--; i--) |
024e9679 AD |
2070 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), |
2071 | NUMA_NO_NODE); | |
2072 | ||
537c00de AD |
2073 | out_no_maps: |
2074 | mutex_unlock(&xps_map_mutex); | |
2075 | } | |
2076 | ||
6234f874 AD |
2077 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2078 | { | |
2079 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2080 | } | |
2081 | ||
01c5f864 AD |
2082 | static struct xps_map *expand_xps_map(struct xps_map *map, |
2083 | int cpu, u16 index) | |
2084 | { | |
2085 | struct xps_map *new_map; | |
2086 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2087 | int i, pos; | |
2088 | ||
2089 | for (pos = 0; map && pos < map->len; pos++) { | |
2090 | if (map->queues[pos] != index) | |
2091 | continue; | |
2092 | return map; | |
2093 | } | |
2094 | ||
2095 | /* Need to add queue to this CPU's existing map */ | |
2096 | if (map) { | |
2097 | if (pos < map->alloc_len) | |
2098 | return map; | |
2099 | ||
2100 | alloc_len = map->alloc_len * 2; | |
2101 | } | |
2102 | ||
2103 | /* Need to allocate new map to store queue on this CPU's map */ | |
2104 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2105 | cpu_to_node(cpu)); | |
2106 | if (!new_map) | |
2107 | return NULL; | |
2108 | ||
2109 | for (i = 0; i < pos; i++) | |
2110 | new_map->queues[i] = map->queues[i]; | |
2111 | new_map->alloc_len = alloc_len; | |
2112 | new_map->len = pos; | |
2113 | ||
2114 | return new_map; | |
2115 | } | |
2116 | ||
3573540c MT |
2117 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
2118 | u16 index) | |
537c00de | 2119 | { |
01c5f864 | 2120 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
184c449f AD |
2121 | int i, cpu, tci, numa_node_id = -2; |
2122 | int maps_sz, num_tc = 1, tc = 0; | |
537c00de | 2123 | struct xps_map *map, *new_map; |
01c5f864 | 2124 | bool active = false; |
537c00de | 2125 | |
184c449f AD |
2126 | if (dev->num_tc) { |
2127 | num_tc = dev->num_tc; | |
2128 | tc = netdev_txq_to_tc(dev, index); | |
2129 | if (tc < 0) | |
2130 | return -EINVAL; | |
2131 | } | |
2132 | ||
2133 | maps_sz = XPS_DEV_MAPS_SIZE(num_tc); | |
2134 | if (maps_sz < L1_CACHE_BYTES) | |
2135 | maps_sz = L1_CACHE_BYTES; | |
2136 | ||
537c00de AD |
2137 | mutex_lock(&xps_map_mutex); |
2138 | ||
2139 | dev_maps = xmap_dereference(dev->xps_maps); | |
2140 | ||
01c5f864 | 2141 | /* allocate memory for queue storage */ |
184c449f | 2142 | for_each_cpu_and(cpu, cpu_online_mask, mask) { |
01c5f864 AD |
2143 | if (!new_dev_maps) |
2144 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2145 | if (!new_dev_maps) { |
2146 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2147 | return -ENOMEM; |
2bb60cb9 | 2148 | } |
01c5f864 | 2149 | |
184c449f AD |
2150 | tci = cpu * num_tc + tc; |
2151 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) : | |
01c5f864 AD |
2152 | NULL; |
2153 | ||
2154 | map = expand_xps_map(map, cpu, index); | |
2155 | if (!map) | |
2156 | goto error; | |
2157 | ||
184c449f | 2158 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); |
01c5f864 AD |
2159 | } |
2160 | ||
2161 | if (!new_dev_maps) | |
2162 | goto out_no_new_maps; | |
2163 | ||
537c00de | 2164 | for_each_possible_cpu(cpu) { |
184c449f AD |
2165 | /* copy maps belonging to foreign traffic classes */ |
2166 | for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) { | |
2167 | /* fill in the new device map from the old device map */ | |
2168 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
2169 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
2170 | } | |
2171 | ||
2172 | /* We need to explicitly update tci as prevous loop | |
2173 | * could break out early if dev_maps is NULL. | |
2174 | */ | |
2175 | tci = cpu * num_tc + tc; | |
2176 | ||
01c5f864 AD |
2177 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
2178 | /* add queue to CPU maps */ | |
2179 | int pos = 0; | |
2180 | ||
184c449f | 2181 | map = xmap_dereference(new_dev_maps->cpu_map[tci]); |
01c5f864 AD |
2182 | while ((pos < map->len) && (map->queues[pos] != index)) |
2183 | pos++; | |
2184 | ||
2185 | if (pos == map->len) | |
2186 | map->queues[map->len++] = index; | |
537c00de | 2187 | #ifdef CONFIG_NUMA |
537c00de AD |
2188 | if (numa_node_id == -2) |
2189 | numa_node_id = cpu_to_node(cpu); | |
2190 | else if (numa_node_id != cpu_to_node(cpu)) | |
2191 | numa_node_id = -1; | |
537c00de | 2192 | #endif |
01c5f864 AD |
2193 | } else if (dev_maps) { |
2194 | /* fill in the new device map from the old device map */ | |
184c449f AD |
2195 | map = xmap_dereference(dev_maps->cpu_map[tci]); |
2196 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
537c00de | 2197 | } |
01c5f864 | 2198 | |
184c449f AD |
2199 | /* copy maps belonging to foreign traffic classes */ |
2200 | for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { | |
2201 | /* fill in the new device map from the old device map */ | |
2202 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
2203 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
2204 | } | |
537c00de AD |
2205 | } |
2206 | ||
01c5f864 AD |
2207 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2208 | ||
537c00de | 2209 | /* Cleanup old maps */ |
184c449f AD |
2210 | if (!dev_maps) |
2211 | goto out_no_old_maps; | |
2212 | ||
2213 | for_each_possible_cpu(cpu) { | |
2214 | for (i = num_tc, tci = cpu * num_tc; i--; tci++) { | |
2215 | new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); | |
2216 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
01c5f864 AD |
2217 | if (map && map != new_map) |
2218 | kfree_rcu(map, rcu); | |
2219 | } | |
537c00de AD |
2220 | } |
2221 | ||
184c449f AD |
2222 | kfree_rcu(dev_maps, rcu); |
2223 | ||
2224 | out_no_old_maps: | |
01c5f864 AD |
2225 | dev_maps = new_dev_maps; |
2226 | active = true; | |
537c00de | 2227 | |
01c5f864 AD |
2228 | out_no_new_maps: |
2229 | /* update Tx queue numa node */ | |
537c00de AD |
2230 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2231 | (numa_node_id >= 0) ? numa_node_id : | |
2232 | NUMA_NO_NODE); | |
2233 | ||
01c5f864 AD |
2234 | if (!dev_maps) |
2235 | goto out_no_maps; | |
2236 | ||
2237 | /* removes queue from unused CPUs */ | |
2238 | for_each_possible_cpu(cpu) { | |
184c449f AD |
2239 | for (i = tc, tci = cpu * num_tc; i--; tci++) |
2240 | active |= remove_xps_queue(dev_maps, tci, index); | |
2241 | if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu)) | |
2242 | active |= remove_xps_queue(dev_maps, tci, index); | |
2243 | for (i = num_tc - tc, tci++; --i; tci++) | |
2244 | active |= remove_xps_queue(dev_maps, tci, index); | |
01c5f864 AD |
2245 | } |
2246 | ||
2247 | /* free map if not active */ | |
2248 | if (!active) { | |
2249 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2250 | kfree_rcu(dev_maps, rcu); | |
2251 | } | |
2252 | ||
2253 | out_no_maps: | |
537c00de AD |
2254 | mutex_unlock(&xps_map_mutex); |
2255 | ||
2256 | return 0; | |
2257 | error: | |
01c5f864 AD |
2258 | /* remove any maps that we added */ |
2259 | for_each_possible_cpu(cpu) { | |
184c449f AD |
2260 | for (i = num_tc, tci = cpu * num_tc; i--; tci++) { |
2261 | new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); | |
2262 | map = dev_maps ? | |
2263 | xmap_dereference(dev_maps->cpu_map[tci]) : | |
2264 | NULL; | |
2265 | if (new_map && new_map != map) | |
2266 | kfree(new_map); | |
2267 | } | |
01c5f864 AD |
2268 | } |
2269 | ||
537c00de AD |
2270 | mutex_unlock(&xps_map_mutex); |
2271 | ||
537c00de AD |
2272 | kfree(new_dev_maps); |
2273 | return -ENOMEM; | |
2274 | } | |
2275 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2276 | ||
2277 | #endif | |
9cf1f6a8 AD |
2278 | void netdev_reset_tc(struct net_device *dev) |
2279 | { | |
6234f874 AD |
2280 | #ifdef CONFIG_XPS |
2281 | netif_reset_xps_queues_gt(dev, 0); | |
2282 | #endif | |
9cf1f6a8 AD |
2283 | dev->num_tc = 0; |
2284 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2285 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2286 | } | |
2287 | EXPORT_SYMBOL(netdev_reset_tc); | |
2288 | ||
2289 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2290 | { | |
2291 | if (tc >= dev->num_tc) | |
2292 | return -EINVAL; | |
2293 | ||
6234f874 AD |
2294 | #ifdef CONFIG_XPS |
2295 | netif_reset_xps_queues(dev, offset, count); | |
2296 | #endif | |
9cf1f6a8 AD |
2297 | dev->tc_to_txq[tc].count = count; |
2298 | dev->tc_to_txq[tc].offset = offset; | |
2299 | return 0; | |
2300 | } | |
2301 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2302 | ||
2303 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2304 | { | |
2305 | if (num_tc > TC_MAX_QUEUE) | |
2306 | return -EINVAL; | |
2307 | ||
6234f874 AD |
2308 | #ifdef CONFIG_XPS |
2309 | netif_reset_xps_queues_gt(dev, 0); | |
2310 | #endif | |
9cf1f6a8 AD |
2311 | dev->num_tc = num_tc; |
2312 | return 0; | |
2313 | } | |
2314 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2315 | ||
f0796d5c JF |
2316 | /* |
2317 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2318 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2319 | */ | |
e6484930 | 2320 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2321 | { |
1d24eb48 TH |
2322 | int rc; |
2323 | ||
e6484930 TH |
2324 | if (txq < 1 || txq > dev->num_tx_queues) |
2325 | return -EINVAL; | |
f0796d5c | 2326 | |
5c56580b BH |
2327 | if (dev->reg_state == NETREG_REGISTERED || |
2328 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2329 | ASSERT_RTNL(); |
2330 | ||
1d24eb48 TH |
2331 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2332 | txq); | |
bf264145 TH |
2333 | if (rc) |
2334 | return rc; | |
2335 | ||
4f57c087 JF |
2336 | if (dev->num_tc) |
2337 | netif_setup_tc(dev, txq); | |
2338 | ||
024e9679 | 2339 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2340 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2341 | #ifdef CONFIG_XPS |
2342 | netif_reset_xps_queues_gt(dev, txq); | |
2343 | #endif | |
2344 | } | |
f0796d5c | 2345 | } |
e6484930 TH |
2346 | |
2347 | dev->real_num_tx_queues = txq; | |
2348 | return 0; | |
f0796d5c JF |
2349 | } |
2350 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2351 | |
a953be53 | 2352 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2353 | /** |
2354 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2355 | * @dev: Network device | |
2356 | * @rxq: Actual number of RX queues | |
2357 | * | |
2358 | * This must be called either with the rtnl_lock held or before | |
2359 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2360 | * negative error code. If called before registration, it always |
2361 | * succeeds. | |
62fe0b40 BH |
2362 | */ |
2363 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2364 | { | |
2365 | int rc; | |
2366 | ||
bd25fa7b TH |
2367 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2368 | return -EINVAL; | |
2369 | ||
62fe0b40 BH |
2370 | if (dev->reg_state == NETREG_REGISTERED) { |
2371 | ASSERT_RTNL(); | |
2372 | ||
62fe0b40 BH |
2373 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2374 | rxq); | |
2375 | if (rc) | |
2376 | return rc; | |
62fe0b40 BH |
2377 | } |
2378 | ||
2379 | dev->real_num_rx_queues = rxq; | |
2380 | return 0; | |
2381 | } | |
2382 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2383 | #endif | |
2384 | ||
2c53040f BH |
2385 | /** |
2386 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2387 | * |
2388 | * This routine should set an upper limit on the number of RSS queues | |
2389 | * used by default by multiqueue devices. | |
2390 | */ | |
a55b138b | 2391 | int netif_get_num_default_rss_queues(void) |
16917b87 | 2392 | { |
40e4e713 HS |
2393 | return is_kdump_kernel() ? |
2394 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
2395 | } |
2396 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2397 | ||
3bcb846c | 2398 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 2399 | { |
def82a1d JP |
2400 | struct softnet_data *sd; |
2401 | unsigned long flags; | |
56079431 | 2402 | |
def82a1d | 2403 | local_irq_save(flags); |
903ceff7 | 2404 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2405 | q->next_sched = NULL; |
2406 | *sd->output_queue_tailp = q; | |
2407 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2408 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2409 | local_irq_restore(flags); | |
2410 | } | |
2411 | ||
2412 | void __netif_schedule(struct Qdisc *q) | |
2413 | { | |
2414 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2415 | __netif_reschedule(q); | |
56079431 DV |
2416 | } |
2417 | EXPORT_SYMBOL(__netif_schedule); | |
2418 | ||
e6247027 ED |
2419 | struct dev_kfree_skb_cb { |
2420 | enum skb_free_reason reason; | |
2421 | }; | |
2422 | ||
2423 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2424 | { |
e6247027 ED |
2425 | return (struct dev_kfree_skb_cb *)skb->cb; |
2426 | } | |
2427 | ||
46e5da40 JF |
2428 | void netif_schedule_queue(struct netdev_queue *txq) |
2429 | { | |
2430 | rcu_read_lock(); | |
2431 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2432 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2433 | ||
2434 | __netif_schedule(q); | |
2435 | } | |
2436 | rcu_read_unlock(); | |
2437 | } | |
2438 | EXPORT_SYMBOL(netif_schedule_queue); | |
2439 | ||
46e5da40 JF |
2440 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
2441 | { | |
2442 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2443 | struct Qdisc *q; | |
2444 | ||
2445 | rcu_read_lock(); | |
2446 | q = rcu_dereference(dev_queue->qdisc); | |
2447 | __netif_schedule(q); | |
2448 | rcu_read_unlock(); | |
2449 | } | |
2450 | } | |
2451 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2452 | ||
e6247027 | 2453 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2454 | { |
e6247027 | 2455 | unsigned long flags; |
56079431 | 2456 | |
9899886d MJ |
2457 | if (unlikely(!skb)) |
2458 | return; | |
2459 | ||
e6247027 ED |
2460 | if (likely(atomic_read(&skb->users) == 1)) { |
2461 | smp_rmb(); | |
2462 | atomic_set(&skb->users, 0); | |
2463 | } else if (likely(!atomic_dec_and_test(&skb->users))) { | |
2464 | return; | |
bea3348e | 2465 | } |
e6247027 ED |
2466 | get_kfree_skb_cb(skb)->reason = reason; |
2467 | local_irq_save(flags); | |
2468 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2469 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2470 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2471 | local_irq_restore(flags); | |
56079431 | 2472 | } |
e6247027 | 2473 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2474 | |
e6247027 | 2475 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2476 | { |
2477 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2478 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2479 | else |
2480 | dev_kfree_skb(skb); | |
2481 | } | |
e6247027 | 2482 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2483 | |
2484 | ||
bea3348e SH |
2485 | /** |
2486 | * netif_device_detach - mark device as removed | |
2487 | * @dev: network device | |
2488 | * | |
2489 | * Mark device as removed from system and therefore no longer available. | |
2490 | */ | |
56079431 DV |
2491 | void netif_device_detach(struct net_device *dev) |
2492 | { | |
2493 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2494 | netif_running(dev)) { | |
d543103a | 2495 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2496 | } |
2497 | } | |
2498 | EXPORT_SYMBOL(netif_device_detach); | |
2499 | ||
bea3348e SH |
2500 | /** |
2501 | * netif_device_attach - mark device as attached | |
2502 | * @dev: network device | |
2503 | * | |
2504 | * Mark device as attached from system and restart if needed. | |
2505 | */ | |
56079431 DV |
2506 | void netif_device_attach(struct net_device *dev) |
2507 | { | |
2508 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2509 | netif_running(dev)) { | |
d543103a | 2510 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2511 | __netdev_watchdog_up(dev); |
56079431 DV |
2512 | } |
2513 | } | |
2514 | EXPORT_SYMBOL(netif_device_attach); | |
2515 | ||
5605c762 JP |
2516 | /* |
2517 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2518 | * to be used as a distribution range. | |
2519 | */ | |
2520 | u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, | |
2521 | unsigned int num_tx_queues) | |
2522 | { | |
2523 | u32 hash; | |
2524 | u16 qoffset = 0; | |
2525 | u16 qcount = num_tx_queues; | |
2526 | ||
2527 | if (skb_rx_queue_recorded(skb)) { | |
2528 | hash = skb_get_rx_queue(skb); | |
2529 | while (unlikely(hash >= num_tx_queues)) | |
2530 | hash -= num_tx_queues; | |
2531 | return hash; | |
2532 | } | |
2533 | ||
2534 | if (dev->num_tc) { | |
2535 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
f4563a75 | 2536 | |
5605c762 JP |
2537 | qoffset = dev->tc_to_txq[tc].offset; |
2538 | qcount = dev->tc_to_txq[tc].count; | |
2539 | } | |
2540 | ||
2541 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
2542 | } | |
2543 | EXPORT_SYMBOL(__skb_tx_hash); | |
2544 | ||
36c92474 BH |
2545 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2546 | { | |
84d15ae5 | 2547 | static const netdev_features_t null_features; |
36c92474 | 2548 | struct net_device *dev = skb->dev; |
88ad4175 | 2549 | const char *name = ""; |
36c92474 | 2550 | |
c846ad9b BG |
2551 | if (!net_ratelimit()) |
2552 | return; | |
2553 | ||
88ad4175 BM |
2554 | if (dev) { |
2555 | if (dev->dev.parent) | |
2556 | name = dev_driver_string(dev->dev.parent); | |
2557 | else | |
2558 | name = netdev_name(dev); | |
2559 | } | |
36c92474 BH |
2560 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " |
2561 | "gso_type=%d ip_summed=%d\n", | |
88ad4175 | 2562 | name, dev ? &dev->features : &null_features, |
65e9d2fa | 2563 | skb->sk ? &skb->sk->sk_route_caps : &null_features, |
36c92474 BH |
2564 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2565 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2566 | } | |
2567 | ||
1da177e4 LT |
2568 | /* |
2569 | * Invalidate hardware checksum when packet is to be mangled, and | |
2570 | * complete checksum manually on outgoing path. | |
2571 | */ | |
84fa7933 | 2572 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2573 | { |
d3bc23e7 | 2574 | __wsum csum; |
663ead3b | 2575 | int ret = 0, offset; |
1da177e4 | 2576 | |
84fa7933 | 2577 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2578 | goto out_set_summed; |
2579 | ||
2580 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2581 | skb_warn_bad_offload(skb); |
2582 | return -EINVAL; | |
1da177e4 LT |
2583 | } |
2584 | ||
cef401de ED |
2585 | /* Before computing a checksum, we should make sure no frag could |
2586 | * be modified by an external entity : checksum could be wrong. | |
2587 | */ | |
2588 | if (skb_has_shared_frag(skb)) { | |
2589 | ret = __skb_linearize(skb); | |
2590 | if (ret) | |
2591 | goto out; | |
2592 | } | |
2593 | ||
55508d60 | 2594 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2595 | BUG_ON(offset >= skb_headlen(skb)); |
2596 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2597 | ||
2598 | offset += skb->csum_offset; | |
2599 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2600 | ||
2601 | if (skb_cloned(skb) && | |
2602 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2603 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2604 | if (ret) | |
2605 | goto out; | |
2606 | } | |
2607 | ||
4f2e4ad5 | 2608 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 2609 | out_set_summed: |
1da177e4 | 2610 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2611 | out: |
1da177e4 LT |
2612 | return ret; |
2613 | } | |
d1b19dff | 2614 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2615 | |
b72b5bf6 DC |
2616 | int skb_crc32c_csum_help(struct sk_buff *skb) |
2617 | { | |
2618 | __le32 crc32c_csum; | |
2619 | int ret = 0, offset, start; | |
2620 | ||
2621 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
2622 | goto out; | |
2623 | ||
2624 | if (unlikely(skb_is_gso(skb))) | |
2625 | goto out; | |
2626 | ||
2627 | /* Before computing a checksum, we should make sure no frag could | |
2628 | * be modified by an external entity : checksum could be wrong. | |
2629 | */ | |
2630 | if (unlikely(skb_has_shared_frag(skb))) { | |
2631 | ret = __skb_linearize(skb); | |
2632 | if (ret) | |
2633 | goto out; | |
2634 | } | |
2635 | start = skb_checksum_start_offset(skb); | |
2636 | offset = start + offsetof(struct sctphdr, checksum); | |
2637 | if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { | |
2638 | ret = -EINVAL; | |
2639 | goto out; | |
2640 | } | |
2641 | if (skb_cloned(skb) && | |
2642 | !skb_clone_writable(skb, offset + sizeof(__le32))) { | |
2643 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); | |
2644 | if (ret) | |
2645 | goto out; | |
2646 | } | |
2647 | crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, | |
2648 | skb->len - start, ~(__u32)0, | |
2649 | crc32c_csum_stub)); | |
2650 | *(__le32 *)(skb->data + offset) = crc32c_csum; | |
2651 | skb->ip_summed = CHECKSUM_NONE; | |
dba00306 | 2652 | skb->csum_not_inet = 0; |
b72b5bf6 DC |
2653 | out: |
2654 | return ret; | |
2655 | } | |
2656 | ||
53d6471c | 2657 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2658 | { |
252e3346 | 2659 | __be16 type = skb->protocol; |
f6a78bfc | 2660 | |
19acc327 PS |
2661 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2662 | if (type == htons(ETH_P_TEB)) { | |
2663 | struct ethhdr *eth; | |
2664 | ||
2665 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2666 | return 0; | |
2667 | ||
2668 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2669 | type = eth->h_proto; | |
2670 | } | |
2671 | ||
d4bcef3f | 2672 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
2673 | } |
2674 | ||
2675 | /** | |
2676 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2677 | * @skb: buffer to segment | |
2678 | * @features: features for the output path (see dev->features) | |
2679 | */ | |
2680 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2681 | netdev_features_t features) | |
2682 | { | |
2683 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2684 | struct packet_offload *ptype; | |
53d6471c VY |
2685 | int vlan_depth = skb->mac_len; |
2686 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2687 | |
2688 | if (unlikely(!type)) | |
2689 | return ERR_PTR(-EINVAL); | |
2690 | ||
53d6471c | 2691 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2692 | |
2693 | rcu_read_lock(); | |
22061d80 | 2694 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2695 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2696 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2697 | break; |
2698 | } | |
2699 | } | |
2700 | rcu_read_unlock(); | |
2701 | ||
98e399f8 | 2702 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2703 | |
f6a78bfc HX |
2704 | return segs; |
2705 | } | |
05e8ef4a PS |
2706 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2707 | ||
2708 | ||
2709 | /* openvswitch calls this on rx path, so we need a different check. | |
2710 | */ | |
2711 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2712 | { | |
2713 | if (tx_path) | |
6e7bc478 ED |
2714 | return skb->ip_summed != CHECKSUM_PARTIAL && |
2715 | skb->ip_summed != CHECKSUM_NONE; | |
2716 | ||
2717 | return skb->ip_summed == CHECKSUM_NONE; | |
05e8ef4a PS |
2718 | } |
2719 | ||
2720 | /** | |
2721 | * __skb_gso_segment - Perform segmentation on skb. | |
2722 | * @skb: buffer to segment | |
2723 | * @features: features for the output path (see dev->features) | |
2724 | * @tx_path: whether it is called in TX path | |
2725 | * | |
2726 | * This function segments the given skb and returns a list of segments. | |
2727 | * | |
2728 | * It may return NULL if the skb requires no segmentation. This is | |
2729 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 KK |
2730 | * |
2731 | * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. | |
05e8ef4a PS |
2732 | */ |
2733 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2734 | netdev_features_t features, bool tx_path) | |
2735 | { | |
b2504a5d ED |
2736 | struct sk_buff *segs; |
2737 | ||
05e8ef4a PS |
2738 | if (unlikely(skb_needs_check(skb, tx_path))) { |
2739 | int err; | |
2740 | ||
b2504a5d | 2741 | /* We're going to init ->check field in TCP or UDP header */ |
a40e0a66 | 2742 | err = skb_cow_head(skb, 0); |
2743 | if (err < 0) | |
05e8ef4a PS |
2744 | return ERR_PTR(err); |
2745 | } | |
2746 | ||
802ab55a AD |
2747 | /* Only report GSO partial support if it will enable us to |
2748 | * support segmentation on this frame without needing additional | |
2749 | * work. | |
2750 | */ | |
2751 | if (features & NETIF_F_GSO_PARTIAL) { | |
2752 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
2753 | struct net_device *dev = skb->dev; | |
2754 | ||
2755 | partial_features |= dev->features & dev->gso_partial_features; | |
2756 | if (!skb_gso_ok(skb, features | partial_features)) | |
2757 | features &= ~NETIF_F_GSO_PARTIAL; | |
2758 | } | |
2759 | ||
9207f9d4 KK |
2760 | BUILD_BUG_ON(SKB_SGO_CB_OFFSET + |
2761 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); | |
2762 | ||
68c33163 | 2763 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2764 | SKB_GSO_CB(skb)->encap_level = 0; |
2765 | ||
05e8ef4a PS |
2766 | skb_reset_mac_header(skb); |
2767 | skb_reset_mac_len(skb); | |
2768 | ||
b2504a5d ED |
2769 | segs = skb_mac_gso_segment(skb, features); |
2770 | ||
2771 | if (unlikely(skb_needs_check(skb, tx_path))) | |
2772 | skb_warn_bad_offload(skb); | |
2773 | ||
2774 | return segs; | |
05e8ef4a | 2775 | } |
12b0004d | 2776 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2777 | |
fb286bb2 HX |
2778 | /* Take action when hardware reception checksum errors are detected. */ |
2779 | #ifdef CONFIG_BUG | |
2780 | void netdev_rx_csum_fault(struct net_device *dev) | |
2781 | { | |
2782 | if (net_ratelimit()) { | |
7b6cd1ce | 2783 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2784 | dump_stack(); |
2785 | } | |
2786 | } | |
2787 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2788 | #endif | |
2789 | ||
1da177e4 LT |
2790 | /* Actually, we should eliminate this check as soon as we know, that: |
2791 | * 1. IOMMU is present and allows to map all the memory. | |
2792 | * 2. No high memory really exists on this machine. | |
2793 | */ | |
2794 | ||
c1e756bf | 2795 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2796 | { |
3d3a8533 | 2797 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2798 | int i; |
f4563a75 | 2799 | |
5acbbd42 | 2800 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2801 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2802 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 2803 | |
ea2ab693 | 2804 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 2805 | return 1; |
ea2ab693 | 2806 | } |
5acbbd42 | 2807 | } |
1da177e4 | 2808 | |
5acbbd42 FT |
2809 | if (PCI_DMA_BUS_IS_PHYS) { |
2810 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2811 | |
9092c658 ED |
2812 | if (!pdev) |
2813 | return 0; | |
5acbbd42 | 2814 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2815 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2816 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
f4563a75 | 2817 | |
5acbbd42 FT |
2818 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2819 | return 1; | |
2820 | } | |
2821 | } | |
3d3a8533 | 2822 | #endif |
1da177e4 LT |
2823 | return 0; |
2824 | } | |
1da177e4 | 2825 | |
3b392ddb SH |
2826 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2827 | * instead of standard features for the netdev. | |
2828 | */ | |
d0edc7bf | 2829 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
2830 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
2831 | netdev_features_t features, | |
2832 | __be16 type) | |
2833 | { | |
25cd9ba0 | 2834 | if (eth_p_mpls(type)) |
3b392ddb SH |
2835 | features &= skb->dev->mpls_features; |
2836 | ||
2837 | return features; | |
2838 | } | |
2839 | #else | |
2840 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2841 | netdev_features_t features, | |
2842 | __be16 type) | |
2843 | { | |
2844 | return features; | |
2845 | } | |
2846 | #endif | |
2847 | ||
c8f44aff | 2848 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2849 | netdev_features_t features) |
f01a5236 | 2850 | { |
53d6471c | 2851 | int tmp; |
3b392ddb SH |
2852 | __be16 type; |
2853 | ||
2854 | type = skb_network_protocol(skb, &tmp); | |
2855 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2856 | |
c0d680e5 | 2857 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2858 | !can_checksum_protocol(features, type)) { |
996e8021 | 2859 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 2860 | } |
7be2c82c ED |
2861 | if (illegal_highdma(skb->dev, skb)) |
2862 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
2863 | |
2864 | return features; | |
2865 | } | |
2866 | ||
e38f3025 TM |
2867 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
2868 | struct net_device *dev, | |
2869 | netdev_features_t features) | |
2870 | { | |
2871 | return features; | |
2872 | } | |
2873 | EXPORT_SYMBOL(passthru_features_check); | |
2874 | ||
8cb65d00 TM |
2875 | static netdev_features_t dflt_features_check(const struct sk_buff *skb, |
2876 | struct net_device *dev, | |
2877 | netdev_features_t features) | |
2878 | { | |
2879 | return vlan_features_check(skb, features); | |
2880 | } | |
2881 | ||
cbc53e08 AD |
2882 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
2883 | struct net_device *dev, | |
2884 | netdev_features_t features) | |
2885 | { | |
2886 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
2887 | ||
2888 | if (gso_segs > dev->gso_max_segs) | |
2889 | return features & ~NETIF_F_GSO_MASK; | |
2890 | ||
802ab55a AD |
2891 | /* Support for GSO partial features requires software |
2892 | * intervention before we can actually process the packets | |
2893 | * so we need to strip support for any partial features now | |
2894 | * and we can pull them back in after we have partially | |
2895 | * segmented the frame. | |
2896 | */ | |
2897 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
2898 | features &= ~dev->gso_partial_features; | |
2899 | ||
2900 | /* Make sure to clear the IPv4 ID mangling feature if the | |
2901 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
2902 | */ |
2903 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
2904 | struct iphdr *iph = skb->encapsulation ? | |
2905 | inner_ip_hdr(skb) : ip_hdr(skb); | |
2906 | ||
2907 | if (!(iph->frag_off & htons(IP_DF))) | |
2908 | features &= ~NETIF_F_TSO_MANGLEID; | |
2909 | } | |
2910 | ||
2911 | return features; | |
2912 | } | |
2913 | ||
c1e756bf | 2914 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 2915 | { |
5f35227e | 2916 | struct net_device *dev = skb->dev; |
fcbeb976 | 2917 | netdev_features_t features = dev->features; |
58e998c6 | 2918 | |
cbc53e08 AD |
2919 | if (skb_is_gso(skb)) |
2920 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 2921 | |
5f35227e JG |
2922 | /* If encapsulation offload request, verify we are testing |
2923 | * hardware encapsulation features instead of standard | |
2924 | * features for the netdev | |
2925 | */ | |
2926 | if (skb->encapsulation) | |
2927 | features &= dev->hw_enc_features; | |
2928 | ||
f5a7fb88 TM |
2929 | if (skb_vlan_tagged(skb)) |
2930 | features = netdev_intersect_features(features, | |
2931 | dev->vlan_features | | |
2932 | NETIF_F_HW_VLAN_CTAG_TX | | |
2933 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2934 | |
5f35227e JG |
2935 | if (dev->netdev_ops->ndo_features_check) |
2936 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
2937 | features); | |
8cb65d00 TM |
2938 | else |
2939 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 2940 | |
c1e756bf | 2941 | return harmonize_features(skb, features); |
58e998c6 | 2942 | } |
c1e756bf | 2943 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2944 | |
2ea25513 | 2945 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2946 | struct netdev_queue *txq, bool more) |
f6a78bfc | 2947 | { |
2ea25513 DM |
2948 | unsigned int len; |
2949 | int rc; | |
00829823 | 2950 | |
7866a621 | 2951 | if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) |
2ea25513 | 2952 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 2953 | |
2ea25513 DM |
2954 | len = skb->len; |
2955 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 2956 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 2957 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 2958 | |
2ea25513 DM |
2959 | return rc; |
2960 | } | |
7b9c6090 | 2961 | |
8dcda22a DM |
2962 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
2963 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
2964 | { |
2965 | struct sk_buff *skb = first; | |
2966 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 2967 | |
7f2e870f DM |
2968 | while (skb) { |
2969 | struct sk_buff *next = skb->next; | |
fc70fb64 | 2970 | |
7f2e870f | 2971 | skb->next = NULL; |
95f6b3dd | 2972 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
2973 | if (unlikely(!dev_xmit_complete(rc))) { |
2974 | skb->next = next; | |
2975 | goto out; | |
2976 | } | |
6afff0ca | 2977 | |
7f2e870f DM |
2978 | skb = next; |
2979 | if (netif_xmit_stopped(txq) && skb) { | |
2980 | rc = NETDEV_TX_BUSY; | |
2981 | break; | |
9ccb8975 | 2982 | } |
7f2e870f | 2983 | } |
9ccb8975 | 2984 | |
7f2e870f DM |
2985 | out: |
2986 | *ret = rc; | |
2987 | return skb; | |
2988 | } | |
b40863c6 | 2989 | |
1ff0dc94 ED |
2990 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
2991 | netdev_features_t features) | |
f6a78bfc | 2992 | { |
df8a39de | 2993 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
2994 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
2995 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
2996 | return skb; |
2997 | } | |
f6a78bfc | 2998 | |
55a93b3e | 2999 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
3000 | { |
3001 | netdev_features_t features; | |
f6a78bfc | 3002 | |
eae3f88e DM |
3003 | features = netif_skb_features(skb); |
3004 | skb = validate_xmit_vlan(skb, features); | |
3005 | if (unlikely(!skb)) | |
3006 | goto out_null; | |
7b9c6090 | 3007 | |
8b86a61d | 3008 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
3009 | struct sk_buff *segs; |
3010 | ||
3011 | segs = skb_gso_segment(skb, features); | |
cecda693 | 3012 | if (IS_ERR(segs)) { |
af6dabc9 | 3013 | goto out_kfree_skb; |
cecda693 JW |
3014 | } else if (segs) { |
3015 | consume_skb(skb); | |
3016 | skb = segs; | |
f6a78bfc | 3017 | } |
eae3f88e DM |
3018 | } else { |
3019 | if (skb_needs_linearize(skb, features) && | |
3020 | __skb_linearize(skb)) | |
3021 | goto out_kfree_skb; | |
4ec93edb | 3022 | |
f6e27114 SK |
3023 | if (validate_xmit_xfrm(skb, features)) |
3024 | goto out_kfree_skb; | |
3025 | ||
eae3f88e DM |
3026 | /* If packet is not checksummed and device does not |
3027 | * support checksumming for this protocol, complete | |
3028 | * checksumming here. | |
3029 | */ | |
3030 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
3031 | if (skb->encapsulation) | |
3032 | skb_set_inner_transport_header(skb, | |
3033 | skb_checksum_start_offset(skb)); | |
3034 | else | |
3035 | skb_set_transport_header(skb, | |
3036 | skb_checksum_start_offset(skb)); | |
a188222b | 3037 | if (!(features & NETIF_F_CSUM_MASK) && |
eae3f88e DM |
3038 | skb_checksum_help(skb)) |
3039 | goto out_kfree_skb; | |
7b9c6090 | 3040 | } |
0c772159 | 3041 | } |
7b9c6090 | 3042 | |
eae3f88e | 3043 | return skb; |
fc70fb64 | 3044 | |
f6a78bfc HX |
3045 | out_kfree_skb: |
3046 | kfree_skb(skb); | |
eae3f88e | 3047 | out_null: |
d21fd63e | 3048 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3049 | return NULL; |
3050 | } | |
6afff0ca | 3051 | |
55a93b3e ED |
3052 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
3053 | { | |
3054 | struct sk_buff *next, *head = NULL, *tail; | |
3055 | ||
bec3cfdc | 3056 | for (; skb != NULL; skb = next) { |
55a93b3e ED |
3057 | next = skb->next; |
3058 | skb->next = NULL; | |
bec3cfdc ED |
3059 | |
3060 | /* in case skb wont be segmented, point to itself */ | |
3061 | skb->prev = skb; | |
3062 | ||
55a93b3e | 3063 | skb = validate_xmit_skb(skb, dev); |
bec3cfdc ED |
3064 | if (!skb) |
3065 | continue; | |
55a93b3e | 3066 | |
bec3cfdc ED |
3067 | if (!head) |
3068 | head = skb; | |
3069 | else | |
3070 | tail->next = skb; | |
3071 | /* If skb was segmented, skb->prev points to | |
3072 | * the last segment. If not, it still contains skb. | |
3073 | */ | |
3074 | tail = skb->prev; | |
55a93b3e ED |
3075 | } |
3076 | return head; | |
f6a78bfc | 3077 | } |
104ba78c | 3078 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3079 | |
1def9238 ED |
3080 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3081 | { | |
3082 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3083 | ||
3084 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3085 | ||
3086 | /* To get more precise estimation of bytes sent on wire, | |
3087 | * we add to pkt_len the headers size of all segments | |
3088 | */ | |
3089 | if (shinfo->gso_size) { | |
757b8b1d | 3090 | unsigned int hdr_len; |
15e5a030 | 3091 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3092 | |
757b8b1d ED |
3093 | /* mac layer + network layer */ |
3094 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3095 | ||
3096 | /* + transport layer */ | |
1def9238 ED |
3097 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
3098 | hdr_len += tcp_hdrlen(skb); | |
3099 | else | |
3100 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
3101 | |
3102 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3103 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3104 | shinfo->gso_size); | |
3105 | ||
3106 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3107 | } |
3108 | } | |
3109 | ||
bbd8a0d3 KK |
3110 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3111 | struct net_device *dev, | |
3112 | struct netdev_queue *txq) | |
3113 | { | |
3114 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3115 | struct sk_buff *to_free = NULL; |
a2da570d | 3116 | bool contended; |
bbd8a0d3 KK |
3117 | int rc; |
3118 | ||
a2da570d | 3119 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
3120 | /* |
3121 | * Heuristic to force contended enqueues to serialize on a | |
3122 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3123 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3124 | * often and dequeue packets faster. |
79640a4c | 3125 | */ |
a2da570d | 3126 | contended = qdisc_is_running(q); |
79640a4c ED |
3127 | if (unlikely(contended)) |
3128 | spin_lock(&q->busylock); | |
3129 | ||
bbd8a0d3 KK |
3130 | spin_lock(root_lock); |
3131 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3132 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3133 | rc = NET_XMIT_DROP; |
3134 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3135 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3136 | /* |
3137 | * This is a work-conserving queue; there are no old skbs | |
3138 | * waiting to be sent out; and the qdisc is not running - | |
3139 | * xmit the skb directly. | |
3140 | */ | |
bfe0d029 | 3141 | |
bfe0d029 ED |
3142 | qdisc_bstats_update(q, skb); |
3143 | ||
55a93b3e | 3144 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3145 | if (unlikely(contended)) { |
3146 | spin_unlock(&q->busylock); | |
3147 | contended = false; | |
3148 | } | |
bbd8a0d3 | 3149 | __qdisc_run(q); |
79640a4c | 3150 | } else |
bc135b23 | 3151 | qdisc_run_end(q); |
bbd8a0d3 KK |
3152 | |
3153 | rc = NET_XMIT_SUCCESS; | |
3154 | } else { | |
520ac30f | 3155 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3156 | if (qdisc_run_begin(q)) { |
3157 | if (unlikely(contended)) { | |
3158 | spin_unlock(&q->busylock); | |
3159 | contended = false; | |
3160 | } | |
3161 | __qdisc_run(q); | |
3162 | } | |
bbd8a0d3 KK |
3163 | } |
3164 | spin_unlock(root_lock); | |
520ac30f ED |
3165 | if (unlikely(to_free)) |
3166 | kfree_skb_list(to_free); | |
79640a4c ED |
3167 | if (unlikely(contended)) |
3168 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3169 | return rc; |
3170 | } | |
3171 | ||
86f8515f | 3172 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3173 | static void skb_update_prio(struct sk_buff *skb) |
3174 | { | |
6977a79d | 3175 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 3176 | |
91c68ce2 | 3177 | if (!skb->priority && skb->sk && map) { |
2a56a1fe TH |
3178 | unsigned int prioidx = |
3179 | sock_cgroup_prioidx(&skb->sk->sk_cgrp_data); | |
91c68ce2 ED |
3180 | |
3181 | if (prioidx < map->priomap_len) | |
3182 | skb->priority = map->priomap[prioidx]; | |
3183 | } | |
5bc1421e NH |
3184 | } |
3185 | #else | |
3186 | #define skb_update_prio(skb) | |
3187 | #endif | |
3188 | ||
f60e5990 | 3189 | DEFINE_PER_CPU(int, xmit_recursion); |
3190 | EXPORT_SYMBOL(xmit_recursion); | |
3191 | ||
95603e22 MM |
3192 | /** |
3193 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3194 | * @net: network namespace this loopback is happening in |
3195 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3196 | * @skb: buffer to transmit |
3197 | */ | |
0c4b51f0 | 3198 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3199 | { |
3200 | skb_reset_mac_header(skb); | |
3201 | __skb_pull(skb, skb_network_offset(skb)); | |
3202 | skb->pkt_type = PACKET_LOOPBACK; | |
3203 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3204 | WARN_ON(!skb_dst(skb)); | |
3205 | skb_dst_force(skb); | |
3206 | netif_rx_ni(skb); | |
3207 | return 0; | |
3208 | } | |
3209 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3210 | ||
1f211a1b DB |
3211 | #ifdef CONFIG_NET_EGRESS |
3212 | static struct sk_buff * | |
3213 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3214 | { | |
3215 | struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list); | |
3216 | struct tcf_result cl_res; | |
3217 | ||
3218 | if (!cl) | |
3219 | return skb; | |
3220 | ||
8dc07fdb | 3221 | /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ |
1f211a1b DB |
3222 | qdisc_bstats_cpu_update(cl->q, skb); |
3223 | ||
87d83093 | 3224 | switch (tcf_classify(skb, cl, &cl_res, false)) { |
1f211a1b DB |
3225 | case TC_ACT_OK: |
3226 | case TC_ACT_RECLASSIFY: | |
3227 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3228 | break; | |
3229 | case TC_ACT_SHOT: | |
3230 | qdisc_qstats_cpu_drop(cl->q); | |
3231 | *ret = NET_XMIT_DROP; | |
7e2c3aea DB |
3232 | kfree_skb(skb); |
3233 | return NULL; | |
1f211a1b DB |
3234 | case TC_ACT_STOLEN: |
3235 | case TC_ACT_QUEUED: | |
3236 | *ret = NET_XMIT_SUCCESS; | |
7e2c3aea | 3237 | consume_skb(skb); |
1f211a1b DB |
3238 | return NULL; |
3239 | case TC_ACT_REDIRECT: | |
3240 | /* No need to push/pop skb's mac_header here on egress! */ | |
3241 | skb_do_redirect(skb); | |
3242 | *ret = NET_XMIT_SUCCESS; | |
3243 | return NULL; | |
3244 | default: | |
3245 | break; | |
3246 | } | |
3247 | ||
3248 | return skb; | |
3249 | } | |
3250 | #endif /* CONFIG_NET_EGRESS */ | |
3251 | ||
638b2a69 JP |
3252 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
3253 | { | |
3254 | #ifdef CONFIG_XPS | |
3255 | struct xps_dev_maps *dev_maps; | |
3256 | struct xps_map *map; | |
3257 | int queue_index = -1; | |
3258 | ||
3259 | rcu_read_lock(); | |
3260 | dev_maps = rcu_dereference(dev->xps_maps); | |
3261 | if (dev_maps) { | |
184c449f AD |
3262 | unsigned int tci = skb->sender_cpu - 1; |
3263 | ||
3264 | if (dev->num_tc) { | |
3265 | tci *= dev->num_tc; | |
3266 | tci += netdev_get_prio_tc_map(dev, skb->priority); | |
3267 | } | |
3268 | ||
3269 | map = rcu_dereference(dev_maps->cpu_map[tci]); | |
638b2a69 JP |
3270 | if (map) { |
3271 | if (map->len == 1) | |
3272 | queue_index = map->queues[0]; | |
3273 | else | |
3274 | queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), | |
3275 | map->len)]; | |
3276 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3277 | queue_index = -1; | |
3278 | } | |
3279 | } | |
3280 | rcu_read_unlock(); | |
3281 | ||
3282 | return queue_index; | |
3283 | #else | |
3284 | return -1; | |
3285 | #endif | |
3286 | } | |
3287 | ||
3288 | static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) | |
3289 | { | |
3290 | struct sock *sk = skb->sk; | |
3291 | int queue_index = sk_tx_queue_get(sk); | |
3292 | ||
3293 | if (queue_index < 0 || skb->ooo_okay || | |
3294 | queue_index >= dev->real_num_tx_queues) { | |
3295 | int new_index = get_xps_queue(dev, skb); | |
f4563a75 | 3296 | |
638b2a69 JP |
3297 | if (new_index < 0) |
3298 | new_index = skb_tx_hash(dev, skb); | |
3299 | ||
3300 | if (queue_index != new_index && sk && | |
004a5d01 | 3301 | sk_fullsock(sk) && |
638b2a69 JP |
3302 | rcu_access_pointer(sk->sk_dst_cache)) |
3303 | sk_tx_queue_set(sk, new_index); | |
3304 | ||
3305 | queue_index = new_index; | |
3306 | } | |
3307 | ||
3308 | return queue_index; | |
3309 | } | |
3310 | ||
3311 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, | |
3312 | struct sk_buff *skb, | |
3313 | void *accel_priv) | |
3314 | { | |
3315 | int queue_index = 0; | |
3316 | ||
3317 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
3318 | u32 sender_cpu = skb->sender_cpu - 1; |
3319 | ||
3320 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
3321 | skb->sender_cpu = raw_smp_processor_id() + 1; |
3322 | #endif | |
3323 | ||
3324 | if (dev->real_num_tx_queues != 1) { | |
3325 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 3326 | |
638b2a69 JP |
3327 | if (ops->ndo_select_queue) |
3328 | queue_index = ops->ndo_select_queue(dev, skb, accel_priv, | |
3329 | __netdev_pick_tx); | |
3330 | else | |
3331 | queue_index = __netdev_pick_tx(dev, skb); | |
3332 | ||
3333 | if (!accel_priv) | |
3334 | queue_index = netdev_cap_txqueue(dev, queue_index); | |
3335 | } | |
3336 | ||
3337 | skb_set_queue_mapping(skb, queue_index); | |
3338 | return netdev_get_tx_queue(dev, queue_index); | |
3339 | } | |
3340 | ||
d29f749e | 3341 | /** |
9d08dd3d | 3342 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3343 | * @skb: buffer to transmit |
9d08dd3d | 3344 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
3345 | * |
3346 | * Queue a buffer for transmission to a network device. The caller must | |
3347 | * have set the device and priority and built the buffer before calling | |
3348 | * this function. The function can be called from an interrupt. | |
3349 | * | |
3350 | * A negative errno code is returned on a failure. A success does not | |
3351 | * guarantee the frame will be transmitted as it may be dropped due | |
3352 | * to congestion or traffic shaping. | |
3353 | * | |
3354 | * ----------------------------------------------------------------------------------- | |
3355 | * I notice this method can also return errors from the queue disciplines, | |
3356 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3357 | * be positive. | |
3358 | * | |
3359 | * Regardless of the return value, the skb is consumed, so it is currently | |
3360 | * difficult to retry a send to this method. (You can bump the ref count | |
3361 | * before sending to hold a reference for retry if you are careful.) | |
3362 | * | |
3363 | * When calling this method, interrupts MUST be enabled. This is because | |
3364 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3365 | * --BLG | |
3366 | */ | |
0a59f3a9 | 3367 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
3368 | { |
3369 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3370 | struct netdev_queue *txq; |
1da177e4 LT |
3371 | struct Qdisc *q; |
3372 | int rc = -ENOMEM; | |
3373 | ||
6d1ccff6 ED |
3374 | skb_reset_mac_header(skb); |
3375 | ||
e7fd2885 WB |
3376 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3377 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3378 | ||
4ec93edb YH |
3379 | /* Disable soft irqs for various locks below. Also |
3380 | * stops preemption for RCU. | |
1da177e4 | 3381 | */ |
4ec93edb | 3382 | rcu_read_lock_bh(); |
1da177e4 | 3383 | |
5bc1421e NH |
3384 | skb_update_prio(skb); |
3385 | ||
1f211a1b DB |
3386 | qdisc_pkt_len_init(skb); |
3387 | #ifdef CONFIG_NET_CLS_ACT | |
8dc07fdb | 3388 | skb->tc_at_ingress = 0; |
1f211a1b DB |
3389 | # ifdef CONFIG_NET_EGRESS |
3390 | if (static_key_false(&egress_needed)) { | |
3391 | skb = sch_handle_egress(skb, &rc, dev); | |
3392 | if (!skb) | |
3393 | goto out; | |
3394 | } | |
3395 | # endif | |
3396 | #endif | |
02875878 ED |
3397 | /* If device/qdisc don't need skb->dst, release it right now while |
3398 | * its hot in this cpu cache. | |
3399 | */ | |
3400 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
3401 | skb_dst_drop(skb); | |
3402 | else | |
3403 | skb_dst_force(skb); | |
3404 | ||
f663dd9a | 3405 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 3406 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 3407 | |
cf66ba58 | 3408 | trace_net_dev_queue(skb); |
1da177e4 | 3409 | if (q->enqueue) { |
bbd8a0d3 | 3410 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 3411 | goto out; |
1da177e4 LT |
3412 | } |
3413 | ||
3414 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 3415 | * loopback, all the sorts of tunnels... |
1da177e4 | 3416 | |
eb13da1a | 3417 | * Really, it is unlikely that netif_tx_lock protection is necessary |
3418 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
3419 | * counters.) | |
3420 | * However, it is possible, that they rely on protection | |
3421 | * made by us here. | |
1da177e4 | 3422 | |
eb13da1a | 3423 | * Check this and shot the lock. It is not prone from deadlocks. |
3424 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
3425 | */ |
3426 | if (dev->flags & IFF_UP) { | |
3427 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
3428 | ||
c773e847 | 3429 | if (txq->xmit_lock_owner != cpu) { |
a70b506e DB |
3430 | if (unlikely(__this_cpu_read(xmit_recursion) > |
3431 | XMIT_RECURSION_LIMIT)) | |
745e20f1 ED |
3432 | goto recursion_alert; |
3433 | ||
1f59533f JDB |
3434 | skb = validate_xmit_skb(skb, dev); |
3435 | if (!skb) | |
d21fd63e | 3436 | goto out; |
1f59533f | 3437 | |
c773e847 | 3438 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 3439 | |
73466498 | 3440 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 3441 | __this_cpu_inc(xmit_recursion); |
ce93718f | 3442 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 3443 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 3444 | if (dev_xmit_complete(rc)) { |
c773e847 | 3445 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
3446 | goto out; |
3447 | } | |
3448 | } | |
c773e847 | 3449 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
3450 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
3451 | dev->name); | |
1da177e4 LT |
3452 | } else { |
3453 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
3454 | * unfortunately |
3455 | */ | |
3456 | recursion_alert: | |
e87cc472 JP |
3457 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
3458 | dev->name); | |
1da177e4 LT |
3459 | } |
3460 | } | |
3461 | ||
3462 | rc = -ENETDOWN; | |
d4828d85 | 3463 | rcu_read_unlock_bh(); |
1da177e4 | 3464 | |
015f0688 | 3465 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 3466 | kfree_skb_list(skb); |
1da177e4 LT |
3467 | return rc; |
3468 | out: | |
d4828d85 | 3469 | rcu_read_unlock_bh(); |
1da177e4 LT |
3470 | return rc; |
3471 | } | |
f663dd9a | 3472 | |
2b4aa3ce | 3473 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
3474 | { |
3475 | return __dev_queue_xmit(skb, NULL); | |
3476 | } | |
2b4aa3ce | 3477 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3478 | |
f663dd9a JW |
3479 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3480 | { | |
3481 | return __dev_queue_xmit(skb, accel_priv); | |
3482 | } | |
3483 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3484 | ||
1da177e4 | 3485 | |
eb13da1a | 3486 | /************************************************************************* |
3487 | * Receiver routines | |
3488 | *************************************************************************/ | |
1da177e4 | 3489 | |
6b2bedc3 | 3490 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3491 | EXPORT_SYMBOL(netdev_max_backlog); |
3492 | ||
3b098e2d | 3493 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 | 3494 | int netdev_budget __read_mostly = 300; |
7acf8a1e | 3495 | unsigned int __read_mostly netdev_budget_usecs = 2000; |
3d48b53f MT |
3496 | int weight_p __read_mostly = 64; /* old backlog weight */ |
3497 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
3498 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
3499 | int dev_rx_weight __read_mostly = 64; | |
3500 | int dev_tx_weight __read_mostly = 64; | |
1da177e4 | 3501 | |
eecfd7c4 ED |
3502 | /* Called with irq disabled */ |
3503 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3504 | struct napi_struct *napi) | |
3505 | { | |
3506 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3507 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3508 | } | |
3509 | ||
bfb564e7 KK |
3510 | #ifdef CONFIG_RPS |
3511 | ||
3512 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3513 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 3514 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
3515 | u32 rps_cpu_mask __read_mostly; |
3516 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 3517 | |
c5905afb | 3518 | struct static_key rps_needed __read_mostly; |
3df97ba8 | 3519 | EXPORT_SYMBOL(rps_needed); |
13bfff25 ED |
3520 | struct static_key rfs_needed __read_mostly; |
3521 | EXPORT_SYMBOL(rfs_needed); | |
adc9300e | 3522 | |
c445477d BH |
3523 | static struct rps_dev_flow * |
3524 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3525 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3526 | { | |
a31196b0 | 3527 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
3528 | #ifdef CONFIG_RFS_ACCEL |
3529 | struct netdev_rx_queue *rxqueue; | |
3530 | struct rps_dev_flow_table *flow_table; | |
3531 | struct rps_dev_flow *old_rflow; | |
3532 | u32 flow_id; | |
3533 | u16 rxq_index; | |
3534 | int rc; | |
3535 | ||
3536 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3537 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3538 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3539 | goto out; |
3540 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3541 | if (rxq_index == skb_get_rx_queue(skb)) | |
3542 | goto out; | |
3543 | ||
3544 | rxqueue = dev->_rx + rxq_index; | |
3545 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3546 | if (!flow_table) | |
3547 | goto out; | |
61b905da | 3548 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3549 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3550 | rxq_index, flow_id); | |
3551 | if (rc < 0) | |
3552 | goto out; | |
3553 | old_rflow = rflow; | |
3554 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3555 | rflow->filter = rc; |
3556 | if (old_rflow->filter == rflow->filter) | |
3557 | old_rflow->filter = RPS_NO_FILTER; | |
3558 | out: | |
3559 | #endif | |
3560 | rflow->last_qtail = | |
09994d1b | 3561 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3562 | } |
3563 | ||
09994d1b | 3564 | rflow->cpu = next_cpu; |
c445477d BH |
3565 | return rflow; |
3566 | } | |
3567 | ||
bfb564e7 KK |
3568 | /* |
3569 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3570 | * CPU from the RPS map of the receiving queue for a given skb. | |
3571 | * rcu_read_lock must be held on entry. | |
3572 | */ | |
3573 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3574 | struct rps_dev_flow **rflowp) | |
3575 | { | |
567e4b79 ED |
3576 | const struct rps_sock_flow_table *sock_flow_table; |
3577 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 3578 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 3579 | struct rps_map *map; |
bfb564e7 | 3580 | int cpu = -1; |
567e4b79 | 3581 | u32 tcpu; |
61b905da | 3582 | u32 hash; |
bfb564e7 KK |
3583 | |
3584 | if (skb_rx_queue_recorded(skb)) { | |
3585 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 3586 | |
62fe0b40 BH |
3587 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3588 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3589 | "%s received packet on queue %u, but number " | |
3590 | "of RX queues is %u\n", | |
3591 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3592 | goto done; |
3593 | } | |
567e4b79 ED |
3594 | rxqueue += index; |
3595 | } | |
bfb564e7 | 3596 | |
567e4b79 ED |
3597 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
3598 | ||
3599 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 3600 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 3601 | if (!flow_table && !map) |
bfb564e7 KK |
3602 | goto done; |
3603 | ||
2d47b459 | 3604 | skb_reset_network_header(skb); |
61b905da TH |
3605 | hash = skb_get_hash(skb); |
3606 | if (!hash) | |
bfb564e7 KK |
3607 | goto done; |
3608 | ||
fec5e652 TH |
3609 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
3610 | if (flow_table && sock_flow_table) { | |
fec5e652 | 3611 | struct rps_dev_flow *rflow; |
567e4b79 ED |
3612 | u32 next_cpu; |
3613 | u32 ident; | |
3614 | ||
3615 | /* First check into global flow table if there is a match */ | |
3616 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
3617 | if ((ident ^ hash) & ~rps_cpu_mask) | |
3618 | goto try_rps; | |
fec5e652 | 3619 | |
567e4b79 ED |
3620 | next_cpu = ident & rps_cpu_mask; |
3621 | ||
3622 | /* OK, now we know there is a match, | |
3623 | * we can look at the local (per receive queue) flow table | |
3624 | */ | |
61b905da | 3625 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3626 | tcpu = rflow->cpu; |
3627 | ||
fec5e652 TH |
3628 | /* |
3629 | * If the desired CPU (where last recvmsg was done) is | |
3630 | * different from current CPU (one in the rx-queue flow | |
3631 | * table entry), switch if one of the following holds: | |
a31196b0 | 3632 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
3633 | * - Current CPU is offline. |
3634 | * - The current CPU's queue tail has advanced beyond the | |
3635 | * last packet that was enqueued using this table entry. | |
3636 | * This guarantees that all previous packets for the flow | |
3637 | * have been dequeued, thus preserving in order delivery. | |
3638 | */ | |
3639 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 3640 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 3641 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
3642 | rflow->last_qtail)) >= 0)) { |
3643 | tcpu = next_cpu; | |
c445477d | 3644 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3645 | } |
c445477d | 3646 | |
a31196b0 | 3647 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
3648 | *rflowp = rflow; |
3649 | cpu = tcpu; | |
3650 | goto done; | |
3651 | } | |
3652 | } | |
3653 | ||
567e4b79 ED |
3654 | try_rps: |
3655 | ||
0a9627f2 | 3656 | if (map) { |
8fc54f68 | 3657 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3658 | if (cpu_online(tcpu)) { |
3659 | cpu = tcpu; | |
3660 | goto done; | |
3661 | } | |
3662 | } | |
3663 | ||
3664 | done: | |
0a9627f2 TH |
3665 | return cpu; |
3666 | } | |
3667 | ||
c445477d BH |
3668 | #ifdef CONFIG_RFS_ACCEL |
3669 | ||
3670 | /** | |
3671 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3672 | * @dev: Device on which the filter was set | |
3673 | * @rxq_index: RX queue index | |
3674 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3675 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3676 | * | |
3677 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3678 | * this function for each installed filter and remove the filters for | |
3679 | * which it returns %true. | |
3680 | */ | |
3681 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3682 | u32 flow_id, u16 filter_id) | |
3683 | { | |
3684 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3685 | struct rps_dev_flow_table *flow_table; | |
3686 | struct rps_dev_flow *rflow; | |
3687 | bool expire = true; | |
a31196b0 | 3688 | unsigned int cpu; |
c445477d BH |
3689 | |
3690 | rcu_read_lock(); | |
3691 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3692 | if (flow_table && flow_id <= flow_table->mask) { | |
3693 | rflow = &flow_table->flows[flow_id]; | |
3694 | cpu = ACCESS_ONCE(rflow->cpu); | |
a31196b0 | 3695 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
3696 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
3697 | rflow->last_qtail) < | |
3698 | (int)(10 * flow_table->mask))) | |
3699 | expire = false; | |
3700 | } | |
3701 | rcu_read_unlock(); | |
3702 | return expire; | |
3703 | } | |
3704 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3705 | ||
3706 | #endif /* CONFIG_RFS_ACCEL */ | |
3707 | ||
0a9627f2 | 3708 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3709 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3710 | { |
e36fa2f7 ED |
3711 | struct softnet_data *sd = data; |
3712 | ||
eecfd7c4 | 3713 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3714 | sd->received_rps++; |
0a9627f2 | 3715 | } |
e36fa2f7 | 3716 | |
fec5e652 | 3717 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3718 | |
e36fa2f7 ED |
3719 | /* |
3720 | * Check if this softnet_data structure is another cpu one | |
3721 | * If yes, queue it to our IPI list and return 1 | |
3722 | * If no, return 0 | |
3723 | */ | |
3724 | static int rps_ipi_queued(struct softnet_data *sd) | |
3725 | { | |
3726 | #ifdef CONFIG_RPS | |
903ceff7 | 3727 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
3728 | |
3729 | if (sd != mysd) { | |
3730 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3731 | mysd->rps_ipi_list = sd; | |
3732 | ||
3733 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3734 | return 1; | |
3735 | } | |
3736 | #endif /* CONFIG_RPS */ | |
3737 | return 0; | |
3738 | } | |
3739 | ||
99bbc707 WB |
3740 | #ifdef CONFIG_NET_FLOW_LIMIT |
3741 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3742 | #endif | |
3743 | ||
3744 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3745 | { | |
3746 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3747 | struct sd_flow_limit *fl; | |
3748 | struct softnet_data *sd; | |
3749 | unsigned int old_flow, new_flow; | |
3750 | ||
3751 | if (qlen < (netdev_max_backlog >> 1)) | |
3752 | return false; | |
3753 | ||
903ceff7 | 3754 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
3755 | |
3756 | rcu_read_lock(); | |
3757 | fl = rcu_dereference(sd->flow_limit); | |
3758 | if (fl) { | |
3958afa1 | 3759 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3760 | old_flow = fl->history[fl->history_head]; |
3761 | fl->history[fl->history_head] = new_flow; | |
3762 | ||
3763 | fl->history_head++; | |
3764 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3765 | ||
3766 | if (likely(fl->buckets[old_flow])) | |
3767 | fl->buckets[old_flow]--; | |
3768 | ||
3769 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3770 | fl->count++; | |
3771 | rcu_read_unlock(); | |
3772 | return true; | |
3773 | } | |
3774 | } | |
3775 | rcu_read_unlock(); | |
3776 | #endif | |
3777 | return false; | |
3778 | } | |
3779 | ||
0a9627f2 TH |
3780 | /* |
3781 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3782 | * queue (may be a remote CPU queue). | |
3783 | */ | |
fec5e652 TH |
3784 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3785 | unsigned int *qtail) | |
0a9627f2 | 3786 | { |
e36fa2f7 | 3787 | struct softnet_data *sd; |
0a9627f2 | 3788 | unsigned long flags; |
99bbc707 | 3789 | unsigned int qlen; |
0a9627f2 | 3790 | |
e36fa2f7 | 3791 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3792 | |
3793 | local_irq_save(flags); | |
0a9627f2 | 3794 | |
e36fa2f7 | 3795 | rps_lock(sd); |
e9e4dd32 JA |
3796 | if (!netif_running(skb->dev)) |
3797 | goto drop; | |
99bbc707 WB |
3798 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3799 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 3800 | if (qlen) { |
0a9627f2 | 3801 | enqueue: |
e36fa2f7 | 3802 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3803 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3804 | rps_unlock(sd); |
152102c7 | 3805 | local_irq_restore(flags); |
0a9627f2 TH |
3806 | return NET_RX_SUCCESS; |
3807 | } | |
3808 | ||
ebda37c2 ED |
3809 | /* Schedule NAPI for backlog device |
3810 | * We can use non atomic operation since we own the queue lock | |
3811 | */ | |
3812 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3813 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3814 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3815 | } |
3816 | goto enqueue; | |
3817 | } | |
3818 | ||
e9e4dd32 | 3819 | drop: |
dee42870 | 3820 | sd->dropped++; |
e36fa2f7 | 3821 | rps_unlock(sd); |
0a9627f2 | 3822 | |
0a9627f2 TH |
3823 | local_irq_restore(flags); |
3824 | ||
caf586e5 | 3825 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3826 | kfree_skb(skb); |
3827 | return NET_RX_DROP; | |
3828 | } | |
1da177e4 | 3829 | |
ae78dbfa | 3830 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3831 | { |
b0e28f1e | 3832 | int ret; |
1da177e4 | 3833 | |
588f0330 | 3834 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3835 | |
cf66ba58 | 3836 | trace_netif_rx(skb); |
df334545 | 3837 | #ifdef CONFIG_RPS |
c5905afb | 3838 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3839 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3840 | int cpu; |
3841 | ||
cece1945 | 3842 | preempt_disable(); |
b0e28f1e | 3843 | rcu_read_lock(); |
fec5e652 TH |
3844 | |
3845 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3846 | if (cpu < 0) |
3847 | cpu = smp_processor_id(); | |
fec5e652 TH |
3848 | |
3849 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3850 | ||
b0e28f1e | 3851 | rcu_read_unlock(); |
cece1945 | 3852 | preempt_enable(); |
adc9300e ED |
3853 | } else |
3854 | #endif | |
fec5e652 TH |
3855 | { |
3856 | unsigned int qtail; | |
f4563a75 | 3857 | |
fec5e652 TH |
3858 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); |
3859 | put_cpu(); | |
3860 | } | |
b0e28f1e | 3861 | return ret; |
1da177e4 | 3862 | } |
ae78dbfa BH |
3863 | |
3864 | /** | |
3865 | * netif_rx - post buffer to the network code | |
3866 | * @skb: buffer to post | |
3867 | * | |
3868 | * This function receives a packet from a device driver and queues it for | |
3869 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3870 | * may be dropped during processing for congestion control or by the | |
3871 | * protocol layers. | |
3872 | * | |
3873 | * return values: | |
3874 | * NET_RX_SUCCESS (no congestion) | |
3875 | * NET_RX_DROP (packet was dropped) | |
3876 | * | |
3877 | */ | |
3878 | ||
3879 | int netif_rx(struct sk_buff *skb) | |
3880 | { | |
3881 | trace_netif_rx_entry(skb); | |
3882 | ||
3883 | return netif_rx_internal(skb); | |
3884 | } | |
d1b19dff | 3885 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3886 | |
3887 | int netif_rx_ni(struct sk_buff *skb) | |
3888 | { | |
3889 | int err; | |
3890 | ||
ae78dbfa BH |
3891 | trace_netif_rx_ni_entry(skb); |
3892 | ||
1da177e4 | 3893 | preempt_disable(); |
ae78dbfa | 3894 | err = netif_rx_internal(skb); |
1da177e4 LT |
3895 | if (local_softirq_pending()) |
3896 | do_softirq(); | |
3897 | preempt_enable(); | |
3898 | ||
3899 | return err; | |
3900 | } | |
1da177e4 LT |
3901 | EXPORT_SYMBOL(netif_rx_ni); |
3902 | ||
0766f788 | 3903 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 3904 | { |
903ceff7 | 3905 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
3906 | |
3907 | if (sd->completion_queue) { | |
3908 | struct sk_buff *clist; | |
3909 | ||
3910 | local_irq_disable(); | |
3911 | clist = sd->completion_queue; | |
3912 | sd->completion_queue = NULL; | |
3913 | local_irq_enable(); | |
3914 | ||
3915 | while (clist) { | |
3916 | struct sk_buff *skb = clist; | |
f4563a75 | 3917 | |
1da177e4 LT |
3918 | clist = clist->next; |
3919 | ||
547b792c | 3920 | WARN_ON(atomic_read(&skb->users)); |
e6247027 ED |
3921 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
3922 | trace_consume_skb(skb); | |
3923 | else | |
3924 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
3925 | |
3926 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
3927 | __kfree_skb(skb); | |
3928 | else | |
3929 | __kfree_skb_defer(skb); | |
1da177e4 | 3930 | } |
15fad714 JDB |
3931 | |
3932 | __kfree_skb_flush(); | |
1da177e4 LT |
3933 | } |
3934 | ||
3935 | if (sd->output_queue) { | |
37437bb2 | 3936 | struct Qdisc *head; |
1da177e4 LT |
3937 | |
3938 | local_irq_disable(); | |
3939 | head = sd->output_queue; | |
3940 | sd->output_queue = NULL; | |
a9cbd588 | 3941 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3942 | local_irq_enable(); |
3943 | ||
3944 | while (head) { | |
37437bb2 DM |
3945 | struct Qdisc *q = head; |
3946 | spinlock_t *root_lock; | |
3947 | ||
1da177e4 LT |
3948 | head = head->next_sched; |
3949 | ||
5fb66229 | 3950 | root_lock = qdisc_lock(q); |
3bcb846c ED |
3951 | spin_lock(root_lock); |
3952 | /* We need to make sure head->next_sched is read | |
3953 | * before clearing __QDISC_STATE_SCHED | |
3954 | */ | |
3955 | smp_mb__before_atomic(); | |
3956 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
3957 | qdisc_run(q); | |
3958 | spin_unlock(root_lock); | |
1da177e4 LT |
3959 | } |
3960 | } | |
3961 | } | |
3962 | ||
181402a5 | 3963 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
3964 | /* This hook is defined here for ATM LANE */ |
3965 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3966 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3967 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3968 | #endif |
1da177e4 | 3969 | |
1f211a1b DB |
3970 | static inline struct sk_buff * |
3971 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
3972 | struct net_device *orig_dev) | |
f697c3e8 | 3973 | { |
e7582bab | 3974 | #ifdef CONFIG_NET_CLS_ACT |
d2788d34 DB |
3975 | struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list); |
3976 | struct tcf_result cl_res; | |
24824a09 | 3977 | |
c9e99fd0 DB |
3978 | /* If there's at least one ingress present somewhere (so |
3979 | * we get here via enabled static key), remaining devices | |
3980 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 3981 | * out here. |
c9e99fd0 | 3982 | */ |
d2788d34 | 3983 | if (!cl) |
4577139b | 3984 | return skb; |
f697c3e8 HX |
3985 | if (*pt_prev) { |
3986 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3987 | *pt_prev = NULL; | |
1da177e4 LT |
3988 | } |
3989 | ||
3365495c | 3990 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
8dc07fdb | 3991 | skb->tc_at_ingress = 1; |
24ea591d | 3992 | qdisc_bstats_cpu_update(cl->q, skb); |
c9e99fd0 | 3993 | |
87d83093 | 3994 | switch (tcf_classify(skb, cl, &cl_res, false)) { |
d2788d34 DB |
3995 | case TC_ACT_OK: |
3996 | case TC_ACT_RECLASSIFY: | |
3997 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3998 | break; | |
3999 | case TC_ACT_SHOT: | |
24ea591d | 4000 | qdisc_qstats_cpu_drop(cl->q); |
8a3a4c6e ED |
4001 | kfree_skb(skb); |
4002 | return NULL; | |
d2788d34 DB |
4003 | case TC_ACT_STOLEN: |
4004 | case TC_ACT_QUEUED: | |
8a3a4c6e | 4005 | consume_skb(skb); |
d2788d34 | 4006 | return NULL; |
27b29f63 AS |
4007 | case TC_ACT_REDIRECT: |
4008 | /* skb_mac_header check was done by cls/act_bpf, so | |
4009 | * we can safely push the L2 header back before | |
4010 | * redirecting to another netdev | |
4011 | */ | |
4012 | __skb_push(skb, skb->mac_len); | |
4013 | skb_do_redirect(skb); | |
4014 | return NULL; | |
d2788d34 DB |
4015 | default: |
4016 | break; | |
f697c3e8 | 4017 | } |
e7582bab | 4018 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
4019 | return skb; |
4020 | } | |
1da177e4 | 4021 | |
24b27fc4 MB |
4022 | /** |
4023 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
4024 | * @dev: device to check | |
4025 | * | |
4026 | * Check if a receive handler is already registered for a given device. | |
4027 | * Return true if there one. | |
4028 | * | |
4029 | * The caller must hold the rtnl_mutex. | |
4030 | */ | |
4031 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
4032 | { | |
4033 | ASSERT_RTNL(); | |
4034 | return dev && rtnl_dereference(dev->rx_handler); | |
4035 | } | |
4036 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
4037 | ||
ab95bfe0 JP |
4038 | /** |
4039 | * netdev_rx_handler_register - register receive handler | |
4040 | * @dev: device to register a handler for | |
4041 | * @rx_handler: receive handler to register | |
93e2c32b | 4042 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 4043 | * |
e227867f | 4044 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
4045 | * called from __netif_receive_skb. A negative errno code is returned |
4046 | * on a failure. | |
4047 | * | |
4048 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
4049 | * |
4050 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
4051 | */ |
4052 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
4053 | rx_handler_func_t *rx_handler, |
4054 | void *rx_handler_data) | |
ab95bfe0 | 4055 | { |
1b7cd004 | 4056 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
4057 | return -EBUSY; |
4058 | ||
00cfec37 | 4059 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 4060 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
4061 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
4062 | ||
4063 | return 0; | |
4064 | } | |
4065 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
4066 | ||
4067 | /** | |
4068 | * netdev_rx_handler_unregister - unregister receive handler | |
4069 | * @dev: device to unregister a handler from | |
4070 | * | |
166ec369 | 4071 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
4072 | * |
4073 | * The caller must hold the rtnl_mutex. | |
4074 | */ | |
4075 | void netdev_rx_handler_unregister(struct net_device *dev) | |
4076 | { | |
4077 | ||
4078 | ASSERT_RTNL(); | |
a9b3cd7f | 4079 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
4080 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
4081 | * section has a guarantee to see a non NULL rx_handler_data | |
4082 | * as well. | |
4083 | */ | |
4084 | synchronize_net(); | |
a9b3cd7f | 4085 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
4086 | } |
4087 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
4088 | ||
b4b9e355 MG |
4089 | /* |
4090 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
4091 | * the special handling of PFMEMALLOC skbs. | |
4092 | */ | |
4093 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
4094 | { | |
4095 | switch (skb->protocol) { | |
2b8837ae JP |
4096 | case htons(ETH_P_ARP): |
4097 | case htons(ETH_P_IP): | |
4098 | case htons(ETH_P_IPV6): | |
4099 | case htons(ETH_P_8021Q): | |
4100 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
4101 | return true; |
4102 | default: | |
4103 | return false; | |
4104 | } | |
4105 | } | |
4106 | ||
e687ad60 PN |
4107 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
4108 | int *ret, struct net_device *orig_dev) | |
4109 | { | |
e7582bab | 4110 | #ifdef CONFIG_NETFILTER_INGRESS |
e687ad60 | 4111 | if (nf_hook_ingress_active(skb)) { |
2c1e2703 AC |
4112 | int ingress_retval; |
4113 | ||
e687ad60 PN |
4114 | if (*pt_prev) { |
4115 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4116 | *pt_prev = NULL; | |
4117 | } | |
4118 | ||
2c1e2703 AC |
4119 | rcu_read_lock(); |
4120 | ingress_retval = nf_hook_ingress(skb); | |
4121 | rcu_read_unlock(); | |
4122 | return ingress_retval; | |
e687ad60 | 4123 | } |
e7582bab | 4124 | #endif /* CONFIG_NETFILTER_INGRESS */ |
e687ad60 PN |
4125 | return 0; |
4126 | } | |
e687ad60 | 4127 | |
9754e293 | 4128 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
4129 | { |
4130 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 4131 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 4132 | struct net_device *orig_dev; |
8a4eb573 | 4133 | bool deliver_exact = false; |
1da177e4 | 4134 | int ret = NET_RX_DROP; |
252e3346 | 4135 | __be16 type; |
1da177e4 | 4136 | |
588f0330 | 4137 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 4138 | |
cf66ba58 | 4139 | trace_netif_receive_skb(skb); |
9b22ea56 | 4140 | |
cc9bd5ce | 4141 | orig_dev = skb->dev; |
8f903c70 | 4142 | |
c1d2bbe1 | 4143 | skb_reset_network_header(skb); |
fda55eca ED |
4144 | if (!skb_transport_header_was_set(skb)) |
4145 | skb_reset_transport_header(skb); | |
0b5c9db1 | 4146 | skb_reset_mac_len(skb); |
1da177e4 LT |
4147 | |
4148 | pt_prev = NULL; | |
4149 | ||
63d8ea7f | 4150 | another_round: |
b6858177 | 4151 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
4152 | |
4153 | __this_cpu_inc(softnet_data.processed); | |
4154 | ||
8ad227ff PM |
4155 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
4156 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 4157 | skb = skb_vlan_untag(skb); |
bcc6d479 | 4158 | if (unlikely(!skb)) |
2c17d27c | 4159 | goto out; |
bcc6d479 JP |
4160 | } |
4161 | ||
e7246e12 WB |
4162 | if (skb_skip_tc_classify(skb)) |
4163 | goto skip_classify; | |
1da177e4 | 4164 | |
9754e293 | 4165 | if (pfmemalloc) |
b4b9e355 MG |
4166 | goto skip_taps; |
4167 | ||
1da177e4 | 4168 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
4169 | if (pt_prev) |
4170 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4171 | pt_prev = ptype; | |
4172 | } | |
4173 | ||
4174 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
4175 | if (pt_prev) | |
4176 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4177 | pt_prev = ptype; | |
1da177e4 LT |
4178 | } |
4179 | ||
b4b9e355 | 4180 | skip_taps: |
1cf51900 | 4181 | #ifdef CONFIG_NET_INGRESS |
4577139b | 4182 | if (static_key_false(&ingress_needed)) { |
1f211a1b | 4183 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); |
4577139b | 4184 | if (!skb) |
2c17d27c | 4185 | goto out; |
e687ad60 PN |
4186 | |
4187 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 4188 | goto out; |
4577139b | 4189 | } |
1cf51900 | 4190 | #endif |
a5135bcf | 4191 | skb_reset_tc(skb); |
e7246e12 | 4192 | skip_classify: |
9754e293 | 4193 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
4194 | goto drop; |
4195 | ||
df8a39de | 4196 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
4197 | if (pt_prev) { |
4198 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4199 | pt_prev = NULL; | |
4200 | } | |
48cc32d3 | 4201 | if (vlan_do_receive(&skb)) |
2425717b JF |
4202 | goto another_round; |
4203 | else if (unlikely(!skb)) | |
2c17d27c | 4204 | goto out; |
2425717b JF |
4205 | } |
4206 | ||
48cc32d3 | 4207 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
4208 | if (rx_handler) { |
4209 | if (pt_prev) { | |
4210 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4211 | pt_prev = NULL; | |
4212 | } | |
8a4eb573 JP |
4213 | switch (rx_handler(&skb)) { |
4214 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 4215 | ret = NET_RX_SUCCESS; |
2c17d27c | 4216 | goto out; |
8a4eb573 | 4217 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 4218 | goto another_round; |
8a4eb573 JP |
4219 | case RX_HANDLER_EXACT: |
4220 | deliver_exact = true; | |
4221 | case RX_HANDLER_PASS: | |
4222 | break; | |
4223 | default: | |
4224 | BUG(); | |
4225 | } | |
ab95bfe0 | 4226 | } |
1da177e4 | 4227 | |
df8a39de JP |
4228 | if (unlikely(skb_vlan_tag_present(skb))) { |
4229 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
4230 | skb->pkt_type = PACKET_OTHERHOST; |
4231 | /* Note: we might in the future use prio bits | |
4232 | * and set skb->priority like in vlan_do_receive() | |
4233 | * For the time being, just ignore Priority Code Point | |
4234 | */ | |
4235 | skb->vlan_tci = 0; | |
4236 | } | |
48cc32d3 | 4237 | |
7866a621 SN |
4238 | type = skb->protocol; |
4239 | ||
63d8ea7f | 4240 | /* deliver only exact match when indicated */ |
7866a621 SN |
4241 | if (likely(!deliver_exact)) { |
4242 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4243 | &ptype_base[ntohs(type) & | |
4244 | PTYPE_HASH_MASK]); | |
4245 | } | |
1f3c8804 | 4246 | |
7866a621 SN |
4247 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
4248 | &orig_dev->ptype_specific); | |
4249 | ||
4250 | if (unlikely(skb->dev != orig_dev)) { | |
4251 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4252 | &skb->dev->ptype_specific); | |
1da177e4 LT |
4253 | } |
4254 | ||
4255 | if (pt_prev) { | |
1080e512 | 4256 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 4257 | goto drop; |
1080e512 MT |
4258 | else |
4259 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 4260 | } else { |
b4b9e355 | 4261 | drop: |
6e7333d3 JW |
4262 | if (!deliver_exact) |
4263 | atomic_long_inc(&skb->dev->rx_dropped); | |
4264 | else | |
4265 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
4266 | kfree_skb(skb); |
4267 | /* Jamal, now you will not able to escape explaining | |
4268 | * me how you were going to use this. :-) | |
4269 | */ | |
4270 | ret = NET_RX_DROP; | |
4271 | } | |
4272 | ||
2c17d27c | 4273 | out: |
9754e293 DM |
4274 | return ret; |
4275 | } | |
4276 | ||
4277 | static int __netif_receive_skb(struct sk_buff *skb) | |
4278 | { | |
4279 | int ret; | |
4280 | ||
4281 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 4282 | unsigned int noreclaim_flag; |
9754e293 DM |
4283 | |
4284 | /* | |
4285 | * PFMEMALLOC skbs are special, they should | |
4286 | * - be delivered to SOCK_MEMALLOC sockets only | |
4287 | * - stay away from userspace | |
4288 | * - have bounded memory usage | |
4289 | * | |
4290 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
4291 | * context down to all allocation sites. | |
4292 | */ | |
f1083048 | 4293 | noreclaim_flag = memalloc_noreclaim_save(); |
9754e293 | 4294 | ret = __netif_receive_skb_core(skb, true); |
f1083048 | 4295 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 DM |
4296 | } else |
4297 | ret = __netif_receive_skb_core(skb, false); | |
4298 | ||
1da177e4 LT |
4299 | return ret; |
4300 | } | |
0a9627f2 | 4301 | |
b5cdae32 DM |
4302 | static struct static_key generic_xdp_needed __read_mostly; |
4303 | ||
4304 | static int generic_xdp_install(struct net_device *dev, struct netdev_xdp *xdp) | |
4305 | { | |
4306 | struct bpf_prog *new = xdp->prog; | |
4307 | int ret = 0; | |
4308 | ||
4309 | switch (xdp->command) { | |
4310 | case XDP_SETUP_PROG: { | |
4311 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); | |
4312 | ||
4313 | rcu_assign_pointer(dev->xdp_prog, new); | |
4314 | if (old) | |
4315 | bpf_prog_put(old); | |
4316 | ||
4317 | if (old && !new) { | |
4318 | static_key_slow_dec(&generic_xdp_needed); | |
4319 | } else if (new && !old) { | |
4320 | static_key_slow_inc(&generic_xdp_needed); | |
4321 | dev_disable_lro(dev); | |
4322 | } | |
4323 | break; | |
4324 | } | |
4325 | ||
4326 | case XDP_QUERY_PROG: | |
4327 | xdp->prog_attached = !!rcu_access_pointer(dev->xdp_prog); | |
4328 | break; | |
4329 | ||
4330 | default: | |
4331 | ret = -EINVAL; | |
4332 | break; | |
4333 | } | |
4334 | ||
4335 | return ret; | |
4336 | } | |
4337 | ||
4338 | static u32 netif_receive_generic_xdp(struct sk_buff *skb, | |
4339 | struct bpf_prog *xdp_prog) | |
4340 | { | |
4341 | struct xdp_buff xdp; | |
4342 | u32 act = XDP_DROP; | |
4343 | void *orig_data; | |
4344 | int hlen, off; | |
4345 | u32 mac_len; | |
4346 | ||
4347 | /* Reinjected packets coming from act_mirred or similar should | |
4348 | * not get XDP generic processing. | |
4349 | */ | |
4350 | if (skb_cloned(skb)) | |
4351 | return XDP_PASS; | |
4352 | ||
4353 | if (skb_linearize(skb)) | |
4354 | goto do_drop; | |
4355 | ||
4356 | /* The XDP program wants to see the packet starting at the MAC | |
4357 | * header. | |
4358 | */ | |
4359 | mac_len = skb->data - skb_mac_header(skb); | |
4360 | hlen = skb_headlen(skb) + mac_len; | |
4361 | xdp.data = skb->data - mac_len; | |
4362 | xdp.data_end = xdp.data + hlen; | |
4363 | xdp.data_hard_start = skb->data - skb_headroom(skb); | |
4364 | orig_data = xdp.data; | |
4365 | ||
4366 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
4367 | ||
4368 | off = xdp.data - orig_data; | |
4369 | if (off > 0) | |
4370 | __skb_pull(skb, off); | |
4371 | else if (off < 0) | |
4372 | __skb_push(skb, -off); | |
4373 | ||
4374 | switch (act) { | |
4375 | case XDP_TX: | |
4376 | __skb_push(skb, mac_len); | |
4377 | /* fall through */ | |
4378 | case XDP_PASS: | |
4379 | break; | |
4380 | ||
4381 | default: | |
4382 | bpf_warn_invalid_xdp_action(act); | |
4383 | /* fall through */ | |
4384 | case XDP_ABORTED: | |
4385 | trace_xdp_exception(skb->dev, xdp_prog, act); | |
4386 | /* fall through */ | |
4387 | case XDP_DROP: | |
4388 | do_drop: | |
4389 | kfree_skb(skb); | |
4390 | break; | |
4391 | } | |
4392 | ||
4393 | return act; | |
4394 | } | |
4395 | ||
4396 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
4397 | * network taps in order to match in-driver-XDP behavior. | |
4398 | */ | |
4399 | static void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) | |
4400 | { | |
4401 | struct net_device *dev = skb->dev; | |
4402 | struct netdev_queue *txq; | |
4403 | bool free_skb = true; | |
4404 | int cpu, rc; | |
4405 | ||
4406 | txq = netdev_pick_tx(dev, skb, NULL); | |
4407 | cpu = smp_processor_id(); | |
4408 | HARD_TX_LOCK(dev, txq, cpu); | |
4409 | if (!netif_xmit_stopped(txq)) { | |
4410 | rc = netdev_start_xmit(skb, dev, txq, 0); | |
4411 | if (dev_xmit_complete(rc)) | |
4412 | free_skb = false; | |
4413 | } | |
4414 | HARD_TX_UNLOCK(dev, txq); | |
4415 | if (free_skb) { | |
4416 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
4417 | kfree_skb(skb); | |
4418 | } | |
4419 | } | |
4420 | ||
ae78dbfa | 4421 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 4422 | { |
2c17d27c JA |
4423 | int ret; |
4424 | ||
588f0330 | 4425 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 4426 | |
c1f19b51 RC |
4427 | if (skb_defer_rx_timestamp(skb)) |
4428 | return NET_RX_SUCCESS; | |
4429 | ||
2c17d27c JA |
4430 | rcu_read_lock(); |
4431 | ||
b5cdae32 DM |
4432 | if (static_key_false(&generic_xdp_needed)) { |
4433 | struct bpf_prog *xdp_prog = rcu_dereference(skb->dev->xdp_prog); | |
4434 | ||
4435 | if (xdp_prog) { | |
4436 | u32 act = netif_receive_generic_xdp(skb, xdp_prog); | |
4437 | ||
4438 | if (act != XDP_PASS) { | |
4439 | rcu_read_unlock(); | |
4440 | if (act == XDP_TX) | |
4441 | generic_xdp_tx(skb, xdp_prog); | |
4442 | return NET_RX_DROP; | |
4443 | } | |
4444 | } | |
4445 | } | |
4446 | ||
df334545 | 4447 | #ifdef CONFIG_RPS |
c5905afb | 4448 | if (static_key_false(&rps_needed)) { |
3b098e2d | 4449 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 4450 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 4451 | |
3b098e2d ED |
4452 | if (cpu >= 0) { |
4453 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4454 | rcu_read_unlock(); | |
adc9300e | 4455 | return ret; |
3b098e2d | 4456 | } |
fec5e652 | 4457 | } |
1e94d72f | 4458 | #endif |
2c17d27c JA |
4459 | ret = __netif_receive_skb(skb); |
4460 | rcu_read_unlock(); | |
4461 | return ret; | |
0a9627f2 | 4462 | } |
ae78dbfa BH |
4463 | |
4464 | /** | |
4465 | * netif_receive_skb - process receive buffer from network | |
4466 | * @skb: buffer to process | |
4467 | * | |
4468 | * netif_receive_skb() is the main receive data processing function. | |
4469 | * It always succeeds. The buffer may be dropped during processing | |
4470 | * for congestion control or by the protocol layers. | |
4471 | * | |
4472 | * This function may only be called from softirq context and interrupts | |
4473 | * should be enabled. | |
4474 | * | |
4475 | * Return values (usually ignored): | |
4476 | * NET_RX_SUCCESS: no congestion | |
4477 | * NET_RX_DROP: packet was dropped | |
4478 | */ | |
04eb4489 | 4479 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa BH |
4480 | { |
4481 | trace_netif_receive_skb_entry(skb); | |
4482 | ||
4483 | return netif_receive_skb_internal(skb); | |
4484 | } | |
04eb4489 | 4485 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 4486 | |
41852497 | 4487 | DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
4488 | |
4489 | /* Network device is going away, flush any packets still pending */ | |
4490 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 4491 | { |
6e583ce5 | 4492 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
4493 | struct softnet_data *sd; |
4494 | ||
4495 | local_bh_disable(); | |
4496 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 4497 | |
145dd5f9 | 4498 | local_irq_disable(); |
e36fa2f7 | 4499 | rps_lock(sd); |
6e7676c1 | 4500 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 4501 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 4502 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 4503 | kfree_skb(skb); |
76cc8b13 | 4504 | input_queue_head_incr(sd); |
6e583ce5 | 4505 | } |
6e7676c1 | 4506 | } |
e36fa2f7 | 4507 | rps_unlock(sd); |
145dd5f9 | 4508 | local_irq_enable(); |
6e7676c1 CG |
4509 | |
4510 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 4511 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
4512 | __skb_unlink(skb, &sd->process_queue); |
4513 | kfree_skb(skb); | |
76cc8b13 | 4514 | input_queue_head_incr(sd); |
6e7676c1 CG |
4515 | } |
4516 | } | |
145dd5f9 PA |
4517 | local_bh_enable(); |
4518 | } | |
4519 | ||
41852497 | 4520 | static void flush_all_backlogs(void) |
145dd5f9 PA |
4521 | { |
4522 | unsigned int cpu; | |
4523 | ||
4524 | get_online_cpus(); | |
4525 | ||
41852497 ED |
4526 | for_each_online_cpu(cpu) |
4527 | queue_work_on(cpu, system_highpri_wq, | |
4528 | per_cpu_ptr(&flush_works, cpu)); | |
145dd5f9 PA |
4529 | |
4530 | for_each_online_cpu(cpu) | |
41852497 | 4531 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 PA |
4532 | |
4533 | put_online_cpus(); | |
6e583ce5 SH |
4534 | } |
4535 | ||
d565b0a1 HX |
4536 | static int napi_gro_complete(struct sk_buff *skb) |
4537 | { | |
22061d80 | 4538 | struct packet_offload *ptype; |
d565b0a1 | 4539 | __be16 type = skb->protocol; |
22061d80 | 4540 | struct list_head *head = &offload_base; |
d565b0a1 HX |
4541 | int err = -ENOENT; |
4542 | ||
c3c7c254 ED |
4543 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
4544 | ||
fc59f9a3 HX |
4545 | if (NAPI_GRO_CB(skb)->count == 1) { |
4546 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 4547 | goto out; |
fc59f9a3 | 4548 | } |
d565b0a1 HX |
4549 | |
4550 | rcu_read_lock(); | |
4551 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4552 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
4553 | continue; |
4554 | ||
299603e8 | 4555 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
4556 | break; |
4557 | } | |
4558 | rcu_read_unlock(); | |
4559 | ||
4560 | if (err) { | |
4561 | WARN_ON(&ptype->list == head); | |
4562 | kfree_skb(skb); | |
4563 | return NET_RX_SUCCESS; | |
4564 | } | |
4565 | ||
4566 | out: | |
ae78dbfa | 4567 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
4568 | } |
4569 | ||
2e71a6f8 ED |
4570 | /* napi->gro_list contains packets ordered by age. |
4571 | * youngest packets at the head of it. | |
4572 | * Complete skbs in reverse order to reduce latencies. | |
4573 | */ | |
4574 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 4575 | { |
2e71a6f8 | 4576 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 4577 | |
2e71a6f8 ED |
4578 | /* scan list and build reverse chain */ |
4579 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
4580 | skb->prev = prev; | |
4581 | prev = skb; | |
4582 | } | |
4583 | ||
4584 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 4585 | skb->next = NULL; |
2e71a6f8 ED |
4586 | |
4587 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
4588 | return; | |
4589 | ||
4590 | prev = skb->prev; | |
d565b0a1 | 4591 | napi_gro_complete(skb); |
2e71a6f8 | 4592 | napi->gro_count--; |
d565b0a1 HX |
4593 | } |
4594 | ||
4595 | napi->gro_list = NULL; | |
4596 | } | |
86cac58b | 4597 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 4598 | |
89c5fa33 ED |
4599 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
4600 | { | |
4601 | struct sk_buff *p; | |
4602 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 4603 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
4604 | |
4605 | for (p = napi->gro_list; p; p = p->next) { | |
4606 | unsigned long diffs; | |
4607 | ||
0b4cec8c TH |
4608 | NAPI_GRO_CB(p)->flush = 0; |
4609 | ||
4610 | if (hash != skb_get_hash_raw(p)) { | |
4611 | NAPI_GRO_CB(p)->same_flow = 0; | |
4612 | continue; | |
4613 | } | |
4614 | ||
89c5fa33 ED |
4615 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
4616 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
ce87fc6c | 4617 | diffs |= skb_metadata_dst_cmp(p, skb); |
89c5fa33 ED |
4618 | if (maclen == ETH_HLEN) |
4619 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 4620 | skb_mac_header(skb)); |
89c5fa33 ED |
4621 | else if (!diffs) |
4622 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 4623 | skb_mac_header(skb), |
89c5fa33 ED |
4624 | maclen); |
4625 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
4626 | } |
4627 | } | |
4628 | ||
299603e8 JC |
4629 | static void skb_gro_reset_offset(struct sk_buff *skb) |
4630 | { | |
4631 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4632 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
4633 | ||
4634 | NAPI_GRO_CB(skb)->data_offset = 0; | |
4635 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
4636 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
4637 | ||
4638 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
4639 | pinfo->nr_frags && | |
4640 | !PageHighMem(skb_frag_page(frag0))) { | |
4641 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
7cfd5fd5 ED |
4642 | NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, |
4643 | skb_frag_size(frag0), | |
4644 | skb->end - skb->tail); | |
89c5fa33 ED |
4645 | } |
4646 | } | |
4647 | ||
a50e233c ED |
4648 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
4649 | { | |
4650 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4651 | ||
4652 | BUG_ON(skb->end - skb->tail < grow); | |
4653 | ||
4654 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
4655 | ||
4656 | skb->data_len -= grow; | |
4657 | skb->tail += grow; | |
4658 | ||
4659 | pinfo->frags[0].page_offset += grow; | |
4660 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
4661 | ||
4662 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
4663 | skb_frag_unref(skb, 0); | |
4664 | memmove(pinfo->frags, pinfo->frags + 1, | |
4665 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
4666 | } | |
4667 | } | |
4668 | ||
bb728820 | 4669 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
4670 | { |
4671 | struct sk_buff **pp = NULL; | |
22061d80 | 4672 | struct packet_offload *ptype; |
d565b0a1 | 4673 | __be16 type = skb->protocol; |
22061d80 | 4674 | struct list_head *head = &offload_base; |
0da2afd5 | 4675 | int same_flow; |
5b252f0c | 4676 | enum gro_result ret; |
a50e233c | 4677 | int grow; |
d565b0a1 | 4678 | |
b5cdae32 | 4679 | if (netif_elide_gro(skb->dev)) |
d565b0a1 HX |
4680 | goto normal; |
4681 | ||
89c5fa33 ED |
4682 | gro_list_prepare(napi, skb); |
4683 | ||
d565b0a1 HX |
4684 | rcu_read_lock(); |
4685 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4686 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4687 | continue; |
4688 | ||
86911732 | 4689 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4690 | skb_reset_mac_len(skb); |
d565b0a1 | 4691 | NAPI_GRO_CB(skb)->same_flow = 0; |
d61d072e | 4692 | NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); |
5d38a079 | 4693 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 4694 | NAPI_GRO_CB(skb)->encap_mark = 0; |
fcd91dd4 | 4695 | NAPI_GRO_CB(skb)->recursion_counter = 0; |
a0ca153f | 4696 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 4697 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 4698 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 4699 | |
662880f4 TH |
4700 | /* Setup for GRO checksum validation */ |
4701 | switch (skb->ip_summed) { | |
4702 | case CHECKSUM_COMPLETE: | |
4703 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4704 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4705 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4706 | break; | |
4707 | case CHECKSUM_UNNECESSARY: | |
4708 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4709 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4710 | break; | |
4711 | default: | |
4712 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4713 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4714 | } | |
d565b0a1 | 4715 | |
f191a1d1 | 4716 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4717 | break; |
4718 | } | |
4719 | rcu_read_unlock(); | |
4720 | ||
4721 | if (&ptype->list == head) | |
4722 | goto normal; | |
4723 | ||
25393d3f SK |
4724 | if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) { |
4725 | ret = GRO_CONSUMED; | |
4726 | goto ok; | |
4727 | } | |
4728 | ||
0da2afd5 | 4729 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4730 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4731 | |
d565b0a1 HX |
4732 | if (pp) { |
4733 | struct sk_buff *nskb = *pp; | |
4734 | ||
4735 | *pp = nskb->next; | |
4736 | nskb->next = NULL; | |
4737 | napi_gro_complete(nskb); | |
4ae5544f | 4738 | napi->gro_count--; |
d565b0a1 HX |
4739 | } |
4740 | ||
0da2afd5 | 4741 | if (same_flow) |
d565b0a1 HX |
4742 | goto ok; |
4743 | ||
600adc18 | 4744 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4745 | goto normal; |
d565b0a1 | 4746 | |
600adc18 ED |
4747 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4748 | struct sk_buff *nskb = napi->gro_list; | |
4749 | ||
4750 | /* locate the end of the list to select the 'oldest' flow */ | |
4751 | while (nskb->next) { | |
4752 | pp = &nskb->next; | |
4753 | nskb = *pp; | |
4754 | } | |
4755 | *pp = NULL; | |
4756 | nskb->next = NULL; | |
4757 | napi_gro_complete(nskb); | |
4758 | } else { | |
4759 | napi->gro_count++; | |
4760 | } | |
d565b0a1 | 4761 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4762 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4763 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4764 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4765 | skb->next = napi->gro_list; |
4766 | napi->gro_list = skb; | |
5d0d9be8 | 4767 | ret = GRO_HELD; |
d565b0a1 | 4768 | |
ad0f9904 | 4769 | pull: |
a50e233c ED |
4770 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4771 | if (grow > 0) | |
4772 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4773 | ok: |
5d0d9be8 | 4774 | return ret; |
d565b0a1 HX |
4775 | |
4776 | normal: | |
ad0f9904 HX |
4777 | ret = GRO_NORMAL; |
4778 | goto pull; | |
5d38a079 | 4779 | } |
96e93eab | 4780 | |
bf5a755f JC |
4781 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4782 | { | |
4783 | struct list_head *offload_head = &offload_base; | |
4784 | struct packet_offload *ptype; | |
4785 | ||
4786 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4787 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4788 | continue; | |
4789 | return ptype; | |
4790 | } | |
4791 | return NULL; | |
4792 | } | |
e27a2f83 | 4793 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4794 | |
4795 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4796 | { | |
4797 | struct list_head *offload_head = &offload_base; | |
4798 | struct packet_offload *ptype; | |
4799 | ||
4800 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4801 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4802 | continue; | |
4803 | return ptype; | |
4804 | } | |
4805 | return NULL; | |
4806 | } | |
e27a2f83 | 4807 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4808 | |
bb728820 | 4809 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4810 | { |
5d0d9be8 HX |
4811 | switch (ret) { |
4812 | case GRO_NORMAL: | |
ae78dbfa | 4813 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4814 | ret = GRO_DROP; |
4815 | break; | |
5d38a079 | 4816 | |
5d0d9be8 | 4817 | case GRO_DROP: |
5d38a079 HX |
4818 | kfree_skb(skb); |
4819 | break; | |
5b252f0c | 4820 | |
daa86548 | 4821 | case GRO_MERGED_FREE: |
ce87fc6c JG |
4822 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) { |
4823 | skb_dst_drop(skb); | |
f991bb9d | 4824 | secpath_reset(skb); |
d7e8883c | 4825 | kmem_cache_free(skbuff_head_cache, skb); |
ce87fc6c | 4826 | } else { |
d7e8883c | 4827 | __kfree_skb(skb); |
ce87fc6c | 4828 | } |
daa86548 ED |
4829 | break; |
4830 | ||
5b252f0c BH |
4831 | case GRO_HELD: |
4832 | case GRO_MERGED: | |
25393d3f | 4833 | case GRO_CONSUMED: |
5b252f0c | 4834 | break; |
5d38a079 HX |
4835 | } |
4836 | ||
c7c4b3b6 | 4837 | return ret; |
5d0d9be8 | 4838 | } |
5d0d9be8 | 4839 | |
c7c4b3b6 | 4840 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4841 | { |
93f93a44 | 4842 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 4843 | trace_napi_gro_receive_entry(skb); |
86911732 | 4844 | |
a50e233c ED |
4845 | skb_gro_reset_offset(skb); |
4846 | ||
89c5fa33 | 4847 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4848 | } |
4849 | EXPORT_SYMBOL(napi_gro_receive); | |
4850 | ||
d0c2b0d2 | 4851 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4852 | { |
93a35f59 ED |
4853 | if (unlikely(skb->pfmemalloc)) { |
4854 | consume_skb(skb); | |
4855 | return; | |
4856 | } | |
96e93eab | 4857 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4858 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4859 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4860 | skb->vlan_tci = 0; |
66c46d74 | 4861 | skb->dev = napi->dev; |
6d152e23 | 4862 | skb->skb_iif = 0; |
c3caf119 JC |
4863 | skb->encapsulation = 0; |
4864 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4865 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
f991bb9d | 4866 | secpath_reset(skb); |
96e93eab HX |
4867 | |
4868 | napi->skb = skb; | |
4869 | } | |
96e93eab | 4870 | |
76620aaf | 4871 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4872 | { |
5d38a079 | 4873 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4874 | |
4875 | if (!skb) { | |
fd11a83d | 4876 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
4877 | if (skb) { |
4878 | napi->skb = skb; | |
4879 | skb_mark_napi_id(skb, napi); | |
4880 | } | |
80595d59 | 4881 | } |
96e93eab HX |
4882 | return skb; |
4883 | } | |
76620aaf | 4884 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4885 | |
a50e233c ED |
4886 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4887 | struct sk_buff *skb, | |
4888 | gro_result_t ret) | |
96e93eab | 4889 | { |
5d0d9be8 HX |
4890 | switch (ret) { |
4891 | case GRO_NORMAL: | |
a50e233c ED |
4892 | case GRO_HELD: |
4893 | __skb_push(skb, ETH_HLEN); | |
4894 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4895 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4896 | ret = GRO_DROP; |
86911732 | 4897 | break; |
5d38a079 | 4898 | |
5d0d9be8 | 4899 | case GRO_DROP: |
5d0d9be8 HX |
4900 | case GRO_MERGED_FREE: |
4901 | napi_reuse_skb(napi, skb); | |
4902 | break; | |
5b252f0c BH |
4903 | |
4904 | case GRO_MERGED: | |
25393d3f | 4905 | case GRO_CONSUMED: |
5b252f0c | 4906 | break; |
5d0d9be8 | 4907 | } |
5d38a079 | 4908 | |
c7c4b3b6 | 4909 | return ret; |
5d38a079 | 4910 | } |
5d0d9be8 | 4911 | |
a50e233c ED |
4912 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
4913 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
4914 | * We copy ethernet header into skb->data to have a common layout. | |
4915 | */ | |
4adb9c4a | 4916 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4917 | { |
4918 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
4919 | const struct ethhdr *eth; |
4920 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
4921 | |
4922 | napi->skb = NULL; | |
4923 | ||
a50e233c ED |
4924 | skb_reset_mac_header(skb); |
4925 | skb_gro_reset_offset(skb); | |
4926 | ||
4927 | eth = skb_gro_header_fast(skb, 0); | |
4928 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
4929 | eth = skb_gro_header_slow(skb, hlen, 0); | |
4930 | if (unlikely(!eth)) { | |
4da46ceb AC |
4931 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
4932 | __func__, napi->dev->name); | |
a50e233c ED |
4933 | napi_reuse_skb(napi, skb); |
4934 | return NULL; | |
4935 | } | |
4936 | } else { | |
4937 | gro_pull_from_frag0(skb, hlen); | |
4938 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
4939 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 4940 | } |
a50e233c ED |
4941 | __skb_pull(skb, hlen); |
4942 | ||
4943 | /* | |
4944 | * This works because the only protocols we care about don't require | |
4945 | * special handling. | |
4946 | * We'll fix it up properly in napi_frags_finish() | |
4947 | */ | |
4948 | skb->protocol = eth->h_proto; | |
76620aaf | 4949 | |
76620aaf HX |
4950 | return skb; |
4951 | } | |
76620aaf | 4952 | |
c7c4b3b6 | 4953 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4954 | { |
76620aaf | 4955 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4956 | |
4957 | if (!skb) | |
c7c4b3b6 | 4958 | return GRO_DROP; |
5d0d9be8 | 4959 | |
ae78dbfa BH |
4960 | trace_napi_gro_frags_entry(skb); |
4961 | ||
89c5fa33 | 4962 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4963 | } |
5d38a079 HX |
4964 | EXPORT_SYMBOL(napi_gro_frags); |
4965 | ||
573e8fca TH |
4966 | /* Compute the checksum from gro_offset and return the folded value |
4967 | * after adding in any pseudo checksum. | |
4968 | */ | |
4969 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
4970 | { | |
4971 | __wsum wsum; | |
4972 | __sum16 sum; | |
4973 | ||
4974 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
4975 | ||
4976 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
4977 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
4978 | if (likely(!sum)) { | |
4979 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
4980 | !skb->csum_complete_sw) | |
4981 | netdev_rx_csum_fault(skb->dev); | |
4982 | } | |
4983 | ||
4984 | NAPI_GRO_CB(skb)->csum = wsum; | |
4985 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4986 | ||
4987 | return sum; | |
4988 | } | |
4989 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
4990 | ||
e326bed2 | 4991 | /* |
855abcf0 | 4992 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
4993 | * Note: called with local irq disabled, but exits with local irq enabled. |
4994 | */ | |
4995 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4996 | { | |
4997 | #ifdef CONFIG_RPS | |
4998 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4999 | ||
5000 | if (remsd) { | |
5001 | sd->rps_ipi_list = NULL; | |
5002 | ||
5003 | local_irq_enable(); | |
5004 | ||
5005 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
5006 | while (remsd) { | |
5007 | struct softnet_data *next = remsd->rps_ipi_next; | |
5008 | ||
5009 | if (cpu_online(remsd->cpu)) | |
c46fff2a | 5010 | smp_call_function_single_async(remsd->cpu, |
fce8ad15 | 5011 | &remsd->csd); |
e326bed2 ED |
5012 | remsd = next; |
5013 | } | |
5014 | } else | |
5015 | #endif | |
5016 | local_irq_enable(); | |
5017 | } | |
5018 | ||
d75b1ade ED |
5019 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
5020 | { | |
5021 | #ifdef CONFIG_RPS | |
5022 | return sd->rps_ipi_list != NULL; | |
5023 | #else | |
5024 | return false; | |
5025 | #endif | |
5026 | } | |
5027 | ||
bea3348e | 5028 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 5029 | { |
eecfd7c4 | 5030 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
5031 | bool again = true; |
5032 | int work = 0; | |
1da177e4 | 5033 | |
e326bed2 ED |
5034 | /* Check if we have pending ipi, its better to send them now, |
5035 | * not waiting net_rx_action() end. | |
5036 | */ | |
d75b1ade | 5037 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
5038 | local_irq_disable(); |
5039 | net_rps_action_and_irq_enable(sd); | |
5040 | } | |
d75b1ade | 5041 | |
3d48b53f | 5042 | napi->weight = dev_rx_weight; |
145dd5f9 | 5043 | while (again) { |
1da177e4 | 5044 | struct sk_buff *skb; |
6e7676c1 CG |
5045 | |
5046 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 5047 | rcu_read_lock(); |
6e7676c1 | 5048 | __netif_receive_skb(skb); |
2c17d27c | 5049 | rcu_read_unlock(); |
76cc8b13 | 5050 | input_queue_head_incr(sd); |
145dd5f9 | 5051 | if (++work >= quota) |
76cc8b13 | 5052 | return work; |
145dd5f9 | 5053 | |
6e7676c1 | 5054 | } |
1da177e4 | 5055 | |
145dd5f9 | 5056 | local_irq_disable(); |
e36fa2f7 | 5057 | rps_lock(sd); |
11ef7a89 | 5058 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
5059 | /* |
5060 | * Inline a custom version of __napi_complete(). | |
5061 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
5062 | * and NAPI_STATE_SCHED is the only possible flag set |
5063 | * on backlog. | |
5064 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
5065 | * and we dont need an smp_mb() memory barrier. |
5066 | */ | |
eecfd7c4 | 5067 | napi->state = 0; |
145dd5f9 PA |
5068 | again = false; |
5069 | } else { | |
5070 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
5071 | &sd->process_queue); | |
bea3348e | 5072 | } |
e36fa2f7 | 5073 | rps_unlock(sd); |
145dd5f9 | 5074 | local_irq_enable(); |
6e7676c1 | 5075 | } |
1da177e4 | 5076 | |
bea3348e SH |
5077 | return work; |
5078 | } | |
1da177e4 | 5079 | |
bea3348e SH |
5080 | /** |
5081 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 5082 | * @n: entry to schedule |
bea3348e | 5083 | * |
bc9ad166 ED |
5084 | * The entry's receive function will be scheduled to run. |
5085 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 5086 | */ |
b5606c2d | 5087 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
5088 | { |
5089 | unsigned long flags; | |
1da177e4 | 5090 | |
bea3348e | 5091 | local_irq_save(flags); |
903ceff7 | 5092 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 5093 | local_irq_restore(flags); |
1da177e4 | 5094 | } |
bea3348e SH |
5095 | EXPORT_SYMBOL(__napi_schedule); |
5096 | ||
39e6c820 ED |
5097 | /** |
5098 | * napi_schedule_prep - check if napi can be scheduled | |
5099 | * @n: napi context | |
5100 | * | |
5101 | * Test if NAPI routine is already running, and if not mark | |
5102 | * it as running. This is used as a condition variable | |
5103 | * insure only one NAPI poll instance runs. We also make | |
5104 | * sure there is no pending NAPI disable. | |
5105 | */ | |
5106 | bool napi_schedule_prep(struct napi_struct *n) | |
5107 | { | |
5108 | unsigned long val, new; | |
5109 | ||
5110 | do { | |
5111 | val = READ_ONCE(n->state); | |
5112 | if (unlikely(val & NAPIF_STATE_DISABLE)) | |
5113 | return false; | |
5114 | new = val | NAPIF_STATE_SCHED; | |
5115 | ||
5116 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
5117 | * This was suggested by Alexander Duyck, as compiler | |
5118 | * emits better code than : | |
5119 | * if (val & NAPIF_STATE_SCHED) | |
5120 | * new |= NAPIF_STATE_MISSED; | |
5121 | */ | |
5122 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
5123 | NAPIF_STATE_MISSED; | |
5124 | } while (cmpxchg(&n->state, val, new) != val); | |
5125 | ||
5126 | return !(val & NAPIF_STATE_SCHED); | |
5127 | } | |
5128 | EXPORT_SYMBOL(napi_schedule_prep); | |
5129 | ||
bc9ad166 ED |
5130 | /** |
5131 | * __napi_schedule_irqoff - schedule for receive | |
5132 | * @n: entry to schedule | |
5133 | * | |
5134 | * Variant of __napi_schedule() assuming hard irqs are masked | |
5135 | */ | |
5136 | void __napi_schedule_irqoff(struct napi_struct *n) | |
5137 | { | |
5138 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
5139 | } | |
5140 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
5141 | ||
364b6055 | 5142 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 5143 | { |
39e6c820 | 5144 | unsigned long flags, val, new; |
d565b0a1 HX |
5145 | |
5146 | /* | |
217f6974 ED |
5147 | * 1) Don't let napi dequeue from the cpu poll list |
5148 | * just in case its running on a different cpu. | |
5149 | * 2) If we are busy polling, do nothing here, we have | |
5150 | * the guarantee we will be called later. | |
d565b0a1 | 5151 | */ |
217f6974 ED |
5152 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
5153 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 5154 | return false; |
d565b0a1 | 5155 | |
3b47d303 ED |
5156 | if (n->gro_list) { |
5157 | unsigned long timeout = 0; | |
d75b1ade | 5158 | |
3b47d303 ED |
5159 | if (work_done) |
5160 | timeout = n->dev->gro_flush_timeout; | |
5161 | ||
5162 | if (timeout) | |
5163 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
5164 | HRTIMER_MODE_REL_PINNED); | |
5165 | else | |
5166 | napi_gro_flush(n, false); | |
5167 | } | |
02c1602e | 5168 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
5169 | /* If n->poll_list is not empty, we need to mask irqs */ |
5170 | local_irq_save(flags); | |
02c1602e | 5171 | list_del_init(&n->poll_list); |
d75b1ade ED |
5172 | local_irq_restore(flags); |
5173 | } | |
39e6c820 ED |
5174 | |
5175 | do { | |
5176 | val = READ_ONCE(n->state); | |
5177 | ||
5178 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); | |
5179 | ||
5180 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); | |
5181 | ||
5182 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
5183 | * because we will call napi->poll() one more time. | |
5184 | * This C code was suggested by Alexander Duyck to help gcc. | |
5185 | */ | |
5186 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
5187 | NAPIF_STATE_SCHED; | |
5188 | } while (cmpxchg(&n->state, val, new) != val); | |
5189 | ||
5190 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
5191 | __napi_schedule(n); | |
5192 | return false; | |
5193 | } | |
5194 | ||
364b6055 | 5195 | return true; |
d565b0a1 | 5196 | } |
3b47d303 | 5197 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 5198 | |
af12fa6e | 5199 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 5200 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
5201 | { |
5202 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
5203 | struct napi_struct *napi; | |
5204 | ||
5205 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
5206 | if (napi->napi_id == napi_id) | |
5207 | return napi; | |
5208 | ||
5209 | return NULL; | |
5210 | } | |
02d62e86 ED |
5211 | |
5212 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
217f6974 | 5213 | |
ce6aea93 | 5214 | #define BUSY_POLL_BUDGET 8 |
217f6974 ED |
5215 | |
5216 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) | |
5217 | { | |
5218 | int rc; | |
5219 | ||
39e6c820 ED |
5220 | /* Busy polling means there is a high chance device driver hard irq |
5221 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
5222 | * set in napi_schedule_prep(). | |
5223 | * Since we are about to call napi->poll() once more, we can safely | |
5224 | * clear NAPI_STATE_MISSED. | |
5225 | * | |
5226 | * Note: x86 could use a single "lock and ..." instruction | |
5227 | * to perform these two clear_bit() | |
5228 | */ | |
5229 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
5230 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
5231 | ||
5232 | local_bh_disable(); | |
5233 | ||
5234 | /* All we really want here is to re-enable device interrupts. | |
5235 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
5236 | */ | |
5237 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
5238 | netpoll_poll_unlock(have_poll_lock); | |
5239 | if (rc == BUSY_POLL_BUDGET) | |
5240 | __napi_schedule(napi); | |
5241 | local_bh_enable(); | |
5242 | if (local_softirq_pending()) | |
5243 | do_softirq(); | |
5244 | } | |
5245 | ||
7db6b048 SS |
5246 | void napi_busy_loop(unsigned int napi_id, |
5247 | bool (*loop_end)(void *, unsigned long), | |
5248 | void *loop_end_arg) | |
02d62e86 | 5249 | { |
7db6b048 | 5250 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 5251 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 5252 | void *have_poll_lock = NULL; |
02d62e86 | 5253 | struct napi_struct *napi; |
217f6974 ED |
5254 | |
5255 | restart: | |
217f6974 | 5256 | napi_poll = NULL; |
02d62e86 | 5257 | |
2a028ecb | 5258 | rcu_read_lock(); |
02d62e86 | 5259 | |
545cd5e5 | 5260 | napi = napi_by_id(napi_id); |
02d62e86 ED |
5261 | if (!napi) |
5262 | goto out; | |
5263 | ||
217f6974 ED |
5264 | preempt_disable(); |
5265 | for (;;) { | |
2b5cd0df AD |
5266 | int work = 0; |
5267 | ||
2a028ecb | 5268 | local_bh_disable(); |
217f6974 ED |
5269 | if (!napi_poll) { |
5270 | unsigned long val = READ_ONCE(napi->state); | |
5271 | ||
5272 | /* If multiple threads are competing for this napi, | |
5273 | * we avoid dirtying napi->state as much as we can. | |
5274 | */ | |
5275 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
5276 | NAPIF_STATE_IN_BUSY_POLL)) | |
5277 | goto count; | |
5278 | if (cmpxchg(&napi->state, val, | |
5279 | val | NAPIF_STATE_IN_BUSY_POLL | | |
5280 | NAPIF_STATE_SCHED) != val) | |
5281 | goto count; | |
5282 | have_poll_lock = netpoll_poll_lock(napi); | |
5283 | napi_poll = napi->poll; | |
5284 | } | |
2b5cd0df AD |
5285 | work = napi_poll(napi, BUSY_POLL_BUDGET); |
5286 | trace_napi_poll(napi, work, BUSY_POLL_BUDGET); | |
217f6974 | 5287 | count: |
2b5cd0df | 5288 | if (work > 0) |
7db6b048 | 5289 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 5290 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
2a028ecb | 5291 | local_bh_enable(); |
02d62e86 | 5292 | |
7db6b048 | 5293 | if (!loop_end || loop_end(loop_end_arg, start_time)) |
217f6974 | 5294 | break; |
02d62e86 | 5295 | |
217f6974 ED |
5296 | if (unlikely(need_resched())) { |
5297 | if (napi_poll) | |
5298 | busy_poll_stop(napi, have_poll_lock); | |
5299 | preempt_enable(); | |
5300 | rcu_read_unlock(); | |
5301 | cond_resched(); | |
7db6b048 | 5302 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 5303 | return; |
217f6974 ED |
5304 | goto restart; |
5305 | } | |
6cdf89b1 | 5306 | cpu_relax(); |
217f6974 ED |
5307 | } |
5308 | if (napi_poll) | |
5309 | busy_poll_stop(napi, have_poll_lock); | |
5310 | preempt_enable(); | |
02d62e86 | 5311 | out: |
2a028ecb | 5312 | rcu_read_unlock(); |
02d62e86 | 5313 | } |
7db6b048 | 5314 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
5315 | |
5316 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 5317 | |
149d6ad8 | 5318 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 5319 | { |
d64b5e85 ED |
5320 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
5321 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 5322 | return; |
af12fa6e | 5323 | |
52bd2d62 | 5324 | spin_lock(&napi_hash_lock); |
af12fa6e | 5325 | |
545cd5e5 | 5326 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 5327 | do { |
545cd5e5 AD |
5328 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
5329 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
5330 | } while (napi_by_id(napi_gen_id)); |
5331 | napi->napi_id = napi_gen_id; | |
af12fa6e | 5332 | |
52bd2d62 ED |
5333 | hlist_add_head_rcu(&napi->napi_hash_node, |
5334 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 5335 | |
52bd2d62 | 5336 | spin_unlock(&napi_hash_lock); |
af12fa6e | 5337 | } |
af12fa6e ET |
5338 | |
5339 | /* Warning : caller is responsible to make sure rcu grace period | |
5340 | * is respected before freeing memory containing @napi | |
5341 | */ | |
34cbe27e | 5342 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 5343 | { |
34cbe27e ED |
5344 | bool rcu_sync_needed = false; |
5345 | ||
af12fa6e ET |
5346 | spin_lock(&napi_hash_lock); |
5347 | ||
34cbe27e ED |
5348 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
5349 | rcu_sync_needed = true; | |
af12fa6e | 5350 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 5351 | } |
af12fa6e | 5352 | spin_unlock(&napi_hash_lock); |
34cbe27e | 5353 | return rcu_sync_needed; |
af12fa6e ET |
5354 | } |
5355 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
5356 | ||
3b47d303 ED |
5357 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
5358 | { | |
5359 | struct napi_struct *napi; | |
5360 | ||
5361 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
5362 | |
5363 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
5364 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
5365 | */ | |
5366 | if (napi->gro_list && !napi_disable_pending(napi) && | |
5367 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) | |
5368 | __napi_schedule_irqoff(napi); | |
3b47d303 ED |
5369 | |
5370 | return HRTIMER_NORESTART; | |
5371 | } | |
5372 | ||
d565b0a1 HX |
5373 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
5374 | int (*poll)(struct napi_struct *, int), int weight) | |
5375 | { | |
5376 | INIT_LIST_HEAD(&napi->poll_list); | |
3b47d303 ED |
5377 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
5378 | napi->timer.function = napi_watchdog; | |
4ae5544f | 5379 | napi->gro_count = 0; |
d565b0a1 | 5380 | napi->gro_list = NULL; |
5d38a079 | 5381 | napi->skb = NULL; |
d565b0a1 | 5382 | napi->poll = poll; |
82dc3c63 ED |
5383 | if (weight > NAPI_POLL_WEIGHT) |
5384 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
5385 | weight, dev->name); | |
d565b0a1 HX |
5386 | napi->weight = weight; |
5387 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 5388 | napi->dev = dev; |
5d38a079 | 5389 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
5390 | napi->poll_owner = -1; |
5391 | #endif | |
5392 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 5393 | napi_hash_add(napi); |
d565b0a1 HX |
5394 | } |
5395 | EXPORT_SYMBOL(netif_napi_add); | |
5396 | ||
3b47d303 ED |
5397 | void napi_disable(struct napi_struct *n) |
5398 | { | |
5399 | might_sleep(); | |
5400 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
5401 | ||
5402 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
5403 | msleep(1); | |
2d8bff12 NH |
5404 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
5405 | msleep(1); | |
3b47d303 ED |
5406 | |
5407 | hrtimer_cancel(&n->timer); | |
5408 | ||
5409 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
5410 | } | |
5411 | EXPORT_SYMBOL(napi_disable); | |
5412 | ||
93d05d4a | 5413 | /* Must be called in process context */ |
d565b0a1 HX |
5414 | void netif_napi_del(struct napi_struct *napi) |
5415 | { | |
93d05d4a ED |
5416 | might_sleep(); |
5417 | if (napi_hash_del(napi)) | |
5418 | synchronize_net(); | |
d7b06636 | 5419 | list_del_init(&napi->dev_list); |
76620aaf | 5420 | napi_free_frags(napi); |
d565b0a1 | 5421 | |
289dccbe | 5422 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 5423 | napi->gro_list = NULL; |
4ae5544f | 5424 | napi->gro_count = 0; |
d565b0a1 HX |
5425 | } |
5426 | EXPORT_SYMBOL(netif_napi_del); | |
5427 | ||
726ce70e HX |
5428 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
5429 | { | |
5430 | void *have; | |
5431 | int work, weight; | |
5432 | ||
5433 | list_del_init(&n->poll_list); | |
5434 | ||
5435 | have = netpoll_poll_lock(n); | |
5436 | ||
5437 | weight = n->weight; | |
5438 | ||
5439 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
5440 | * with netpoll's poll_napi(). Only the entity which | |
5441 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
5442 | * actually make the ->poll() call. Therefore we avoid | |
5443 | * accidentally calling ->poll() when NAPI is not scheduled. | |
5444 | */ | |
5445 | work = 0; | |
5446 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
5447 | work = n->poll(n, weight); | |
1db19db7 | 5448 | trace_napi_poll(n, work, weight); |
726ce70e HX |
5449 | } |
5450 | ||
5451 | WARN_ON_ONCE(work > weight); | |
5452 | ||
5453 | if (likely(work < weight)) | |
5454 | goto out_unlock; | |
5455 | ||
5456 | /* Drivers must not modify the NAPI state if they | |
5457 | * consume the entire weight. In such cases this code | |
5458 | * still "owns" the NAPI instance and therefore can | |
5459 | * move the instance around on the list at-will. | |
5460 | */ | |
5461 | if (unlikely(napi_disable_pending(n))) { | |
5462 | napi_complete(n); | |
5463 | goto out_unlock; | |
5464 | } | |
5465 | ||
5466 | if (n->gro_list) { | |
5467 | /* flush too old packets | |
5468 | * If HZ < 1000, flush all packets. | |
5469 | */ | |
5470 | napi_gro_flush(n, HZ >= 1000); | |
5471 | } | |
5472 | ||
001ce546 HX |
5473 | /* Some drivers may have called napi_schedule |
5474 | * prior to exhausting their budget. | |
5475 | */ | |
5476 | if (unlikely(!list_empty(&n->poll_list))) { | |
5477 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
5478 | n->dev ? n->dev->name : "backlog"); | |
5479 | goto out_unlock; | |
5480 | } | |
5481 | ||
726ce70e HX |
5482 | list_add_tail(&n->poll_list, repoll); |
5483 | ||
5484 | out_unlock: | |
5485 | netpoll_poll_unlock(have); | |
5486 | ||
5487 | return work; | |
5488 | } | |
5489 | ||
0766f788 | 5490 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 5491 | { |
903ceff7 | 5492 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e MW |
5493 | unsigned long time_limit = jiffies + |
5494 | usecs_to_jiffies(netdev_budget_usecs); | |
51b0bded | 5495 | int budget = netdev_budget; |
d75b1ade ED |
5496 | LIST_HEAD(list); |
5497 | LIST_HEAD(repoll); | |
53fb95d3 | 5498 | |
1da177e4 | 5499 | local_irq_disable(); |
d75b1ade ED |
5500 | list_splice_init(&sd->poll_list, &list); |
5501 | local_irq_enable(); | |
1da177e4 | 5502 | |
ceb8d5bf | 5503 | for (;;) { |
bea3348e | 5504 | struct napi_struct *n; |
1da177e4 | 5505 | |
ceb8d5bf HX |
5506 | if (list_empty(&list)) { |
5507 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
f52dffe0 | 5508 | goto out; |
ceb8d5bf HX |
5509 | break; |
5510 | } | |
5511 | ||
6bd373eb HX |
5512 | n = list_first_entry(&list, struct napi_struct, poll_list); |
5513 | budget -= napi_poll(n, &repoll); | |
5514 | ||
d75b1ade | 5515 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
5516 | * Allow this to run for 2 jiffies since which will allow |
5517 | * an average latency of 1.5/HZ. | |
bea3348e | 5518 | */ |
ceb8d5bf HX |
5519 | if (unlikely(budget <= 0 || |
5520 | time_after_eq(jiffies, time_limit))) { | |
5521 | sd->time_squeeze++; | |
5522 | break; | |
5523 | } | |
1da177e4 | 5524 | } |
d75b1ade | 5525 | |
d75b1ade ED |
5526 | local_irq_disable(); |
5527 | ||
5528 | list_splice_tail_init(&sd->poll_list, &list); | |
5529 | list_splice_tail(&repoll, &list); | |
5530 | list_splice(&list, &sd->poll_list); | |
5531 | if (!list_empty(&sd->poll_list)) | |
5532 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
5533 | ||
e326bed2 | 5534 | net_rps_action_and_irq_enable(sd); |
f52dffe0 ED |
5535 | out: |
5536 | __kfree_skb_flush(); | |
1da177e4 LT |
5537 | } |
5538 | ||
aa9d8560 | 5539 | struct netdev_adjacent { |
9ff162a8 | 5540 | struct net_device *dev; |
5d261913 VF |
5541 | |
5542 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 5543 | bool master; |
5d261913 | 5544 | |
5d261913 VF |
5545 | /* counter for the number of times this device was added to us */ |
5546 | u16 ref_nr; | |
5547 | ||
402dae96 VF |
5548 | /* private field for the users */ |
5549 | void *private; | |
5550 | ||
9ff162a8 JP |
5551 | struct list_head list; |
5552 | struct rcu_head rcu; | |
9ff162a8 JP |
5553 | }; |
5554 | ||
6ea29da1 | 5555 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 5556 | struct list_head *adj_list) |
9ff162a8 | 5557 | { |
5d261913 | 5558 | struct netdev_adjacent *adj; |
5d261913 | 5559 | |
2f268f12 | 5560 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
5561 | if (adj->dev == adj_dev) |
5562 | return adj; | |
9ff162a8 JP |
5563 | } |
5564 | return NULL; | |
5565 | } | |
5566 | ||
f1170fd4 DA |
5567 | static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data) |
5568 | { | |
5569 | struct net_device *dev = data; | |
5570 | ||
5571 | return upper_dev == dev; | |
5572 | } | |
5573 | ||
9ff162a8 JP |
5574 | /** |
5575 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
5576 | * @dev: device | |
5577 | * @upper_dev: upper device to check | |
5578 | * | |
5579 | * Find out if a device is linked to specified upper device and return true | |
5580 | * in case it is. Note that this checks only immediate upper device, | |
5581 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
5582 | */ | |
5583 | bool netdev_has_upper_dev(struct net_device *dev, | |
5584 | struct net_device *upper_dev) | |
5585 | { | |
5586 | ASSERT_RTNL(); | |
5587 | ||
f1170fd4 DA |
5588 | return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, |
5589 | upper_dev); | |
9ff162a8 JP |
5590 | } |
5591 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
5592 | ||
1a3f060c DA |
5593 | /** |
5594 | * netdev_has_upper_dev_all - Check if device is linked to an upper device | |
5595 | * @dev: device | |
5596 | * @upper_dev: upper device to check | |
5597 | * | |
5598 | * Find out if a device is linked to specified upper device and return true | |
5599 | * in case it is. Note that this checks the entire upper device chain. | |
5600 | * The caller must hold rcu lock. | |
5601 | */ | |
5602 | ||
1a3f060c DA |
5603 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
5604 | struct net_device *upper_dev) | |
5605 | { | |
5606 | return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, | |
5607 | upper_dev); | |
5608 | } | |
5609 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
5610 | ||
9ff162a8 JP |
5611 | /** |
5612 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
5613 | * @dev: device | |
5614 | * | |
5615 | * Find out if a device is linked to an upper device and return true in case | |
5616 | * it is. The caller must hold the RTNL lock. | |
5617 | */ | |
1d143d9f | 5618 | static bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
5619 | { |
5620 | ASSERT_RTNL(); | |
5621 | ||
f1170fd4 | 5622 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 5623 | } |
9ff162a8 JP |
5624 | |
5625 | /** | |
5626 | * netdev_master_upper_dev_get - Get master upper device | |
5627 | * @dev: device | |
5628 | * | |
5629 | * Find a master upper device and return pointer to it or NULL in case | |
5630 | * it's not there. The caller must hold the RTNL lock. | |
5631 | */ | |
5632 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
5633 | { | |
aa9d8560 | 5634 | struct netdev_adjacent *upper; |
9ff162a8 JP |
5635 | |
5636 | ASSERT_RTNL(); | |
5637 | ||
2f268f12 | 5638 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
5639 | return NULL; |
5640 | ||
2f268f12 | 5641 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 5642 | struct netdev_adjacent, list); |
9ff162a8 JP |
5643 | if (likely(upper->master)) |
5644 | return upper->dev; | |
5645 | return NULL; | |
5646 | } | |
5647 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
5648 | ||
0f524a80 DA |
5649 | /** |
5650 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
5651 | * @dev: device | |
5652 | * | |
5653 | * Find out if a device is linked to a lower device and return true in case | |
5654 | * it is. The caller must hold the RTNL lock. | |
5655 | */ | |
5656 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
5657 | { | |
5658 | ASSERT_RTNL(); | |
5659 | ||
5660 | return !list_empty(&dev->adj_list.lower); | |
5661 | } | |
5662 | ||
b6ccba4c VF |
5663 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
5664 | { | |
5665 | struct netdev_adjacent *adj; | |
5666 | ||
5667 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
5668 | ||
5669 | return adj->private; | |
5670 | } | |
5671 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
5672 | ||
44a40855 VY |
5673 | /** |
5674 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
5675 | * @dev: device | |
5676 | * @iter: list_head ** of the current position | |
5677 | * | |
5678 | * Gets the next device from the dev's upper list, starting from iter | |
5679 | * position. The caller must hold RCU read lock. | |
5680 | */ | |
5681 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
5682 | struct list_head **iter) | |
5683 | { | |
5684 | struct netdev_adjacent *upper; | |
5685 | ||
5686 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5687 | ||
5688 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5689 | ||
5690 | if (&upper->list == &dev->adj_list.upper) | |
5691 | return NULL; | |
5692 | ||
5693 | *iter = &upper->list; | |
5694 | ||
5695 | return upper->dev; | |
5696 | } | |
5697 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
5698 | ||
1a3f060c DA |
5699 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
5700 | struct list_head **iter) | |
5701 | { | |
5702 | struct netdev_adjacent *upper; | |
5703 | ||
5704 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5705 | ||
5706 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5707 | ||
5708 | if (&upper->list == &dev->adj_list.upper) | |
5709 | return NULL; | |
5710 | ||
5711 | *iter = &upper->list; | |
5712 | ||
5713 | return upper->dev; | |
5714 | } | |
5715 | ||
5716 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, | |
5717 | int (*fn)(struct net_device *dev, | |
5718 | void *data), | |
5719 | void *data) | |
5720 | { | |
5721 | struct net_device *udev; | |
5722 | struct list_head *iter; | |
5723 | int ret; | |
5724 | ||
5725 | for (iter = &dev->adj_list.upper, | |
5726 | udev = netdev_next_upper_dev_rcu(dev, &iter); | |
5727 | udev; | |
5728 | udev = netdev_next_upper_dev_rcu(dev, &iter)) { | |
5729 | /* first is the upper device itself */ | |
5730 | ret = fn(udev, data); | |
5731 | if (ret) | |
5732 | return ret; | |
5733 | ||
5734 | /* then look at all of its upper devices */ | |
5735 | ret = netdev_walk_all_upper_dev_rcu(udev, fn, data); | |
5736 | if (ret) | |
5737 | return ret; | |
5738 | } | |
5739 | ||
5740 | return 0; | |
5741 | } | |
5742 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
5743 | ||
31088a11 VF |
5744 | /** |
5745 | * netdev_lower_get_next_private - Get the next ->private from the | |
5746 | * lower neighbour list | |
5747 | * @dev: device | |
5748 | * @iter: list_head ** of the current position | |
5749 | * | |
5750 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5751 | * list, starting from iter position. The caller must hold either hold the | |
5752 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 5753 | * list will remain unchanged. |
31088a11 VF |
5754 | */ |
5755 | void *netdev_lower_get_next_private(struct net_device *dev, | |
5756 | struct list_head **iter) | |
5757 | { | |
5758 | struct netdev_adjacent *lower; | |
5759 | ||
5760 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
5761 | ||
5762 | if (&lower->list == &dev->adj_list.lower) | |
5763 | return NULL; | |
5764 | ||
6859e7df | 5765 | *iter = lower->list.next; |
31088a11 VF |
5766 | |
5767 | return lower->private; | |
5768 | } | |
5769 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
5770 | ||
5771 | /** | |
5772 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
5773 | * lower neighbour list, RCU | |
5774 | * variant | |
5775 | * @dev: device | |
5776 | * @iter: list_head ** of the current position | |
5777 | * | |
5778 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5779 | * list, starting from iter position. The caller must hold RCU read lock. | |
5780 | */ | |
5781 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
5782 | struct list_head **iter) | |
5783 | { | |
5784 | struct netdev_adjacent *lower; | |
5785 | ||
5786 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5787 | ||
5788 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5789 | ||
5790 | if (&lower->list == &dev->adj_list.lower) | |
5791 | return NULL; | |
5792 | ||
6859e7df | 5793 | *iter = &lower->list; |
31088a11 VF |
5794 | |
5795 | return lower->private; | |
5796 | } | |
5797 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
5798 | ||
4085ebe8 VY |
5799 | /** |
5800 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
5801 | * list | |
5802 | * @dev: device | |
5803 | * @iter: list_head ** of the current position | |
5804 | * | |
5805 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
5806 | * list, starting from iter position. The caller must hold RTNL lock or | |
5807 | * its own locking that guarantees that the neighbour lower | |
b469139e | 5808 | * list will remain unchanged. |
4085ebe8 VY |
5809 | */ |
5810 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
5811 | { | |
5812 | struct netdev_adjacent *lower; | |
5813 | ||
cfdd28be | 5814 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
5815 | |
5816 | if (&lower->list == &dev->adj_list.lower) | |
5817 | return NULL; | |
5818 | ||
cfdd28be | 5819 | *iter = lower->list.next; |
4085ebe8 VY |
5820 | |
5821 | return lower->dev; | |
5822 | } | |
5823 | EXPORT_SYMBOL(netdev_lower_get_next); | |
5824 | ||
1a3f060c DA |
5825 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
5826 | struct list_head **iter) | |
5827 | { | |
5828 | struct netdev_adjacent *lower; | |
5829 | ||
46b5ab1a | 5830 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
5831 | |
5832 | if (&lower->list == &dev->adj_list.lower) | |
5833 | return NULL; | |
5834 | ||
46b5ab1a | 5835 | *iter = &lower->list; |
1a3f060c DA |
5836 | |
5837 | return lower->dev; | |
5838 | } | |
5839 | ||
5840 | int netdev_walk_all_lower_dev(struct net_device *dev, | |
5841 | int (*fn)(struct net_device *dev, | |
5842 | void *data), | |
5843 | void *data) | |
5844 | { | |
5845 | struct net_device *ldev; | |
5846 | struct list_head *iter; | |
5847 | int ret; | |
5848 | ||
5849 | for (iter = &dev->adj_list.lower, | |
5850 | ldev = netdev_next_lower_dev(dev, &iter); | |
5851 | ldev; | |
5852 | ldev = netdev_next_lower_dev(dev, &iter)) { | |
5853 | /* first is the lower device itself */ | |
5854 | ret = fn(ldev, data); | |
5855 | if (ret) | |
5856 | return ret; | |
5857 | ||
5858 | /* then look at all of its lower devices */ | |
5859 | ret = netdev_walk_all_lower_dev(ldev, fn, data); | |
5860 | if (ret) | |
5861 | return ret; | |
5862 | } | |
5863 | ||
5864 | return 0; | |
5865 | } | |
5866 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
5867 | ||
1a3f060c DA |
5868 | static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
5869 | struct list_head **iter) | |
5870 | { | |
5871 | struct netdev_adjacent *lower; | |
5872 | ||
5873 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5874 | if (&lower->list == &dev->adj_list.lower) | |
5875 | return NULL; | |
5876 | ||
5877 | *iter = &lower->list; | |
5878 | ||
5879 | return lower->dev; | |
5880 | } | |
5881 | ||
5882 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
5883 | int (*fn)(struct net_device *dev, | |
5884 | void *data), | |
5885 | void *data) | |
5886 | { | |
5887 | struct net_device *ldev; | |
5888 | struct list_head *iter; | |
5889 | int ret; | |
5890 | ||
5891 | for (iter = &dev->adj_list.lower, | |
5892 | ldev = netdev_next_lower_dev_rcu(dev, &iter); | |
5893 | ldev; | |
5894 | ldev = netdev_next_lower_dev_rcu(dev, &iter)) { | |
5895 | /* first is the lower device itself */ | |
5896 | ret = fn(ldev, data); | |
5897 | if (ret) | |
5898 | return ret; | |
5899 | ||
5900 | /* then look at all of its lower devices */ | |
5901 | ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data); | |
5902 | if (ret) | |
5903 | return ret; | |
5904 | } | |
5905 | ||
5906 | return 0; | |
5907 | } | |
5908 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
5909 | ||
e001bfad | 5910 | /** |
5911 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
5912 | * lower neighbour list, RCU | |
5913 | * variant | |
5914 | * @dev: device | |
5915 | * | |
5916 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
5917 | * list. The caller must hold RCU read lock. | |
5918 | */ | |
5919 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
5920 | { | |
5921 | struct netdev_adjacent *lower; | |
5922 | ||
5923 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
5924 | struct netdev_adjacent, list); | |
5925 | if (lower) | |
5926 | return lower->private; | |
5927 | return NULL; | |
5928 | } | |
5929 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
5930 | ||
9ff162a8 JP |
5931 | /** |
5932 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
5933 | * @dev: device | |
5934 | * | |
5935 | * Find a master upper device and return pointer to it or NULL in case | |
5936 | * it's not there. The caller must hold the RCU read lock. | |
5937 | */ | |
5938 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
5939 | { | |
aa9d8560 | 5940 | struct netdev_adjacent *upper; |
9ff162a8 | 5941 | |
2f268f12 | 5942 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 5943 | struct netdev_adjacent, list); |
9ff162a8 JP |
5944 | if (upper && likely(upper->master)) |
5945 | return upper->dev; | |
5946 | return NULL; | |
5947 | } | |
5948 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
5949 | ||
0a59f3a9 | 5950 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
5951 | struct net_device *adj_dev, |
5952 | struct list_head *dev_list) | |
5953 | { | |
5954 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 5955 | |
3ee32707 VF |
5956 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
5957 | "upper_%s" : "lower_%s", adj_dev->name); | |
5958 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
5959 | linkname); | |
5960 | } | |
0a59f3a9 | 5961 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
5962 | char *name, |
5963 | struct list_head *dev_list) | |
5964 | { | |
5965 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 5966 | |
3ee32707 VF |
5967 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
5968 | "upper_%s" : "lower_%s", name); | |
5969 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
5970 | } | |
5971 | ||
7ce64c79 AF |
5972 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
5973 | struct net_device *adj_dev, | |
5974 | struct list_head *dev_list) | |
5975 | { | |
5976 | return (dev_list == &dev->adj_list.upper || | |
5977 | dev_list == &dev->adj_list.lower) && | |
5978 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
5979 | } | |
3ee32707 | 5980 | |
5d261913 VF |
5981 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
5982 | struct net_device *adj_dev, | |
7863c054 | 5983 | struct list_head *dev_list, |
402dae96 | 5984 | void *private, bool master) |
5d261913 VF |
5985 | { |
5986 | struct netdev_adjacent *adj; | |
842d67a7 | 5987 | int ret; |
5d261913 | 5988 | |
6ea29da1 | 5989 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
5990 | |
5991 | if (adj) { | |
790510d9 | 5992 | adj->ref_nr += 1; |
67b62f98 DA |
5993 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
5994 | dev->name, adj_dev->name, adj->ref_nr); | |
5995 | ||
5d261913 VF |
5996 | return 0; |
5997 | } | |
5998 | ||
5999 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
6000 | if (!adj) | |
6001 | return -ENOMEM; | |
6002 | ||
6003 | adj->dev = adj_dev; | |
6004 | adj->master = master; | |
790510d9 | 6005 | adj->ref_nr = 1; |
402dae96 | 6006 | adj->private = private; |
5d261913 | 6007 | dev_hold(adj_dev); |
2f268f12 | 6008 | |
67b62f98 DA |
6009 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
6010 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 6011 | |
7ce64c79 | 6012 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 6013 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
6014 | if (ret) |
6015 | goto free_adj; | |
6016 | } | |
6017 | ||
7863c054 | 6018 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
6019 | if (master) { |
6020 | ret = sysfs_create_link(&(dev->dev.kobj), | |
6021 | &(adj_dev->dev.kobj), "master"); | |
6022 | if (ret) | |
5831d66e | 6023 | goto remove_symlinks; |
842d67a7 | 6024 | |
7863c054 | 6025 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 6026 | } else { |
7863c054 | 6027 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 6028 | } |
5d261913 VF |
6029 | |
6030 | return 0; | |
842d67a7 | 6031 | |
5831d66e | 6032 | remove_symlinks: |
7ce64c79 | 6033 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6034 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
6035 | free_adj: |
6036 | kfree(adj); | |
974daef7 | 6037 | dev_put(adj_dev); |
842d67a7 VF |
6038 | |
6039 | return ret; | |
5d261913 VF |
6040 | } |
6041 | ||
1d143d9f | 6042 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
6043 | struct net_device *adj_dev, | |
93409033 | 6044 | u16 ref_nr, |
1d143d9f | 6045 | struct list_head *dev_list) |
5d261913 VF |
6046 | { |
6047 | struct netdev_adjacent *adj; | |
6048 | ||
67b62f98 DA |
6049 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
6050 | dev->name, adj_dev->name, ref_nr); | |
6051 | ||
6ea29da1 | 6052 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 6053 | |
2f268f12 | 6054 | if (!adj) { |
67b62f98 | 6055 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 6056 | dev->name, adj_dev->name); |
67b62f98 DA |
6057 | WARN_ON(1); |
6058 | return; | |
2f268f12 | 6059 | } |
5d261913 | 6060 | |
93409033 | 6061 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
6062 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
6063 | dev->name, adj_dev->name, ref_nr, | |
6064 | adj->ref_nr - ref_nr); | |
93409033 | 6065 | adj->ref_nr -= ref_nr; |
5d261913 VF |
6066 | return; |
6067 | } | |
6068 | ||
842d67a7 VF |
6069 | if (adj->master) |
6070 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
6071 | ||
7ce64c79 | 6072 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6073 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 6074 | |
5d261913 | 6075 | list_del_rcu(&adj->list); |
67b62f98 | 6076 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 6077 | adj_dev->name, dev->name, adj_dev->name); |
5d261913 VF |
6078 | dev_put(adj_dev); |
6079 | kfree_rcu(adj, rcu); | |
6080 | } | |
6081 | ||
1d143d9f | 6082 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
6083 | struct net_device *upper_dev, | |
6084 | struct list_head *up_list, | |
6085 | struct list_head *down_list, | |
6086 | void *private, bool master) | |
5d261913 VF |
6087 | { |
6088 | int ret; | |
6089 | ||
790510d9 | 6090 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 6091 | private, master); |
5d261913 VF |
6092 | if (ret) |
6093 | return ret; | |
6094 | ||
790510d9 | 6095 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 6096 | private, false); |
5d261913 | 6097 | if (ret) { |
790510d9 | 6098 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
6099 | return ret; |
6100 | } | |
6101 | ||
6102 | return 0; | |
6103 | } | |
6104 | ||
1d143d9f | 6105 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
6106 | struct net_device *upper_dev, | |
93409033 | 6107 | u16 ref_nr, |
1d143d9f | 6108 | struct list_head *up_list, |
6109 | struct list_head *down_list) | |
5d261913 | 6110 | { |
93409033 AC |
6111 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
6112 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
6113 | } |
6114 | ||
1d143d9f | 6115 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
6116 | struct net_device *upper_dev, | |
6117 | void *private, bool master) | |
2f268f12 | 6118 | { |
f1170fd4 DA |
6119 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
6120 | &dev->adj_list.upper, | |
6121 | &upper_dev->adj_list.lower, | |
6122 | private, master); | |
5d261913 VF |
6123 | } |
6124 | ||
1d143d9f | 6125 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
6126 | struct net_device *upper_dev) | |
2f268f12 | 6127 | { |
93409033 | 6128 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
6129 | &dev->adj_list.upper, |
6130 | &upper_dev->adj_list.lower); | |
6131 | } | |
5d261913 | 6132 | |
9ff162a8 | 6133 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 6134 | struct net_device *upper_dev, bool master, |
29bf24af | 6135 | void *upper_priv, void *upper_info) |
9ff162a8 | 6136 | { |
0e4ead9d | 6137 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 | 6138 | int ret = 0; |
9ff162a8 JP |
6139 | |
6140 | ASSERT_RTNL(); | |
6141 | ||
6142 | if (dev == upper_dev) | |
6143 | return -EBUSY; | |
6144 | ||
6145 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
f1170fd4 | 6146 | if (netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
6147 | return -EBUSY; |
6148 | ||
f1170fd4 | 6149 | if (netdev_has_upper_dev(dev, upper_dev)) |
9ff162a8 JP |
6150 | return -EEXIST; |
6151 | ||
6152 | if (master && netdev_master_upper_dev_get(dev)) | |
6153 | return -EBUSY; | |
6154 | ||
0e4ead9d JP |
6155 | changeupper_info.upper_dev = upper_dev; |
6156 | changeupper_info.master = master; | |
6157 | changeupper_info.linking = true; | |
29bf24af | 6158 | changeupper_info.upper_info = upper_info; |
0e4ead9d | 6159 | |
573c7ba0 JP |
6160 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
6161 | &changeupper_info.info); | |
6162 | ret = notifier_to_errno(ret); | |
6163 | if (ret) | |
6164 | return ret; | |
6165 | ||
6dffb044 | 6166 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 6167 | master); |
5d261913 VF |
6168 | if (ret) |
6169 | return ret; | |
9ff162a8 | 6170 | |
b03804e7 IS |
6171 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
6172 | &changeupper_info.info); | |
6173 | ret = notifier_to_errno(ret); | |
6174 | if (ret) | |
f1170fd4 | 6175 | goto rollback; |
b03804e7 | 6176 | |
9ff162a8 | 6177 | return 0; |
5d261913 | 6178 | |
f1170fd4 | 6179 | rollback: |
2f268f12 | 6180 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
6181 | |
6182 | return ret; | |
9ff162a8 JP |
6183 | } |
6184 | ||
6185 | /** | |
6186 | * netdev_upper_dev_link - Add a link to the upper device | |
6187 | * @dev: device | |
6188 | * @upper_dev: new upper device | |
6189 | * | |
6190 | * Adds a link to device which is upper to this one. The caller must hold | |
6191 | * the RTNL lock. On a failure a negative errno code is returned. | |
6192 | * On success the reference counts are adjusted and the function | |
6193 | * returns zero. | |
6194 | */ | |
6195 | int netdev_upper_dev_link(struct net_device *dev, | |
6196 | struct net_device *upper_dev) | |
6197 | { | |
29bf24af | 6198 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL); |
9ff162a8 JP |
6199 | } |
6200 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
6201 | ||
6202 | /** | |
6203 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
6204 | * @dev: device | |
6205 | * @upper_dev: new upper device | |
6dffb044 | 6206 | * @upper_priv: upper device private |
29bf24af | 6207 | * @upper_info: upper info to be passed down via notifier |
9ff162a8 JP |
6208 | * |
6209 | * Adds a link to device which is upper to this one. In this case, only | |
6210 | * one master upper device can be linked, although other non-master devices | |
6211 | * might be linked as well. The caller must hold the RTNL lock. | |
6212 | * On a failure a negative errno code is returned. On success the reference | |
6213 | * counts are adjusted and the function returns zero. | |
6214 | */ | |
6215 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 6216 | struct net_device *upper_dev, |
29bf24af | 6217 | void *upper_priv, void *upper_info) |
9ff162a8 | 6218 | { |
29bf24af JP |
6219 | return __netdev_upper_dev_link(dev, upper_dev, true, |
6220 | upper_priv, upper_info); | |
9ff162a8 JP |
6221 | } |
6222 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
6223 | ||
6224 | /** | |
6225 | * netdev_upper_dev_unlink - Removes a link to upper device | |
6226 | * @dev: device | |
6227 | * @upper_dev: new upper device | |
6228 | * | |
6229 | * Removes a link to device which is upper to this one. The caller must hold | |
6230 | * the RTNL lock. | |
6231 | */ | |
6232 | void netdev_upper_dev_unlink(struct net_device *dev, | |
6233 | struct net_device *upper_dev) | |
6234 | { | |
0e4ead9d | 6235 | struct netdev_notifier_changeupper_info changeupper_info; |
f4563a75 | 6236 | |
9ff162a8 JP |
6237 | ASSERT_RTNL(); |
6238 | ||
0e4ead9d JP |
6239 | changeupper_info.upper_dev = upper_dev; |
6240 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; | |
6241 | changeupper_info.linking = false; | |
6242 | ||
573c7ba0 JP |
6243 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
6244 | &changeupper_info.info); | |
6245 | ||
2f268f12 | 6246 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 6247 | |
0e4ead9d JP |
6248 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
6249 | &changeupper_info.info); | |
9ff162a8 JP |
6250 | } |
6251 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
6252 | ||
61bd3857 MS |
6253 | /** |
6254 | * netdev_bonding_info_change - Dispatch event about slave change | |
6255 | * @dev: device | |
4a26e453 | 6256 | * @bonding_info: info to dispatch |
61bd3857 MS |
6257 | * |
6258 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
6259 | * The caller must hold the RTNL lock. | |
6260 | */ | |
6261 | void netdev_bonding_info_change(struct net_device *dev, | |
6262 | struct netdev_bonding_info *bonding_info) | |
6263 | { | |
6264 | struct netdev_notifier_bonding_info info; | |
6265 | ||
6266 | memcpy(&info.bonding_info, bonding_info, | |
6267 | sizeof(struct netdev_bonding_info)); | |
6268 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, | |
6269 | &info.info); | |
6270 | } | |
6271 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
6272 | ||
2ce1ee17 | 6273 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
6274 | { |
6275 | struct netdev_adjacent *iter; | |
6276 | ||
6277 | struct net *net = dev_net(dev); | |
6278 | ||
6279 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 6280 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6281 | continue; |
6282 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6283 | &iter->dev->adj_list.lower); | |
6284 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
6285 | &dev->adj_list.upper); | |
6286 | } | |
6287 | ||
6288 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 6289 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6290 | continue; |
6291 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6292 | &iter->dev->adj_list.upper); | |
6293 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
6294 | &dev->adj_list.lower); | |
6295 | } | |
6296 | } | |
6297 | ||
2ce1ee17 | 6298 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
6299 | { |
6300 | struct netdev_adjacent *iter; | |
6301 | ||
6302 | struct net *net = dev_net(dev); | |
6303 | ||
6304 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 6305 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6306 | continue; |
6307 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6308 | &iter->dev->adj_list.lower); | |
6309 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6310 | &dev->adj_list.upper); | |
6311 | } | |
6312 | ||
6313 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 6314 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6315 | continue; |
6316 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6317 | &iter->dev->adj_list.upper); | |
6318 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6319 | &dev->adj_list.lower); | |
6320 | } | |
6321 | } | |
6322 | ||
5bb025fa | 6323 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 6324 | { |
5bb025fa | 6325 | struct netdev_adjacent *iter; |
402dae96 | 6326 | |
4c75431a AF |
6327 | struct net *net = dev_net(dev); |
6328 | ||
5bb025fa | 6329 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 6330 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6331 | continue; |
5bb025fa VF |
6332 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6333 | &iter->dev->adj_list.lower); | |
6334 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6335 | &iter->dev->adj_list.lower); | |
6336 | } | |
402dae96 | 6337 | |
5bb025fa | 6338 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 6339 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6340 | continue; |
5bb025fa VF |
6341 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6342 | &iter->dev->adj_list.upper); | |
6343 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6344 | &iter->dev->adj_list.upper); | |
6345 | } | |
402dae96 | 6346 | } |
402dae96 VF |
6347 | |
6348 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
6349 | struct net_device *lower_dev) | |
6350 | { | |
6351 | struct netdev_adjacent *lower; | |
6352 | ||
6353 | if (!lower_dev) | |
6354 | return NULL; | |
6ea29da1 | 6355 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
6356 | if (!lower) |
6357 | return NULL; | |
6358 | ||
6359 | return lower->private; | |
6360 | } | |
6361 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
6362 | ||
4085ebe8 | 6363 | |
952fcfd0 | 6364 | int dev_get_nest_level(struct net_device *dev) |
4085ebe8 VY |
6365 | { |
6366 | struct net_device *lower = NULL; | |
6367 | struct list_head *iter; | |
6368 | int max_nest = -1; | |
6369 | int nest; | |
6370 | ||
6371 | ASSERT_RTNL(); | |
6372 | ||
6373 | netdev_for_each_lower_dev(dev, lower, iter) { | |
952fcfd0 | 6374 | nest = dev_get_nest_level(lower); |
4085ebe8 VY |
6375 | if (max_nest < nest) |
6376 | max_nest = nest; | |
6377 | } | |
6378 | ||
952fcfd0 | 6379 | return max_nest + 1; |
4085ebe8 VY |
6380 | } |
6381 | EXPORT_SYMBOL(dev_get_nest_level); | |
6382 | ||
04d48266 JP |
6383 | /** |
6384 | * netdev_lower_change - Dispatch event about lower device state change | |
6385 | * @lower_dev: device | |
6386 | * @lower_state_info: state to dispatch | |
6387 | * | |
6388 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
6389 | * The caller must hold the RTNL lock. | |
6390 | */ | |
6391 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
6392 | void *lower_state_info) | |
6393 | { | |
6394 | struct netdev_notifier_changelowerstate_info changelowerstate_info; | |
6395 | ||
6396 | ASSERT_RTNL(); | |
6397 | changelowerstate_info.lower_state_info = lower_state_info; | |
6398 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev, | |
6399 | &changelowerstate_info.info); | |
6400 | } | |
6401 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
6402 | ||
b6c40d68 PM |
6403 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
6404 | { | |
d314774c SH |
6405 | const struct net_device_ops *ops = dev->netdev_ops; |
6406 | ||
d2615bf4 | 6407 | if (ops->ndo_change_rx_flags) |
d314774c | 6408 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
6409 | } |
6410 | ||
991fb3f7 | 6411 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6412 | { |
b536db93 | 6413 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
6414 | kuid_t uid; |
6415 | kgid_t gid; | |
1da177e4 | 6416 | |
24023451 PM |
6417 | ASSERT_RTNL(); |
6418 | ||
dad9b335 WC |
6419 | dev->flags |= IFF_PROMISC; |
6420 | dev->promiscuity += inc; | |
6421 | if (dev->promiscuity == 0) { | |
6422 | /* | |
6423 | * Avoid overflow. | |
6424 | * If inc causes overflow, untouch promisc and return error. | |
6425 | */ | |
6426 | if (inc < 0) | |
6427 | dev->flags &= ~IFF_PROMISC; | |
6428 | else { | |
6429 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
6430 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
6431 | dev->name); | |
dad9b335 WC |
6432 | return -EOVERFLOW; |
6433 | } | |
6434 | } | |
52609c0b | 6435 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
6436 | pr_info("device %s %s promiscuous mode\n", |
6437 | dev->name, | |
6438 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
6439 | if (audit_enabled) { |
6440 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
6441 | audit_log(current->audit_context, GFP_ATOMIC, |
6442 | AUDIT_ANOM_PROMISCUOUS, | |
6443 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
6444 | dev->name, (dev->flags & IFF_PROMISC), | |
6445 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 6446 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
6447 | from_kuid(&init_user_ns, uid), |
6448 | from_kgid(&init_user_ns, gid), | |
7759db82 | 6449 | audit_get_sessionid(current)); |
8192b0c4 | 6450 | } |
24023451 | 6451 | |
b6c40d68 | 6452 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 6453 | } |
991fb3f7 ND |
6454 | if (notify) |
6455 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 6456 | return 0; |
1da177e4 LT |
6457 | } |
6458 | ||
4417da66 PM |
6459 | /** |
6460 | * dev_set_promiscuity - update promiscuity count on a device | |
6461 | * @dev: device | |
6462 | * @inc: modifier | |
6463 | * | |
6464 | * Add or remove promiscuity from a device. While the count in the device | |
6465 | * remains above zero the interface remains promiscuous. Once it hits zero | |
6466 | * the device reverts back to normal filtering operation. A negative inc | |
6467 | * value is used to drop promiscuity on the device. | |
dad9b335 | 6468 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 6469 | */ |
dad9b335 | 6470 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 6471 | { |
b536db93 | 6472 | unsigned int old_flags = dev->flags; |
dad9b335 | 6473 | int err; |
4417da66 | 6474 | |
991fb3f7 | 6475 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 6476 | if (err < 0) |
dad9b335 | 6477 | return err; |
4417da66 PM |
6478 | if (dev->flags != old_flags) |
6479 | dev_set_rx_mode(dev); | |
dad9b335 | 6480 | return err; |
4417da66 | 6481 | } |
d1b19dff | 6482 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 6483 | |
991fb3f7 | 6484 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6485 | { |
991fb3f7 | 6486 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 6487 | |
24023451 PM |
6488 | ASSERT_RTNL(); |
6489 | ||
1da177e4 | 6490 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
6491 | dev->allmulti += inc; |
6492 | if (dev->allmulti == 0) { | |
6493 | /* | |
6494 | * Avoid overflow. | |
6495 | * If inc causes overflow, untouch allmulti and return error. | |
6496 | */ | |
6497 | if (inc < 0) | |
6498 | dev->flags &= ~IFF_ALLMULTI; | |
6499 | else { | |
6500 | dev->allmulti -= inc; | |
7b6cd1ce JP |
6501 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
6502 | dev->name); | |
dad9b335 WC |
6503 | return -EOVERFLOW; |
6504 | } | |
6505 | } | |
24023451 | 6506 | if (dev->flags ^ old_flags) { |
b6c40d68 | 6507 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 6508 | dev_set_rx_mode(dev); |
991fb3f7 ND |
6509 | if (notify) |
6510 | __dev_notify_flags(dev, old_flags, | |
6511 | dev->gflags ^ old_gflags); | |
24023451 | 6512 | } |
dad9b335 | 6513 | return 0; |
4417da66 | 6514 | } |
991fb3f7 ND |
6515 | |
6516 | /** | |
6517 | * dev_set_allmulti - update allmulti count on a device | |
6518 | * @dev: device | |
6519 | * @inc: modifier | |
6520 | * | |
6521 | * Add or remove reception of all multicast frames to a device. While the | |
6522 | * count in the device remains above zero the interface remains listening | |
6523 | * to all interfaces. Once it hits zero the device reverts back to normal | |
6524 | * filtering operation. A negative @inc value is used to drop the counter | |
6525 | * when releasing a resource needing all multicasts. | |
6526 | * Return 0 if successful or a negative errno code on error. | |
6527 | */ | |
6528 | ||
6529 | int dev_set_allmulti(struct net_device *dev, int inc) | |
6530 | { | |
6531 | return __dev_set_allmulti(dev, inc, true); | |
6532 | } | |
d1b19dff | 6533 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
6534 | |
6535 | /* | |
6536 | * Upload unicast and multicast address lists to device and | |
6537 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 6538 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
6539 | * are present. |
6540 | */ | |
6541 | void __dev_set_rx_mode(struct net_device *dev) | |
6542 | { | |
d314774c SH |
6543 | const struct net_device_ops *ops = dev->netdev_ops; |
6544 | ||
4417da66 PM |
6545 | /* dev_open will call this function so the list will stay sane. */ |
6546 | if (!(dev->flags&IFF_UP)) | |
6547 | return; | |
6548 | ||
6549 | if (!netif_device_present(dev)) | |
40b77c94 | 6550 | return; |
4417da66 | 6551 | |
01789349 | 6552 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
6553 | /* Unicast addresses changes may only happen under the rtnl, |
6554 | * therefore calling __dev_set_promiscuity here is safe. | |
6555 | */ | |
32e7bfc4 | 6556 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 6557 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 6558 | dev->uc_promisc = true; |
32e7bfc4 | 6559 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 6560 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 6561 | dev->uc_promisc = false; |
4417da66 | 6562 | } |
4417da66 | 6563 | } |
01789349 JP |
6564 | |
6565 | if (ops->ndo_set_rx_mode) | |
6566 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
6567 | } |
6568 | ||
6569 | void dev_set_rx_mode(struct net_device *dev) | |
6570 | { | |
b9e40857 | 6571 | netif_addr_lock_bh(dev); |
4417da66 | 6572 | __dev_set_rx_mode(dev); |
b9e40857 | 6573 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
6574 | } |
6575 | ||
f0db275a SH |
6576 | /** |
6577 | * dev_get_flags - get flags reported to userspace | |
6578 | * @dev: device | |
6579 | * | |
6580 | * Get the combination of flag bits exported through APIs to userspace. | |
6581 | */ | |
95c96174 | 6582 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 6583 | { |
95c96174 | 6584 | unsigned int flags; |
1da177e4 LT |
6585 | |
6586 | flags = (dev->flags & ~(IFF_PROMISC | | |
6587 | IFF_ALLMULTI | | |
b00055aa SR |
6588 | IFF_RUNNING | |
6589 | IFF_LOWER_UP | | |
6590 | IFF_DORMANT)) | | |
1da177e4 LT |
6591 | (dev->gflags & (IFF_PROMISC | |
6592 | IFF_ALLMULTI)); | |
6593 | ||
b00055aa SR |
6594 | if (netif_running(dev)) { |
6595 | if (netif_oper_up(dev)) | |
6596 | flags |= IFF_RUNNING; | |
6597 | if (netif_carrier_ok(dev)) | |
6598 | flags |= IFF_LOWER_UP; | |
6599 | if (netif_dormant(dev)) | |
6600 | flags |= IFF_DORMANT; | |
6601 | } | |
1da177e4 LT |
6602 | |
6603 | return flags; | |
6604 | } | |
d1b19dff | 6605 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 6606 | |
bd380811 | 6607 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 6608 | { |
b536db93 | 6609 | unsigned int old_flags = dev->flags; |
bd380811 | 6610 | int ret; |
1da177e4 | 6611 | |
24023451 PM |
6612 | ASSERT_RTNL(); |
6613 | ||
1da177e4 LT |
6614 | /* |
6615 | * Set the flags on our device. | |
6616 | */ | |
6617 | ||
6618 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
6619 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
6620 | IFF_AUTOMEDIA)) | | |
6621 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
6622 | IFF_ALLMULTI)); | |
6623 | ||
6624 | /* | |
6625 | * Load in the correct multicast list now the flags have changed. | |
6626 | */ | |
6627 | ||
b6c40d68 PM |
6628 | if ((old_flags ^ flags) & IFF_MULTICAST) |
6629 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 6630 | |
4417da66 | 6631 | dev_set_rx_mode(dev); |
1da177e4 LT |
6632 | |
6633 | /* | |
6634 | * Have we downed the interface. We handle IFF_UP ourselves | |
6635 | * according to user attempts to set it, rather than blindly | |
6636 | * setting it. | |
6637 | */ | |
6638 | ||
6639 | ret = 0; | |
d215d10f | 6640 | if ((old_flags ^ flags) & IFF_UP) |
bd380811 | 6641 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 | 6642 | |
1da177e4 | 6643 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 6644 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 6645 | unsigned int old_flags = dev->flags; |
d1b19dff | 6646 | |
1da177e4 | 6647 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
6648 | |
6649 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
6650 | if (dev->flags != old_flags) | |
6651 | dev_set_rx_mode(dev); | |
1da177e4 LT |
6652 | } |
6653 | ||
6654 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 6655 | * is important. Some (broken) drivers set IFF_PROMISC, when |
6656 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
6657 | */ |
6658 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
6659 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
6660 | ||
1da177e4 | 6661 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 6662 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
6663 | } |
6664 | ||
bd380811 PM |
6665 | return ret; |
6666 | } | |
6667 | ||
a528c219 ND |
6668 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
6669 | unsigned int gchanges) | |
bd380811 PM |
6670 | { |
6671 | unsigned int changes = dev->flags ^ old_flags; | |
6672 | ||
a528c219 | 6673 | if (gchanges) |
7f294054 | 6674 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 6675 | |
bd380811 PM |
6676 | if (changes & IFF_UP) { |
6677 | if (dev->flags & IFF_UP) | |
6678 | call_netdevice_notifiers(NETDEV_UP, dev); | |
6679 | else | |
6680 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
6681 | } | |
6682 | ||
6683 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
6684 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
6685 | struct netdev_notifier_change_info change_info; | |
6686 | ||
6687 | change_info.flags_changed = changes; | |
6688 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
6689 | &change_info.info); | |
6690 | } | |
bd380811 PM |
6691 | } |
6692 | ||
6693 | /** | |
6694 | * dev_change_flags - change device settings | |
6695 | * @dev: device | |
6696 | * @flags: device state flags | |
6697 | * | |
6698 | * Change settings on device based state flags. The flags are | |
6699 | * in the userspace exported format. | |
6700 | */ | |
b536db93 | 6701 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 6702 | { |
b536db93 | 6703 | int ret; |
991fb3f7 | 6704 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
6705 | |
6706 | ret = __dev_change_flags(dev, flags); | |
6707 | if (ret < 0) | |
6708 | return ret; | |
6709 | ||
991fb3f7 | 6710 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 6711 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
6712 | return ret; |
6713 | } | |
d1b19dff | 6714 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 6715 | |
2315dc91 VF |
6716 | static int __dev_set_mtu(struct net_device *dev, int new_mtu) |
6717 | { | |
6718 | const struct net_device_ops *ops = dev->netdev_ops; | |
6719 | ||
6720 | if (ops->ndo_change_mtu) | |
6721 | return ops->ndo_change_mtu(dev, new_mtu); | |
6722 | ||
6723 | dev->mtu = new_mtu; | |
6724 | return 0; | |
6725 | } | |
6726 | ||
f0db275a SH |
6727 | /** |
6728 | * dev_set_mtu - Change maximum transfer unit | |
6729 | * @dev: device | |
6730 | * @new_mtu: new transfer unit | |
6731 | * | |
6732 | * Change the maximum transfer size of the network device. | |
6733 | */ | |
1da177e4 LT |
6734 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
6735 | { | |
2315dc91 | 6736 | int err, orig_mtu; |
1da177e4 LT |
6737 | |
6738 | if (new_mtu == dev->mtu) | |
6739 | return 0; | |
6740 | ||
61e84623 JW |
6741 | /* MTU must be positive, and in range */ |
6742 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
6743 | net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n", | |
6744 | dev->name, new_mtu, dev->min_mtu); | |
1da177e4 | 6745 | return -EINVAL; |
61e84623 JW |
6746 | } |
6747 | ||
6748 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
6749 | net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n", | |
a0e65de7 | 6750 | dev->name, new_mtu, dev->max_mtu); |
61e84623 JW |
6751 | return -EINVAL; |
6752 | } | |
1da177e4 LT |
6753 | |
6754 | if (!netif_device_present(dev)) | |
6755 | return -ENODEV; | |
6756 | ||
1d486bfb VF |
6757 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
6758 | err = notifier_to_errno(err); | |
6759 | if (err) | |
6760 | return err; | |
d314774c | 6761 | |
2315dc91 VF |
6762 | orig_mtu = dev->mtu; |
6763 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 6764 | |
2315dc91 VF |
6765 | if (!err) { |
6766 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6767 | err = notifier_to_errno(err); | |
6768 | if (err) { | |
6769 | /* setting mtu back and notifying everyone again, | |
6770 | * so that they have a chance to revert changes. | |
6771 | */ | |
6772 | __dev_set_mtu(dev, orig_mtu); | |
6773 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6774 | } | |
6775 | } | |
1da177e4 LT |
6776 | return err; |
6777 | } | |
d1b19dff | 6778 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 6779 | |
cbda10fa VD |
6780 | /** |
6781 | * dev_set_group - Change group this device belongs to | |
6782 | * @dev: device | |
6783 | * @new_group: group this device should belong to | |
6784 | */ | |
6785 | void dev_set_group(struct net_device *dev, int new_group) | |
6786 | { | |
6787 | dev->group = new_group; | |
6788 | } | |
6789 | EXPORT_SYMBOL(dev_set_group); | |
6790 | ||
f0db275a SH |
6791 | /** |
6792 | * dev_set_mac_address - Change Media Access Control Address | |
6793 | * @dev: device | |
6794 | * @sa: new address | |
6795 | * | |
6796 | * Change the hardware (MAC) address of the device | |
6797 | */ | |
1da177e4 LT |
6798 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
6799 | { | |
d314774c | 6800 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
6801 | int err; |
6802 | ||
d314774c | 6803 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
6804 | return -EOPNOTSUPP; |
6805 | if (sa->sa_family != dev->type) | |
6806 | return -EINVAL; | |
6807 | if (!netif_device_present(dev)) | |
6808 | return -ENODEV; | |
d314774c | 6809 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
6810 | if (err) |
6811 | return err; | |
fbdeca2d | 6812 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 6813 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 6814 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 6815 | return 0; |
1da177e4 | 6816 | } |
d1b19dff | 6817 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 6818 | |
4bf84c35 JP |
6819 | /** |
6820 | * dev_change_carrier - Change device carrier | |
6821 | * @dev: device | |
691b3b7e | 6822 | * @new_carrier: new value |
4bf84c35 JP |
6823 | * |
6824 | * Change device carrier | |
6825 | */ | |
6826 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
6827 | { | |
6828 | const struct net_device_ops *ops = dev->netdev_ops; | |
6829 | ||
6830 | if (!ops->ndo_change_carrier) | |
6831 | return -EOPNOTSUPP; | |
6832 | if (!netif_device_present(dev)) | |
6833 | return -ENODEV; | |
6834 | return ops->ndo_change_carrier(dev, new_carrier); | |
6835 | } | |
6836 | EXPORT_SYMBOL(dev_change_carrier); | |
6837 | ||
66b52b0d JP |
6838 | /** |
6839 | * dev_get_phys_port_id - Get device physical port ID | |
6840 | * @dev: device | |
6841 | * @ppid: port ID | |
6842 | * | |
6843 | * Get device physical port ID | |
6844 | */ | |
6845 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 6846 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
6847 | { |
6848 | const struct net_device_ops *ops = dev->netdev_ops; | |
6849 | ||
6850 | if (!ops->ndo_get_phys_port_id) | |
6851 | return -EOPNOTSUPP; | |
6852 | return ops->ndo_get_phys_port_id(dev, ppid); | |
6853 | } | |
6854 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
6855 | ||
db24a904 DA |
6856 | /** |
6857 | * dev_get_phys_port_name - Get device physical port name | |
6858 | * @dev: device | |
6859 | * @name: port name | |
ed49e650 | 6860 | * @len: limit of bytes to copy to name |
db24a904 DA |
6861 | * |
6862 | * Get device physical port name | |
6863 | */ | |
6864 | int dev_get_phys_port_name(struct net_device *dev, | |
6865 | char *name, size_t len) | |
6866 | { | |
6867 | const struct net_device_ops *ops = dev->netdev_ops; | |
6868 | ||
6869 | if (!ops->ndo_get_phys_port_name) | |
6870 | return -EOPNOTSUPP; | |
6871 | return ops->ndo_get_phys_port_name(dev, name, len); | |
6872 | } | |
6873 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
6874 | ||
d746d707 AK |
6875 | /** |
6876 | * dev_change_proto_down - update protocol port state information | |
6877 | * @dev: device | |
6878 | * @proto_down: new value | |
6879 | * | |
6880 | * This info can be used by switch drivers to set the phys state of the | |
6881 | * port. | |
6882 | */ | |
6883 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
6884 | { | |
6885 | const struct net_device_ops *ops = dev->netdev_ops; | |
6886 | ||
6887 | if (!ops->ndo_change_proto_down) | |
6888 | return -EOPNOTSUPP; | |
6889 | if (!netif_device_present(dev)) | |
6890 | return -ENODEV; | |
6891 | return ops->ndo_change_proto_down(dev, proto_down); | |
6892 | } | |
6893 | EXPORT_SYMBOL(dev_change_proto_down); | |
6894 | ||
d67b9cd2 DB |
6895 | bool __dev_xdp_attached(struct net_device *dev, xdp_op_t xdp_op) |
6896 | { | |
6897 | struct netdev_xdp xdp; | |
6898 | ||
6899 | memset(&xdp, 0, sizeof(xdp)); | |
6900 | xdp.command = XDP_QUERY_PROG; | |
6901 | ||
6902 | /* Query must always succeed. */ | |
6903 | WARN_ON(xdp_op(dev, &xdp) < 0); | |
6904 | return xdp.prog_attached; | |
6905 | } | |
6906 | ||
6907 | static int dev_xdp_install(struct net_device *dev, xdp_op_t xdp_op, | |
6908 | struct netlink_ext_ack *extack, | |
6909 | struct bpf_prog *prog) | |
6910 | { | |
6911 | struct netdev_xdp xdp; | |
6912 | ||
6913 | memset(&xdp, 0, sizeof(xdp)); | |
6914 | xdp.command = XDP_SETUP_PROG; | |
6915 | xdp.extack = extack; | |
6916 | xdp.prog = prog; | |
6917 | ||
6918 | return xdp_op(dev, &xdp); | |
6919 | } | |
6920 | ||
a7862b45 BB |
6921 | /** |
6922 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
6923 | * @dev: device | |
b5d60989 | 6924 | * @extack: netlink extended ack |
a7862b45 | 6925 | * @fd: new program fd or negative value to clear |
85de8576 | 6926 | * @flags: xdp-related flags |
a7862b45 BB |
6927 | * |
6928 | * Set or clear a bpf program for a device | |
6929 | */ | |
ddf9f970 JK |
6930 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, |
6931 | int fd, u32 flags) | |
a7862b45 BB |
6932 | { |
6933 | const struct net_device_ops *ops = dev->netdev_ops; | |
6934 | struct bpf_prog *prog = NULL; | |
d67b9cd2 | 6935 | xdp_op_t xdp_op, xdp_chk; |
a7862b45 BB |
6936 | int err; |
6937 | ||
85de8576 DB |
6938 | ASSERT_RTNL(); |
6939 | ||
d67b9cd2 | 6940 | xdp_op = xdp_chk = ops->ndo_xdp; |
0489df9a DB |
6941 | if (!xdp_op && (flags & XDP_FLAGS_DRV_MODE)) |
6942 | return -EOPNOTSUPP; | |
b5cdae32 DM |
6943 | if (!xdp_op || (flags & XDP_FLAGS_SKB_MODE)) |
6944 | xdp_op = generic_xdp_install; | |
d67b9cd2 DB |
6945 | if (xdp_op == xdp_chk) |
6946 | xdp_chk = generic_xdp_install; | |
b5cdae32 | 6947 | |
a7862b45 | 6948 | if (fd >= 0) { |
d67b9cd2 DB |
6949 | if (xdp_chk && __dev_xdp_attached(dev, xdp_chk)) |
6950 | return -EEXIST; | |
6951 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && | |
6952 | __dev_xdp_attached(dev, xdp_op)) | |
6953 | return -EBUSY; | |
85de8576 | 6954 | |
a7862b45 BB |
6955 | prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP); |
6956 | if (IS_ERR(prog)) | |
6957 | return PTR_ERR(prog); | |
6958 | } | |
6959 | ||
d67b9cd2 | 6960 | err = dev_xdp_install(dev, xdp_op, extack, prog); |
a7862b45 BB |
6961 | if (err < 0 && prog) |
6962 | bpf_prog_put(prog); | |
6963 | ||
6964 | return err; | |
6965 | } | |
a7862b45 | 6966 | |
1da177e4 LT |
6967 | /** |
6968 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 6969 | * @net: the applicable net namespace |
1da177e4 LT |
6970 | * |
6971 | * Returns a suitable unique value for a new device interface | |
6972 | * number. The caller must hold the rtnl semaphore or the | |
6973 | * dev_base_lock to be sure it remains unique. | |
6974 | */ | |
881d966b | 6975 | static int dev_new_index(struct net *net) |
1da177e4 | 6976 | { |
aa79e66e | 6977 | int ifindex = net->ifindex; |
f4563a75 | 6978 | |
1da177e4 LT |
6979 | for (;;) { |
6980 | if (++ifindex <= 0) | |
6981 | ifindex = 1; | |
881d966b | 6982 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 6983 | return net->ifindex = ifindex; |
1da177e4 LT |
6984 | } |
6985 | } | |
6986 | ||
1da177e4 | 6987 | /* Delayed registration/unregisteration */ |
3b5b34fd | 6988 | static LIST_HEAD(net_todo_list); |
200b916f | 6989 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 6990 | |
6f05f629 | 6991 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 6992 | { |
1da177e4 | 6993 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 6994 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
6995 | } |
6996 | ||
9b5e383c | 6997 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 6998 | { |
e93737b0 | 6999 | struct net_device *dev, *tmp; |
5cde2829 | 7000 | LIST_HEAD(close_head); |
9b5e383c | 7001 | |
93ee31f1 DL |
7002 | BUG_ON(dev_boot_phase); |
7003 | ASSERT_RTNL(); | |
7004 | ||
e93737b0 | 7005 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 7006 | /* Some devices call without registering |
e93737b0 KK |
7007 | * for initialization unwind. Remove those |
7008 | * devices and proceed with the remaining. | |
9b5e383c ED |
7009 | */ |
7010 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
7011 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
7012 | dev->name, dev); | |
93ee31f1 | 7013 | |
9b5e383c | 7014 | WARN_ON(1); |
e93737b0 KK |
7015 | list_del(&dev->unreg_list); |
7016 | continue; | |
9b5e383c | 7017 | } |
449f4544 | 7018 | dev->dismantle = true; |
9b5e383c | 7019 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 7020 | } |
93ee31f1 | 7021 | |
44345724 | 7022 | /* If device is running, close it first. */ |
5cde2829 EB |
7023 | list_for_each_entry(dev, head, unreg_list) |
7024 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 7025 | dev_close_many(&close_head, true); |
93ee31f1 | 7026 | |
44345724 | 7027 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
7028 | /* And unlink it from device chain. */ |
7029 | unlist_netdevice(dev); | |
93ee31f1 | 7030 | |
9b5e383c ED |
7031 | dev->reg_state = NETREG_UNREGISTERING; |
7032 | } | |
41852497 | 7033 | flush_all_backlogs(); |
93ee31f1 DL |
7034 | |
7035 | synchronize_net(); | |
7036 | ||
9b5e383c | 7037 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
7038 | struct sk_buff *skb = NULL; |
7039 | ||
9b5e383c ED |
7040 | /* Shutdown queueing discipline. */ |
7041 | dev_shutdown(dev); | |
93ee31f1 DL |
7042 | |
7043 | ||
9b5e383c | 7044 | /* Notify protocols, that we are about to destroy |
eb13da1a | 7045 | * this device. They should clean all the things. |
7046 | */ | |
9b5e383c | 7047 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
93ee31f1 | 7048 | |
395eea6c MB |
7049 | if (!dev->rtnl_link_ops || |
7050 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7051 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, | |
7052 | GFP_KERNEL); | |
7053 | ||
9b5e383c ED |
7054 | /* |
7055 | * Flush the unicast and multicast chains | |
7056 | */ | |
a748ee24 | 7057 | dev_uc_flush(dev); |
22bedad3 | 7058 | dev_mc_flush(dev); |
93ee31f1 | 7059 | |
9b5e383c ED |
7060 | if (dev->netdev_ops->ndo_uninit) |
7061 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 7062 | |
395eea6c MB |
7063 | if (skb) |
7064 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 7065 | |
9ff162a8 JP |
7066 | /* Notifier chain MUST detach us all upper devices. */ |
7067 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
0f524a80 | 7068 | WARN_ON(netdev_has_any_lower_dev(dev)); |
93ee31f1 | 7069 | |
9b5e383c ED |
7070 | /* Remove entries from kobject tree */ |
7071 | netdev_unregister_kobject(dev); | |
024e9679 AD |
7072 | #ifdef CONFIG_XPS |
7073 | /* Remove XPS queueing entries */ | |
7074 | netif_reset_xps_queues_gt(dev, 0); | |
7075 | #endif | |
9b5e383c | 7076 | } |
93ee31f1 | 7077 | |
850a545b | 7078 | synchronize_net(); |
395264d5 | 7079 | |
a5ee1551 | 7080 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
7081 | dev_put(dev); |
7082 | } | |
7083 | ||
7084 | static void rollback_registered(struct net_device *dev) | |
7085 | { | |
7086 | LIST_HEAD(single); | |
7087 | ||
7088 | list_add(&dev->unreg_list, &single); | |
7089 | rollback_registered_many(&single); | |
ceaaec98 | 7090 | list_del(&single); |
93ee31f1 DL |
7091 | } |
7092 | ||
fd867d51 JW |
7093 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
7094 | struct net_device *upper, netdev_features_t features) | |
7095 | { | |
7096 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
7097 | netdev_features_t feature; | |
5ba3f7d6 | 7098 | int feature_bit; |
fd867d51 | 7099 | |
5ba3f7d6 JW |
7100 | for_each_netdev_feature(&upper_disables, feature_bit) { |
7101 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
7102 | if (!(upper->wanted_features & feature) |
7103 | && (features & feature)) { | |
7104 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
7105 | &feature, upper->name); | |
7106 | features &= ~feature; | |
7107 | } | |
7108 | } | |
7109 | ||
7110 | return features; | |
7111 | } | |
7112 | ||
7113 | static void netdev_sync_lower_features(struct net_device *upper, | |
7114 | struct net_device *lower, netdev_features_t features) | |
7115 | { | |
7116 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
7117 | netdev_features_t feature; | |
5ba3f7d6 | 7118 | int feature_bit; |
fd867d51 | 7119 | |
5ba3f7d6 JW |
7120 | for_each_netdev_feature(&upper_disables, feature_bit) { |
7121 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
7122 | if (!(features & feature) && (lower->features & feature)) { |
7123 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
7124 | &feature, lower->name); | |
7125 | lower->wanted_features &= ~feature; | |
7126 | netdev_update_features(lower); | |
7127 | ||
7128 | if (unlikely(lower->features & feature)) | |
7129 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
7130 | &feature, lower->name); | |
7131 | } | |
7132 | } | |
7133 | } | |
7134 | ||
c8f44aff MM |
7135 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
7136 | netdev_features_t features) | |
b63365a2 | 7137 | { |
57422dc5 MM |
7138 | /* Fix illegal checksum combinations */ |
7139 | if ((features & NETIF_F_HW_CSUM) && | |
7140 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 7141 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
7142 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
7143 | } | |
7144 | ||
b63365a2 | 7145 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 7146 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 7147 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 7148 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
7149 | } |
7150 | ||
ec5f0615 PS |
7151 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
7152 | !(features & NETIF_F_IP_CSUM)) { | |
7153 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
7154 | features &= ~NETIF_F_TSO; | |
7155 | features &= ~NETIF_F_TSO_ECN; | |
7156 | } | |
7157 | ||
7158 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
7159 | !(features & NETIF_F_IPV6_CSUM)) { | |
7160 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
7161 | features &= ~NETIF_F_TSO6; | |
7162 | } | |
7163 | ||
b1dc497b AD |
7164 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
7165 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
7166 | features &= ~NETIF_F_TSO_MANGLEID; | |
7167 | ||
31d8b9e0 BH |
7168 | /* TSO ECN requires that TSO is present as well. */ |
7169 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
7170 | features &= ~NETIF_F_TSO_ECN; | |
7171 | ||
212b573f MM |
7172 | /* Software GSO depends on SG. */ |
7173 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 7174 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
7175 | features &= ~NETIF_F_GSO; |
7176 | } | |
7177 | ||
acd1130e | 7178 | /* UFO needs SG and checksumming */ |
b63365a2 | 7179 | if (features & NETIF_F_UFO) { |
79032644 | 7180 | /* maybe split UFO into V4 and V6? */ |
c8cd0989 TH |
7181 | if (!(features & NETIF_F_HW_CSUM) && |
7182 | ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) != | |
7183 | (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 7184 | netdev_dbg(dev, |
acd1130e | 7185 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
7186 | features &= ~NETIF_F_UFO; |
7187 | } | |
7188 | ||
7189 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 7190 | netdev_dbg(dev, |
acd1130e | 7191 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
7192 | features &= ~NETIF_F_UFO; |
7193 | } | |
7194 | } | |
7195 | ||
802ab55a AD |
7196 | /* GSO partial features require GSO partial be set */ |
7197 | if ((features & dev->gso_partial_features) && | |
7198 | !(features & NETIF_F_GSO_PARTIAL)) { | |
7199 | netdev_dbg(dev, | |
7200 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
7201 | features &= ~dev->gso_partial_features; | |
7202 | } | |
7203 | ||
b63365a2 HX |
7204 | return features; |
7205 | } | |
b63365a2 | 7206 | |
6cb6a27c | 7207 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 7208 | { |
fd867d51 | 7209 | struct net_device *upper, *lower; |
c8f44aff | 7210 | netdev_features_t features; |
fd867d51 | 7211 | struct list_head *iter; |
e7868a85 | 7212 | int err = -1; |
5455c699 | 7213 | |
87267485 MM |
7214 | ASSERT_RTNL(); |
7215 | ||
5455c699 MM |
7216 | features = netdev_get_wanted_features(dev); |
7217 | ||
7218 | if (dev->netdev_ops->ndo_fix_features) | |
7219 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
7220 | ||
7221 | /* driver might be less strict about feature dependencies */ | |
7222 | features = netdev_fix_features(dev, features); | |
7223 | ||
fd867d51 JW |
7224 | /* some features can't be enabled if they're off an an upper device */ |
7225 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
7226 | features = netdev_sync_upper_features(dev, upper, features); | |
7227 | ||
5455c699 | 7228 | if (dev->features == features) |
e7868a85 | 7229 | goto sync_lower; |
5455c699 | 7230 | |
c8f44aff MM |
7231 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
7232 | &dev->features, &features); | |
5455c699 MM |
7233 | |
7234 | if (dev->netdev_ops->ndo_set_features) | |
7235 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
7236 | else |
7237 | err = 0; | |
5455c699 | 7238 | |
6cb6a27c | 7239 | if (unlikely(err < 0)) { |
5455c699 | 7240 | netdev_err(dev, |
c8f44aff MM |
7241 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
7242 | err, &features, &dev->features); | |
17b85d29 NA |
7243 | /* return non-0 since some features might have changed and |
7244 | * it's better to fire a spurious notification than miss it | |
7245 | */ | |
7246 | return -1; | |
6cb6a27c MM |
7247 | } |
7248 | ||
e7868a85 | 7249 | sync_lower: |
fd867d51 JW |
7250 | /* some features must be disabled on lower devices when disabled |
7251 | * on an upper device (think: bonding master or bridge) | |
7252 | */ | |
7253 | netdev_for_each_lower_dev(dev, lower, iter) | |
7254 | netdev_sync_lower_features(dev, lower, features); | |
7255 | ||
6cb6a27c MM |
7256 | if (!err) |
7257 | dev->features = features; | |
7258 | ||
e7868a85 | 7259 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
7260 | } |
7261 | ||
afe12cc8 MM |
7262 | /** |
7263 | * netdev_update_features - recalculate device features | |
7264 | * @dev: the device to check | |
7265 | * | |
7266 | * Recalculate dev->features set and send notifications if it | |
7267 | * has changed. Should be called after driver or hardware dependent | |
7268 | * conditions might have changed that influence the features. | |
7269 | */ | |
6cb6a27c MM |
7270 | void netdev_update_features(struct net_device *dev) |
7271 | { | |
7272 | if (__netdev_update_features(dev)) | |
7273 | netdev_features_change(dev); | |
5455c699 MM |
7274 | } |
7275 | EXPORT_SYMBOL(netdev_update_features); | |
7276 | ||
afe12cc8 MM |
7277 | /** |
7278 | * netdev_change_features - recalculate device features | |
7279 | * @dev: the device to check | |
7280 | * | |
7281 | * Recalculate dev->features set and send notifications even | |
7282 | * if they have not changed. Should be called instead of | |
7283 | * netdev_update_features() if also dev->vlan_features might | |
7284 | * have changed to allow the changes to be propagated to stacked | |
7285 | * VLAN devices. | |
7286 | */ | |
7287 | void netdev_change_features(struct net_device *dev) | |
7288 | { | |
7289 | __netdev_update_features(dev); | |
7290 | netdev_features_change(dev); | |
7291 | } | |
7292 | EXPORT_SYMBOL(netdev_change_features); | |
7293 | ||
fc4a7489 PM |
7294 | /** |
7295 | * netif_stacked_transfer_operstate - transfer operstate | |
7296 | * @rootdev: the root or lower level device to transfer state from | |
7297 | * @dev: the device to transfer operstate to | |
7298 | * | |
7299 | * Transfer operational state from root to device. This is normally | |
7300 | * called when a stacking relationship exists between the root | |
7301 | * device and the device(a leaf device). | |
7302 | */ | |
7303 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
7304 | struct net_device *dev) | |
7305 | { | |
7306 | if (rootdev->operstate == IF_OPER_DORMANT) | |
7307 | netif_dormant_on(dev); | |
7308 | else | |
7309 | netif_dormant_off(dev); | |
7310 | ||
0575c86b ZS |
7311 | if (netif_carrier_ok(rootdev)) |
7312 | netif_carrier_on(dev); | |
7313 | else | |
7314 | netif_carrier_off(dev); | |
fc4a7489 PM |
7315 | } |
7316 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
7317 | ||
a953be53 | 7318 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
7319 | static int netif_alloc_rx_queues(struct net_device *dev) |
7320 | { | |
1b4bf461 | 7321 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 7322 | struct netdev_rx_queue *rx; |
10595902 | 7323 | size_t sz = count * sizeof(*rx); |
1b4bf461 | 7324 | |
bd25fa7b | 7325 | BUG_ON(count < 1); |
1b4bf461 | 7326 | |
da6bc57a MH |
7327 | rx = kvzalloc(sz, GFP_KERNEL | __GFP_REPEAT); |
7328 | if (!rx) | |
7329 | return -ENOMEM; | |
7330 | ||
bd25fa7b TH |
7331 | dev->_rx = rx; |
7332 | ||
bd25fa7b | 7333 | for (i = 0; i < count; i++) |
fe822240 | 7334 | rx[i].dev = dev; |
1b4bf461 ED |
7335 | return 0; |
7336 | } | |
bf264145 | 7337 | #endif |
1b4bf461 | 7338 | |
aa942104 CG |
7339 | static void netdev_init_one_queue(struct net_device *dev, |
7340 | struct netdev_queue *queue, void *_unused) | |
7341 | { | |
7342 | /* Initialize queue lock */ | |
7343 | spin_lock_init(&queue->_xmit_lock); | |
7344 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
7345 | queue->xmit_lock_owner = -1; | |
b236da69 | 7346 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 7347 | queue->dev = dev; |
114cf580 TH |
7348 | #ifdef CONFIG_BQL |
7349 | dql_init(&queue->dql, HZ); | |
7350 | #endif | |
aa942104 CG |
7351 | } |
7352 | ||
60877a32 ED |
7353 | static void netif_free_tx_queues(struct net_device *dev) |
7354 | { | |
4cb28970 | 7355 | kvfree(dev->_tx); |
60877a32 ED |
7356 | } |
7357 | ||
e6484930 TH |
7358 | static int netif_alloc_netdev_queues(struct net_device *dev) |
7359 | { | |
7360 | unsigned int count = dev->num_tx_queues; | |
7361 | struct netdev_queue *tx; | |
60877a32 | 7362 | size_t sz = count * sizeof(*tx); |
e6484930 | 7363 | |
d339727c ED |
7364 | if (count < 1 || count > 0xffff) |
7365 | return -EINVAL; | |
62b5942a | 7366 | |
da6bc57a MH |
7367 | tx = kvzalloc(sz, GFP_KERNEL | __GFP_REPEAT); |
7368 | if (!tx) | |
7369 | return -ENOMEM; | |
7370 | ||
e6484930 | 7371 | dev->_tx = tx; |
1d24eb48 | 7372 | |
e6484930 TH |
7373 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
7374 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
7375 | |
7376 | return 0; | |
e6484930 TH |
7377 | } |
7378 | ||
a2029240 DV |
7379 | void netif_tx_stop_all_queues(struct net_device *dev) |
7380 | { | |
7381 | unsigned int i; | |
7382 | ||
7383 | for (i = 0; i < dev->num_tx_queues; i++) { | |
7384 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 7385 | |
a2029240 DV |
7386 | netif_tx_stop_queue(txq); |
7387 | } | |
7388 | } | |
7389 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
7390 | ||
1da177e4 LT |
7391 | /** |
7392 | * register_netdevice - register a network device | |
7393 | * @dev: device to register | |
7394 | * | |
7395 | * Take a completed network device structure and add it to the kernel | |
7396 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7397 | * chain. 0 is returned on success. A negative errno code is returned | |
7398 | * on a failure to set up the device, or if the name is a duplicate. | |
7399 | * | |
7400 | * Callers must hold the rtnl semaphore. You may want | |
7401 | * register_netdev() instead of this. | |
7402 | * | |
7403 | * BUGS: | |
7404 | * The locking appears insufficient to guarantee two parallel registers | |
7405 | * will not get the same name. | |
7406 | */ | |
7407 | ||
7408 | int register_netdevice(struct net_device *dev) | |
7409 | { | |
1da177e4 | 7410 | int ret; |
d314774c | 7411 | struct net *net = dev_net(dev); |
1da177e4 LT |
7412 | |
7413 | BUG_ON(dev_boot_phase); | |
7414 | ASSERT_RTNL(); | |
7415 | ||
b17a7c17 SH |
7416 | might_sleep(); |
7417 | ||
1da177e4 LT |
7418 | /* When net_device's are persistent, this will be fatal. */ |
7419 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 7420 | BUG_ON(!net); |
1da177e4 | 7421 | |
f1f28aa3 | 7422 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 7423 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 7424 | |
828de4f6 | 7425 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
7426 | if (ret < 0) |
7427 | goto out; | |
7428 | ||
1da177e4 | 7429 | /* Init, if this function is available */ |
d314774c SH |
7430 | if (dev->netdev_ops->ndo_init) { |
7431 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
7432 | if (ret) { |
7433 | if (ret > 0) | |
7434 | ret = -EIO; | |
90833aa4 | 7435 | goto out; |
1da177e4 LT |
7436 | } |
7437 | } | |
4ec93edb | 7438 | |
f646968f PM |
7439 | if (((dev->hw_features | dev->features) & |
7440 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
7441 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
7442 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
7443 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
7444 | ret = -EINVAL; | |
7445 | goto err_uninit; | |
7446 | } | |
7447 | ||
9c7dafbf PE |
7448 | ret = -EBUSY; |
7449 | if (!dev->ifindex) | |
7450 | dev->ifindex = dev_new_index(net); | |
7451 | else if (__dev_get_by_index(net, dev->ifindex)) | |
7452 | goto err_uninit; | |
7453 | ||
5455c699 MM |
7454 | /* Transfer changeable features to wanted_features and enable |
7455 | * software offloads (GSO and GRO). | |
7456 | */ | |
7457 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
7458 | dev->features |= NETIF_F_SOFT_FEATURES; |
7459 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 7460 | |
cbc53e08 | 7461 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 7462 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 7463 | |
7f348a60 AD |
7464 | /* If IPv4 TCP segmentation offload is supported we should also |
7465 | * allow the device to enable segmenting the frame with the option | |
7466 | * of ignoring a static IP ID value. This doesn't enable the | |
7467 | * feature itself but allows the user to enable it later. | |
7468 | */ | |
cbc53e08 AD |
7469 | if (dev->hw_features & NETIF_F_TSO) |
7470 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
7471 | if (dev->vlan_features & NETIF_F_TSO) |
7472 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
7473 | if (dev->mpls_features & NETIF_F_TSO) | |
7474 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
7475 | if (dev->hw_enc_features & NETIF_F_TSO) | |
7476 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 7477 | |
1180e7d6 | 7478 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 7479 | */ |
1180e7d6 | 7480 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 7481 | |
ee579677 PS |
7482 | /* Make NETIF_F_SG inheritable to tunnel devices. |
7483 | */ | |
802ab55a | 7484 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 7485 | |
0d89d203 SH |
7486 | /* Make NETIF_F_SG inheritable to MPLS. |
7487 | */ | |
7488 | dev->mpls_features |= NETIF_F_SG; | |
7489 | ||
7ffbe3fd JB |
7490 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
7491 | ret = notifier_to_errno(ret); | |
7492 | if (ret) | |
7493 | goto err_uninit; | |
7494 | ||
8b41d188 | 7495 | ret = netdev_register_kobject(dev); |
b17a7c17 | 7496 | if (ret) |
7ce1b0ed | 7497 | goto err_uninit; |
b17a7c17 SH |
7498 | dev->reg_state = NETREG_REGISTERED; |
7499 | ||
6cb6a27c | 7500 | __netdev_update_features(dev); |
8e9b59b2 | 7501 | |
1da177e4 LT |
7502 | /* |
7503 | * Default initial state at registry is that the | |
7504 | * device is present. | |
7505 | */ | |
7506 | ||
7507 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7508 | ||
8f4cccbb BH |
7509 | linkwatch_init_dev(dev); |
7510 | ||
1da177e4 | 7511 | dev_init_scheduler(dev); |
1da177e4 | 7512 | dev_hold(dev); |
ce286d32 | 7513 | list_netdevice(dev); |
7bf23575 | 7514 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 7515 | |
948b337e JP |
7516 | /* If the device has permanent device address, driver should |
7517 | * set dev_addr and also addr_assign_type should be set to | |
7518 | * NET_ADDR_PERM (default value). | |
7519 | */ | |
7520 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
7521 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
7522 | ||
1da177e4 | 7523 | /* Notify protocols, that a new device appeared. */ |
056925ab | 7524 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 7525 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
7526 | if (ret) { |
7527 | rollback_registered(dev); | |
7528 | dev->reg_state = NETREG_UNREGISTERED; | |
7529 | } | |
d90a909e EB |
7530 | /* |
7531 | * Prevent userspace races by waiting until the network | |
7532 | * device is fully setup before sending notifications. | |
7533 | */ | |
a2835763 PM |
7534 | if (!dev->rtnl_link_ops || |
7535 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 7536 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
7537 | |
7538 | out: | |
7539 | return ret; | |
7ce1b0ed HX |
7540 | |
7541 | err_uninit: | |
d314774c SH |
7542 | if (dev->netdev_ops->ndo_uninit) |
7543 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 7544 | goto out; |
1da177e4 | 7545 | } |
d1b19dff | 7546 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 7547 | |
937f1ba5 BH |
7548 | /** |
7549 | * init_dummy_netdev - init a dummy network device for NAPI | |
7550 | * @dev: device to init | |
7551 | * | |
7552 | * This takes a network device structure and initialize the minimum | |
7553 | * amount of fields so it can be used to schedule NAPI polls without | |
7554 | * registering a full blown interface. This is to be used by drivers | |
7555 | * that need to tie several hardware interfaces to a single NAPI | |
7556 | * poll scheduler due to HW limitations. | |
7557 | */ | |
7558 | int init_dummy_netdev(struct net_device *dev) | |
7559 | { | |
7560 | /* Clear everything. Note we don't initialize spinlocks | |
7561 | * are they aren't supposed to be taken by any of the | |
7562 | * NAPI code and this dummy netdev is supposed to be | |
7563 | * only ever used for NAPI polls | |
7564 | */ | |
7565 | memset(dev, 0, sizeof(struct net_device)); | |
7566 | ||
7567 | /* make sure we BUG if trying to hit standard | |
7568 | * register/unregister code path | |
7569 | */ | |
7570 | dev->reg_state = NETREG_DUMMY; | |
7571 | ||
937f1ba5 BH |
7572 | /* NAPI wants this */ |
7573 | INIT_LIST_HEAD(&dev->napi_list); | |
7574 | ||
7575 | /* a dummy interface is started by default */ | |
7576 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7577 | set_bit(__LINK_STATE_START, &dev->state); | |
7578 | ||
29b4433d ED |
7579 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
7580 | * because users of this 'device' dont need to change | |
7581 | * its refcount. | |
7582 | */ | |
7583 | ||
937f1ba5 BH |
7584 | return 0; |
7585 | } | |
7586 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
7587 | ||
7588 | ||
1da177e4 LT |
7589 | /** |
7590 | * register_netdev - register a network device | |
7591 | * @dev: device to register | |
7592 | * | |
7593 | * Take a completed network device structure and add it to the kernel | |
7594 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7595 | * chain. 0 is returned on success. A negative errno code is returned | |
7596 | * on a failure to set up the device, or if the name is a duplicate. | |
7597 | * | |
38b4da38 | 7598 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
7599 | * and expands the device name if you passed a format string to |
7600 | * alloc_netdev. | |
7601 | */ | |
7602 | int register_netdev(struct net_device *dev) | |
7603 | { | |
7604 | int err; | |
7605 | ||
7606 | rtnl_lock(); | |
1da177e4 | 7607 | err = register_netdevice(dev); |
1da177e4 LT |
7608 | rtnl_unlock(); |
7609 | return err; | |
7610 | } | |
7611 | EXPORT_SYMBOL(register_netdev); | |
7612 | ||
29b4433d ED |
7613 | int netdev_refcnt_read(const struct net_device *dev) |
7614 | { | |
7615 | int i, refcnt = 0; | |
7616 | ||
7617 | for_each_possible_cpu(i) | |
7618 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
7619 | return refcnt; | |
7620 | } | |
7621 | EXPORT_SYMBOL(netdev_refcnt_read); | |
7622 | ||
2c53040f | 7623 | /** |
1da177e4 | 7624 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 7625 | * @dev: target net_device |
1da177e4 LT |
7626 | * |
7627 | * This is called when unregistering network devices. | |
7628 | * | |
7629 | * Any protocol or device that holds a reference should register | |
7630 | * for netdevice notification, and cleanup and put back the | |
7631 | * reference if they receive an UNREGISTER event. | |
7632 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 7633 | * call dev_put. |
1da177e4 LT |
7634 | */ |
7635 | static void netdev_wait_allrefs(struct net_device *dev) | |
7636 | { | |
7637 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 7638 | int refcnt; |
1da177e4 | 7639 | |
e014debe ED |
7640 | linkwatch_forget_dev(dev); |
7641 | ||
1da177e4 | 7642 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
7643 | refcnt = netdev_refcnt_read(dev); |
7644 | ||
7645 | while (refcnt != 0) { | |
1da177e4 | 7646 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 7647 | rtnl_lock(); |
1da177e4 LT |
7648 | |
7649 | /* Rebroadcast unregister notification */ | |
056925ab | 7650 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 7651 | |
748e2d93 | 7652 | __rtnl_unlock(); |
0115e8e3 | 7653 | rcu_barrier(); |
748e2d93 ED |
7654 | rtnl_lock(); |
7655 | ||
0115e8e3 | 7656 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
7657 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
7658 | &dev->state)) { | |
7659 | /* We must not have linkwatch events | |
7660 | * pending on unregister. If this | |
7661 | * happens, we simply run the queue | |
7662 | * unscheduled, resulting in a noop | |
7663 | * for this device. | |
7664 | */ | |
7665 | linkwatch_run_queue(); | |
7666 | } | |
7667 | ||
6756ae4b | 7668 | __rtnl_unlock(); |
1da177e4 LT |
7669 | |
7670 | rebroadcast_time = jiffies; | |
7671 | } | |
7672 | ||
7673 | msleep(250); | |
7674 | ||
29b4433d ED |
7675 | refcnt = netdev_refcnt_read(dev); |
7676 | ||
1da177e4 | 7677 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
7678 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
7679 | dev->name, refcnt); | |
1da177e4 LT |
7680 | warning_time = jiffies; |
7681 | } | |
7682 | } | |
7683 | } | |
7684 | ||
7685 | /* The sequence is: | |
7686 | * | |
7687 | * rtnl_lock(); | |
7688 | * ... | |
7689 | * register_netdevice(x1); | |
7690 | * register_netdevice(x2); | |
7691 | * ... | |
7692 | * unregister_netdevice(y1); | |
7693 | * unregister_netdevice(y2); | |
7694 | * ... | |
7695 | * rtnl_unlock(); | |
7696 | * free_netdev(y1); | |
7697 | * free_netdev(y2); | |
7698 | * | |
58ec3b4d | 7699 | * We are invoked by rtnl_unlock(). |
1da177e4 | 7700 | * This allows us to deal with problems: |
b17a7c17 | 7701 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
7702 | * without deadlocking with linkwatch via keventd. |
7703 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
7704 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
7705 | * |
7706 | * We must not return until all unregister events added during | |
7707 | * the interval the lock was held have been completed. | |
1da177e4 | 7708 | */ |
1da177e4 LT |
7709 | void netdev_run_todo(void) |
7710 | { | |
626ab0e6 | 7711 | struct list_head list; |
1da177e4 | 7712 | |
1da177e4 | 7713 | /* Snapshot list, allow later requests */ |
626ab0e6 | 7714 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
7715 | |
7716 | __rtnl_unlock(); | |
626ab0e6 | 7717 | |
0115e8e3 ED |
7718 | |
7719 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
7720 | if (!list_empty(&list)) |
7721 | rcu_barrier(); | |
7722 | ||
1da177e4 LT |
7723 | while (!list_empty(&list)) { |
7724 | struct net_device *dev | |
e5e26d75 | 7725 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
7726 | list_del(&dev->todo_list); |
7727 | ||
748e2d93 | 7728 | rtnl_lock(); |
0115e8e3 | 7729 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 7730 | __rtnl_unlock(); |
0115e8e3 | 7731 | |
b17a7c17 | 7732 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 7733 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
7734 | dev->name, dev->reg_state); |
7735 | dump_stack(); | |
7736 | continue; | |
7737 | } | |
1da177e4 | 7738 | |
b17a7c17 | 7739 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 7740 | |
b17a7c17 | 7741 | netdev_wait_allrefs(dev); |
1da177e4 | 7742 | |
b17a7c17 | 7743 | /* paranoia */ |
29b4433d | 7744 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
7745 | BUG_ON(!list_empty(&dev->ptype_all)); |
7746 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
7747 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
7748 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 7749 | WARN_ON(dev->dn_ptr); |
1da177e4 | 7750 | |
b17a7c17 SH |
7751 | if (dev->destructor) |
7752 | dev->destructor(dev); | |
9093bbb2 | 7753 | |
50624c93 EB |
7754 | /* Report a network device has been unregistered */ |
7755 | rtnl_lock(); | |
7756 | dev_net(dev)->dev_unreg_count--; | |
7757 | __rtnl_unlock(); | |
7758 | wake_up(&netdev_unregistering_wq); | |
7759 | ||
9093bbb2 SH |
7760 | /* Free network device */ |
7761 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 7762 | } |
1da177e4 LT |
7763 | } |
7764 | ||
9256645a JW |
7765 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
7766 | * all the same fields in the same order as net_device_stats, with only | |
7767 | * the type differing, but rtnl_link_stats64 may have additional fields | |
7768 | * at the end for newer counters. | |
3cfde79c | 7769 | */ |
77a1abf5 ED |
7770 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
7771 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
7772 | { |
7773 | #if BITS_PER_LONG == 64 | |
9256645a | 7774 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
77a1abf5 | 7775 | memcpy(stats64, netdev_stats, sizeof(*stats64)); |
9256645a JW |
7776 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7777 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
7778 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 7779 | #else |
9256645a | 7780 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
7781 | const unsigned long *src = (const unsigned long *)netdev_stats; |
7782 | u64 *dst = (u64 *)stats64; | |
7783 | ||
9256645a | 7784 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
7785 | for (i = 0; i < n; i++) |
7786 | dst[i] = src[i]; | |
9256645a JW |
7787 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7788 | memset((char *)stats64 + n * sizeof(u64), 0, | |
7789 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
7790 | #endif |
7791 | } | |
77a1abf5 | 7792 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 7793 | |
eeda3fd6 SH |
7794 | /** |
7795 | * dev_get_stats - get network device statistics | |
7796 | * @dev: device to get statistics from | |
28172739 | 7797 | * @storage: place to store stats |
eeda3fd6 | 7798 | * |
d7753516 BH |
7799 | * Get network statistics from device. Return @storage. |
7800 | * The device driver may provide its own method by setting | |
7801 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
7802 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 7803 | */ |
d7753516 BH |
7804 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
7805 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 7806 | { |
eeda3fd6 SH |
7807 | const struct net_device_ops *ops = dev->netdev_ops; |
7808 | ||
28172739 ED |
7809 | if (ops->ndo_get_stats64) { |
7810 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
7811 | ops->ndo_get_stats64(dev, storage); |
7812 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 7813 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
7814 | } else { |
7815 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 7816 | } |
caf586e5 | 7817 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
015f0688 | 7818 | storage->tx_dropped += atomic_long_read(&dev->tx_dropped); |
6e7333d3 | 7819 | storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler); |
28172739 | 7820 | return storage; |
c45d286e | 7821 | } |
eeda3fd6 | 7822 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 7823 | |
24824a09 | 7824 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 7825 | { |
24824a09 | 7826 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 7827 | |
24824a09 ED |
7828 | #ifdef CONFIG_NET_CLS_ACT |
7829 | if (queue) | |
7830 | return queue; | |
7831 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
7832 | if (!queue) | |
7833 | return NULL; | |
7834 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 7835 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
7836 | queue->qdisc_sleeping = &noop_qdisc; |
7837 | rcu_assign_pointer(dev->ingress_queue, queue); | |
7838 | #endif | |
7839 | return queue; | |
bb949fbd DM |
7840 | } |
7841 | ||
2c60db03 ED |
7842 | static const struct ethtool_ops default_ethtool_ops; |
7843 | ||
d07d7507 SG |
7844 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
7845 | const struct ethtool_ops *ops) | |
7846 | { | |
7847 | if (dev->ethtool_ops == &default_ethtool_ops) | |
7848 | dev->ethtool_ops = ops; | |
7849 | } | |
7850 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
7851 | ||
74d332c1 ED |
7852 | void netdev_freemem(struct net_device *dev) |
7853 | { | |
7854 | char *addr = (char *)dev - dev->padded; | |
7855 | ||
4cb28970 | 7856 | kvfree(addr); |
74d332c1 ED |
7857 | } |
7858 | ||
1da177e4 | 7859 | /** |
722c9a0c | 7860 | * alloc_netdev_mqs - allocate network device |
7861 | * @sizeof_priv: size of private data to allocate space for | |
7862 | * @name: device name format string | |
7863 | * @name_assign_type: origin of device name | |
7864 | * @setup: callback to initialize device | |
7865 | * @txqs: the number of TX subqueues to allocate | |
7866 | * @rxqs: the number of RX subqueues to allocate | |
7867 | * | |
7868 | * Allocates a struct net_device with private data area for driver use | |
7869 | * and performs basic initialization. Also allocates subqueue structs | |
7870 | * for each queue on the device. | |
1da177e4 | 7871 | */ |
36909ea4 | 7872 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 7873 | unsigned char name_assign_type, |
36909ea4 TH |
7874 | void (*setup)(struct net_device *), |
7875 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 7876 | { |
1da177e4 | 7877 | struct net_device *dev; |
7943986c | 7878 | size_t alloc_size; |
1ce8e7b5 | 7879 | struct net_device *p; |
1da177e4 | 7880 | |
b6fe17d6 SH |
7881 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
7882 | ||
36909ea4 | 7883 | if (txqs < 1) { |
7b6cd1ce | 7884 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
7885 | return NULL; |
7886 | } | |
7887 | ||
a953be53 | 7888 | #ifdef CONFIG_SYSFS |
36909ea4 | 7889 | if (rxqs < 1) { |
7b6cd1ce | 7890 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
7891 | return NULL; |
7892 | } | |
7893 | #endif | |
7894 | ||
fd2ea0a7 | 7895 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
7896 | if (sizeof_priv) { |
7897 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 7898 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
7899 | alloc_size += sizeof_priv; |
7900 | } | |
7901 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 7902 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 7903 | |
da6bc57a | 7904 | p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_REPEAT); |
62b5942a | 7905 | if (!p) |
1da177e4 | 7906 | return NULL; |
1da177e4 | 7907 | |
1ce8e7b5 | 7908 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 7909 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 7910 | |
29b4433d ED |
7911 | dev->pcpu_refcnt = alloc_percpu(int); |
7912 | if (!dev->pcpu_refcnt) | |
74d332c1 | 7913 | goto free_dev; |
ab9c73cc | 7914 | |
ab9c73cc | 7915 | if (dev_addr_init(dev)) |
29b4433d | 7916 | goto free_pcpu; |
ab9c73cc | 7917 | |
22bedad3 | 7918 | dev_mc_init(dev); |
a748ee24 | 7919 | dev_uc_init(dev); |
ccffad25 | 7920 | |
c346dca1 | 7921 | dev_net_set(dev, &init_net); |
1da177e4 | 7922 | |
8d3bdbd5 | 7923 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 7924 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 7925 | |
8d3bdbd5 DM |
7926 | INIT_LIST_HEAD(&dev->napi_list); |
7927 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 7928 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 7929 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
7930 | INIT_LIST_HEAD(&dev->adj_list.upper); |
7931 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
7932 | INIT_LIST_HEAD(&dev->ptype_all); |
7933 | INIT_LIST_HEAD(&dev->ptype_specific); | |
59cc1f61 JK |
7934 | #ifdef CONFIG_NET_SCHED |
7935 | hash_init(dev->qdisc_hash); | |
7936 | #endif | |
02875878 | 7937 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
7938 | setup(dev); |
7939 | ||
a813104d | 7940 | if (!dev->tx_queue_len) { |
f84bb1ea | 7941 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 7942 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 7943 | } |
906470c1 | 7944 | |
36909ea4 TH |
7945 | dev->num_tx_queues = txqs; |
7946 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 7947 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 7948 | goto free_all; |
e8a0464c | 7949 | |
a953be53 | 7950 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
7951 | dev->num_rx_queues = rxqs; |
7952 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 7953 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 7954 | goto free_all; |
df334545 | 7955 | #endif |
0a9627f2 | 7956 | |
1da177e4 | 7957 | strcpy(dev->name, name); |
c835a677 | 7958 | dev->name_assign_type = name_assign_type; |
cbda10fa | 7959 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
7960 | if (!dev->ethtool_ops) |
7961 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 PN |
7962 | |
7963 | nf_hook_ingress_init(dev); | |
7964 | ||
1da177e4 | 7965 | return dev; |
ab9c73cc | 7966 | |
8d3bdbd5 DM |
7967 | free_all: |
7968 | free_netdev(dev); | |
7969 | return NULL; | |
7970 | ||
29b4433d ED |
7971 | free_pcpu: |
7972 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
7973 | free_dev: |
7974 | netdev_freemem(dev); | |
ab9c73cc | 7975 | return NULL; |
1da177e4 | 7976 | } |
36909ea4 | 7977 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
7978 | |
7979 | /** | |
722c9a0c | 7980 | * free_netdev - free network device |
7981 | * @dev: device | |
1da177e4 | 7982 | * |
722c9a0c | 7983 | * This function does the last stage of destroying an allocated device |
7984 | * interface. The reference to the device object is released. If this | |
7985 | * is the last reference then it will be freed.Must be called in process | |
7986 | * context. | |
1da177e4 LT |
7987 | */ |
7988 | void free_netdev(struct net_device *dev) | |
7989 | { | |
d565b0a1 | 7990 | struct napi_struct *p, *n; |
b5cdae32 | 7991 | struct bpf_prog *prog; |
d565b0a1 | 7992 | |
93d05d4a | 7993 | might_sleep(); |
60877a32 | 7994 | netif_free_tx_queues(dev); |
a953be53 | 7995 | #ifdef CONFIG_SYSFS |
10595902 | 7996 | kvfree(dev->_rx); |
fe822240 | 7997 | #endif |
e8a0464c | 7998 | |
33d480ce | 7999 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 8000 | |
f001fde5 JP |
8001 | /* Flush device addresses */ |
8002 | dev_addr_flush(dev); | |
8003 | ||
d565b0a1 HX |
8004 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
8005 | netif_napi_del(p); | |
8006 | ||
29b4433d ED |
8007 | free_percpu(dev->pcpu_refcnt); |
8008 | dev->pcpu_refcnt = NULL; | |
8009 | ||
b5cdae32 DM |
8010 | prog = rcu_dereference_protected(dev->xdp_prog, 1); |
8011 | if (prog) { | |
8012 | bpf_prog_put(prog); | |
8013 | static_key_slow_dec(&generic_xdp_needed); | |
8014 | } | |
8015 | ||
3041a069 | 8016 | /* Compatibility with error handling in drivers */ |
1da177e4 | 8017 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 8018 | netdev_freemem(dev); |
1da177e4 LT |
8019 | return; |
8020 | } | |
8021 | ||
8022 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
8023 | dev->reg_state = NETREG_RELEASED; | |
8024 | ||
43cb76d9 GKH |
8025 | /* will free via device release */ |
8026 | put_device(&dev->dev); | |
1da177e4 | 8027 | } |
d1b19dff | 8028 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 8029 | |
f0db275a SH |
8030 | /** |
8031 | * synchronize_net - Synchronize with packet receive processing | |
8032 | * | |
8033 | * Wait for packets currently being received to be done. | |
8034 | * Does not block later packets from starting. | |
8035 | */ | |
4ec93edb | 8036 | void synchronize_net(void) |
1da177e4 LT |
8037 | { |
8038 | might_sleep(); | |
be3fc413 ED |
8039 | if (rtnl_is_locked()) |
8040 | synchronize_rcu_expedited(); | |
8041 | else | |
8042 | synchronize_rcu(); | |
1da177e4 | 8043 | } |
d1b19dff | 8044 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
8045 | |
8046 | /** | |
44a0873d | 8047 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 8048 | * @dev: device |
44a0873d | 8049 | * @head: list |
6ebfbc06 | 8050 | * |
1da177e4 | 8051 | * This function shuts down a device interface and removes it |
d59b54b1 | 8052 | * from the kernel tables. |
44a0873d | 8053 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
8054 | * |
8055 | * Callers must hold the rtnl semaphore. You may want | |
8056 | * unregister_netdev() instead of this. | |
8057 | */ | |
8058 | ||
44a0873d | 8059 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 8060 | { |
a6620712 HX |
8061 | ASSERT_RTNL(); |
8062 | ||
44a0873d | 8063 | if (head) { |
9fdce099 | 8064 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
8065 | } else { |
8066 | rollback_registered(dev); | |
8067 | /* Finish processing unregister after unlock */ | |
8068 | net_set_todo(dev); | |
8069 | } | |
1da177e4 | 8070 | } |
44a0873d | 8071 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 8072 | |
9b5e383c ED |
8073 | /** |
8074 | * unregister_netdevice_many - unregister many devices | |
8075 | * @head: list of devices | |
87757a91 ED |
8076 | * |
8077 | * Note: As most callers use a stack allocated list_head, | |
8078 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
8079 | */ |
8080 | void unregister_netdevice_many(struct list_head *head) | |
8081 | { | |
8082 | struct net_device *dev; | |
8083 | ||
8084 | if (!list_empty(head)) { | |
8085 | rollback_registered_many(head); | |
8086 | list_for_each_entry(dev, head, unreg_list) | |
8087 | net_set_todo(dev); | |
87757a91 | 8088 | list_del(head); |
9b5e383c ED |
8089 | } |
8090 | } | |
63c8099d | 8091 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 8092 | |
1da177e4 LT |
8093 | /** |
8094 | * unregister_netdev - remove device from the kernel | |
8095 | * @dev: device | |
8096 | * | |
8097 | * This function shuts down a device interface and removes it | |
d59b54b1 | 8098 | * from the kernel tables. |
1da177e4 LT |
8099 | * |
8100 | * This is just a wrapper for unregister_netdevice that takes | |
8101 | * the rtnl semaphore. In general you want to use this and not | |
8102 | * unregister_netdevice. | |
8103 | */ | |
8104 | void unregister_netdev(struct net_device *dev) | |
8105 | { | |
8106 | rtnl_lock(); | |
8107 | unregister_netdevice(dev); | |
8108 | rtnl_unlock(); | |
8109 | } | |
1da177e4 LT |
8110 | EXPORT_SYMBOL(unregister_netdev); |
8111 | ||
ce286d32 EB |
8112 | /** |
8113 | * dev_change_net_namespace - move device to different nethost namespace | |
8114 | * @dev: device | |
8115 | * @net: network namespace | |
8116 | * @pat: If not NULL name pattern to try if the current device name | |
8117 | * is already taken in the destination network namespace. | |
8118 | * | |
8119 | * This function shuts down a device interface and moves it | |
8120 | * to a new network namespace. On success 0 is returned, on | |
8121 | * a failure a netagive errno code is returned. | |
8122 | * | |
8123 | * Callers must hold the rtnl semaphore. | |
8124 | */ | |
8125 | ||
8126 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
8127 | { | |
ce286d32 EB |
8128 | int err; |
8129 | ||
8130 | ASSERT_RTNL(); | |
8131 | ||
8132 | /* Don't allow namespace local devices to be moved. */ | |
8133 | err = -EINVAL; | |
8134 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8135 | goto out; | |
8136 | ||
8137 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
8138 | if (dev->reg_state != NETREG_REGISTERED) |
8139 | goto out; | |
8140 | ||
8141 | /* Get out if there is nothing todo */ | |
8142 | err = 0; | |
878628fb | 8143 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
8144 | goto out; |
8145 | ||
8146 | /* Pick the destination device name, and ensure | |
8147 | * we can use it in the destination network namespace. | |
8148 | */ | |
8149 | err = -EEXIST; | |
d9031024 | 8150 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
8151 | /* We get here if we can't use the current device name */ |
8152 | if (!pat) | |
8153 | goto out; | |
828de4f6 | 8154 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
8155 | goto out; |
8156 | } | |
8157 | ||
8158 | /* | |
8159 | * And now a mini version of register_netdevice unregister_netdevice. | |
8160 | */ | |
8161 | ||
8162 | /* If device is running close it first. */ | |
9b772652 | 8163 | dev_close(dev); |
ce286d32 EB |
8164 | |
8165 | /* And unlink it from device chain */ | |
8166 | err = -ENODEV; | |
8167 | unlist_netdevice(dev); | |
8168 | ||
8169 | synchronize_net(); | |
8170 | ||
8171 | /* Shutdown queueing discipline. */ | |
8172 | dev_shutdown(dev); | |
8173 | ||
8174 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 8175 | * this device. They should clean all the things. |
8176 | * | |
8177 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
8178 | * This is wanted because this way 8021q and macvlan know | |
8179 | * the device is just moving and can keep their slaves up. | |
8180 | */ | |
ce286d32 | 8181 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 G |
8182 | rcu_barrier(); |
8183 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 8184 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
8185 | |
8186 | /* | |
8187 | * Flush the unicast and multicast chains | |
8188 | */ | |
a748ee24 | 8189 | dev_uc_flush(dev); |
22bedad3 | 8190 | dev_mc_flush(dev); |
ce286d32 | 8191 | |
4e66ae2e SH |
8192 | /* Send a netdev-removed uevent to the old namespace */ |
8193 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 8194 | netdev_adjacent_del_links(dev); |
4e66ae2e | 8195 | |
ce286d32 | 8196 | /* Actually switch the network namespace */ |
c346dca1 | 8197 | dev_net_set(dev, net); |
ce286d32 | 8198 | |
ce286d32 | 8199 | /* If there is an ifindex conflict assign a new one */ |
7a66bbc9 | 8200 | if (__dev_get_by_index(net, dev->ifindex)) |
ce286d32 | 8201 | dev->ifindex = dev_new_index(net); |
ce286d32 | 8202 | |
4e66ae2e SH |
8203 | /* Send a netdev-add uevent to the new namespace */ |
8204 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 8205 | netdev_adjacent_add_links(dev); |
4e66ae2e | 8206 | |
8b41d188 | 8207 | /* Fixup kobjects */ |
a1b3f594 | 8208 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 8209 | WARN_ON(err); |
ce286d32 EB |
8210 | |
8211 | /* Add the device back in the hashes */ | |
8212 | list_netdevice(dev); | |
8213 | ||
8214 | /* Notify protocols, that a new device appeared. */ | |
8215 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
8216 | ||
d90a909e EB |
8217 | /* |
8218 | * Prevent userspace races by waiting until the network | |
8219 | * device is fully setup before sending notifications. | |
8220 | */ | |
7f294054 | 8221 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 8222 | |
ce286d32 EB |
8223 | synchronize_net(); |
8224 | err = 0; | |
8225 | out: | |
8226 | return err; | |
8227 | } | |
463d0183 | 8228 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 8229 | |
f0bf90de | 8230 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
8231 | { |
8232 | struct sk_buff **list_skb; | |
1da177e4 | 8233 | struct sk_buff *skb; |
f0bf90de | 8234 | unsigned int cpu; |
1da177e4 LT |
8235 | struct softnet_data *sd, *oldsd; |
8236 | ||
1da177e4 LT |
8237 | local_irq_disable(); |
8238 | cpu = smp_processor_id(); | |
8239 | sd = &per_cpu(softnet_data, cpu); | |
8240 | oldsd = &per_cpu(softnet_data, oldcpu); | |
8241 | ||
8242 | /* Find end of our completion_queue. */ | |
8243 | list_skb = &sd->completion_queue; | |
8244 | while (*list_skb) | |
8245 | list_skb = &(*list_skb)->next; | |
8246 | /* Append completion queue from offline CPU. */ | |
8247 | *list_skb = oldsd->completion_queue; | |
8248 | oldsd->completion_queue = NULL; | |
8249 | ||
1da177e4 | 8250 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
8251 | if (oldsd->output_queue) { |
8252 | *sd->output_queue_tailp = oldsd->output_queue; | |
8253 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
8254 | oldsd->output_queue = NULL; | |
8255 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
8256 | } | |
ac64da0b ED |
8257 | /* Append NAPI poll list from offline CPU, with one exception : |
8258 | * process_backlog() must be called by cpu owning percpu backlog. | |
8259 | * We properly handle process_queue & input_pkt_queue later. | |
8260 | */ | |
8261 | while (!list_empty(&oldsd->poll_list)) { | |
8262 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
8263 | struct napi_struct, | |
8264 | poll_list); | |
8265 | ||
8266 | list_del_init(&napi->poll_list); | |
8267 | if (napi->poll == process_backlog) | |
8268 | napi->state = 0; | |
8269 | else | |
8270 | ____napi_schedule(sd, napi); | |
264524d5 | 8271 | } |
1da177e4 LT |
8272 | |
8273 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
8274 | local_irq_enable(); | |
8275 | ||
8276 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 8277 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 8278 | netif_rx_ni(skb); |
76cc8b13 | 8279 | input_queue_head_incr(oldsd); |
fec5e652 | 8280 | } |
ac64da0b | 8281 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 8282 | netif_rx_ni(skb); |
76cc8b13 TH |
8283 | input_queue_head_incr(oldsd); |
8284 | } | |
1da177e4 | 8285 | |
f0bf90de | 8286 | return 0; |
1da177e4 | 8287 | } |
1da177e4 | 8288 | |
7f353bf2 | 8289 | /** |
b63365a2 HX |
8290 | * netdev_increment_features - increment feature set by one |
8291 | * @all: current feature set | |
8292 | * @one: new feature set | |
8293 | * @mask: mask feature set | |
7f353bf2 HX |
8294 | * |
8295 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
8296 | * @one to the master device with current feature set @all. Will not |
8297 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 8298 | */ |
c8f44aff MM |
8299 | netdev_features_t netdev_increment_features(netdev_features_t all, |
8300 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 8301 | { |
c8cd0989 | 8302 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 8303 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 8304 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 8305 | |
a188222b | 8306 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 8307 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 8308 | |
1742f183 | 8309 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
8310 | if (all & NETIF_F_HW_CSUM) |
8311 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
8312 | |
8313 | return all; | |
8314 | } | |
b63365a2 | 8315 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 8316 | |
430f03cd | 8317 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
8318 | { |
8319 | int i; | |
8320 | struct hlist_head *hash; | |
8321 | ||
8322 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
8323 | if (hash != NULL) | |
8324 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
8325 | INIT_HLIST_HEAD(&hash[i]); | |
8326 | ||
8327 | return hash; | |
8328 | } | |
8329 | ||
881d966b | 8330 | /* Initialize per network namespace state */ |
4665079c | 8331 | static int __net_init netdev_init(struct net *net) |
881d966b | 8332 | { |
734b6541 RM |
8333 | if (net != &init_net) |
8334 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 8335 | |
30d97d35 PE |
8336 | net->dev_name_head = netdev_create_hash(); |
8337 | if (net->dev_name_head == NULL) | |
8338 | goto err_name; | |
881d966b | 8339 | |
30d97d35 PE |
8340 | net->dev_index_head = netdev_create_hash(); |
8341 | if (net->dev_index_head == NULL) | |
8342 | goto err_idx; | |
881d966b EB |
8343 | |
8344 | return 0; | |
30d97d35 PE |
8345 | |
8346 | err_idx: | |
8347 | kfree(net->dev_name_head); | |
8348 | err_name: | |
8349 | return -ENOMEM; | |
881d966b EB |
8350 | } |
8351 | ||
f0db275a SH |
8352 | /** |
8353 | * netdev_drivername - network driver for the device | |
8354 | * @dev: network device | |
f0db275a SH |
8355 | * |
8356 | * Determine network driver for device. | |
8357 | */ | |
3019de12 | 8358 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 8359 | { |
cf04a4c7 SH |
8360 | const struct device_driver *driver; |
8361 | const struct device *parent; | |
3019de12 | 8362 | const char *empty = ""; |
6579e57b AV |
8363 | |
8364 | parent = dev->dev.parent; | |
6579e57b | 8365 | if (!parent) |
3019de12 | 8366 | return empty; |
6579e57b AV |
8367 | |
8368 | driver = parent->driver; | |
8369 | if (driver && driver->name) | |
3019de12 DM |
8370 | return driver->name; |
8371 | return empty; | |
6579e57b AV |
8372 | } |
8373 | ||
6ea754eb JP |
8374 | static void __netdev_printk(const char *level, const struct net_device *dev, |
8375 | struct va_format *vaf) | |
256df2f3 | 8376 | { |
b004ff49 | 8377 | if (dev && dev->dev.parent) { |
6ea754eb JP |
8378 | dev_printk_emit(level[1] - '0', |
8379 | dev->dev.parent, | |
8380 | "%s %s %s%s: %pV", | |
8381 | dev_driver_string(dev->dev.parent), | |
8382 | dev_name(dev->dev.parent), | |
8383 | netdev_name(dev), netdev_reg_state(dev), | |
8384 | vaf); | |
b004ff49 | 8385 | } else if (dev) { |
6ea754eb JP |
8386 | printk("%s%s%s: %pV", |
8387 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 8388 | } else { |
6ea754eb | 8389 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 8390 | } |
256df2f3 JP |
8391 | } |
8392 | ||
6ea754eb JP |
8393 | void netdev_printk(const char *level, const struct net_device *dev, |
8394 | const char *format, ...) | |
256df2f3 JP |
8395 | { |
8396 | struct va_format vaf; | |
8397 | va_list args; | |
256df2f3 JP |
8398 | |
8399 | va_start(args, format); | |
8400 | ||
8401 | vaf.fmt = format; | |
8402 | vaf.va = &args; | |
8403 | ||
6ea754eb | 8404 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 8405 | |
256df2f3 | 8406 | va_end(args); |
256df2f3 JP |
8407 | } |
8408 | EXPORT_SYMBOL(netdev_printk); | |
8409 | ||
8410 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 8411 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 8412 | { \ |
256df2f3 JP |
8413 | struct va_format vaf; \ |
8414 | va_list args; \ | |
8415 | \ | |
8416 | va_start(args, fmt); \ | |
8417 | \ | |
8418 | vaf.fmt = fmt; \ | |
8419 | vaf.va = &args; \ | |
8420 | \ | |
6ea754eb | 8421 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 8422 | \ |
256df2f3 | 8423 | va_end(args); \ |
256df2f3 JP |
8424 | } \ |
8425 | EXPORT_SYMBOL(func); | |
8426 | ||
8427 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
8428 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
8429 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
8430 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
8431 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
8432 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
8433 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
8434 | ||
4665079c | 8435 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
8436 | { |
8437 | kfree(net->dev_name_head); | |
8438 | kfree(net->dev_index_head); | |
8439 | } | |
8440 | ||
022cbae6 | 8441 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
8442 | .init = netdev_init, |
8443 | .exit = netdev_exit, | |
8444 | }; | |
8445 | ||
4665079c | 8446 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 8447 | { |
e008b5fc | 8448 | struct net_device *dev, *aux; |
ce286d32 | 8449 | /* |
e008b5fc | 8450 | * Push all migratable network devices back to the |
ce286d32 EB |
8451 | * initial network namespace |
8452 | */ | |
8453 | rtnl_lock(); | |
e008b5fc | 8454 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 8455 | int err; |
aca51397 | 8456 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
8457 | |
8458 | /* Ignore unmoveable devices (i.e. loopback) */ | |
8459 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8460 | continue; | |
8461 | ||
e008b5fc EB |
8462 | /* Leave virtual devices for the generic cleanup */ |
8463 | if (dev->rtnl_link_ops) | |
8464 | continue; | |
d0c082ce | 8465 | |
25985edc | 8466 | /* Push remaining network devices to init_net */ |
aca51397 PE |
8467 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
8468 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 8469 | if (err) { |
7b6cd1ce JP |
8470 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
8471 | __func__, dev->name, err); | |
aca51397 | 8472 | BUG(); |
ce286d32 EB |
8473 | } |
8474 | } | |
8475 | rtnl_unlock(); | |
8476 | } | |
8477 | ||
50624c93 EB |
8478 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
8479 | { | |
8480 | /* Return with the rtnl_lock held when there are no network | |
8481 | * devices unregistering in any network namespace in net_list. | |
8482 | */ | |
8483 | struct net *net; | |
8484 | bool unregistering; | |
ff960a73 | 8485 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 8486 | |
ff960a73 | 8487 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 8488 | for (;;) { |
50624c93 EB |
8489 | unregistering = false; |
8490 | rtnl_lock(); | |
8491 | list_for_each_entry(net, net_list, exit_list) { | |
8492 | if (net->dev_unreg_count > 0) { | |
8493 | unregistering = true; | |
8494 | break; | |
8495 | } | |
8496 | } | |
8497 | if (!unregistering) | |
8498 | break; | |
8499 | __rtnl_unlock(); | |
ff960a73 PZ |
8500 | |
8501 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 8502 | } |
ff960a73 | 8503 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
8504 | } |
8505 | ||
04dc7f6b EB |
8506 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
8507 | { | |
8508 | /* At exit all network devices most be removed from a network | |
b595076a | 8509 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
8510 | * Do this across as many network namespaces as possible to |
8511 | * improve batching efficiency. | |
8512 | */ | |
8513 | struct net_device *dev; | |
8514 | struct net *net; | |
8515 | LIST_HEAD(dev_kill_list); | |
8516 | ||
50624c93 EB |
8517 | /* To prevent network device cleanup code from dereferencing |
8518 | * loopback devices or network devices that have been freed | |
8519 | * wait here for all pending unregistrations to complete, | |
8520 | * before unregistring the loopback device and allowing the | |
8521 | * network namespace be freed. | |
8522 | * | |
8523 | * The netdev todo list containing all network devices | |
8524 | * unregistrations that happen in default_device_exit_batch | |
8525 | * will run in the rtnl_unlock() at the end of | |
8526 | * default_device_exit_batch. | |
8527 | */ | |
8528 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
8529 | list_for_each_entry(net, net_list, exit_list) { |
8530 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 8531 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
8532 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
8533 | else | |
8534 | unregister_netdevice_queue(dev, &dev_kill_list); | |
8535 | } | |
8536 | } | |
8537 | unregister_netdevice_many(&dev_kill_list); | |
8538 | rtnl_unlock(); | |
8539 | } | |
8540 | ||
022cbae6 | 8541 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 8542 | .exit = default_device_exit, |
04dc7f6b | 8543 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
8544 | }; |
8545 | ||
1da177e4 LT |
8546 | /* |
8547 | * Initialize the DEV module. At boot time this walks the device list and | |
8548 | * unhooks any devices that fail to initialise (normally hardware not | |
8549 | * present) and leaves us with a valid list of present and active devices. | |
8550 | * | |
8551 | */ | |
8552 | ||
8553 | /* | |
8554 | * This is called single threaded during boot, so no need | |
8555 | * to take the rtnl semaphore. | |
8556 | */ | |
8557 | static int __init net_dev_init(void) | |
8558 | { | |
8559 | int i, rc = -ENOMEM; | |
8560 | ||
8561 | BUG_ON(!dev_boot_phase); | |
8562 | ||
1da177e4 LT |
8563 | if (dev_proc_init()) |
8564 | goto out; | |
8565 | ||
8b41d188 | 8566 | if (netdev_kobject_init()) |
1da177e4 LT |
8567 | goto out; |
8568 | ||
8569 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 8570 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
8571 | INIT_LIST_HEAD(&ptype_base[i]); |
8572 | ||
62532da9 VY |
8573 | INIT_LIST_HEAD(&offload_base); |
8574 | ||
881d966b EB |
8575 | if (register_pernet_subsys(&netdev_net_ops)) |
8576 | goto out; | |
1da177e4 LT |
8577 | |
8578 | /* | |
8579 | * Initialise the packet receive queues. | |
8580 | */ | |
8581 | ||
6f912042 | 8582 | for_each_possible_cpu(i) { |
41852497 | 8583 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 8584 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 8585 | |
41852497 ED |
8586 | INIT_WORK(flush, flush_backlog); |
8587 | ||
e36fa2f7 | 8588 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 8589 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 8590 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 8591 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 8592 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
8593 | sd->csd.func = rps_trigger_softirq; |
8594 | sd->csd.info = sd; | |
e36fa2f7 | 8595 | sd->cpu = i; |
1e94d72f | 8596 | #endif |
0a9627f2 | 8597 | |
e36fa2f7 ED |
8598 | sd->backlog.poll = process_backlog; |
8599 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
8600 | } |
8601 | ||
1da177e4 LT |
8602 | dev_boot_phase = 0; |
8603 | ||
505d4f73 EB |
8604 | /* The loopback device is special if any other network devices |
8605 | * is present in a network namespace the loopback device must | |
8606 | * be present. Since we now dynamically allocate and free the | |
8607 | * loopback device ensure this invariant is maintained by | |
8608 | * keeping the loopback device as the first device on the | |
8609 | * list of network devices. Ensuring the loopback devices | |
8610 | * is the first device that appears and the last network device | |
8611 | * that disappears. | |
8612 | */ | |
8613 | if (register_pernet_device(&loopback_net_ops)) | |
8614 | goto out; | |
8615 | ||
8616 | if (register_pernet_device(&default_device_ops)) | |
8617 | goto out; | |
8618 | ||
962cf36c CM |
8619 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
8620 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 8621 | |
f0bf90de SAS |
8622 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
8623 | NULL, dev_cpu_dead); | |
8624 | WARN_ON(rc < 0); | |
f38a9eb1 | 8625 | dst_subsys_init(); |
1da177e4 LT |
8626 | rc = 0; |
8627 | out: | |
8628 | return rc; | |
8629 | } | |
8630 | ||
8631 | subsys_initcall(net_dev_init); |