]>
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 LT |
107 | #include <net/pkt_sched.h> |
108 | #include <net/checksum.h> | |
44540960 | 109 | #include <net/xfrm.h> |
1da177e4 LT |
110 | #include <linux/highmem.h> |
111 | #include <linux/init.h> | |
1da177e4 | 112 | #include <linux/module.h> |
1da177e4 LT |
113 | #include <linux/netpoll.h> |
114 | #include <linux/rcupdate.h> | |
115 | #include <linux/delay.h> | |
1da177e4 | 116 | #include <net/iw_handler.h> |
1da177e4 | 117 | #include <asm/current.h> |
5bdb9886 | 118 | #include <linux/audit.h> |
db217334 | 119 | #include <linux/dmaengine.h> |
f6a78bfc | 120 | #include <linux/err.h> |
c7fa9d18 | 121 | #include <linux/ctype.h> |
723e98b7 | 122 | #include <linux/if_arp.h> |
6de329e2 | 123 | #include <linux/if_vlan.h> |
8f0f2223 | 124 | #include <linux/ip.h> |
ad55dcaf | 125 | #include <net/ip.h> |
25cd9ba0 | 126 | #include <net/mpls.h> |
8f0f2223 DM |
127 | #include <linux/ipv6.h> |
128 | #include <linux/in.h> | |
b6b2fed1 DM |
129 | #include <linux/jhash.h> |
130 | #include <linux/random.h> | |
9cbc1cb8 | 131 | #include <trace/events/napi.h> |
cf66ba58 | 132 | #include <trace/events/net.h> |
07dc22e7 | 133 | #include <trace/events/skb.h> |
5acbbd42 | 134 | #include <linux/pci.h> |
caeda9b9 | 135 | #include <linux/inetdevice.h> |
c445477d | 136 | #include <linux/cpu_rmap.h> |
c5905afb | 137 | #include <linux/static_key.h> |
af12fa6e | 138 | #include <linux/hashtable.h> |
60877a32 | 139 | #include <linux/vmalloc.h> |
529d0489 | 140 | #include <linux/if_macvlan.h> |
e7fd2885 | 141 | #include <linux/errqueue.h> |
3b47d303 | 142 | #include <linux/hrtimer.h> |
e687ad60 | 143 | #include <linux/netfilter_ingress.h> |
40e4e713 | 144 | #include <linux/crash_dump.h> |
1da177e4 | 145 | |
342709ef PE |
146 | #include "net-sysfs.h" |
147 | ||
d565b0a1 HX |
148 | /* Instead of increasing this, you should create a hash table. */ |
149 | #define MAX_GRO_SKBS 8 | |
150 | ||
5d38a079 HX |
151 | /* This should be increased if a protocol with a bigger head is added. */ |
152 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
153 | ||
1da177e4 | 154 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 155 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
156 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
157 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 158 | static struct list_head offload_base __read_mostly; |
1da177e4 | 159 | |
ae78dbfa | 160 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
161 | static int call_netdevice_notifiers_info(unsigned long val, |
162 | struct net_device *dev, | |
163 | struct netdev_notifier_info *info); | |
ae78dbfa | 164 | |
1da177e4 | 165 | /* |
7562f876 | 166 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
167 | * semaphore. |
168 | * | |
c6d14c84 | 169 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
170 | * |
171 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 172 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
173 | * actual updates. This allows pure readers to access the list even |
174 | * while a writer is preparing to update it. | |
175 | * | |
176 | * To put it another way, dev_base_lock is held for writing only to | |
177 | * protect against pure readers; the rtnl semaphore provides the | |
178 | * protection against other writers. | |
179 | * | |
180 | * See, for example usages, register_netdevice() and | |
181 | * unregister_netdevice(), which must be called with the rtnl | |
182 | * semaphore held. | |
183 | */ | |
1da177e4 | 184 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
185 | EXPORT_SYMBOL(dev_base_lock); |
186 | ||
af12fa6e ET |
187 | /* protects napi_hash addition/deletion and napi_gen_id */ |
188 | static DEFINE_SPINLOCK(napi_hash_lock); | |
189 | ||
52bd2d62 | 190 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 191 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 192 | |
18afa4b0 | 193 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 194 | |
4e985ada TG |
195 | static inline void dev_base_seq_inc(struct net *net) |
196 | { | |
643aa9cb | 197 | while (++net->dev_base_seq == 0) |
198 | ; | |
4e985ada TG |
199 | } |
200 | ||
881d966b | 201 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 202 | { |
8387ff25 | 203 | unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); |
95c96174 | 204 | |
08e9897d | 205 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
206 | } |
207 | ||
881d966b | 208 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 209 | { |
7c28bd0b | 210 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
211 | } |
212 | ||
e36fa2f7 | 213 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
214 | { |
215 | #ifdef CONFIG_RPS | |
e36fa2f7 | 216 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
217 | #endif |
218 | } | |
219 | ||
e36fa2f7 | 220 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
221 | { |
222 | #ifdef CONFIG_RPS | |
e36fa2f7 | 223 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
224 | #endif |
225 | } | |
226 | ||
ce286d32 | 227 | /* Device list insertion */ |
53759be9 | 228 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 229 | { |
c346dca1 | 230 | struct net *net = dev_net(dev); |
ce286d32 EB |
231 | |
232 | ASSERT_RTNL(); | |
233 | ||
234 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 235 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 236 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
237 | hlist_add_head_rcu(&dev->index_hlist, |
238 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 239 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
240 | |
241 | dev_base_seq_inc(net); | |
ce286d32 EB |
242 | } |
243 | ||
fb699dfd ED |
244 | /* Device list removal |
245 | * caller must respect a RCU grace period before freeing/reusing dev | |
246 | */ | |
ce286d32 EB |
247 | static void unlist_netdevice(struct net_device *dev) |
248 | { | |
249 | ASSERT_RTNL(); | |
250 | ||
251 | /* Unlink dev from the device chain */ | |
252 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 253 | list_del_rcu(&dev->dev_list); |
72c9528b | 254 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 255 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 256 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
257 | |
258 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
259 | } |
260 | ||
1da177e4 LT |
261 | /* |
262 | * Our notifier list | |
263 | */ | |
264 | ||
f07d5b94 | 265 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
266 | |
267 | /* | |
268 | * Device drivers call our routines to queue packets here. We empty the | |
269 | * queue in the local softnet handler. | |
270 | */ | |
bea3348e | 271 | |
9958da05 | 272 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 273 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 274 | |
cf508b12 | 275 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 276 | /* |
c773e847 | 277 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
278 | * according to dev->type |
279 | */ | |
643aa9cb | 280 | static const unsigned short netdev_lock_type[] = { |
281 | ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
723e98b7 JP |
282 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, |
283 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
284 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
285 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
286 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
287 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
288 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
289 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
290 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
291 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
292 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
293 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
294 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
295 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 296 | |
643aa9cb | 297 | static const char *const netdev_lock_name[] = { |
298 | "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
299 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
300 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
301 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
302 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
303 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
304 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
305 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
306 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
307 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
308 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
309 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
310 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", | |
311 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
312 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
313 | |
314 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 315 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
316 | |
317 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
318 | { | |
319 | int i; | |
320 | ||
321 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
322 | if (netdev_lock_type[i] == dev_type) | |
323 | return i; | |
324 | /* the last key is used by default */ | |
325 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
326 | } | |
327 | ||
cf508b12 DM |
328 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
329 | unsigned short dev_type) | |
723e98b7 JP |
330 | { |
331 | int i; | |
332 | ||
333 | i = netdev_lock_pos(dev_type); | |
334 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
335 | netdev_lock_name[i]); | |
336 | } | |
cf508b12 DM |
337 | |
338 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
339 | { | |
340 | int i; | |
341 | ||
342 | i = netdev_lock_pos(dev->type); | |
343 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
344 | &netdev_addr_lock_key[i], | |
345 | netdev_lock_name[i]); | |
346 | } | |
723e98b7 | 347 | #else |
cf508b12 DM |
348 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
349 | unsigned short dev_type) | |
350 | { | |
351 | } | |
352 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
353 | { |
354 | } | |
355 | #endif | |
1da177e4 LT |
356 | |
357 | /******************************************************************************* | |
eb13da1a | 358 | * |
359 | * Protocol management and registration routines | |
360 | * | |
361 | *******************************************************************************/ | |
1da177e4 | 362 | |
1da177e4 | 363 | |
1da177e4 LT |
364 | /* |
365 | * Add a protocol ID to the list. Now that the input handler is | |
366 | * smarter we can dispense with all the messy stuff that used to be | |
367 | * here. | |
368 | * | |
369 | * BEWARE!!! Protocol handlers, mangling input packets, | |
370 | * MUST BE last in hash buckets and checking protocol handlers | |
371 | * MUST start from promiscuous ptype_all chain in net_bh. | |
372 | * It is true now, do not change it. | |
373 | * Explanation follows: if protocol handler, mangling packet, will | |
374 | * be the first on list, it is not able to sense, that packet | |
375 | * is cloned and should be copied-on-write, so that it will | |
376 | * change it and subsequent readers will get broken packet. | |
377 | * --ANK (980803) | |
378 | */ | |
379 | ||
c07b68e8 ED |
380 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
381 | { | |
382 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 383 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 384 | else |
7866a621 SN |
385 | return pt->dev ? &pt->dev->ptype_specific : |
386 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
387 | } |
388 | ||
1da177e4 LT |
389 | /** |
390 | * dev_add_pack - add packet handler | |
391 | * @pt: packet type declaration | |
392 | * | |
393 | * Add a protocol handler to the networking stack. The passed &packet_type | |
394 | * is linked into kernel lists and may not be freed until it has been | |
395 | * removed from the kernel lists. | |
396 | * | |
4ec93edb | 397 | * This call does not sleep therefore it can not |
1da177e4 LT |
398 | * guarantee all CPU's that are in middle of receiving packets |
399 | * will see the new packet type (until the next received packet). | |
400 | */ | |
401 | ||
402 | void dev_add_pack(struct packet_type *pt) | |
403 | { | |
c07b68e8 | 404 | struct list_head *head = ptype_head(pt); |
1da177e4 | 405 | |
c07b68e8 ED |
406 | spin_lock(&ptype_lock); |
407 | list_add_rcu(&pt->list, head); | |
408 | spin_unlock(&ptype_lock); | |
1da177e4 | 409 | } |
d1b19dff | 410 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 411 | |
1da177e4 LT |
412 | /** |
413 | * __dev_remove_pack - remove packet handler | |
414 | * @pt: packet type declaration | |
415 | * | |
416 | * Remove a protocol handler that was previously added to the kernel | |
417 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
418 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 419 | * returns. |
1da177e4 LT |
420 | * |
421 | * The packet type might still be in use by receivers | |
422 | * and must not be freed until after all the CPU's have gone | |
423 | * through a quiescent state. | |
424 | */ | |
425 | void __dev_remove_pack(struct packet_type *pt) | |
426 | { | |
c07b68e8 | 427 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
428 | struct packet_type *pt1; |
429 | ||
c07b68e8 | 430 | spin_lock(&ptype_lock); |
1da177e4 LT |
431 | |
432 | list_for_each_entry(pt1, head, list) { | |
433 | if (pt == pt1) { | |
434 | list_del_rcu(&pt->list); | |
435 | goto out; | |
436 | } | |
437 | } | |
438 | ||
7b6cd1ce | 439 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 440 | out: |
c07b68e8 | 441 | spin_unlock(&ptype_lock); |
1da177e4 | 442 | } |
d1b19dff ED |
443 | EXPORT_SYMBOL(__dev_remove_pack); |
444 | ||
1da177e4 LT |
445 | /** |
446 | * dev_remove_pack - remove packet handler | |
447 | * @pt: packet type declaration | |
448 | * | |
449 | * Remove a protocol handler that was previously added to the kernel | |
450 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
451 | * from the kernel lists and can be freed or reused once this function | |
452 | * returns. | |
453 | * | |
454 | * This call sleeps to guarantee that no CPU is looking at the packet | |
455 | * type after return. | |
456 | */ | |
457 | void dev_remove_pack(struct packet_type *pt) | |
458 | { | |
459 | __dev_remove_pack(pt); | |
4ec93edb | 460 | |
1da177e4 LT |
461 | synchronize_net(); |
462 | } | |
d1b19dff | 463 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 464 | |
62532da9 VY |
465 | |
466 | /** | |
467 | * dev_add_offload - register offload handlers | |
468 | * @po: protocol offload declaration | |
469 | * | |
470 | * Add protocol offload handlers to the networking stack. The passed | |
471 | * &proto_offload is linked into kernel lists and may not be freed until | |
472 | * it has been removed from the kernel lists. | |
473 | * | |
474 | * This call does not sleep therefore it can not | |
475 | * guarantee all CPU's that are in middle of receiving packets | |
476 | * will see the new offload handlers (until the next received packet). | |
477 | */ | |
478 | void dev_add_offload(struct packet_offload *po) | |
479 | { | |
bdef7de4 | 480 | struct packet_offload *elem; |
62532da9 VY |
481 | |
482 | spin_lock(&offload_lock); | |
bdef7de4 DM |
483 | list_for_each_entry(elem, &offload_base, list) { |
484 | if (po->priority < elem->priority) | |
485 | break; | |
486 | } | |
487 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
488 | spin_unlock(&offload_lock); |
489 | } | |
490 | EXPORT_SYMBOL(dev_add_offload); | |
491 | ||
492 | /** | |
493 | * __dev_remove_offload - remove offload handler | |
494 | * @po: packet offload declaration | |
495 | * | |
496 | * Remove a protocol offload handler that was previously added to the | |
497 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
498 | * is removed from the kernel lists and can be freed or reused once this | |
499 | * function returns. | |
500 | * | |
501 | * The packet type might still be in use by receivers | |
502 | * and must not be freed until after all the CPU's have gone | |
503 | * through a quiescent state. | |
504 | */ | |
1d143d9f | 505 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
506 | { |
507 | struct list_head *head = &offload_base; | |
508 | struct packet_offload *po1; | |
509 | ||
c53aa505 | 510 | spin_lock(&offload_lock); |
62532da9 VY |
511 | |
512 | list_for_each_entry(po1, head, list) { | |
513 | if (po == po1) { | |
514 | list_del_rcu(&po->list); | |
515 | goto out; | |
516 | } | |
517 | } | |
518 | ||
519 | pr_warn("dev_remove_offload: %p not found\n", po); | |
520 | out: | |
c53aa505 | 521 | spin_unlock(&offload_lock); |
62532da9 | 522 | } |
62532da9 VY |
523 | |
524 | /** | |
525 | * dev_remove_offload - remove packet offload handler | |
526 | * @po: packet offload declaration | |
527 | * | |
528 | * Remove a packet offload handler that was previously added to the kernel | |
529 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
530 | * removed from the kernel lists and can be freed or reused once this | |
531 | * function returns. | |
532 | * | |
533 | * This call sleeps to guarantee that no CPU is looking at the packet | |
534 | * type after return. | |
535 | */ | |
536 | void dev_remove_offload(struct packet_offload *po) | |
537 | { | |
538 | __dev_remove_offload(po); | |
539 | ||
540 | synchronize_net(); | |
541 | } | |
542 | EXPORT_SYMBOL(dev_remove_offload); | |
543 | ||
1da177e4 | 544 | /****************************************************************************** |
eb13da1a | 545 | * |
546 | * Device Boot-time Settings Routines | |
547 | * | |
548 | ******************************************************************************/ | |
1da177e4 LT |
549 | |
550 | /* Boot time configuration table */ | |
551 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
552 | ||
553 | /** | |
554 | * netdev_boot_setup_add - add new setup entry | |
555 | * @name: name of the device | |
556 | * @map: configured settings for the device | |
557 | * | |
558 | * Adds new setup entry to the dev_boot_setup list. The function | |
559 | * returns 0 on error and 1 on success. This is a generic routine to | |
560 | * all netdevices. | |
561 | */ | |
562 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
563 | { | |
564 | struct netdev_boot_setup *s; | |
565 | int i; | |
566 | ||
567 | s = dev_boot_setup; | |
568 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
569 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
570 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 571 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
572 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
573 | break; | |
574 | } | |
575 | } | |
576 | ||
577 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
578 | } | |
579 | ||
580 | /** | |
722c9a0c | 581 | * netdev_boot_setup_check - check boot time settings |
582 | * @dev: the netdevice | |
1da177e4 | 583 | * |
722c9a0c | 584 | * Check boot time settings for the device. |
585 | * The found settings are set for the device to be used | |
586 | * later in the device probing. | |
587 | * Returns 0 if no settings found, 1 if they are. | |
1da177e4 LT |
588 | */ |
589 | int netdev_boot_setup_check(struct net_device *dev) | |
590 | { | |
591 | struct netdev_boot_setup *s = dev_boot_setup; | |
592 | int i; | |
593 | ||
594 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
595 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 596 | !strcmp(dev->name, s[i].name)) { |
722c9a0c | 597 | dev->irq = s[i].map.irq; |
598 | dev->base_addr = s[i].map.base_addr; | |
599 | dev->mem_start = s[i].map.mem_start; | |
600 | dev->mem_end = s[i].map.mem_end; | |
1da177e4 LT |
601 | return 1; |
602 | } | |
603 | } | |
604 | return 0; | |
605 | } | |
d1b19dff | 606 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
607 | |
608 | ||
609 | /** | |
722c9a0c | 610 | * netdev_boot_base - get address from boot time settings |
611 | * @prefix: prefix for network device | |
612 | * @unit: id for network device | |
613 | * | |
614 | * Check boot time settings for the base address of device. | |
615 | * The found settings are set for the device to be used | |
616 | * later in the device probing. | |
617 | * Returns 0 if no settings found. | |
1da177e4 LT |
618 | */ |
619 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
620 | { | |
621 | const struct netdev_boot_setup *s = dev_boot_setup; | |
622 | char name[IFNAMSIZ]; | |
623 | int i; | |
624 | ||
625 | sprintf(name, "%s%d", prefix, unit); | |
626 | ||
627 | /* | |
628 | * If device already registered then return base of 1 | |
629 | * to indicate not to probe for this interface | |
630 | */ | |
881d966b | 631 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
632 | return 1; |
633 | ||
634 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
635 | if (!strcmp(name, s[i].name)) | |
636 | return s[i].map.base_addr; | |
637 | return 0; | |
638 | } | |
639 | ||
640 | /* | |
641 | * Saves at boot time configured settings for any netdevice. | |
642 | */ | |
643 | int __init netdev_boot_setup(char *str) | |
644 | { | |
645 | int ints[5]; | |
646 | struct ifmap map; | |
647 | ||
648 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
649 | if (!str || !*str) | |
650 | return 0; | |
651 | ||
652 | /* Save settings */ | |
653 | memset(&map, 0, sizeof(map)); | |
654 | if (ints[0] > 0) | |
655 | map.irq = ints[1]; | |
656 | if (ints[0] > 1) | |
657 | map.base_addr = ints[2]; | |
658 | if (ints[0] > 2) | |
659 | map.mem_start = ints[3]; | |
660 | if (ints[0] > 3) | |
661 | map.mem_end = ints[4]; | |
662 | ||
663 | /* Add new entry to the list */ | |
664 | return netdev_boot_setup_add(str, &map); | |
665 | } | |
666 | ||
667 | __setup("netdev=", netdev_boot_setup); | |
668 | ||
669 | /******************************************************************************* | |
eb13da1a | 670 | * |
671 | * Device Interface Subroutines | |
672 | * | |
673 | *******************************************************************************/ | |
1da177e4 | 674 | |
a54acb3a ND |
675 | /** |
676 | * dev_get_iflink - get 'iflink' value of a interface | |
677 | * @dev: targeted interface | |
678 | * | |
679 | * Indicates the ifindex the interface is linked to. | |
680 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
681 | */ | |
682 | ||
683 | int dev_get_iflink(const struct net_device *dev) | |
684 | { | |
685 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
686 | return dev->netdev_ops->ndo_get_iflink(dev); | |
687 | ||
7a66bbc9 | 688 | return dev->ifindex; |
a54acb3a ND |
689 | } |
690 | EXPORT_SYMBOL(dev_get_iflink); | |
691 | ||
fc4099f1 PS |
692 | /** |
693 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
694 | * @dev: targeted interface | |
695 | * @skb: The packet. | |
696 | * | |
697 | * For better visibility of tunnel traffic OVS needs to retrieve | |
698 | * egress tunnel information for a packet. Following API allows | |
699 | * user to get this info. | |
700 | */ | |
701 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
702 | { | |
703 | struct ip_tunnel_info *info; | |
704 | ||
705 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
706 | return -EINVAL; | |
707 | ||
708 | info = skb_tunnel_info_unclone(skb); | |
709 | if (!info) | |
710 | return -ENOMEM; | |
711 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
712 | return -EINVAL; | |
713 | ||
714 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
715 | } | |
716 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
717 | ||
1da177e4 LT |
718 | /** |
719 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 720 | * @net: the applicable net namespace |
1da177e4 LT |
721 | * @name: name to find |
722 | * | |
723 | * Find an interface by name. Must be called under RTNL semaphore | |
724 | * or @dev_base_lock. If the name is found a pointer to the device | |
725 | * is returned. If the name is not found then %NULL is returned. The | |
726 | * reference counters are not incremented so the caller must be | |
727 | * careful with locks. | |
728 | */ | |
729 | ||
881d966b | 730 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 731 | { |
0bd8d536 ED |
732 | struct net_device *dev; |
733 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 734 | |
b67bfe0d | 735 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
736 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
737 | return dev; | |
0bd8d536 | 738 | |
1da177e4 LT |
739 | return NULL; |
740 | } | |
d1b19dff | 741 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 742 | |
72c9528b | 743 | /** |
722c9a0c | 744 | * dev_get_by_name_rcu - find a device by its name |
745 | * @net: the applicable net namespace | |
746 | * @name: name to find | |
747 | * | |
748 | * Find an interface by name. | |
749 | * If the name is found a pointer to the device is returned. | |
750 | * If the name is not found then %NULL is returned. | |
751 | * The reference counters are not incremented so the caller must be | |
752 | * careful with locks. The caller must hold RCU lock. | |
72c9528b ED |
753 | */ |
754 | ||
755 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
756 | { | |
72c9528b ED |
757 | struct net_device *dev; |
758 | struct hlist_head *head = dev_name_hash(net, name); | |
759 | ||
b67bfe0d | 760 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
761 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
762 | return dev; | |
763 | ||
764 | return NULL; | |
765 | } | |
766 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
767 | ||
1da177e4 LT |
768 | /** |
769 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 770 | * @net: the applicable net namespace |
1da177e4 LT |
771 | * @name: name to find |
772 | * | |
773 | * Find an interface by name. This can be called from any | |
774 | * context and does its own locking. The returned handle has | |
775 | * the usage count incremented and the caller must use dev_put() to | |
776 | * release it when it is no longer needed. %NULL is returned if no | |
777 | * matching device is found. | |
778 | */ | |
779 | ||
881d966b | 780 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
781 | { |
782 | struct net_device *dev; | |
783 | ||
72c9528b ED |
784 | rcu_read_lock(); |
785 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
786 | if (dev) |
787 | dev_hold(dev); | |
72c9528b | 788 | rcu_read_unlock(); |
1da177e4 LT |
789 | return dev; |
790 | } | |
d1b19dff | 791 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
792 | |
793 | /** | |
794 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 795 | * @net: the applicable net namespace |
1da177e4 LT |
796 | * @ifindex: index of device |
797 | * | |
798 | * Search for an interface by index. Returns %NULL if the device | |
799 | * is not found or a pointer to the device. The device has not | |
800 | * had its reference counter increased so the caller must be careful | |
801 | * about locking. The caller must hold either the RTNL semaphore | |
802 | * or @dev_base_lock. | |
803 | */ | |
804 | ||
881d966b | 805 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 806 | { |
0bd8d536 ED |
807 | struct net_device *dev; |
808 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 809 | |
b67bfe0d | 810 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
811 | if (dev->ifindex == ifindex) |
812 | return dev; | |
0bd8d536 | 813 | |
1da177e4 LT |
814 | return NULL; |
815 | } | |
d1b19dff | 816 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 817 | |
fb699dfd ED |
818 | /** |
819 | * dev_get_by_index_rcu - find a device by its ifindex | |
820 | * @net: the applicable net namespace | |
821 | * @ifindex: index of device | |
822 | * | |
823 | * Search for an interface by index. Returns %NULL if the device | |
824 | * is not found or a pointer to the device. The device has not | |
825 | * had its reference counter increased so the caller must be careful | |
826 | * about locking. The caller must hold RCU lock. | |
827 | */ | |
828 | ||
829 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
830 | { | |
fb699dfd ED |
831 | struct net_device *dev; |
832 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
833 | ||
b67bfe0d | 834 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
835 | if (dev->ifindex == ifindex) |
836 | return dev; | |
837 | ||
838 | return NULL; | |
839 | } | |
840 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
841 | ||
1da177e4 LT |
842 | |
843 | /** | |
844 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 845 | * @net: the applicable net namespace |
1da177e4 LT |
846 | * @ifindex: index of device |
847 | * | |
848 | * Search for an interface by index. Returns NULL if the device | |
849 | * is not found or a pointer to the device. The device returned has | |
850 | * had a reference added and the pointer is safe until the user calls | |
851 | * dev_put to indicate they have finished with it. | |
852 | */ | |
853 | ||
881d966b | 854 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
855 | { |
856 | struct net_device *dev; | |
857 | ||
fb699dfd ED |
858 | rcu_read_lock(); |
859 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
860 | if (dev) |
861 | dev_hold(dev); | |
fb699dfd | 862 | rcu_read_unlock(); |
1da177e4 LT |
863 | return dev; |
864 | } | |
d1b19dff | 865 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 866 | |
5dbe7c17 NS |
867 | /** |
868 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
869 | * @net: network namespace | |
870 | * @name: a pointer to the buffer where the name will be stored. | |
871 | * @ifindex: the ifindex of the interface to get the name from. | |
872 | * | |
873 | * The use of raw_seqcount_begin() and cond_resched() before | |
874 | * retrying is required as we want to give the writers a chance | |
875 | * to complete when CONFIG_PREEMPT is not set. | |
876 | */ | |
877 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
878 | { | |
879 | struct net_device *dev; | |
880 | unsigned int seq; | |
881 | ||
882 | retry: | |
883 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
884 | rcu_read_lock(); | |
885 | dev = dev_get_by_index_rcu(net, ifindex); | |
886 | if (!dev) { | |
887 | rcu_read_unlock(); | |
888 | return -ENODEV; | |
889 | } | |
890 | ||
891 | strcpy(name, dev->name); | |
892 | rcu_read_unlock(); | |
893 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
894 | cond_resched(); | |
895 | goto retry; | |
896 | } | |
897 | ||
898 | return 0; | |
899 | } | |
900 | ||
1da177e4 | 901 | /** |
941666c2 | 902 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 903 | * @net: the applicable net namespace |
1da177e4 LT |
904 | * @type: media type of device |
905 | * @ha: hardware address | |
906 | * | |
907 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
908 | * is not found or a pointer to the device. |
909 | * The caller must hold RCU or RTNL. | |
941666c2 | 910 | * The returned device has not had its ref count increased |
1da177e4 LT |
911 | * and the caller must therefore be careful about locking |
912 | * | |
1da177e4 LT |
913 | */ |
914 | ||
941666c2 ED |
915 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
916 | const char *ha) | |
1da177e4 LT |
917 | { |
918 | struct net_device *dev; | |
919 | ||
941666c2 | 920 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
921 | if (dev->type == type && |
922 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
923 | return dev; |
924 | ||
925 | return NULL; | |
1da177e4 | 926 | } |
941666c2 | 927 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 928 | |
881d966b | 929 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
930 | { |
931 | struct net_device *dev; | |
932 | ||
4e9cac2b | 933 | ASSERT_RTNL(); |
881d966b | 934 | for_each_netdev(net, dev) |
4e9cac2b | 935 | if (dev->type == type) |
7562f876 PE |
936 | return dev; |
937 | ||
938 | return NULL; | |
4e9cac2b | 939 | } |
4e9cac2b PM |
940 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
941 | ||
881d966b | 942 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 943 | { |
99fe3c39 | 944 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 945 | |
99fe3c39 ED |
946 | rcu_read_lock(); |
947 | for_each_netdev_rcu(net, dev) | |
948 | if (dev->type == type) { | |
949 | dev_hold(dev); | |
950 | ret = dev; | |
951 | break; | |
952 | } | |
953 | rcu_read_unlock(); | |
954 | return ret; | |
1da177e4 | 955 | } |
1da177e4 LT |
956 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
957 | ||
958 | /** | |
6c555490 | 959 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 960 | * @net: the applicable net namespace |
1da177e4 LT |
961 | * @if_flags: IFF_* values |
962 | * @mask: bitmask of bits in if_flags to check | |
963 | * | |
964 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 965 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 966 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
967 | */ |
968 | ||
6c555490 WC |
969 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
970 | unsigned short mask) | |
1da177e4 | 971 | { |
7562f876 | 972 | struct net_device *dev, *ret; |
1da177e4 | 973 | |
6c555490 WC |
974 | ASSERT_RTNL(); |
975 | ||
7562f876 | 976 | ret = NULL; |
6c555490 | 977 | for_each_netdev(net, dev) { |
1da177e4 | 978 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 979 | ret = dev; |
1da177e4 LT |
980 | break; |
981 | } | |
982 | } | |
7562f876 | 983 | return ret; |
1da177e4 | 984 | } |
6c555490 | 985 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
986 | |
987 | /** | |
988 | * dev_valid_name - check if name is okay for network device | |
989 | * @name: name string | |
990 | * | |
991 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
992 | * to allow sysfs to work. We also disallow any kind of |
993 | * whitespace. | |
1da177e4 | 994 | */ |
95f050bf | 995 | bool dev_valid_name(const char *name) |
1da177e4 | 996 | { |
c7fa9d18 | 997 | if (*name == '\0') |
95f050bf | 998 | return false; |
b6fe17d6 | 999 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 1000 | return false; |
c7fa9d18 | 1001 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1002 | return false; |
c7fa9d18 DM |
1003 | |
1004 | while (*name) { | |
a4176a93 | 1005 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1006 | return false; |
c7fa9d18 DM |
1007 | name++; |
1008 | } | |
95f050bf | 1009 | return true; |
1da177e4 | 1010 | } |
d1b19dff | 1011 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1012 | |
1013 | /** | |
b267b179 EB |
1014 | * __dev_alloc_name - allocate a name for a device |
1015 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1016 | * @name: name format string |
b267b179 | 1017 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1018 | * |
1019 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1020 | * id. It scans list of devices to build up a free map, then chooses |
1021 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1022 | * while allocating the name and adding the device in order to avoid | |
1023 | * duplicates. | |
1024 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1025 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1026 | */ |
1027 | ||
b267b179 | 1028 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1029 | { |
1030 | int i = 0; | |
1da177e4 LT |
1031 | const char *p; |
1032 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1033 | unsigned long *inuse; |
1da177e4 LT |
1034 | struct net_device *d; |
1035 | ||
1036 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
1037 | if (p) { | |
1038 | /* | |
1039 | * Verify the string as this thing may have come from | |
1040 | * the user. There must be either one "%d" and no other "%" | |
1041 | * characters. | |
1042 | */ | |
1043 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1044 | return -EINVAL; | |
1045 | ||
1046 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1047 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1048 | if (!inuse) |
1049 | return -ENOMEM; | |
1050 | ||
881d966b | 1051 | for_each_netdev(net, d) { |
1da177e4 LT |
1052 | if (!sscanf(d->name, name, &i)) |
1053 | continue; | |
1054 | if (i < 0 || i >= max_netdevices) | |
1055 | continue; | |
1056 | ||
1057 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1058 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1059 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1060 | set_bit(i, inuse); | |
1061 | } | |
1062 | ||
1063 | i = find_first_zero_bit(inuse, max_netdevices); | |
1064 | free_page((unsigned long) inuse); | |
1065 | } | |
1066 | ||
d9031024 OP |
1067 | if (buf != name) |
1068 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1069 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1070 | return i; |
1da177e4 LT |
1071 | |
1072 | /* It is possible to run out of possible slots | |
1073 | * when the name is long and there isn't enough space left | |
1074 | * for the digits, or if all bits are used. | |
1075 | */ | |
1076 | return -ENFILE; | |
1077 | } | |
1078 | ||
b267b179 EB |
1079 | /** |
1080 | * dev_alloc_name - allocate a name for a device | |
1081 | * @dev: device | |
1082 | * @name: name format string | |
1083 | * | |
1084 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1085 | * id. It scans list of devices to build up a free map, then chooses | |
1086 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1087 | * while allocating the name and adding the device in order to avoid | |
1088 | * duplicates. | |
1089 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1090 | * Returns the number of the unit assigned or a negative errno code. | |
1091 | */ | |
1092 | ||
1093 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1094 | { | |
1095 | char buf[IFNAMSIZ]; | |
1096 | struct net *net; | |
1097 | int ret; | |
1098 | ||
c346dca1 YH |
1099 | BUG_ON(!dev_net(dev)); |
1100 | net = dev_net(dev); | |
b267b179 EB |
1101 | ret = __dev_alloc_name(net, name, buf); |
1102 | if (ret >= 0) | |
1103 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1104 | return ret; | |
1105 | } | |
d1b19dff | 1106 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1107 | |
828de4f6 G |
1108 | static int dev_alloc_name_ns(struct net *net, |
1109 | struct net_device *dev, | |
1110 | const char *name) | |
d9031024 | 1111 | { |
828de4f6 G |
1112 | char buf[IFNAMSIZ]; |
1113 | int ret; | |
8ce6cebc | 1114 | |
828de4f6 G |
1115 | ret = __dev_alloc_name(net, name, buf); |
1116 | if (ret >= 0) | |
1117 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1118 | return ret; | |
1119 | } | |
1120 | ||
1121 | static int dev_get_valid_name(struct net *net, | |
1122 | struct net_device *dev, | |
1123 | const char *name) | |
1124 | { | |
1125 | BUG_ON(!net); | |
8ce6cebc | 1126 | |
d9031024 OP |
1127 | if (!dev_valid_name(name)) |
1128 | return -EINVAL; | |
1129 | ||
1c5cae81 | 1130 | if (strchr(name, '%')) |
828de4f6 | 1131 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1132 | else if (__dev_get_by_name(net, name)) |
1133 | return -EEXIST; | |
8ce6cebc DL |
1134 | else if (dev->name != name) |
1135 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1136 | |
1137 | return 0; | |
1138 | } | |
1da177e4 LT |
1139 | |
1140 | /** | |
1141 | * dev_change_name - change name of a device | |
1142 | * @dev: device | |
1143 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1144 | * | |
1145 | * Change name of a device, can pass format strings "eth%d". | |
1146 | * for wildcarding. | |
1147 | */ | |
cf04a4c7 | 1148 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1149 | { |
238fa362 | 1150 | unsigned char old_assign_type; |
fcc5a03a | 1151 | char oldname[IFNAMSIZ]; |
1da177e4 | 1152 | int err = 0; |
fcc5a03a | 1153 | int ret; |
881d966b | 1154 | struct net *net; |
1da177e4 LT |
1155 | |
1156 | ASSERT_RTNL(); | |
c346dca1 | 1157 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1158 | |
c346dca1 | 1159 | net = dev_net(dev); |
1da177e4 LT |
1160 | if (dev->flags & IFF_UP) |
1161 | return -EBUSY; | |
1162 | ||
30e6c9fa | 1163 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1164 | |
1165 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1166 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1167 | return 0; |
c91f6df2 | 1168 | } |
c8d90dca | 1169 | |
fcc5a03a HX |
1170 | memcpy(oldname, dev->name, IFNAMSIZ); |
1171 | ||
828de4f6 | 1172 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1173 | if (err < 0) { |
30e6c9fa | 1174 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1175 | return err; |
c91f6df2 | 1176 | } |
1da177e4 | 1177 | |
6fe82a39 VF |
1178 | if (oldname[0] && !strchr(oldname, '%')) |
1179 | netdev_info(dev, "renamed from %s\n", oldname); | |
1180 | ||
238fa362 TG |
1181 | old_assign_type = dev->name_assign_type; |
1182 | dev->name_assign_type = NET_NAME_RENAMED; | |
1183 | ||
fcc5a03a | 1184 | rollback: |
a1b3f594 EB |
1185 | ret = device_rename(&dev->dev, dev->name); |
1186 | if (ret) { | |
1187 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1188 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1189 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1190 | return ret; |
dcc99773 | 1191 | } |
7f988eab | 1192 | |
30e6c9fa | 1193 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1194 | |
5bb025fa VF |
1195 | netdev_adjacent_rename_links(dev, oldname); |
1196 | ||
7f988eab | 1197 | write_lock_bh(&dev_base_lock); |
372b2312 | 1198 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1199 | write_unlock_bh(&dev_base_lock); |
1200 | ||
1201 | synchronize_rcu(); | |
1202 | ||
1203 | write_lock_bh(&dev_base_lock); | |
1204 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1205 | write_unlock_bh(&dev_base_lock); |
1206 | ||
056925ab | 1207 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1208 | ret = notifier_to_errno(ret); |
1209 | ||
1210 | if (ret) { | |
91e9c07b ED |
1211 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1212 | if (err >= 0) { | |
fcc5a03a | 1213 | err = ret; |
30e6c9fa | 1214 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1215 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1216 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1217 | dev->name_assign_type = old_assign_type; |
1218 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1219 | goto rollback; |
91e9c07b | 1220 | } else { |
7b6cd1ce | 1221 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1222 | dev->name, ret); |
fcc5a03a HX |
1223 | } |
1224 | } | |
1da177e4 LT |
1225 | |
1226 | return err; | |
1227 | } | |
1228 | ||
0b815a1a SH |
1229 | /** |
1230 | * dev_set_alias - change ifalias of a device | |
1231 | * @dev: device | |
1232 | * @alias: name up to IFALIASZ | |
f0db275a | 1233 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1234 | * |
1235 | * Set ifalias for a device, | |
1236 | */ | |
1237 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1238 | { | |
7364e445 AK |
1239 | char *new_ifalias; |
1240 | ||
0b815a1a SH |
1241 | ASSERT_RTNL(); |
1242 | ||
1243 | if (len >= IFALIASZ) | |
1244 | return -EINVAL; | |
1245 | ||
96ca4a2c | 1246 | if (!len) { |
388dfc2d SK |
1247 | kfree(dev->ifalias); |
1248 | dev->ifalias = NULL; | |
96ca4a2c OH |
1249 | return 0; |
1250 | } | |
1251 | ||
7364e445 AK |
1252 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1253 | if (!new_ifalias) | |
0b815a1a | 1254 | return -ENOMEM; |
7364e445 | 1255 | dev->ifalias = new_ifalias; |
c28294b9 AP |
1256 | memcpy(dev->ifalias, alias, len); |
1257 | dev->ifalias[len] = 0; | |
0b815a1a | 1258 | |
0b815a1a SH |
1259 | return len; |
1260 | } | |
1261 | ||
1262 | ||
d8a33ac4 | 1263 | /** |
3041a069 | 1264 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1265 | * @dev: device to cause notification |
1266 | * | |
1267 | * Called to indicate a device has changed features. | |
1268 | */ | |
1269 | void netdev_features_change(struct net_device *dev) | |
1270 | { | |
056925ab | 1271 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1272 | } |
1273 | EXPORT_SYMBOL(netdev_features_change); | |
1274 | ||
1da177e4 LT |
1275 | /** |
1276 | * netdev_state_change - device changes state | |
1277 | * @dev: device to cause notification | |
1278 | * | |
1279 | * Called to indicate a device has changed state. This function calls | |
1280 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1281 | * to the routing socket. | |
1282 | */ | |
1283 | void netdev_state_change(struct net_device *dev) | |
1284 | { | |
1285 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1286 | struct netdev_notifier_change_info change_info; |
1287 | ||
1288 | change_info.flags_changed = 0; | |
1289 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1290 | &change_info.info); | |
7f294054 | 1291 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1292 | } |
1293 | } | |
d1b19dff | 1294 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1295 | |
ee89bab1 | 1296 | /** |
722c9a0c | 1297 | * netdev_notify_peers - notify network peers about existence of @dev |
1298 | * @dev: network device | |
ee89bab1 AW |
1299 | * |
1300 | * Generate traffic such that interested network peers are aware of | |
1301 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1302 | * a device wants to inform the rest of the network about some sort of | |
1303 | * reconfiguration such as a failover event or virtual machine | |
1304 | * migration. | |
1305 | */ | |
1306 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1307 | { |
ee89bab1 AW |
1308 | rtnl_lock(); |
1309 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
37c343b4 | 1310 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); |
ee89bab1 | 1311 | rtnl_unlock(); |
c1da4ac7 | 1312 | } |
ee89bab1 | 1313 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1314 | |
bd380811 | 1315 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1316 | { |
d314774c | 1317 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1318 | int ret; |
1da177e4 | 1319 | |
e46b66bc BH |
1320 | ASSERT_RTNL(); |
1321 | ||
1da177e4 LT |
1322 | if (!netif_device_present(dev)) |
1323 | return -ENODEV; | |
1324 | ||
ca99ca14 NH |
1325 | /* Block netpoll from trying to do any rx path servicing. |
1326 | * If we don't do this there is a chance ndo_poll_controller | |
1327 | * or ndo_poll may be running while we open the device | |
1328 | */ | |
66b5552f | 1329 | netpoll_poll_disable(dev); |
ca99ca14 | 1330 | |
3b8bcfd5 JB |
1331 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1332 | ret = notifier_to_errno(ret); | |
1333 | if (ret) | |
1334 | return ret; | |
1335 | ||
1da177e4 | 1336 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1337 | |
d314774c SH |
1338 | if (ops->ndo_validate_addr) |
1339 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1340 | |
d314774c SH |
1341 | if (!ret && ops->ndo_open) |
1342 | ret = ops->ndo_open(dev); | |
1da177e4 | 1343 | |
66b5552f | 1344 | netpoll_poll_enable(dev); |
ca99ca14 | 1345 | |
bada339b JG |
1346 | if (ret) |
1347 | clear_bit(__LINK_STATE_START, &dev->state); | |
1348 | else { | |
1da177e4 | 1349 | dev->flags |= IFF_UP; |
4417da66 | 1350 | dev_set_rx_mode(dev); |
1da177e4 | 1351 | dev_activate(dev); |
7bf23575 | 1352 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1353 | } |
bada339b | 1354 | |
1da177e4 LT |
1355 | return ret; |
1356 | } | |
1357 | ||
1358 | /** | |
bd380811 PM |
1359 | * dev_open - prepare an interface for use. |
1360 | * @dev: device to open | |
1da177e4 | 1361 | * |
bd380811 PM |
1362 | * Takes a device from down to up state. The device's private open |
1363 | * function is invoked and then the multicast lists are loaded. Finally | |
1364 | * the device is moved into the up state and a %NETDEV_UP message is | |
1365 | * sent to the netdev notifier chain. | |
1366 | * | |
1367 | * Calling this function on an active interface is a nop. On a failure | |
1368 | * a negative errno code is returned. | |
1da177e4 | 1369 | */ |
bd380811 PM |
1370 | int dev_open(struct net_device *dev) |
1371 | { | |
1372 | int ret; | |
1373 | ||
bd380811 PM |
1374 | if (dev->flags & IFF_UP) |
1375 | return 0; | |
1376 | ||
bd380811 PM |
1377 | ret = __dev_open(dev); |
1378 | if (ret < 0) | |
1379 | return ret; | |
1380 | ||
7f294054 | 1381 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1382 | call_netdevice_notifiers(NETDEV_UP, dev); |
1383 | ||
1384 | return ret; | |
1385 | } | |
1386 | EXPORT_SYMBOL(dev_open); | |
1387 | ||
44345724 | 1388 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1389 | { |
44345724 | 1390 | struct net_device *dev; |
e46b66bc | 1391 | |
bd380811 | 1392 | ASSERT_RTNL(); |
9d5010db DM |
1393 | might_sleep(); |
1394 | ||
5cde2829 | 1395 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1396 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1397 | netpoll_poll_disable(dev); |
3f4df206 | 1398 | |
44345724 | 1399 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1400 | |
44345724 | 1401 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1402 | |
44345724 OP |
1403 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1404 | * can be even on different cpu. So just clear netif_running(). | |
1405 | * | |
1406 | * dev->stop() will invoke napi_disable() on all of it's | |
1407 | * napi_struct instances on this device. | |
1408 | */ | |
4e857c58 | 1409 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1410 | } |
1da177e4 | 1411 | |
44345724 | 1412 | dev_deactivate_many(head); |
d8b2a4d2 | 1413 | |
5cde2829 | 1414 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1415 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1416 | |
44345724 OP |
1417 | /* |
1418 | * Call the device specific close. This cannot fail. | |
1419 | * Only if device is UP | |
1420 | * | |
1421 | * We allow it to be called even after a DETACH hot-plug | |
1422 | * event. | |
1423 | */ | |
1424 | if (ops->ndo_stop) | |
1425 | ops->ndo_stop(dev); | |
1426 | ||
44345724 | 1427 | dev->flags &= ~IFF_UP; |
66b5552f | 1428 | netpoll_poll_enable(dev); |
44345724 OP |
1429 | } |
1430 | ||
1431 | return 0; | |
1432 | } | |
1433 | ||
1434 | static int __dev_close(struct net_device *dev) | |
1435 | { | |
f87e6f47 | 1436 | int retval; |
44345724 OP |
1437 | LIST_HEAD(single); |
1438 | ||
5cde2829 | 1439 | list_add(&dev->close_list, &single); |
f87e6f47 LT |
1440 | retval = __dev_close_many(&single); |
1441 | list_del(&single); | |
ca99ca14 | 1442 | |
f87e6f47 | 1443 | return retval; |
44345724 OP |
1444 | } |
1445 | ||
99c4a26a | 1446 | int dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1447 | { |
1448 | struct net_device *dev, *tmp; | |
1da177e4 | 1449 | |
5cde2829 EB |
1450 | /* Remove the devices that don't need to be closed */ |
1451 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1452 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1453 | list_del_init(&dev->close_list); |
44345724 OP |
1454 | |
1455 | __dev_close_many(head); | |
1da177e4 | 1456 | |
5cde2829 | 1457 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1458 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1459 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1460 | if (unlink) |
1461 | list_del_init(&dev->close_list); | |
44345724 | 1462 | } |
bd380811 PM |
1463 | |
1464 | return 0; | |
1465 | } | |
99c4a26a | 1466 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1467 | |
1468 | /** | |
1469 | * dev_close - shutdown an interface. | |
1470 | * @dev: device to shutdown | |
1471 | * | |
1472 | * This function moves an active device into down state. A | |
1473 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1474 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1475 | * chain. | |
1476 | */ | |
1477 | int dev_close(struct net_device *dev) | |
1478 | { | |
e14a5993 ED |
1479 | if (dev->flags & IFF_UP) { |
1480 | LIST_HEAD(single); | |
1da177e4 | 1481 | |
5cde2829 | 1482 | list_add(&dev->close_list, &single); |
99c4a26a | 1483 | dev_close_many(&single, true); |
e14a5993 ED |
1484 | list_del(&single); |
1485 | } | |
da6e378b | 1486 | return 0; |
1da177e4 | 1487 | } |
d1b19dff | 1488 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1489 | |
1490 | ||
0187bdfb BH |
1491 | /** |
1492 | * dev_disable_lro - disable Large Receive Offload on a device | |
1493 | * @dev: device | |
1494 | * | |
1495 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1496 | * called under RTNL. This is needed if received packets may be | |
1497 | * forwarded to another interface. | |
1498 | */ | |
1499 | void dev_disable_lro(struct net_device *dev) | |
1500 | { | |
fbe168ba MK |
1501 | struct net_device *lower_dev; |
1502 | struct list_head *iter; | |
529d0489 | 1503 | |
bc5787c6 MM |
1504 | dev->wanted_features &= ~NETIF_F_LRO; |
1505 | netdev_update_features(dev); | |
27660515 | 1506 | |
22d5969f MM |
1507 | if (unlikely(dev->features & NETIF_F_LRO)) |
1508 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1509 | |
1510 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1511 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1512 | } |
1513 | EXPORT_SYMBOL(dev_disable_lro); | |
1514 | ||
351638e7 JP |
1515 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1516 | struct net_device *dev) | |
1517 | { | |
1518 | struct netdev_notifier_info info; | |
1519 | ||
1520 | netdev_notifier_info_init(&info, dev); | |
1521 | return nb->notifier_call(nb, val, &info); | |
1522 | } | |
0187bdfb | 1523 | |
881d966b EB |
1524 | static int dev_boot_phase = 1; |
1525 | ||
1da177e4 | 1526 | /** |
722c9a0c | 1527 | * register_netdevice_notifier - register a network notifier block |
1528 | * @nb: notifier | |
1da177e4 | 1529 | * |
722c9a0c | 1530 | * Register a notifier to be called when network device events occur. |
1531 | * The notifier passed is linked into the kernel structures and must | |
1532 | * not be reused until it has been unregistered. A negative errno code | |
1533 | * is returned on a failure. | |
1da177e4 | 1534 | * |
722c9a0c | 1535 | * When registered all registration and up events are replayed |
1536 | * to the new notifier to allow device to have a race free | |
1537 | * view of the network device list. | |
1da177e4 LT |
1538 | */ |
1539 | ||
1540 | int register_netdevice_notifier(struct notifier_block *nb) | |
1541 | { | |
1542 | struct net_device *dev; | |
fcc5a03a | 1543 | struct net_device *last; |
881d966b | 1544 | struct net *net; |
1da177e4 LT |
1545 | int err; |
1546 | ||
1547 | rtnl_lock(); | |
f07d5b94 | 1548 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1549 | if (err) |
1550 | goto unlock; | |
881d966b EB |
1551 | if (dev_boot_phase) |
1552 | goto unlock; | |
1553 | for_each_net(net) { | |
1554 | for_each_netdev(net, dev) { | |
351638e7 | 1555 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1556 | err = notifier_to_errno(err); |
1557 | if (err) | |
1558 | goto rollback; | |
1559 | ||
1560 | if (!(dev->flags & IFF_UP)) | |
1561 | continue; | |
1da177e4 | 1562 | |
351638e7 | 1563 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1564 | } |
1da177e4 | 1565 | } |
fcc5a03a HX |
1566 | |
1567 | unlock: | |
1da177e4 LT |
1568 | rtnl_unlock(); |
1569 | return err; | |
fcc5a03a HX |
1570 | |
1571 | rollback: | |
1572 | last = dev; | |
881d966b EB |
1573 | for_each_net(net) { |
1574 | for_each_netdev(net, dev) { | |
1575 | if (dev == last) | |
8f891489 | 1576 | goto outroll; |
fcc5a03a | 1577 | |
881d966b | 1578 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1579 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1580 | dev); | |
1581 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1582 | } |
351638e7 | 1583 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1584 | } |
fcc5a03a | 1585 | } |
c67625a1 | 1586 | |
8f891489 | 1587 | outroll: |
c67625a1 | 1588 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1589 | goto unlock; |
1da177e4 | 1590 | } |
d1b19dff | 1591 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1592 | |
1593 | /** | |
722c9a0c | 1594 | * unregister_netdevice_notifier - unregister a network notifier block |
1595 | * @nb: notifier | |
1da177e4 | 1596 | * |
722c9a0c | 1597 | * Unregister a notifier previously registered by |
1598 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1599 | * kernel structures and may then be reused. A negative errno code | |
1600 | * is returned on a failure. | |
7d3d43da | 1601 | * |
722c9a0c | 1602 | * After unregistering unregister and down device events are synthesized |
1603 | * for all devices on the device list to the removed notifier to remove | |
1604 | * the need for special case cleanup code. | |
1da177e4 LT |
1605 | */ |
1606 | ||
1607 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1608 | { | |
7d3d43da EB |
1609 | struct net_device *dev; |
1610 | struct net *net; | |
9f514950 HX |
1611 | int err; |
1612 | ||
1613 | rtnl_lock(); | |
f07d5b94 | 1614 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1615 | if (err) |
1616 | goto unlock; | |
1617 | ||
1618 | for_each_net(net) { | |
1619 | for_each_netdev(net, dev) { | |
1620 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1621 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1622 | dev); | |
1623 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1624 | } |
351638e7 | 1625 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1626 | } |
1627 | } | |
1628 | unlock: | |
9f514950 HX |
1629 | rtnl_unlock(); |
1630 | return err; | |
1da177e4 | 1631 | } |
d1b19dff | 1632 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1633 | |
351638e7 JP |
1634 | /** |
1635 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1636 | * @val: value passed unmodified to notifier function | |
1637 | * @dev: net_device pointer passed unmodified to notifier function | |
1638 | * @info: notifier information data | |
1639 | * | |
1640 | * Call all network notifier blocks. Parameters and return value | |
1641 | * are as for raw_notifier_call_chain(). | |
1642 | */ | |
1643 | ||
1d143d9f | 1644 | static int call_netdevice_notifiers_info(unsigned long val, |
1645 | struct net_device *dev, | |
1646 | struct netdev_notifier_info *info) | |
351638e7 JP |
1647 | { |
1648 | ASSERT_RTNL(); | |
1649 | netdev_notifier_info_init(info, dev); | |
1650 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1651 | } | |
351638e7 | 1652 | |
1da177e4 LT |
1653 | /** |
1654 | * call_netdevice_notifiers - call all network notifier blocks | |
1655 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1656 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1657 | * |
1658 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1659 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1660 | */ |
1661 | ||
ad7379d4 | 1662 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1663 | { |
351638e7 JP |
1664 | struct netdev_notifier_info info; |
1665 | ||
1666 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1667 | } |
edf947f1 | 1668 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1669 | |
1cf51900 | 1670 | #ifdef CONFIG_NET_INGRESS |
4577139b DB |
1671 | static struct static_key ingress_needed __read_mostly; |
1672 | ||
1673 | void net_inc_ingress_queue(void) | |
1674 | { | |
1675 | static_key_slow_inc(&ingress_needed); | |
1676 | } | |
1677 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
1678 | ||
1679 | void net_dec_ingress_queue(void) | |
1680 | { | |
1681 | static_key_slow_dec(&ingress_needed); | |
1682 | } | |
1683 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
1684 | #endif | |
1685 | ||
1f211a1b DB |
1686 | #ifdef CONFIG_NET_EGRESS |
1687 | static struct static_key egress_needed __read_mostly; | |
1688 | ||
1689 | void net_inc_egress_queue(void) | |
1690 | { | |
1691 | static_key_slow_inc(&egress_needed); | |
1692 | } | |
1693 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
1694 | ||
1695 | void net_dec_egress_queue(void) | |
1696 | { | |
1697 | static_key_slow_dec(&egress_needed); | |
1698 | } | |
1699 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
1700 | #endif | |
1701 | ||
c5905afb | 1702 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1703 | #ifdef HAVE_JUMP_LABEL |
b90e5794 | 1704 | static atomic_t netstamp_needed_deferred; |
13baa00a | 1705 | static atomic_t netstamp_wanted; |
5fa8bbda | 1706 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 1707 | { |
b90e5794 | 1708 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 1709 | int wanted; |
b90e5794 | 1710 | |
13baa00a ED |
1711 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
1712 | if (wanted > 0) | |
1713 | static_key_enable(&netstamp_needed); | |
1714 | else | |
1715 | static_key_disable(&netstamp_needed); | |
5fa8bbda ED |
1716 | } |
1717 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 1718 | #endif |
5fa8bbda ED |
1719 | |
1720 | void net_enable_timestamp(void) | |
1721 | { | |
13baa00a ED |
1722 | #ifdef HAVE_JUMP_LABEL |
1723 | int wanted; | |
1724 | ||
1725 | while (1) { | |
1726 | wanted = atomic_read(&netstamp_wanted); | |
1727 | if (wanted <= 0) | |
1728 | break; | |
1729 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) | |
1730 | return; | |
1731 | } | |
1732 | atomic_inc(&netstamp_needed_deferred); | |
1733 | schedule_work(&netstamp_work); | |
1734 | #else | |
c5905afb | 1735 | static_key_slow_inc(&netstamp_needed); |
13baa00a | 1736 | #endif |
1da177e4 | 1737 | } |
d1b19dff | 1738 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1739 | |
1740 | void net_disable_timestamp(void) | |
1741 | { | |
b90e5794 | 1742 | #ifdef HAVE_JUMP_LABEL |
13baa00a ED |
1743 | int wanted; |
1744 | ||
1745 | while (1) { | |
1746 | wanted = atomic_read(&netstamp_wanted); | |
1747 | if (wanted <= 1) | |
1748 | break; | |
1749 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) | |
1750 | return; | |
1751 | } | |
1752 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
1753 | schedule_work(&netstamp_work); |
1754 | #else | |
c5905afb | 1755 | static_key_slow_dec(&netstamp_needed); |
5fa8bbda | 1756 | #endif |
1da177e4 | 1757 | } |
d1b19dff | 1758 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1759 | |
3b098e2d | 1760 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1761 | { |
2456e855 | 1762 | skb->tstamp = 0; |
c5905afb | 1763 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1764 | __net_timestamp(skb); |
1da177e4 LT |
1765 | } |
1766 | ||
588f0330 | 1767 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1768 | if (static_key_false(&netstamp_needed)) { \ |
2456e855 | 1769 | if ((COND) && !(SKB)->tstamp) \ |
588f0330 ED |
1770 | __net_timestamp(SKB); \ |
1771 | } \ | |
3b098e2d | 1772 | |
f4b05d27 | 1773 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 DL |
1774 | { |
1775 | unsigned int len; | |
1776 | ||
1777 | if (!(dev->flags & IFF_UP)) | |
1778 | return false; | |
1779 | ||
1780 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1781 | if (skb->len <= len) | |
1782 | return true; | |
1783 | ||
1784 | /* if TSO is enabled, we don't care about the length as the packet | |
1785 | * could be forwarded without being segmented before | |
1786 | */ | |
1787 | if (skb_is_gso(skb)) | |
1788 | return true; | |
1789 | ||
1790 | return false; | |
1791 | } | |
1ee481fb | 1792 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1793 | |
a0265d28 HX |
1794 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1795 | { | |
4e3264d2 | 1796 | int ret = ____dev_forward_skb(dev, skb); |
a0265d28 | 1797 | |
4e3264d2 MKL |
1798 | if (likely(!ret)) { |
1799 | skb->protocol = eth_type_trans(skb, dev); | |
1800 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
1801 | } | |
a0265d28 | 1802 | |
4e3264d2 | 1803 | return ret; |
a0265d28 HX |
1804 | } |
1805 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1806 | ||
44540960 AB |
1807 | /** |
1808 | * dev_forward_skb - loopback an skb to another netif | |
1809 | * | |
1810 | * @dev: destination network device | |
1811 | * @skb: buffer to forward | |
1812 | * | |
1813 | * return values: | |
1814 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1815 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1816 | * |
1817 | * dev_forward_skb can be used for injecting an skb from the | |
1818 | * start_xmit function of one device into the receive queue | |
1819 | * of another device. | |
1820 | * | |
1821 | * The receiving device may be in another namespace, so | |
1822 | * we have to clear all information in the skb that could | |
1823 | * impact namespace isolation. | |
1824 | */ | |
1825 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1826 | { | |
a0265d28 | 1827 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1828 | } |
1829 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1830 | ||
71d9dec2 CG |
1831 | static inline int deliver_skb(struct sk_buff *skb, |
1832 | struct packet_type *pt_prev, | |
1833 | struct net_device *orig_dev) | |
1834 | { | |
1080e512 MT |
1835 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1836 | return -ENOMEM; | |
71d9dec2 CG |
1837 | atomic_inc(&skb->users); |
1838 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1839 | } | |
1840 | ||
7866a621 SN |
1841 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
1842 | struct packet_type **pt, | |
fbcb2170 JP |
1843 | struct net_device *orig_dev, |
1844 | __be16 type, | |
7866a621 SN |
1845 | struct list_head *ptype_list) |
1846 | { | |
1847 | struct packet_type *ptype, *pt_prev = *pt; | |
1848 | ||
1849 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1850 | if (ptype->type != type) | |
1851 | continue; | |
1852 | if (pt_prev) | |
fbcb2170 | 1853 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
1854 | pt_prev = ptype; |
1855 | } | |
1856 | *pt = pt_prev; | |
1857 | } | |
1858 | ||
c0de08d0 EL |
1859 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1860 | { | |
a3d744e9 | 1861 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1862 | return false; |
1863 | ||
1864 | if (ptype->id_match) | |
1865 | return ptype->id_match(ptype, skb->sk); | |
1866 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1867 | return true; | |
1868 | ||
1869 | return false; | |
1870 | } | |
1871 | ||
1da177e4 LT |
1872 | /* |
1873 | * Support routine. Sends outgoing frames to any network | |
1874 | * taps currently in use. | |
1875 | */ | |
1876 | ||
74b20582 | 1877 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1878 | { |
1879 | struct packet_type *ptype; | |
71d9dec2 CG |
1880 | struct sk_buff *skb2 = NULL; |
1881 | struct packet_type *pt_prev = NULL; | |
7866a621 | 1882 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 1883 | |
1da177e4 | 1884 | rcu_read_lock(); |
7866a621 SN |
1885 | again: |
1886 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1da177e4 LT |
1887 | /* Never send packets back to the socket |
1888 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1889 | */ | |
7866a621 SN |
1890 | if (skb_loop_sk(ptype, skb)) |
1891 | continue; | |
71d9dec2 | 1892 | |
7866a621 SN |
1893 | if (pt_prev) { |
1894 | deliver_skb(skb2, pt_prev, skb->dev); | |
1895 | pt_prev = ptype; | |
1896 | continue; | |
1897 | } | |
1da177e4 | 1898 | |
7866a621 SN |
1899 | /* need to clone skb, done only once */ |
1900 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1901 | if (!skb2) | |
1902 | goto out_unlock; | |
70978182 | 1903 | |
7866a621 | 1904 | net_timestamp_set(skb2); |
1da177e4 | 1905 | |
7866a621 SN |
1906 | /* skb->nh should be correctly |
1907 | * set by sender, so that the second statement is | |
1908 | * just protection against buggy protocols. | |
1909 | */ | |
1910 | skb_reset_mac_header(skb2); | |
1911 | ||
1912 | if (skb_network_header(skb2) < skb2->data || | |
1913 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
1914 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
1915 | ntohs(skb2->protocol), | |
1916 | dev->name); | |
1917 | skb_reset_network_header(skb2); | |
1da177e4 | 1918 | } |
7866a621 SN |
1919 | |
1920 | skb2->transport_header = skb2->network_header; | |
1921 | skb2->pkt_type = PACKET_OUTGOING; | |
1922 | pt_prev = ptype; | |
1923 | } | |
1924 | ||
1925 | if (ptype_list == &ptype_all) { | |
1926 | ptype_list = &dev->ptype_all; | |
1927 | goto again; | |
1da177e4 | 1928 | } |
7866a621 | 1929 | out_unlock: |
71d9dec2 CG |
1930 | if (pt_prev) |
1931 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1932 | rcu_read_unlock(); |
1933 | } | |
74b20582 | 1934 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 1935 | |
2c53040f BH |
1936 | /** |
1937 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1938 | * @dev: Network device |
1939 | * @txq: number of queues available | |
1940 | * | |
1941 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1942 | * valid. To resolve this verify the tc mapping remains valid and if | |
1943 | * not NULL the mapping. With no priorities mapping to this | |
1944 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1945 | * is invalid nothing can be done so disable priority mappings. If is | |
1946 | * expected that drivers will fix this mapping if they can before | |
1947 | * calling netif_set_real_num_tx_queues. | |
1948 | */ | |
bb134d22 | 1949 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1950 | { |
1951 | int i; | |
1952 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1953 | ||
1954 | /* If TC0 is invalidated disable TC mapping */ | |
1955 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1956 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1957 | dev->num_tc = 0; |
1958 | return; | |
1959 | } | |
1960 | ||
1961 | /* Invalidated prio to tc mappings set to TC0 */ | |
1962 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1963 | int q = netdev_get_prio_tc_map(dev, i); | |
1964 | ||
1965 | tc = &dev->tc_to_txq[q]; | |
1966 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1967 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1968 | i, q); | |
4f57c087 JF |
1969 | netdev_set_prio_tc_map(dev, i, 0); |
1970 | } | |
1971 | } | |
1972 | } | |
1973 | ||
8d059b0f AD |
1974 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
1975 | { | |
1976 | if (dev->num_tc) { | |
1977 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1978 | int i; | |
1979 | ||
1980 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { | |
1981 | if ((txq - tc->offset) < tc->count) | |
1982 | return i; | |
1983 | } | |
1984 | ||
1985 | return -1; | |
1986 | } | |
1987 | ||
1988 | return 0; | |
1989 | } | |
1990 | ||
537c00de AD |
1991 | #ifdef CONFIG_XPS |
1992 | static DEFINE_MUTEX(xps_map_mutex); | |
1993 | #define xmap_dereference(P) \ | |
1994 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1995 | ||
6234f874 AD |
1996 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
1997 | int tci, u16 index) | |
537c00de | 1998 | { |
10cdc3f3 AD |
1999 | struct xps_map *map = NULL; |
2000 | int pos; | |
537c00de | 2001 | |
10cdc3f3 | 2002 | if (dev_maps) |
6234f874 AD |
2003 | map = xmap_dereference(dev_maps->cpu_map[tci]); |
2004 | if (!map) | |
2005 | return false; | |
537c00de | 2006 | |
6234f874 AD |
2007 | for (pos = map->len; pos--;) { |
2008 | if (map->queues[pos] != index) | |
2009 | continue; | |
2010 | ||
2011 | if (map->len > 1) { | |
2012 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2013 | break; |
537c00de | 2014 | } |
6234f874 AD |
2015 | |
2016 | RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL); | |
2017 | kfree_rcu(map, rcu); | |
2018 | return false; | |
537c00de AD |
2019 | } |
2020 | ||
6234f874 | 2021 | return true; |
10cdc3f3 AD |
2022 | } |
2023 | ||
6234f874 AD |
2024 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2025 | struct xps_dev_maps *dev_maps, | |
2026 | int cpu, u16 offset, u16 count) | |
2027 | { | |
184c449f AD |
2028 | int num_tc = dev->num_tc ? : 1; |
2029 | bool active = false; | |
2030 | int tci; | |
6234f874 | 2031 | |
184c449f AD |
2032 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2033 | int i, j; | |
2034 | ||
2035 | for (i = count, j = offset; i--; j++) { | |
2036 | if (!remove_xps_queue(dev_maps, cpu, j)) | |
2037 | break; | |
2038 | } | |
2039 | ||
2040 | active |= i < 0; | |
6234f874 AD |
2041 | } |
2042 | ||
184c449f | 2043 | return active; |
6234f874 AD |
2044 | } |
2045 | ||
2046 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, | |
2047 | u16 count) | |
10cdc3f3 AD |
2048 | { |
2049 | struct xps_dev_maps *dev_maps; | |
024e9679 | 2050 | int cpu, i; |
10cdc3f3 AD |
2051 | bool active = false; |
2052 | ||
2053 | mutex_lock(&xps_map_mutex); | |
2054 | dev_maps = xmap_dereference(dev->xps_maps); | |
2055 | ||
2056 | if (!dev_maps) | |
2057 | goto out_no_maps; | |
2058 | ||
6234f874 AD |
2059 | for_each_possible_cpu(cpu) |
2060 | active |= remove_xps_queue_cpu(dev, dev_maps, cpu, | |
2061 | offset, count); | |
10cdc3f3 AD |
2062 | |
2063 | if (!active) { | |
537c00de AD |
2064 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
2065 | kfree_rcu(dev_maps, rcu); | |
2066 | } | |
2067 | ||
6234f874 | 2068 | for (i = offset + (count - 1); count--; i--) |
024e9679 AD |
2069 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), |
2070 | NUMA_NO_NODE); | |
2071 | ||
537c00de AD |
2072 | out_no_maps: |
2073 | mutex_unlock(&xps_map_mutex); | |
2074 | } | |
2075 | ||
6234f874 AD |
2076 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2077 | { | |
2078 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2079 | } | |
2080 | ||
01c5f864 AD |
2081 | static struct xps_map *expand_xps_map(struct xps_map *map, |
2082 | int cpu, u16 index) | |
2083 | { | |
2084 | struct xps_map *new_map; | |
2085 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2086 | int i, pos; | |
2087 | ||
2088 | for (pos = 0; map && pos < map->len; pos++) { | |
2089 | if (map->queues[pos] != index) | |
2090 | continue; | |
2091 | return map; | |
2092 | } | |
2093 | ||
2094 | /* Need to add queue to this CPU's existing map */ | |
2095 | if (map) { | |
2096 | if (pos < map->alloc_len) | |
2097 | return map; | |
2098 | ||
2099 | alloc_len = map->alloc_len * 2; | |
2100 | } | |
2101 | ||
2102 | /* Need to allocate new map to store queue on this CPU's map */ | |
2103 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2104 | cpu_to_node(cpu)); | |
2105 | if (!new_map) | |
2106 | return NULL; | |
2107 | ||
2108 | for (i = 0; i < pos; i++) | |
2109 | new_map->queues[i] = map->queues[i]; | |
2110 | new_map->alloc_len = alloc_len; | |
2111 | new_map->len = pos; | |
2112 | ||
2113 | return new_map; | |
2114 | } | |
2115 | ||
3573540c MT |
2116 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
2117 | u16 index) | |
537c00de | 2118 | { |
01c5f864 | 2119 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
184c449f AD |
2120 | int i, cpu, tci, numa_node_id = -2; |
2121 | int maps_sz, num_tc = 1, tc = 0; | |
537c00de | 2122 | struct xps_map *map, *new_map; |
01c5f864 | 2123 | bool active = false; |
537c00de | 2124 | |
184c449f AD |
2125 | if (dev->num_tc) { |
2126 | num_tc = dev->num_tc; | |
2127 | tc = netdev_txq_to_tc(dev, index); | |
2128 | if (tc < 0) | |
2129 | return -EINVAL; | |
2130 | } | |
2131 | ||
2132 | maps_sz = XPS_DEV_MAPS_SIZE(num_tc); | |
2133 | if (maps_sz < L1_CACHE_BYTES) | |
2134 | maps_sz = L1_CACHE_BYTES; | |
2135 | ||
537c00de AD |
2136 | mutex_lock(&xps_map_mutex); |
2137 | ||
2138 | dev_maps = xmap_dereference(dev->xps_maps); | |
2139 | ||
01c5f864 | 2140 | /* allocate memory for queue storage */ |
184c449f | 2141 | for_each_cpu_and(cpu, cpu_online_mask, mask) { |
01c5f864 AD |
2142 | if (!new_dev_maps) |
2143 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2144 | if (!new_dev_maps) { |
2145 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2146 | return -ENOMEM; |
2bb60cb9 | 2147 | } |
01c5f864 | 2148 | |
184c449f AD |
2149 | tci = cpu * num_tc + tc; |
2150 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) : | |
01c5f864 AD |
2151 | NULL; |
2152 | ||
2153 | map = expand_xps_map(map, cpu, index); | |
2154 | if (!map) | |
2155 | goto error; | |
2156 | ||
184c449f | 2157 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); |
01c5f864 AD |
2158 | } |
2159 | ||
2160 | if (!new_dev_maps) | |
2161 | goto out_no_new_maps; | |
2162 | ||
537c00de | 2163 | for_each_possible_cpu(cpu) { |
184c449f AD |
2164 | /* copy maps belonging to foreign traffic classes */ |
2165 | for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) { | |
2166 | /* fill in the new device map from the old device map */ | |
2167 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
2168 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
2169 | } | |
2170 | ||
2171 | /* We need to explicitly update tci as prevous loop | |
2172 | * could break out early if dev_maps is NULL. | |
2173 | */ | |
2174 | tci = cpu * num_tc + tc; | |
2175 | ||
01c5f864 AD |
2176 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
2177 | /* add queue to CPU maps */ | |
2178 | int pos = 0; | |
2179 | ||
184c449f | 2180 | map = xmap_dereference(new_dev_maps->cpu_map[tci]); |
01c5f864 AD |
2181 | while ((pos < map->len) && (map->queues[pos] != index)) |
2182 | pos++; | |
2183 | ||
2184 | if (pos == map->len) | |
2185 | map->queues[map->len++] = index; | |
537c00de | 2186 | #ifdef CONFIG_NUMA |
537c00de AD |
2187 | if (numa_node_id == -2) |
2188 | numa_node_id = cpu_to_node(cpu); | |
2189 | else if (numa_node_id != cpu_to_node(cpu)) | |
2190 | numa_node_id = -1; | |
537c00de | 2191 | #endif |
01c5f864 AD |
2192 | } else if (dev_maps) { |
2193 | /* fill in the new device map from the old device map */ | |
184c449f AD |
2194 | map = xmap_dereference(dev_maps->cpu_map[tci]); |
2195 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
537c00de | 2196 | } |
01c5f864 | 2197 | |
184c449f AD |
2198 | /* copy maps belonging to foreign traffic classes */ |
2199 | for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { | |
2200 | /* fill in the new device map from the old device map */ | |
2201 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
2202 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
2203 | } | |
537c00de AD |
2204 | } |
2205 | ||
01c5f864 AD |
2206 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2207 | ||
537c00de | 2208 | /* Cleanup old maps */ |
184c449f AD |
2209 | if (!dev_maps) |
2210 | goto out_no_old_maps; | |
2211 | ||
2212 | for_each_possible_cpu(cpu) { | |
2213 | for (i = num_tc, tci = cpu * num_tc; i--; tci++) { | |
2214 | new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); | |
2215 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
01c5f864 AD |
2216 | if (map && map != new_map) |
2217 | kfree_rcu(map, rcu); | |
2218 | } | |
537c00de AD |
2219 | } |
2220 | ||
184c449f AD |
2221 | kfree_rcu(dev_maps, rcu); |
2222 | ||
2223 | out_no_old_maps: | |
01c5f864 AD |
2224 | dev_maps = new_dev_maps; |
2225 | active = true; | |
537c00de | 2226 | |
01c5f864 AD |
2227 | out_no_new_maps: |
2228 | /* update Tx queue numa node */ | |
537c00de AD |
2229 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2230 | (numa_node_id >= 0) ? numa_node_id : | |
2231 | NUMA_NO_NODE); | |
2232 | ||
01c5f864 AD |
2233 | if (!dev_maps) |
2234 | goto out_no_maps; | |
2235 | ||
2236 | /* removes queue from unused CPUs */ | |
2237 | for_each_possible_cpu(cpu) { | |
184c449f AD |
2238 | for (i = tc, tci = cpu * num_tc; i--; tci++) |
2239 | active |= remove_xps_queue(dev_maps, tci, index); | |
2240 | if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu)) | |
2241 | active |= remove_xps_queue(dev_maps, tci, index); | |
2242 | for (i = num_tc - tc, tci++; --i; tci++) | |
2243 | active |= remove_xps_queue(dev_maps, tci, index); | |
01c5f864 AD |
2244 | } |
2245 | ||
2246 | /* free map if not active */ | |
2247 | if (!active) { | |
2248 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2249 | kfree_rcu(dev_maps, rcu); | |
2250 | } | |
2251 | ||
2252 | out_no_maps: | |
537c00de AD |
2253 | mutex_unlock(&xps_map_mutex); |
2254 | ||
2255 | return 0; | |
2256 | error: | |
01c5f864 AD |
2257 | /* remove any maps that we added */ |
2258 | for_each_possible_cpu(cpu) { | |
184c449f AD |
2259 | for (i = num_tc, tci = cpu * num_tc; i--; tci++) { |
2260 | new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); | |
2261 | map = dev_maps ? | |
2262 | xmap_dereference(dev_maps->cpu_map[tci]) : | |
2263 | NULL; | |
2264 | if (new_map && new_map != map) | |
2265 | kfree(new_map); | |
2266 | } | |
01c5f864 AD |
2267 | } |
2268 | ||
537c00de AD |
2269 | mutex_unlock(&xps_map_mutex); |
2270 | ||
537c00de AD |
2271 | kfree(new_dev_maps); |
2272 | return -ENOMEM; | |
2273 | } | |
2274 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2275 | ||
2276 | #endif | |
9cf1f6a8 AD |
2277 | void netdev_reset_tc(struct net_device *dev) |
2278 | { | |
6234f874 AD |
2279 | #ifdef CONFIG_XPS |
2280 | netif_reset_xps_queues_gt(dev, 0); | |
2281 | #endif | |
9cf1f6a8 AD |
2282 | dev->num_tc = 0; |
2283 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2284 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2285 | } | |
2286 | EXPORT_SYMBOL(netdev_reset_tc); | |
2287 | ||
2288 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2289 | { | |
2290 | if (tc >= dev->num_tc) | |
2291 | return -EINVAL; | |
2292 | ||
6234f874 AD |
2293 | #ifdef CONFIG_XPS |
2294 | netif_reset_xps_queues(dev, offset, count); | |
2295 | #endif | |
9cf1f6a8 AD |
2296 | dev->tc_to_txq[tc].count = count; |
2297 | dev->tc_to_txq[tc].offset = offset; | |
2298 | return 0; | |
2299 | } | |
2300 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2301 | ||
2302 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2303 | { | |
2304 | if (num_tc > TC_MAX_QUEUE) | |
2305 | return -EINVAL; | |
2306 | ||
6234f874 AD |
2307 | #ifdef CONFIG_XPS |
2308 | netif_reset_xps_queues_gt(dev, 0); | |
2309 | #endif | |
9cf1f6a8 AD |
2310 | dev->num_tc = num_tc; |
2311 | return 0; | |
2312 | } | |
2313 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2314 | ||
f0796d5c JF |
2315 | /* |
2316 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2317 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2318 | */ | |
e6484930 | 2319 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2320 | { |
1d24eb48 TH |
2321 | int rc; |
2322 | ||
e6484930 TH |
2323 | if (txq < 1 || txq > dev->num_tx_queues) |
2324 | return -EINVAL; | |
f0796d5c | 2325 | |
5c56580b BH |
2326 | if (dev->reg_state == NETREG_REGISTERED || |
2327 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2328 | ASSERT_RTNL(); |
2329 | ||
1d24eb48 TH |
2330 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2331 | txq); | |
bf264145 TH |
2332 | if (rc) |
2333 | return rc; | |
2334 | ||
4f57c087 JF |
2335 | if (dev->num_tc) |
2336 | netif_setup_tc(dev, txq); | |
2337 | ||
024e9679 | 2338 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2339 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2340 | #ifdef CONFIG_XPS |
2341 | netif_reset_xps_queues_gt(dev, txq); | |
2342 | #endif | |
2343 | } | |
f0796d5c | 2344 | } |
e6484930 TH |
2345 | |
2346 | dev->real_num_tx_queues = txq; | |
2347 | return 0; | |
f0796d5c JF |
2348 | } |
2349 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2350 | |
a953be53 | 2351 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2352 | /** |
2353 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2354 | * @dev: Network device | |
2355 | * @rxq: Actual number of RX queues | |
2356 | * | |
2357 | * This must be called either with the rtnl_lock held or before | |
2358 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2359 | * negative error code. If called before registration, it always |
2360 | * succeeds. | |
62fe0b40 BH |
2361 | */ |
2362 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2363 | { | |
2364 | int rc; | |
2365 | ||
bd25fa7b TH |
2366 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2367 | return -EINVAL; | |
2368 | ||
62fe0b40 BH |
2369 | if (dev->reg_state == NETREG_REGISTERED) { |
2370 | ASSERT_RTNL(); | |
2371 | ||
62fe0b40 BH |
2372 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2373 | rxq); | |
2374 | if (rc) | |
2375 | return rc; | |
62fe0b40 BH |
2376 | } |
2377 | ||
2378 | dev->real_num_rx_queues = rxq; | |
2379 | return 0; | |
2380 | } | |
2381 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2382 | #endif | |
2383 | ||
2c53040f BH |
2384 | /** |
2385 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2386 | * |
2387 | * This routine should set an upper limit on the number of RSS queues | |
2388 | * used by default by multiqueue devices. | |
2389 | */ | |
a55b138b | 2390 | int netif_get_num_default_rss_queues(void) |
16917b87 | 2391 | { |
40e4e713 HS |
2392 | return is_kdump_kernel() ? |
2393 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
2394 | } |
2395 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2396 | ||
3bcb846c | 2397 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 2398 | { |
def82a1d JP |
2399 | struct softnet_data *sd; |
2400 | unsigned long flags; | |
56079431 | 2401 | |
def82a1d | 2402 | local_irq_save(flags); |
903ceff7 | 2403 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2404 | q->next_sched = NULL; |
2405 | *sd->output_queue_tailp = q; | |
2406 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2407 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2408 | local_irq_restore(flags); | |
2409 | } | |
2410 | ||
2411 | void __netif_schedule(struct Qdisc *q) | |
2412 | { | |
2413 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2414 | __netif_reschedule(q); | |
56079431 DV |
2415 | } |
2416 | EXPORT_SYMBOL(__netif_schedule); | |
2417 | ||
e6247027 ED |
2418 | struct dev_kfree_skb_cb { |
2419 | enum skb_free_reason reason; | |
2420 | }; | |
2421 | ||
2422 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2423 | { |
e6247027 ED |
2424 | return (struct dev_kfree_skb_cb *)skb->cb; |
2425 | } | |
2426 | ||
46e5da40 JF |
2427 | void netif_schedule_queue(struct netdev_queue *txq) |
2428 | { | |
2429 | rcu_read_lock(); | |
2430 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2431 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2432 | ||
2433 | __netif_schedule(q); | |
2434 | } | |
2435 | rcu_read_unlock(); | |
2436 | } | |
2437 | EXPORT_SYMBOL(netif_schedule_queue); | |
2438 | ||
46e5da40 JF |
2439 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
2440 | { | |
2441 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2442 | struct Qdisc *q; | |
2443 | ||
2444 | rcu_read_lock(); | |
2445 | q = rcu_dereference(dev_queue->qdisc); | |
2446 | __netif_schedule(q); | |
2447 | rcu_read_unlock(); | |
2448 | } | |
2449 | } | |
2450 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2451 | ||
e6247027 | 2452 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2453 | { |
e6247027 | 2454 | unsigned long flags; |
56079431 | 2455 | |
9899886d MJ |
2456 | if (unlikely(!skb)) |
2457 | return; | |
2458 | ||
e6247027 ED |
2459 | if (likely(atomic_read(&skb->users) == 1)) { |
2460 | smp_rmb(); | |
2461 | atomic_set(&skb->users, 0); | |
2462 | } else if (likely(!atomic_dec_and_test(&skb->users))) { | |
2463 | return; | |
bea3348e | 2464 | } |
e6247027 ED |
2465 | get_kfree_skb_cb(skb)->reason = reason; |
2466 | local_irq_save(flags); | |
2467 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2468 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2469 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2470 | local_irq_restore(flags); | |
56079431 | 2471 | } |
e6247027 | 2472 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2473 | |
e6247027 | 2474 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2475 | { |
2476 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2477 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2478 | else |
2479 | dev_kfree_skb(skb); | |
2480 | } | |
e6247027 | 2481 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2482 | |
2483 | ||
bea3348e SH |
2484 | /** |
2485 | * netif_device_detach - mark device as removed | |
2486 | * @dev: network device | |
2487 | * | |
2488 | * Mark device as removed from system and therefore no longer available. | |
2489 | */ | |
56079431 DV |
2490 | void netif_device_detach(struct net_device *dev) |
2491 | { | |
2492 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2493 | netif_running(dev)) { | |
d543103a | 2494 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2495 | } |
2496 | } | |
2497 | EXPORT_SYMBOL(netif_device_detach); | |
2498 | ||
bea3348e SH |
2499 | /** |
2500 | * netif_device_attach - mark device as attached | |
2501 | * @dev: network device | |
2502 | * | |
2503 | * Mark device as attached from system and restart if needed. | |
2504 | */ | |
56079431 DV |
2505 | void netif_device_attach(struct net_device *dev) |
2506 | { | |
2507 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2508 | netif_running(dev)) { | |
d543103a | 2509 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2510 | __netdev_watchdog_up(dev); |
56079431 DV |
2511 | } |
2512 | } | |
2513 | EXPORT_SYMBOL(netif_device_attach); | |
2514 | ||
5605c762 JP |
2515 | /* |
2516 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2517 | * to be used as a distribution range. | |
2518 | */ | |
2519 | u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, | |
2520 | unsigned int num_tx_queues) | |
2521 | { | |
2522 | u32 hash; | |
2523 | u16 qoffset = 0; | |
2524 | u16 qcount = num_tx_queues; | |
2525 | ||
2526 | if (skb_rx_queue_recorded(skb)) { | |
2527 | hash = skb_get_rx_queue(skb); | |
2528 | while (unlikely(hash >= num_tx_queues)) | |
2529 | hash -= num_tx_queues; | |
2530 | return hash; | |
2531 | } | |
2532 | ||
2533 | if (dev->num_tc) { | |
2534 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
f4563a75 | 2535 | |
5605c762 JP |
2536 | qoffset = dev->tc_to_txq[tc].offset; |
2537 | qcount = dev->tc_to_txq[tc].count; | |
2538 | } | |
2539 | ||
2540 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
2541 | } | |
2542 | EXPORT_SYMBOL(__skb_tx_hash); | |
2543 | ||
36c92474 BH |
2544 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2545 | { | |
84d15ae5 | 2546 | static const netdev_features_t null_features; |
36c92474 | 2547 | struct net_device *dev = skb->dev; |
88ad4175 | 2548 | const char *name = ""; |
36c92474 | 2549 | |
c846ad9b BG |
2550 | if (!net_ratelimit()) |
2551 | return; | |
2552 | ||
88ad4175 BM |
2553 | if (dev) { |
2554 | if (dev->dev.parent) | |
2555 | name = dev_driver_string(dev->dev.parent); | |
2556 | else | |
2557 | name = netdev_name(dev); | |
2558 | } | |
36c92474 BH |
2559 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " |
2560 | "gso_type=%d ip_summed=%d\n", | |
88ad4175 | 2561 | name, dev ? &dev->features : &null_features, |
65e9d2fa | 2562 | skb->sk ? &skb->sk->sk_route_caps : &null_features, |
36c92474 BH |
2563 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2564 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2565 | } | |
2566 | ||
1da177e4 LT |
2567 | /* |
2568 | * Invalidate hardware checksum when packet is to be mangled, and | |
2569 | * complete checksum manually on outgoing path. | |
2570 | */ | |
84fa7933 | 2571 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2572 | { |
d3bc23e7 | 2573 | __wsum csum; |
663ead3b | 2574 | int ret = 0, offset; |
1da177e4 | 2575 | |
84fa7933 | 2576 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2577 | goto out_set_summed; |
2578 | ||
2579 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2580 | skb_warn_bad_offload(skb); |
2581 | return -EINVAL; | |
1da177e4 LT |
2582 | } |
2583 | ||
cef401de ED |
2584 | /* Before computing a checksum, we should make sure no frag could |
2585 | * be modified by an external entity : checksum could be wrong. | |
2586 | */ | |
2587 | if (skb_has_shared_frag(skb)) { | |
2588 | ret = __skb_linearize(skb); | |
2589 | if (ret) | |
2590 | goto out; | |
2591 | } | |
2592 | ||
55508d60 | 2593 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2594 | BUG_ON(offset >= skb_headlen(skb)); |
2595 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2596 | ||
2597 | offset += skb->csum_offset; | |
2598 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2599 | ||
2600 | if (skb_cloned(skb) && | |
2601 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2602 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2603 | if (ret) | |
2604 | goto out; | |
2605 | } | |
2606 | ||
4f2e4ad5 | 2607 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 2608 | out_set_summed: |
1da177e4 | 2609 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2610 | out: |
1da177e4 LT |
2611 | return ret; |
2612 | } | |
d1b19dff | 2613 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2614 | |
53d6471c | 2615 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2616 | { |
252e3346 | 2617 | __be16 type = skb->protocol; |
f6a78bfc | 2618 | |
19acc327 PS |
2619 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2620 | if (type == htons(ETH_P_TEB)) { | |
2621 | struct ethhdr *eth; | |
2622 | ||
2623 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2624 | return 0; | |
2625 | ||
2626 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2627 | type = eth->h_proto; | |
2628 | } | |
2629 | ||
d4bcef3f | 2630 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
2631 | } |
2632 | ||
2633 | /** | |
2634 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2635 | * @skb: buffer to segment | |
2636 | * @features: features for the output path (see dev->features) | |
2637 | */ | |
2638 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2639 | netdev_features_t features) | |
2640 | { | |
2641 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2642 | struct packet_offload *ptype; | |
53d6471c VY |
2643 | int vlan_depth = skb->mac_len; |
2644 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2645 | |
2646 | if (unlikely(!type)) | |
2647 | return ERR_PTR(-EINVAL); | |
2648 | ||
53d6471c | 2649 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2650 | |
2651 | rcu_read_lock(); | |
22061d80 | 2652 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2653 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2654 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2655 | break; |
2656 | } | |
2657 | } | |
2658 | rcu_read_unlock(); | |
2659 | ||
98e399f8 | 2660 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2661 | |
f6a78bfc HX |
2662 | return segs; |
2663 | } | |
05e8ef4a PS |
2664 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2665 | ||
2666 | ||
2667 | /* openvswitch calls this on rx path, so we need a different check. | |
2668 | */ | |
2669 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2670 | { | |
2671 | if (tx_path) | |
6e7bc478 ED |
2672 | return skb->ip_summed != CHECKSUM_PARTIAL && |
2673 | skb->ip_summed != CHECKSUM_NONE; | |
2674 | ||
2675 | return skb->ip_summed == CHECKSUM_NONE; | |
05e8ef4a PS |
2676 | } |
2677 | ||
2678 | /** | |
2679 | * __skb_gso_segment - Perform segmentation on skb. | |
2680 | * @skb: buffer to segment | |
2681 | * @features: features for the output path (see dev->features) | |
2682 | * @tx_path: whether it is called in TX path | |
2683 | * | |
2684 | * This function segments the given skb and returns a list of segments. | |
2685 | * | |
2686 | * It may return NULL if the skb requires no segmentation. This is | |
2687 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 KK |
2688 | * |
2689 | * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. | |
05e8ef4a PS |
2690 | */ |
2691 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2692 | netdev_features_t features, bool tx_path) | |
2693 | { | |
b2504a5d ED |
2694 | struct sk_buff *segs; |
2695 | ||
05e8ef4a PS |
2696 | if (unlikely(skb_needs_check(skb, tx_path))) { |
2697 | int err; | |
2698 | ||
b2504a5d | 2699 | /* We're going to init ->check field in TCP or UDP header */ |
a40e0a66 | 2700 | err = skb_cow_head(skb, 0); |
2701 | if (err < 0) | |
05e8ef4a PS |
2702 | return ERR_PTR(err); |
2703 | } | |
2704 | ||
802ab55a AD |
2705 | /* Only report GSO partial support if it will enable us to |
2706 | * support segmentation on this frame without needing additional | |
2707 | * work. | |
2708 | */ | |
2709 | if (features & NETIF_F_GSO_PARTIAL) { | |
2710 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
2711 | struct net_device *dev = skb->dev; | |
2712 | ||
2713 | partial_features |= dev->features & dev->gso_partial_features; | |
2714 | if (!skb_gso_ok(skb, features | partial_features)) | |
2715 | features &= ~NETIF_F_GSO_PARTIAL; | |
2716 | } | |
2717 | ||
9207f9d4 KK |
2718 | BUILD_BUG_ON(SKB_SGO_CB_OFFSET + |
2719 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); | |
2720 | ||
68c33163 | 2721 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2722 | SKB_GSO_CB(skb)->encap_level = 0; |
2723 | ||
05e8ef4a PS |
2724 | skb_reset_mac_header(skb); |
2725 | skb_reset_mac_len(skb); | |
2726 | ||
b2504a5d ED |
2727 | segs = skb_mac_gso_segment(skb, features); |
2728 | ||
2729 | if (unlikely(skb_needs_check(skb, tx_path))) | |
2730 | skb_warn_bad_offload(skb); | |
2731 | ||
2732 | return segs; | |
05e8ef4a | 2733 | } |
12b0004d | 2734 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2735 | |
fb286bb2 HX |
2736 | /* Take action when hardware reception checksum errors are detected. */ |
2737 | #ifdef CONFIG_BUG | |
2738 | void netdev_rx_csum_fault(struct net_device *dev) | |
2739 | { | |
2740 | if (net_ratelimit()) { | |
7b6cd1ce | 2741 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2742 | dump_stack(); |
2743 | } | |
2744 | } | |
2745 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2746 | #endif | |
2747 | ||
1da177e4 LT |
2748 | /* Actually, we should eliminate this check as soon as we know, that: |
2749 | * 1. IOMMU is present and allows to map all the memory. | |
2750 | * 2. No high memory really exists on this machine. | |
2751 | */ | |
2752 | ||
c1e756bf | 2753 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2754 | { |
3d3a8533 | 2755 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2756 | int i; |
f4563a75 | 2757 | |
5acbbd42 | 2758 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2759 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2760 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 2761 | |
ea2ab693 | 2762 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 2763 | return 1; |
ea2ab693 | 2764 | } |
5acbbd42 | 2765 | } |
1da177e4 | 2766 | |
5acbbd42 FT |
2767 | if (PCI_DMA_BUS_IS_PHYS) { |
2768 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2769 | |
9092c658 ED |
2770 | if (!pdev) |
2771 | return 0; | |
5acbbd42 | 2772 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2773 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2774 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
f4563a75 | 2775 | |
5acbbd42 FT |
2776 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2777 | return 1; | |
2778 | } | |
2779 | } | |
3d3a8533 | 2780 | #endif |
1da177e4 LT |
2781 | return 0; |
2782 | } | |
1da177e4 | 2783 | |
3b392ddb SH |
2784 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2785 | * instead of standard features for the netdev. | |
2786 | */ | |
d0edc7bf | 2787 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
2788 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
2789 | netdev_features_t features, | |
2790 | __be16 type) | |
2791 | { | |
25cd9ba0 | 2792 | if (eth_p_mpls(type)) |
3b392ddb SH |
2793 | features &= skb->dev->mpls_features; |
2794 | ||
2795 | return features; | |
2796 | } | |
2797 | #else | |
2798 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2799 | netdev_features_t features, | |
2800 | __be16 type) | |
2801 | { | |
2802 | return features; | |
2803 | } | |
2804 | #endif | |
2805 | ||
c8f44aff | 2806 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2807 | netdev_features_t features) |
f01a5236 | 2808 | { |
53d6471c | 2809 | int tmp; |
3b392ddb SH |
2810 | __be16 type; |
2811 | ||
2812 | type = skb_network_protocol(skb, &tmp); | |
2813 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2814 | |
c0d680e5 | 2815 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2816 | !can_checksum_protocol(features, type)) { |
996e8021 | 2817 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 2818 | } |
7be2c82c ED |
2819 | if (illegal_highdma(skb->dev, skb)) |
2820 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
2821 | |
2822 | return features; | |
2823 | } | |
2824 | ||
e38f3025 TM |
2825 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
2826 | struct net_device *dev, | |
2827 | netdev_features_t features) | |
2828 | { | |
2829 | return features; | |
2830 | } | |
2831 | EXPORT_SYMBOL(passthru_features_check); | |
2832 | ||
8cb65d00 TM |
2833 | static netdev_features_t dflt_features_check(const struct sk_buff *skb, |
2834 | struct net_device *dev, | |
2835 | netdev_features_t features) | |
2836 | { | |
2837 | return vlan_features_check(skb, features); | |
2838 | } | |
2839 | ||
cbc53e08 AD |
2840 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
2841 | struct net_device *dev, | |
2842 | netdev_features_t features) | |
2843 | { | |
2844 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
2845 | ||
2846 | if (gso_segs > dev->gso_max_segs) | |
2847 | return features & ~NETIF_F_GSO_MASK; | |
2848 | ||
802ab55a AD |
2849 | /* Support for GSO partial features requires software |
2850 | * intervention before we can actually process the packets | |
2851 | * so we need to strip support for any partial features now | |
2852 | * and we can pull them back in after we have partially | |
2853 | * segmented the frame. | |
2854 | */ | |
2855 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
2856 | features &= ~dev->gso_partial_features; | |
2857 | ||
2858 | /* Make sure to clear the IPv4 ID mangling feature if the | |
2859 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
2860 | */ |
2861 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
2862 | struct iphdr *iph = skb->encapsulation ? | |
2863 | inner_ip_hdr(skb) : ip_hdr(skb); | |
2864 | ||
2865 | if (!(iph->frag_off & htons(IP_DF))) | |
2866 | features &= ~NETIF_F_TSO_MANGLEID; | |
2867 | } | |
2868 | ||
2869 | return features; | |
2870 | } | |
2871 | ||
c1e756bf | 2872 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 2873 | { |
5f35227e | 2874 | struct net_device *dev = skb->dev; |
fcbeb976 | 2875 | netdev_features_t features = dev->features; |
58e998c6 | 2876 | |
cbc53e08 AD |
2877 | if (skb_is_gso(skb)) |
2878 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 2879 | |
5f35227e JG |
2880 | /* If encapsulation offload request, verify we are testing |
2881 | * hardware encapsulation features instead of standard | |
2882 | * features for the netdev | |
2883 | */ | |
2884 | if (skb->encapsulation) | |
2885 | features &= dev->hw_enc_features; | |
2886 | ||
f5a7fb88 TM |
2887 | if (skb_vlan_tagged(skb)) |
2888 | features = netdev_intersect_features(features, | |
2889 | dev->vlan_features | | |
2890 | NETIF_F_HW_VLAN_CTAG_TX | | |
2891 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2892 | |
5f35227e JG |
2893 | if (dev->netdev_ops->ndo_features_check) |
2894 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
2895 | features); | |
8cb65d00 TM |
2896 | else |
2897 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 2898 | |
c1e756bf | 2899 | return harmonize_features(skb, features); |
58e998c6 | 2900 | } |
c1e756bf | 2901 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2902 | |
2ea25513 | 2903 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2904 | struct netdev_queue *txq, bool more) |
f6a78bfc | 2905 | { |
2ea25513 DM |
2906 | unsigned int len; |
2907 | int rc; | |
00829823 | 2908 | |
7866a621 | 2909 | if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) |
2ea25513 | 2910 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 2911 | |
2ea25513 DM |
2912 | len = skb->len; |
2913 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 2914 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 2915 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 2916 | |
2ea25513 DM |
2917 | return rc; |
2918 | } | |
7b9c6090 | 2919 | |
8dcda22a DM |
2920 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
2921 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
2922 | { |
2923 | struct sk_buff *skb = first; | |
2924 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 2925 | |
7f2e870f DM |
2926 | while (skb) { |
2927 | struct sk_buff *next = skb->next; | |
fc70fb64 | 2928 | |
7f2e870f | 2929 | skb->next = NULL; |
95f6b3dd | 2930 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
2931 | if (unlikely(!dev_xmit_complete(rc))) { |
2932 | skb->next = next; | |
2933 | goto out; | |
2934 | } | |
6afff0ca | 2935 | |
7f2e870f DM |
2936 | skb = next; |
2937 | if (netif_xmit_stopped(txq) && skb) { | |
2938 | rc = NETDEV_TX_BUSY; | |
2939 | break; | |
9ccb8975 | 2940 | } |
7f2e870f | 2941 | } |
9ccb8975 | 2942 | |
7f2e870f DM |
2943 | out: |
2944 | *ret = rc; | |
2945 | return skb; | |
2946 | } | |
b40863c6 | 2947 | |
1ff0dc94 ED |
2948 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
2949 | netdev_features_t features) | |
f6a78bfc | 2950 | { |
df8a39de | 2951 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
2952 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
2953 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
2954 | return skb; |
2955 | } | |
f6a78bfc | 2956 | |
55a93b3e | 2957 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
2958 | { |
2959 | netdev_features_t features; | |
f6a78bfc | 2960 | |
eae3f88e DM |
2961 | features = netif_skb_features(skb); |
2962 | skb = validate_xmit_vlan(skb, features); | |
2963 | if (unlikely(!skb)) | |
2964 | goto out_null; | |
7b9c6090 | 2965 | |
8b86a61d | 2966 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
2967 | struct sk_buff *segs; |
2968 | ||
2969 | segs = skb_gso_segment(skb, features); | |
cecda693 | 2970 | if (IS_ERR(segs)) { |
af6dabc9 | 2971 | goto out_kfree_skb; |
cecda693 JW |
2972 | } else if (segs) { |
2973 | consume_skb(skb); | |
2974 | skb = segs; | |
f6a78bfc | 2975 | } |
eae3f88e DM |
2976 | } else { |
2977 | if (skb_needs_linearize(skb, features) && | |
2978 | __skb_linearize(skb)) | |
2979 | goto out_kfree_skb; | |
4ec93edb | 2980 | |
f6e27114 SK |
2981 | if (validate_xmit_xfrm(skb, features)) |
2982 | goto out_kfree_skb; | |
2983 | ||
eae3f88e DM |
2984 | /* If packet is not checksummed and device does not |
2985 | * support checksumming for this protocol, complete | |
2986 | * checksumming here. | |
2987 | */ | |
2988 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
2989 | if (skb->encapsulation) | |
2990 | skb_set_inner_transport_header(skb, | |
2991 | skb_checksum_start_offset(skb)); | |
2992 | else | |
2993 | skb_set_transport_header(skb, | |
2994 | skb_checksum_start_offset(skb)); | |
a188222b | 2995 | if (!(features & NETIF_F_CSUM_MASK) && |
eae3f88e DM |
2996 | skb_checksum_help(skb)) |
2997 | goto out_kfree_skb; | |
7b9c6090 | 2998 | } |
0c772159 | 2999 | } |
7b9c6090 | 3000 | |
eae3f88e | 3001 | return skb; |
fc70fb64 | 3002 | |
f6a78bfc HX |
3003 | out_kfree_skb: |
3004 | kfree_skb(skb); | |
eae3f88e | 3005 | out_null: |
d21fd63e | 3006 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3007 | return NULL; |
3008 | } | |
6afff0ca | 3009 | |
55a93b3e ED |
3010 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
3011 | { | |
3012 | struct sk_buff *next, *head = NULL, *tail; | |
3013 | ||
bec3cfdc | 3014 | for (; skb != NULL; skb = next) { |
55a93b3e ED |
3015 | next = skb->next; |
3016 | skb->next = NULL; | |
bec3cfdc ED |
3017 | |
3018 | /* in case skb wont be segmented, point to itself */ | |
3019 | skb->prev = skb; | |
3020 | ||
55a93b3e | 3021 | skb = validate_xmit_skb(skb, dev); |
bec3cfdc ED |
3022 | if (!skb) |
3023 | continue; | |
55a93b3e | 3024 | |
bec3cfdc ED |
3025 | if (!head) |
3026 | head = skb; | |
3027 | else | |
3028 | tail->next = skb; | |
3029 | /* If skb was segmented, skb->prev points to | |
3030 | * the last segment. If not, it still contains skb. | |
3031 | */ | |
3032 | tail = skb->prev; | |
55a93b3e ED |
3033 | } |
3034 | return head; | |
f6a78bfc | 3035 | } |
104ba78c | 3036 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3037 | |
1def9238 ED |
3038 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3039 | { | |
3040 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3041 | ||
3042 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3043 | ||
3044 | /* To get more precise estimation of bytes sent on wire, | |
3045 | * we add to pkt_len the headers size of all segments | |
3046 | */ | |
3047 | if (shinfo->gso_size) { | |
757b8b1d | 3048 | unsigned int hdr_len; |
15e5a030 | 3049 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3050 | |
757b8b1d ED |
3051 | /* mac layer + network layer */ |
3052 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3053 | ||
3054 | /* + transport layer */ | |
1def9238 ED |
3055 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
3056 | hdr_len += tcp_hdrlen(skb); | |
3057 | else | |
3058 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
3059 | |
3060 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3061 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3062 | shinfo->gso_size); | |
3063 | ||
3064 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3065 | } |
3066 | } | |
3067 | ||
bbd8a0d3 KK |
3068 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3069 | struct net_device *dev, | |
3070 | struct netdev_queue *txq) | |
3071 | { | |
3072 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3073 | struct sk_buff *to_free = NULL; |
a2da570d | 3074 | bool contended; |
bbd8a0d3 KK |
3075 | int rc; |
3076 | ||
a2da570d | 3077 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
3078 | /* |
3079 | * Heuristic to force contended enqueues to serialize on a | |
3080 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3081 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3082 | * often and dequeue packets faster. |
79640a4c | 3083 | */ |
a2da570d | 3084 | contended = qdisc_is_running(q); |
79640a4c ED |
3085 | if (unlikely(contended)) |
3086 | spin_lock(&q->busylock); | |
3087 | ||
bbd8a0d3 KK |
3088 | spin_lock(root_lock); |
3089 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3090 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3091 | rc = NET_XMIT_DROP; |
3092 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3093 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3094 | /* |
3095 | * This is a work-conserving queue; there are no old skbs | |
3096 | * waiting to be sent out; and the qdisc is not running - | |
3097 | * xmit the skb directly. | |
3098 | */ | |
bfe0d029 | 3099 | |
bfe0d029 ED |
3100 | qdisc_bstats_update(q, skb); |
3101 | ||
55a93b3e | 3102 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3103 | if (unlikely(contended)) { |
3104 | spin_unlock(&q->busylock); | |
3105 | contended = false; | |
3106 | } | |
bbd8a0d3 | 3107 | __qdisc_run(q); |
79640a4c | 3108 | } else |
bc135b23 | 3109 | qdisc_run_end(q); |
bbd8a0d3 KK |
3110 | |
3111 | rc = NET_XMIT_SUCCESS; | |
3112 | } else { | |
520ac30f | 3113 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3114 | if (qdisc_run_begin(q)) { |
3115 | if (unlikely(contended)) { | |
3116 | spin_unlock(&q->busylock); | |
3117 | contended = false; | |
3118 | } | |
3119 | __qdisc_run(q); | |
3120 | } | |
bbd8a0d3 KK |
3121 | } |
3122 | spin_unlock(root_lock); | |
520ac30f ED |
3123 | if (unlikely(to_free)) |
3124 | kfree_skb_list(to_free); | |
79640a4c ED |
3125 | if (unlikely(contended)) |
3126 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3127 | return rc; |
3128 | } | |
3129 | ||
86f8515f | 3130 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3131 | static void skb_update_prio(struct sk_buff *skb) |
3132 | { | |
6977a79d | 3133 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 3134 | |
91c68ce2 | 3135 | if (!skb->priority && skb->sk && map) { |
2a56a1fe TH |
3136 | unsigned int prioidx = |
3137 | sock_cgroup_prioidx(&skb->sk->sk_cgrp_data); | |
91c68ce2 ED |
3138 | |
3139 | if (prioidx < map->priomap_len) | |
3140 | skb->priority = map->priomap[prioidx]; | |
3141 | } | |
5bc1421e NH |
3142 | } |
3143 | #else | |
3144 | #define skb_update_prio(skb) | |
3145 | #endif | |
3146 | ||
f60e5990 | 3147 | DEFINE_PER_CPU(int, xmit_recursion); |
3148 | EXPORT_SYMBOL(xmit_recursion); | |
3149 | ||
95603e22 MM |
3150 | /** |
3151 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3152 | * @net: network namespace this loopback is happening in |
3153 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3154 | * @skb: buffer to transmit |
3155 | */ | |
0c4b51f0 | 3156 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3157 | { |
3158 | skb_reset_mac_header(skb); | |
3159 | __skb_pull(skb, skb_network_offset(skb)); | |
3160 | skb->pkt_type = PACKET_LOOPBACK; | |
3161 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3162 | WARN_ON(!skb_dst(skb)); | |
3163 | skb_dst_force(skb); | |
3164 | netif_rx_ni(skb); | |
3165 | return 0; | |
3166 | } | |
3167 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3168 | ||
1f211a1b DB |
3169 | #ifdef CONFIG_NET_EGRESS |
3170 | static struct sk_buff * | |
3171 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3172 | { | |
3173 | struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list); | |
3174 | struct tcf_result cl_res; | |
3175 | ||
3176 | if (!cl) | |
3177 | return skb; | |
3178 | ||
8dc07fdb | 3179 | /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ |
1f211a1b DB |
3180 | qdisc_bstats_cpu_update(cl->q, skb); |
3181 | ||
3182 | switch (tc_classify(skb, cl, &cl_res, false)) { | |
3183 | case TC_ACT_OK: | |
3184 | case TC_ACT_RECLASSIFY: | |
3185 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3186 | break; | |
3187 | case TC_ACT_SHOT: | |
3188 | qdisc_qstats_cpu_drop(cl->q); | |
3189 | *ret = NET_XMIT_DROP; | |
7e2c3aea DB |
3190 | kfree_skb(skb); |
3191 | return NULL; | |
1f211a1b DB |
3192 | case TC_ACT_STOLEN: |
3193 | case TC_ACT_QUEUED: | |
3194 | *ret = NET_XMIT_SUCCESS; | |
7e2c3aea | 3195 | consume_skb(skb); |
1f211a1b DB |
3196 | return NULL; |
3197 | case TC_ACT_REDIRECT: | |
3198 | /* No need to push/pop skb's mac_header here on egress! */ | |
3199 | skb_do_redirect(skb); | |
3200 | *ret = NET_XMIT_SUCCESS; | |
3201 | return NULL; | |
3202 | default: | |
3203 | break; | |
3204 | } | |
3205 | ||
3206 | return skb; | |
3207 | } | |
3208 | #endif /* CONFIG_NET_EGRESS */ | |
3209 | ||
638b2a69 JP |
3210 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
3211 | { | |
3212 | #ifdef CONFIG_XPS | |
3213 | struct xps_dev_maps *dev_maps; | |
3214 | struct xps_map *map; | |
3215 | int queue_index = -1; | |
3216 | ||
3217 | rcu_read_lock(); | |
3218 | dev_maps = rcu_dereference(dev->xps_maps); | |
3219 | if (dev_maps) { | |
184c449f AD |
3220 | unsigned int tci = skb->sender_cpu - 1; |
3221 | ||
3222 | if (dev->num_tc) { | |
3223 | tci *= dev->num_tc; | |
3224 | tci += netdev_get_prio_tc_map(dev, skb->priority); | |
3225 | } | |
3226 | ||
3227 | map = rcu_dereference(dev_maps->cpu_map[tci]); | |
638b2a69 JP |
3228 | if (map) { |
3229 | if (map->len == 1) | |
3230 | queue_index = map->queues[0]; | |
3231 | else | |
3232 | queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), | |
3233 | map->len)]; | |
3234 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3235 | queue_index = -1; | |
3236 | } | |
3237 | } | |
3238 | rcu_read_unlock(); | |
3239 | ||
3240 | return queue_index; | |
3241 | #else | |
3242 | return -1; | |
3243 | #endif | |
3244 | } | |
3245 | ||
3246 | static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) | |
3247 | { | |
3248 | struct sock *sk = skb->sk; | |
3249 | int queue_index = sk_tx_queue_get(sk); | |
3250 | ||
3251 | if (queue_index < 0 || skb->ooo_okay || | |
3252 | queue_index >= dev->real_num_tx_queues) { | |
3253 | int new_index = get_xps_queue(dev, skb); | |
f4563a75 | 3254 | |
638b2a69 JP |
3255 | if (new_index < 0) |
3256 | new_index = skb_tx_hash(dev, skb); | |
3257 | ||
3258 | if (queue_index != new_index && sk && | |
004a5d01 | 3259 | sk_fullsock(sk) && |
638b2a69 JP |
3260 | rcu_access_pointer(sk->sk_dst_cache)) |
3261 | sk_tx_queue_set(sk, new_index); | |
3262 | ||
3263 | queue_index = new_index; | |
3264 | } | |
3265 | ||
3266 | return queue_index; | |
3267 | } | |
3268 | ||
3269 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, | |
3270 | struct sk_buff *skb, | |
3271 | void *accel_priv) | |
3272 | { | |
3273 | int queue_index = 0; | |
3274 | ||
3275 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
3276 | u32 sender_cpu = skb->sender_cpu - 1; |
3277 | ||
3278 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
3279 | skb->sender_cpu = raw_smp_processor_id() + 1; |
3280 | #endif | |
3281 | ||
3282 | if (dev->real_num_tx_queues != 1) { | |
3283 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 3284 | |
638b2a69 JP |
3285 | if (ops->ndo_select_queue) |
3286 | queue_index = ops->ndo_select_queue(dev, skb, accel_priv, | |
3287 | __netdev_pick_tx); | |
3288 | else | |
3289 | queue_index = __netdev_pick_tx(dev, skb); | |
3290 | ||
3291 | if (!accel_priv) | |
3292 | queue_index = netdev_cap_txqueue(dev, queue_index); | |
3293 | } | |
3294 | ||
3295 | skb_set_queue_mapping(skb, queue_index); | |
3296 | return netdev_get_tx_queue(dev, queue_index); | |
3297 | } | |
3298 | ||
d29f749e | 3299 | /** |
9d08dd3d | 3300 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3301 | * @skb: buffer to transmit |
9d08dd3d | 3302 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
3303 | * |
3304 | * Queue a buffer for transmission to a network device. The caller must | |
3305 | * have set the device and priority and built the buffer before calling | |
3306 | * this function. The function can be called from an interrupt. | |
3307 | * | |
3308 | * A negative errno code is returned on a failure. A success does not | |
3309 | * guarantee the frame will be transmitted as it may be dropped due | |
3310 | * to congestion or traffic shaping. | |
3311 | * | |
3312 | * ----------------------------------------------------------------------------------- | |
3313 | * I notice this method can also return errors from the queue disciplines, | |
3314 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3315 | * be positive. | |
3316 | * | |
3317 | * Regardless of the return value, the skb is consumed, so it is currently | |
3318 | * difficult to retry a send to this method. (You can bump the ref count | |
3319 | * before sending to hold a reference for retry if you are careful.) | |
3320 | * | |
3321 | * When calling this method, interrupts MUST be enabled. This is because | |
3322 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3323 | * --BLG | |
3324 | */ | |
0a59f3a9 | 3325 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
3326 | { |
3327 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3328 | struct netdev_queue *txq; |
1da177e4 LT |
3329 | struct Qdisc *q; |
3330 | int rc = -ENOMEM; | |
3331 | ||
6d1ccff6 ED |
3332 | skb_reset_mac_header(skb); |
3333 | ||
e7fd2885 WB |
3334 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3335 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3336 | ||
4ec93edb YH |
3337 | /* Disable soft irqs for various locks below. Also |
3338 | * stops preemption for RCU. | |
1da177e4 | 3339 | */ |
4ec93edb | 3340 | rcu_read_lock_bh(); |
1da177e4 | 3341 | |
5bc1421e NH |
3342 | skb_update_prio(skb); |
3343 | ||
1f211a1b DB |
3344 | qdisc_pkt_len_init(skb); |
3345 | #ifdef CONFIG_NET_CLS_ACT | |
8dc07fdb | 3346 | skb->tc_at_ingress = 0; |
1f211a1b DB |
3347 | # ifdef CONFIG_NET_EGRESS |
3348 | if (static_key_false(&egress_needed)) { | |
3349 | skb = sch_handle_egress(skb, &rc, dev); | |
3350 | if (!skb) | |
3351 | goto out; | |
3352 | } | |
3353 | # endif | |
3354 | #endif | |
02875878 ED |
3355 | /* If device/qdisc don't need skb->dst, release it right now while |
3356 | * its hot in this cpu cache. | |
3357 | */ | |
3358 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
3359 | skb_dst_drop(skb); | |
3360 | else | |
3361 | skb_dst_force(skb); | |
3362 | ||
f663dd9a | 3363 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 3364 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 3365 | |
cf66ba58 | 3366 | trace_net_dev_queue(skb); |
1da177e4 | 3367 | if (q->enqueue) { |
bbd8a0d3 | 3368 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 3369 | goto out; |
1da177e4 LT |
3370 | } |
3371 | ||
3372 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 3373 | * loopback, all the sorts of tunnels... |
1da177e4 | 3374 | |
eb13da1a | 3375 | * Really, it is unlikely that netif_tx_lock protection is necessary |
3376 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
3377 | * counters.) | |
3378 | * However, it is possible, that they rely on protection | |
3379 | * made by us here. | |
1da177e4 | 3380 | |
eb13da1a | 3381 | * Check this and shot the lock. It is not prone from deadlocks. |
3382 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
3383 | */ |
3384 | if (dev->flags & IFF_UP) { | |
3385 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
3386 | ||
c773e847 | 3387 | if (txq->xmit_lock_owner != cpu) { |
a70b506e DB |
3388 | if (unlikely(__this_cpu_read(xmit_recursion) > |
3389 | XMIT_RECURSION_LIMIT)) | |
745e20f1 ED |
3390 | goto recursion_alert; |
3391 | ||
1f59533f JDB |
3392 | skb = validate_xmit_skb(skb, dev); |
3393 | if (!skb) | |
d21fd63e | 3394 | goto out; |
1f59533f | 3395 | |
c773e847 | 3396 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 3397 | |
73466498 | 3398 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 3399 | __this_cpu_inc(xmit_recursion); |
ce93718f | 3400 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 3401 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 3402 | if (dev_xmit_complete(rc)) { |
c773e847 | 3403 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
3404 | goto out; |
3405 | } | |
3406 | } | |
c773e847 | 3407 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
3408 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
3409 | dev->name); | |
1da177e4 LT |
3410 | } else { |
3411 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
3412 | * unfortunately |
3413 | */ | |
3414 | recursion_alert: | |
e87cc472 JP |
3415 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
3416 | dev->name); | |
1da177e4 LT |
3417 | } |
3418 | } | |
3419 | ||
3420 | rc = -ENETDOWN; | |
d4828d85 | 3421 | rcu_read_unlock_bh(); |
1da177e4 | 3422 | |
015f0688 | 3423 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 3424 | kfree_skb_list(skb); |
1da177e4 LT |
3425 | return rc; |
3426 | out: | |
d4828d85 | 3427 | rcu_read_unlock_bh(); |
1da177e4 LT |
3428 | return rc; |
3429 | } | |
f663dd9a | 3430 | |
2b4aa3ce | 3431 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
3432 | { |
3433 | return __dev_queue_xmit(skb, NULL); | |
3434 | } | |
2b4aa3ce | 3435 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3436 | |
f663dd9a JW |
3437 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3438 | { | |
3439 | return __dev_queue_xmit(skb, accel_priv); | |
3440 | } | |
3441 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3442 | ||
1da177e4 | 3443 | |
eb13da1a | 3444 | /************************************************************************* |
3445 | * Receiver routines | |
3446 | *************************************************************************/ | |
1da177e4 | 3447 | |
6b2bedc3 | 3448 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3449 | EXPORT_SYMBOL(netdev_max_backlog); |
3450 | ||
3b098e2d | 3451 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 | 3452 | int netdev_budget __read_mostly = 300; |
7acf8a1e | 3453 | unsigned int __read_mostly netdev_budget_usecs = 2000; |
3d48b53f MT |
3454 | int weight_p __read_mostly = 64; /* old backlog weight */ |
3455 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
3456 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
3457 | int dev_rx_weight __read_mostly = 64; | |
3458 | int dev_tx_weight __read_mostly = 64; | |
1da177e4 | 3459 | |
eecfd7c4 ED |
3460 | /* Called with irq disabled */ |
3461 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3462 | struct napi_struct *napi) | |
3463 | { | |
3464 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3465 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3466 | } | |
3467 | ||
bfb564e7 KK |
3468 | #ifdef CONFIG_RPS |
3469 | ||
3470 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3471 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 3472 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
3473 | u32 rps_cpu_mask __read_mostly; |
3474 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 3475 | |
c5905afb | 3476 | struct static_key rps_needed __read_mostly; |
3df97ba8 | 3477 | EXPORT_SYMBOL(rps_needed); |
13bfff25 ED |
3478 | struct static_key rfs_needed __read_mostly; |
3479 | EXPORT_SYMBOL(rfs_needed); | |
adc9300e | 3480 | |
c445477d BH |
3481 | static struct rps_dev_flow * |
3482 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3483 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3484 | { | |
a31196b0 | 3485 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
3486 | #ifdef CONFIG_RFS_ACCEL |
3487 | struct netdev_rx_queue *rxqueue; | |
3488 | struct rps_dev_flow_table *flow_table; | |
3489 | struct rps_dev_flow *old_rflow; | |
3490 | u32 flow_id; | |
3491 | u16 rxq_index; | |
3492 | int rc; | |
3493 | ||
3494 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3495 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3496 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3497 | goto out; |
3498 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3499 | if (rxq_index == skb_get_rx_queue(skb)) | |
3500 | goto out; | |
3501 | ||
3502 | rxqueue = dev->_rx + rxq_index; | |
3503 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3504 | if (!flow_table) | |
3505 | goto out; | |
61b905da | 3506 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3507 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3508 | rxq_index, flow_id); | |
3509 | if (rc < 0) | |
3510 | goto out; | |
3511 | old_rflow = rflow; | |
3512 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3513 | rflow->filter = rc; |
3514 | if (old_rflow->filter == rflow->filter) | |
3515 | old_rflow->filter = RPS_NO_FILTER; | |
3516 | out: | |
3517 | #endif | |
3518 | rflow->last_qtail = | |
09994d1b | 3519 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3520 | } |
3521 | ||
09994d1b | 3522 | rflow->cpu = next_cpu; |
c445477d BH |
3523 | return rflow; |
3524 | } | |
3525 | ||
bfb564e7 KK |
3526 | /* |
3527 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3528 | * CPU from the RPS map of the receiving queue for a given skb. | |
3529 | * rcu_read_lock must be held on entry. | |
3530 | */ | |
3531 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3532 | struct rps_dev_flow **rflowp) | |
3533 | { | |
567e4b79 ED |
3534 | const struct rps_sock_flow_table *sock_flow_table; |
3535 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 3536 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 3537 | struct rps_map *map; |
bfb564e7 | 3538 | int cpu = -1; |
567e4b79 | 3539 | u32 tcpu; |
61b905da | 3540 | u32 hash; |
bfb564e7 KK |
3541 | |
3542 | if (skb_rx_queue_recorded(skb)) { | |
3543 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 3544 | |
62fe0b40 BH |
3545 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3546 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3547 | "%s received packet on queue %u, but number " | |
3548 | "of RX queues is %u\n", | |
3549 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3550 | goto done; |
3551 | } | |
567e4b79 ED |
3552 | rxqueue += index; |
3553 | } | |
bfb564e7 | 3554 | |
567e4b79 ED |
3555 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
3556 | ||
3557 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 3558 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 3559 | if (!flow_table && !map) |
bfb564e7 KK |
3560 | goto done; |
3561 | ||
2d47b459 | 3562 | skb_reset_network_header(skb); |
61b905da TH |
3563 | hash = skb_get_hash(skb); |
3564 | if (!hash) | |
bfb564e7 KK |
3565 | goto done; |
3566 | ||
fec5e652 TH |
3567 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
3568 | if (flow_table && sock_flow_table) { | |
fec5e652 | 3569 | struct rps_dev_flow *rflow; |
567e4b79 ED |
3570 | u32 next_cpu; |
3571 | u32 ident; | |
3572 | ||
3573 | /* First check into global flow table if there is a match */ | |
3574 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
3575 | if ((ident ^ hash) & ~rps_cpu_mask) | |
3576 | goto try_rps; | |
fec5e652 | 3577 | |
567e4b79 ED |
3578 | next_cpu = ident & rps_cpu_mask; |
3579 | ||
3580 | /* OK, now we know there is a match, | |
3581 | * we can look at the local (per receive queue) flow table | |
3582 | */ | |
61b905da | 3583 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3584 | tcpu = rflow->cpu; |
3585 | ||
fec5e652 TH |
3586 | /* |
3587 | * If the desired CPU (where last recvmsg was done) is | |
3588 | * different from current CPU (one in the rx-queue flow | |
3589 | * table entry), switch if one of the following holds: | |
a31196b0 | 3590 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
3591 | * - Current CPU is offline. |
3592 | * - The current CPU's queue tail has advanced beyond the | |
3593 | * last packet that was enqueued using this table entry. | |
3594 | * This guarantees that all previous packets for the flow | |
3595 | * have been dequeued, thus preserving in order delivery. | |
3596 | */ | |
3597 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 3598 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 3599 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
3600 | rflow->last_qtail)) >= 0)) { |
3601 | tcpu = next_cpu; | |
c445477d | 3602 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3603 | } |
c445477d | 3604 | |
a31196b0 | 3605 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
3606 | *rflowp = rflow; |
3607 | cpu = tcpu; | |
3608 | goto done; | |
3609 | } | |
3610 | } | |
3611 | ||
567e4b79 ED |
3612 | try_rps: |
3613 | ||
0a9627f2 | 3614 | if (map) { |
8fc54f68 | 3615 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3616 | if (cpu_online(tcpu)) { |
3617 | cpu = tcpu; | |
3618 | goto done; | |
3619 | } | |
3620 | } | |
3621 | ||
3622 | done: | |
0a9627f2 TH |
3623 | return cpu; |
3624 | } | |
3625 | ||
c445477d BH |
3626 | #ifdef CONFIG_RFS_ACCEL |
3627 | ||
3628 | /** | |
3629 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3630 | * @dev: Device on which the filter was set | |
3631 | * @rxq_index: RX queue index | |
3632 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3633 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3634 | * | |
3635 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3636 | * this function for each installed filter and remove the filters for | |
3637 | * which it returns %true. | |
3638 | */ | |
3639 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3640 | u32 flow_id, u16 filter_id) | |
3641 | { | |
3642 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3643 | struct rps_dev_flow_table *flow_table; | |
3644 | struct rps_dev_flow *rflow; | |
3645 | bool expire = true; | |
a31196b0 | 3646 | unsigned int cpu; |
c445477d BH |
3647 | |
3648 | rcu_read_lock(); | |
3649 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3650 | if (flow_table && flow_id <= flow_table->mask) { | |
3651 | rflow = &flow_table->flows[flow_id]; | |
3652 | cpu = ACCESS_ONCE(rflow->cpu); | |
a31196b0 | 3653 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
3654 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
3655 | rflow->last_qtail) < | |
3656 | (int)(10 * flow_table->mask))) | |
3657 | expire = false; | |
3658 | } | |
3659 | rcu_read_unlock(); | |
3660 | return expire; | |
3661 | } | |
3662 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3663 | ||
3664 | #endif /* CONFIG_RFS_ACCEL */ | |
3665 | ||
0a9627f2 | 3666 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3667 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3668 | { |
e36fa2f7 ED |
3669 | struct softnet_data *sd = data; |
3670 | ||
eecfd7c4 | 3671 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3672 | sd->received_rps++; |
0a9627f2 | 3673 | } |
e36fa2f7 | 3674 | |
fec5e652 | 3675 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3676 | |
e36fa2f7 ED |
3677 | /* |
3678 | * Check if this softnet_data structure is another cpu one | |
3679 | * If yes, queue it to our IPI list and return 1 | |
3680 | * If no, return 0 | |
3681 | */ | |
3682 | static int rps_ipi_queued(struct softnet_data *sd) | |
3683 | { | |
3684 | #ifdef CONFIG_RPS | |
903ceff7 | 3685 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
3686 | |
3687 | if (sd != mysd) { | |
3688 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3689 | mysd->rps_ipi_list = sd; | |
3690 | ||
3691 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3692 | return 1; | |
3693 | } | |
3694 | #endif /* CONFIG_RPS */ | |
3695 | return 0; | |
3696 | } | |
3697 | ||
99bbc707 WB |
3698 | #ifdef CONFIG_NET_FLOW_LIMIT |
3699 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3700 | #endif | |
3701 | ||
3702 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3703 | { | |
3704 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3705 | struct sd_flow_limit *fl; | |
3706 | struct softnet_data *sd; | |
3707 | unsigned int old_flow, new_flow; | |
3708 | ||
3709 | if (qlen < (netdev_max_backlog >> 1)) | |
3710 | return false; | |
3711 | ||
903ceff7 | 3712 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
3713 | |
3714 | rcu_read_lock(); | |
3715 | fl = rcu_dereference(sd->flow_limit); | |
3716 | if (fl) { | |
3958afa1 | 3717 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3718 | old_flow = fl->history[fl->history_head]; |
3719 | fl->history[fl->history_head] = new_flow; | |
3720 | ||
3721 | fl->history_head++; | |
3722 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3723 | ||
3724 | if (likely(fl->buckets[old_flow])) | |
3725 | fl->buckets[old_flow]--; | |
3726 | ||
3727 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3728 | fl->count++; | |
3729 | rcu_read_unlock(); | |
3730 | return true; | |
3731 | } | |
3732 | } | |
3733 | rcu_read_unlock(); | |
3734 | #endif | |
3735 | return false; | |
3736 | } | |
3737 | ||
0a9627f2 TH |
3738 | /* |
3739 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3740 | * queue (may be a remote CPU queue). | |
3741 | */ | |
fec5e652 TH |
3742 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3743 | unsigned int *qtail) | |
0a9627f2 | 3744 | { |
e36fa2f7 | 3745 | struct softnet_data *sd; |
0a9627f2 | 3746 | unsigned long flags; |
99bbc707 | 3747 | unsigned int qlen; |
0a9627f2 | 3748 | |
e36fa2f7 | 3749 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3750 | |
3751 | local_irq_save(flags); | |
0a9627f2 | 3752 | |
e36fa2f7 | 3753 | rps_lock(sd); |
e9e4dd32 JA |
3754 | if (!netif_running(skb->dev)) |
3755 | goto drop; | |
99bbc707 WB |
3756 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3757 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 3758 | if (qlen) { |
0a9627f2 | 3759 | enqueue: |
e36fa2f7 | 3760 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3761 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3762 | rps_unlock(sd); |
152102c7 | 3763 | local_irq_restore(flags); |
0a9627f2 TH |
3764 | return NET_RX_SUCCESS; |
3765 | } | |
3766 | ||
ebda37c2 ED |
3767 | /* Schedule NAPI for backlog device |
3768 | * We can use non atomic operation since we own the queue lock | |
3769 | */ | |
3770 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3771 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3772 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3773 | } |
3774 | goto enqueue; | |
3775 | } | |
3776 | ||
e9e4dd32 | 3777 | drop: |
dee42870 | 3778 | sd->dropped++; |
e36fa2f7 | 3779 | rps_unlock(sd); |
0a9627f2 | 3780 | |
0a9627f2 TH |
3781 | local_irq_restore(flags); |
3782 | ||
caf586e5 | 3783 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3784 | kfree_skb(skb); |
3785 | return NET_RX_DROP; | |
3786 | } | |
1da177e4 | 3787 | |
ae78dbfa | 3788 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3789 | { |
b0e28f1e | 3790 | int ret; |
1da177e4 | 3791 | |
588f0330 | 3792 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3793 | |
cf66ba58 | 3794 | trace_netif_rx(skb); |
df334545 | 3795 | #ifdef CONFIG_RPS |
c5905afb | 3796 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3797 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3798 | int cpu; |
3799 | ||
cece1945 | 3800 | preempt_disable(); |
b0e28f1e | 3801 | rcu_read_lock(); |
fec5e652 TH |
3802 | |
3803 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3804 | if (cpu < 0) |
3805 | cpu = smp_processor_id(); | |
fec5e652 TH |
3806 | |
3807 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3808 | ||
b0e28f1e | 3809 | rcu_read_unlock(); |
cece1945 | 3810 | preempt_enable(); |
adc9300e ED |
3811 | } else |
3812 | #endif | |
fec5e652 TH |
3813 | { |
3814 | unsigned int qtail; | |
f4563a75 | 3815 | |
fec5e652 TH |
3816 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); |
3817 | put_cpu(); | |
3818 | } | |
b0e28f1e | 3819 | return ret; |
1da177e4 | 3820 | } |
ae78dbfa BH |
3821 | |
3822 | /** | |
3823 | * netif_rx - post buffer to the network code | |
3824 | * @skb: buffer to post | |
3825 | * | |
3826 | * This function receives a packet from a device driver and queues it for | |
3827 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3828 | * may be dropped during processing for congestion control or by the | |
3829 | * protocol layers. | |
3830 | * | |
3831 | * return values: | |
3832 | * NET_RX_SUCCESS (no congestion) | |
3833 | * NET_RX_DROP (packet was dropped) | |
3834 | * | |
3835 | */ | |
3836 | ||
3837 | int netif_rx(struct sk_buff *skb) | |
3838 | { | |
3839 | trace_netif_rx_entry(skb); | |
3840 | ||
3841 | return netif_rx_internal(skb); | |
3842 | } | |
d1b19dff | 3843 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3844 | |
3845 | int netif_rx_ni(struct sk_buff *skb) | |
3846 | { | |
3847 | int err; | |
3848 | ||
ae78dbfa BH |
3849 | trace_netif_rx_ni_entry(skb); |
3850 | ||
1da177e4 | 3851 | preempt_disable(); |
ae78dbfa | 3852 | err = netif_rx_internal(skb); |
1da177e4 LT |
3853 | if (local_softirq_pending()) |
3854 | do_softirq(); | |
3855 | preempt_enable(); | |
3856 | ||
3857 | return err; | |
3858 | } | |
1da177e4 LT |
3859 | EXPORT_SYMBOL(netif_rx_ni); |
3860 | ||
0766f788 | 3861 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 3862 | { |
903ceff7 | 3863 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
3864 | |
3865 | if (sd->completion_queue) { | |
3866 | struct sk_buff *clist; | |
3867 | ||
3868 | local_irq_disable(); | |
3869 | clist = sd->completion_queue; | |
3870 | sd->completion_queue = NULL; | |
3871 | local_irq_enable(); | |
3872 | ||
3873 | while (clist) { | |
3874 | struct sk_buff *skb = clist; | |
f4563a75 | 3875 | |
1da177e4 LT |
3876 | clist = clist->next; |
3877 | ||
547b792c | 3878 | WARN_ON(atomic_read(&skb->users)); |
e6247027 ED |
3879 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
3880 | trace_consume_skb(skb); | |
3881 | else | |
3882 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
3883 | |
3884 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
3885 | __kfree_skb(skb); | |
3886 | else | |
3887 | __kfree_skb_defer(skb); | |
1da177e4 | 3888 | } |
15fad714 JDB |
3889 | |
3890 | __kfree_skb_flush(); | |
1da177e4 LT |
3891 | } |
3892 | ||
3893 | if (sd->output_queue) { | |
37437bb2 | 3894 | struct Qdisc *head; |
1da177e4 LT |
3895 | |
3896 | local_irq_disable(); | |
3897 | head = sd->output_queue; | |
3898 | sd->output_queue = NULL; | |
a9cbd588 | 3899 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3900 | local_irq_enable(); |
3901 | ||
3902 | while (head) { | |
37437bb2 DM |
3903 | struct Qdisc *q = head; |
3904 | spinlock_t *root_lock; | |
3905 | ||
1da177e4 LT |
3906 | head = head->next_sched; |
3907 | ||
5fb66229 | 3908 | root_lock = qdisc_lock(q); |
3bcb846c ED |
3909 | spin_lock(root_lock); |
3910 | /* We need to make sure head->next_sched is read | |
3911 | * before clearing __QDISC_STATE_SCHED | |
3912 | */ | |
3913 | smp_mb__before_atomic(); | |
3914 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
3915 | qdisc_run(q); | |
3916 | spin_unlock(root_lock); | |
1da177e4 LT |
3917 | } |
3918 | } | |
3919 | } | |
3920 | ||
181402a5 | 3921 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
3922 | /* This hook is defined here for ATM LANE */ |
3923 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3924 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3925 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3926 | #endif |
1da177e4 | 3927 | |
1f211a1b DB |
3928 | static inline struct sk_buff * |
3929 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
3930 | struct net_device *orig_dev) | |
f697c3e8 | 3931 | { |
e7582bab | 3932 | #ifdef CONFIG_NET_CLS_ACT |
d2788d34 DB |
3933 | struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list); |
3934 | struct tcf_result cl_res; | |
24824a09 | 3935 | |
c9e99fd0 DB |
3936 | /* If there's at least one ingress present somewhere (so |
3937 | * we get here via enabled static key), remaining devices | |
3938 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 3939 | * out here. |
c9e99fd0 | 3940 | */ |
d2788d34 | 3941 | if (!cl) |
4577139b | 3942 | return skb; |
f697c3e8 HX |
3943 | if (*pt_prev) { |
3944 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3945 | *pt_prev = NULL; | |
1da177e4 LT |
3946 | } |
3947 | ||
3365495c | 3948 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
8dc07fdb | 3949 | skb->tc_at_ingress = 1; |
24ea591d | 3950 | qdisc_bstats_cpu_update(cl->q, skb); |
c9e99fd0 | 3951 | |
3b3ae880 | 3952 | switch (tc_classify(skb, cl, &cl_res, false)) { |
d2788d34 DB |
3953 | case TC_ACT_OK: |
3954 | case TC_ACT_RECLASSIFY: | |
3955 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3956 | break; | |
3957 | case TC_ACT_SHOT: | |
24ea591d | 3958 | qdisc_qstats_cpu_drop(cl->q); |
8a3a4c6e ED |
3959 | kfree_skb(skb); |
3960 | return NULL; | |
d2788d34 DB |
3961 | case TC_ACT_STOLEN: |
3962 | case TC_ACT_QUEUED: | |
8a3a4c6e | 3963 | consume_skb(skb); |
d2788d34 | 3964 | return NULL; |
27b29f63 AS |
3965 | case TC_ACT_REDIRECT: |
3966 | /* skb_mac_header check was done by cls/act_bpf, so | |
3967 | * we can safely push the L2 header back before | |
3968 | * redirecting to another netdev | |
3969 | */ | |
3970 | __skb_push(skb, skb->mac_len); | |
3971 | skb_do_redirect(skb); | |
3972 | return NULL; | |
d2788d34 DB |
3973 | default: |
3974 | break; | |
f697c3e8 | 3975 | } |
e7582bab | 3976 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
3977 | return skb; |
3978 | } | |
1da177e4 | 3979 | |
24b27fc4 MB |
3980 | /** |
3981 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
3982 | * @dev: device to check | |
3983 | * | |
3984 | * Check if a receive handler is already registered for a given device. | |
3985 | * Return true if there one. | |
3986 | * | |
3987 | * The caller must hold the rtnl_mutex. | |
3988 | */ | |
3989 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
3990 | { | |
3991 | ASSERT_RTNL(); | |
3992 | return dev && rtnl_dereference(dev->rx_handler); | |
3993 | } | |
3994 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
3995 | ||
ab95bfe0 JP |
3996 | /** |
3997 | * netdev_rx_handler_register - register receive handler | |
3998 | * @dev: device to register a handler for | |
3999 | * @rx_handler: receive handler to register | |
93e2c32b | 4000 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 4001 | * |
e227867f | 4002 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
4003 | * called from __netif_receive_skb. A negative errno code is returned |
4004 | * on a failure. | |
4005 | * | |
4006 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
4007 | * |
4008 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
4009 | */ |
4010 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
4011 | rx_handler_func_t *rx_handler, |
4012 | void *rx_handler_data) | |
ab95bfe0 | 4013 | { |
1b7cd004 | 4014 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
4015 | return -EBUSY; |
4016 | ||
00cfec37 | 4017 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 4018 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
4019 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
4020 | ||
4021 | return 0; | |
4022 | } | |
4023 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
4024 | ||
4025 | /** | |
4026 | * netdev_rx_handler_unregister - unregister receive handler | |
4027 | * @dev: device to unregister a handler from | |
4028 | * | |
166ec369 | 4029 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
4030 | * |
4031 | * The caller must hold the rtnl_mutex. | |
4032 | */ | |
4033 | void netdev_rx_handler_unregister(struct net_device *dev) | |
4034 | { | |
4035 | ||
4036 | ASSERT_RTNL(); | |
a9b3cd7f | 4037 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
4038 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
4039 | * section has a guarantee to see a non NULL rx_handler_data | |
4040 | * as well. | |
4041 | */ | |
4042 | synchronize_net(); | |
a9b3cd7f | 4043 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
4044 | } |
4045 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
4046 | ||
b4b9e355 MG |
4047 | /* |
4048 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
4049 | * the special handling of PFMEMALLOC skbs. | |
4050 | */ | |
4051 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
4052 | { | |
4053 | switch (skb->protocol) { | |
2b8837ae JP |
4054 | case htons(ETH_P_ARP): |
4055 | case htons(ETH_P_IP): | |
4056 | case htons(ETH_P_IPV6): | |
4057 | case htons(ETH_P_8021Q): | |
4058 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
4059 | return true; |
4060 | default: | |
4061 | return false; | |
4062 | } | |
4063 | } | |
4064 | ||
e687ad60 PN |
4065 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
4066 | int *ret, struct net_device *orig_dev) | |
4067 | { | |
e7582bab | 4068 | #ifdef CONFIG_NETFILTER_INGRESS |
e687ad60 | 4069 | if (nf_hook_ingress_active(skb)) { |
2c1e2703 AC |
4070 | int ingress_retval; |
4071 | ||
e687ad60 PN |
4072 | if (*pt_prev) { |
4073 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4074 | *pt_prev = NULL; | |
4075 | } | |
4076 | ||
2c1e2703 AC |
4077 | rcu_read_lock(); |
4078 | ingress_retval = nf_hook_ingress(skb); | |
4079 | rcu_read_unlock(); | |
4080 | return ingress_retval; | |
e687ad60 | 4081 | } |
e7582bab | 4082 | #endif /* CONFIG_NETFILTER_INGRESS */ |
e687ad60 PN |
4083 | return 0; |
4084 | } | |
e687ad60 | 4085 | |
9754e293 | 4086 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
4087 | { |
4088 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 4089 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 4090 | struct net_device *orig_dev; |
8a4eb573 | 4091 | bool deliver_exact = false; |
1da177e4 | 4092 | int ret = NET_RX_DROP; |
252e3346 | 4093 | __be16 type; |
1da177e4 | 4094 | |
588f0330 | 4095 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 4096 | |
cf66ba58 | 4097 | trace_netif_receive_skb(skb); |
9b22ea56 | 4098 | |
cc9bd5ce | 4099 | orig_dev = skb->dev; |
8f903c70 | 4100 | |
c1d2bbe1 | 4101 | skb_reset_network_header(skb); |
fda55eca ED |
4102 | if (!skb_transport_header_was_set(skb)) |
4103 | skb_reset_transport_header(skb); | |
0b5c9db1 | 4104 | skb_reset_mac_len(skb); |
1da177e4 LT |
4105 | |
4106 | pt_prev = NULL; | |
4107 | ||
63d8ea7f | 4108 | another_round: |
b6858177 | 4109 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
4110 | |
4111 | __this_cpu_inc(softnet_data.processed); | |
4112 | ||
8ad227ff PM |
4113 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
4114 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 4115 | skb = skb_vlan_untag(skb); |
bcc6d479 | 4116 | if (unlikely(!skb)) |
2c17d27c | 4117 | goto out; |
bcc6d479 JP |
4118 | } |
4119 | ||
e7246e12 WB |
4120 | if (skb_skip_tc_classify(skb)) |
4121 | goto skip_classify; | |
1da177e4 | 4122 | |
9754e293 | 4123 | if (pfmemalloc) |
b4b9e355 MG |
4124 | goto skip_taps; |
4125 | ||
1da177e4 | 4126 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
4127 | if (pt_prev) |
4128 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4129 | pt_prev = ptype; | |
4130 | } | |
4131 | ||
4132 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
4133 | if (pt_prev) | |
4134 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4135 | pt_prev = ptype; | |
1da177e4 LT |
4136 | } |
4137 | ||
b4b9e355 | 4138 | skip_taps: |
1cf51900 | 4139 | #ifdef CONFIG_NET_INGRESS |
4577139b | 4140 | if (static_key_false(&ingress_needed)) { |
1f211a1b | 4141 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); |
4577139b | 4142 | if (!skb) |
2c17d27c | 4143 | goto out; |
e687ad60 PN |
4144 | |
4145 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 4146 | goto out; |
4577139b | 4147 | } |
1cf51900 | 4148 | #endif |
a5135bcf | 4149 | skb_reset_tc(skb); |
e7246e12 | 4150 | skip_classify: |
9754e293 | 4151 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
4152 | goto drop; |
4153 | ||
df8a39de | 4154 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
4155 | if (pt_prev) { |
4156 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4157 | pt_prev = NULL; | |
4158 | } | |
48cc32d3 | 4159 | if (vlan_do_receive(&skb)) |
2425717b JF |
4160 | goto another_round; |
4161 | else if (unlikely(!skb)) | |
2c17d27c | 4162 | goto out; |
2425717b JF |
4163 | } |
4164 | ||
48cc32d3 | 4165 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
4166 | if (rx_handler) { |
4167 | if (pt_prev) { | |
4168 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4169 | pt_prev = NULL; | |
4170 | } | |
8a4eb573 JP |
4171 | switch (rx_handler(&skb)) { |
4172 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 4173 | ret = NET_RX_SUCCESS; |
2c17d27c | 4174 | goto out; |
8a4eb573 | 4175 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 4176 | goto another_round; |
8a4eb573 JP |
4177 | case RX_HANDLER_EXACT: |
4178 | deliver_exact = true; | |
4179 | case RX_HANDLER_PASS: | |
4180 | break; | |
4181 | default: | |
4182 | BUG(); | |
4183 | } | |
ab95bfe0 | 4184 | } |
1da177e4 | 4185 | |
df8a39de JP |
4186 | if (unlikely(skb_vlan_tag_present(skb))) { |
4187 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
4188 | skb->pkt_type = PACKET_OTHERHOST; |
4189 | /* Note: we might in the future use prio bits | |
4190 | * and set skb->priority like in vlan_do_receive() | |
4191 | * For the time being, just ignore Priority Code Point | |
4192 | */ | |
4193 | skb->vlan_tci = 0; | |
4194 | } | |
48cc32d3 | 4195 | |
7866a621 SN |
4196 | type = skb->protocol; |
4197 | ||
63d8ea7f | 4198 | /* deliver only exact match when indicated */ |
7866a621 SN |
4199 | if (likely(!deliver_exact)) { |
4200 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4201 | &ptype_base[ntohs(type) & | |
4202 | PTYPE_HASH_MASK]); | |
4203 | } | |
1f3c8804 | 4204 | |
7866a621 SN |
4205 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
4206 | &orig_dev->ptype_specific); | |
4207 | ||
4208 | if (unlikely(skb->dev != orig_dev)) { | |
4209 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4210 | &skb->dev->ptype_specific); | |
1da177e4 LT |
4211 | } |
4212 | ||
4213 | if (pt_prev) { | |
1080e512 | 4214 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 4215 | goto drop; |
1080e512 MT |
4216 | else |
4217 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 4218 | } else { |
b4b9e355 | 4219 | drop: |
6e7333d3 JW |
4220 | if (!deliver_exact) |
4221 | atomic_long_inc(&skb->dev->rx_dropped); | |
4222 | else | |
4223 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
4224 | kfree_skb(skb); |
4225 | /* Jamal, now you will not able to escape explaining | |
4226 | * me how you were going to use this. :-) | |
4227 | */ | |
4228 | ret = NET_RX_DROP; | |
4229 | } | |
4230 | ||
2c17d27c | 4231 | out: |
9754e293 DM |
4232 | return ret; |
4233 | } | |
4234 | ||
4235 | static int __netif_receive_skb(struct sk_buff *skb) | |
4236 | { | |
4237 | int ret; | |
4238 | ||
4239 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 4240 | unsigned int noreclaim_flag; |
9754e293 DM |
4241 | |
4242 | /* | |
4243 | * PFMEMALLOC skbs are special, they should | |
4244 | * - be delivered to SOCK_MEMALLOC sockets only | |
4245 | * - stay away from userspace | |
4246 | * - have bounded memory usage | |
4247 | * | |
4248 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
4249 | * context down to all allocation sites. | |
4250 | */ | |
f1083048 | 4251 | noreclaim_flag = memalloc_noreclaim_save(); |
9754e293 | 4252 | ret = __netif_receive_skb_core(skb, true); |
f1083048 | 4253 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 DM |
4254 | } else |
4255 | ret = __netif_receive_skb_core(skb, false); | |
4256 | ||
1da177e4 LT |
4257 | return ret; |
4258 | } | |
0a9627f2 | 4259 | |
b5cdae32 DM |
4260 | static struct static_key generic_xdp_needed __read_mostly; |
4261 | ||
4262 | static int generic_xdp_install(struct net_device *dev, struct netdev_xdp *xdp) | |
4263 | { | |
4264 | struct bpf_prog *new = xdp->prog; | |
4265 | int ret = 0; | |
4266 | ||
4267 | switch (xdp->command) { | |
4268 | case XDP_SETUP_PROG: { | |
4269 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); | |
4270 | ||
4271 | rcu_assign_pointer(dev->xdp_prog, new); | |
4272 | if (old) | |
4273 | bpf_prog_put(old); | |
4274 | ||
4275 | if (old && !new) { | |
4276 | static_key_slow_dec(&generic_xdp_needed); | |
4277 | } else if (new && !old) { | |
4278 | static_key_slow_inc(&generic_xdp_needed); | |
4279 | dev_disable_lro(dev); | |
4280 | } | |
4281 | break; | |
4282 | } | |
4283 | ||
4284 | case XDP_QUERY_PROG: | |
4285 | xdp->prog_attached = !!rcu_access_pointer(dev->xdp_prog); | |
4286 | break; | |
4287 | ||
4288 | default: | |
4289 | ret = -EINVAL; | |
4290 | break; | |
4291 | } | |
4292 | ||
4293 | return ret; | |
4294 | } | |
4295 | ||
4296 | static u32 netif_receive_generic_xdp(struct sk_buff *skb, | |
4297 | struct bpf_prog *xdp_prog) | |
4298 | { | |
4299 | struct xdp_buff xdp; | |
4300 | u32 act = XDP_DROP; | |
4301 | void *orig_data; | |
4302 | int hlen, off; | |
4303 | u32 mac_len; | |
4304 | ||
4305 | /* Reinjected packets coming from act_mirred or similar should | |
4306 | * not get XDP generic processing. | |
4307 | */ | |
4308 | if (skb_cloned(skb)) | |
4309 | return XDP_PASS; | |
4310 | ||
4311 | if (skb_linearize(skb)) | |
4312 | goto do_drop; | |
4313 | ||
4314 | /* The XDP program wants to see the packet starting at the MAC | |
4315 | * header. | |
4316 | */ | |
4317 | mac_len = skb->data - skb_mac_header(skb); | |
4318 | hlen = skb_headlen(skb) + mac_len; | |
4319 | xdp.data = skb->data - mac_len; | |
4320 | xdp.data_end = xdp.data + hlen; | |
4321 | xdp.data_hard_start = skb->data - skb_headroom(skb); | |
4322 | orig_data = xdp.data; | |
4323 | ||
4324 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
4325 | ||
4326 | off = xdp.data - orig_data; | |
4327 | if (off > 0) | |
4328 | __skb_pull(skb, off); | |
4329 | else if (off < 0) | |
4330 | __skb_push(skb, -off); | |
4331 | ||
4332 | switch (act) { | |
4333 | case XDP_TX: | |
4334 | __skb_push(skb, mac_len); | |
4335 | /* fall through */ | |
4336 | case XDP_PASS: | |
4337 | break; | |
4338 | ||
4339 | default: | |
4340 | bpf_warn_invalid_xdp_action(act); | |
4341 | /* fall through */ | |
4342 | case XDP_ABORTED: | |
4343 | trace_xdp_exception(skb->dev, xdp_prog, act); | |
4344 | /* fall through */ | |
4345 | case XDP_DROP: | |
4346 | do_drop: | |
4347 | kfree_skb(skb); | |
4348 | break; | |
4349 | } | |
4350 | ||
4351 | return act; | |
4352 | } | |
4353 | ||
4354 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
4355 | * network taps in order to match in-driver-XDP behavior. | |
4356 | */ | |
4357 | static void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) | |
4358 | { | |
4359 | struct net_device *dev = skb->dev; | |
4360 | struct netdev_queue *txq; | |
4361 | bool free_skb = true; | |
4362 | int cpu, rc; | |
4363 | ||
4364 | txq = netdev_pick_tx(dev, skb, NULL); | |
4365 | cpu = smp_processor_id(); | |
4366 | HARD_TX_LOCK(dev, txq, cpu); | |
4367 | if (!netif_xmit_stopped(txq)) { | |
4368 | rc = netdev_start_xmit(skb, dev, txq, 0); | |
4369 | if (dev_xmit_complete(rc)) | |
4370 | free_skb = false; | |
4371 | } | |
4372 | HARD_TX_UNLOCK(dev, txq); | |
4373 | if (free_skb) { | |
4374 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
4375 | kfree_skb(skb); | |
4376 | } | |
4377 | } | |
4378 | ||
ae78dbfa | 4379 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 4380 | { |
2c17d27c JA |
4381 | int ret; |
4382 | ||
588f0330 | 4383 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 4384 | |
c1f19b51 RC |
4385 | if (skb_defer_rx_timestamp(skb)) |
4386 | return NET_RX_SUCCESS; | |
4387 | ||
2c17d27c JA |
4388 | rcu_read_lock(); |
4389 | ||
b5cdae32 DM |
4390 | if (static_key_false(&generic_xdp_needed)) { |
4391 | struct bpf_prog *xdp_prog = rcu_dereference(skb->dev->xdp_prog); | |
4392 | ||
4393 | if (xdp_prog) { | |
4394 | u32 act = netif_receive_generic_xdp(skb, xdp_prog); | |
4395 | ||
4396 | if (act != XDP_PASS) { | |
4397 | rcu_read_unlock(); | |
4398 | if (act == XDP_TX) | |
4399 | generic_xdp_tx(skb, xdp_prog); | |
4400 | return NET_RX_DROP; | |
4401 | } | |
4402 | } | |
4403 | } | |
4404 | ||
df334545 | 4405 | #ifdef CONFIG_RPS |
c5905afb | 4406 | if (static_key_false(&rps_needed)) { |
3b098e2d | 4407 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 4408 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 4409 | |
3b098e2d ED |
4410 | if (cpu >= 0) { |
4411 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4412 | rcu_read_unlock(); | |
adc9300e | 4413 | return ret; |
3b098e2d | 4414 | } |
fec5e652 | 4415 | } |
1e94d72f | 4416 | #endif |
2c17d27c JA |
4417 | ret = __netif_receive_skb(skb); |
4418 | rcu_read_unlock(); | |
4419 | return ret; | |
0a9627f2 | 4420 | } |
ae78dbfa BH |
4421 | |
4422 | /** | |
4423 | * netif_receive_skb - process receive buffer from network | |
4424 | * @skb: buffer to process | |
4425 | * | |
4426 | * netif_receive_skb() is the main receive data processing function. | |
4427 | * It always succeeds. The buffer may be dropped during processing | |
4428 | * for congestion control or by the protocol layers. | |
4429 | * | |
4430 | * This function may only be called from softirq context and interrupts | |
4431 | * should be enabled. | |
4432 | * | |
4433 | * Return values (usually ignored): | |
4434 | * NET_RX_SUCCESS: no congestion | |
4435 | * NET_RX_DROP: packet was dropped | |
4436 | */ | |
04eb4489 | 4437 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa BH |
4438 | { |
4439 | trace_netif_receive_skb_entry(skb); | |
4440 | ||
4441 | return netif_receive_skb_internal(skb); | |
4442 | } | |
04eb4489 | 4443 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 4444 | |
41852497 | 4445 | DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
4446 | |
4447 | /* Network device is going away, flush any packets still pending */ | |
4448 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 4449 | { |
6e583ce5 | 4450 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
4451 | struct softnet_data *sd; |
4452 | ||
4453 | local_bh_disable(); | |
4454 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 4455 | |
145dd5f9 | 4456 | local_irq_disable(); |
e36fa2f7 | 4457 | rps_lock(sd); |
6e7676c1 | 4458 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 4459 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 4460 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 4461 | kfree_skb(skb); |
76cc8b13 | 4462 | input_queue_head_incr(sd); |
6e583ce5 | 4463 | } |
6e7676c1 | 4464 | } |
e36fa2f7 | 4465 | rps_unlock(sd); |
145dd5f9 | 4466 | local_irq_enable(); |
6e7676c1 CG |
4467 | |
4468 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 4469 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
4470 | __skb_unlink(skb, &sd->process_queue); |
4471 | kfree_skb(skb); | |
76cc8b13 | 4472 | input_queue_head_incr(sd); |
6e7676c1 CG |
4473 | } |
4474 | } | |
145dd5f9 PA |
4475 | local_bh_enable(); |
4476 | } | |
4477 | ||
41852497 | 4478 | static void flush_all_backlogs(void) |
145dd5f9 PA |
4479 | { |
4480 | unsigned int cpu; | |
4481 | ||
4482 | get_online_cpus(); | |
4483 | ||
41852497 ED |
4484 | for_each_online_cpu(cpu) |
4485 | queue_work_on(cpu, system_highpri_wq, | |
4486 | per_cpu_ptr(&flush_works, cpu)); | |
145dd5f9 PA |
4487 | |
4488 | for_each_online_cpu(cpu) | |
41852497 | 4489 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 PA |
4490 | |
4491 | put_online_cpus(); | |
6e583ce5 SH |
4492 | } |
4493 | ||
d565b0a1 HX |
4494 | static int napi_gro_complete(struct sk_buff *skb) |
4495 | { | |
22061d80 | 4496 | struct packet_offload *ptype; |
d565b0a1 | 4497 | __be16 type = skb->protocol; |
22061d80 | 4498 | struct list_head *head = &offload_base; |
d565b0a1 HX |
4499 | int err = -ENOENT; |
4500 | ||
c3c7c254 ED |
4501 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
4502 | ||
fc59f9a3 HX |
4503 | if (NAPI_GRO_CB(skb)->count == 1) { |
4504 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 4505 | goto out; |
fc59f9a3 | 4506 | } |
d565b0a1 HX |
4507 | |
4508 | rcu_read_lock(); | |
4509 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4510 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
4511 | continue; |
4512 | ||
299603e8 | 4513 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
4514 | break; |
4515 | } | |
4516 | rcu_read_unlock(); | |
4517 | ||
4518 | if (err) { | |
4519 | WARN_ON(&ptype->list == head); | |
4520 | kfree_skb(skb); | |
4521 | return NET_RX_SUCCESS; | |
4522 | } | |
4523 | ||
4524 | out: | |
ae78dbfa | 4525 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
4526 | } |
4527 | ||
2e71a6f8 ED |
4528 | /* napi->gro_list contains packets ordered by age. |
4529 | * youngest packets at the head of it. | |
4530 | * Complete skbs in reverse order to reduce latencies. | |
4531 | */ | |
4532 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 4533 | { |
2e71a6f8 | 4534 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 4535 | |
2e71a6f8 ED |
4536 | /* scan list and build reverse chain */ |
4537 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
4538 | skb->prev = prev; | |
4539 | prev = skb; | |
4540 | } | |
4541 | ||
4542 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 4543 | skb->next = NULL; |
2e71a6f8 ED |
4544 | |
4545 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
4546 | return; | |
4547 | ||
4548 | prev = skb->prev; | |
d565b0a1 | 4549 | napi_gro_complete(skb); |
2e71a6f8 | 4550 | napi->gro_count--; |
d565b0a1 HX |
4551 | } |
4552 | ||
4553 | napi->gro_list = NULL; | |
4554 | } | |
86cac58b | 4555 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 4556 | |
89c5fa33 ED |
4557 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
4558 | { | |
4559 | struct sk_buff *p; | |
4560 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 4561 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
4562 | |
4563 | for (p = napi->gro_list; p; p = p->next) { | |
4564 | unsigned long diffs; | |
4565 | ||
0b4cec8c TH |
4566 | NAPI_GRO_CB(p)->flush = 0; |
4567 | ||
4568 | if (hash != skb_get_hash_raw(p)) { | |
4569 | NAPI_GRO_CB(p)->same_flow = 0; | |
4570 | continue; | |
4571 | } | |
4572 | ||
89c5fa33 ED |
4573 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
4574 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
ce87fc6c | 4575 | diffs |= skb_metadata_dst_cmp(p, skb); |
89c5fa33 ED |
4576 | if (maclen == ETH_HLEN) |
4577 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 4578 | skb_mac_header(skb)); |
89c5fa33 ED |
4579 | else if (!diffs) |
4580 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 4581 | skb_mac_header(skb), |
89c5fa33 ED |
4582 | maclen); |
4583 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
4584 | } |
4585 | } | |
4586 | ||
299603e8 JC |
4587 | static void skb_gro_reset_offset(struct sk_buff *skb) |
4588 | { | |
4589 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4590 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
4591 | ||
4592 | NAPI_GRO_CB(skb)->data_offset = 0; | |
4593 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
4594 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
4595 | ||
4596 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
4597 | pinfo->nr_frags && | |
4598 | !PageHighMem(skb_frag_page(frag0))) { | |
4599 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
7cfd5fd5 ED |
4600 | NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, |
4601 | skb_frag_size(frag0), | |
4602 | skb->end - skb->tail); | |
89c5fa33 ED |
4603 | } |
4604 | } | |
4605 | ||
a50e233c ED |
4606 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
4607 | { | |
4608 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4609 | ||
4610 | BUG_ON(skb->end - skb->tail < grow); | |
4611 | ||
4612 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
4613 | ||
4614 | skb->data_len -= grow; | |
4615 | skb->tail += grow; | |
4616 | ||
4617 | pinfo->frags[0].page_offset += grow; | |
4618 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
4619 | ||
4620 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
4621 | skb_frag_unref(skb, 0); | |
4622 | memmove(pinfo->frags, pinfo->frags + 1, | |
4623 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
4624 | } | |
4625 | } | |
4626 | ||
bb728820 | 4627 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
4628 | { |
4629 | struct sk_buff **pp = NULL; | |
22061d80 | 4630 | struct packet_offload *ptype; |
d565b0a1 | 4631 | __be16 type = skb->protocol; |
22061d80 | 4632 | struct list_head *head = &offload_base; |
0da2afd5 | 4633 | int same_flow; |
5b252f0c | 4634 | enum gro_result ret; |
a50e233c | 4635 | int grow; |
d565b0a1 | 4636 | |
b5cdae32 | 4637 | if (netif_elide_gro(skb->dev)) |
d565b0a1 HX |
4638 | goto normal; |
4639 | ||
d61d072e | 4640 | if (skb->csum_bad) |
f17f5c91 HX |
4641 | goto normal; |
4642 | ||
89c5fa33 ED |
4643 | gro_list_prepare(napi, skb); |
4644 | ||
d565b0a1 HX |
4645 | rcu_read_lock(); |
4646 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4647 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4648 | continue; |
4649 | ||
86911732 | 4650 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4651 | skb_reset_mac_len(skb); |
d565b0a1 | 4652 | NAPI_GRO_CB(skb)->same_flow = 0; |
d61d072e | 4653 | NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); |
5d38a079 | 4654 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 4655 | NAPI_GRO_CB(skb)->encap_mark = 0; |
fcd91dd4 | 4656 | NAPI_GRO_CB(skb)->recursion_counter = 0; |
a0ca153f | 4657 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 4658 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 4659 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 4660 | |
662880f4 TH |
4661 | /* Setup for GRO checksum validation */ |
4662 | switch (skb->ip_summed) { | |
4663 | case CHECKSUM_COMPLETE: | |
4664 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4665 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4666 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4667 | break; | |
4668 | case CHECKSUM_UNNECESSARY: | |
4669 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4670 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4671 | break; | |
4672 | default: | |
4673 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4674 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4675 | } | |
d565b0a1 | 4676 | |
f191a1d1 | 4677 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4678 | break; |
4679 | } | |
4680 | rcu_read_unlock(); | |
4681 | ||
4682 | if (&ptype->list == head) | |
4683 | goto normal; | |
4684 | ||
25393d3f SK |
4685 | if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) { |
4686 | ret = GRO_CONSUMED; | |
4687 | goto ok; | |
4688 | } | |
4689 | ||
0da2afd5 | 4690 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4691 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4692 | |
d565b0a1 HX |
4693 | if (pp) { |
4694 | struct sk_buff *nskb = *pp; | |
4695 | ||
4696 | *pp = nskb->next; | |
4697 | nskb->next = NULL; | |
4698 | napi_gro_complete(nskb); | |
4ae5544f | 4699 | napi->gro_count--; |
d565b0a1 HX |
4700 | } |
4701 | ||
0da2afd5 | 4702 | if (same_flow) |
d565b0a1 HX |
4703 | goto ok; |
4704 | ||
600adc18 | 4705 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4706 | goto normal; |
d565b0a1 | 4707 | |
600adc18 ED |
4708 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4709 | struct sk_buff *nskb = napi->gro_list; | |
4710 | ||
4711 | /* locate the end of the list to select the 'oldest' flow */ | |
4712 | while (nskb->next) { | |
4713 | pp = &nskb->next; | |
4714 | nskb = *pp; | |
4715 | } | |
4716 | *pp = NULL; | |
4717 | nskb->next = NULL; | |
4718 | napi_gro_complete(nskb); | |
4719 | } else { | |
4720 | napi->gro_count++; | |
4721 | } | |
d565b0a1 | 4722 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4723 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4724 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4725 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4726 | skb->next = napi->gro_list; |
4727 | napi->gro_list = skb; | |
5d0d9be8 | 4728 | ret = GRO_HELD; |
d565b0a1 | 4729 | |
ad0f9904 | 4730 | pull: |
a50e233c ED |
4731 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4732 | if (grow > 0) | |
4733 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4734 | ok: |
5d0d9be8 | 4735 | return ret; |
d565b0a1 HX |
4736 | |
4737 | normal: | |
ad0f9904 HX |
4738 | ret = GRO_NORMAL; |
4739 | goto pull; | |
5d38a079 | 4740 | } |
96e93eab | 4741 | |
bf5a755f JC |
4742 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4743 | { | |
4744 | struct list_head *offload_head = &offload_base; | |
4745 | struct packet_offload *ptype; | |
4746 | ||
4747 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4748 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4749 | continue; | |
4750 | return ptype; | |
4751 | } | |
4752 | return NULL; | |
4753 | } | |
e27a2f83 | 4754 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4755 | |
4756 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4757 | { | |
4758 | struct list_head *offload_head = &offload_base; | |
4759 | struct packet_offload *ptype; | |
4760 | ||
4761 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4762 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4763 | continue; | |
4764 | return ptype; | |
4765 | } | |
4766 | return NULL; | |
4767 | } | |
e27a2f83 | 4768 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4769 | |
bb728820 | 4770 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4771 | { |
5d0d9be8 HX |
4772 | switch (ret) { |
4773 | case GRO_NORMAL: | |
ae78dbfa | 4774 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4775 | ret = GRO_DROP; |
4776 | break; | |
5d38a079 | 4777 | |
5d0d9be8 | 4778 | case GRO_DROP: |
5d38a079 HX |
4779 | kfree_skb(skb); |
4780 | break; | |
5b252f0c | 4781 | |
daa86548 | 4782 | case GRO_MERGED_FREE: |
ce87fc6c JG |
4783 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) { |
4784 | skb_dst_drop(skb); | |
f991bb9d | 4785 | secpath_reset(skb); |
d7e8883c | 4786 | kmem_cache_free(skbuff_head_cache, skb); |
ce87fc6c | 4787 | } else { |
d7e8883c | 4788 | __kfree_skb(skb); |
ce87fc6c | 4789 | } |
daa86548 ED |
4790 | break; |
4791 | ||
5b252f0c BH |
4792 | case GRO_HELD: |
4793 | case GRO_MERGED: | |
25393d3f | 4794 | case GRO_CONSUMED: |
5b252f0c | 4795 | break; |
5d38a079 HX |
4796 | } |
4797 | ||
c7c4b3b6 | 4798 | return ret; |
5d0d9be8 | 4799 | } |
5d0d9be8 | 4800 | |
c7c4b3b6 | 4801 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4802 | { |
93f93a44 | 4803 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 4804 | trace_napi_gro_receive_entry(skb); |
86911732 | 4805 | |
a50e233c ED |
4806 | skb_gro_reset_offset(skb); |
4807 | ||
89c5fa33 | 4808 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4809 | } |
4810 | EXPORT_SYMBOL(napi_gro_receive); | |
4811 | ||
d0c2b0d2 | 4812 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4813 | { |
93a35f59 ED |
4814 | if (unlikely(skb->pfmemalloc)) { |
4815 | consume_skb(skb); | |
4816 | return; | |
4817 | } | |
96e93eab | 4818 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4819 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4820 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4821 | skb->vlan_tci = 0; |
66c46d74 | 4822 | skb->dev = napi->dev; |
6d152e23 | 4823 | skb->skb_iif = 0; |
c3caf119 JC |
4824 | skb->encapsulation = 0; |
4825 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4826 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
f991bb9d | 4827 | secpath_reset(skb); |
96e93eab HX |
4828 | |
4829 | napi->skb = skb; | |
4830 | } | |
96e93eab | 4831 | |
76620aaf | 4832 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4833 | { |
5d38a079 | 4834 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4835 | |
4836 | if (!skb) { | |
fd11a83d | 4837 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
4838 | if (skb) { |
4839 | napi->skb = skb; | |
4840 | skb_mark_napi_id(skb, napi); | |
4841 | } | |
80595d59 | 4842 | } |
96e93eab HX |
4843 | return skb; |
4844 | } | |
76620aaf | 4845 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4846 | |
a50e233c ED |
4847 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4848 | struct sk_buff *skb, | |
4849 | gro_result_t ret) | |
96e93eab | 4850 | { |
5d0d9be8 HX |
4851 | switch (ret) { |
4852 | case GRO_NORMAL: | |
a50e233c ED |
4853 | case GRO_HELD: |
4854 | __skb_push(skb, ETH_HLEN); | |
4855 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4856 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4857 | ret = GRO_DROP; |
86911732 | 4858 | break; |
5d38a079 | 4859 | |
5d0d9be8 | 4860 | case GRO_DROP: |
5d0d9be8 HX |
4861 | case GRO_MERGED_FREE: |
4862 | napi_reuse_skb(napi, skb); | |
4863 | break; | |
5b252f0c BH |
4864 | |
4865 | case GRO_MERGED: | |
25393d3f | 4866 | case GRO_CONSUMED: |
5b252f0c | 4867 | break; |
5d0d9be8 | 4868 | } |
5d38a079 | 4869 | |
c7c4b3b6 | 4870 | return ret; |
5d38a079 | 4871 | } |
5d0d9be8 | 4872 | |
a50e233c ED |
4873 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
4874 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
4875 | * We copy ethernet header into skb->data to have a common layout. | |
4876 | */ | |
4adb9c4a | 4877 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4878 | { |
4879 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
4880 | const struct ethhdr *eth; |
4881 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
4882 | |
4883 | napi->skb = NULL; | |
4884 | ||
a50e233c ED |
4885 | skb_reset_mac_header(skb); |
4886 | skb_gro_reset_offset(skb); | |
4887 | ||
4888 | eth = skb_gro_header_fast(skb, 0); | |
4889 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
4890 | eth = skb_gro_header_slow(skb, hlen, 0); | |
4891 | if (unlikely(!eth)) { | |
4da46ceb AC |
4892 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
4893 | __func__, napi->dev->name); | |
a50e233c ED |
4894 | napi_reuse_skb(napi, skb); |
4895 | return NULL; | |
4896 | } | |
4897 | } else { | |
4898 | gro_pull_from_frag0(skb, hlen); | |
4899 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
4900 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 4901 | } |
a50e233c ED |
4902 | __skb_pull(skb, hlen); |
4903 | ||
4904 | /* | |
4905 | * This works because the only protocols we care about don't require | |
4906 | * special handling. | |
4907 | * We'll fix it up properly in napi_frags_finish() | |
4908 | */ | |
4909 | skb->protocol = eth->h_proto; | |
76620aaf | 4910 | |
76620aaf HX |
4911 | return skb; |
4912 | } | |
76620aaf | 4913 | |
c7c4b3b6 | 4914 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4915 | { |
76620aaf | 4916 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4917 | |
4918 | if (!skb) | |
c7c4b3b6 | 4919 | return GRO_DROP; |
5d0d9be8 | 4920 | |
ae78dbfa BH |
4921 | trace_napi_gro_frags_entry(skb); |
4922 | ||
89c5fa33 | 4923 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4924 | } |
5d38a079 HX |
4925 | EXPORT_SYMBOL(napi_gro_frags); |
4926 | ||
573e8fca TH |
4927 | /* Compute the checksum from gro_offset and return the folded value |
4928 | * after adding in any pseudo checksum. | |
4929 | */ | |
4930 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
4931 | { | |
4932 | __wsum wsum; | |
4933 | __sum16 sum; | |
4934 | ||
4935 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
4936 | ||
4937 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
4938 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
4939 | if (likely(!sum)) { | |
4940 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
4941 | !skb->csum_complete_sw) | |
4942 | netdev_rx_csum_fault(skb->dev); | |
4943 | } | |
4944 | ||
4945 | NAPI_GRO_CB(skb)->csum = wsum; | |
4946 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4947 | ||
4948 | return sum; | |
4949 | } | |
4950 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
4951 | ||
773fc8f6 | 4952 | static void net_rps_send_ipi(struct softnet_data *remsd) |
4953 | { | |
4954 | #ifdef CONFIG_RPS | |
4955 | while (remsd) { | |
4956 | struct softnet_data *next = remsd->rps_ipi_next; | |
4957 | ||
4958 | if (cpu_online(remsd->cpu)) | |
4959 | smp_call_function_single_async(remsd->cpu, &remsd->csd); | |
4960 | remsd = next; | |
4961 | } | |
4962 | #endif | |
4963 | } | |
4964 | ||
e326bed2 | 4965 | /* |
855abcf0 | 4966 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
4967 | * Note: called with local irq disabled, but exits with local irq enabled. |
4968 | */ | |
4969 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4970 | { | |
4971 | #ifdef CONFIG_RPS | |
4972 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4973 | ||
4974 | if (remsd) { | |
4975 | sd->rps_ipi_list = NULL; | |
4976 | ||
4977 | local_irq_enable(); | |
4978 | ||
4979 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
773fc8f6 | 4980 | net_rps_send_ipi(remsd); |
e326bed2 ED |
4981 | } else |
4982 | #endif | |
4983 | local_irq_enable(); | |
4984 | } | |
4985 | ||
d75b1ade ED |
4986 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
4987 | { | |
4988 | #ifdef CONFIG_RPS | |
4989 | return sd->rps_ipi_list != NULL; | |
4990 | #else | |
4991 | return false; | |
4992 | #endif | |
4993 | } | |
4994 | ||
bea3348e | 4995 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 4996 | { |
eecfd7c4 | 4997 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
4998 | bool again = true; |
4999 | int work = 0; | |
1da177e4 | 5000 | |
e326bed2 ED |
5001 | /* Check if we have pending ipi, its better to send them now, |
5002 | * not waiting net_rx_action() end. | |
5003 | */ | |
d75b1ade | 5004 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
5005 | local_irq_disable(); |
5006 | net_rps_action_and_irq_enable(sd); | |
5007 | } | |
d75b1ade | 5008 | |
3d48b53f | 5009 | napi->weight = dev_rx_weight; |
145dd5f9 | 5010 | while (again) { |
1da177e4 | 5011 | struct sk_buff *skb; |
6e7676c1 CG |
5012 | |
5013 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 5014 | rcu_read_lock(); |
6e7676c1 | 5015 | __netif_receive_skb(skb); |
2c17d27c | 5016 | rcu_read_unlock(); |
76cc8b13 | 5017 | input_queue_head_incr(sd); |
145dd5f9 | 5018 | if (++work >= quota) |
76cc8b13 | 5019 | return work; |
145dd5f9 | 5020 | |
6e7676c1 | 5021 | } |
1da177e4 | 5022 | |
145dd5f9 | 5023 | local_irq_disable(); |
e36fa2f7 | 5024 | rps_lock(sd); |
11ef7a89 | 5025 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
5026 | /* |
5027 | * Inline a custom version of __napi_complete(). | |
5028 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
5029 | * and NAPI_STATE_SCHED is the only possible flag set |
5030 | * on backlog. | |
5031 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
5032 | * and we dont need an smp_mb() memory barrier. |
5033 | */ | |
eecfd7c4 | 5034 | napi->state = 0; |
145dd5f9 PA |
5035 | again = false; |
5036 | } else { | |
5037 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
5038 | &sd->process_queue); | |
bea3348e | 5039 | } |
e36fa2f7 | 5040 | rps_unlock(sd); |
145dd5f9 | 5041 | local_irq_enable(); |
6e7676c1 | 5042 | } |
1da177e4 | 5043 | |
bea3348e SH |
5044 | return work; |
5045 | } | |
1da177e4 | 5046 | |
bea3348e SH |
5047 | /** |
5048 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 5049 | * @n: entry to schedule |
bea3348e | 5050 | * |
bc9ad166 ED |
5051 | * The entry's receive function will be scheduled to run. |
5052 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 5053 | */ |
b5606c2d | 5054 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
5055 | { |
5056 | unsigned long flags; | |
1da177e4 | 5057 | |
bea3348e | 5058 | local_irq_save(flags); |
903ceff7 | 5059 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 5060 | local_irq_restore(flags); |
1da177e4 | 5061 | } |
bea3348e SH |
5062 | EXPORT_SYMBOL(__napi_schedule); |
5063 | ||
39e6c820 ED |
5064 | /** |
5065 | * napi_schedule_prep - check if napi can be scheduled | |
5066 | * @n: napi context | |
5067 | * | |
5068 | * Test if NAPI routine is already running, and if not mark | |
5069 | * it as running. This is used as a condition variable | |
5070 | * insure only one NAPI poll instance runs. We also make | |
5071 | * sure there is no pending NAPI disable. | |
5072 | */ | |
5073 | bool napi_schedule_prep(struct napi_struct *n) | |
5074 | { | |
5075 | unsigned long val, new; | |
5076 | ||
5077 | do { | |
5078 | val = READ_ONCE(n->state); | |
5079 | if (unlikely(val & NAPIF_STATE_DISABLE)) | |
5080 | return false; | |
5081 | new = val | NAPIF_STATE_SCHED; | |
5082 | ||
5083 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
5084 | * This was suggested by Alexander Duyck, as compiler | |
5085 | * emits better code than : | |
5086 | * if (val & NAPIF_STATE_SCHED) | |
5087 | * new |= NAPIF_STATE_MISSED; | |
5088 | */ | |
5089 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
5090 | NAPIF_STATE_MISSED; | |
5091 | } while (cmpxchg(&n->state, val, new) != val); | |
5092 | ||
5093 | return !(val & NAPIF_STATE_SCHED); | |
5094 | } | |
5095 | EXPORT_SYMBOL(napi_schedule_prep); | |
5096 | ||
bc9ad166 ED |
5097 | /** |
5098 | * __napi_schedule_irqoff - schedule for receive | |
5099 | * @n: entry to schedule | |
5100 | * | |
5101 | * Variant of __napi_schedule() assuming hard irqs are masked | |
5102 | */ | |
5103 | void __napi_schedule_irqoff(struct napi_struct *n) | |
5104 | { | |
5105 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
5106 | } | |
5107 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
5108 | ||
364b6055 | 5109 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 5110 | { |
39e6c820 | 5111 | unsigned long flags, val, new; |
d565b0a1 HX |
5112 | |
5113 | /* | |
217f6974 ED |
5114 | * 1) Don't let napi dequeue from the cpu poll list |
5115 | * just in case its running on a different cpu. | |
5116 | * 2) If we are busy polling, do nothing here, we have | |
5117 | * the guarantee we will be called later. | |
d565b0a1 | 5118 | */ |
217f6974 ED |
5119 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
5120 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 5121 | return false; |
d565b0a1 | 5122 | |
3b47d303 ED |
5123 | if (n->gro_list) { |
5124 | unsigned long timeout = 0; | |
d75b1ade | 5125 | |
3b47d303 ED |
5126 | if (work_done) |
5127 | timeout = n->dev->gro_flush_timeout; | |
5128 | ||
5129 | if (timeout) | |
5130 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
5131 | HRTIMER_MODE_REL_PINNED); | |
5132 | else | |
5133 | napi_gro_flush(n, false); | |
5134 | } | |
02c1602e | 5135 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
5136 | /* If n->poll_list is not empty, we need to mask irqs */ |
5137 | local_irq_save(flags); | |
02c1602e | 5138 | list_del_init(&n->poll_list); |
d75b1ade ED |
5139 | local_irq_restore(flags); |
5140 | } | |
39e6c820 ED |
5141 | |
5142 | do { | |
5143 | val = READ_ONCE(n->state); | |
5144 | ||
5145 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); | |
5146 | ||
5147 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); | |
5148 | ||
5149 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
5150 | * because we will call napi->poll() one more time. | |
5151 | * This C code was suggested by Alexander Duyck to help gcc. | |
5152 | */ | |
5153 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
5154 | NAPIF_STATE_SCHED; | |
5155 | } while (cmpxchg(&n->state, val, new) != val); | |
5156 | ||
5157 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
5158 | __napi_schedule(n); | |
5159 | return false; | |
5160 | } | |
5161 | ||
364b6055 | 5162 | return true; |
d565b0a1 | 5163 | } |
3b47d303 | 5164 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 5165 | |
af12fa6e | 5166 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 5167 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
5168 | { |
5169 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
5170 | struct napi_struct *napi; | |
5171 | ||
5172 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
5173 | if (napi->napi_id == napi_id) | |
5174 | return napi; | |
5175 | ||
5176 | return NULL; | |
5177 | } | |
02d62e86 ED |
5178 | |
5179 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
217f6974 | 5180 | |
ce6aea93 | 5181 | #define BUSY_POLL_BUDGET 8 |
217f6974 ED |
5182 | |
5183 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) | |
5184 | { | |
5185 | int rc; | |
5186 | ||
39e6c820 ED |
5187 | /* Busy polling means there is a high chance device driver hard irq |
5188 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
5189 | * set in napi_schedule_prep(). | |
5190 | * Since we are about to call napi->poll() once more, we can safely | |
5191 | * clear NAPI_STATE_MISSED. | |
5192 | * | |
5193 | * Note: x86 could use a single "lock and ..." instruction | |
5194 | * to perform these two clear_bit() | |
5195 | */ | |
5196 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
5197 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
5198 | ||
5199 | local_bh_disable(); | |
5200 | ||
5201 | /* All we really want here is to re-enable device interrupts. | |
5202 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
5203 | */ | |
5204 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
5205 | netpoll_poll_unlock(have_poll_lock); | |
5206 | if (rc == BUSY_POLL_BUDGET) | |
5207 | __napi_schedule(napi); | |
5208 | local_bh_enable(); | |
217f6974 ED |
5209 | } |
5210 | ||
7db6b048 SS |
5211 | void napi_busy_loop(unsigned int napi_id, |
5212 | bool (*loop_end)(void *, unsigned long), | |
5213 | void *loop_end_arg) | |
02d62e86 | 5214 | { |
7db6b048 | 5215 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 5216 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 5217 | void *have_poll_lock = NULL; |
02d62e86 | 5218 | struct napi_struct *napi; |
217f6974 ED |
5219 | |
5220 | restart: | |
217f6974 | 5221 | napi_poll = NULL; |
02d62e86 | 5222 | |
2a028ecb | 5223 | rcu_read_lock(); |
02d62e86 | 5224 | |
545cd5e5 | 5225 | napi = napi_by_id(napi_id); |
02d62e86 ED |
5226 | if (!napi) |
5227 | goto out; | |
5228 | ||
217f6974 ED |
5229 | preempt_disable(); |
5230 | for (;;) { | |
2b5cd0df AD |
5231 | int work = 0; |
5232 | ||
2a028ecb | 5233 | local_bh_disable(); |
217f6974 ED |
5234 | if (!napi_poll) { |
5235 | unsigned long val = READ_ONCE(napi->state); | |
5236 | ||
5237 | /* If multiple threads are competing for this napi, | |
5238 | * we avoid dirtying napi->state as much as we can. | |
5239 | */ | |
5240 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
5241 | NAPIF_STATE_IN_BUSY_POLL)) | |
5242 | goto count; | |
5243 | if (cmpxchg(&napi->state, val, | |
5244 | val | NAPIF_STATE_IN_BUSY_POLL | | |
5245 | NAPIF_STATE_SCHED) != val) | |
5246 | goto count; | |
5247 | have_poll_lock = netpoll_poll_lock(napi); | |
5248 | napi_poll = napi->poll; | |
5249 | } | |
2b5cd0df AD |
5250 | work = napi_poll(napi, BUSY_POLL_BUDGET); |
5251 | trace_napi_poll(napi, work, BUSY_POLL_BUDGET); | |
217f6974 | 5252 | count: |
2b5cd0df | 5253 | if (work > 0) |
7db6b048 | 5254 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 5255 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
2a028ecb | 5256 | local_bh_enable(); |
02d62e86 | 5257 | |
7db6b048 | 5258 | if (!loop_end || loop_end(loop_end_arg, start_time)) |
217f6974 | 5259 | break; |
02d62e86 | 5260 | |
217f6974 ED |
5261 | if (unlikely(need_resched())) { |
5262 | if (napi_poll) | |
5263 | busy_poll_stop(napi, have_poll_lock); | |
5264 | preempt_enable(); | |
5265 | rcu_read_unlock(); | |
5266 | cond_resched(); | |
7db6b048 | 5267 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 5268 | return; |
217f6974 ED |
5269 | goto restart; |
5270 | } | |
6cdf89b1 | 5271 | cpu_relax(); |
217f6974 ED |
5272 | } |
5273 | if (napi_poll) | |
5274 | busy_poll_stop(napi, have_poll_lock); | |
5275 | preempt_enable(); | |
02d62e86 | 5276 | out: |
2a028ecb | 5277 | rcu_read_unlock(); |
02d62e86 | 5278 | } |
7db6b048 | 5279 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
5280 | |
5281 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 5282 | |
149d6ad8 | 5283 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 5284 | { |
d64b5e85 ED |
5285 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
5286 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 5287 | return; |
af12fa6e | 5288 | |
52bd2d62 | 5289 | spin_lock(&napi_hash_lock); |
af12fa6e | 5290 | |
545cd5e5 | 5291 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 5292 | do { |
545cd5e5 AD |
5293 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
5294 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
5295 | } while (napi_by_id(napi_gen_id)); |
5296 | napi->napi_id = napi_gen_id; | |
af12fa6e | 5297 | |
52bd2d62 ED |
5298 | hlist_add_head_rcu(&napi->napi_hash_node, |
5299 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 5300 | |
52bd2d62 | 5301 | spin_unlock(&napi_hash_lock); |
af12fa6e | 5302 | } |
af12fa6e ET |
5303 | |
5304 | /* Warning : caller is responsible to make sure rcu grace period | |
5305 | * is respected before freeing memory containing @napi | |
5306 | */ | |
34cbe27e | 5307 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 5308 | { |
34cbe27e ED |
5309 | bool rcu_sync_needed = false; |
5310 | ||
af12fa6e ET |
5311 | spin_lock(&napi_hash_lock); |
5312 | ||
34cbe27e ED |
5313 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
5314 | rcu_sync_needed = true; | |
af12fa6e | 5315 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 5316 | } |
af12fa6e | 5317 | spin_unlock(&napi_hash_lock); |
34cbe27e | 5318 | return rcu_sync_needed; |
af12fa6e ET |
5319 | } |
5320 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
5321 | ||
3b47d303 ED |
5322 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
5323 | { | |
5324 | struct napi_struct *napi; | |
5325 | ||
5326 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
5327 | |
5328 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
5329 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
5330 | */ | |
5331 | if (napi->gro_list && !napi_disable_pending(napi) && | |
5332 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) | |
5333 | __napi_schedule_irqoff(napi); | |
3b47d303 ED |
5334 | |
5335 | return HRTIMER_NORESTART; | |
5336 | } | |
5337 | ||
d565b0a1 HX |
5338 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
5339 | int (*poll)(struct napi_struct *, int), int weight) | |
5340 | { | |
5341 | INIT_LIST_HEAD(&napi->poll_list); | |
3b47d303 ED |
5342 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
5343 | napi->timer.function = napi_watchdog; | |
4ae5544f | 5344 | napi->gro_count = 0; |
d565b0a1 | 5345 | napi->gro_list = NULL; |
5d38a079 | 5346 | napi->skb = NULL; |
d565b0a1 | 5347 | napi->poll = poll; |
82dc3c63 ED |
5348 | if (weight > NAPI_POLL_WEIGHT) |
5349 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
5350 | weight, dev->name); | |
d565b0a1 HX |
5351 | napi->weight = weight; |
5352 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 5353 | napi->dev = dev; |
5d38a079 | 5354 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
5355 | napi->poll_owner = -1; |
5356 | #endif | |
5357 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 5358 | napi_hash_add(napi); |
d565b0a1 HX |
5359 | } |
5360 | EXPORT_SYMBOL(netif_napi_add); | |
5361 | ||
3b47d303 ED |
5362 | void napi_disable(struct napi_struct *n) |
5363 | { | |
5364 | might_sleep(); | |
5365 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
5366 | ||
5367 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
5368 | msleep(1); | |
2d8bff12 NH |
5369 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
5370 | msleep(1); | |
3b47d303 ED |
5371 | |
5372 | hrtimer_cancel(&n->timer); | |
5373 | ||
5374 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
5375 | } | |
5376 | EXPORT_SYMBOL(napi_disable); | |
5377 | ||
93d05d4a | 5378 | /* Must be called in process context */ |
d565b0a1 HX |
5379 | void netif_napi_del(struct napi_struct *napi) |
5380 | { | |
93d05d4a ED |
5381 | might_sleep(); |
5382 | if (napi_hash_del(napi)) | |
5383 | synchronize_net(); | |
d7b06636 | 5384 | list_del_init(&napi->dev_list); |
76620aaf | 5385 | napi_free_frags(napi); |
d565b0a1 | 5386 | |
289dccbe | 5387 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 5388 | napi->gro_list = NULL; |
4ae5544f | 5389 | napi->gro_count = 0; |
d565b0a1 HX |
5390 | } |
5391 | EXPORT_SYMBOL(netif_napi_del); | |
5392 | ||
726ce70e HX |
5393 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
5394 | { | |
5395 | void *have; | |
5396 | int work, weight; | |
5397 | ||
5398 | list_del_init(&n->poll_list); | |
5399 | ||
5400 | have = netpoll_poll_lock(n); | |
5401 | ||
5402 | weight = n->weight; | |
5403 | ||
5404 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
5405 | * with netpoll's poll_napi(). Only the entity which | |
5406 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
5407 | * actually make the ->poll() call. Therefore we avoid | |
5408 | * accidentally calling ->poll() when NAPI is not scheduled. | |
5409 | */ | |
5410 | work = 0; | |
5411 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
5412 | work = n->poll(n, weight); | |
1db19db7 | 5413 | trace_napi_poll(n, work, weight); |
726ce70e HX |
5414 | } |
5415 | ||
5416 | WARN_ON_ONCE(work > weight); | |
5417 | ||
5418 | if (likely(work < weight)) | |
5419 | goto out_unlock; | |
5420 | ||
5421 | /* Drivers must not modify the NAPI state if they | |
5422 | * consume the entire weight. In such cases this code | |
5423 | * still "owns" the NAPI instance and therefore can | |
5424 | * move the instance around on the list at-will. | |
5425 | */ | |
5426 | if (unlikely(napi_disable_pending(n))) { | |
5427 | napi_complete(n); | |
5428 | goto out_unlock; | |
5429 | } | |
5430 | ||
5431 | if (n->gro_list) { | |
5432 | /* flush too old packets | |
5433 | * If HZ < 1000, flush all packets. | |
5434 | */ | |
5435 | napi_gro_flush(n, HZ >= 1000); | |
5436 | } | |
5437 | ||
001ce546 HX |
5438 | /* Some drivers may have called napi_schedule |
5439 | * prior to exhausting their budget. | |
5440 | */ | |
5441 | if (unlikely(!list_empty(&n->poll_list))) { | |
5442 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
5443 | n->dev ? n->dev->name : "backlog"); | |
5444 | goto out_unlock; | |
5445 | } | |
5446 | ||
726ce70e HX |
5447 | list_add_tail(&n->poll_list, repoll); |
5448 | ||
5449 | out_unlock: | |
5450 | netpoll_poll_unlock(have); | |
5451 | ||
5452 | return work; | |
5453 | } | |
5454 | ||
0766f788 | 5455 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 5456 | { |
903ceff7 | 5457 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e MW |
5458 | unsigned long time_limit = jiffies + |
5459 | usecs_to_jiffies(netdev_budget_usecs); | |
51b0bded | 5460 | int budget = netdev_budget; |
d75b1ade ED |
5461 | LIST_HEAD(list); |
5462 | LIST_HEAD(repoll); | |
53fb95d3 | 5463 | |
1da177e4 | 5464 | local_irq_disable(); |
d75b1ade ED |
5465 | list_splice_init(&sd->poll_list, &list); |
5466 | local_irq_enable(); | |
1da177e4 | 5467 | |
ceb8d5bf | 5468 | for (;;) { |
bea3348e | 5469 | struct napi_struct *n; |
1da177e4 | 5470 | |
ceb8d5bf HX |
5471 | if (list_empty(&list)) { |
5472 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
f52dffe0 | 5473 | goto out; |
ceb8d5bf HX |
5474 | break; |
5475 | } | |
5476 | ||
6bd373eb HX |
5477 | n = list_first_entry(&list, struct napi_struct, poll_list); |
5478 | budget -= napi_poll(n, &repoll); | |
5479 | ||
d75b1ade | 5480 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
5481 | * Allow this to run for 2 jiffies since which will allow |
5482 | * an average latency of 1.5/HZ. | |
bea3348e | 5483 | */ |
ceb8d5bf HX |
5484 | if (unlikely(budget <= 0 || |
5485 | time_after_eq(jiffies, time_limit))) { | |
5486 | sd->time_squeeze++; | |
5487 | break; | |
5488 | } | |
1da177e4 | 5489 | } |
d75b1ade | 5490 | |
d75b1ade ED |
5491 | local_irq_disable(); |
5492 | ||
5493 | list_splice_tail_init(&sd->poll_list, &list); | |
5494 | list_splice_tail(&repoll, &list); | |
5495 | list_splice(&list, &sd->poll_list); | |
5496 | if (!list_empty(&sd->poll_list)) | |
5497 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
5498 | ||
e326bed2 | 5499 | net_rps_action_and_irq_enable(sd); |
f52dffe0 ED |
5500 | out: |
5501 | __kfree_skb_flush(); | |
1da177e4 LT |
5502 | } |
5503 | ||
aa9d8560 | 5504 | struct netdev_adjacent { |
9ff162a8 | 5505 | struct net_device *dev; |
5d261913 VF |
5506 | |
5507 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 5508 | bool master; |
5d261913 | 5509 | |
5d261913 VF |
5510 | /* counter for the number of times this device was added to us */ |
5511 | u16 ref_nr; | |
5512 | ||
402dae96 VF |
5513 | /* private field for the users */ |
5514 | void *private; | |
5515 | ||
9ff162a8 JP |
5516 | struct list_head list; |
5517 | struct rcu_head rcu; | |
9ff162a8 JP |
5518 | }; |
5519 | ||
6ea29da1 | 5520 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 5521 | struct list_head *adj_list) |
9ff162a8 | 5522 | { |
5d261913 | 5523 | struct netdev_adjacent *adj; |
5d261913 | 5524 | |
2f268f12 | 5525 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
5526 | if (adj->dev == adj_dev) |
5527 | return adj; | |
9ff162a8 JP |
5528 | } |
5529 | return NULL; | |
5530 | } | |
5531 | ||
f1170fd4 DA |
5532 | static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data) |
5533 | { | |
5534 | struct net_device *dev = data; | |
5535 | ||
5536 | return upper_dev == dev; | |
5537 | } | |
5538 | ||
9ff162a8 JP |
5539 | /** |
5540 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
5541 | * @dev: device | |
5542 | * @upper_dev: upper device to check | |
5543 | * | |
5544 | * Find out if a device is linked to specified upper device and return true | |
5545 | * in case it is. Note that this checks only immediate upper device, | |
5546 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
5547 | */ | |
5548 | bool netdev_has_upper_dev(struct net_device *dev, | |
5549 | struct net_device *upper_dev) | |
5550 | { | |
5551 | ASSERT_RTNL(); | |
5552 | ||
f1170fd4 DA |
5553 | return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, |
5554 | upper_dev); | |
9ff162a8 JP |
5555 | } |
5556 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
5557 | ||
1a3f060c DA |
5558 | /** |
5559 | * netdev_has_upper_dev_all - Check if device is linked to an upper device | |
5560 | * @dev: device | |
5561 | * @upper_dev: upper device to check | |
5562 | * | |
5563 | * Find out if a device is linked to specified upper device and return true | |
5564 | * in case it is. Note that this checks the entire upper device chain. | |
5565 | * The caller must hold rcu lock. | |
5566 | */ | |
5567 | ||
1a3f060c DA |
5568 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
5569 | struct net_device *upper_dev) | |
5570 | { | |
5571 | return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, | |
5572 | upper_dev); | |
5573 | } | |
5574 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
5575 | ||
9ff162a8 JP |
5576 | /** |
5577 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
5578 | * @dev: device | |
5579 | * | |
5580 | * Find out if a device is linked to an upper device and return true in case | |
5581 | * it is. The caller must hold the RTNL lock. | |
5582 | */ | |
1d143d9f | 5583 | static bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
5584 | { |
5585 | ASSERT_RTNL(); | |
5586 | ||
f1170fd4 | 5587 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 5588 | } |
9ff162a8 JP |
5589 | |
5590 | /** | |
5591 | * netdev_master_upper_dev_get - Get master upper device | |
5592 | * @dev: device | |
5593 | * | |
5594 | * Find a master upper device and return pointer to it or NULL in case | |
5595 | * it's not there. The caller must hold the RTNL lock. | |
5596 | */ | |
5597 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
5598 | { | |
aa9d8560 | 5599 | struct netdev_adjacent *upper; |
9ff162a8 JP |
5600 | |
5601 | ASSERT_RTNL(); | |
5602 | ||
2f268f12 | 5603 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
5604 | return NULL; |
5605 | ||
2f268f12 | 5606 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 5607 | struct netdev_adjacent, list); |
9ff162a8 JP |
5608 | if (likely(upper->master)) |
5609 | return upper->dev; | |
5610 | return NULL; | |
5611 | } | |
5612 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
5613 | ||
0f524a80 DA |
5614 | /** |
5615 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
5616 | * @dev: device | |
5617 | * | |
5618 | * Find out if a device is linked to a lower device and return true in case | |
5619 | * it is. The caller must hold the RTNL lock. | |
5620 | */ | |
5621 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
5622 | { | |
5623 | ASSERT_RTNL(); | |
5624 | ||
5625 | return !list_empty(&dev->adj_list.lower); | |
5626 | } | |
5627 | ||
b6ccba4c VF |
5628 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
5629 | { | |
5630 | struct netdev_adjacent *adj; | |
5631 | ||
5632 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
5633 | ||
5634 | return adj->private; | |
5635 | } | |
5636 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
5637 | ||
44a40855 VY |
5638 | /** |
5639 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
5640 | * @dev: device | |
5641 | * @iter: list_head ** of the current position | |
5642 | * | |
5643 | * Gets the next device from the dev's upper list, starting from iter | |
5644 | * position. The caller must hold RCU read lock. | |
5645 | */ | |
5646 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
5647 | struct list_head **iter) | |
5648 | { | |
5649 | struct netdev_adjacent *upper; | |
5650 | ||
5651 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5652 | ||
5653 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5654 | ||
5655 | if (&upper->list == &dev->adj_list.upper) | |
5656 | return NULL; | |
5657 | ||
5658 | *iter = &upper->list; | |
5659 | ||
5660 | return upper->dev; | |
5661 | } | |
5662 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
5663 | ||
1a3f060c DA |
5664 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
5665 | struct list_head **iter) | |
5666 | { | |
5667 | struct netdev_adjacent *upper; | |
5668 | ||
5669 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5670 | ||
5671 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5672 | ||
5673 | if (&upper->list == &dev->adj_list.upper) | |
5674 | return NULL; | |
5675 | ||
5676 | *iter = &upper->list; | |
5677 | ||
5678 | return upper->dev; | |
5679 | } | |
5680 | ||
5681 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, | |
5682 | int (*fn)(struct net_device *dev, | |
5683 | void *data), | |
5684 | void *data) | |
5685 | { | |
5686 | struct net_device *udev; | |
5687 | struct list_head *iter; | |
5688 | int ret; | |
5689 | ||
5690 | for (iter = &dev->adj_list.upper, | |
5691 | udev = netdev_next_upper_dev_rcu(dev, &iter); | |
5692 | udev; | |
5693 | udev = netdev_next_upper_dev_rcu(dev, &iter)) { | |
5694 | /* first is the upper device itself */ | |
5695 | ret = fn(udev, data); | |
5696 | if (ret) | |
5697 | return ret; | |
5698 | ||
5699 | /* then look at all of its upper devices */ | |
5700 | ret = netdev_walk_all_upper_dev_rcu(udev, fn, data); | |
5701 | if (ret) | |
5702 | return ret; | |
5703 | } | |
5704 | ||
5705 | return 0; | |
5706 | } | |
5707 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
5708 | ||
31088a11 VF |
5709 | /** |
5710 | * netdev_lower_get_next_private - Get the next ->private from the | |
5711 | * lower neighbour list | |
5712 | * @dev: device | |
5713 | * @iter: list_head ** of the current position | |
5714 | * | |
5715 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5716 | * list, starting from iter position. The caller must hold either hold the | |
5717 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 5718 | * list will remain unchanged. |
31088a11 VF |
5719 | */ |
5720 | void *netdev_lower_get_next_private(struct net_device *dev, | |
5721 | struct list_head **iter) | |
5722 | { | |
5723 | struct netdev_adjacent *lower; | |
5724 | ||
5725 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
5726 | ||
5727 | if (&lower->list == &dev->adj_list.lower) | |
5728 | return NULL; | |
5729 | ||
6859e7df | 5730 | *iter = lower->list.next; |
31088a11 VF |
5731 | |
5732 | return lower->private; | |
5733 | } | |
5734 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
5735 | ||
5736 | /** | |
5737 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
5738 | * lower neighbour list, RCU | |
5739 | * variant | |
5740 | * @dev: device | |
5741 | * @iter: list_head ** of the current position | |
5742 | * | |
5743 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5744 | * list, starting from iter position. The caller must hold RCU read lock. | |
5745 | */ | |
5746 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
5747 | struct list_head **iter) | |
5748 | { | |
5749 | struct netdev_adjacent *lower; | |
5750 | ||
5751 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5752 | ||
5753 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5754 | ||
5755 | if (&lower->list == &dev->adj_list.lower) | |
5756 | return NULL; | |
5757 | ||
6859e7df | 5758 | *iter = &lower->list; |
31088a11 VF |
5759 | |
5760 | return lower->private; | |
5761 | } | |
5762 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
5763 | ||
4085ebe8 VY |
5764 | /** |
5765 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
5766 | * list | |
5767 | * @dev: device | |
5768 | * @iter: list_head ** of the current position | |
5769 | * | |
5770 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
5771 | * list, starting from iter position. The caller must hold RTNL lock or | |
5772 | * its own locking that guarantees that the neighbour lower | |
b469139e | 5773 | * list will remain unchanged. |
4085ebe8 VY |
5774 | */ |
5775 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
5776 | { | |
5777 | struct netdev_adjacent *lower; | |
5778 | ||
cfdd28be | 5779 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
5780 | |
5781 | if (&lower->list == &dev->adj_list.lower) | |
5782 | return NULL; | |
5783 | ||
cfdd28be | 5784 | *iter = lower->list.next; |
4085ebe8 VY |
5785 | |
5786 | return lower->dev; | |
5787 | } | |
5788 | EXPORT_SYMBOL(netdev_lower_get_next); | |
5789 | ||
1a3f060c DA |
5790 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
5791 | struct list_head **iter) | |
5792 | { | |
5793 | struct netdev_adjacent *lower; | |
5794 | ||
46b5ab1a | 5795 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
5796 | |
5797 | if (&lower->list == &dev->adj_list.lower) | |
5798 | return NULL; | |
5799 | ||
46b5ab1a | 5800 | *iter = &lower->list; |
1a3f060c DA |
5801 | |
5802 | return lower->dev; | |
5803 | } | |
5804 | ||
5805 | int netdev_walk_all_lower_dev(struct net_device *dev, | |
5806 | int (*fn)(struct net_device *dev, | |
5807 | void *data), | |
5808 | void *data) | |
5809 | { | |
5810 | struct net_device *ldev; | |
5811 | struct list_head *iter; | |
5812 | int ret; | |
5813 | ||
5814 | for (iter = &dev->adj_list.lower, | |
5815 | ldev = netdev_next_lower_dev(dev, &iter); | |
5816 | ldev; | |
5817 | ldev = netdev_next_lower_dev(dev, &iter)) { | |
5818 | /* first is the lower device itself */ | |
5819 | ret = fn(ldev, data); | |
5820 | if (ret) | |
5821 | return ret; | |
5822 | ||
5823 | /* then look at all of its lower devices */ | |
5824 | ret = netdev_walk_all_lower_dev(ldev, fn, data); | |
5825 | if (ret) | |
5826 | return ret; | |
5827 | } | |
5828 | ||
5829 | return 0; | |
5830 | } | |
5831 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
5832 | ||
1a3f060c DA |
5833 | static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
5834 | struct list_head **iter) | |
5835 | { | |
5836 | struct netdev_adjacent *lower; | |
5837 | ||
5838 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5839 | if (&lower->list == &dev->adj_list.lower) | |
5840 | return NULL; | |
5841 | ||
5842 | *iter = &lower->list; | |
5843 | ||
5844 | return lower->dev; | |
5845 | } | |
5846 | ||
5847 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
5848 | int (*fn)(struct net_device *dev, | |
5849 | void *data), | |
5850 | void *data) | |
5851 | { | |
5852 | struct net_device *ldev; | |
5853 | struct list_head *iter; | |
5854 | int ret; | |
5855 | ||
5856 | for (iter = &dev->adj_list.lower, | |
5857 | ldev = netdev_next_lower_dev_rcu(dev, &iter); | |
5858 | ldev; | |
5859 | ldev = netdev_next_lower_dev_rcu(dev, &iter)) { | |
5860 | /* first is the lower device itself */ | |
5861 | ret = fn(ldev, data); | |
5862 | if (ret) | |
5863 | return ret; | |
5864 | ||
5865 | /* then look at all of its lower devices */ | |
5866 | ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data); | |
5867 | if (ret) | |
5868 | return ret; | |
5869 | } | |
5870 | ||
5871 | return 0; | |
5872 | } | |
5873 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
5874 | ||
e001bfad | 5875 | /** |
5876 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
5877 | * lower neighbour list, RCU | |
5878 | * variant | |
5879 | * @dev: device | |
5880 | * | |
5881 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
5882 | * list. The caller must hold RCU read lock. | |
5883 | */ | |
5884 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
5885 | { | |
5886 | struct netdev_adjacent *lower; | |
5887 | ||
5888 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
5889 | struct netdev_adjacent, list); | |
5890 | if (lower) | |
5891 | return lower->private; | |
5892 | return NULL; | |
5893 | } | |
5894 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
5895 | ||
9ff162a8 JP |
5896 | /** |
5897 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
5898 | * @dev: device | |
5899 | * | |
5900 | * Find a master upper device and return pointer to it or NULL in case | |
5901 | * it's not there. The caller must hold the RCU read lock. | |
5902 | */ | |
5903 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
5904 | { | |
aa9d8560 | 5905 | struct netdev_adjacent *upper; |
9ff162a8 | 5906 | |
2f268f12 | 5907 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 5908 | struct netdev_adjacent, list); |
9ff162a8 JP |
5909 | if (upper && likely(upper->master)) |
5910 | return upper->dev; | |
5911 | return NULL; | |
5912 | } | |
5913 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
5914 | ||
0a59f3a9 | 5915 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
5916 | struct net_device *adj_dev, |
5917 | struct list_head *dev_list) | |
5918 | { | |
5919 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 5920 | |
3ee32707 VF |
5921 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
5922 | "upper_%s" : "lower_%s", adj_dev->name); | |
5923 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
5924 | linkname); | |
5925 | } | |
0a59f3a9 | 5926 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
5927 | char *name, |
5928 | struct list_head *dev_list) | |
5929 | { | |
5930 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 5931 | |
3ee32707 VF |
5932 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
5933 | "upper_%s" : "lower_%s", name); | |
5934 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
5935 | } | |
5936 | ||
7ce64c79 AF |
5937 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
5938 | struct net_device *adj_dev, | |
5939 | struct list_head *dev_list) | |
5940 | { | |
5941 | return (dev_list == &dev->adj_list.upper || | |
5942 | dev_list == &dev->adj_list.lower) && | |
5943 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
5944 | } | |
3ee32707 | 5945 | |
5d261913 VF |
5946 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
5947 | struct net_device *adj_dev, | |
7863c054 | 5948 | struct list_head *dev_list, |
402dae96 | 5949 | void *private, bool master) |
5d261913 VF |
5950 | { |
5951 | struct netdev_adjacent *adj; | |
842d67a7 | 5952 | int ret; |
5d261913 | 5953 | |
6ea29da1 | 5954 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
5955 | |
5956 | if (adj) { | |
790510d9 | 5957 | adj->ref_nr += 1; |
67b62f98 DA |
5958 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
5959 | dev->name, adj_dev->name, adj->ref_nr); | |
5960 | ||
5d261913 VF |
5961 | return 0; |
5962 | } | |
5963 | ||
5964 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
5965 | if (!adj) | |
5966 | return -ENOMEM; | |
5967 | ||
5968 | adj->dev = adj_dev; | |
5969 | adj->master = master; | |
790510d9 | 5970 | adj->ref_nr = 1; |
402dae96 | 5971 | adj->private = private; |
5d261913 | 5972 | dev_hold(adj_dev); |
2f268f12 | 5973 | |
67b62f98 DA |
5974 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
5975 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 5976 | |
7ce64c79 | 5977 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 5978 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
5979 | if (ret) |
5980 | goto free_adj; | |
5981 | } | |
5982 | ||
7863c054 | 5983 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
5984 | if (master) { |
5985 | ret = sysfs_create_link(&(dev->dev.kobj), | |
5986 | &(adj_dev->dev.kobj), "master"); | |
5987 | if (ret) | |
5831d66e | 5988 | goto remove_symlinks; |
842d67a7 | 5989 | |
7863c054 | 5990 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 5991 | } else { |
7863c054 | 5992 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 5993 | } |
5d261913 VF |
5994 | |
5995 | return 0; | |
842d67a7 | 5996 | |
5831d66e | 5997 | remove_symlinks: |
7ce64c79 | 5998 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5999 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
6000 | free_adj: |
6001 | kfree(adj); | |
974daef7 | 6002 | dev_put(adj_dev); |
842d67a7 VF |
6003 | |
6004 | return ret; | |
5d261913 VF |
6005 | } |
6006 | ||
1d143d9f | 6007 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
6008 | struct net_device *adj_dev, | |
93409033 | 6009 | u16 ref_nr, |
1d143d9f | 6010 | struct list_head *dev_list) |
5d261913 VF |
6011 | { |
6012 | struct netdev_adjacent *adj; | |
6013 | ||
67b62f98 DA |
6014 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
6015 | dev->name, adj_dev->name, ref_nr); | |
6016 | ||
6ea29da1 | 6017 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 6018 | |
2f268f12 | 6019 | if (!adj) { |
67b62f98 | 6020 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 6021 | dev->name, adj_dev->name); |
67b62f98 DA |
6022 | WARN_ON(1); |
6023 | return; | |
2f268f12 | 6024 | } |
5d261913 | 6025 | |
93409033 | 6026 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
6027 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
6028 | dev->name, adj_dev->name, ref_nr, | |
6029 | adj->ref_nr - ref_nr); | |
93409033 | 6030 | adj->ref_nr -= ref_nr; |
5d261913 VF |
6031 | return; |
6032 | } | |
6033 | ||
842d67a7 VF |
6034 | if (adj->master) |
6035 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
6036 | ||
7ce64c79 | 6037 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6038 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 6039 | |
5d261913 | 6040 | list_del_rcu(&adj->list); |
67b62f98 | 6041 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 6042 | adj_dev->name, dev->name, adj_dev->name); |
5d261913 VF |
6043 | dev_put(adj_dev); |
6044 | kfree_rcu(adj, rcu); | |
6045 | } | |
6046 | ||
1d143d9f | 6047 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
6048 | struct net_device *upper_dev, | |
6049 | struct list_head *up_list, | |
6050 | struct list_head *down_list, | |
6051 | void *private, bool master) | |
5d261913 VF |
6052 | { |
6053 | int ret; | |
6054 | ||
790510d9 | 6055 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 6056 | private, master); |
5d261913 VF |
6057 | if (ret) |
6058 | return ret; | |
6059 | ||
790510d9 | 6060 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 6061 | private, false); |
5d261913 | 6062 | if (ret) { |
790510d9 | 6063 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
6064 | return ret; |
6065 | } | |
6066 | ||
6067 | return 0; | |
6068 | } | |
6069 | ||
1d143d9f | 6070 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
6071 | struct net_device *upper_dev, | |
93409033 | 6072 | u16 ref_nr, |
1d143d9f | 6073 | struct list_head *up_list, |
6074 | struct list_head *down_list) | |
5d261913 | 6075 | { |
93409033 AC |
6076 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
6077 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
6078 | } |
6079 | ||
1d143d9f | 6080 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
6081 | struct net_device *upper_dev, | |
6082 | void *private, bool master) | |
2f268f12 | 6083 | { |
f1170fd4 DA |
6084 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
6085 | &dev->adj_list.upper, | |
6086 | &upper_dev->adj_list.lower, | |
6087 | private, master); | |
5d261913 VF |
6088 | } |
6089 | ||
1d143d9f | 6090 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
6091 | struct net_device *upper_dev) | |
2f268f12 | 6092 | { |
93409033 | 6093 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
6094 | &dev->adj_list.upper, |
6095 | &upper_dev->adj_list.lower); | |
6096 | } | |
5d261913 | 6097 | |
9ff162a8 | 6098 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 6099 | struct net_device *upper_dev, bool master, |
29bf24af | 6100 | void *upper_priv, void *upper_info) |
9ff162a8 | 6101 | { |
0e4ead9d | 6102 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 | 6103 | int ret = 0; |
9ff162a8 JP |
6104 | |
6105 | ASSERT_RTNL(); | |
6106 | ||
6107 | if (dev == upper_dev) | |
6108 | return -EBUSY; | |
6109 | ||
6110 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
f1170fd4 | 6111 | if (netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
6112 | return -EBUSY; |
6113 | ||
f1170fd4 | 6114 | if (netdev_has_upper_dev(dev, upper_dev)) |
9ff162a8 JP |
6115 | return -EEXIST; |
6116 | ||
6117 | if (master && netdev_master_upper_dev_get(dev)) | |
6118 | return -EBUSY; | |
6119 | ||
0e4ead9d JP |
6120 | changeupper_info.upper_dev = upper_dev; |
6121 | changeupper_info.master = master; | |
6122 | changeupper_info.linking = true; | |
29bf24af | 6123 | changeupper_info.upper_info = upper_info; |
0e4ead9d | 6124 | |
573c7ba0 JP |
6125 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
6126 | &changeupper_info.info); | |
6127 | ret = notifier_to_errno(ret); | |
6128 | if (ret) | |
6129 | return ret; | |
6130 | ||
6dffb044 | 6131 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 6132 | master); |
5d261913 VF |
6133 | if (ret) |
6134 | return ret; | |
9ff162a8 | 6135 | |
b03804e7 IS |
6136 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
6137 | &changeupper_info.info); | |
6138 | ret = notifier_to_errno(ret); | |
6139 | if (ret) | |
f1170fd4 | 6140 | goto rollback; |
b03804e7 | 6141 | |
9ff162a8 | 6142 | return 0; |
5d261913 | 6143 | |
f1170fd4 | 6144 | rollback: |
2f268f12 | 6145 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
6146 | |
6147 | return ret; | |
9ff162a8 JP |
6148 | } |
6149 | ||
6150 | /** | |
6151 | * netdev_upper_dev_link - Add a link to the upper device | |
6152 | * @dev: device | |
6153 | * @upper_dev: new upper device | |
6154 | * | |
6155 | * Adds a link to device which is upper to this one. The caller must hold | |
6156 | * the RTNL lock. On a failure a negative errno code is returned. | |
6157 | * On success the reference counts are adjusted and the function | |
6158 | * returns zero. | |
6159 | */ | |
6160 | int netdev_upper_dev_link(struct net_device *dev, | |
6161 | struct net_device *upper_dev) | |
6162 | { | |
29bf24af | 6163 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL); |
9ff162a8 JP |
6164 | } |
6165 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
6166 | ||
6167 | /** | |
6168 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
6169 | * @dev: device | |
6170 | * @upper_dev: new upper device | |
6dffb044 | 6171 | * @upper_priv: upper device private |
29bf24af | 6172 | * @upper_info: upper info to be passed down via notifier |
9ff162a8 JP |
6173 | * |
6174 | * Adds a link to device which is upper to this one. In this case, only | |
6175 | * one master upper device can be linked, although other non-master devices | |
6176 | * might be linked as well. The caller must hold the RTNL lock. | |
6177 | * On a failure a negative errno code is returned. On success the reference | |
6178 | * counts are adjusted and the function returns zero. | |
6179 | */ | |
6180 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 6181 | struct net_device *upper_dev, |
29bf24af | 6182 | void *upper_priv, void *upper_info) |
9ff162a8 | 6183 | { |
29bf24af JP |
6184 | return __netdev_upper_dev_link(dev, upper_dev, true, |
6185 | upper_priv, upper_info); | |
9ff162a8 JP |
6186 | } |
6187 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
6188 | ||
6189 | /** | |
6190 | * netdev_upper_dev_unlink - Removes a link to upper device | |
6191 | * @dev: device | |
6192 | * @upper_dev: new upper device | |
6193 | * | |
6194 | * Removes a link to device which is upper to this one. The caller must hold | |
6195 | * the RTNL lock. | |
6196 | */ | |
6197 | void netdev_upper_dev_unlink(struct net_device *dev, | |
6198 | struct net_device *upper_dev) | |
6199 | { | |
0e4ead9d | 6200 | struct netdev_notifier_changeupper_info changeupper_info; |
f4563a75 | 6201 | |
9ff162a8 JP |
6202 | ASSERT_RTNL(); |
6203 | ||
0e4ead9d JP |
6204 | changeupper_info.upper_dev = upper_dev; |
6205 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; | |
6206 | changeupper_info.linking = false; | |
6207 | ||
573c7ba0 JP |
6208 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
6209 | &changeupper_info.info); | |
6210 | ||
2f268f12 | 6211 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 6212 | |
0e4ead9d JP |
6213 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
6214 | &changeupper_info.info); | |
9ff162a8 JP |
6215 | } |
6216 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
6217 | ||
61bd3857 MS |
6218 | /** |
6219 | * netdev_bonding_info_change - Dispatch event about slave change | |
6220 | * @dev: device | |
4a26e453 | 6221 | * @bonding_info: info to dispatch |
61bd3857 MS |
6222 | * |
6223 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
6224 | * The caller must hold the RTNL lock. | |
6225 | */ | |
6226 | void netdev_bonding_info_change(struct net_device *dev, | |
6227 | struct netdev_bonding_info *bonding_info) | |
6228 | { | |
6229 | struct netdev_notifier_bonding_info info; | |
6230 | ||
6231 | memcpy(&info.bonding_info, bonding_info, | |
6232 | sizeof(struct netdev_bonding_info)); | |
6233 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, | |
6234 | &info.info); | |
6235 | } | |
6236 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
6237 | ||
2ce1ee17 | 6238 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
6239 | { |
6240 | struct netdev_adjacent *iter; | |
6241 | ||
6242 | struct net *net = dev_net(dev); | |
6243 | ||
6244 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 6245 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6246 | continue; |
6247 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6248 | &iter->dev->adj_list.lower); | |
6249 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
6250 | &dev->adj_list.upper); | |
6251 | } | |
6252 | ||
6253 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 6254 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6255 | continue; |
6256 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6257 | &iter->dev->adj_list.upper); | |
6258 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
6259 | &dev->adj_list.lower); | |
6260 | } | |
6261 | } | |
6262 | ||
2ce1ee17 | 6263 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
6264 | { |
6265 | struct netdev_adjacent *iter; | |
6266 | ||
6267 | struct net *net = dev_net(dev); | |
6268 | ||
6269 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 6270 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6271 | continue; |
6272 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6273 | &iter->dev->adj_list.lower); | |
6274 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6275 | &dev->adj_list.upper); | |
6276 | } | |
6277 | ||
6278 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 6279 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6280 | continue; |
6281 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6282 | &iter->dev->adj_list.upper); | |
6283 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6284 | &dev->adj_list.lower); | |
6285 | } | |
6286 | } | |
6287 | ||
5bb025fa | 6288 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 6289 | { |
5bb025fa | 6290 | struct netdev_adjacent *iter; |
402dae96 | 6291 | |
4c75431a AF |
6292 | struct net *net = dev_net(dev); |
6293 | ||
5bb025fa | 6294 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 6295 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6296 | continue; |
5bb025fa VF |
6297 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6298 | &iter->dev->adj_list.lower); | |
6299 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6300 | &iter->dev->adj_list.lower); | |
6301 | } | |
402dae96 | 6302 | |
5bb025fa | 6303 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 6304 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6305 | continue; |
5bb025fa VF |
6306 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6307 | &iter->dev->adj_list.upper); | |
6308 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6309 | &iter->dev->adj_list.upper); | |
6310 | } | |
402dae96 | 6311 | } |
402dae96 VF |
6312 | |
6313 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
6314 | struct net_device *lower_dev) | |
6315 | { | |
6316 | struct netdev_adjacent *lower; | |
6317 | ||
6318 | if (!lower_dev) | |
6319 | return NULL; | |
6ea29da1 | 6320 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
6321 | if (!lower) |
6322 | return NULL; | |
6323 | ||
6324 | return lower->private; | |
6325 | } | |
6326 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
6327 | ||
4085ebe8 | 6328 | |
952fcfd0 | 6329 | int dev_get_nest_level(struct net_device *dev) |
4085ebe8 VY |
6330 | { |
6331 | struct net_device *lower = NULL; | |
6332 | struct list_head *iter; | |
6333 | int max_nest = -1; | |
6334 | int nest; | |
6335 | ||
6336 | ASSERT_RTNL(); | |
6337 | ||
6338 | netdev_for_each_lower_dev(dev, lower, iter) { | |
952fcfd0 | 6339 | nest = dev_get_nest_level(lower); |
4085ebe8 VY |
6340 | if (max_nest < nest) |
6341 | max_nest = nest; | |
6342 | } | |
6343 | ||
952fcfd0 | 6344 | return max_nest + 1; |
4085ebe8 VY |
6345 | } |
6346 | EXPORT_SYMBOL(dev_get_nest_level); | |
6347 | ||
04d48266 JP |
6348 | /** |
6349 | * netdev_lower_change - Dispatch event about lower device state change | |
6350 | * @lower_dev: device | |
6351 | * @lower_state_info: state to dispatch | |
6352 | * | |
6353 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
6354 | * The caller must hold the RTNL lock. | |
6355 | */ | |
6356 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
6357 | void *lower_state_info) | |
6358 | { | |
6359 | struct netdev_notifier_changelowerstate_info changelowerstate_info; | |
6360 | ||
6361 | ASSERT_RTNL(); | |
6362 | changelowerstate_info.lower_state_info = lower_state_info; | |
6363 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev, | |
6364 | &changelowerstate_info.info); | |
6365 | } | |
6366 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
6367 | ||
b6c40d68 PM |
6368 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
6369 | { | |
d314774c SH |
6370 | const struct net_device_ops *ops = dev->netdev_ops; |
6371 | ||
d2615bf4 | 6372 | if (ops->ndo_change_rx_flags) |
d314774c | 6373 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
6374 | } |
6375 | ||
991fb3f7 | 6376 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6377 | { |
b536db93 | 6378 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
6379 | kuid_t uid; |
6380 | kgid_t gid; | |
1da177e4 | 6381 | |
24023451 PM |
6382 | ASSERT_RTNL(); |
6383 | ||
dad9b335 WC |
6384 | dev->flags |= IFF_PROMISC; |
6385 | dev->promiscuity += inc; | |
6386 | if (dev->promiscuity == 0) { | |
6387 | /* | |
6388 | * Avoid overflow. | |
6389 | * If inc causes overflow, untouch promisc and return error. | |
6390 | */ | |
6391 | if (inc < 0) | |
6392 | dev->flags &= ~IFF_PROMISC; | |
6393 | else { | |
6394 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
6395 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
6396 | dev->name); | |
dad9b335 WC |
6397 | return -EOVERFLOW; |
6398 | } | |
6399 | } | |
52609c0b | 6400 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
6401 | pr_info("device %s %s promiscuous mode\n", |
6402 | dev->name, | |
6403 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
6404 | if (audit_enabled) { |
6405 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
6406 | audit_log(current->audit_context, GFP_ATOMIC, |
6407 | AUDIT_ANOM_PROMISCUOUS, | |
6408 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
6409 | dev->name, (dev->flags & IFF_PROMISC), | |
6410 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 6411 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
6412 | from_kuid(&init_user_ns, uid), |
6413 | from_kgid(&init_user_ns, gid), | |
7759db82 | 6414 | audit_get_sessionid(current)); |
8192b0c4 | 6415 | } |
24023451 | 6416 | |
b6c40d68 | 6417 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 6418 | } |
991fb3f7 ND |
6419 | if (notify) |
6420 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 6421 | return 0; |
1da177e4 LT |
6422 | } |
6423 | ||
4417da66 PM |
6424 | /** |
6425 | * dev_set_promiscuity - update promiscuity count on a device | |
6426 | * @dev: device | |
6427 | * @inc: modifier | |
6428 | * | |
6429 | * Add or remove promiscuity from a device. While the count in the device | |
6430 | * remains above zero the interface remains promiscuous. Once it hits zero | |
6431 | * the device reverts back to normal filtering operation. A negative inc | |
6432 | * value is used to drop promiscuity on the device. | |
dad9b335 | 6433 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 6434 | */ |
dad9b335 | 6435 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 6436 | { |
b536db93 | 6437 | unsigned int old_flags = dev->flags; |
dad9b335 | 6438 | int err; |
4417da66 | 6439 | |
991fb3f7 | 6440 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 6441 | if (err < 0) |
dad9b335 | 6442 | return err; |
4417da66 PM |
6443 | if (dev->flags != old_flags) |
6444 | dev_set_rx_mode(dev); | |
dad9b335 | 6445 | return err; |
4417da66 | 6446 | } |
d1b19dff | 6447 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 6448 | |
991fb3f7 | 6449 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6450 | { |
991fb3f7 | 6451 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 6452 | |
24023451 PM |
6453 | ASSERT_RTNL(); |
6454 | ||
1da177e4 | 6455 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
6456 | dev->allmulti += inc; |
6457 | if (dev->allmulti == 0) { | |
6458 | /* | |
6459 | * Avoid overflow. | |
6460 | * If inc causes overflow, untouch allmulti and return error. | |
6461 | */ | |
6462 | if (inc < 0) | |
6463 | dev->flags &= ~IFF_ALLMULTI; | |
6464 | else { | |
6465 | dev->allmulti -= inc; | |
7b6cd1ce JP |
6466 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
6467 | dev->name); | |
dad9b335 WC |
6468 | return -EOVERFLOW; |
6469 | } | |
6470 | } | |
24023451 | 6471 | if (dev->flags ^ old_flags) { |
b6c40d68 | 6472 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 6473 | dev_set_rx_mode(dev); |
991fb3f7 ND |
6474 | if (notify) |
6475 | __dev_notify_flags(dev, old_flags, | |
6476 | dev->gflags ^ old_gflags); | |
24023451 | 6477 | } |
dad9b335 | 6478 | return 0; |
4417da66 | 6479 | } |
991fb3f7 ND |
6480 | |
6481 | /** | |
6482 | * dev_set_allmulti - update allmulti count on a device | |
6483 | * @dev: device | |
6484 | * @inc: modifier | |
6485 | * | |
6486 | * Add or remove reception of all multicast frames to a device. While the | |
6487 | * count in the device remains above zero the interface remains listening | |
6488 | * to all interfaces. Once it hits zero the device reverts back to normal | |
6489 | * filtering operation. A negative @inc value is used to drop the counter | |
6490 | * when releasing a resource needing all multicasts. | |
6491 | * Return 0 if successful or a negative errno code on error. | |
6492 | */ | |
6493 | ||
6494 | int dev_set_allmulti(struct net_device *dev, int inc) | |
6495 | { | |
6496 | return __dev_set_allmulti(dev, inc, true); | |
6497 | } | |
d1b19dff | 6498 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
6499 | |
6500 | /* | |
6501 | * Upload unicast and multicast address lists to device and | |
6502 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 6503 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
6504 | * are present. |
6505 | */ | |
6506 | void __dev_set_rx_mode(struct net_device *dev) | |
6507 | { | |
d314774c SH |
6508 | const struct net_device_ops *ops = dev->netdev_ops; |
6509 | ||
4417da66 PM |
6510 | /* dev_open will call this function so the list will stay sane. */ |
6511 | if (!(dev->flags&IFF_UP)) | |
6512 | return; | |
6513 | ||
6514 | if (!netif_device_present(dev)) | |
40b77c94 | 6515 | return; |
4417da66 | 6516 | |
01789349 | 6517 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
6518 | /* Unicast addresses changes may only happen under the rtnl, |
6519 | * therefore calling __dev_set_promiscuity here is safe. | |
6520 | */ | |
32e7bfc4 | 6521 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 6522 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 6523 | dev->uc_promisc = true; |
32e7bfc4 | 6524 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 6525 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 6526 | dev->uc_promisc = false; |
4417da66 | 6527 | } |
4417da66 | 6528 | } |
01789349 JP |
6529 | |
6530 | if (ops->ndo_set_rx_mode) | |
6531 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
6532 | } |
6533 | ||
6534 | void dev_set_rx_mode(struct net_device *dev) | |
6535 | { | |
b9e40857 | 6536 | netif_addr_lock_bh(dev); |
4417da66 | 6537 | __dev_set_rx_mode(dev); |
b9e40857 | 6538 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
6539 | } |
6540 | ||
f0db275a SH |
6541 | /** |
6542 | * dev_get_flags - get flags reported to userspace | |
6543 | * @dev: device | |
6544 | * | |
6545 | * Get the combination of flag bits exported through APIs to userspace. | |
6546 | */ | |
95c96174 | 6547 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 6548 | { |
95c96174 | 6549 | unsigned int flags; |
1da177e4 LT |
6550 | |
6551 | flags = (dev->flags & ~(IFF_PROMISC | | |
6552 | IFF_ALLMULTI | | |
b00055aa SR |
6553 | IFF_RUNNING | |
6554 | IFF_LOWER_UP | | |
6555 | IFF_DORMANT)) | | |
1da177e4 LT |
6556 | (dev->gflags & (IFF_PROMISC | |
6557 | IFF_ALLMULTI)); | |
6558 | ||
b00055aa SR |
6559 | if (netif_running(dev)) { |
6560 | if (netif_oper_up(dev)) | |
6561 | flags |= IFF_RUNNING; | |
6562 | if (netif_carrier_ok(dev)) | |
6563 | flags |= IFF_LOWER_UP; | |
6564 | if (netif_dormant(dev)) | |
6565 | flags |= IFF_DORMANT; | |
6566 | } | |
1da177e4 LT |
6567 | |
6568 | return flags; | |
6569 | } | |
d1b19dff | 6570 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 6571 | |
bd380811 | 6572 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 6573 | { |
b536db93 | 6574 | unsigned int old_flags = dev->flags; |
bd380811 | 6575 | int ret; |
1da177e4 | 6576 | |
24023451 PM |
6577 | ASSERT_RTNL(); |
6578 | ||
1da177e4 LT |
6579 | /* |
6580 | * Set the flags on our device. | |
6581 | */ | |
6582 | ||
6583 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
6584 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
6585 | IFF_AUTOMEDIA)) | | |
6586 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
6587 | IFF_ALLMULTI)); | |
6588 | ||
6589 | /* | |
6590 | * Load in the correct multicast list now the flags have changed. | |
6591 | */ | |
6592 | ||
b6c40d68 PM |
6593 | if ((old_flags ^ flags) & IFF_MULTICAST) |
6594 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 6595 | |
4417da66 | 6596 | dev_set_rx_mode(dev); |
1da177e4 LT |
6597 | |
6598 | /* | |
6599 | * Have we downed the interface. We handle IFF_UP ourselves | |
6600 | * according to user attempts to set it, rather than blindly | |
6601 | * setting it. | |
6602 | */ | |
6603 | ||
6604 | ret = 0; | |
d215d10f | 6605 | if ((old_flags ^ flags) & IFF_UP) |
bd380811 | 6606 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 | 6607 | |
1da177e4 | 6608 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 6609 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 6610 | unsigned int old_flags = dev->flags; |
d1b19dff | 6611 | |
1da177e4 | 6612 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
6613 | |
6614 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
6615 | if (dev->flags != old_flags) | |
6616 | dev_set_rx_mode(dev); | |
1da177e4 LT |
6617 | } |
6618 | ||
6619 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 6620 | * is important. Some (broken) drivers set IFF_PROMISC, when |
6621 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
6622 | */ |
6623 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
6624 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
6625 | ||
1da177e4 | 6626 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 6627 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
6628 | } |
6629 | ||
bd380811 PM |
6630 | return ret; |
6631 | } | |
6632 | ||
a528c219 ND |
6633 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
6634 | unsigned int gchanges) | |
bd380811 PM |
6635 | { |
6636 | unsigned int changes = dev->flags ^ old_flags; | |
6637 | ||
a528c219 | 6638 | if (gchanges) |
7f294054 | 6639 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 6640 | |
bd380811 PM |
6641 | if (changes & IFF_UP) { |
6642 | if (dev->flags & IFF_UP) | |
6643 | call_netdevice_notifiers(NETDEV_UP, dev); | |
6644 | else | |
6645 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
6646 | } | |
6647 | ||
6648 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
6649 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
6650 | struct netdev_notifier_change_info change_info; | |
6651 | ||
6652 | change_info.flags_changed = changes; | |
6653 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
6654 | &change_info.info); | |
6655 | } | |
bd380811 PM |
6656 | } |
6657 | ||
6658 | /** | |
6659 | * dev_change_flags - change device settings | |
6660 | * @dev: device | |
6661 | * @flags: device state flags | |
6662 | * | |
6663 | * Change settings on device based state flags. The flags are | |
6664 | * in the userspace exported format. | |
6665 | */ | |
b536db93 | 6666 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 6667 | { |
b536db93 | 6668 | int ret; |
991fb3f7 | 6669 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
6670 | |
6671 | ret = __dev_change_flags(dev, flags); | |
6672 | if (ret < 0) | |
6673 | return ret; | |
6674 | ||
991fb3f7 | 6675 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 6676 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
6677 | return ret; |
6678 | } | |
d1b19dff | 6679 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 6680 | |
2315dc91 VF |
6681 | static int __dev_set_mtu(struct net_device *dev, int new_mtu) |
6682 | { | |
6683 | const struct net_device_ops *ops = dev->netdev_ops; | |
6684 | ||
6685 | if (ops->ndo_change_mtu) | |
6686 | return ops->ndo_change_mtu(dev, new_mtu); | |
6687 | ||
6688 | dev->mtu = new_mtu; | |
6689 | return 0; | |
6690 | } | |
6691 | ||
f0db275a SH |
6692 | /** |
6693 | * dev_set_mtu - Change maximum transfer unit | |
6694 | * @dev: device | |
6695 | * @new_mtu: new transfer unit | |
6696 | * | |
6697 | * Change the maximum transfer size of the network device. | |
6698 | */ | |
1da177e4 LT |
6699 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
6700 | { | |
2315dc91 | 6701 | int err, orig_mtu; |
1da177e4 LT |
6702 | |
6703 | if (new_mtu == dev->mtu) | |
6704 | return 0; | |
6705 | ||
61e84623 JW |
6706 | /* MTU must be positive, and in range */ |
6707 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
6708 | net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n", | |
6709 | dev->name, new_mtu, dev->min_mtu); | |
1da177e4 | 6710 | return -EINVAL; |
61e84623 JW |
6711 | } |
6712 | ||
6713 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
6714 | net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n", | |
a0e65de7 | 6715 | dev->name, new_mtu, dev->max_mtu); |
61e84623 JW |
6716 | return -EINVAL; |
6717 | } | |
1da177e4 LT |
6718 | |
6719 | if (!netif_device_present(dev)) | |
6720 | return -ENODEV; | |
6721 | ||
1d486bfb VF |
6722 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
6723 | err = notifier_to_errno(err); | |
6724 | if (err) | |
6725 | return err; | |
d314774c | 6726 | |
2315dc91 VF |
6727 | orig_mtu = dev->mtu; |
6728 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 6729 | |
2315dc91 VF |
6730 | if (!err) { |
6731 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6732 | err = notifier_to_errno(err); | |
6733 | if (err) { | |
6734 | /* setting mtu back and notifying everyone again, | |
6735 | * so that they have a chance to revert changes. | |
6736 | */ | |
6737 | __dev_set_mtu(dev, orig_mtu); | |
6738 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6739 | } | |
6740 | } | |
1da177e4 LT |
6741 | return err; |
6742 | } | |
d1b19dff | 6743 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 6744 | |
cbda10fa VD |
6745 | /** |
6746 | * dev_set_group - Change group this device belongs to | |
6747 | * @dev: device | |
6748 | * @new_group: group this device should belong to | |
6749 | */ | |
6750 | void dev_set_group(struct net_device *dev, int new_group) | |
6751 | { | |
6752 | dev->group = new_group; | |
6753 | } | |
6754 | EXPORT_SYMBOL(dev_set_group); | |
6755 | ||
f0db275a SH |
6756 | /** |
6757 | * dev_set_mac_address - Change Media Access Control Address | |
6758 | * @dev: device | |
6759 | * @sa: new address | |
6760 | * | |
6761 | * Change the hardware (MAC) address of the device | |
6762 | */ | |
1da177e4 LT |
6763 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
6764 | { | |
d314774c | 6765 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
6766 | int err; |
6767 | ||
d314774c | 6768 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
6769 | return -EOPNOTSUPP; |
6770 | if (sa->sa_family != dev->type) | |
6771 | return -EINVAL; | |
6772 | if (!netif_device_present(dev)) | |
6773 | return -ENODEV; | |
d314774c | 6774 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
6775 | if (err) |
6776 | return err; | |
fbdeca2d | 6777 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 6778 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 6779 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 6780 | return 0; |
1da177e4 | 6781 | } |
d1b19dff | 6782 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 6783 | |
4bf84c35 JP |
6784 | /** |
6785 | * dev_change_carrier - Change device carrier | |
6786 | * @dev: device | |
691b3b7e | 6787 | * @new_carrier: new value |
4bf84c35 JP |
6788 | * |
6789 | * Change device carrier | |
6790 | */ | |
6791 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
6792 | { | |
6793 | const struct net_device_ops *ops = dev->netdev_ops; | |
6794 | ||
6795 | if (!ops->ndo_change_carrier) | |
6796 | return -EOPNOTSUPP; | |
6797 | if (!netif_device_present(dev)) | |
6798 | return -ENODEV; | |
6799 | return ops->ndo_change_carrier(dev, new_carrier); | |
6800 | } | |
6801 | EXPORT_SYMBOL(dev_change_carrier); | |
6802 | ||
66b52b0d JP |
6803 | /** |
6804 | * dev_get_phys_port_id - Get device physical port ID | |
6805 | * @dev: device | |
6806 | * @ppid: port ID | |
6807 | * | |
6808 | * Get device physical port ID | |
6809 | */ | |
6810 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 6811 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
6812 | { |
6813 | const struct net_device_ops *ops = dev->netdev_ops; | |
6814 | ||
6815 | if (!ops->ndo_get_phys_port_id) | |
6816 | return -EOPNOTSUPP; | |
6817 | return ops->ndo_get_phys_port_id(dev, ppid); | |
6818 | } | |
6819 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
6820 | ||
db24a904 DA |
6821 | /** |
6822 | * dev_get_phys_port_name - Get device physical port name | |
6823 | * @dev: device | |
6824 | * @name: port name | |
ed49e650 | 6825 | * @len: limit of bytes to copy to name |
db24a904 DA |
6826 | * |
6827 | * Get device physical port name | |
6828 | */ | |
6829 | int dev_get_phys_port_name(struct net_device *dev, | |
6830 | char *name, size_t len) | |
6831 | { | |
6832 | const struct net_device_ops *ops = dev->netdev_ops; | |
6833 | ||
6834 | if (!ops->ndo_get_phys_port_name) | |
6835 | return -EOPNOTSUPP; | |
6836 | return ops->ndo_get_phys_port_name(dev, name, len); | |
6837 | } | |
6838 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
6839 | ||
d746d707 AK |
6840 | /** |
6841 | * dev_change_proto_down - update protocol port state information | |
6842 | * @dev: device | |
6843 | * @proto_down: new value | |
6844 | * | |
6845 | * This info can be used by switch drivers to set the phys state of the | |
6846 | * port. | |
6847 | */ | |
6848 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
6849 | { | |
6850 | const struct net_device_ops *ops = dev->netdev_ops; | |
6851 | ||
6852 | if (!ops->ndo_change_proto_down) | |
6853 | return -EOPNOTSUPP; | |
6854 | if (!netif_device_present(dev)) | |
6855 | return -ENODEV; | |
6856 | return ops->ndo_change_proto_down(dev, proto_down); | |
6857 | } | |
6858 | EXPORT_SYMBOL(dev_change_proto_down); | |
6859 | ||
d67b9cd2 DB |
6860 | bool __dev_xdp_attached(struct net_device *dev, xdp_op_t xdp_op) |
6861 | { | |
6862 | struct netdev_xdp xdp; | |
6863 | ||
6864 | memset(&xdp, 0, sizeof(xdp)); | |
6865 | xdp.command = XDP_QUERY_PROG; | |
6866 | ||
6867 | /* Query must always succeed. */ | |
6868 | WARN_ON(xdp_op(dev, &xdp) < 0); | |
6869 | return xdp.prog_attached; | |
6870 | } | |
6871 | ||
6872 | static int dev_xdp_install(struct net_device *dev, xdp_op_t xdp_op, | |
6873 | struct netlink_ext_ack *extack, | |
6874 | struct bpf_prog *prog) | |
6875 | { | |
6876 | struct netdev_xdp xdp; | |
6877 | ||
6878 | memset(&xdp, 0, sizeof(xdp)); | |
6879 | xdp.command = XDP_SETUP_PROG; | |
6880 | xdp.extack = extack; | |
6881 | xdp.prog = prog; | |
6882 | ||
6883 | return xdp_op(dev, &xdp); | |
6884 | } | |
6885 | ||
a7862b45 BB |
6886 | /** |
6887 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
6888 | * @dev: device | |
b5d60989 | 6889 | * @extack: netlink extended ack |
a7862b45 | 6890 | * @fd: new program fd or negative value to clear |
85de8576 | 6891 | * @flags: xdp-related flags |
a7862b45 BB |
6892 | * |
6893 | * Set or clear a bpf program for a device | |
6894 | */ | |
ddf9f970 JK |
6895 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, |
6896 | int fd, u32 flags) | |
a7862b45 BB |
6897 | { |
6898 | const struct net_device_ops *ops = dev->netdev_ops; | |
6899 | struct bpf_prog *prog = NULL; | |
d67b9cd2 | 6900 | xdp_op_t xdp_op, xdp_chk; |
a7862b45 BB |
6901 | int err; |
6902 | ||
85de8576 DB |
6903 | ASSERT_RTNL(); |
6904 | ||
d67b9cd2 | 6905 | xdp_op = xdp_chk = ops->ndo_xdp; |
0489df9a DB |
6906 | if (!xdp_op && (flags & XDP_FLAGS_DRV_MODE)) |
6907 | return -EOPNOTSUPP; | |
b5cdae32 DM |
6908 | if (!xdp_op || (flags & XDP_FLAGS_SKB_MODE)) |
6909 | xdp_op = generic_xdp_install; | |
d67b9cd2 DB |
6910 | if (xdp_op == xdp_chk) |
6911 | xdp_chk = generic_xdp_install; | |
b5cdae32 | 6912 | |
a7862b45 | 6913 | if (fd >= 0) { |
d67b9cd2 DB |
6914 | if (xdp_chk && __dev_xdp_attached(dev, xdp_chk)) |
6915 | return -EEXIST; | |
6916 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && | |
6917 | __dev_xdp_attached(dev, xdp_op)) | |
6918 | return -EBUSY; | |
85de8576 | 6919 | |
a7862b45 BB |
6920 | prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP); |
6921 | if (IS_ERR(prog)) | |
6922 | return PTR_ERR(prog); | |
6923 | } | |
6924 | ||
d67b9cd2 | 6925 | err = dev_xdp_install(dev, xdp_op, extack, prog); |
a7862b45 BB |
6926 | if (err < 0 && prog) |
6927 | bpf_prog_put(prog); | |
6928 | ||
6929 | return err; | |
6930 | } | |
a7862b45 | 6931 | |
1da177e4 LT |
6932 | /** |
6933 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 6934 | * @net: the applicable net namespace |
1da177e4 LT |
6935 | * |
6936 | * Returns a suitable unique value for a new device interface | |
6937 | * number. The caller must hold the rtnl semaphore or the | |
6938 | * dev_base_lock to be sure it remains unique. | |
6939 | */ | |
881d966b | 6940 | static int dev_new_index(struct net *net) |
1da177e4 | 6941 | { |
aa79e66e | 6942 | int ifindex = net->ifindex; |
f4563a75 | 6943 | |
1da177e4 LT |
6944 | for (;;) { |
6945 | if (++ifindex <= 0) | |
6946 | ifindex = 1; | |
881d966b | 6947 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 6948 | return net->ifindex = ifindex; |
1da177e4 LT |
6949 | } |
6950 | } | |
6951 | ||
1da177e4 | 6952 | /* Delayed registration/unregisteration */ |
3b5b34fd | 6953 | static LIST_HEAD(net_todo_list); |
200b916f | 6954 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 6955 | |
6f05f629 | 6956 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 6957 | { |
1da177e4 | 6958 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 6959 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
6960 | } |
6961 | ||
9b5e383c | 6962 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 6963 | { |
e93737b0 | 6964 | struct net_device *dev, *tmp; |
5cde2829 | 6965 | LIST_HEAD(close_head); |
9b5e383c | 6966 | |
93ee31f1 DL |
6967 | BUG_ON(dev_boot_phase); |
6968 | ASSERT_RTNL(); | |
6969 | ||
e93737b0 | 6970 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 6971 | /* Some devices call without registering |
e93737b0 KK |
6972 | * for initialization unwind. Remove those |
6973 | * devices and proceed with the remaining. | |
9b5e383c ED |
6974 | */ |
6975 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
6976 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
6977 | dev->name, dev); | |
93ee31f1 | 6978 | |
9b5e383c | 6979 | WARN_ON(1); |
e93737b0 KK |
6980 | list_del(&dev->unreg_list); |
6981 | continue; | |
9b5e383c | 6982 | } |
449f4544 | 6983 | dev->dismantle = true; |
9b5e383c | 6984 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 6985 | } |
93ee31f1 | 6986 | |
44345724 | 6987 | /* If device is running, close it first. */ |
5cde2829 EB |
6988 | list_for_each_entry(dev, head, unreg_list) |
6989 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 6990 | dev_close_many(&close_head, true); |
93ee31f1 | 6991 | |
44345724 | 6992 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
6993 | /* And unlink it from device chain. */ |
6994 | unlist_netdevice(dev); | |
93ee31f1 | 6995 | |
9b5e383c ED |
6996 | dev->reg_state = NETREG_UNREGISTERING; |
6997 | } | |
41852497 | 6998 | flush_all_backlogs(); |
93ee31f1 DL |
6999 | |
7000 | synchronize_net(); | |
7001 | ||
9b5e383c | 7002 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
7003 | struct sk_buff *skb = NULL; |
7004 | ||
9b5e383c ED |
7005 | /* Shutdown queueing discipline. */ |
7006 | dev_shutdown(dev); | |
93ee31f1 DL |
7007 | |
7008 | ||
9b5e383c | 7009 | /* Notify protocols, that we are about to destroy |
eb13da1a | 7010 | * this device. They should clean all the things. |
7011 | */ | |
9b5e383c | 7012 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
93ee31f1 | 7013 | |
395eea6c MB |
7014 | if (!dev->rtnl_link_ops || |
7015 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7016 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, | |
7017 | GFP_KERNEL); | |
7018 | ||
9b5e383c ED |
7019 | /* |
7020 | * Flush the unicast and multicast chains | |
7021 | */ | |
a748ee24 | 7022 | dev_uc_flush(dev); |
22bedad3 | 7023 | dev_mc_flush(dev); |
93ee31f1 | 7024 | |
9b5e383c ED |
7025 | if (dev->netdev_ops->ndo_uninit) |
7026 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 7027 | |
395eea6c MB |
7028 | if (skb) |
7029 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 7030 | |
9ff162a8 JP |
7031 | /* Notifier chain MUST detach us all upper devices. */ |
7032 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
0f524a80 | 7033 | WARN_ON(netdev_has_any_lower_dev(dev)); |
93ee31f1 | 7034 | |
9b5e383c ED |
7035 | /* Remove entries from kobject tree */ |
7036 | netdev_unregister_kobject(dev); | |
024e9679 AD |
7037 | #ifdef CONFIG_XPS |
7038 | /* Remove XPS queueing entries */ | |
7039 | netif_reset_xps_queues_gt(dev, 0); | |
7040 | #endif | |
9b5e383c | 7041 | } |
93ee31f1 | 7042 | |
850a545b | 7043 | synchronize_net(); |
395264d5 | 7044 | |
a5ee1551 | 7045 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
7046 | dev_put(dev); |
7047 | } | |
7048 | ||
7049 | static void rollback_registered(struct net_device *dev) | |
7050 | { | |
7051 | LIST_HEAD(single); | |
7052 | ||
7053 | list_add(&dev->unreg_list, &single); | |
7054 | rollback_registered_many(&single); | |
ceaaec98 | 7055 | list_del(&single); |
93ee31f1 DL |
7056 | } |
7057 | ||
fd867d51 JW |
7058 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
7059 | struct net_device *upper, netdev_features_t features) | |
7060 | { | |
7061 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
7062 | netdev_features_t feature; | |
5ba3f7d6 | 7063 | int feature_bit; |
fd867d51 | 7064 | |
5ba3f7d6 JW |
7065 | for_each_netdev_feature(&upper_disables, feature_bit) { |
7066 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
7067 | if (!(upper->wanted_features & feature) |
7068 | && (features & feature)) { | |
7069 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
7070 | &feature, upper->name); | |
7071 | features &= ~feature; | |
7072 | } | |
7073 | } | |
7074 | ||
7075 | return features; | |
7076 | } | |
7077 | ||
7078 | static void netdev_sync_lower_features(struct net_device *upper, | |
7079 | struct net_device *lower, netdev_features_t features) | |
7080 | { | |
7081 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
7082 | netdev_features_t feature; | |
5ba3f7d6 | 7083 | int feature_bit; |
fd867d51 | 7084 | |
5ba3f7d6 JW |
7085 | for_each_netdev_feature(&upper_disables, feature_bit) { |
7086 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
7087 | if (!(features & feature) && (lower->features & feature)) { |
7088 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
7089 | &feature, lower->name); | |
7090 | lower->wanted_features &= ~feature; | |
7091 | netdev_update_features(lower); | |
7092 | ||
7093 | if (unlikely(lower->features & feature)) | |
7094 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
7095 | &feature, lower->name); | |
7096 | } | |
7097 | } | |
7098 | } | |
7099 | ||
c8f44aff MM |
7100 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
7101 | netdev_features_t features) | |
b63365a2 | 7102 | { |
57422dc5 MM |
7103 | /* Fix illegal checksum combinations */ |
7104 | if ((features & NETIF_F_HW_CSUM) && | |
7105 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 7106 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
7107 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
7108 | } | |
7109 | ||
b63365a2 | 7110 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 7111 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 7112 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 7113 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
7114 | } |
7115 | ||
ec5f0615 PS |
7116 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
7117 | !(features & NETIF_F_IP_CSUM)) { | |
7118 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
7119 | features &= ~NETIF_F_TSO; | |
7120 | features &= ~NETIF_F_TSO_ECN; | |
7121 | } | |
7122 | ||
7123 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
7124 | !(features & NETIF_F_IPV6_CSUM)) { | |
7125 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
7126 | features &= ~NETIF_F_TSO6; | |
7127 | } | |
7128 | ||
b1dc497b AD |
7129 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
7130 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
7131 | features &= ~NETIF_F_TSO_MANGLEID; | |
7132 | ||
31d8b9e0 BH |
7133 | /* TSO ECN requires that TSO is present as well. */ |
7134 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
7135 | features &= ~NETIF_F_TSO_ECN; | |
7136 | ||
212b573f MM |
7137 | /* Software GSO depends on SG. */ |
7138 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 7139 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
7140 | features &= ~NETIF_F_GSO; |
7141 | } | |
7142 | ||
acd1130e | 7143 | /* UFO needs SG and checksumming */ |
b63365a2 | 7144 | if (features & NETIF_F_UFO) { |
79032644 | 7145 | /* maybe split UFO into V4 and V6? */ |
c8cd0989 TH |
7146 | if (!(features & NETIF_F_HW_CSUM) && |
7147 | ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) != | |
7148 | (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 7149 | netdev_dbg(dev, |
acd1130e | 7150 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
7151 | features &= ~NETIF_F_UFO; |
7152 | } | |
7153 | ||
7154 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 7155 | netdev_dbg(dev, |
acd1130e | 7156 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
7157 | features &= ~NETIF_F_UFO; |
7158 | } | |
7159 | } | |
7160 | ||
802ab55a AD |
7161 | /* GSO partial features require GSO partial be set */ |
7162 | if ((features & dev->gso_partial_features) && | |
7163 | !(features & NETIF_F_GSO_PARTIAL)) { | |
7164 | netdev_dbg(dev, | |
7165 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
7166 | features &= ~dev->gso_partial_features; | |
7167 | } | |
7168 | ||
b63365a2 HX |
7169 | return features; |
7170 | } | |
b63365a2 | 7171 | |
6cb6a27c | 7172 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 7173 | { |
fd867d51 | 7174 | struct net_device *upper, *lower; |
c8f44aff | 7175 | netdev_features_t features; |
fd867d51 | 7176 | struct list_head *iter; |
e7868a85 | 7177 | int err = -1; |
5455c699 | 7178 | |
87267485 MM |
7179 | ASSERT_RTNL(); |
7180 | ||
5455c699 MM |
7181 | features = netdev_get_wanted_features(dev); |
7182 | ||
7183 | if (dev->netdev_ops->ndo_fix_features) | |
7184 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
7185 | ||
7186 | /* driver might be less strict about feature dependencies */ | |
7187 | features = netdev_fix_features(dev, features); | |
7188 | ||
fd867d51 JW |
7189 | /* some features can't be enabled if they're off an an upper device */ |
7190 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
7191 | features = netdev_sync_upper_features(dev, upper, features); | |
7192 | ||
5455c699 | 7193 | if (dev->features == features) |
e7868a85 | 7194 | goto sync_lower; |
5455c699 | 7195 | |
c8f44aff MM |
7196 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
7197 | &dev->features, &features); | |
5455c699 MM |
7198 | |
7199 | if (dev->netdev_ops->ndo_set_features) | |
7200 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
7201 | else |
7202 | err = 0; | |
5455c699 | 7203 | |
6cb6a27c | 7204 | if (unlikely(err < 0)) { |
5455c699 | 7205 | netdev_err(dev, |
c8f44aff MM |
7206 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
7207 | err, &features, &dev->features); | |
17b85d29 NA |
7208 | /* return non-0 since some features might have changed and |
7209 | * it's better to fire a spurious notification than miss it | |
7210 | */ | |
7211 | return -1; | |
6cb6a27c MM |
7212 | } |
7213 | ||
e7868a85 | 7214 | sync_lower: |
fd867d51 JW |
7215 | /* some features must be disabled on lower devices when disabled |
7216 | * on an upper device (think: bonding master or bridge) | |
7217 | */ | |
7218 | netdev_for_each_lower_dev(dev, lower, iter) | |
7219 | netdev_sync_lower_features(dev, lower, features); | |
7220 | ||
6cb6a27c MM |
7221 | if (!err) |
7222 | dev->features = features; | |
7223 | ||
e7868a85 | 7224 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
7225 | } |
7226 | ||
afe12cc8 MM |
7227 | /** |
7228 | * netdev_update_features - recalculate device features | |
7229 | * @dev: the device to check | |
7230 | * | |
7231 | * Recalculate dev->features set and send notifications if it | |
7232 | * has changed. Should be called after driver or hardware dependent | |
7233 | * conditions might have changed that influence the features. | |
7234 | */ | |
6cb6a27c MM |
7235 | void netdev_update_features(struct net_device *dev) |
7236 | { | |
7237 | if (__netdev_update_features(dev)) | |
7238 | netdev_features_change(dev); | |
5455c699 MM |
7239 | } |
7240 | EXPORT_SYMBOL(netdev_update_features); | |
7241 | ||
afe12cc8 MM |
7242 | /** |
7243 | * netdev_change_features - recalculate device features | |
7244 | * @dev: the device to check | |
7245 | * | |
7246 | * Recalculate dev->features set and send notifications even | |
7247 | * if they have not changed. Should be called instead of | |
7248 | * netdev_update_features() if also dev->vlan_features might | |
7249 | * have changed to allow the changes to be propagated to stacked | |
7250 | * VLAN devices. | |
7251 | */ | |
7252 | void netdev_change_features(struct net_device *dev) | |
7253 | { | |
7254 | __netdev_update_features(dev); | |
7255 | netdev_features_change(dev); | |
7256 | } | |
7257 | EXPORT_SYMBOL(netdev_change_features); | |
7258 | ||
fc4a7489 PM |
7259 | /** |
7260 | * netif_stacked_transfer_operstate - transfer operstate | |
7261 | * @rootdev: the root or lower level device to transfer state from | |
7262 | * @dev: the device to transfer operstate to | |
7263 | * | |
7264 | * Transfer operational state from root to device. This is normally | |
7265 | * called when a stacking relationship exists between the root | |
7266 | * device and the device(a leaf device). | |
7267 | */ | |
7268 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
7269 | struct net_device *dev) | |
7270 | { | |
7271 | if (rootdev->operstate == IF_OPER_DORMANT) | |
7272 | netif_dormant_on(dev); | |
7273 | else | |
7274 | netif_dormant_off(dev); | |
7275 | ||
0575c86b ZS |
7276 | if (netif_carrier_ok(rootdev)) |
7277 | netif_carrier_on(dev); | |
7278 | else | |
7279 | netif_carrier_off(dev); | |
fc4a7489 PM |
7280 | } |
7281 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
7282 | ||
a953be53 | 7283 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
7284 | static int netif_alloc_rx_queues(struct net_device *dev) |
7285 | { | |
1b4bf461 | 7286 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 7287 | struct netdev_rx_queue *rx; |
10595902 | 7288 | size_t sz = count * sizeof(*rx); |
1b4bf461 | 7289 | |
bd25fa7b | 7290 | BUG_ON(count < 1); |
1b4bf461 | 7291 | |
da6bc57a MH |
7292 | rx = kvzalloc(sz, GFP_KERNEL | __GFP_REPEAT); |
7293 | if (!rx) | |
7294 | return -ENOMEM; | |
7295 | ||
bd25fa7b TH |
7296 | dev->_rx = rx; |
7297 | ||
bd25fa7b | 7298 | for (i = 0; i < count; i++) |
fe822240 | 7299 | rx[i].dev = dev; |
1b4bf461 ED |
7300 | return 0; |
7301 | } | |
bf264145 | 7302 | #endif |
1b4bf461 | 7303 | |
aa942104 CG |
7304 | static void netdev_init_one_queue(struct net_device *dev, |
7305 | struct netdev_queue *queue, void *_unused) | |
7306 | { | |
7307 | /* Initialize queue lock */ | |
7308 | spin_lock_init(&queue->_xmit_lock); | |
7309 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
7310 | queue->xmit_lock_owner = -1; | |
b236da69 | 7311 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 7312 | queue->dev = dev; |
114cf580 TH |
7313 | #ifdef CONFIG_BQL |
7314 | dql_init(&queue->dql, HZ); | |
7315 | #endif | |
aa942104 CG |
7316 | } |
7317 | ||
60877a32 ED |
7318 | static void netif_free_tx_queues(struct net_device *dev) |
7319 | { | |
4cb28970 | 7320 | kvfree(dev->_tx); |
60877a32 ED |
7321 | } |
7322 | ||
e6484930 TH |
7323 | static int netif_alloc_netdev_queues(struct net_device *dev) |
7324 | { | |
7325 | unsigned int count = dev->num_tx_queues; | |
7326 | struct netdev_queue *tx; | |
60877a32 | 7327 | size_t sz = count * sizeof(*tx); |
e6484930 | 7328 | |
d339727c ED |
7329 | if (count < 1 || count > 0xffff) |
7330 | return -EINVAL; | |
62b5942a | 7331 | |
da6bc57a MH |
7332 | tx = kvzalloc(sz, GFP_KERNEL | __GFP_REPEAT); |
7333 | if (!tx) | |
7334 | return -ENOMEM; | |
7335 | ||
e6484930 | 7336 | dev->_tx = tx; |
1d24eb48 | 7337 | |
e6484930 TH |
7338 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
7339 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
7340 | |
7341 | return 0; | |
e6484930 TH |
7342 | } |
7343 | ||
a2029240 DV |
7344 | void netif_tx_stop_all_queues(struct net_device *dev) |
7345 | { | |
7346 | unsigned int i; | |
7347 | ||
7348 | for (i = 0; i < dev->num_tx_queues; i++) { | |
7349 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 7350 | |
a2029240 DV |
7351 | netif_tx_stop_queue(txq); |
7352 | } | |
7353 | } | |
7354 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
7355 | ||
1da177e4 LT |
7356 | /** |
7357 | * register_netdevice - register a network device | |
7358 | * @dev: device to register | |
7359 | * | |
7360 | * Take a completed network device structure and add it to the kernel | |
7361 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7362 | * chain. 0 is returned on success. A negative errno code is returned | |
7363 | * on a failure to set up the device, or if the name is a duplicate. | |
7364 | * | |
7365 | * Callers must hold the rtnl semaphore. You may want | |
7366 | * register_netdev() instead of this. | |
7367 | * | |
7368 | * BUGS: | |
7369 | * The locking appears insufficient to guarantee two parallel registers | |
7370 | * will not get the same name. | |
7371 | */ | |
7372 | ||
7373 | int register_netdevice(struct net_device *dev) | |
7374 | { | |
1da177e4 | 7375 | int ret; |
d314774c | 7376 | struct net *net = dev_net(dev); |
1da177e4 LT |
7377 | |
7378 | BUG_ON(dev_boot_phase); | |
7379 | ASSERT_RTNL(); | |
7380 | ||
b17a7c17 SH |
7381 | might_sleep(); |
7382 | ||
1da177e4 LT |
7383 | /* When net_device's are persistent, this will be fatal. */ |
7384 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 7385 | BUG_ON(!net); |
1da177e4 | 7386 | |
f1f28aa3 | 7387 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 7388 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 7389 | |
828de4f6 | 7390 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
7391 | if (ret < 0) |
7392 | goto out; | |
7393 | ||
1da177e4 | 7394 | /* Init, if this function is available */ |
d314774c SH |
7395 | if (dev->netdev_ops->ndo_init) { |
7396 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
7397 | if (ret) { |
7398 | if (ret > 0) | |
7399 | ret = -EIO; | |
90833aa4 | 7400 | goto out; |
1da177e4 LT |
7401 | } |
7402 | } | |
4ec93edb | 7403 | |
f646968f PM |
7404 | if (((dev->hw_features | dev->features) & |
7405 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
7406 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
7407 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
7408 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
7409 | ret = -EINVAL; | |
7410 | goto err_uninit; | |
7411 | } | |
7412 | ||
9c7dafbf PE |
7413 | ret = -EBUSY; |
7414 | if (!dev->ifindex) | |
7415 | dev->ifindex = dev_new_index(net); | |
7416 | else if (__dev_get_by_index(net, dev->ifindex)) | |
7417 | goto err_uninit; | |
7418 | ||
5455c699 MM |
7419 | /* Transfer changeable features to wanted_features and enable |
7420 | * software offloads (GSO and GRO). | |
7421 | */ | |
7422 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
7423 | dev->features |= NETIF_F_SOFT_FEATURES; |
7424 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 7425 | |
cbc53e08 | 7426 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 7427 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 7428 | |
7f348a60 AD |
7429 | /* If IPv4 TCP segmentation offload is supported we should also |
7430 | * allow the device to enable segmenting the frame with the option | |
7431 | * of ignoring a static IP ID value. This doesn't enable the | |
7432 | * feature itself but allows the user to enable it later. | |
7433 | */ | |
cbc53e08 AD |
7434 | if (dev->hw_features & NETIF_F_TSO) |
7435 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
7436 | if (dev->vlan_features & NETIF_F_TSO) |
7437 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
7438 | if (dev->mpls_features & NETIF_F_TSO) | |
7439 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
7440 | if (dev->hw_enc_features & NETIF_F_TSO) | |
7441 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 7442 | |
1180e7d6 | 7443 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 7444 | */ |
1180e7d6 | 7445 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 7446 | |
ee579677 PS |
7447 | /* Make NETIF_F_SG inheritable to tunnel devices. |
7448 | */ | |
802ab55a | 7449 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 7450 | |
0d89d203 SH |
7451 | /* Make NETIF_F_SG inheritable to MPLS. |
7452 | */ | |
7453 | dev->mpls_features |= NETIF_F_SG; | |
7454 | ||
7ffbe3fd JB |
7455 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
7456 | ret = notifier_to_errno(ret); | |
7457 | if (ret) | |
7458 | goto err_uninit; | |
7459 | ||
8b41d188 | 7460 | ret = netdev_register_kobject(dev); |
b17a7c17 | 7461 | if (ret) |
7ce1b0ed | 7462 | goto err_uninit; |
b17a7c17 SH |
7463 | dev->reg_state = NETREG_REGISTERED; |
7464 | ||
6cb6a27c | 7465 | __netdev_update_features(dev); |
8e9b59b2 | 7466 | |
1da177e4 LT |
7467 | /* |
7468 | * Default initial state at registry is that the | |
7469 | * device is present. | |
7470 | */ | |
7471 | ||
7472 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7473 | ||
8f4cccbb BH |
7474 | linkwatch_init_dev(dev); |
7475 | ||
1da177e4 | 7476 | dev_init_scheduler(dev); |
1da177e4 | 7477 | dev_hold(dev); |
ce286d32 | 7478 | list_netdevice(dev); |
7bf23575 | 7479 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 7480 | |
948b337e JP |
7481 | /* If the device has permanent device address, driver should |
7482 | * set dev_addr and also addr_assign_type should be set to | |
7483 | * NET_ADDR_PERM (default value). | |
7484 | */ | |
7485 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
7486 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
7487 | ||
1da177e4 | 7488 | /* Notify protocols, that a new device appeared. */ |
056925ab | 7489 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 7490 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
7491 | if (ret) { |
7492 | rollback_registered(dev); | |
7493 | dev->reg_state = NETREG_UNREGISTERED; | |
7494 | } | |
d90a909e EB |
7495 | /* |
7496 | * Prevent userspace races by waiting until the network | |
7497 | * device is fully setup before sending notifications. | |
7498 | */ | |
a2835763 PM |
7499 | if (!dev->rtnl_link_ops || |
7500 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 7501 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
7502 | |
7503 | out: | |
7504 | return ret; | |
7ce1b0ed HX |
7505 | |
7506 | err_uninit: | |
d314774c SH |
7507 | if (dev->netdev_ops->ndo_uninit) |
7508 | dev->netdev_ops->ndo_uninit(dev); | |
cf124db5 DM |
7509 | if (dev->priv_destructor) |
7510 | dev->priv_destructor(dev); | |
7ce1b0ed | 7511 | goto out; |
1da177e4 | 7512 | } |
d1b19dff | 7513 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 7514 | |
937f1ba5 BH |
7515 | /** |
7516 | * init_dummy_netdev - init a dummy network device for NAPI | |
7517 | * @dev: device to init | |
7518 | * | |
7519 | * This takes a network device structure and initialize the minimum | |
7520 | * amount of fields so it can be used to schedule NAPI polls without | |
7521 | * registering a full blown interface. This is to be used by drivers | |
7522 | * that need to tie several hardware interfaces to a single NAPI | |
7523 | * poll scheduler due to HW limitations. | |
7524 | */ | |
7525 | int init_dummy_netdev(struct net_device *dev) | |
7526 | { | |
7527 | /* Clear everything. Note we don't initialize spinlocks | |
7528 | * are they aren't supposed to be taken by any of the | |
7529 | * NAPI code and this dummy netdev is supposed to be | |
7530 | * only ever used for NAPI polls | |
7531 | */ | |
7532 | memset(dev, 0, sizeof(struct net_device)); | |
7533 | ||
7534 | /* make sure we BUG if trying to hit standard | |
7535 | * register/unregister code path | |
7536 | */ | |
7537 | dev->reg_state = NETREG_DUMMY; | |
7538 | ||
937f1ba5 BH |
7539 | /* NAPI wants this */ |
7540 | INIT_LIST_HEAD(&dev->napi_list); | |
7541 | ||
7542 | /* a dummy interface is started by default */ | |
7543 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7544 | set_bit(__LINK_STATE_START, &dev->state); | |
7545 | ||
29b4433d ED |
7546 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
7547 | * because users of this 'device' dont need to change | |
7548 | * its refcount. | |
7549 | */ | |
7550 | ||
937f1ba5 BH |
7551 | return 0; |
7552 | } | |
7553 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
7554 | ||
7555 | ||
1da177e4 LT |
7556 | /** |
7557 | * register_netdev - register a network device | |
7558 | * @dev: device to register | |
7559 | * | |
7560 | * Take a completed network device structure and add it to the kernel | |
7561 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7562 | * chain. 0 is returned on success. A negative errno code is returned | |
7563 | * on a failure to set up the device, or if the name is a duplicate. | |
7564 | * | |
38b4da38 | 7565 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
7566 | * and expands the device name if you passed a format string to |
7567 | * alloc_netdev. | |
7568 | */ | |
7569 | int register_netdev(struct net_device *dev) | |
7570 | { | |
7571 | int err; | |
7572 | ||
7573 | rtnl_lock(); | |
1da177e4 | 7574 | err = register_netdevice(dev); |
1da177e4 LT |
7575 | rtnl_unlock(); |
7576 | return err; | |
7577 | } | |
7578 | EXPORT_SYMBOL(register_netdev); | |
7579 | ||
29b4433d ED |
7580 | int netdev_refcnt_read(const struct net_device *dev) |
7581 | { | |
7582 | int i, refcnt = 0; | |
7583 | ||
7584 | for_each_possible_cpu(i) | |
7585 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
7586 | return refcnt; | |
7587 | } | |
7588 | EXPORT_SYMBOL(netdev_refcnt_read); | |
7589 | ||
2c53040f | 7590 | /** |
1da177e4 | 7591 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 7592 | * @dev: target net_device |
1da177e4 LT |
7593 | * |
7594 | * This is called when unregistering network devices. | |
7595 | * | |
7596 | * Any protocol or device that holds a reference should register | |
7597 | * for netdevice notification, and cleanup and put back the | |
7598 | * reference if they receive an UNREGISTER event. | |
7599 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 7600 | * call dev_put. |
1da177e4 LT |
7601 | */ |
7602 | static void netdev_wait_allrefs(struct net_device *dev) | |
7603 | { | |
7604 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 7605 | int refcnt; |
1da177e4 | 7606 | |
e014debe ED |
7607 | linkwatch_forget_dev(dev); |
7608 | ||
1da177e4 | 7609 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
7610 | refcnt = netdev_refcnt_read(dev); |
7611 | ||
7612 | while (refcnt != 0) { | |
1da177e4 | 7613 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 7614 | rtnl_lock(); |
1da177e4 LT |
7615 | |
7616 | /* Rebroadcast unregister notification */ | |
056925ab | 7617 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 7618 | |
748e2d93 | 7619 | __rtnl_unlock(); |
0115e8e3 | 7620 | rcu_barrier(); |
748e2d93 ED |
7621 | rtnl_lock(); |
7622 | ||
0115e8e3 | 7623 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
7624 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
7625 | &dev->state)) { | |
7626 | /* We must not have linkwatch events | |
7627 | * pending on unregister. If this | |
7628 | * happens, we simply run the queue | |
7629 | * unscheduled, resulting in a noop | |
7630 | * for this device. | |
7631 | */ | |
7632 | linkwatch_run_queue(); | |
7633 | } | |
7634 | ||
6756ae4b | 7635 | __rtnl_unlock(); |
1da177e4 LT |
7636 | |
7637 | rebroadcast_time = jiffies; | |
7638 | } | |
7639 | ||
7640 | msleep(250); | |
7641 | ||
29b4433d ED |
7642 | refcnt = netdev_refcnt_read(dev); |
7643 | ||
1da177e4 | 7644 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
7645 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
7646 | dev->name, refcnt); | |
1da177e4 LT |
7647 | warning_time = jiffies; |
7648 | } | |
7649 | } | |
7650 | } | |
7651 | ||
7652 | /* The sequence is: | |
7653 | * | |
7654 | * rtnl_lock(); | |
7655 | * ... | |
7656 | * register_netdevice(x1); | |
7657 | * register_netdevice(x2); | |
7658 | * ... | |
7659 | * unregister_netdevice(y1); | |
7660 | * unregister_netdevice(y2); | |
7661 | * ... | |
7662 | * rtnl_unlock(); | |
7663 | * free_netdev(y1); | |
7664 | * free_netdev(y2); | |
7665 | * | |
58ec3b4d | 7666 | * We are invoked by rtnl_unlock(). |
1da177e4 | 7667 | * This allows us to deal with problems: |
b17a7c17 | 7668 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
7669 | * without deadlocking with linkwatch via keventd. |
7670 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
7671 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
7672 | * |
7673 | * We must not return until all unregister events added during | |
7674 | * the interval the lock was held have been completed. | |
1da177e4 | 7675 | */ |
1da177e4 LT |
7676 | void netdev_run_todo(void) |
7677 | { | |
626ab0e6 | 7678 | struct list_head list; |
1da177e4 | 7679 | |
1da177e4 | 7680 | /* Snapshot list, allow later requests */ |
626ab0e6 | 7681 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
7682 | |
7683 | __rtnl_unlock(); | |
626ab0e6 | 7684 | |
0115e8e3 ED |
7685 | |
7686 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
7687 | if (!list_empty(&list)) |
7688 | rcu_barrier(); | |
7689 | ||
1da177e4 LT |
7690 | while (!list_empty(&list)) { |
7691 | struct net_device *dev | |
e5e26d75 | 7692 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
7693 | list_del(&dev->todo_list); |
7694 | ||
748e2d93 | 7695 | rtnl_lock(); |
0115e8e3 | 7696 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 7697 | __rtnl_unlock(); |
0115e8e3 | 7698 | |
b17a7c17 | 7699 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 7700 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
7701 | dev->name, dev->reg_state); |
7702 | dump_stack(); | |
7703 | continue; | |
7704 | } | |
1da177e4 | 7705 | |
b17a7c17 | 7706 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 7707 | |
b17a7c17 | 7708 | netdev_wait_allrefs(dev); |
1da177e4 | 7709 | |
b17a7c17 | 7710 | /* paranoia */ |
29b4433d | 7711 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
7712 | BUG_ON(!list_empty(&dev->ptype_all)); |
7713 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
7714 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
7715 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 7716 | WARN_ON(dev->dn_ptr); |
1da177e4 | 7717 | |
cf124db5 DM |
7718 | if (dev->priv_destructor) |
7719 | dev->priv_destructor(dev); | |
7720 | if (dev->needs_free_netdev) | |
7721 | free_netdev(dev); | |
9093bbb2 | 7722 | |
50624c93 EB |
7723 | /* Report a network device has been unregistered */ |
7724 | rtnl_lock(); | |
7725 | dev_net(dev)->dev_unreg_count--; | |
7726 | __rtnl_unlock(); | |
7727 | wake_up(&netdev_unregistering_wq); | |
7728 | ||
9093bbb2 SH |
7729 | /* Free network device */ |
7730 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 7731 | } |
1da177e4 LT |
7732 | } |
7733 | ||
9256645a JW |
7734 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
7735 | * all the same fields in the same order as net_device_stats, with only | |
7736 | * the type differing, but rtnl_link_stats64 may have additional fields | |
7737 | * at the end for newer counters. | |
3cfde79c | 7738 | */ |
77a1abf5 ED |
7739 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
7740 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
7741 | { |
7742 | #if BITS_PER_LONG == 64 | |
9256645a | 7743 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
77a1abf5 | 7744 | memcpy(stats64, netdev_stats, sizeof(*stats64)); |
9256645a JW |
7745 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7746 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
7747 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 7748 | #else |
9256645a | 7749 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
7750 | const unsigned long *src = (const unsigned long *)netdev_stats; |
7751 | u64 *dst = (u64 *)stats64; | |
7752 | ||
9256645a | 7753 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
7754 | for (i = 0; i < n; i++) |
7755 | dst[i] = src[i]; | |
9256645a JW |
7756 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7757 | memset((char *)stats64 + n * sizeof(u64), 0, | |
7758 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
7759 | #endif |
7760 | } | |
77a1abf5 | 7761 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 7762 | |
eeda3fd6 SH |
7763 | /** |
7764 | * dev_get_stats - get network device statistics | |
7765 | * @dev: device to get statistics from | |
28172739 | 7766 | * @storage: place to store stats |
eeda3fd6 | 7767 | * |
d7753516 BH |
7768 | * Get network statistics from device. Return @storage. |
7769 | * The device driver may provide its own method by setting | |
7770 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
7771 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 7772 | */ |
d7753516 BH |
7773 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
7774 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 7775 | { |
eeda3fd6 SH |
7776 | const struct net_device_ops *ops = dev->netdev_ops; |
7777 | ||
28172739 ED |
7778 | if (ops->ndo_get_stats64) { |
7779 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
7780 | ops->ndo_get_stats64(dev, storage); |
7781 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 7782 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
7783 | } else { |
7784 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 7785 | } |
6f64ec74 ED |
7786 | storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); |
7787 | storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); | |
7788 | storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); | |
28172739 | 7789 | return storage; |
c45d286e | 7790 | } |
eeda3fd6 | 7791 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 7792 | |
24824a09 | 7793 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 7794 | { |
24824a09 | 7795 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 7796 | |
24824a09 ED |
7797 | #ifdef CONFIG_NET_CLS_ACT |
7798 | if (queue) | |
7799 | return queue; | |
7800 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
7801 | if (!queue) | |
7802 | return NULL; | |
7803 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 7804 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
7805 | queue->qdisc_sleeping = &noop_qdisc; |
7806 | rcu_assign_pointer(dev->ingress_queue, queue); | |
7807 | #endif | |
7808 | return queue; | |
bb949fbd DM |
7809 | } |
7810 | ||
2c60db03 ED |
7811 | static const struct ethtool_ops default_ethtool_ops; |
7812 | ||
d07d7507 SG |
7813 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
7814 | const struct ethtool_ops *ops) | |
7815 | { | |
7816 | if (dev->ethtool_ops == &default_ethtool_ops) | |
7817 | dev->ethtool_ops = ops; | |
7818 | } | |
7819 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
7820 | ||
74d332c1 ED |
7821 | void netdev_freemem(struct net_device *dev) |
7822 | { | |
7823 | char *addr = (char *)dev - dev->padded; | |
7824 | ||
4cb28970 | 7825 | kvfree(addr); |
74d332c1 ED |
7826 | } |
7827 | ||
1da177e4 | 7828 | /** |
722c9a0c | 7829 | * alloc_netdev_mqs - allocate network device |
7830 | * @sizeof_priv: size of private data to allocate space for | |
7831 | * @name: device name format string | |
7832 | * @name_assign_type: origin of device name | |
7833 | * @setup: callback to initialize device | |
7834 | * @txqs: the number of TX subqueues to allocate | |
7835 | * @rxqs: the number of RX subqueues to allocate | |
7836 | * | |
7837 | * Allocates a struct net_device with private data area for driver use | |
7838 | * and performs basic initialization. Also allocates subqueue structs | |
7839 | * for each queue on the device. | |
1da177e4 | 7840 | */ |
36909ea4 | 7841 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 7842 | unsigned char name_assign_type, |
36909ea4 TH |
7843 | void (*setup)(struct net_device *), |
7844 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 7845 | { |
1da177e4 | 7846 | struct net_device *dev; |
7943986c | 7847 | size_t alloc_size; |
1ce8e7b5 | 7848 | struct net_device *p; |
1da177e4 | 7849 | |
b6fe17d6 SH |
7850 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
7851 | ||
36909ea4 | 7852 | if (txqs < 1) { |
7b6cd1ce | 7853 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
7854 | return NULL; |
7855 | } | |
7856 | ||
a953be53 | 7857 | #ifdef CONFIG_SYSFS |
36909ea4 | 7858 | if (rxqs < 1) { |
7b6cd1ce | 7859 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
7860 | return NULL; |
7861 | } | |
7862 | #endif | |
7863 | ||
fd2ea0a7 | 7864 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
7865 | if (sizeof_priv) { |
7866 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 7867 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
7868 | alloc_size += sizeof_priv; |
7869 | } | |
7870 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 7871 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 7872 | |
da6bc57a | 7873 | p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_REPEAT); |
62b5942a | 7874 | if (!p) |
1da177e4 | 7875 | return NULL; |
1da177e4 | 7876 | |
1ce8e7b5 | 7877 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 7878 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 7879 | |
29b4433d ED |
7880 | dev->pcpu_refcnt = alloc_percpu(int); |
7881 | if (!dev->pcpu_refcnt) | |
74d332c1 | 7882 | goto free_dev; |
ab9c73cc | 7883 | |
ab9c73cc | 7884 | if (dev_addr_init(dev)) |
29b4433d | 7885 | goto free_pcpu; |
ab9c73cc | 7886 | |
22bedad3 | 7887 | dev_mc_init(dev); |
a748ee24 | 7888 | dev_uc_init(dev); |
ccffad25 | 7889 | |
c346dca1 | 7890 | dev_net_set(dev, &init_net); |
1da177e4 | 7891 | |
8d3bdbd5 | 7892 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 7893 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 7894 | |
8d3bdbd5 DM |
7895 | INIT_LIST_HEAD(&dev->napi_list); |
7896 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 7897 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 7898 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
7899 | INIT_LIST_HEAD(&dev->adj_list.upper); |
7900 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
7901 | INIT_LIST_HEAD(&dev->ptype_all); |
7902 | INIT_LIST_HEAD(&dev->ptype_specific); | |
59cc1f61 JK |
7903 | #ifdef CONFIG_NET_SCHED |
7904 | hash_init(dev->qdisc_hash); | |
7905 | #endif | |
02875878 | 7906 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
7907 | setup(dev); |
7908 | ||
a813104d | 7909 | if (!dev->tx_queue_len) { |
f84bb1ea | 7910 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 7911 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 7912 | } |
906470c1 | 7913 | |
36909ea4 TH |
7914 | dev->num_tx_queues = txqs; |
7915 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 7916 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 7917 | goto free_all; |
e8a0464c | 7918 | |
a953be53 | 7919 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
7920 | dev->num_rx_queues = rxqs; |
7921 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 7922 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 7923 | goto free_all; |
df334545 | 7924 | #endif |
0a9627f2 | 7925 | |
1da177e4 | 7926 | strcpy(dev->name, name); |
c835a677 | 7927 | dev->name_assign_type = name_assign_type; |
cbda10fa | 7928 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
7929 | if (!dev->ethtool_ops) |
7930 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 PN |
7931 | |
7932 | nf_hook_ingress_init(dev); | |
7933 | ||
1da177e4 | 7934 | return dev; |
ab9c73cc | 7935 | |
8d3bdbd5 DM |
7936 | free_all: |
7937 | free_netdev(dev); | |
7938 | return NULL; | |
7939 | ||
29b4433d ED |
7940 | free_pcpu: |
7941 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
7942 | free_dev: |
7943 | netdev_freemem(dev); | |
ab9c73cc | 7944 | return NULL; |
1da177e4 | 7945 | } |
36909ea4 | 7946 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
7947 | |
7948 | /** | |
722c9a0c | 7949 | * free_netdev - free network device |
7950 | * @dev: device | |
1da177e4 | 7951 | * |
722c9a0c | 7952 | * This function does the last stage of destroying an allocated device |
7953 | * interface. The reference to the device object is released. If this | |
7954 | * is the last reference then it will be freed.Must be called in process | |
7955 | * context. | |
1da177e4 LT |
7956 | */ |
7957 | void free_netdev(struct net_device *dev) | |
7958 | { | |
d565b0a1 | 7959 | struct napi_struct *p, *n; |
b5cdae32 | 7960 | struct bpf_prog *prog; |
d565b0a1 | 7961 | |
93d05d4a | 7962 | might_sleep(); |
60877a32 | 7963 | netif_free_tx_queues(dev); |
a953be53 | 7964 | #ifdef CONFIG_SYSFS |
10595902 | 7965 | kvfree(dev->_rx); |
fe822240 | 7966 | #endif |
e8a0464c | 7967 | |
33d480ce | 7968 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 7969 | |
f001fde5 JP |
7970 | /* Flush device addresses */ |
7971 | dev_addr_flush(dev); | |
7972 | ||
d565b0a1 HX |
7973 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
7974 | netif_napi_del(p); | |
7975 | ||
29b4433d ED |
7976 | free_percpu(dev->pcpu_refcnt); |
7977 | dev->pcpu_refcnt = NULL; | |
7978 | ||
b5cdae32 DM |
7979 | prog = rcu_dereference_protected(dev->xdp_prog, 1); |
7980 | if (prog) { | |
7981 | bpf_prog_put(prog); | |
7982 | static_key_slow_dec(&generic_xdp_needed); | |
7983 | } | |
7984 | ||
3041a069 | 7985 | /* Compatibility with error handling in drivers */ |
1da177e4 | 7986 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 7987 | netdev_freemem(dev); |
1da177e4 LT |
7988 | return; |
7989 | } | |
7990 | ||
7991 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
7992 | dev->reg_state = NETREG_RELEASED; | |
7993 | ||
43cb76d9 GKH |
7994 | /* will free via device release */ |
7995 | put_device(&dev->dev); | |
1da177e4 | 7996 | } |
d1b19dff | 7997 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 7998 | |
f0db275a SH |
7999 | /** |
8000 | * synchronize_net - Synchronize with packet receive processing | |
8001 | * | |
8002 | * Wait for packets currently being received to be done. | |
8003 | * Does not block later packets from starting. | |
8004 | */ | |
4ec93edb | 8005 | void synchronize_net(void) |
1da177e4 LT |
8006 | { |
8007 | might_sleep(); | |
be3fc413 ED |
8008 | if (rtnl_is_locked()) |
8009 | synchronize_rcu_expedited(); | |
8010 | else | |
8011 | synchronize_rcu(); | |
1da177e4 | 8012 | } |
d1b19dff | 8013 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
8014 | |
8015 | /** | |
44a0873d | 8016 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 8017 | * @dev: device |
44a0873d | 8018 | * @head: list |
6ebfbc06 | 8019 | * |
1da177e4 | 8020 | * This function shuts down a device interface and removes it |
d59b54b1 | 8021 | * from the kernel tables. |
44a0873d | 8022 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
8023 | * |
8024 | * Callers must hold the rtnl semaphore. You may want | |
8025 | * unregister_netdev() instead of this. | |
8026 | */ | |
8027 | ||
44a0873d | 8028 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 8029 | { |
a6620712 HX |
8030 | ASSERT_RTNL(); |
8031 | ||
44a0873d | 8032 | if (head) { |
9fdce099 | 8033 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
8034 | } else { |
8035 | rollback_registered(dev); | |
8036 | /* Finish processing unregister after unlock */ | |
8037 | net_set_todo(dev); | |
8038 | } | |
1da177e4 | 8039 | } |
44a0873d | 8040 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 8041 | |
9b5e383c ED |
8042 | /** |
8043 | * unregister_netdevice_many - unregister many devices | |
8044 | * @head: list of devices | |
87757a91 ED |
8045 | * |
8046 | * Note: As most callers use a stack allocated list_head, | |
8047 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
8048 | */ |
8049 | void unregister_netdevice_many(struct list_head *head) | |
8050 | { | |
8051 | struct net_device *dev; | |
8052 | ||
8053 | if (!list_empty(head)) { | |
8054 | rollback_registered_many(head); | |
8055 | list_for_each_entry(dev, head, unreg_list) | |
8056 | net_set_todo(dev); | |
87757a91 | 8057 | list_del(head); |
9b5e383c ED |
8058 | } |
8059 | } | |
63c8099d | 8060 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 8061 | |
1da177e4 LT |
8062 | /** |
8063 | * unregister_netdev - remove device from the kernel | |
8064 | * @dev: device | |
8065 | * | |
8066 | * This function shuts down a device interface and removes it | |
d59b54b1 | 8067 | * from the kernel tables. |
1da177e4 LT |
8068 | * |
8069 | * This is just a wrapper for unregister_netdevice that takes | |
8070 | * the rtnl semaphore. In general you want to use this and not | |
8071 | * unregister_netdevice. | |
8072 | */ | |
8073 | void unregister_netdev(struct net_device *dev) | |
8074 | { | |
8075 | rtnl_lock(); | |
8076 | unregister_netdevice(dev); | |
8077 | rtnl_unlock(); | |
8078 | } | |
1da177e4 LT |
8079 | EXPORT_SYMBOL(unregister_netdev); |
8080 | ||
ce286d32 EB |
8081 | /** |
8082 | * dev_change_net_namespace - move device to different nethost namespace | |
8083 | * @dev: device | |
8084 | * @net: network namespace | |
8085 | * @pat: If not NULL name pattern to try if the current device name | |
8086 | * is already taken in the destination network namespace. | |
8087 | * | |
8088 | * This function shuts down a device interface and moves it | |
8089 | * to a new network namespace. On success 0 is returned, on | |
8090 | * a failure a netagive errno code is returned. | |
8091 | * | |
8092 | * Callers must hold the rtnl semaphore. | |
8093 | */ | |
8094 | ||
8095 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
8096 | { | |
ce286d32 EB |
8097 | int err; |
8098 | ||
8099 | ASSERT_RTNL(); | |
8100 | ||
8101 | /* Don't allow namespace local devices to be moved. */ | |
8102 | err = -EINVAL; | |
8103 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8104 | goto out; | |
8105 | ||
8106 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
8107 | if (dev->reg_state != NETREG_REGISTERED) |
8108 | goto out; | |
8109 | ||
8110 | /* Get out if there is nothing todo */ | |
8111 | err = 0; | |
878628fb | 8112 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
8113 | goto out; |
8114 | ||
8115 | /* Pick the destination device name, and ensure | |
8116 | * we can use it in the destination network namespace. | |
8117 | */ | |
8118 | err = -EEXIST; | |
d9031024 | 8119 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
8120 | /* We get here if we can't use the current device name */ |
8121 | if (!pat) | |
8122 | goto out; | |
828de4f6 | 8123 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
8124 | goto out; |
8125 | } | |
8126 | ||
8127 | /* | |
8128 | * And now a mini version of register_netdevice unregister_netdevice. | |
8129 | */ | |
8130 | ||
8131 | /* If device is running close it first. */ | |
9b772652 | 8132 | dev_close(dev); |
ce286d32 EB |
8133 | |
8134 | /* And unlink it from device chain */ | |
8135 | err = -ENODEV; | |
8136 | unlist_netdevice(dev); | |
8137 | ||
8138 | synchronize_net(); | |
8139 | ||
8140 | /* Shutdown queueing discipline. */ | |
8141 | dev_shutdown(dev); | |
8142 | ||
8143 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 8144 | * this device. They should clean all the things. |
8145 | * | |
8146 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
8147 | * This is wanted because this way 8021q and macvlan know | |
8148 | * the device is just moving and can keep their slaves up. | |
8149 | */ | |
ce286d32 | 8150 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 G |
8151 | rcu_barrier(); |
8152 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 8153 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
8154 | |
8155 | /* | |
8156 | * Flush the unicast and multicast chains | |
8157 | */ | |
a748ee24 | 8158 | dev_uc_flush(dev); |
22bedad3 | 8159 | dev_mc_flush(dev); |
ce286d32 | 8160 | |
4e66ae2e SH |
8161 | /* Send a netdev-removed uevent to the old namespace */ |
8162 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 8163 | netdev_adjacent_del_links(dev); |
4e66ae2e | 8164 | |
ce286d32 | 8165 | /* Actually switch the network namespace */ |
c346dca1 | 8166 | dev_net_set(dev, net); |
ce286d32 | 8167 | |
ce286d32 | 8168 | /* If there is an ifindex conflict assign a new one */ |
7a66bbc9 | 8169 | if (__dev_get_by_index(net, dev->ifindex)) |
ce286d32 | 8170 | dev->ifindex = dev_new_index(net); |
ce286d32 | 8171 | |
4e66ae2e SH |
8172 | /* Send a netdev-add uevent to the new namespace */ |
8173 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 8174 | netdev_adjacent_add_links(dev); |
4e66ae2e | 8175 | |
8b41d188 | 8176 | /* Fixup kobjects */ |
a1b3f594 | 8177 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 8178 | WARN_ON(err); |
ce286d32 EB |
8179 | |
8180 | /* Add the device back in the hashes */ | |
8181 | list_netdevice(dev); | |
8182 | ||
8183 | /* Notify protocols, that a new device appeared. */ | |
8184 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
8185 | ||
d90a909e EB |
8186 | /* |
8187 | * Prevent userspace races by waiting until the network | |
8188 | * device is fully setup before sending notifications. | |
8189 | */ | |
7f294054 | 8190 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 8191 | |
ce286d32 EB |
8192 | synchronize_net(); |
8193 | err = 0; | |
8194 | out: | |
8195 | return err; | |
8196 | } | |
463d0183 | 8197 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 8198 | |
f0bf90de | 8199 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
8200 | { |
8201 | struct sk_buff **list_skb; | |
1da177e4 | 8202 | struct sk_buff *skb; |
f0bf90de | 8203 | unsigned int cpu; |
97d8b6e3 | 8204 | struct softnet_data *sd, *oldsd, *remsd = NULL; |
1da177e4 | 8205 | |
1da177e4 LT |
8206 | local_irq_disable(); |
8207 | cpu = smp_processor_id(); | |
8208 | sd = &per_cpu(softnet_data, cpu); | |
8209 | oldsd = &per_cpu(softnet_data, oldcpu); | |
8210 | ||
8211 | /* Find end of our completion_queue. */ | |
8212 | list_skb = &sd->completion_queue; | |
8213 | while (*list_skb) | |
8214 | list_skb = &(*list_skb)->next; | |
8215 | /* Append completion queue from offline CPU. */ | |
8216 | *list_skb = oldsd->completion_queue; | |
8217 | oldsd->completion_queue = NULL; | |
8218 | ||
1da177e4 | 8219 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
8220 | if (oldsd->output_queue) { |
8221 | *sd->output_queue_tailp = oldsd->output_queue; | |
8222 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
8223 | oldsd->output_queue = NULL; | |
8224 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
8225 | } | |
ac64da0b ED |
8226 | /* Append NAPI poll list from offline CPU, with one exception : |
8227 | * process_backlog() must be called by cpu owning percpu backlog. | |
8228 | * We properly handle process_queue & input_pkt_queue later. | |
8229 | */ | |
8230 | while (!list_empty(&oldsd->poll_list)) { | |
8231 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
8232 | struct napi_struct, | |
8233 | poll_list); | |
8234 | ||
8235 | list_del_init(&napi->poll_list); | |
8236 | if (napi->poll == process_backlog) | |
8237 | napi->state = 0; | |
8238 | else | |
8239 | ____napi_schedule(sd, napi); | |
264524d5 | 8240 | } |
1da177e4 LT |
8241 | |
8242 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
8243 | local_irq_enable(); | |
8244 | ||
773fc8f6 | 8245 | #ifdef CONFIG_RPS |
8246 | remsd = oldsd->rps_ipi_list; | |
8247 | oldsd->rps_ipi_list = NULL; | |
8248 | #endif | |
8249 | /* send out pending IPI's on offline CPU */ | |
8250 | net_rps_send_ipi(remsd); | |
8251 | ||
1da177e4 | 8252 | /* Process offline CPU's input_pkt_queue */ |
76cc8b13 | 8253 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 8254 | netif_rx_ni(skb); |
76cc8b13 | 8255 | input_queue_head_incr(oldsd); |
fec5e652 | 8256 | } |
ac64da0b | 8257 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 8258 | netif_rx_ni(skb); |
76cc8b13 TH |
8259 | input_queue_head_incr(oldsd); |
8260 | } | |
1da177e4 | 8261 | |
f0bf90de | 8262 | return 0; |
1da177e4 | 8263 | } |
1da177e4 | 8264 | |
7f353bf2 | 8265 | /** |
b63365a2 HX |
8266 | * netdev_increment_features - increment feature set by one |
8267 | * @all: current feature set | |
8268 | * @one: new feature set | |
8269 | * @mask: mask feature set | |
7f353bf2 HX |
8270 | * |
8271 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
8272 | * @one to the master device with current feature set @all. Will not |
8273 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 8274 | */ |
c8f44aff MM |
8275 | netdev_features_t netdev_increment_features(netdev_features_t all, |
8276 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 8277 | { |
c8cd0989 | 8278 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 8279 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 8280 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 8281 | |
a188222b | 8282 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 8283 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 8284 | |
1742f183 | 8285 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
8286 | if (all & NETIF_F_HW_CSUM) |
8287 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
8288 | |
8289 | return all; | |
8290 | } | |
b63365a2 | 8291 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 8292 | |
430f03cd | 8293 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
8294 | { |
8295 | int i; | |
8296 | struct hlist_head *hash; | |
8297 | ||
8298 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
8299 | if (hash != NULL) | |
8300 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
8301 | INIT_HLIST_HEAD(&hash[i]); | |
8302 | ||
8303 | return hash; | |
8304 | } | |
8305 | ||
881d966b | 8306 | /* Initialize per network namespace state */ |
4665079c | 8307 | static int __net_init netdev_init(struct net *net) |
881d966b | 8308 | { |
734b6541 RM |
8309 | if (net != &init_net) |
8310 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 8311 | |
30d97d35 PE |
8312 | net->dev_name_head = netdev_create_hash(); |
8313 | if (net->dev_name_head == NULL) | |
8314 | goto err_name; | |
881d966b | 8315 | |
30d97d35 PE |
8316 | net->dev_index_head = netdev_create_hash(); |
8317 | if (net->dev_index_head == NULL) | |
8318 | goto err_idx; | |
881d966b EB |
8319 | |
8320 | return 0; | |
30d97d35 PE |
8321 | |
8322 | err_idx: | |
8323 | kfree(net->dev_name_head); | |
8324 | err_name: | |
8325 | return -ENOMEM; | |
881d966b EB |
8326 | } |
8327 | ||
f0db275a SH |
8328 | /** |
8329 | * netdev_drivername - network driver for the device | |
8330 | * @dev: network device | |
f0db275a SH |
8331 | * |
8332 | * Determine network driver for device. | |
8333 | */ | |
3019de12 | 8334 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 8335 | { |
cf04a4c7 SH |
8336 | const struct device_driver *driver; |
8337 | const struct device *parent; | |
3019de12 | 8338 | const char *empty = ""; |
6579e57b AV |
8339 | |
8340 | parent = dev->dev.parent; | |
6579e57b | 8341 | if (!parent) |
3019de12 | 8342 | return empty; |
6579e57b AV |
8343 | |
8344 | driver = parent->driver; | |
8345 | if (driver && driver->name) | |
3019de12 DM |
8346 | return driver->name; |
8347 | return empty; | |
6579e57b AV |
8348 | } |
8349 | ||
6ea754eb JP |
8350 | static void __netdev_printk(const char *level, const struct net_device *dev, |
8351 | struct va_format *vaf) | |
256df2f3 | 8352 | { |
b004ff49 | 8353 | if (dev && dev->dev.parent) { |
6ea754eb JP |
8354 | dev_printk_emit(level[1] - '0', |
8355 | dev->dev.parent, | |
8356 | "%s %s %s%s: %pV", | |
8357 | dev_driver_string(dev->dev.parent), | |
8358 | dev_name(dev->dev.parent), | |
8359 | netdev_name(dev), netdev_reg_state(dev), | |
8360 | vaf); | |
b004ff49 | 8361 | } else if (dev) { |
6ea754eb JP |
8362 | printk("%s%s%s: %pV", |
8363 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 8364 | } else { |
6ea754eb | 8365 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 8366 | } |
256df2f3 JP |
8367 | } |
8368 | ||
6ea754eb JP |
8369 | void netdev_printk(const char *level, const struct net_device *dev, |
8370 | const char *format, ...) | |
256df2f3 JP |
8371 | { |
8372 | struct va_format vaf; | |
8373 | va_list args; | |
256df2f3 JP |
8374 | |
8375 | va_start(args, format); | |
8376 | ||
8377 | vaf.fmt = format; | |
8378 | vaf.va = &args; | |
8379 | ||
6ea754eb | 8380 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 8381 | |
256df2f3 | 8382 | va_end(args); |
256df2f3 JP |
8383 | } |
8384 | EXPORT_SYMBOL(netdev_printk); | |
8385 | ||
8386 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 8387 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 8388 | { \ |
256df2f3 JP |
8389 | struct va_format vaf; \ |
8390 | va_list args; \ | |
8391 | \ | |
8392 | va_start(args, fmt); \ | |
8393 | \ | |
8394 | vaf.fmt = fmt; \ | |
8395 | vaf.va = &args; \ | |
8396 | \ | |
6ea754eb | 8397 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 8398 | \ |
256df2f3 | 8399 | va_end(args); \ |
256df2f3 JP |
8400 | } \ |
8401 | EXPORT_SYMBOL(func); | |
8402 | ||
8403 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
8404 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
8405 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
8406 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
8407 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
8408 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
8409 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
8410 | ||
4665079c | 8411 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
8412 | { |
8413 | kfree(net->dev_name_head); | |
8414 | kfree(net->dev_index_head); | |
8415 | } | |
8416 | ||
022cbae6 | 8417 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
8418 | .init = netdev_init, |
8419 | .exit = netdev_exit, | |
8420 | }; | |
8421 | ||
4665079c | 8422 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 8423 | { |
e008b5fc | 8424 | struct net_device *dev, *aux; |
ce286d32 | 8425 | /* |
e008b5fc | 8426 | * Push all migratable network devices back to the |
ce286d32 EB |
8427 | * initial network namespace |
8428 | */ | |
8429 | rtnl_lock(); | |
e008b5fc | 8430 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 8431 | int err; |
aca51397 | 8432 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
8433 | |
8434 | /* Ignore unmoveable devices (i.e. loopback) */ | |
8435 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8436 | continue; | |
8437 | ||
e008b5fc EB |
8438 | /* Leave virtual devices for the generic cleanup */ |
8439 | if (dev->rtnl_link_ops) | |
8440 | continue; | |
d0c082ce | 8441 | |
25985edc | 8442 | /* Push remaining network devices to init_net */ |
aca51397 PE |
8443 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
8444 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 8445 | if (err) { |
7b6cd1ce JP |
8446 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
8447 | __func__, dev->name, err); | |
aca51397 | 8448 | BUG(); |
ce286d32 EB |
8449 | } |
8450 | } | |
8451 | rtnl_unlock(); | |
8452 | } | |
8453 | ||
50624c93 EB |
8454 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
8455 | { | |
8456 | /* Return with the rtnl_lock held when there are no network | |
8457 | * devices unregistering in any network namespace in net_list. | |
8458 | */ | |
8459 | struct net *net; | |
8460 | bool unregistering; | |
ff960a73 | 8461 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 8462 | |
ff960a73 | 8463 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 8464 | for (;;) { |
50624c93 EB |
8465 | unregistering = false; |
8466 | rtnl_lock(); | |
8467 | list_for_each_entry(net, net_list, exit_list) { | |
8468 | if (net->dev_unreg_count > 0) { | |
8469 | unregistering = true; | |
8470 | break; | |
8471 | } | |
8472 | } | |
8473 | if (!unregistering) | |
8474 | break; | |
8475 | __rtnl_unlock(); | |
ff960a73 PZ |
8476 | |
8477 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 8478 | } |
ff960a73 | 8479 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
8480 | } |
8481 | ||
04dc7f6b EB |
8482 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
8483 | { | |
8484 | /* At exit all network devices most be removed from a network | |
b595076a | 8485 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
8486 | * Do this across as many network namespaces as possible to |
8487 | * improve batching efficiency. | |
8488 | */ | |
8489 | struct net_device *dev; | |
8490 | struct net *net; | |
8491 | LIST_HEAD(dev_kill_list); | |
8492 | ||
50624c93 EB |
8493 | /* To prevent network device cleanup code from dereferencing |
8494 | * loopback devices or network devices that have been freed | |
8495 | * wait here for all pending unregistrations to complete, | |
8496 | * before unregistring the loopback device and allowing the | |
8497 | * network namespace be freed. | |
8498 | * | |
8499 | * The netdev todo list containing all network devices | |
8500 | * unregistrations that happen in default_device_exit_batch | |
8501 | * will run in the rtnl_unlock() at the end of | |
8502 | * default_device_exit_batch. | |
8503 | */ | |
8504 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
8505 | list_for_each_entry(net, net_list, exit_list) { |
8506 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 8507 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
8508 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
8509 | else | |
8510 | unregister_netdevice_queue(dev, &dev_kill_list); | |
8511 | } | |
8512 | } | |
8513 | unregister_netdevice_many(&dev_kill_list); | |
8514 | rtnl_unlock(); | |
8515 | } | |
8516 | ||
022cbae6 | 8517 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 8518 | .exit = default_device_exit, |
04dc7f6b | 8519 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
8520 | }; |
8521 | ||
1da177e4 LT |
8522 | /* |
8523 | * Initialize the DEV module. At boot time this walks the device list and | |
8524 | * unhooks any devices that fail to initialise (normally hardware not | |
8525 | * present) and leaves us with a valid list of present and active devices. | |
8526 | * | |
8527 | */ | |
8528 | ||
8529 | /* | |
8530 | * This is called single threaded during boot, so no need | |
8531 | * to take the rtnl semaphore. | |
8532 | */ | |
8533 | static int __init net_dev_init(void) | |
8534 | { | |
8535 | int i, rc = -ENOMEM; | |
8536 | ||
8537 | BUG_ON(!dev_boot_phase); | |
8538 | ||
1da177e4 LT |
8539 | if (dev_proc_init()) |
8540 | goto out; | |
8541 | ||
8b41d188 | 8542 | if (netdev_kobject_init()) |
1da177e4 LT |
8543 | goto out; |
8544 | ||
8545 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 8546 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
8547 | INIT_LIST_HEAD(&ptype_base[i]); |
8548 | ||
62532da9 VY |
8549 | INIT_LIST_HEAD(&offload_base); |
8550 | ||
881d966b EB |
8551 | if (register_pernet_subsys(&netdev_net_ops)) |
8552 | goto out; | |
1da177e4 LT |
8553 | |
8554 | /* | |
8555 | * Initialise the packet receive queues. | |
8556 | */ | |
8557 | ||
6f912042 | 8558 | for_each_possible_cpu(i) { |
41852497 | 8559 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 8560 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 8561 | |
41852497 ED |
8562 | INIT_WORK(flush, flush_backlog); |
8563 | ||
e36fa2f7 | 8564 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 8565 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 8566 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 8567 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 8568 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
8569 | sd->csd.func = rps_trigger_softirq; |
8570 | sd->csd.info = sd; | |
e36fa2f7 | 8571 | sd->cpu = i; |
1e94d72f | 8572 | #endif |
0a9627f2 | 8573 | |
e36fa2f7 ED |
8574 | sd->backlog.poll = process_backlog; |
8575 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
8576 | } |
8577 | ||
1da177e4 LT |
8578 | dev_boot_phase = 0; |
8579 | ||
505d4f73 EB |
8580 | /* The loopback device is special if any other network devices |
8581 | * is present in a network namespace the loopback device must | |
8582 | * be present. Since we now dynamically allocate and free the | |
8583 | * loopback device ensure this invariant is maintained by | |
8584 | * keeping the loopback device as the first device on the | |
8585 | * list of network devices. Ensuring the loopback devices | |
8586 | * is the first device that appears and the last network device | |
8587 | * that disappears. | |
8588 | */ | |
8589 | if (register_pernet_device(&loopback_net_ops)) | |
8590 | goto out; | |
8591 | ||
8592 | if (register_pernet_device(&default_device_ops)) | |
8593 | goto out; | |
8594 | ||
962cf36c CM |
8595 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
8596 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 8597 | |
f0bf90de SAS |
8598 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
8599 | NULL, dev_cpu_dead); | |
8600 | WARN_ON(rc < 0); | |
f38a9eb1 | 8601 | dst_subsys_init(); |
1da177e4 LT |
8602 | rc = 0; |
8603 | out: | |
8604 | return rc; | |
8605 | } | |
8606 | ||
8607 | subsys_initcall(net_dev_init); |