Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 | 2 | /* |
722c9a0c | 3 | * NET3 Protocol independent device support routines. |
1da177e4 | 4 | * |
1da177e4 | 5 | * Derived from the non IP parts of dev.c 1.0.19 |
722c9a0c | 6 | * Authors: Ross Biro |
1da177e4 LT |
7 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
8 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
9 | * | |
10 | * Additional Authors: | |
11 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
12 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
13 | * David Hinds <dahinds@users.sourceforge.net> | |
14 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
15 | * Adam Sulmicki <adam@cfar.umd.edu> | |
16 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
17 | * | |
18 | * Changes: | |
19 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
722c9a0c | 20 | * to 2 if register_netdev gets called |
21 | * before net_dev_init & also removed a | |
22 | * few lines of code in the process. | |
1da177e4 LT |
23 | * Alan Cox : device private ioctl copies fields back. |
24 | * Alan Cox : Transmit queue code does relevant | |
25 | * stunts to keep the queue safe. | |
26 | * Alan Cox : Fixed double lock. | |
27 | * Alan Cox : Fixed promisc NULL pointer trap | |
28 | * ???????? : Support the full private ioctl range | |
29 | * Alan Cox : Moved ioctl permission check into | |
30 | * drivers | |
31 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
32 | * Alan Cox : 100 backlog just doesn't cut it when | |
33 | * you start doing multicast video 8) | |
34 | * Alan Cox : Rewrote net_bh and list manager. | |
722c9a0c | 35 | * Alan Cox : Fix ETH_P_ALL echoback lengths. |
1da177e4 LT |
36 | * Alan Cox : Took out transmit every packet pass |
37 | * Saved a few bytes in the ioctl handler | |
38 | * Alan Cox : Network driver sets packet type before | |
39 | * calling netif_rx. Saves a function | |
40 | * call a packet. | |
41 | * Alan Cox : Hashed net_bh() | |
42 | * Richard Kooijman: Timestamp fixes. | |
43 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
44 | * Alan Cox : Device lock protection. | |
722c9a0c | 45 | * Alan Cox : Fixed nasty side effect of device close |
1da177e4 LT |
46 | * changes. |
47 | * Rudi Cilibrasi : Pass the right thing to | |
48 | * set_mac_address() | |
49 | * Dave Miller : 32bit quantity for the device lock to | |
50 | * make it work out on a Sparc. | |
51 | * Bjorn Ekwall : Added KERNELD hack. | |
52 | * Alan Cox : Cleaned up the backlog initialise. | |
53 | * Craig Metz : SIOCGIFCONF fix if space for under | |
54 | * 1 device. | |
55 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
56 | * is no device open function. | |
57 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
58 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
59 | * Cyrus Durgin : Cleaned for KMOD | |
60 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
61 | * A network device unload needs to purge | |
62 | * the backlog queue. | |
63 | * Paul Rusty Russell : SIOCSIFNAME | |
64 | * Pekka Riikonen : Netdev boot-time settings code | |
65 | * Andrew Morton : Make unregister_netdevice wait | |
722c9a0c | 66 | * indefinitely on dev->refcnt |
67 | * J Hadi Salim : - Backlog queue sampling | |
1da177e4 LT |
68 | * - netif_rx() feedback |
69 | */ | |
70 | ||
7c0f6ba6 | 71 | #include <linux/uaccess.h> |
1da177e4 | 72 | #include <linux/bitops.h> |
4fc268d2 | 73 | #include <linux/capability.h> |
1da177e4 LT |
74 | #include <linux/cpu.h> |
75 | #include <linux/types.h> | |
76 | #include <linux/kernel.h> | |
08e9897d | 77 | #include <linux/hash.h> |
5a0e3ad6 | 78 | #include <linux/slab.h> |
1da177e4 | 79 | #include <linux/sched.h> |
f1083048 | 80 | #include <linux/sched/mm.h> |
4a3e2f71 | 81 | #include <linux/mutex.h> |
11d6011c | 82 | #include <linux/rwsem.h> |
1da177e4 LT |
83 | #include <linux/string.h> |
84 | #include <linux/mm.h> | |
85 | #include <linux/socket.h> | |
86 | #include <linux/sockios.h> | |
87 | #include <linux/errno.h> | |
88 | #include <linux/interrupt.h> | |
89 | #include <linux/if_ether.h> | |
90 | #include <linux/netdevice.h> | |
91 | #include <linux/etherdevice.h> | |
0187bdfb | 92 | #include <linux/ethtool.h> |
1da177e4 | 93 | #include <linux/skbuff.h> |
29863d41 | 94 | #include <linux/kthread.h> |
a7862b45 | 95 | #include <linux/bpf.h> |
b5cdae32 | 96 | #include <linux/bpf_trace.h> |
457c4cbc | 97 | #include <net/net_namespace.h> |
1da177e4 | 98 | #include <net/sock.h> |
02d62e86 | 99 | #include <net/busy_poll.h> |
1da177e4 | 100 | #include <linux/rtnetlink.h> |
1da177e4 | 101 | #include <linux/stat.h> |
b14a9fc4 | 102 | #include <net/dsa.h> |
1da177e4 | 103 | #include <net/dst.h> |
fc4099f1 | 104 | #include <net/dst_metadata.h> |
04f00ab2 | 105 | #include <net/gro.h> |
1da177e4 | 106 | #include <net/pkt_sched.h> |
87d83093 | 107 | #include <net/pkt_cls.h> |
1da177e4 | 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> |
caeda9b9 | 134 | #include <linux/inetdevice.h> |
c445477d | 135 | #include <linux/cpu_rmap.h> |
c5905afb | 136 | #include <linux/static_key.h> |
af12fa6e | 137 | #include <linux/hashtable.h> |
60877a32 | 138 | #include <linux/vmalloc.h> |
529d0489 | 139 | #include <linux/if_macvlan.h> |
e7fd2885 | 140 | #include <linux/errqueue.h> |
3b47d303 | 141 | #include <linux/hrtimer.h> |
357b6cc5 | 142 | #include <linux/netfilter_ingress.h> |
40e4e713 | 143 | #include <linux/crash_dump.h> |
b72b5bf6 | 144 | #include <linux/sctp.h> |
ae847f40 | 145 | #include <net/udp_tunnel.h> |
6621dd29 | 146 | #include <linux/net_namespace.h> |
aaa5d90b | 147 | #include <linux/indirect_call_wrapper.h> |
af3836df | 148 | #include <net/devlink.h> |
bd869245 | 149 | #include <linux/pm_runtime.h> |
3744741a | 150 | #include <linux/prandom.h> |
1da177e4 | 151 | |
342709ef PE |
152 | #include "net-sysfs.h" |
153 | ||
d565b0a1 HX |
154 | #define MAX_GRO_SKBS 8 |
155 | ||
5d38a079 HX |
156 | /* This should be increased if a protocol with a bigger head is added. */ |
157 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
158 | ||
1da177e4 | 159 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 160 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
161 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
162 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 163 | static struct list_head offload_base __read_mostly; |
1da177e4 | 164 | |
ae78dbfa | 165 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 | 166 | static int call_netdevice_notifiers_info(unsigned long val, |
54951194 | 167 | struct netdev_notifier_info *info); |
26372605 PM |
168 | static int call_netdevice_notifiers_extack(unsigned long val, |
169 | struct net_device *dev, | |
170 | struct netlink_ext_ack *extack); | |
90b602f8 | 171 | static struct napi_struct *napi_by_id(unsigned int napi_id); |
ae78dbfa | 172 | |
1da177e4 | 173 | /* |
7562f876 | 174 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
175 | * semaphore. |
176 | * | |
c6d14c84 | 177 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
178 | * |
179 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 180 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
181 | * actual updates. This allows pure readers to access the list even |
182 | * while a writer is preparing to update it. | |
183 | * | |
184 | * To put it another way, dev_base_lock is held for writing only to | |
185 | * protect against pure readers; the rtnl semaphore provides the | |
186 | * protection against other writers. | |
187 | * | |
188 | * See, for example usages, register_netdevice() and | |
189 | * unregister_netdevice(), which must be called with the rtnl | |
190 | * semaphore held. | |
191 | */ | |
1da177e4 | 192 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
193 | EXPORT_SYMBOL(dev_base_lock); |
194 | ||
6c557001 FW |
195 | static DEFINE_MUTEX(ifalias_mutex); |
196 | ||
af12fa6e ET |
197 | /* protects napi_hash addition/deletion and napi_gen_id */ |
198 | static DEFINE_SPINLOCK(napi_hash_lock); | |
199 | ||
52bd2d62 | 200 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 201 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 202 | |
11d6011c | 203 | static DECLARE_RWSEM(devnet_rename_sem); |
c91f6df2 | 204 | |
4e985ada TG |
205 | static inline void dev_base_seq_inc(struct net *net) |
206 | { | |
643aa9cb | 207 | while (++net->dev_base_seq == 0) |
208 | ; | |
4e985ada TG |
209 | } |
210 | ||
881d966b | 211 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 212 | { |
8387ff25 | 213 | unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); |
95c96174 | 214 | |
08e9897d | 215 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
216 | } |
217 | ||
881d966b | 218 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 219 | { |
7c28bd0b | 220 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
221 | } |
222 | ||
e36fa2f7 | 223 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
224 | { |
225 | #ifdef CONFIG_RPS | |
e36fa2f7 | 226 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
227 | #endif |
228 | } | |
229 | ||
e36fa2f7 | 230 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
231 | { |
232 | #ifdef CONFIG_RPS | |
e36fa2f7 | 233 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
234 | #endif |
235 | } | |
236 | ||
ff927412 JP |
237 | static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, |
238 | const char *name) | |
239 | { | |
240 | struct netdev_name_node *name_node; | |
241 | ||
242 | name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); | |
243 | if (!name_node) | |
244 | return NULL; | |
245 | INIT_HLIST_NODE(&name_node->hlist); | |
246 | name_node->dev = dev; | |
247 | name_node->name = name; | |
248 | return name_node; | |
249 | } | |
250 | ||
251 | static struct netdev_name_node * | |
252 | netdev_name_node_head_alloc(struct net_device *dev) | |
253 | { | |
36fbf1e5 JP |
254 | struct netdev_name_node *name_node; |
255 | ||
256 | name_node = netdev_name_node_alloc(dev, dev->name); | |
257 | if (!name_node) | |
258 | return NULL; | |
259 | INIT_LIST_HEAD(&name_node->list); | |
260 | return name_node; | |
ff927412 JP |
261 | } |
262 | ||
263 | static void netdev_name_node_free(struct netdev_name_node *name_node) | |
264 | { | |
265 | kfree(name_node); | |
266 | } | |
267 | ||
268 | static void netdev_name_node_add(struct net *net, | |
269 | struct netdev_name_node *name_node) | |
270 | { | |
271 | hlist_add_head_rcu(&name_node->hlist, | |
272 | dev_name_hash(net, name_node->name)); | |
273 | } | |
274 | ||
275 | static void netdev_name_node_del(struct netdev_name_node *name_node) | |
276 | { | |
277 | hlist_del_rcu(&name_node->hlist); | |
278 | } | |
279 | ||
280 | static struct netdev_name_node *netdev_name_node_lookup(struct net *net, | |
281 | const char *name) | |
282 | { | |
283 | struct hlist_head *head = dev_name_hash(net, name); | |
284 | struct netdev_name_node *name_node; | |
285 | ||
286 | hlist_for_each_entry(name_node, head, hlist) | |
287 | if (!strcmp(name_node->name, name)) | |
288 | return name_node; | |
289 | return NULL; | |
290 | } | |
291 | ||
292 | static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, | |
293 | const char *name) | |
294 | { | |
295 | struct hlist_head *head = dev_name_hash(net, name); | |
296 | struct netdev_name_node *name_node; | |
297 | ||
298 | hlist_for_each_entry_rcu(name_node, head, hlist) | |
299 | if (!strcmp(name_node->name, name)) | |
300 | return name_node; | |
301 | return NULL; | |
302 | } | |
303 | ||
36fbf1e5 JP |
304 | int netdev_name_node_alt_create(struct net_device *dev, const char *name) |
305 | { | |
306 | struct netdev_name_node *name_node; | |
307 | struct net *net = dev_net(dev); | |
308 | ||
309 | name_node = netdev_name_node_lookup(net, name); | |
310 | if (name_node) | |
311 | return -EEXIST; | |
312 | name_node = netdev_name_node_alloc(dev, name); | |
313 | if (!name_node) | |
314 | return -ENOMEM; | |
315 | netdev_name_node_add(net, name_node); | |
316 | /* The node that holds dev->name acts as a head of per-device list. */ | |
317 | list_add_tail(&name_node->list, &dev->name_node->list); | |
318 | ||
319 | return 0; | |
320 | } | |
321 | EXPORT_SYMBOL(netdev_name_node_alt_create); | |
322 | ||
323 | static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) | |
324 | { | |
325 | list_del(&name_node->list); | |
326 | netdev_name_node_del(name_node); | |
327 | kfree(name_node->name); | |
328 | netdev_name_node_free(name_node); | |
329 | } | |
330 | ||
331 | int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) | |
332 | { | |
333 | struct netdev_name_node *name_node; | |
334 | struct net *net = dev_net(dev); | |
335 | ||
336 | name_node = netdev_name_node_lookup(net, name); | |
337 | if (!name_node) | |
338 | return -ENOENT; | |
e08ad805 ED |
339 | /* lookup might have found our primary name or a name belonging |
340 | * to another device. | |
341 | */ | |
342 | if (name_node == dev->name_node || name_node->dev != dev) | |
343 | return -EINVAL; | |
344 | ||
36fbf1e5 JP |
345 | __netdev_name_node_alt_destroy(name_node); |
346 | ||
347 | return 0; | |
348 | } | |
349 | EXPORT_SYMBOL(netdev_name_node_alt_destroy); | |
350 | ||
351 | static void netdev_name_node_alt_flush(struct net_device *dev) | |
352 | { | |
353 | struct netdev_name_node *name_node, *tmp; | |
354 | ||
355 | list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) | |
356 | __netdev_name_node_alt_destroy(name_node); | |
357 | } | |
358 | ||
ce286d32 | 359 | /* Device list insertion */ |
53759be9 | 360 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 361 | { |
c346dca1 | 362 | struct net *net = dev_net(dev); |
ce286d32 EB |
363 | |
364 | ASSERT_RTNL(); | |
365 | ||
366 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 367 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
ff927412 | 368 | netdev_name_node_add(net, dev->name_node); |
fb699dfd ED |
369 | hlist_add_head_rcu(&dev->index_hlist, |
370 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 371 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
372 | |
373 | dev_base_seq_inc(net); | |
ce286d32 EB |
374 | } |
375 | ||
fb699dfd ED |
376 | /* Device list removal |
377 | * caller must respect a RCU grace period before freeing/reusing dev | |
378 | */ | |
ce286d32 EB |
379 | static void unlist_netdevice(struct net_device *dev) |
380 | { | |
381 | ASSERT_RTNL(); | |
382 | ||
383 | /* Unlink dev from the device chain */ | |
384 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 385 | list_del_rcu(&dev->dev_list); |
ff927412 | 386 | netdev_name_node_del(dev->name_node); |
fb699dfd | 387 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 388 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
389 | |
390 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
391 | } |
392 | ||
1da177e4 LT |
393 | /* |
394 | * Our notifier list | |
395 | */ | |
396 | ||
f07d5b94 | 397 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
398 | |
399 | /* | |
400 | * Device drivers call our routines to queue packets here. We empty the | |
401 | * queue in the local softnet handler. | |
402 | */ | |
bea3348e | 403 | |
9958da05 | 404 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 405 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 406 | |
1a33e10e CW |
407 | #ifdef CONFIG_LOCKDEP |
408 | /* | |
409 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class | |
410 | * according to dev->type | |
411 | */ | |
412 | static const unsigned short netdev_lock_type[] = { | |
413 | ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
414 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
415 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
416 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
417 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
418 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
419 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
420 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
421 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
422 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
423 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
424 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
425 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, | |
426 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
427 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
428 | ||
429 | static const char *const netdev_lock_name[] = { | |
430 | "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
431 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
432 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
433 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
434 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
435 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
436 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
437 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
438 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
439 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
440 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
441 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
442 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", | |
443 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
444 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
445 | ||
446 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
845e0ebb | 447 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
1a33e10e CW |
448 | |
449 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
450 | { | |
451 | int i; | |
452 | ||
453 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
454 | if (netdev_lock_type[i] == dev_type) | |
455 | return i; | |
456 | /* the last key is used by default */ | |
457 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
458 | } | |
459 | ||
460 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, | |
461 | unsigned short dev_type) | |
462 | { | |
463 | int i; | |
464 | ||
465 | i = netdev_lock_pos(dev_type); | |
466 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
467 | netdev_lock_name[i]); | |
468 | } | |
845e0ebb CW |
469 | |
470 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
471 | { | |
472 | int i; | |
473 | ||
474 | i = netdev_lock_pos(dev->type); | |
475 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
476 | &netdev_addr_lock_key[i], | |
477 | netdev_lock_name[i]); | |
478 | } | |
1a33e10e CW |
479 | #else |
480 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, | |
481 | unsigned short dev_type) | |
482 | { | |
483 | } | |
845e0ebb CW |
484 | |
485 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
486 | { | |
487 | } | |
1a33e10e CW |
488 | #endif |
489 | ||
1da177e4 | 490 | /******************************************************************************* |
eb13da1a | 491 | * |
492 | * Protocol management and registration routines | |
493 | * | |
494 | *******************************************************************************/ | |
1da177e4 | 495 | |
1da177e4 | 496 | |
1da177e4 LT |
497 | /* |
498 | * Add a protocol ID to the list. Now that the input handler is | |
499 | * smarter we can dispense with all the messy stuff that used to be | |
500 | * here. | |
501 | * | |
502 | * BEWARE!!! Protocol handlers, mangling input packets, | |
503 | * MUST BE last in hash buckets and checking protocol handlers | |
504 | * MUST start from promiscuous ptype_all chain in net_bh. | |
505 | * It is true now, do not change it. | |
506 | * Explanation follows: if protocol handler, mangling packet, will | |
507 | * be the first on list, it is not able to sense, that packet | |
508 | * is cloned and should be copied-on-write, so that it will | |
509 | * change it and subsequent readers will get broken packet. | |
510 | * --ANK (980803) | |
511 | */ | |
512 | ||
c07b68e8 ED |
513 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
514 | { | |
515 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 516 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 517 | else |
7866a621 SN |
518 | return pt->dev ? &pt->dev->ptype_specific : |
519 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
520 | } |
521 | ||
1da177e4 LT |
522 | /** |
523 | * dev_add_pack - add packet handler | |
524 | * @pt: packet type declaration | |
525 | * | |
526 | * Add a protocol handler to the networking stack. The passed &packet_type | |
527 | * is linked into kernel lists and may not be freed until it has been | |
528 | * removed from the kernel lists. | |
529 | * | |
4ec93edb | 530 | * This call does not sleep therefore it can not |
1da177e4 LT |
531 | * guarantee all CPU's that are in middle of receiving packets |
532 | * will see the new packet type (until the next received packet). | |
533 | */ | |
534 | ||
535 | void dev_add_pack(struct packet_type *pt) | |
536 | { | |
c07b68e8 | 537 | struct list_head *head = ptype_head(pt); |
1da177e4 | 538 | |
c07b68e8 ED |
539 | spin_lock(&ptype_lock); |
540 | list_add_rcu(&pt->list, head); | |
541 | spin_unlock(&ptype_lock); | |
1da177e4 | 542 | } |
d1b19dff | 543 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 544 | |
1da177e4 LT |
545 | /** |
546 | * __dev_remove_pack - remove packet handler | |
547 | * @pt: packet type declaration | |
548 | * | |
549 | * Remove a protocol handler that was previously added to the kernel | |
550 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
551 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 552 | * returns. |
1da177e4 LT |
553 | * |
554 | * The packet type might still be in use by receivers | |
555 | * and must not be freed until after all the CPU's have gone | |
556 | * through a quiescent state. | |
557 | */ | |
558 | void __dev_remove_pack(struct packet_type *pt) | |
559 | { | |
c07b68e8 | 560 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
561 | struct packet_type *pt1; |
562 | ||
c07b68e8 | 563 | spin_lock(&ptype_lock); |
1da177e4 LT |
564 | |
565 | list_for_each_entry(pt1, head, list) { | |
566 | if (pt == pt1) { | |
567 | list_del_rcu(&pt->list); | |
568 | goto out; | |
569 | } | |
570 | } | |
571 | ||
7b6cd1ce | 572 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 573 | out: |
c07b68e8 | 574 | spin_unlock(&ptype_lock); |
1da177e4 | 575 | } |
d1b19dff ED |
576 | EXPORT_SYMBOL(__dev_remove_pack); |
577 | ||
1da177e4 LT |
578 | /** |
579 | * dev_remove_pack - remove packet handler | |
580 | * @pt: packet type declaration | |
581 | * | |
582 | * Remove a protocol handler that was previously added to the kernel | |
583 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
584 | * from the kernel lists and can be freed or reused once this function | |
585 | * returns. | |
586 | * | |
587 | * This call sleeps to guarantee that no CPU is looking at the packet | |
588 | * type after return. | |
589 | */ | |
590 | void dev_remove_pack(struct packet_type *pt) | |
591 | { | |
592 | __dev_remove_pack(pt); | |
4ec93edb | 593 | |
1da177e4 LT |
594 | synchronize_net(); |
595 | } | |
d1b19dff | 596 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 597 | |
62532da9 VY |
598 | |
599 | /** | |
600 | * dev_add_offload - register offload handlers | |
601 | * @po: protocol offload declaration | |
602 | * | |
603 | * Add protocol offload handlers to the networking stack. The passed | |
604 | * &proto_offload is linked into kernel lists and may not be freed until | |
605 | * it has been removed from the kernel lists. | |
606 | * | |
607 | * This call does not sleep therefore it can not | |
608 | * guarantee all CPU's that are in middle of receiving packets | |
609 | * will see the new offload handlers (until the next received packet). | |
610 | */ | |
611 | void dev_add_offload(struct packet_offload *po) | |
612 | { | |
bdef7de4 | 613 | struct packet_offload *elem; |
62532da9 VY |
614 | |
615 | spin_lock(&offload_lock); | |
bdef7de4 DM |
616 | list_for_each_entry(elem, &offload_base, list) { |
617 | if (po->priority < elem->priority) | |
618 | break; | |
619 | } | |
620 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
621 | spin_unlock(&offload_lock); |
622 | } | |
623 | EXPORT_SYMBOL(dev_add_offload); | |
624 | ||
625 | /** | |
626 | * __dev_remove_offload - remove offload handler | |
627 | * @po: packet offload declaration | |
628 | * | |
629 | * Remove a protocol offload handler that was previously added to the | |
630 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
631 | * is removed from the kernel lists and can be freed or reused once this | |
632 | * function returns. | |
633 | * | |
634 | * The packet type might still be in use by receivers | |
635 | * and must not be freed until after all the CPU's have gone | |
636 | * through a quiescent state. | |
637 | */ | |
1d143d9f | 638 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
639 | { |
640 | struct list_head *head = &offload_base; | |
641 | struct packet_offload *po1; | |
642 | ||
c53aa505 | 643 | spin_lock(&offload_lock); |
62532da9 VY |
644 | |
645 | list_for_each_entry(po1, head, list) { | |
646 | if (po == po1) { | |
647 | list_del_rcu(&po->list); | |
648 | goto out; | |
649 | } | |
650 | } | |
651 | ||
652 | pr_warn("dev_remove_offload: %p not found\n", po); | |
653 | out: | |
c53aa505 | 654 | spin_unlock(&offload_lock); |
62532da9 | 655 | } |
62532da9 VY |
656 | |
657 | /** | |
658 | * dev_remove_offload - remove packet offload handler | |
659 | * @po: packet offload declaration | |
660 | * | |
661 | * Remove a packet offload handler that was previously added to the kernel | |
662 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
663 | * removed from the kernel lists and can be freed or reused once this | |
664 | * function returns. | |
665 | * | |
666 | * This call sleeps to guarantee that no CPU is looking at the packet | |
667 | * type after return. | |
668 | */ | |
669 | void dev_remove_offload(struct packet_offload *po) | |
670 | { | |
671 | __dev_remove_offload(po); | |
672 | ||
673 | synchronize_net(); | |
674 | } | |
675 | EXPORT_SYMBOL(dev_remove_offload); | |
676 | ||
1da177e4 | 677 | /****************************************************************************** |
eb13da1a | 678 | * |
679 | * Device Boot-time Settings Routines | |
680 | * | |
681 | ******************************************************************************/ | |
1da177e4 LT |
682 | |
683 | /* Boot time configuration table */ | |
684 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
685 | ||
686 | /** | |
687 | * netdev_boot_setup_add - add new setup entry | |
688 | * @name: name of the device | |
689 | * @map: configured settings for the device | |
690 | * | |
691 | * Adds new setup entry to the dev_boot_setup list. The function | |
692 | * returns 0 on error and 1 on success. This is a generic routine to | |
693 | * all netdevices. | |
694 | */ | |
695 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
696 | { | |
697 | struct netdev_boot_setup *s; | |
698 | int i; | |
699 | ||
700 | s = dev_boot_setup; | |
701 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
702 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
703 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 704 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
705 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
706 | break; | |
707 | } | |
708 | } | |
709 | ||
710 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
711 | } | |
712 | ||
713 | /** | |
722c9a0c | 714 | * netdev_boot_setup_check - check boot time settings |
715 | * @dev: the netdevice | |
1da177e4 | 716 | * |
722c9a0c | 717 | * Check boot time settings for the device. |
718 | * The found settings are set for the device to be used | |
719 | * later in the device probing. | |
720 | * Returns 0 if no settings found, 1 if they are. | |
1da177e4 LT |
721 | */ |
722 | int netdev_boot_setup_check(struct net_device *dev) | |
723 | { | |
724 | struct netdev_boot_setup *s = dev_boot_setup; | |
725 | int i; | |
726 | ||
727 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
728 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 729 | !strcmp(dev->name, s[i].name)) { |
722c9a0c | 730 | dev->irq = s[i].map.irq; |
731 | dev->base_addr = s[i].map.base_addr; | |
732 | dev->mem_start = s[i].map.mem_start; | |
733 | dev->mem_end = s[i].map.mem_end; | |
1da177e4 LT |
734 | return 1; |
735 | } | |
736 | } | |
737 | return 0; | |
738 | } | |
d1b19dff | 739 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
740 | |
741 | ||
742 | /** | |
722c9a0c | 743 | * netdev_boot_base - get address from boot time settings |
744 | * @prefix: prefix for network device | |
745 | * @unit: id for network device | |
746 | * | |
747 | * Check boot time settings for the base address of device. | |
748 | * The found settings are set for the device to be used | |
749 | * later in the device probing. | |
750 | * Returns 0 if no settings found. | |
1da177e4 LT |
751 | */ |
752 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
753 | { | |
754 | const struct netdev_boot_setup *s = dev_boot_setup; | |
755 | char name[IFNAMSIZ]; | |
756 | int i; | |
757 | ||
758 | sprintf(name, "%s%d", prefix, unit); | |
759 | ||
760 | /* | |
761 | * If device already registered then return base of 1 | |
762 | * to indicate not to probe for this interface | |
763 | */ | |
881d966b | 764 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
765 | return 1; |
766 | ||
767 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
768 | if (!strcmp(name, s[i].name)) | |
769 | return s[i].map.base_addr; | |
770 | return 0; | |
771 | } | |
772 | ||
773 | /* | |
774 | * Saves at boot time configured settings for any netdevice. | |
775 | */ | |
776 | int __init netdev_boot_setup(char *str) | |
777 | { | |
778 | int ints[5]; | |
779 | struct ifmap map; | |
780 | ||
781 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
782 | if (!str || !*str) | |
783 | return 0; | |
784 | ||
785 | /* Save settings */ | |
786 | memset(&map, 0, sizeof(map)); | |
787 | if (ints[0] > 0) | |
788 | map.irq = ints[1]; | |
789 | if (ints[0] > 1) | |
790 | map.base_addr = ints[2]; | |
791 | if (ints[0] > 2) | |
792 | map.mem_start = ints[3]; | |
793 | if (ints[0] > 3) | |
794 | map.mem_end = ints[4]; | |
795 | ||
796 | /* Add new entry to the list */ | |
797 | return netdev_boot_setup_add(str, &map); | |
798 | } | |
799 | ||
800 | __setup("netdev=", netdev_boot_setup); | |
801 | ||
802 | /******************************************************************************* | |
eb13da1a | 803 | * |
804 | * Device Interface Subroutines | |
805 | * | |
806 | *******************************************************************************/ | |
1da177e4 | 807 | |
a54acb3a ND |
808 | /** |
809 | * dev_get_iflink - get 'iflink' value of a interface | |
810 | * @dev: targeted interface | |
811 | * | |
812 | * Indicates the ifindex the interface is linked to. | |
813 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
814 | */ | |
815 | ||
816 | int dev_get_iflink(const struct net_device *dev) | |
817 | { | |
818 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
819 | return dev->netdev_ops->ndo_get_iflink(dev); | |
820 | ||
7a66bbc9 | 821 | return dev->ifindex; |
a54acb3a ND |
822 | } |
823 | EXPORT_SYMBOL(dev_get_iflink); | |
824 | ||
fc4099f1 PS |
825 | /** |
826 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
827 | * @dev: targeted interface | |
828 | * @skb: The packet. | |
829 | * | |
830 | * For better visibility of tunnel traffic OVS needs to retrieve | |
831 | * egress tunnel information for a packet. Following API allows | |
832 | * user to get this info. | |
833 | */ | |
834 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
835 | { | |
836 | struct ip_tunnel_info *info; | |
837 | ||
838 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
839 | return -EINVAL; | |
840 | ||
841 | info = skb_tunnel_info_unclone(skb); | |
842 | if (!info) | |
843 | return -ENOMEM; | |
844 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
845 | return -EINVAL; | |
846 | ||
847 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
848 | } | |
849 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
850 | ||
1da177e4 LT |
851 | /** |
852 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 853 | * @net: the applicable net namespace |
1da177e4 LT |
854 | * @name: name to find |
855 | * | |
856 | * Find an interface by name. Must be called under RTNL semaphore | |
857 | * or @dev_base_lock. If the name is found a pointer to the device | |
858 | * is returned. If the name is not found then %NULL is returned. The | |
859 | * reference counters are not incremented so the caller must be | |
860 | * careful with locks. | |
861 | */ | |
862 | ||
881d966b | 863 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 864 | { |
ff927412 | 865 | struct netdev_name_node *node_name; |
1da177e4 | 866 | |
ff927412 JP |
867 | node_name = netdev_name_node_lookup(net, name); |
868 | return node_name ? node_name->dev : NULL; | |
1da177e4 | 869 | } |
d1b19dff | 870 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 871 | |
72c9528b | 872 | /** |
722c9a0c | 873 | * dev_get_by_name_rcu - find a device by its name |
874 | * @net: the applicable net namespace | |
875 | * @name: name to find | |
876 | * | |
877 | * Find an interface by name. | |
878 | * If the name is found a pointer to the device is returned. | |
879 | * If the name is not found then %NULL is returned. | |
880 | * The reference counters are not incremented so the caller must be | |
881 | * careful with locks. The caller must hold RCU lock. | |
72c9528b ED |
882 | */ |
883 | ||
884 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
885 | { | |
ff927412 | 886 | struct netdev_name_node *node_name; |
72c9528b | 887 | |
ff927412 JP |
888 | node_name = netdev_name_node_lookup_rcu(net, name); |
889 | return node_name ? node_name->dev : NULL; | |
72c9528b ED |
890 | } |
891 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
892 | ||
1da177e4 LT |
893 | /** |
894 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 895 | * @net: the applicable net namespace |
1da177e4 LT |
896 | * @name: name to find |
897 | * | |
898 | * Find an interface by name. This can be called from any | |
899 | * context and does its own locking. The returned handle has | |
900 | * the usage count incremented and the caller must use dev_put() to | |
901 | * release it when it is no longer needed. %NULL is returned if no | |
902 | * matching device is found. | |
903 | */ | |
904 | ||
881d966b | 905 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
906 | { |
907 | struct net_device *dev; | |
908 | ||
72c9528b ED |
909 | rcu_read_lock(); |
910 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
911 | if (dev) |
912 | dev_hold(dev); | |
72c9528b | 913 | rcu_read_unlock(); |
1da177e4 LT |
914 | return dev; |
915 | } | |
d1b19dff | 916 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
917 | |
918 | /** | |
919 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 920 | * @net: the applicable net namespace |
1da177e4 LT |
921 | * @ifindex: index of device |
922 | * | |
923 | * Search for an interface by index. Returns %NULL if the device | |
924 | * is not found or a pointer to the device. The device has not | |
925 | * had its reference counter increased so the caller must be careful | |
926 | * about locking. The caller must hold either the RTNL semaphore | |
927 | * or @dev_base_lock. | |
928 | */ | |
929 | ||
881d966b | 930 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 931 | { |
0bd8d536 ED |
932 | struct net_device *dev; |
933 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 934 | |
b67bfe0d | 935 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
936 | if (dev->ifindex == ifindex) |
937 | return dev; | |
0bd8d536 | 938 | |
1da177e4 LT |
939 | return NULL; |
940 | } | |
d1b19dff | 941 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 942 | |
fb699dfd ED |
943 | /** |
944 | * dev_get_by_index_rcu - find a device by its ifindex | |
945 | * @net: the applicable net namespace | |
946 | * @ifindex: index of device | |
947 | * | |
948 | * Search for an interface by index. Returns %NULL if the device | |
949 | * is not found or a pointer to the device. The device has not | |
950 | * had its reference counter increased so the caller must be careful | |
951 | * about locking. The caller must hold RCU lock. | |
952 | */ | |
953 | ||
954 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
955 | { | |
fb699dfd ED |
956 | struct net_device *dev; |
957 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
958 | ||
b67bfe0d | 959 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
960 | if (dev->ifindex == ifindex) |
961 | return dev; | |
962 | ||
963 | return NULL; | |
964 | } | |
965 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
966 | ||
1da177e4 LT |
967 | |
968 | /** | |
969 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 970 | * @net: the applicable net namespace |
1da177e4 LT |
971 | * @ifindex: index of device |
972 | * | |
973 | * Search for an interface by index. Returns NULL if the device | |
974 | * is not found or a pointer to the device. The device returned has | |
975 | * had a reference added and the pointer is safe until the user calls | |
976 | * dev_put to indicate they have finished with it. | |
977 | */ | |
978 | ||
881d966b | 979 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
980 | { |
981 | struct net_device *dev; | |
982 | ||
fb699dfd ED |
983 | rcu_read_lock(); |
984 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
985 | if (dev) |
986 | dev_hold(dev); | |
fb699dfd | 987 | rcu_read_unlock(); |
1da177e4 LT |
988 | return dev; |
989 | } | |
d1b19dff | 990 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 991 | |
90b602f8 ML |
992 | /** |
993 | * dev_get_by_napi_id - find a device by napi_id | |
994 | * @napi_id: ID of the NAPI struct | |
995 | * | |
996 | * Search for an interface by NAPI ID. Returns %NULL if the device | |
997 | * is not found or a pointer to the device. The device has not had | |
998 | * its reference counter increased so the caller must be careful | |
999 | * about locking. The caller must hold RCU lock. | |
1000 | */ | |
1001 | ||
1002 | struct net_device *dev_get_by_napi_id(unsigned int napi_id) | |
1003 | { | |
1004 | struct napi_struct *napi; | |
1005 | ||
1006 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
1007 | ||
1008 | if (napi_id < MIN_NAPI_ID) | |
1009 | return NULL; | |
1010 | ||
1011 | napi = napi_by_id(napi_id); | |
1012 | ||
1013 | return napi ? napi->dev : NULL; | |
1014 | } | |
1015 | EXPORT_SYMBOL(dev_get_by_napi_id); | |
1016 | ||
5dbe7c17 NS |
1017 | /** |
1018 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
1019 | * @net: network namespace | |
1020 | * @name: a pointer to the buffer where the name will be stored. | |
1021 | * @ifindex: the ifindex of the interface to get the name from. | |
5dbe7c17 NS |
1022 | */ |
1023 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
1024 | { | |
1025 | struct net_device *dev; | |
11d6011c | 1026 | int ret; |
5dbe7c17 | 1027 | |
11d6011c | 1028 | down_read(&devnet_rename_sem); |
5dbe7c17 | 1029 | rcu_read_lock(); |
11d6011c | 1030 | |
5dbe7c17 NS |
1031 | dev = dev_get_by_index_rcu(net, ifindex); |
1032 | if (!dev) { | |
11d6011c AD |
1033 | ret = -ENODEV; |
1034 | goto out; | |
5dbe7c17 NS |
1035 | } |
1036 | ||
1037 | strcpy(name, dev->name); | |
5dbe7c17 | 1038 | |
11d6011c AD |
1039 | ret = 0; |
1040 | out: | |
1041 | rcu_read_unlock(); | |
1042 | up_read(&devnet_rename_sem); | |
1043 | return ret; | |
5dbe7c17 NS |
1044 | } |
1045 | ||
1da177e4 | 1046 | /** |
941666c2 | 1047 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 1048 | * @net: the applicable net namespace |
1da177e4 LT |
1049 | * @type: media type of device |
1050 | * @ha: hardware address | |
1051 | * | |
1052 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
1053 | * is not found or a pointer to the device. |
1054 | * The caller must hold RCU or RTNL. | |
941666c2 | 1055 | * The returned device has not had its ref count increased |
1da177e4 LT |
1056 | * and the caller must therefore be careful about locking |
1057 | * | |
1da177e4 LT |
1058 | */ |
1059 | ||
941666c2 ED |
1060 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
1061 | const char *ha) | |
1da177e4 LT |
1062 | { |
1063 | struct net_device *dev; | |
1064 | ||
941666c2 | 1065 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
1066 | if (dev->type == type && |
1067 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
1068 | return dev; |
1069 | ||
1070 | return NULL; | |
1da177e4 | 1071 | } |
941666c2 | 1072 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 1073 | |
881d966b | 1074 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 1075 | { |
99fe3c39 | 1076 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 1077 | |
99fe3c39 ED |
1078 | rcu_read_lock(); |
1079 | for_each_netdev_rcu(net, dev) | |
1080 | if (dev->type == type) { | |
1081 | dev_hold(dev); | |
1082 | ret = dev; | |
1083 | break; | |
1084 | } | |
1085 | rcu_read_unlock(); | |
1086 | return ret; | |
1da177e4 | 1087 | } |
1da177e4 LT |
1088 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
1089 | ||
1090 | /** | |
6c555490 | 1091 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 1092 | * @net: the applicable net namespace |
1da177e4 LT |
1093 | * @if_flags: IFF_* values |
1094 | * @mask: bitmask of bits in if_flags to check | |
1095 | * | |
1096 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 1097 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 1098 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
1099 | */ |
1100 | ||
6c555490 WC |
1101 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
1102 | unsigned short mask) | |
1da177e4 | 1103 | { |
7562f876 | 1104 | struct net_device *dev, *ret; |
1da177e4 | 1105 | |
6c555490 WC |
1106 | ASSERT_RTNL(); |
1107 | ||
7562f876 | 1108 | ret = NULL; |
6c555490 | 1109 | for_each_netdev(net, dev) { |
1da177e4 | 1110 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 1111 | ret = dev; |
1da177e4 LT |
1112 | break; |
1113 | } | |
1114 | } | |
7562f876 | 1115 | return ret; |
1da177e4 | 1116 | } |
6c555490 | 1117 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
1118 | |
1119 | /** | |
1120 | * dev_valid_name - check if name is okay for network device | |
1121 | * @name: name string | |
1122 | * | |
1123 | * Network device names need to be valid file names to | |
4250b75b | 1124 | * allow sysfs to work. We also disallow any kind of |
c7fa9d18 | 1125 | * whitespace. |
1da177e4 | 1126 | */ |
95f050bf | 1127 | bool dev_valid_name(const char *name) |
1da177e4 | 1128 | { |
c7fa9d18 | 1129 | if (*name == '\0') |
95f050bf | 1130 | return false; |
a9d48205 | 1131 | if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) |
95f050bf | 1132 | return false; |
c7fa9d18 | 1133 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1134 | return false; |
c7fa9d18 DM |
1135 | |
1136 | while (*name) { | |
a4176a93 | 1137 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1138 | return false; |
c7fa9d18 DM |
1139 | name++; |
1140 | } | |
95f050bf | 1141 | return true; |
1da177e4 | 1142 | } |
d1b19dff | 1143 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1144 | |
1145 | /** | |
b267b179 EB |
1146 | * __dev_alloc_name - allocate a name for a device |
1147 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1148 | * @name: name format string |
b267b179 | 1149 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1150 | * |
1151 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1152 | * id. It scans list of devices to build up a free map, then chooses |
1153 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1154 | * while allocating the name and adding the device in order to avoid | |
1155 | * duplicates. | |
1156 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1157 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1158 | */ |
1159 | ||
b267b179 | 1160 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1161 | { |
1162 | int i = 0; | |
1da177e4 LT |
1163 | const char *p; |
1164 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1165 | unsigned long *inuse; |
1da177e4 LT |
1166 | struct net_device *d; |
1167 | ||
93809105 RV |
1168 | if (!dev_valid_name(name)) |
1169 | return -EINVAL; | |
1170 | ||
51f299dd | 1171 | p = strchr(name, '%'); |
1da177e4 LT |
1172 | if (p) { |
1173 | /* | |
1174 | * Verify the string as this thing may have come from | |
1175 | * the user. There must be either one "%d" and no other "%" | |
1176 | * characters. | |
1177 | */ | |
1178 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1179 | return -EINVAL; | |
1180 | ||
1181 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1182 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1183 | if (!inuse) |
1184 | return -ENOMEM; | |
1185 | ||
881d966b | 1186 | for_each_netdev(net, d) { |
1da177e4 LT |
1187 | if (!sscanf(d->name, name, &i)) |
1188 | continue; | |
1189 | if (i < 0 || i >= max_netdevices) | |
1190 | continue; | |
1191 | ||
1192 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1193 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1194 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1195 | set_bit(i, inuse); | |
1196 | } | |
1197 | ||
1198 | i = find_first_zero_bit(inuse, max_netdevices); | |
1199 | free_page((unsigned long) inuse); | |
1200 | } | |
1201 | ||
6224abda | 1202 | snprintf(buf, IFNAMSIZ, name, i); |
b267b179 | 1203 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1204 | return i; |
1da177e4 LT |
1205 | |
1206 | /* It is possible to run out of possible slots | |
1207 | * when the name is long and there isn't enough space left | |
1208 | * for the digits, or if all bits are used. | |
1209 | */ | |
029b6d14 | 1210 | return -ENFILE; |
1da177e4 LT |
1211 | } |
1212 | ||
2c88b855 RV |
1213 | static int dev_alloc_name_ns(struct net *net, |
1214 | struct net_device *dev, | |
1215 | const char *name) | |
1216 | { | |
1217 | char buf[IFNAMSIZ]; | |
1218 | int ret; | |
1219 | ||
c46d7642 | 1220 | BUG_ON(!net); |
2c88b855 RV |
1221 | ret = __dev_alloc_name(net, name, buf); |
1222 | if (ret >= 0) | |
1223 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1224 | return ret; | |
1da177e4 LT |
1225 | } |
1226 | ||
b267b179 EB |
1227 | /** |
1228 | * dev_alloc_name - allocate a name for a device | |
1229 | * @dev: device | |
1230 | * @name: name format string | |
1231 | * | |
1232 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1233 | * id. It scans list of devices to build up a free map, then chooses | |
1234 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1235 | * while allocating the name and adding the device in order to avoid | |
1236 | * duplicates. | |
1237 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1238 | * Returns the number of the unit assigned or a negative errno code. | |
1239 | */ | |
1240 | ||
1241 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1242 | { | |
c46d7642 | 1243 | return dev_alloc_name_ns(dev_net(dev), dev, name); |
b267b179 | 1244 | } |
d1b19dff | 1245 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1246 | |
bacb7e18 ED |
1247 | static int dev_get_valid_name(struct net *net, struct net_device *dev, |
1248 | const char *name) | |
828de4f6 | 1249 | { |
55a5ec9b DM |
1250 | BUG_ON(!net); |
1251 | ||
1252 | if (!dev_valid_name(name)) | |
1253 | return -EINVAL; | |
1254 | ||
1255 | if (strchr(name, '%')) | |
1256 | return dev_alloc_name_ns(net, dev, name); | |
1257 | else if (__dev_get_by_name(net, name)) | |
1258 | return -EEXIST; | |
1259 | else if (dev->name != name) | |
1260 | strlcpy(dev->name, name, IFNAMSIZ); | |
1261 | ||
1262 | return 0; | |
d9031024 | 1263 | } |
1da177e4 LT |
1264 | |
1265 | /** | |
1266 | * dev_change_name - change name of a device | |
1267 | * @dev: device | |
1268 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1269 | * | |
1270 | * Change name of a device, can pass format strings "eth%d". | |
1271 | * for wildcarding. | |
1272 | */ | |
cf04a4c7 | 1273 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1274 | { |
238fa362 | 1275 | unsigned char old_assign_type; |
fcc5a03a | 1276 | char oldname[IFNAMSIZ]; |
1da177e4 | 1277 | int err = 0; |
fcc5a03a | 1278 | int ret; |
881d966b | 1279 | struct net *net; |
1da177e4 LT |
1280 | |
1281 | ASSERT_RTNL(); | |
c346dca1 | 1282 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1283 | |
c346dca1 | 1284 | net = dev_net(dev); |
8065a779 SWL |
1285 | |
1286 | /* Some auto-enslaved devices e.g. failover slaves are | |
1287 | * special, as userspace might rename the device after | |
1288 | * the interface had been brought up and running since | |
1289 | * the point kernel initiated auto-enslavement. Allow | |
1290 | * live name change even when these slave devices are | |
1291 | * up and running. | |
1292 | * | |
1293 | * Typically, users of these auto-enslaving devices | |
1294 | * don't actually care about slave name change, as | |
1295 | * they are supposed to operate on master interface | |
1296 | * directly. | |
1297 | */ | |
1298 | if (dev->flags & IFF_UP && | |
1299 | likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) | |
1da177e4 LT |
1300 | return -EBUSY; |
1301 | ||
11d6011c | 1302 | down_write(&devnet_rename_sem); |
c91f6df2 BH |
1303 | |
1304 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
11d6011c | 1305 | up_write(&devnet_rename_sem); |
c8d90dca | 1306 | return 0; |
c91f6df2 | 1307 | } |
c8d90dca | 1308 | |
fcc5a03a HX |
1309 | memcpy(oldname, dev->name, IFNAMSIZ); |
1310 | ||
828de4f6 | 1311 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1312 | if (err < 0) { |
11d6011c | 1313 | up_write(&devnet_rename_sem); |
d9031024 | 1314 | return err; |
c91f6df2 | 1315 | } |
1da177e4 | 1316 | |
6fe82a39 VF |
1317 | if (oldname[0] && !strchr(oldname, '%')) |
1318 | netdev_info(dev, "renamed from %s\n", oldname); | |
1319 | ||
238fa362 TG |
1320 | old_assign_type = dev->name_assign_type; |
1321 | dev->name_assign_type = NET_NAME_RENAMED; | |
1322 | ||
fcc5a03a | 1323 | rollback: |
a1b3f594 EB |
1324 | ret = device_rename(&dev->dev, dev->name); |
1325 | if (ret) { | |
1326 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1327 | dev->name_assign_type = old_assign_type; |
11d6011c | 1328 | up_write(&devnet_rename_sem); |
a1b3f594 | 1329 | return ret; |
dcc99773 | 1330 | } |
7f988eab | 1331 | |
11d6011c | 1332 | up_write(&devnet_rename_sem); |
c91f6df2 | 1333 | |
5bb025fa VF |
1334 | netdev_adjacent_rename_links(dev, oldname); |
1335 | ||
7f988eab | 1336 | write_lock_bh(&dev_base_lock); |
ff927412 | 1337 | netdev_name_node_del(dev->name_node); |
72c9528b ED |
1338 | write_unlock_bh(&dev_base_lock); |
1339 | ||
1340 | synchronize_rcu(); | |
1341 | ||
1342 | write_lock_bh(&dev_base_lock); | |
ff927412 | 1343 | netdev_name_node_add(net, dev->name_node); |
7f988eab HX |
1344 | write_unlock_bh(&dev_base_lock); |
1345 | ||
056925ab | 1346 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1347 | ret = notifier_to_errno(ret); |
1348 | ||
1349 | if (ret) { | |
91e9c07b ED |
1350 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1351 | if (err >= 0) { | |
fcc5a03a | 1352 | err = ret; |
11d6011c | 1353 | down_write(&devnet_rename_sem); |
fcc5a03a | 1354 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1355 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1356 | dev->name_assign_type = old_assign_type; |
1357 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1358 | goto rollback; |
91e9c07b | 1359 | } else { |
7b6cd1ce | 1360 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1361 | dev->name, ret); |
fcc5a03a HX |
1362 | } |
1363 | } | |
1da177e4 LT |
1364 | |
1365 | return err; | |
1366 | } | |
1367 | ||
0b815a1a SH |
1368 | /** |
1369 | * dev_set_alias - change ifalias of a device | |
1370 | * @dev: device | |
1371 | * @alias: name up to IFALIASZ | |
f0db275a | 1372 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1373 | * |
1374 | * Set ifalias for a device, | |
1375 | */ | |
1376 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1377 | { | |
6c557001 | 1378 | struct dev_ifalias *new_alias = NULL; |
0b815a1a SH |
1379 | |
1380 | if (len >= IFALIASZ) | |
1381 | return -EINVAL; | |
1382 | ||
6c557001 FW |
1383 | if (len) { |
1384 | new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); | |
1385 | if (!new_alias) | |
1386 | return -ENOMEM; | |
1387 | ||
1388 | memcpy(new_alias->ifalias, alias, len); | |
1389 | new_alias->ifalias[len] = 0; | |
96ca4a2c OH |
1390 | } |
1391 | ||
6c557001 | 1392 | mutex_lock(&ifalias_mutex); |
e3f0d761 PM |
1393 | new_alias = rcu_replace_pointer(dev->ifalias, new_alias, |
1394 | mutex_is_locked(&ifalias_mutex)); | |
6c557001 FW |
1395 | mutex_unlock(&ifalias_mutex); |
1396 | ||
1397 | if (new_alias) | |
1398 | kfree_rcu(new_alias, rcuhead); | |
0b815a1a | 1399 | |
0b815a1a SH |
1400 | return len; |
1401 | } | |
0fe554a4 | 1402 | EXPORT_SYMBOL(dev_set_alias); |
0b815a1a | 1403 | |
6c557001 FW |
1404 | /** |
1405 | * dev_get_alias - get ifalias of a device | |
1406 | * @dev: device | |
20e88320 | 1407 | * @name: buffer to store name of ifalias |
6c557001 FW |
1408 | * @len: size of buffer |
1409 | * | |
1410 | * get ifalias for a device. Caller must make sure dev cannot go | |
1411 | * away, e.g. rcu read lock or own a reference count to device. | |
1412 | */ | |
1413 | int dev_get_alias(const struct net_device *dev, char *name, size_t len) | |
1414 | { | |
1415 | const struct dev_ifalias *alias; | |
1416 | int ret = 0; | |
1417 | ||
1418 | rcu_read_lock(); | |
1419 | alias = rcu_dereference(dev->ifalias); | |
1420 | if (alias) | |
1421 | ret = snprintf(name, len, "%s", alias->ifalias); | |
1422 | rcu_read_unlock(); | |
1423 | ||
1424 | return ret; | |
1425 | } | |
0b815a1a | 1426 | |
d8a33ac4 | 1427 | /** |
3041a069 | 1428 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1429 | * @dev: device to cause notification |
1430 | * | |
1431 | * Called to indicate a device has changed features. | |
1432 | */ | |
1433 | void netdev_features_change(struct net_device *dev) | |
1434 | { | |
056925ab | 1435 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1436 | } |
1437 | EXPORT_SYMBOL(netdev_features_change); | |
1438 | ||
1da177e4 LT |
1439 | /** |
1440 | * netdev_state_change - device changes state | |
1441 | * @dev: device to cause notification | |
1442 | * | |
1443 | * Called to indicate a device has changed state. This function calls | |
1444 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1445 | * to the routing socket. | |
1446 | */ | |
1447 | void netdev_state_change(struct net_device *dev) | |
1448 | { | |
1449 | if (dev->flags & IFF_UP) { | |
51d0c047 DA |
1450 | struct netdev_notifier_change_info change_info = { |
1451 | .info.dev = dev, | |
1452 | }; | |
54951194 | 1453 | |
51d0c047 | 1454 | call_netdevice_notifiers_info(NETDEV_CHANGE, |
54951194 | 1455 | &change_info.info); |
7f294054 | 1456 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1457 | } |
1458 | } | |
d1b19dff | 1459 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1460 | |
7061eb8c LP |
1461 | /** |
1462 | * __netdev_notify_peers - notify network peers about existence of @dev, | |
1463 | * to be called when rtnl lock is already held. | |
1464 | * @dev: network device | |
1465 | * | |
1466 | * Generate traffic such that interested network peers are aware of | |
1467 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1468 | * a device wants to inform the rest of the network about some sort of | |
1469 | * reconfiguration such as a failover event or virtual machine | |
1470 | * migration. | |
1471 | */ | |
1472 | void __netdev_notify_peers(struct net_device *dev) | |
1473 | { | |
1474 | ASSERT_RTNL(); | |
1475 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1476 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); | |
1477 | } | |
1478 | EXPORT_SYMBOL(__netdev_notify_peers); | |
1479 | ||
ee89bab1 | 1480 | /** |
722c9a0c | 1481 | * netdev_notify_peers - notify network peers about existence of @dev |
1482 | * @dev: network device | |
ee89bab1 AW |
1483 | * |
1484 | * Generate traffic such that interested network peers are aware of | |
1485 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1486 | * a device wants to inform the rest of the network about some sort of | |
1487 | * reconfiguration such as a failover event or virtual machine | |
1488 | * migration. | |
1489 | */ | |
1490 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1491 | { |
ee89bab1 | 1492 | rtnl_lock(); |
7061eb8c | 1493 | __netdev_notify_peers(dev); |
ee89bab1 | 1494 | rtnl_unlock(); |
c1da4ac7 | 1495 | } |
ee89bab1 | 1496 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1497 | |
29863d41 WW |
1498 | static int napi_threaded_poll(void *data); |
1499 | ||
1500 | static int napi_kthread_create(struct napi_struct *n) | |
1501 | { | |
1502 | int err = 0; | |
1503 | ||
1504 | /* Create and wake up the kthread once to put it in | |
1505 | * TASK_INTERRUPTIBLE mode to avoid the blocked task | |
1506 | * warning and work with loadavg. | |
1507 | */ | |
1508 | n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", | |
1509 | n->dev->name, n->napi_id); | |
1510 | if (IS_ERR(n->thread)) { | |
1511 | err = PTR_ERR(n->thread); | |
1512 | pr_err("kthread_run failed with err %d\n", err); | |
1513 | n->thread = NULL; | |
1514 | } | |
1515 | ||
1516 | return err; | |
1517 | } | |
1518 | ||
40c900aa | 1519 | static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
1da177e4 | 1520 | { |
d314774c | 1521 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1522 | int ret; |
1da177e4 | 1523 | |
e46b66bc BH |
1524 | ASSERT_RTNL(); |
1525 | ||
bd869245 HK |
1526 | if (!netif_device_present(dev)) { |
1527 | /* may be detached because parent is runtime-suspended */ | |
1528 | if (dev->dev.parent) | |
1529 | pm_runtime_resume(dev->dev.parent); | |
1530 | if (!netif_device_present(dev)) | |
1531 | return -ENODEV; | |
1532 | } | |
1da177e4 | 1533 | |
ca99ca14 NH |
1534 | /* Block netpoll from trying to do any rx path servicing. |
1535 | * If we don't do this there is a chance ndo_poll_controller | |
1536 | * or ndo_poll may be running while we open the device | |
1537 | */ | |
66b5552f | 1538 | netpoll_poll_disable(dev); |
ca99ca14 | 1539 | |
40c900aa | 1540 | ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); |
3b8bcfd5 JB |
1541 | ret = notifier_to_errno(ret); |
1542 | if (ret) | |
1543 | return ret; | |
1544 | ||
1da177e4 | 1545 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1546 | |
d314774c SH |
1547 | if (ops->ndo_validate_addr) |
1548 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1549 | |
d314774c SH |
1550 | if (!ret && ops->ndo_open) |
1551 | ret = ops->ndo_open(dev); | |
1da177e4 | 1552 | |
66b5552f | 1553 | netpoll_poll_enable(dev); |
ca99ca14 | 1554 | |
bada339b JG |
1555 | if (ret) |
1556 | clear_bit(__LINK_STATE_START, &dev->state); | |
1557 | else { | |
1da177e4 | 1558 | dev->flags |= IFF_UP; |
4417da66 | 1559 | dev_set_rx_mode(dev); |
1da177e4 | 1560 | dev_activate(dev); |
7bf23575 | 1561 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1562 | } |
bada339b | 1563 | |
1da177e4 LT |
1564 | return ret; |
1565 | } | |
1566 | ||
1567 | /** | |
bd380811 | 1568 | * dev_open - prepare an interface for use. |
00f54e68 PM |
1569 | * @dev: device to open |
1570 | * @extack: netlink extended ack | |
1da177e4 | 1571 | * |
bd380811 PM |
1572 | * Takes a device from down to up state. The device's private open |
1573 | * function is invoked and then the multicast lists are loaded. Finally | |
1574 | * the device is moved into the up state and a %NETDEV_UP message is | |
1575 | * sent to the netdev notifier chain. | |
1576 | * | |
1577 | * Calling this function on an active interface is a nop. On a failure | |
1578 | * a negative errno code is returned. | |
1da177e4 | 1579 | */ |
00f54e68 | 1580 | int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
bd380811 PM |
1581 | { |
1582 | int ret; | |
1583 | ||
bd380811 PM |
1584 | if (dev->flags & IFF_UP) |
1585 | return 0; | |
1586 | ||
40c900aa | 1587 | ret = __dev_open(dev, extack); |
bd380811 PM |
1588 | if (ret < 0) |
1589 | return ret; | |
1590 | ||
7f294054 | 1591 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1592 | call_netdevice_notifiers(NETDEV_UP, dev); |
1593 | ||
1594 | return ret; | |
1595 | } | |
1596 | EXPORT_SYMBOL(dev_open); | |
1597 | ||
7051b88a | 1598 | static void __dev_close_many(struct list_head *head) |
1da177e4 | 1599 | { |
44345724 | 1600 | struct net_device *dev; |
e46b66bc | 1601 | |
bd380811 | 1602 | ASSERT_RTNL(); |
9d5010db DM |
1603 | might_sleep(); |
1604 | ||
5cde2829 | 1605 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1606 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1607 | netpoll_poll_disable(dev); |
3f4df206 | 1608 | |
44345724 | 1609 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1610 | |
44345724 | 1611 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1612 | |
44345724 OP |
1613 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1614 | * can be even on different cpu. So just clear netif_running(). | |
1615 | * | |
1616 | * dev->stop() will invoke napi_disable() on all of it's | |
1617 | * napi_struct instances on this device. | |
1618 | */ | |
4e857c58 | 1619 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1620 | } |
1da177e4 | 1621 | |
44345724 | 1622 | dev_deactivate_many(head); |
d8b2a4d2 | 1623 | |
5cde2829 | 1624 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1625 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1626 | |
44345724 OP |
1627 | /* |
1628 | * Call the device specific close. This cannot fail. | |
1629 | * Only if device is UP | |
1630 | * | |
1631 | * We allow it to be called even after a DETACH hot-plug | |
1632 | * event. | |
1633 | */ | |
1634 | if (ops->ndo_stop) | |
1635 | ops->ndo_stop(dev); | |
1636 | ||
44345724 | 1637 | dev->flags &= ~IFF_UP; |
66b5552f | 1638 | netpoll_poll_enable(dev); |
44345724 | 1639 | } |
44345724 OP |
1640 | } |
1641 | ||
7051b88a | 1642 | static void __dev_close(struct net_device *dev) |
44345724 OP |
1643 | { |
1644 | LIST_HEAD(single); | |
1645 | ||
5cde2829 | 1646 | list_add(&dev->close_list, &single); |
7051b88a | 1647 | __dev_close_many(&single); |
f87e6f47 | 1648 | list_del(&single); |
44345724 OP |
1649 | } |
1650 | ||
7051b88a | 1651 | void dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1652 | { |
1653 | struct net_device *dev, *tmp; | |
1da177e4 | 1654 | |
5cde2829 EB |
1655 | /* Remove the devices that don't need to be closed */ |
1656 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1657 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1658 | list_del_init(&dev->close_list); |
44345724 OP |
1659 | |
1660 | __dev_close_many(head); | |
1da177e4 | 1661 | |
5cde2829 | 1662 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1663 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1664 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1665 | if (unlink) |
1666 | list_del_init(&dev->close_list); | |
44345724 | 1667 | } |
bd380811 | 1668 | } |
99c4a26a | 1669 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1670 | |
1671 | /** | |
1672 | * dev_close - shutdown an interface. | |
1673 | * @dev: device to shutdown | |
1674 | * | |
1675 | * This function moves an active device into down state. A | |
1676 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1677 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1678 | * chain. | |
1679 | */ | |
7051b88a | 1680 | void dev_close(struct net_device *dev) |
bd380811 | 1681 | { |
e14a5993 ED |
1682 | if (dev->flags & IFF_UP) { |
1683 | LIST_HEAD(single); | |
1da177e4 | 1684 | |
5cde2829 | 1685 | list_add(&dev->close_list, &single); |
99c4a26a | 1686 | dev_close_many(&single, true); |
e14a5993 ED |
1687 | list_del(&single); |
1688 | } | |
1da177e4 | 1689 | } |
d1b19dff | 1690 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1691 | |
1692 | ||
0187bdfb BH |
1693 | /** |
1694 | * dev_disable_lro - disable Large Receive Offload on a device | |
1695 | * @dev: device | |
1696 | * | |
1697 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1698 | * called under RTNL. This is needed if received packets may be | |
1699 | * forwarded to another interface. | |
1700 | */ | |
1701 | void dev_disable_lro(struct net_device *dev) | |
1702 | { | |
fbe168ba MK |
1703 | struct net_device *lower_dev; |
1704 | struct list_head *iter; | |
529d0489 | 1705 | |
bc5787c6 MM |
1706 | dev->wanted_features &= ~NETIF_F_LRO; |
1707 | netdev_update_features(dev); | |
27660515 | 1708 | |
22d5969f MM |
1709 | if (unlikely(dev->features & NETIF_F_LRO)) |
1710 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1711 | |
1712 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1713 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1714 | } |
1715 | EXPORT_SYMBOL(dev_disable_lro); | |
1716 | ||
56f5aa77 MC |
1717 | /** |
1718 | * dev_disable_gro_hw - disable HW Generic Receive Offload on a device | |
1719 | * @dev: device | |
1720 | * | |
1721 | * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be | |
1722 | * called under RTNL. This is needed if Generic XDP is installed on | |
1723 | * the device. | |
1724 | */ | |
1725 | static void dev_disable_gro_hw(struct net_device *dev) | |
1726 | { | |
1727 | dev->wanted_features &= ~NETIF_F_GRO_HW; | |
1728 | netdev_update_features(dev); | |
1729 | ||
1730 | if (unlikely(dev->features & NETIF_F_GRO_HW)) | |
1731 | netdev_WARN(dev, "failed to disable GRO_HW!\n"); | |
1732 | } | |
1733 | ||
ede2762d KT |
1734 | const char *netdev_cmd_to_name(enum netdev_cmd cmd) |
1735 | { | |
1736 | #define N(val) \ | |
1737 | case NETDEV_##val: \ | |
1738 | return "NETDEV_" __stringify(val); | |
1739 | switch (cmd) { | |
1740 | N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) | |
1741 | N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) | |
1742 | N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) | |
1743 | N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) | |
1744 | N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) | |
1745 | N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) | |
1746 | N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) | |
9daae9bd GP |
1747 | N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) |
1748 | N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) | |
1570415f | 1749 | N(PRE_CHANGEADDR) |
3f5ecd8a | 1750 | } |
ede2762d KT |
1751 | #undef N |
1752 | return "UNKNOWN_NETDEV_EVENT"; | |
1753 | } | |
1754 | EXPORT_SYMBOL_GPL(netdev_cmd_to_name); | |
1755 | ||
351638e7 JP |
1756 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1757 | struct net_device *dev) | |
1758 | { | |
51d0c047 DA |
1759 | struct netdev_notifier_info info = { |
1760 | .dev = dev, | |
1761 | }; | |
351638e7 | 1762 | |
351638e7 JP |
1763 | return nb->notifier_call(nb, val, &info); |
1764 | } | |
0187bdfb | 1765 | |
afa0df59 JP |
1766 | static int call_netdevice_register_notifiers(struct notifier_block *nb, |
1767 | struct net_device *dev) | |
1768 | { | |
1769 | int err; | |
1770 | ||
1771 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); | |
1772 | err = notifier_to_errno(err); | |
1773 | if (err) | |
1774 | return err; | |
1775 | ||
1776 | if (!(dev->flags & IFF_UP)) | |
1777 | return 0; | |
1778 | ||
1779 | call_netdevice_notifier(nb, NETDEV_UP, dev); | |
1780 | return 0; | |
1781 | } | |
1782 | ||
1783 | static void call_netdevice_unregister_notifiers(struct notifier_block *nb, | |
1784 | struct net_device *dev) | |
1785 | { | |
1786 | if (dev->flags & IFF_UP) { | |
1787 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, | |
1788 | dev); | |
1789 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
1790 | } | |
1791 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); | |
1792 | } | |
1793 | ||
1794 | static int call_netdevice_register_net_notifiers(struct notifier_block *nb, | |
1795 | struct net *net) | |
1796 | { | |
1797 | struct net_device *dev; | |
1798 | int err; | |
1799 | ||
1800 | for_each_netdev(net, dev) { | |
1801 | err = call_netdevice_register_notifiers(nb, dev); | |
1802 | if (err) | |
1803 | goto rollback; | |
1804 | } | |
1805 | return 0; | |
1806 | ||
1807 | rollback: | |
1808 | for_each_netdev_continue_reverse(net, dev) | |
1809 | call_netdevice_unregister_notifiers(nb, dev); | |
1810 | return err; | |
1811 | } | |
1812 | ||
1813 | static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, | |
1814 | struct net *net) | |
1815 | { | |
1816 | struct net_device *dev; | |
1817 | ||
1818 | for_each_netdev(net, dev) | |
1819 | call_netdevice_unregister_notifiers(nb, dev); | |
1820 | } | |
1821 | ||
881d966b EB |
1822 | static int dev_boot_phase = 1; |
1823 | ||
1da177e4 | 1824 | /** |
722c9a0c | 1825 | * register_netdevice_notifier - register a network notifier block |
1826 | * @nb: notifier | |
1da177e4 | 1827 | * |
722c9a0c | 1828 | * Register a notifier to be called when network device events occur. |
1829 | * The notifier passed is linked into the kernel structures and must | |
1830 | * not be reused until it has been unregistered. A negative errno code | |
1831 | * is returned on a failure. | |
1da177e4 | 1832 | * |
722c9a0c | 1833 | * When registered all registration and up events are replayed |
1834 | * to the new notifier to allow device to have a race free | |
1835 | * view of the network device list. | |
1da177e4 LT |
1836 | */ |
1837 | ||
1838 | int register_netdevice_notifier(struct notifier_block *nb) | |
1839 | { | |
881d966b | 1840 | struct net *net; |
1da177e4 LT |
1841 | int err; |
1842 | ||
328fbe74 KT |
1843 | /* Close race with setup_net() and cleanup_net() */ |
1844 | down_write(&pernet_ops_rwsem); | |
1da177e4 | 1845 | rtnl_lock(); |
f07d5b94 | 1846 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1847 | if (err) |
1848 | goto unlock; | |
881d966b EB |
1849 | if (dev_boot_phase) |
1850 | goto unlock; | |
1851 | for_each_net(net) { | |
afa0df59 JP |
1852 | err = call_netdevice_register_net_notifiers(nb, net); |
1853 | if (err) | |
1854 | goto rollback; | |
1da177e4 | 1855 | } |
fcc5a03a HX |
1856 | |
1857 | unlock: | |
1da177e4 | 1858 | rtnl_unlock(); |
328fbe74 | 1859 | up_write(&pernet_ops_rwsem); |
1da177e4 | 1860 | return err; |
fcc5a03a HX |
1861 | |
1862 | rollback: | |
afa0df59 JP |
1863 | for_each_net_continue_reverse(net) |
1864 | call_netdevice_unregister_net_notifiers(nb, net); | |
c67625a1 PE |
1865 | |
1866 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1867 | goto unlock; |
1da177e4 | 1868 | } |
d1b19dff | 1869 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1870 | |
1871 | /** | |
722c9a0c | 1872 | * unregister_netdevice_notifier - unregister a network notifier block |
1873 | * @nb: notifier | |
1da177e4 | 1874 | * |
722c9a0c | 1875 | * Unregister a notifier previously registered by |
1876 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1877 | * kernel structures and may then be reused. A negative errno code | |
1878 | * is returned on a failure. | |
7d3d43da | 1879 | * |
722c9a0c | 1880 | * After unregistering unregister and down device events are synthesized |
1881 | * for all devices on the device list to the removed notifier to remove | |
1882 | * the need for special case cleanup code. | |
1da177e4 LT |
1883 | */ |
1884 | ||
1885 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1886 | { | |
7d3d43da | 1887 | struct net *net; |
9f514950 HX |
1888 | int err; |
1889 | ||
328fbe74 KT |
1890 | /* Close race with setup_net() and cleanup_net() */ |
1891 | down_write(&pernet_ops_rwsem); | |
9f514950 | 1892 | rtnl_lock(); |
f07d5b94 | 1893 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1894 | if (err) |
1895 | goto unlock; | |
1896 | ||
48b3a137 JP |
1897 | for_each_net(net) |
1898 | call_netdevice_unregister_net_notifiers(nb, net); | |
1899 | ||
7d3d43da | 1900 | unlock: |
9f514950 | 1901 | rtnl_unlock(); |
328fbe74 | 1902 | up_write(&pernet_ops_rwsem); |
9f514950 | 1903 | return err; |
1da177e4 | 1904 | } |
d1b19dff | 1905 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1906 | |
1f637703 JP |
1907 | static int __register_netdevice_notifier_net(struct net *net, |
1908 | struct notifier_block *nb, | |
1909 | bool ignore_call_fail) | |
1910 | { | |
1911 | int err; | |
1912 | ||
1913 | err = raw_notifier_chain_register(&net->netdev_chain, nb); | |
1914 | if (err) | |
1915 | return err; | |
1916 | if (dev_boot_phase) | |
1917 | return 0; | |
1918 | ||
1919 | err = call_netdevice_register_net_notifiers(nb, net); | |
1920 | if (err && !ignore_call_fail) | |
1921 | goto chain_unregister; | |
1922 | ||
1923 | return 0; | |
1924 | ||
1925 | chain_unregister: | |
1926 | raw_notifier_chain_unregister(&net->netdev_chain, nb); | |
1927 | return err; | |
1928 | } | |
1929 | ||
1930 | static int __unregister_netdevice_notifier_net(struct net *net, | |
1931 | struct notifier_block *nb) | |
1932 | { | |
1933 | int err; | |
1934 | ||
1935 | err = raw_notifier_chain_unregister(&net->netdev_chain, nb); | |
1936 | if (err) | |
1937 | return err; | |
1938 | ||
1939 | call_netdevice_unregister_net_notifiers(nb, net); | |
1940 | return 0; | |
1941 | } | |
1942 | ||
a30c7b42 JP |
1943 | /** |
1944 | * register_netdevice_notifier_net - register a per-netns network notifier block | |
1945 | * @net: network namespace | |
1946 | * @nb: notifier | |
1947 | * | |
1948 | * Register a notifier to be called when network device events occur. | |
1949 | * The notifier passed is linked into the kernel structures and must | |
1950 | * not be reused until it has been unregistered. A negative errno code | |
1951 | * is returned on a failure. | |
1952 | * | |
1953 | * When registered all registration and up events are replayed | |
1954 | * to the new notifier to allow device to have a race free | |
1955 | * view of the network device list. | |
1956 | */ | |
1957 | ||
1958 | int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) | |
1959 | { | |
1960 | int err; | |
1961 | ||
1962 | rtnl_lock(); | |
1f637703 | 1963 | err = __register_netdevice_notifier_net(net, nb, false); |
a30c7b42 JP |
1964 | rtnl_unlock(); |
1965 | return err; | |
a30c7b42 JP |
1966 | } |
1967 | EXPORT_SYMBOL(register_netdevice_notifier_net); | |
1968 | ||
1969 | /** | |
1970 | * unregister_netdevice_notifier_net - unregister a per-netns | |
1971 | * network notifier block | |
1972 | * @net: network namespace | |
1973 | * @nb: notifier | |
1974 | * | |
1975 | * Unregister a notifier previously registered by | |
1976 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1977 | * kernel structures and may then be reused. A negative errno code | |
1978 | * is returned on a failure. | |
1979 | * | |
1980 | * After unregistering unregister and down device events are synthesized | |
1981 | * for all devices on the device list to the removed notifier to remove | |
1982 | * the need for special case cleanup code. | |
1983 | */ | |
1984 | ||
1985 | int unregister_netdevice_notifier_net(struct net *net, | |
1986 | struct notifier_block *nb) | |
1987 | { | |
1988 | int err; | |
1989 | ||
1990 | rtnl_lock(); | |
1f637703 | 1991 | err = __unregister_netdevice_notifier_net(net, nb); |
a30c7b42 JP |
1992 | rtnl_unlock(); |
1993 | return err; | |
1994 | } | |
1995 | EXPORT_SYMBOL(unregister_netdevice_notifier_net); | |
a30c7b42 | 1996 | |
93642e14 JP |
1997 | int register_netdevice_notifier_dev_net(struct net_device *dev, |
1998 | struct notifier_block *nb, | |
1999 | struct netdev_net_notifier *nn) | |
2000 | { | |
2001 | int err; | |
a30c7b42 | 2002 | |
93642e14 JP |
2003 | rtnl_lock(); |
2004 | err = __register_netdevice_notifier_net(dev_net(dev), nb, false); | |
2005 | if (!err) { | |
2006 | nn->nb = nb; | |
2007 | list_add(&nn->list, &dev->net_notifier_list); | |
2008 | } | |
a30c7b42 JP |
2009 | rtnl_unlock(); |
2010 | return err; | |
2011 | } | |
93642e14 JP |
2012 | EXPORT_SYMBOL(register_netdevice_notifier_dev_net); |
2013 | ||
2014 | int unregister_netdevice_notifier_dev_net(struct net_device *dev, | |
2015 | struct notifier_block *nb, | |
2016 | struct netdev_net_notifier *nn) | |
2017 | { | |
2018 | int err; | |
2019 | ||
2020 | rtnl_lock(); | |
2021 | list_del(&nn->list); | |
2022 | err = __unregister_netdevice_notifier_net(dev_net(dev), nb); | |
2023 | rtnl_unlock(); | |
2024 | return err; | |
2025 | } | |
2026 | EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); | |
2027 | ||
2028 | static void move_netdevice_notifiers_dev_net(struct net_device *dev, | |
2029 | struct net *net) | |
2030 | { | |
2031 | struct netdev_net_notifier *nn; | |
2032 | ||
2033 | list_for_each_entry(nn, &dev->net_notifier_list, list) { | |
2034 | __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); | |
2035 | __register_netdevice_notifier_net(net, nn->nb, true); | |
2036 | } | |
2037 | } | |
a30c7b42 | 2038 | |
351638e7 JP |
2039 | /** |
2040 | * call_netdevice_notifiers_info - call all network notifier blocks | |
2041 | * @val: value passed unmodified to notifier function | |
351638e7 JP |
2042 | * @info: notifier information data |
2043 | * | |
2044 | * Call all network notifier blocks. Parameters and return value | |
2045 | * are as for raw_notifier_call_chain(). | |
2046 | */ | |
2047 | ||
1d143d9f | 2048 | static int call_netdevice_notifiers_info(unsigned long val, |
1d143d9f | 2049 | struct netdev_notifier_info *info) |
351638e7 | 2050 | { |
a30c7b42 JP |
2051 | struct net *net = dev_net(info->dev); |
2052 | int ret; | |
2053 | ||
351638e7 | 2054 | ASSERT_RTNL(); |
a30c7b42 JP |
2055 | |
2056 | /* Run per-netns notifier block chain first, then run the global one. | |
2057 | * Hopefully, one day, the global one is going to be removed after | |
2058 | * all notifier block registrators get converted to be per-netns. | |
2059 | */ | |
2060 | ret = raw_notifier_call_chain(&net->netdev_chain, val, info); | |
2061 | if (ret & NOTIFY_STOP_MASK) | |
2062 | return ret; | |
351638e7 JP |
2063 | return raw_notifier_call_chain(&netdev_chain, val, info); |
2064 | } | |
351638e7 | 2065 | |
26372605 PM |
2066 | static int call_netdevice_notifiers_extack(unsigned long val, |
2067 | struct net_device *dev, | |
2068 | struct netlink_ext_ack *extack) | |
2069 | { | |
2070 | struct netdev_notifier_info info = { | |
2071 | .dev = dev, | |
2072 | .extack = extack, | |
2073 | }; | |
2074 | ||
2075 | return call_netdevice_notifiers_info(val, &info); | |
2076 | } | |
2077 | ||
1da177e4 LT |
2078 | /** |
2079 | * call_netdevice_notifiers - call all network notifier blocks | |
2080 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 2081 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
2082 | * |
2083 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 2084 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
2085 | */ |
2086 | ||
ad7379d4 | 2087 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 2088 | { |
26372605 | 2089 | return call_netdevice_notifiers_extack(val, dev, NULL); |
1da177e4 | 2090 | } |
edf947f1 | 2091 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 2092 | |
af7d6cce SD |
2093 | /** |
2094 | * call_netdevice_notifiers_mtu - call all network notifier blocks | |
2095 | * @val: value passed unmodified to notifier function | |
2096 | * @dev: net_device pointer passed unmodified to notifier function | |
2097 | * @arg: additional u32 argument passed to the notifier function | |
2098 | * | |
2099 | * Call all network notifier blocks. Parameters and return value | |
2100 | * are as for raw_notifier_call_chain(). | |
2101 | */ | |
2102 | static int call_netdevice_notifiers_mtu(unsigned long val, | |
2103 | struct net_device *dev, u32 arg) | |
2104 | { | |
2105 | struct netdev_notifier_info_ext info = { | |
2106 | .info.dev = dev, | |
2107 | .ext.mtu = arg, | |
2108 | }; | |
2109 | ||
2110 | BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); | |
2111 | ||
2112 | return call_netdevice_notifiers_info(val, &info.info); | |
2113 | } | |
2114 | ||
1cf51900 | 2115 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 2116 | static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); |
4577139b DB |
2117 | |
2118 | void net_inc_ingress_queue(void) | |
2119 | { | |
aabf6772 | 2120 | static_branch_inc(&ingress_needed_key); |
4577139b DB |
2121 | } |
2122 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
2123 | ||
2124 | void net_dec_ingress_queue(void) | |
2125 | { | |
aabf6772 | 2126 | static_branch_dec(&ingress_needed_key); |
4577139b DB |
2127 | } |
2128 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
2129 | #endif | |
2130 | ||
1f211a1b | 2131 | #ifdef CONFIG_NET_EGRESS |
aabf6772 | 2132 | static DEFINE_STATIC_KEY_FALSE(egress_needed_key); |
1f211a1b DB |
2133 | |
2134 | void net_inc_egress_queue(void) | |
2135 | { | |
aabf6772 | 2136 | static_branch_inc(&egress_needed_key); |
1f211a1b DB |
2137 | } |
2138 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
2139 | ||
2140 | void net_dec_egress_queue(void) | |
2141 | { | |
aabf6772 | 2142 | static_branch_dec(&egress_needed_key); |
1f211a1b DB |
2143 | } |
2144 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
2145 | #endif | |
2146 | ||
39e83922 | 2147 | static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); |
e9666d10 | 2148 | #ifdef CONFIG_JUMP_LABEL |
b90e5794 | 2149 | static atomic_t netstamp_needed_deferred; |
13baa00a | 2150 | static atomic_t netstamp_wanted; |
5fa8bbda | 2151 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 2152 | { |
b90e5794 | 2153 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 2154 | int wanted; |
b90e5794 | 2155 | |
13baa00a ED |
2156 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
2157 | if (wanted > 0) | |
39e83922 | 2158 | static_branch_enable(&netstamp_needed_key); |
13baa00a | 2159 | else |
39e83922 | 2160 | static_branch_disable(&netstamp_needed_key); |
5fa8bbda ED |
2161 | } |
2162 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 2163 | #endif |
5fa8bbda ED |
2164 | |
2165 | void net_enable_timestamp(void) | |
2166 | { | |
e9666d10 | 2167 | #ifdef CONFIG_JUMP_LABEL |
13baa00a ED |
2168 | int wanted; |
2169 | ||
2170 | while (1) { | |
2171 | wanted = atomic_read(&netstamp_wanted); | |
2172 | if (wanted <= 0) | |
2173 | break; | |
2174 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) | |
2175 | return; | |
2176 | } | |
2177 | atomic_inc(&netstamp_needed_deferred); | |
2178 | schedule_work(&netstamp_work); | |
2179 | #else | |
39e83922 | 2180 | static_branch_inc(&netstamp_needed_key); |
13baa00a | 2181 | #endif |
1da177e4 | 2182 | } |
d1b19dff | 2183 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
2184 | |
2185 | void net_disable_timestamp(void) | |
2186 | { | |
e9666d10 | 2187 | #ifdef CONFIG_JUMP_LABEL |
13baa00a ED |
2188 | int wanted; |
2189 | ||
2190 | while (1) { | |
2191 | wanted = atomic_read(&netstamp_wanted); | |
2192 | if (wanted <= 1) | |
2193 | break; | |
2194 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) | |
2195 | return; | |
2196 | } | |
2197 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
2198 | schedule_work(&netstamp_work); |
2199 | #else | |
39e83922 | 2200 | static_branch_dec(&netstamp_needed_key); |
5fa8bbda | 2201 | #endif |
1da177e4 | 2202 | } |
d1b19dff | 2203 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 2204 | |
3b098e2d | 2205 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 2206 | { |
2456e855 | 2207 | skb->tstamp = 0; |
39e83922 | 2208 | if (static_branch_unlikely(&netstamp_needed_key)) |
a61bbcf2 | 2209 | __net_timestamp(skb); |
1da177e4 LT |
2210 | } |
2211 | ||
39e83922 DB |
2212 | #define net_timestamp_check(COND, SKB) \ |
2213 | if (static_branch_unlikely(&netstamp_needed_key)) { \ | |
2214 | if ((COND) && !(SKB)->tstamp) \ | |
2215 | __net_timestamp(SKB); \ | |
2216 | } \ | |
3b098e2d | 2217 | |
f4b05d27 | 2218 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 | 2219 | { |
5f7d5728 | 2220 | return __is_skb_forwardable(dev, skb, true); |
79b569f0 | 2221 | } |
1ee481fb | 2222 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 2223 | |
5f7d5728 JDB |
2224 | static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, |
2225 | bool check_mtu) | |
a0265d28 | 2226 | { |
5f7d5728 | 2227 | int ret = ____dev_forward_skb(dev, skb, check_mtu); |
a0265d28 | 2228 | |
4e3264d2 MKL |
2229 | if (likely(!ret)) { |
2230 | skb->protocol = eth_type_trans(skb, dev); | |
2231 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
2232 | } | |
a0265d28 | 2233 | |
4e3264d2 | 2234 | return ret; |
a0265d28 | 2235 | } |
5f7d5728 JDB |
2236 | |
2237 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
2238 | { | |
2239 | return __dev_forward_skb2(dev, skb, true); | |
2240 | } | |
a0265d28 HX |
2241 | EXPORT_SYMBOL_GPL(__dev_forward_skb); |
2242 | ||
44540960 AB |
2243 | /** |
2244 | * dev_forward_skb - loopback an skb to another netif | |
2245 | * | |
2246 | * @dev: destination network device | |
2247 | * @skb: buffer to forward | |
2248 | * | |
2249 | * return values: | |
2250 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 2251 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
2252 | * |
2253 | * dev_forward_skb can be used for injecting an skb from the | |
2254 | * start_xmit function of one device into the receive queue | |
2255 | * of another device. | |
2256 | * | |
2257 | * The receiving device may be in another namespace, so | |
2258 | * we have to clear all information in the skb that could | |
2259 | * impact namespace isolation. | |
2260 | */ | |
2261 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
2262 | { | |
a0265d28 | 2263 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
2264 | } |
2265 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
2266 | ||
5f7d5728 JDB |
2267 | int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) |
2268 | { | |
2269 | return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); | |
2270 | } | |
2271 | ||
71d9dec2 CG |
2272 | static inline int deliver_skb(struct sk_buff *skb, |
2273 | struct packet_type *pt_prev, | |
2274 | struct net_device *orig_dev) | |
2275 | { | |
1f8b977a | 2276 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
1080e512 | 2277 | return -ENOMEM; |
63354797 | 2278 | refcount_inc(&skb->users); |
71d9dec2 CG |
2279 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
2280 | } | |
2281 | ||
7866a621 SN |
2282 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
2283 | struct packet_type **pt, | |
fbcb2170 JP |
2284 | struct net_device *orig_dev, |
2285 | __be16 type, | |
7866a621 SN |
2286 | struct list_head *ptype_list) |
2287 | { | |
2288 | struct packet_type *ptype, *pt_prev = *pt; | |
2289 | ||
2290 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
2291 | if (ptype->type != type) | |
2292 | continue; | |
2293 | if (pt_prev) | |
fbcb2170 | 2294 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
2295 | pt_prev = ptype; |
2296 | } | |
2297 | *pt = pt_prev; | |
2298 | } | |
2299 | ||
c0de08d0 EL |
2300 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
2301 | { | |
a3d744e9 | 2302 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
2303 | return false; |
2304 | ||
2305 | if (ptype->id_match) | |
2306 | return ptype->id_match(ptype, skb->sk); | |
2307 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
2308 | return true; | |
2309 | ||
2310 | return false; | |
2311 | } | |
2312 | ||
9f9a742d MR |
2313 | /** |
2314 | * dev_nit_active - return true if any network interface taps are in use | |
2315 | * | |
2316 | * @dev: network device to check for the presence of taps | |
2317 | */ | |
2318 | bool dev_nit_active(struct net_device *dev) | |
2319 | { | |
2320 | return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); | |
2321 | } | |
2322 | EXPORT_SYMBOL_GPL(dev_nit_active); | |
2323 | ||
1da177e4 LT |
2324 | /* |
2325 | * Support routine. Sends outgoing frames to any network | |
2326 | * taps currently in use. | |
2327 | */ | |
2328 | ||
74b20582 | 2329 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
2330 | { |
2331 | struct packet_type *ptype; | |
71d9dec2 CG |
2332 | struct sk_buff *skb2 = NULL; |
2333 | struct packet_type *pt_prev = NULL; | |
7866a621 | 2334 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 2335 | |
1da177e4 | 2336 | rcu_read_lock(); |
7866a621 SN |
2337 | again: |
2338 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
fa788d98 VW |
2339 | if (ptype->ignore_outgoing) |
2340 | continue; | |
2341 | ||
1da177e4 LT |
2342 | /* Never send packets back to the socket |
2343 | * they originated from - MvS (miquels@drinkel.ow.org) | |
2344 | */ | |
7866a621 SN |
2345 | if (skb_loop_sk(ptype, skb)) |
2346 | continue; | |
71d9dec2 | 2347 | |
7866a621 SN |
2348 | if (pt_prev) { |
2349 | deliver_skb(skb2, pt_prev, skb->dev); | |
2350 | pt_prev = ptype; | |
2351 | continue; | |
2352 | } | |
1da177e4 | 2353 | |
7866a621 SN |
2354 | /* need to clone skb, done only once */ |
2355 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
2356 | if (!skb2) | |
2357 | goto out_unlock; | |
70978182 | 2358 | |
7866a621 | 2359 | net_timestamp_set(skb2); |
1da177e4 | 2360 | |
7866a621 SN |
2361 | /* skb->nh should be correctly |
2362 | * set by sender, so that the second statement is | |
2363 | * just protection against buggy protocols. | |
2364 | */ | |
2365 | skb_reset_mac_header(skb2); | |
2366 | ||
2367 | if (skb_network_header(skb2) < skb2->data || | |
2368 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
2369 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
2370 | ntohs(skb2->protocol), | |
2371 | dev->name); | |
2372 | skb_reset_network_header(skb2); | |
1da177e4 | 2373 | } |
7866a621 SN |
2374 | |
2375 | skb2->transport_header = skb2->network_header; | |
2376 | skb2->pkt_type = PACKET_OUTGOING; | |
2377 | pt_prev = ptype; | |
2378 | } | |
2379 | ||
2380 | if (ptype_list == &ptype_all) { | |
2381 | ptype_list = &dev->ptype_all; | |
2382 | goto again; | |
1da177e4 | 2383 | } |
7866a621 | 2384 | out_unlock: |
581fe0ea WB |
2385 | if (pt_prev) { |
2386 | if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) | |
2387 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
2388 | else | |
2389 | kfree_skb(skb2); | |
2390 | } | |
1da177e4 LT |
2391 | rcu_read_unlock(); |
2392 | } | |
74b20582 | 2393 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 2394 | |
2c53040f BH |
2395 | /** |
2396 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
2397 | * @dev: Network device |
2398 | * @txq: number of queues available | |
2399 | * | |
2400 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
2401 | * valid. To resolve this verify the tc mapping remains valid and if | |
2402 | * not NULL the mapping. With no priorities mapping to this | |
2403 | * offset/count pair it will no longer be used. In the worst case TC0 | |
2404 | * is invalid nothing can be done so disable priority mappings. If is | |
2405 | * expected that drivers will fix this mapping if they can before | |
2406 | * calling netif_set_real_num_tx_queues. | |
2407 | */ | |
bb134d22 | 2408 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
2409 | { |
2410 | int i; | |
2411 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2412 | ||
2413 | /* If TC0 is invalidated disable TC mapping */ | |
2414 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 2415 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
2416 | dev->num_tc = 0; |
2417 | return; | |
2418 | } | |
2419 | ||
2420 | /* Invalidated prio to tc mappings set to TC0 */ | |
2421 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
2422 | int q = netdev_get_prio_tc_map(dev, i); | |
2423 | ||
2424 | tc = &dev->tc_to_txq[q]; | |
2425 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
2426 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
2427 | i, q); | |
4f57c087 JF |
2428 | netdev_set_prio_tc_map(dev, i, 0); |
2429 | } | |
2430 | } | |
2431 | } | |
2432 | ||
8d059b0f AD |
2433 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
2434 | { | |
2435 | if (dev->num_tc) { | |
2436 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2437 | int i; | |
2438 | ||
ffcfe25b | 2439 | /* walk through the TCs and see if it falls into any of them */ |
8d059b0f AD |
2440 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { |
2441 | if ((txq - tc->offset) < tc->count) | |
2442 | return i; | |
2443 | } | |
2444 | ||
ffcfe25b | 2445 | /* didn't find it, just return -1 to indicate no match */ |
8d059b0f AD |
2446 | return -1; |
2447 | } | |
2448 | ||
2449 | return 0; | |
2450 | } | |
8a5f2166 | 2451 | EXPORT_SYMBOL(netdev_txq_to_tc); |
8d059b0f | 2452 | |
537c00de | 2453 | #ifdef CONFIG_XPS |
04157469 AN |
2454 | struct static_key xps_needed __read_mostly; |
2455 | EXPORT_SYMBOL(xps_needed); | |
2456 | struct static_key xps_rxqs_needed __read_mostly; | |
2457 | EXPORT_SYMBOL(xps_rxqs_needed); | |
537c00de AD |
2458 | static DEFINE_MUTEX(xps_map_mutex); |
2459 | #define xmap_dereference(P) \ | |
2460 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
2461 | ||
6234f874 AD |
2462 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
2463 | int tci, u16 index) | |
537c00de | 2464 | { |
10cdc3f3 AD |
2465 | struct xps_map *map = NULL; |
2466 | int pos; | |
537c00de | 2467 | |
10cdc3f3 | 2468 | if (dev_maps) |
80d19669 | 2469 | map = xmap_dereference(dev_maps->attr_map[tci]); |
6234f874 AD |
2470 | if (!map) |
2471 | return false; | |
537c00de | 2472 | |
6234f874 AD |
2473 | for (pos = map->len; pos--;) { |
2474 | if (map->queues[pos] != index) | |
2475 | continue; | |
2476 | ||
2477 | if (map->len > 1) { | |
2478 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2479 | break; |
537c00de | 2480 | } |
6234f874 | 2481 | |
80d19669 | 2482 | RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); |
6234f874 AD |
2483 | kfree_rcu(map, rcu); |
2484 | return false; | |
537c00de AD |
2485 | } |
2486 | ||
6234f874 | 2487 | return true; |
10cdc3f3 AD |
2488 | } |
2489 | ||
6234f874 AD |
2490 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2491 | struct xps_dev_maps *dev_maps, | |
2492 | int cpu, u16 offset, u16 count) | |
2493 | { | |
184c449f AD |
2494 | int num_tc = dev->num_tc ? : 1; |
2495 | bool active = false; | |
2496 | int tci; | |
6234f874 | 2497 | |
184c449f AD |
2498 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2499 | int i, j; | |
2500 | ||
2501 | for (i = count, j = offset; i--; j++) { | |
6358d49a | 2502 | if (!remove_xps_queue(dev_maps, tci, j)) |
184c449f AD |
2503 | break; |
2504 | } | |
2505 | ||
2506 | active |= i < 0; | |
6234f874 AD |
2507 | } |
2508 | ||
184c449f | 2509 | return active; |
6234f874 AD |
2510 | } |
2511 | ||
867d0ad4 SD |
2512 | static void reset_xps_maps(struct net_device *dev, |
2513 | struct xps_dev_maps *dev_maps, | |
2514 | bool is_rxqs_map) | |
2515 | { | |
2516 | if (is_rxqs_map) { | |
2517 | static_key_slow_dec_cpuslocked(&xps_rxqs_needed); | |
2518 | RCU_INIT_POINTER(dev->xps_rxqs_map, NULL); | |
2519 | } else { | |
2520 | RCU_INIT_POINTER(dev->xps_cpus_map, NULL); | |
2521 | } | |
2522 | static_key_slow_dec_cpuslocked(&xps_needed); | |
2523 | kfree_rcu(dev_maps, rcu); | |
2524 | } | |
2525 | ||
80d19669 AN |
2526 | static void clean_xps_maps(struct net_device *dev, const unsigned long *mask, |
2527 | struct xps_dev_maps *dev_maps, unsigned int nr_ids, | |
2528 | u16 offset, u16 count, bool is_rxqs_map) | |
2529 | { | |
2530 | bool active = false; | |
2531 | int i, j; | |
2532 | ||
2533 | for (j = -1; j = netif_attrmask_next(j, mask, nr_ids), | |
2534 | j < nr_ids;) | |
2535 | active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, | |
2536 | count); | |
867d0ad4 SD |
2537 | if (!active) |
2538 | reset_xps_maps(dev, dev_maps, is_rxqs_map); | |
80d19669 | 2539 | |
f28c020f SD |
2540 | if (!is_rxqs_map) { |
2541 | for (i = offset + (count - 1); count--; i--) { | |
2542 | netdev_queue_numa_node_write( | |
2543 | netdev_get_tx_queue(dev, i), | |
2544 | NUMA_NO_NODE); | |
80d19669 | 2545 | } |
80d19669 AN |
2546 | } |
2547 | } | |
2548 | ||
6234f874 AD |
2549 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, |
2550 | u16 count) | |
10cdc3f3 | 2551 | { |
80d19669 | 2552 | const unsigned long *possible_mask = NULL; |
10cdc3f3 | 2553 | struct xps_dev_maps *dev_maps; |
80d19669 | 2554 | unsigned int nr_ids; |
10cdc3f3 | 2555 | |
04157469 AN |
2556 | if (!static_key_false(&xps_needed)) |
2557 | return; | |
10cdc3f3 | 2558 | |
4d99f660 | 2559 | cpus_read_lock(); |
04157469 | 2560 | mutex_lock(&xps_map_mutex); |
10cdc3f3 | 2561 | |
04157469 AN |
2562 | if (static_key_false(&xps_rxqs_needed)) { |
2563 | dev_maps = xmap_dereference(dev->xps_rxqs_map); | |
2564 | if (dev_maps) { | |
2565 | nr_ids = dev->num_rx_queues; | |
2566 | clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, | |
2567 | offset, count, true); | |
2568 | } | |
537c00de AD |
2569 | } |
2570 | ||
80d19669 AN |
2571 | dev_maps = xmap_dereference(dev->xps_cpus_map); |
2572 | if (!dev_maps) | |
2573 | goto out_no_maps; | |
2574 | ||
2575 | if (num_possible_cpus() > 1) | |
2576 | possible_mask = cpumask_bits(cpu_possible_mask); | |
2577 | nr_ids = nr_cpu_ids; | |
2578 | clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count, | |
2579 | false); | |
024e9679 | 2580 | |
537c00de AD |
2581 | out_no_maps: |
2582 | mutex_unlock(&xps_map_mutex); | |
4d99f660 | 2583 | cpus_read_unlock(); |
537c00de AD |
2584 | } |
2585 | ||
6234f874 AD |
2586 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2587 | { | |
2588 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2589 | } | |
2590 | ||
80d19669 AN |
2591 | static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, |
2592 | u16 index, bool is_rxqs_map) | |
01c5f864 AD |
2593 | { |
2594 | struct xps_map *new_map; | |
2595 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2596 | int i, pos; | |
2597 | ||
2598 | for (pos = 0; map && pos < map->len; pos++) { | |
2599 | if (map->queues[pos] != index) | |
2600 | continue; | |
2601 | return map; | |
2602 | } | |
2603 | ||
80d19669 | 2604 | /* Need to add tx-queue to this CPU's/rx-queue's existing map */ |
01c5f864 AD |
2605 | if (map) { |
2606 | if (pos < map->alloc_len) | |
2607 | return map; | |
2608 | ||
2609 | alloc_len = map->alloc_len * 2; | |
2610 | } | |
2611 | ||
80d19669 AN |
2612 | /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's |
2613 | * map | |
2614 | */ | |
2615 | if (is_rxqs_map) | |
2616 | new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); | |
2617 | else | |
2618 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2619 | cpu_to_node(attr_index)); | |
01c5f864 AD |
2620 | if (!new_map) |
2621 | return NULL; | |
2622 | ||
2623 | for (i = 0; i < pos; i++) | |
2624 | new_map->queues[i] = map->queues[i]; | |
2625 | new_map->alloc_len = alloc_len; | |
2626 | new_map->len = pos; | |
2627 | ||
2628 | return new_map; | |
2629 | } | |
2630 | ||
4d99f660 | 2631 | /* Must be called under cpus_read_lock */ |
80d19669 AN |
2632 | int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, |
2633 | u16 index, bool is_rxqs_map) | |
537c00de | 2634 | { |
80d19669 | 2635 | const unsigned long *online_mask = NULL, *possible_mask = NULL; |
01c5f864 | 2636 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
80d19669 | 2637 | int i, j, tci, numa_node_id = -2; |
184c449f | 2638 | int maps_sz, num_tc = 1, tc = 0; |
537c00de | 2639 | struct xps_map *map, *new_map; |
01c5f864 | 2640 | bool active = false; |
80d19669 | 2641 | unsigned int nr_ids; |
537c00de | 2642 | |
184c449f | 2643 | if (dev->num_tc) { |
ffcfe25b | 2644 | /* Do not allow XPS on subordinate device directly */ |
184c449f | 2645 | num_tc = dev->num_tc; |
ffcfe25b AD |
2646 | if (num_tc < 0) |
2647 | return -EINVAL; | |
2648 | ||
2649 | /* If queue belongs to subordinate dev use its map */ | |
2650 | dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; | |
2651 | ||
184c449f AD |
2652 | tc = netdev_txq_to_tc(dev, index); |
2653 | if (tc < 0) | |
2654 | return -EINVAL; | |
2655 | } | |
2656 | ||
537c00de | 2657 | mutex_lock(&xps_map_mutex); |
80d19669 AN |
2658 | if (is_rxqs_map) { |
2659 | maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); | |
2660 | dev_maps = xmap_dereference(dev->xps_rxqs_map); | |
2661 | nr_ids = dev->num_rx_queues; | |
2662 | } else { | |
2663 | maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); | |
2664 | if (num_possible_cpus() > 1) { | |
2665 | online_mask = cpumask_bits(cpu_online_mask); | |
2666 | possible_mask = cpumask_bits(cpu_possible_mask); | |
2667 | } | |
2668 | dev_maps = xmap_dereference(dev->xps_cpus_map); | |
2669 | nr_ids = nr_cpu_ids; | |
2670 | } | |
537c00de | 2671 | |
80d19669 AN |
2672 | if (maps_sz < L1_CACHE_BYTES) |
2673 | maps_sz = L1_CACHE_BYTES; | |
537c00de | 2674 | |
01c5f864 | 2675 | /* allocate memory for queue storage */ |
80d19669 AN |
2676 | for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), |
2677 | j < nr_ids;) { | |
01c5f864 AD |
2678 | if (!new_dev_maps) |
2679 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2680 | if (!new_dev_maps) { |
2681 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2682 | return -ENOMEM; |
2bb60cb9 | 2683 | } |
01c5f864 | 2684 | |
80d19669 AN |
2685 | tci = j * num_tc + tc; |
2686 | map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) : | |
01c5f864 AD |
2687 | NULL; |
2688 | ||
80d19669 | 2689 | map = expand_xps_map(map, j, index, is_rxqs_map); |
01c5f864 AD |
2690 | if (!map) |
2691 | goto error; | |
2692 | ||
80d19669 | 2693 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); |
01c5f864 AD |
2694 | } |
2695 | ||
2696 | if (!new_dev_maps) | |
2697 | goto out_no_new_maps; | |
2698 | ||
867d0ad4 SD |
2699 | if (!dev_maps) { |
2700 | /* Increment static keys at most once per type */ | |
2701 | static_key_slow_inc_cpuslocked(&xps_needed); | |
2702 | if (is_rxqs_map) | |
2703 | static_key_slow_inc_cpuslocked(&xps_rxqs_needed); | |
2704 | } | |
04157469 | 2705 | |
80d19669 AN |
2706 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2707 | j < nr_ids;) { | |
184c449f | 2708 | /* copy maps belonging to foreign traffic classes */ |
80d19669 | 2709 | for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) { |
184c449f | 2710 | /* fill in the new device map from the old device map */ |
80d19669 AN |
2711 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2712 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
184c449f AD |
2713 | } |
2714 | ||
2715 | /* We need to explicitly update tci as prevous loop | |
2716 | * could break out early if dev_maps is NULL. | |
2717 | */ | |
80d19669 | 2718 | tci = j * num_tc + tc; |
184c449f | 2719 | |
80d19669 AN |
2720 | if (netif_attr_test_mask(j, mask, nr_ids) && |
2721 | netif_attr_test_online(j, online_mask, nr_ids)) { | |
2722 | /* add tx-queue to CPU/rx-queue maps */ | |
01c5f864 AD |
2723 | int pos = 0; |
2724 | ||
80d19669 | 2725 | map = xmap_dereference(new_dev_maps->attr_map[tci]); |
01c5f864 AD |
2726 | while ((pos < map->len) && (map->queues[pos] != index)) |
2727 | pos++; | |
2728 | ||
2729 | if (pos == map->len) | |
2730 | map->queues[map->len++] = index; | |
537c00de | 2731 | #ifdef CONFIG_NUMA |
80d19669 AN |
2732 | if (!is_rxqs_map) { |
2733 | if (numa_node_id == -2) | |
2734 | numa_node_id = cpu_to_node(j); | |
2735 | else if (numa_node_id != cpu_to_node(j)) | |
2736 | numa_node_id = -1; | |
2737 | } | |
537c00de | 2738 | #endif |
01c5f864 AD |
2739 | } else if (dev_maps) { |
2740 | /* fill in the new device map from the old device map */ | |
80d19669 AN |
2741 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2742 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
537c00de | 2743 | } |
01c5f864 | 2744 | |
184c449f AD |
2745 | /* copy maps belonging to foreign traffic classes */ |
2746 | for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { | |
2747 | /* fill in the new device map from the old device map */ | |
80d19669 AN |
2748 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2749 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
184c449f | 2750 | } |
537c00de AD |
2751 | } |
2752 | ||
80d19669 AN |
2753 | if (is_rxqs_map) |
2754 | rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps); | |
2755 | else | |
2756 | rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps); | |
01c5f864 | 2757 | |
537c00de | 2758 | /* Cleanup old maps */ |
184c449f AD |
2759 | if (!dev_maps) |
2760 | goto out_no_old_maps; | |
2761 | ||
80d19669 AN |
2762 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2763 | j < nr_ids;) { | |
2764 | for (i = num_tc, tci = j * num_tc; i--; tci++) { | |
2765 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
2766 | map = xmap_dereference(dev_maps->attr_map[tci]); | |
01c5f864 AD |
2767 | if (map && map != new_map) |
2768 | kfree_rcu(map, rcu); | |
2769 | } | |
537c00de AD |
2770 | } |
2771 | ||
184c449f AD |
2772 | kfree_rcu(dev_maps, rcu); |
2773 | ||
2774 | out_no_old_maps: | |
01c5f864 AD |
2775 | dev_maps = new_dev_maps; |
2776 | active = true; | |
537c00de | 2777 | |
01c5f864 | 2778 | out_no_new_maps: |
80d19669 AN |
2779 | if (!is_rxqs_map) { |
2780 | /* update Tx queue numa node */ | |
2781 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), | |
2782 | (numa_node_id >= 0) ? | |
2783 | numa_node_id : NUMA_NO_NODE); | |
2784 | } | |
537c00de | 2785 | |
01c5f864 AD |
2786 | if (!dev_maps) |
2787 | goto out_no_maps; | |
2788 | ||
80d19669 AN |
2789 | /* removes tx-queue from unused CPUs/rx-queues */ |
2790 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), | |
2791 | j < nr_ids;) { | |
2792 | for (i = tc, tci = j * num_tc; i--; tci++) | |
184c449f | 2793 | active |= remove_xps_queue(dev_maps, tci, index); |
80d19669 AN |
2794 | if (!netif_attr_test_mask(j, mask, nr_ids) || |
2795 | !netif_attr_test_online(j, online_mask, nr_ids)) | |
184c449f AD |
2796 | active |= remove_xps_queue(dev_maps, tci, index); |
2797 | for (i = num_tc - tc, tci++; --i; tci++) | |
2798 | active |= remove_xps_queue(dev_maps, tci, index); | |
01c5f864 AD |
2799 | } |
2800 | ||
2801 | /* free map if not active */ | |
867d0ad4 SD |
2802 | if (!active) |
2803 | reset_xps_maps(dev, dev_maps, is_rxqs_map); | |
01c5f864 AD |
2804 | |
2805 | out_no_maps: | |
537c00de AD |
2806 | mutex_unlock(&xps_map_mutex); |
2807 | ||
2808 | return 0; | |
2809 | error: | |
01c5f864 | 2810 | /* remove any maps that we added */ |
80d19669 AN |
2811 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2812 | j < nr_ids;) { | |
2813 | for (i = num_tc, tci = j * num_tc; i--; tci++) { | |
2814 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
184c449f | 2815 | map = dev_maps ? |
80d19669 | 2816 | xmap_dereference(dev_maps->attr_map[tci]) : |
184c449f AD |
2817 | NULL; |
2818 | if (new_map && new_map != map) | |
2819 | kfree(new_map); | |
2820 | } | |
01c5f864 AD |
2821 | } |
2822 | ||
537c00de AD |
2823 | mutex_unlock(&xps_map_mutex); |
2824 | ||
537c00de AD |
2825 | kfree(new_dev_maps); |
2826 | return -ENOMEM; | |
2827 | } | |
4d99f660 | 2828 | EXPORT_SYMBOL_GPL(__netif_set_xps_queue); |
80d19669 AN |
2829 | |
2830 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, | |
2831 | u16 index) | |
2832 | { | |
4d99f660 AV |
2833 | int ret; |
2834 | ||
2835 | cpus_read_lock(); | |
2836 | ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false); | |
2837 | cpus_read_unlock(); | |
2838 | ||
2839 | return ret; | |
80d19669 | 2840 | } |
537c00de AD |
2841 | EXPORT_SYMBOL(netif_set_xps_queue); |
2842 | ||
2843 | #endif | |
ffcfe25b AD |
2844 | static void netdev_unbind_all_sb_channels(struct net_device *dev) |
2845 | { | |
2846 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2847 | ||
2848 | /* Unbind any subordinate channels */ | |
2849 | while (txq-- != &dev->_tx[0]) { | |
2850 | if (txq->sb_dev) | |
2851 | netdev_unbind_sb_channel(dev, txq->sb_dev); | |
2852 | } | |
2853 | } | |
2854 | ||
9cf1f6a8 AD |
2855 | void netdev_reset_tc(struct net_device *dev) |
2856 | { | |
6234f874 AD |
2857 | #ifdef CONFIG_XPS |
2858 | netif_reset_xps_queues_gt(dev, 0); | |
2859 | #endif | |
ffcfe25b AD |
2860 | netdev_unbind_all_sb_channels(dev); |
2861 | ||
2862 | /* Reset TC configuration of device */ | |
9cf1f6a8 AD |
2863 | dev->num_tc = 0; |
2864 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2865 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2866 | } | |
2867 | EXPORT_SYMBOL(netdev_reset_tc); | |
2868 | ||
2869 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2870 | { | |
2871 | if (tc >= dev->num_tc) | |
2872 | return -EINVAL; | |
2873 | ||
6234f874 AD |
2874 | #ifdef CONFIG_XPS |
2875 | netif_reset_xps_queues(dev, offset, count); | |
2876 | #endif | |
9cf1f6a8 AD |
2877 | dev->tc_to_txq[tc].count = count; |
2878 | dev->tc_to_txq[tc].offset = offset; | |
2879 | return 0; | |
2880 | } | |
2881 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2882 | ||
2883 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2884 | { | |
2885 | if (num_tc > TC_MAX_QUEUE) | |
2886 | return -EINVAL; | |
2887 | ||
6234f874 AD |
2888 | #ifdef CONFIG_XPS |
2889 | netif_reset_xps_queues_gt(dev, 0); | |
2890 | #endif | |
ffcfe25b AD |
2891 | netdev_unbind_all_sb_channels(dev); |
2892 | ||
9cf1f6a8 AD |
2893 | dev->num_tc = num_tc; |
2894 | return 0; | |
2895 | } | |
2896 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2897 | ||
ffcfe25b AD |
2898 | void netdev_unbind_sb_channel(struct net_device *dev, |
2899 | struct net_device *sb_dev) | |
2900 | { | |
2901 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2902 | ||
2903 | #ifdef CONFIG_XPS | |
2904 | netif_reset_xps_queues_gt(sb_dev, 0); | |
2905 | #endif | |
2906 | memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); | |
2907 | memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); | |
2908 | ||
2909 | while (txq-- != &dev->_tx[0]) { | |
2910 | if (txq->sb_dev == sb_dev) | |
2911 | txq->sb_dev = NULL; | |
2912 | } | |
2913 | } | |
2914 | EXPORT_SYMBOL(netdev_unbind_sb_channel); | |
2915 | ||
2916 | int netdev_bind_sb_channel_queue(struct net_device *dev, | |
2917 | struct net_device *sb_dev, | |
2918 | u8 tc, u16 count, u16 offset) | |
2919 | { | |
2920 | /* Make certain the sb_dev and dev are already configured */ | |
2921 | if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) | |
2922 | return -EINVAL; | |
2923 | ||
2924 | /* We cannot hand out queues we don't have */ | |
2925 | if ((offset + count) > dev->real_num_tx_queues) | |
2926 | return -EINVAL; | |
2927 | ||
2928 | /* Record the mapping */ | |
2929 | sb_dev->tc_to_txq[tc].count = count; | |
2930 | sb_dev->tc_to_txq[tc].offset = offset; | |
2931 | ||
2932 | /* Provide a way for Tx queue to find the tc_to_txq map or | |
2933 | * XPS map for itself. | |
2934 | */ | |
2935 | while (count--) | |
2936 | netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; | |
2937 | ||
2938 | return 0; | |
2939 | } | |
2940 | EXPORT_SYMBOL(netdev_bind_sb_channel_queue); | |
2941 | ||
2942 | int netdev_set_sb_channel(struct net_device *dev, u16 channel) | |
2943 | { | |
2944 | /* Do not use a multiqueue device to represent a subordinate channel */ | |
2945 | if (netif_is_multiqueue(dev)) | |
2946 | return -ENODEV; | |
2947 | ||
2948 | /* We allow channels 1 - 32767 to be used for subordinate channels. | |
2949 | * Channel 0 is meant to be "native" mode and used only to represent | |
2950 | * the main root device. We allow writing 0 to reset the device back | |
2951 | * to normal mode after being used as a subordinate channel. | |
2952 | */ | |
2953 | if (channel > S16_MAX) | |
2954 | return -EINVAL; | |
2955 | ||
2956 | dev->num_tc = -channel; | |
2957 | ||
2958 | return 0; | |
2959 | } | |
2960 | EXPORT_SYMBOL(netdev_set_sb_channel); | |
2961 | ||
f0796d5c JF |
2962 | /* |
2963 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
3a053b1a | 2964 | * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. |
f0796d5c | 2965 | */ |
e6484930 | 2966 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2967 | { |
ac5b7019 | 2968 | bool disabling; |
1d24eb48 TH |
2969 | int rc; |
2970 | ||
ac5b7019 JK |
2971 | disabling = txq < dev->real_num_tx_queues; |
2972 | ||
e6484930 TH |
2973 | if (txq < 1 || txq > dev->num_tx_queues) |
2974 | return -EINVAL; | |
f0796d5c | 2975 | |
5c56580b BH |
2976 | if (dev->reg_state == NETREG_REGISTERED || |
2977 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2978 | ASSERT_RTNL(); |
2979 | ||
1d24eb48 TH |
2980 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2981 | txq); | |
bf264145 TH |
2982 | if (rc) |
2983 | return rc; | |
2984 | ||
4f57c087 JF |
2985 | if (dev->num_tc) |
2986 | netif_setup_tc(dev, txq); | |
2987 | ||
ac5b7019 JK |
2988 | dev->real_num_tx_queues = txq; |
2989 | ||
2990 | if (disabling) { | |
2991 | synchronize_net(); | |
e6484930 | 2992 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2993 | #ifdef CONFIG_XPS |
2994 | netif_reset_xps_queues_gt(dev, txq); | |
2995 | #endif | |
2996 | } | |
ac5b7019 JK |
2997 | } else { |
2998 | dev->real_num_tx_queues = txq; | |
f0796d5c | 2999 | } |
e6484930 | 3000 | |
e6484930 | 3001 | return 0; |
f0796d5c JF |
3002 | } |
3003 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 3004 | |
a953be53 | 3005 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
3006 | /** |
3007 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
3008 | * @dev: Network device | |
3009 | * @rxq: Actual number of RX queues | |
3010 | * | |
3011 | * This must be called either with the rtnl_lock held or before | |
3012 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
3013 | * negative error code. If called before registration, it always |
3014 | * succeeds. | |
62fe0b40 BH |
3015 | */ |
3016 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
3017 | { | |
3018 | int rc; | |
3019 | ||
bd25fa7b TH |
3020 | if (rxq < 1 || rxq > dev->num_rx_queues) |
3021 | return -EINVAL; | |
3022 | ||
62fe0b40 BH |
3023 | if (dev->reg_state == NETREG_REGISTERED) { |
3024 | ASSERT_RTNL(); | |
3025 | ||
62fe0b40 BH |
3026 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
3027 | rxq); | |
3028 | if (rc) | |
3029 | return rc; | |
62fe0b40 BH |
3030 | } |
3031 | ||
3032 | dev->real_num_rx_queues = rxq; | |
3033 | return 0; | |
3034 | } | |
3035 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
3036 | #endif | |
3037 | ||
2c53040f BH |
3038 | /** |
3039 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
3040 | * |
3041 | * This routine should set an upper limit on the number of RSS queues | |
3042 | * used by default by multiqueue devices. | |
3043 | */ | |
a55b138b | 3044 | int netif_get_num_default_rss_queues(void) |
16917b87 | 3045 | { |
40e4e713 HS |
3046 | return is_kdump_kernel() ? |
3047 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
3048 | } |
3049 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
3050 | ||
3bcb846c | 3051 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 3052 | { |
def82a1d JP |
3053 | struct softnet_data *sd; |
3054 | unsigned long flags; | |
56079431 | 3055 | |
def82a1d | 3056 | local_irq_save(flags); |
903ceff7 | 3057 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
3058 | q->next_sched = NULL; |
3059 | *sd->output_queue_tailp = q; | |
3060 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
3061 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
3062 | local_irq_restore(flags); | |
3063 | } | |
3064 | ||
3065 | void __netif_schedule(struct Qdisc *q) | |
3066 | { | |
3067 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
3068 | __netif_reschedule(q); | |
56079431 DV |
3069 | } |
3070 | EXPORT_SYMBOL(__netif_schedule); | |
3071 | ||
e6247027 ED |
3072 | struct dev_kfree_skb_cb { |
3073 | enum skb_free_reason reason; | |
3074 | }; | |
3075 | ||
3076 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 3077 | { |
e6247027 ED |
3078 | return (struct dev_kfree_skb_cb *)skb->cb; |
3079 | } | |
3080 | ||
46e5da40 JF |
3081 | void netif_schedule_queue(struct netdev_queue *txq) |
3082 | { | |
3083 | rcu_read_lock(); | |
5be5515a | 3084 | if (!netif_xmit_stopped(txq)) { |
46e5da40 JF |
3085 | struct Qdisc *q = rcu_dereference(txq->qdisc); |
3086 | ||
3087 | __netif_schedule(q); | |
3088 | } | |
3089 | rcu_read_unlock(); | |
3090 | } | |
3091 | EXPORT_SYMBOL(netif_schedule_queue); | |
3092 | ||
46e5da40 JF |
3093 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
3094 | { | |
3095 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
3096 | struct Qdisc *q; | |
3097 | ||
3098 | rcu_read_lock(); | |
3099 | q = rcu_dereference(dev_queue->qdisc); | |
3100 | __netif_schedule(q); | |
3101 | rcu_read_unlock(); | |
3102 | } | |
3103 | } | |
3104 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
3105 | ||
e6247027 | 3106 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 3107 | { |
e6247027 | 3108 | unsigned long flags; |
56079431 | 3109 | |
9899886d MJ |
3110 | if (unlikely(!skb)) |
3111 | return; | |
3112 | ||
63354797 | 3113 | if (likely(refcount_read(&skb->users) == 1)) { |
e6247027 | 3114 | smp_rmb(); |
63354797 RE |
3115 | refcount_set(&skb->users, 0); |
3116 | } else if (likely(!refcount_dec_and_test(&skb->users))) { | |
e6247027 | 3117 | return; |
bea3348e | 3118 | } |
e6247027 ED |
3119 | get_kfree_skb_cb(skb)->reason = reason; |
3120 | local_irq_save(flags); | |
3121 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
3122 | __this_cpu_write(softnet_data.completion_queue, skb); | |
3123 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
3124 | local_irq_restore(flags); | |
56079431 | 3125 | } |
e6247027 | 3126 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 3127 | |
e6247027 | 3128 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
3129 | { |
3130 | if (in_irq() || irqs_disabled()) | |
e6247027 | 3131 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
3132 | else |
3133 | dev_kfree_skb(skb); | |
3134 | } | |
e6247027 | 3135 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
3136 | |
3137 | ||
bea3348e SH |
3138 | /** |
3139 | * netif_device_detach - mark device as removed | |
3140 | * @dev: network device | |
3141 | * | |
3142 | * Mark device as removed from system and therefore no longer available. | |
3143 | */ | |
56079431 DV |
3144 | void netif_device_detach(struct net_device *dev) |
3145 | { | |
3146 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
3147 | netif_running(dev)) { | |
d543103a | 3148 | netif_tx_stop_all_queues(dev); |
56079431 DV |
3149 | } |
3150 | } | |
3151 | EXPORT_SYMBOL(netif_device_detach); | |
3152 | ||
bea3348e SH |
3153 | /** |
3154 | * netif_device_attach - mark device as attached | |
3155 | * @dev: network device | |
3156 | * | |
3157 | * Mark device as attached from system and restart if needed. | |
3158 | */ | |
56079431 DV |
3159 | void netif_device_attach(struct net_device *dev) |
3160 | { | |
3161 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
3162 | netif_running(dev)) { | |
d543103a | 3163 | netif_tx_wake_all_queues(dev); |
4ec93edb | 3164 | __netdev_watchdog_up(dev); |
56079431 DV |
3165 | } |
3166 | } | |
3167 | EXPORT_SYMBOL(netif_device_attach); | |
3168 | ||
5605c762 JP |
3169 | /* |
3170 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
3171 | * to be used as a distribution range. | |
3172 | */ | |
eadec877 AD |
3173 | static u16 skb_tx_hash(const struct net_device *dev, |
3174 | const struct net_device *sb_dev, | |
3175 | struct sk_buff *skb) | |
5605c762 JP |
3176 | { |
3177 | u32 hash; | |
3178 | u16 qoffset = 0; | |
1b837d48 | 3179 | u16 qcount = dev->real_num_tx_queues; |
5605c762 | 3180 | |
eadec877 AD |
3181 | if (dev->num_tc) { |
3182 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
3183 | ||
3184 | qoffset = sb_dev->tc_to_txq[tc].offset; | |
3185 | qcount = sb_dev->tc_to_txq[tc].count; | |
3186 | } | |
3187 | ||
5605c762 JP |
3188 | if (skb_rx_queue_recorded(skb)) { |
3189 | hash = skb_get_rx_queue(skb); | |
6e11d157 AN |
3190 | if (hash >= qoffset) |
3191 | hash -= qoffset; | |
1b837d48 AD |
3192 | while (unlikely(hash >= qcount)) |
3193 | hash -= qcount; | |
eadec877 | 3194 | return hash + qoffset; |
5605c762 JP |
3195 | } |
3196 | ||
3197 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
3198 | } | |
5605c762 | 3199 | |
36c92474 BH |
3200 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
3201 | { | |
84d15ae5 | 3202 | static const netdev_features_t null_features; |
36c92474 | 3203 | struct net_device *dev = skb->dev; |
88ad4175 | 3204 | const char *name = ""; |
36c92474 | 3205 | |
c846ad9b BG |
3206 | if (!net_ratelimit()) |
3207 | return; | |
3208 | ||
88ad4175 BM |
3209 | if (dev) { |
3210 | if (dev->dev.parent) | |
3211 | name = dev_driver_string(dev->dev.parent); | |
3212 | else | |
3213 | name = netdev_name(dev); | |
3214 | } | |
6413139d WB |
3215 | skb_dump(KERN_WARNING, skb, false); |
3216 | WARN(1, "%s: caps=(%pNF, %pNF)\n", | |
88ad4175 | 3217 | name, dev ? &dev->features : &null_features, |
6413139d | 3218 | skb->sk ? &skb->sk->sk_route_caps : &null_features); |
36c92474 BH |
3219 | } |
3220 | ||
1da177e4 LT |
3221 | /* |
3222 | * Invalidate hardware checksum when packet is to be mangled, and | |
3223 | * complete checksum manually on outgoing path. | |
3224 | */ | |
84fa7933 | 3225 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 3226 | { |
d3bc23e7 | 3227 | __wsum csum; |
663ead3b | 3228 | int ret = 0, offset; |
1da177e4 | 3229 | |
84fa7933 | 3230 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
3231 | goto out_set_summed; |
3232 | ||
3aefd7d6 | 3233 | if (unlikely(skb_is_gso(skb))) { |
36c92474 BH |
3234 | skb_warn_bad_offload(skb); |
3235 | return -EINVAL; | |
1da177e4 LT |
3236 | } |
3237 | ||
cef401de ED |
3238 | /* Before computing a checksum, we should make sure no frag could |
3239 | * be modified by an external entity : checksum could be wrong. | |
3240 | */ | |
3241 | if (skb_has_shared_frag(skb)) { | |
3242 | ret = __skb_linearize(skb); | |
3243 | if (ret) | |
3244 | goto out; | |
3245 | } | |
3246 | ||
55508d60 | 3247 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
3248 | BUG_ON(offset >= skb_headlen(skb)); |
3249 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
3250 | ||
3251 | offset += skb->csum_offset; | |
3252 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
3253 | ||
8211fbfa HK |
3254 | ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); |
3255 | if (ret) | |
3256 | goto out; | |
1da177e4 | 3257 | |
4f2e4ad5 | 3258 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 3259 | out_set_summed: |
1da177e4 | 3260 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 3261 | out: |
1da177e4 LT |
3262 | return ret; |
3263 | } | |
d1b19dff | 3264 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 3265 | |
b72b5bf6 DC |
3266 | int skb_crc32c_csum_help(struct sk_buff *skb) |
3267 | { | |
3268 | __le32 crc32c_csum; | |
3269 | int ret = 0, offset, start; | |
3270 | ||
3271 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
3272 | goto out; | |
3273 | ||
3274 | if (unlikely(skb_is_gso(skb))) | |
3275 | goto out; | |
3276 | ||
3277 | /* Before computing a checksum, we should make sure no frag could | |
3278 | * be modified by an external entity : checksum could be wrong. | |
3279 | */ | |
3280 | if (unlikely(skb_has_shared_frag(skb))) { | |
3281 | ret = __skb_linearize(skb); | |
3282 | if (ret) | |
3283 | goto out; | |
3284 | } | |
3285 | start = skb_checksum_start_offset(skb); | |
3286 | offset = start + offsetof(struct sctphdr, checksum); | |
3287 | if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { | |
3288 | ret = -EINVAL; | |
3289 | goto out; | |
3290 | } | |
8211fbfa HK |
3291 | |
3292 | ret = skb_ensure_writable(skb, offset + sizeof(__le32)); | |
3293 | if (ret) | |
3294 | goto out; | |
3295 | ||
b72b5bf6 DC |
3296 | crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, |
3297 | skb->len - start, ~(__u32)0, | |
3298 | crc32c_csum_stub)); | |
3299 | *(__le32 *)(skb->data + offset) = crc32c_csum; | |
3300 | skb->ip_summed = CHECKSUM_NONE; | |
dba00306 | 3301 | skb->csum_not_inet = 0; |
b72b5bf6 DC |
3302 | out: |
3303 | return ret; | |
3304 | } | |
3305 | ||
53d6471c | 3306 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 3307 | { |
252e3346 | 3308 | __be16 type = skb->protocol; |
f6a78bfc | 3309 | |
19acc327 PS |
3310 | /* Tunnel gso handlers can set protocol to ethernet. */ |
3311 | if (type == htons(ETH_P_TEB)) { | |
3312 | struct ethhdr *eth; | |
3313 | ||
3314 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
3315 | return 0; | |
3316 | ||
1dfe82eb | 3317 | eth = (struct ethhdr *)skb->data; |
19acc327 PS |
3318 | type = eth->h_proto; |
3319 | } | |
3320 | ||
d4bcef3f | 3321 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
3322 | } |
3323 | ||
3324 | /** | |
3325 | * skb_mac_gso_segment - mac layer segmentation handler. | |
3326 | * @skb: buffer to segment | |
3327 | * @features: features for the output path (see dev->features) | |
3328 | */ | |
3329 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
3330 | netdev_features_t features) | |
3331 | { | |
3332 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
3333 | struct packet_offload *ptype; | |
53d6471c VY |
3334 | int vlan_depth = skb->mac_len; |
3335 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
3336 | |
3337 | if (unlikely(!type)) | |
3338 | return ERR_PTR(-EINVAL); | |
3339 | ||
53d6471c | 3340 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
3341 | |
3342 | rcu_read_lock(); | |
22061d80 | 3343 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 3344 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 3345 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
3346 | break; |
3347 | } | |
3348 | } | |
3349 | rcu_read_unlock(); | |
3350 | ||
98e399f8 | 3351 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 3352 | |
f6a78bfc HX |
3353 | return segs; |
3354 | } | |
05e8ef4a PS |
3355 | EXPORT_SYMBOL(skb_mac_gso_segment); |
3356 | ||
3357 | ||
3358 | /* openvswitch calls this on rx path, so we need a different check. | |
3359 | */ | |
3360 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
3361 | { | |
3362 | if (tx_path) | |
0c19f846 WB |
3363 | return skb->ip_summed != CHECKSUM_PARTIAL && |
3364 | skb->ip_summed != CHECKSUM_UNNECESSARY; | |
6e7bc478 ED |
3365 | |
3366 | return skb->ip_summed == CHECKSUM_NONE; | |
05e8ef4a PS |
3367 | } |
3368 | ||
3369 | /** | |
3370 | * __skb_gso_segment - Perform segmentation on skb. | |
3371 | * @skb: buffer to segment | |
3372 | * @features: features for the output path (see dev->features) | |
3373 | * @tx_path: whether it is called in TX path | |
3374 | * | |
3375 | * This function segments the given skb and returns a list of segments. | |
3376 | * | |
3377 | * It may return NULL if the skb requires no segmentation. This is | |
3378 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 | 3379 | * |
a08e7fd9 | 3380 | * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. |
05e8ef4a PS |
3381 | */ |
3382 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
3383 | netdev_features_t features, bool tx_path) | |
3384 | { | |
b2504a5d ED |
3385 | struct sk_buff *segs; |
3386 | ||
05e8ef4a PS |
3387 | if (unlikely(skb_needs_check(skb, tx_path))) { |
3388 | int err; | |
3389 | ||
b2504a5d | 3390 | /* We're going to init ->check field in TCP or UDP header */ |
a40e0a66 | 3391 | err = skb_cow_head(skb, 0); |
3392 | if (err < 0) | |
05e8ef4a PS |
3393 | return ERR_PTR(err); |
3394 | } | |
3395 | ||
802ab55a AD |
3396 | /* Only report GSO partial support if it will enable us to |
3397 | * support segmentation on this frame without needing additional | |
3398 | * work. | |
3399 | */ | |
3400 | if (features & NETIF_F_GSO_PARTIAL) { | |
3401 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
3402 | struct net_device *dev = skb->dev; | |
3403 | ||
3404 | partial_features |= dev->features & dev->gso_partial_features; | |
3405 | if (!skb_gso_ok(skb, features | partial_features)) | |
3406 | features &= ~NETIF_F_GSO_PARTIAL; | |
3407 | } | |
3408 | ||
a08e7fd9 | 3409 | BUILD_BUG_ON(SKB_GSO_CB_OFFSET + |
9207f9d4 KK |
3410 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); |
3411 | ||
68c33163 | 3412 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
3413 | SKB_GSO_CB(skb)->encap_level = 0; |
3414 | ||
05e8ef4a PS |
3415 | skb_reset_mac_header(skb); |
3416 | skb_reset_mac_len(skb); | |
3417 | ||
b2504a5d ED |
3418 | segs = skb_mac_gso_segment(skb, features); |
3419 | ||
3a1296a3 | 3420 | if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) |
b2504a5d ED |
3421 | skb_warn_bad_offload(skb); |
3422 | ||
3423 | return segs; | |
05e8ef4a | 3424 | } |
12b0004d | 3425 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 3426 | |
fb286bb2 HX |
3427 | /* Take action when hardware reception checksum errors are detected. */ |
3428 | #ifdef CONFIG_BUG | |
7fe50ac8 | 3429 | void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) |
fb286bb2 HX |
3430 | { |
3431 | if (net_ratelimit()) { | |
7b6cd1ce | 3432 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
6413139d | 3433 | skb_dump(KERN_ERR, skb, true); |
fb286bb2 HX |
3434 | dump_stack(); |
3435 | } | |
3436 | } | |
3437 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
3438 | #endif | |
3439 | ||
ab74cfeb | 3440 | /* XXX: check that highmem exists at all on the given machine. */ |
c1e756bf | 3441 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 3442 | { |
3d3a8533 | 3443 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 3444 | int i; |
f4563a75 | 3445 | |
5acbbd42 | 3446 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
3447 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
3448 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 3449 | |
ea2ab693 | 3450 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 3451 | return 1; |
ea2ab693 | 3452 | } |
5acbbd42 | 3453 | } |
3d3a8533 | 3454 | #endif |
1da177e4 LT |
3455 | return 0; |
3456 | } | |
1da177e4 | 3457 | |
3b392ddb SH |
3458 | /* If MPLS offload request, verify we are testing hardware MPLS features |
3459 | * instead of standard features for the netdev. | |
3460 | */ | |
d0edc7bf | 3461 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
3462 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
3463 | netdev_features_t features, | |
3464 | __be16 type) | |
3465 | { | |
25cd9ba0 | 3466 | if (eth_p_mpls(type)) |
3b392ddb SH |
3467 | features &= skb->dev->mpls_features; |
3468 | ||
3469 | return features; | |
3470 | } | |
3471 | #else | |
3472 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
3473 | netdev_features_t features, | |
3474 | __be16 type) | |
3475 | { | |
3476 | return features; | |
3477 | } | |
3478 | #endif | |
3479 | ||
c8f44aff | 3480 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 3481 | netdev_features_t features) |
f01a5236 | 3482 | { |
3b392ddb SH |
3483 | __be16 type; |
3484 | ||
9fc95f50 | 3485 | type = skb_network_protocol(skb, NULL); |
3b392ddb | 3486 | features = net_mpls_features(skb, features, type); |
53d6471c | 3487 | |
c0d680e5 | 3488 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 3489 | !can_checksum_protocol(features, type)) { |
996e8021 | 3490 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 3491 | } |
7be2c82c ED |
3492 | if (illegal_highdma(skb->dev, skb)) |
3493 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
3494 | |
3495 | return features; | |
3496 | } | |
3497 | ||
e38f3025 TM |
3498 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
3499 | struct net_device *dev, | |
3500 | netdev_features_t features) | |
3501 | { | |
3502 | return features; | |
3503 | } | |
3504 | EXPORT_SYMBOL(passthru_features_check); | |
3505 | ||
7ce23672 | 3506 | static netdev_features_t dflt_features_check(struct sk_buff *skb, |
8cb65d00 TM |
3507 | struct net_device *dev, |
3508 | netdev_features_t features) | |
3509 | { | |
3510 | return vlan_features_check(skb, features); | |
3511 | } | |
3512 | ||
cbc53e08 AD |
3513 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
3514 | struct net_device *dev, | |
3515 | netdev_features_t features) | |
3516 | { | |
3517 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
3518 | ||
3519 | if (gso_segs > dev->gso_max_segs) | |
3520 | return features & ~NETIF_F_GSO_MASK; | |
3521 | ||
1d155dfd HK |
3522 | if (!skb_shinfo(skb)->gso_type) { |
3523 | skb_warn_bad_offload(skb); | |
3524 | return features & ~NETIF_F_GSO_MASK; | |
3525 | } | |
3526 | ||
802ab55a AD |
3527 | /* Support for GSO partial features requires software |
3528 | * intervention before we can actually process the packets | |
3529 | * so we need to strip support for any partial features now | |
3530 | * and we can pull them back in after we have partially | |
3531 | * segmented the frame. | |
3532 | */ | |
3533 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
3534 | features &= ~dev->gso_partial_features; | |
3535 | ||
3536 | /* Make sure to clear the IPv4 ID mangling feature if the | |
3537 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
3538 | */ |
3539 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
3540 | struct iphdr *iph = skb->encapsulation ? | |
3541 | inner_ip_hdr(skb) : ip_hdr(skb); | |
3542 | ||
3543 | if (!(iph->frag_off & htons(IP_DF))) | |
3544 | features &= ~NETIF_F_TSO_MANGLEID; | |
3545 | } | |
3546 | ||
3547 | return features; | |
3548 | } | |
3549 | ||
c1e756bf | 3550 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 3551 | { |
5f35227e | 3552 | struct net_device *dev = skb->dev; |
fcbeb976 | 3553 | netdev_features_t features = dev->features; |
58e998c6 | 3554 | |
cbc53e08 AD |
3555 | if (skb_is_gso(skb)) |
3556 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 3557 | |
5f35227e JG |
3558 | /* If encapsulation offload request, verify we are testing |
3559 | * hardware encapsulation features instead of standard | |
3560 | * features for the netdev | |
3561 | */ | |
3562 | if (skb->encapsulation) | |
3563 | features &= dev->hw_enc_features; | |
3564 | ||
f5a7fb88 TM |
3565 | if (skb_vlan_tagged(skb)) |
3566 | features = netdev_intersect_features(features, | |
3567 | dev->vlan_features | | |
3568 | NETIF_F_HW_VLAN_CTAG_TX | | |
3569 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 3570 | |
5f35227e JG |
3571 | if (dev->netdev_ops->ndo_features_check) |
3572 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
3573 | features); | |
8cb65d00 TM |
3574 | else |
3575 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 3576 | |
c1e756bf | 3577 | return harmonize_features(skb, features); |
58e998c6 | 3578 | } |
c1e756bf | 3579 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 3580 | |
2ea25513 | 3581 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 3582 | struct netdev_queue *txq, bool more) |
f6a78bfc | 3583 | { |
2ea25513 DM |
3584 | unsigned int len; |
3585 | int rc; | |
00829823 | 3586 | |
9f9a742d | 3587 | if (dev_nit_active(dev)) |
2ea25513 | 3588 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 3589 | |
2ea25513 | 3590 | len = skb->len; |
3744741a | 3591 | PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); |
2ea25513 | 3592 | trace_net_dev_start_xmit(skb, dev); |
95f6b3dd | 3593 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 3594 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 3595 | |
2ea25513 DM |
3596 | return rc; |
3597 | } | |
7b9c6090 | 3598 | |
8dcda22a DM |
3599 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
3600 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
3601 | { |
3602 | struct sk_buff *skb = first; | |
3603 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 3604 | |
7f2e870f DM |
3605 | while (skb) { |
3606 | struct sk_buff *next = skb->next; | |
fc70fb64 | 3607 | |
a8305bff | 3608 | skb_mark_not_on_list(skb); |
95f6b3dd | 3609 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
3610 | if (unlikely(!dev_xmit_complete(rc))) { |
3611 | skb->next = next; | |
3612 | goto out; | |
3613 | } | |
6afff0ca | 3614 | |
7f2e870f | 3615 | skb = next; |
fe60faa5 | 3616 | if (netif_tx_queue_stopped(txq) && skb) { |
7f2e870f DM |
3617 | rc = NETDEV_TX_BUSY; |
3618 | break; | |
9ccb8975 | 3619 | } |
7f2e870f | 3620 | } |
9ccb8975 | 3621 | |
7f2e870f DM |
3622 | out: |
3623 | *ret = rc; | |
3624 | return skb; | |
3625 | } | |
b40863c6 | 3626 | |
1ff0dc94 ED |
3627 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
3628 | netdev_features_t features) | |
f6a78bfc | 3629 | { |
df8a39de | 3630 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
3631 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
3632 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
3633 | return skb; |
3634 | } | |
f6a78bfc | 3635 | |
43c26a1a DC |
3636 | int skb_csum_hwoffload_help(struct sk_buff *skb, |
3637 | const netdev_features_t features) | |
3638 | { | |
fa821170 | 3639 | if (unlikely(skb_csum_is_sctp(skb))) |
43c26a1a DC |
3640 | return !!(features & NETIF_F_SCTP_CRC) ? 0 : |
3641 | skb_crc32c_csum_help(skb); | |
3642 | ||
62fafcd6 XL |
3643 | if (features & NETIF_F_HW_CSUM) |
3644 | return 0; | |
3645 | ||
3646 | if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { | |
3647 | switch (skb->csum_offset) { | |
3648 | case offsetof(struct tcphdr, check): | |
3649 | case offsetof(struct udphdr, check): | |
3650 | return 0; | |
3651 | } | |
3652 | } | |
3653 | ||
3654 | return skb_checksum_help(skb); | |
43c26a1a DC |
3655 | } |
3656 | EXPORT_SYMBOL(skb_csum_hwoffload_help); | |
3657 | ||
f53c7239 | 3658 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) |
eae3f88e DM |
3659 | { |
3660 | netdev_features_t features; | |
f6a78bfc | 3661 | |
eae3f88e DM |
3662 | features = netif_skb_features(skb); |
3663 | skb = validate_xmit_vlan(skb, features); | |
3664 | if (unlikely(!skb)) | |
3665 | goto out_null; | |
7b9c6090 | 3666 | |
ebf4e808 IL |
3667 | skb = sk_validate_xmit_skb(skb, dev); |
3668 | if (unlikely(!skb)) | |
3669 | goto out_null; | |
3670 | ||
8b86a61d | 3671 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
3672 | struct sk_buff *segs; |
3673 | ||
3674 | segs = skb_gso_segment(skb, features); | |
cecda693 | 3675 | if (IS_ERR(segs)) { |
af6dabc9 | 3676 | goto out_kfree_skb; |
cecda693 JW |
3677 | } else if (segs) { |
3678 | consume_skb(skb); | |
3679 | skb = segs; | |
f6a78bfc | 3680 | } |
eae3f88e DM |
3681 | } else { |
3682 | if (skb_needs_linearize(skb, features) && | |
3683 | __skb_linearize(skb)) | |
3684 | goto out_kfree_skb; | |
4ec93edb | 3685 | |
eae3f88e DM |
3686 | /* If packet is not checksummed and device does not |
3687 | * support checksumming for this protocol, complete | |
3688 | * checksumming here. | |
3689 | */ | |
3690 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
3691 | if (skb->encapsulation) | |
3692 | skb_set_inner_transport_header(skb, | |
3693 | skb_checksum_start_offset(skb)); | |
3694 | else | |
3695 | skb_set_transport_header(skb, | |
3696 | skb_checksum_start_offset(skb)); | |
43c26a1a | 3697 | if (skb_csum_hwoffload_help(skb, features)) |
eae3f88e | 3698 | goto out_kfree_skb; |
7b9c6090 | 3699 | } |
0c772159 | 3700 | } |
7b9c6090 | 3701 | |
f53c7239 | 3702 | skb = validate_xmit_xfrm(skb, features, again); |
3dca3f38 | 3703 | |
eae3f88e | 3704 | return skb; |
fc70fb64 | 3705 | |
f6a78bfc HX |
3706 | out_kfree_skb: |
3707 | kfree_skb(skb); | |
eae3f88e | 3708 | out_null: |
d21fd63e | 3709 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3710 | return NULL; |
3711 | } | |
6afff0ca | 3712 | |
f53c7239 | 3713 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) |
55a93b3e ED |
3714 | { |
3715 | struct sk_buff *next, *head = NULL, *tail; | |
3716 | ||
bec3cfdc | 3717 | for (; skb != NULL; skb = next) { |
55a93b3e | 3718 | next = skb->next; |
a8305bff | 3719 | skb_mark_not_on_list(skb); |
bec3cfdc ED |
3720 | |
3721 | /* in case skb wont be segmented, point to itself */ | |
3722 | skb->prev = skb; | |
3723 | ||
f53c7239 | 3724 | skb = validate_xmit_skb(skb, dev, again); |
bec3cfdc ED |
3725 | if (!skb) |
3726 | continue; | |
55a93b3e | 3727 | |
bec3cfdc ED |
3728 | if (!head) |
3729 | head = skb; | |
3730 | else | |
3731 | tail->next = skb; | |
3732 | /* If skb was segmented, skb->prev points to | |
3733 | * the last segment. If not, it still contains skb. | |
3734 | */ | |
3735 | tail = skb->prev; | |
55a93b3e ED |
3736 | } |
3737 | return head; | |
f6a78bfc | 3738 | } |
104ba78c | 3739 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3740 | |
1def9238 ED |
3741 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3742 | { | |
3743 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3744 | ||
3745 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3746 | ||
3747 | /* To get more precise estimation of bytes sent on wire, | |
3748 | * we add to pkt_len the headers size of all segments | |
3749 | */ | |
a0dce875 | 3750 | if (shinfo->gso_size && skb_transport_header_was_set(skb)) { |
757b8b1d | 3751 | unsigned int hdr_len; |
15e5a030 | 3752 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3753 | |
757b8b1d ED |
3754 | /* mac layer + network layer */ |
3755 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3756 | ||
3757 | /* + transport layer */ | |
7c68d1a6 ED |
3758 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { |
3759 | const struct tcphdr *th; | |
3760 | struct tcphdr _tcphdr; | |
3761 | ||
3762 | th = skb_header_pointer(skb, skb_transport_offset(skb), | |
3763 | sizeof(_tcphdr), &_tcphdr); | |
3764 | if (likely(th)) | |
3765 | hdr_len += __tcp_hdrlen(th); | |
3766 | } else { | |
3767 | struct udphdr _udphdr; | |
3768 | ||
3769 | if (skb_header_pointer(skb, skb_transport_offset(skb), | |
3770 | sizeof(_udphdr), &_udphdr)) | |
3771 | hdr_len += sizeof(struct udphdr); | |
3772 | } | |
15e5a030 JW |
3773 | |
3774 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3775 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3776 | shinfo->gso_size); | |
3777 | ||
3778 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3779 | } |
3780 | } | |
3781 | ||
bbd8a0d3 KK |
3782 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3783 | struct net_device *dev, | |
3784 | struct netdev_queue *txq) | |
3785 | { | |
3786 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3787 | struct sk_buff *to_free = NULL; |
a2da570d | 3788 | bool contended; |
bbd8a0d3 KK |
3789 | int rc; |
3790 | ||
a2da570d | 3791 | qdisc_calculate_pkt_len(skb, q); |
6b3ba914 JF |
3792 | |
3793 | if (q->flags & TCQ_F_NOLOCK) { | |
ac5c66f2 | 3794 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
379349e9 | 3795 | qdisc_run(q); |
6b3ba914 JF |
3796 | |
3797 | if (unlikely(to_free)) | |
3798 | kfree_skb_list(to_free); | |
3799 | return rc; | |
3800 | } | |
3801 | ||
79640a4c ED |
3802 | /* |
3803 | * Heuristic to force contended enqueues to serialize on a | |
3804 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3805 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3806 | * often and dequeue packets faster. |
79640a4c | 3807 | */ |
a2da570d | 3808 | contended = qdisc_is_running(q); |
79640a4c ED |
3809 | if (unlikely(contended)) |
3810 | spin_lock(&q->busylock); | |
3811 | ||
bbd8a0d3 KK |
3812 | spin_lock(root_lock); |
3813 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3814 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3815 | rc = NET_XMIT_DROP; |
3816 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3817 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3818 | /* |
3819 | * This is a work-conserving queue; there are no old skbs | |
3820 | * waiting to be sent out; and the qdisc is not running - | |
3821 | * xmit the skb directly. | |
3822 | */ | |
bfe0d029 | 3823 | |
bfe0d029 ED |
3824 | qdisc_bstats_update(q, skb); |
3825 | ||
55a93b3e | 3826 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3827 | if (unlikely(contended)) { |
3828 | spin_unlock(&q->busylock); | |
3829 | contended = false; | |
3830 | } | |
bbd8a0d3 | 3831 | __qdisc_run(q); |
6c148184 | 3832 | } |
bbd8a0d3 | 3833 | |
6c148184 | 3834 | qdisc_run_end(q); |
bbd8a0d3 KK |
3835 | rc = NET_XMIT_SUCCESS; |
3836 | } else { | |
ac5c66f2 | 3837 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3838 | if (qdisc_run_begin(q)) { |
3839 | if (unlikely(contended)) { | |
3840 | spin_unlock(&q->busylock); | |
3841 | contended = false; | |
3842 | } | |
3843 | __qdisc_run(q); | |
6c148184 | 3844 | qdisc_run_end(q); |
79640a4c | 3845 | } |
bbd8a0d3 KK |
3846 | } |
3847 | spin_unlock(root_lock); | |
520ac30f ED |
3848 | if (unlikely(to_free)) |
3849 | kfree_skb_list(to_free); | |
79640a4c ED |
3850 | if (unlikely(contended)) |
3851 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3852 | return rc; |
3853 | } | |
3854 | ||
86f8515f | 3855 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3856 | static void skb_update_prio(struct sk_buff *skb) |
3857 | { | |
4dcb31d4 ED |
3858 | const struct netprio_map *map; |
3859 | const struct sock *sk; | |
3860 | unsigned int prioidx; | |
5bc1421e | 3861 | |
4dcb31d4 ED |
3862 | if (skb->priority) |
3863 | return; | |
3864 | map = rcu_dereference_bh(skb->dev->priomap); | |
3865 | if (!map) | |
3866 | return; | |
3867 | sk = skb_to_full_sk(skb); | |
3868 | if (!sk) | |
3869 | return; | |
91c68ce2 | 3870 | |
4dcb31d4 ED |
3871 | prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); |
3872 | ||
3873 | if (prioidx < map->priomap_len) | |
3874 | skb->priority = map->priomap[prioidx]; | |
5bc1421e NH |
3875 | } |
3876 | #else | |
3877 | #define skb_update_prio(skb) | |
3878 | #endif | |
3879 | ||
95603e22 MM |
3880 | /** |
3881 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3882 | * @net: network namespace this loopback is happening in |
3883 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3884 | * @skb: buffer to transmit |
3885 | */ | |
0c4b51f0 | 3886 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3887 | { |
3888 | skb_reset_mac_header(skb); | |
3889 | __skb_pull(skb, skb_network_offset(skb)); | |
3890 | skb->pkt_type = PACKET_LOOPBACK; | |
3891 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3892 | WARN_ON(!skb_dst(skb)); | |
3893 | skb_dst_force(skb); | |
3894 | netif_rx_ni(skb); | |
3895 | return 0; | |
3896 | } | |
3897 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3898 | ||
1f211a1b DB |
3899 | #ifdef CONFIG_NET_EGRESS |
3900 | static struct sk_buff * | |
3901 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3902 | { | |
46209401 | 3903 | struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); |
1f211a1b DB |
3904 | struct tcf_result cl_res; |
3905 | ||
46209401 | 3906 | if (!miniq) |
1f211a1b DB |
3907 | return skb; |
3908 | ||
8dc07fdb | 3909 | /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ |
aadaca9e | 3910 | qdisc_skb_cb(skb)->mru = 0; |
7baf2429 | 3911 | qdisc_skb_cb(skb)->post_ct = false; |
46209401 | 3912 | mini_qdisc_bstats_cpu_update(miniq, skb); |
1f211a1b | 3913 | |
46209401 | 3914 | switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { |
1f211a1b DB |
3915 | case TC_ACT_OK: |
3916 | case TC_ACT_RECLASSIFY: | |
3917 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3918 | break; | |
3919 | case TC_ACT_SHOT: | |
46209401 | 3920 | mini_qdisc_qstats_cpu_drop(miniq); |
1f211a1b | 3921 | *ret = NET_XMIT_DROP; |
7e2c3aea DB |
3922 | kfree_skb(skb); |
3923 | return NULL; | |
1f211a1b DB |
3924 | case TC_ACT_STOLEN: |
3925 | case TC_ACT_QUEUED: | |
e25ea21f | 3926 | case TC_ACT_TRAP: |
1f211a1b | 3927 | *ret = NET_XMIT_SUCCESS; |
7e2c3aea | 3928 | consume_skb(skb); |
1f211a1b DB |
3929 | return NULL; |
3930 | case TC_ACT_REDIRECT: | |
3931 | /* No need to push/pop skb's mac_header here on egress! */ | |
3932 | skb_do_redirect(skb); | |
3933 | *ret = NET_XMIT_SUCCESS; | |
3934 | return NULL; | |
3935 | default: | |
3936 | break; | |
3937 | } | |
357b6cc5 | 3938 | |
1f211a1b DB |
3939 | return skb; |
3940 | } | |
3941 | #endif /* CONFIG_NET_EGRESS */ | |
3942 | ||
fc9bab24 AN |
3943 | #ifdef CONFIG_XPS |
3944 | static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, | |
3945 | struct xps_dev_maps *dev_maps, unsigned int tci) | |
3946 | { | |
3947 | struct xps_map *map; | |
3948 | int queue_index = -1; | |
3949 | ||
3950 | if (dev->num_tc) { | |
3951 | tci *= dev->num_tc; | |
3952 | tci += netdev_get_prio_tc_map(dev, skb->priority); | |
3953 | } | |
3954 | ||
3955 | map = rcu_dereference(dev_maps->attr_map[tci]); | |
3956 | if (map) { | |
3957 | if (map->len == 1) | |
3958 | queue_index = map->queues[0]; | |
3959 | else | |
3960 | queue_index = map->queues[reciprocal_scale( | |
3961 | skb_get_hash(skb), map->len)]; | |
3962 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3963 | queue_index = -1; | |
3964 | } | |
3965 | return queue_index; | |
3966 | } | |
3967 | #endif | |
3968 | ||
eadec877 AD |
3969 | static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, |
3970 | struct sk_buff *skb) | |
638b2a69 JP |
3971 | { |
3972 | #ifdef CONFIG_XPS | |
3973 | struct xps_dev_maps *dev_maps; | |
fc9bab24 | 3974 | struct sock *sk = skb->sk; |
638b2a69 JP |
3975 | int queue_index = -1; |
3976 | ||
04157469 AN |
3977 | if (!static_key_false(&xps_needed)) |
3978 | return -1; | |
3979 | ||
638b2a69 | 3980 | rcu_read_lock(); |
fc9bab24 AN |
3981 | if (!static_key_false(&xps_rxqs_needed)) |
3982 | goto get_cpus_map; | |
3983 | ||
eadec877 | 3984 | dev_maps = rcu_dereference(sb_dev->xps_rxqs_map); |
638b2a69 | 3985 | if (dev_maps) { |
fc9bab24 | 3986 | int tci = sk_rx_queue_get(sk); |
184c449f | 3987 | |
fc9bab24 AN |
3988 | if (tci >= 0 && tci < dev->num_rx_queues) |
3989 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
3990 | tci); | |
3991 | } | |
184c449f | 3992 | |
fc9bab24 AN |
3993 | get_cpus_map: |
3994 | if (queue_index < 0) { | |
eadec877 | 3995 | dev_maps = rcu_dereference(sb_dev->xps_cpus_map); |
fc9bab24 AN |
3996 | if (dev_maps) { |
3997 | unsigned int tci = skb->sender_cpu - 1; | |
3998 | ||
3999 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
4000 | tci); | |
638b2a69 JP |
4001 | } |
4002 | } | |
4003 | rcu_read_unlock(); | |
4004 | ||
4005 | return queue_index; | |
4006 | #else | |
4007 | return -1; | |
4008 | #endif | |
4009 | } | |
4010 | ||
a4ea8a3d | 4011 | u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, |
a350ecce | 4012 | struct net_device *sb_dev) |
a4ea8a3d AD |
4013 | { |
4014 | return 0; | |
4015 | } | |
4016 | EXPORT_SYMBOL(dev_pick_tx_zero); | |
4017 | ||
4018 | u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, | |
a350ecce | 4019 | struct net_device *sb_dev) |
a4ea8a3d AD |
4020 | { |
4021 | return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; | |
4022 | } | |
4023 | EXPORT_SYMBOL(dev_pick_tx_cpu_id); | |
4024 | ||
b71b5837 PA |
4025 | u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, |
4026 | struct net_device *sb_dev) | |
638b2a69 JP |
4027 | { |
4028 | struct sock *sk = skb->sk; | |
4029 | int queue_index = sk_tx_queue_get(sk); | |
4030 | ||
eadec877 AD |
4031 | sb_dev = sb_dev ? : dev; |
4032 | ||
638b2a69 JP |
4033 | if (queue_index < 0 || skb->ooo_okay || |
4034 | queue_index >= dev->real_num_tx_queues) { | |
eadec877 | 4035 | int new_index = get_xps_queue(dev, sb_dev, skb); |
f4563a75 | 4036 | |
638b2a69 | 4037 | if (new_index < 0) |
eadec877 | 4038 | new_index = skb_tx_hash(dev, sb_dev, skb); |
638b2a69 JP |
4039 | |
4040 | if (queue_index != new_index && sk && | |
004a5d01 | 4041 | sk_fullsock(sk) && |
638b2a69 JP |
4042 | rcu_access_pointer(sk->sk_dst_cache)) |
4043 | sk_tx_queue_set(sk, new_index); | |
4044 | ||
4045 | queue_index = new_index; | |
4046 | } | |
4047 | ||
4048 | return queue_index; | |
4049 | } | |
b71b5837 | 4050 | EXPORT_SYMBOL(netdev_pick_tx); |
638b2a69 | 4051 | |
4bd97d51 PA |
4052 | struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, |
4053 | struct sk_buff *skb, | |
4054 | struct net_device *sb_dev) | |
638b2a69 JP |
4055 | { |
4056 | int queue_index = 0; | |
4057 | ||
4058 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
4059 | u32 sender_cpu = skb->sender_cpu - 1; |
4060 | ||
4061 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
4062 | skb->sender_cpu = raw_smp_processor_id() + 1; |
4063 | #endif | |
4064 | ||
4065 | if (dev->real_num_tx_queues != 1) { | |
4066 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 4067 | |
638b2a69 | 4068 | if (ops->ndo_select_queue) |
a350ecce | 4069 | queue_index = ops->ndo_select_queue(dev, skb, sb_dev); |
638b2a69 | 4070 | else |
4bd97d51 | 4071 | queue_index = netdev_pick_tx(dev, skb, sb_dev); |
638b2a69 | 4072 | |
d584527c | 4073 | queue_index = netdev_cap_txqueue(dev, queue_index); |
638b2a69 JP |
4074 | } |
4075 | ||
4076 | skb_set_queue_mapping(skb, queue_index); | |
4077 | return netdev_get_tx_queue(dev, queue_index); | |
4078 | } | |
4079 | ||
d29f749e | 4080 | /** |
9d08dd3d | 4081 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 4082 | * @skb: buffer to transmit |
eadec877 | 4083 | * @sb_dev: suboordinate device used for L2 forwarding offload |
d29f749e DJ |
4084 | * |
4085 | * Queue a buffer for transmission to a network device. The caller must | |
4086 | * have set the device and priority and built the buffer before calling | |
4087 | * this function. The function can be called from an interrupt. | |
4088 | * | |
4089 | * A negative errno code is returned on a failure. A success does not | |
4090 | * guarantee the frame will be transmitted as it may be dropped due | |
4091 | * to congestion or traffic shaping. | |
4092 | * | |
4093 | * ----------------------------------------------------------------------------------- | |
4094 | * I notice this method can also return errors from the queue disciplines, | |
4095 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
4096 | * be positive. | |
4097 | * | |
4098 | * Regardless of the return value, the skb is consumed, so it is currently | |
4099 | * difficult to retry a send to this method. (You can bump the ref count | |
4100 | * before sending to hold a reference for retry if you are careful.) | |
4101 | * | |
4102 | * When calling this method, interrupts MUST be enabled. This is because | |
4103 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
4104 | * --BLG | |
4105 | */ | |
eadec877 | 4106 | static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) |
1da177e4 LT |
4107 | { |
4108 | struct net_device *dev = skb->dev; | |
dc2b4847 | 4109 | struct netdev_queue *txq; |
1da177e4 LT |
4110 | struct Qdisc *q; |
4111 | int rc = -ENOMEM; | |
f53c7239 | 4112 | bool again = false; |
1da177e4 | 4113 | |
6d1ccff6 ED |
4114 | skb_reset_mac_header(skb); |
4115 | ||
e7fd2885 | 4116 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
e7ed11ee | 4117 | __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); |
e7fd2885 | 4118 | |
4ec93edb YH |
4119 | /* Disable soft irqs for various locks below. Also |
4120 | * stops preemption for RCU. | |
1da177e4 | 4121 | */ |
4ec93edb | 4122 | rcu_read_lock_bh(); |
1da177e4 | 4123 | |
5bc1421e NH |
4124 | skb_update_prio(skb); |
4125 | ||
1f211a1b DB |
4126 | qdisc_pkt_len_init(skb); |
4127 | #ifdef CONFIG_NET_CLS_ACT | |
8dc07fdb | 4128 | skb->tc_at_ingress = 0; |
357b6cc5 | 4129 | # ifdef CONFIG_NET_EGRESS |
aabf6772 | 4130 | if (static_branch_unlikely(&egress_needed_key)) { |
1f211a1b DB |
4131 | skb = sch_handle_egress(skb, &rc, dev); |
4132 | if (!skb) | |
4133 | goto out; | |
4134 | } | |
357b6cc5 | 4135 | # endif |
1f211a1b | 4136 | #endif |
02875878 ED |
4137 | /* If device/qdisc don't need skb->dst, release it right now while |
4138 | * its hot in this cpu cache. | |
4139 | */ | |
4140 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
4141 | skb_dst_drop(skb); | |
4142 | else | |
4143 | skb_dst_force(skb); | |
4144 | ||
4bd97d51 | 4145 | txq = netdev_core_pick_tx(dev, skb, sb_dev); |
a898def2 | 4146 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 4147 | |
cf66ba58 | 4148 | trace_net_dev_queue(skb); |
1da177e4 | 4149 | if (q->enqueue) { |
bbd8a0d3 | 4150 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 4151 | goto out; |
1da177e4 LT |
4152 | } |
4153 | ||
4154 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 4155 | * loopback, all the sorts of tunnels... |
1da177e4 | 4156 | |
eb13da1a | 4157 | * Really, it is unlikely that netif_tx_lock protection is necessary |
4158 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
4159 | * counters.) | |
4160 | * However, it is possible, that they rely on protection | |
4161 | * made by us here. | |
1da177e4 | 4162 | |
eb13da1a | 4163 | * Check this and shot the lock. It is not prone from deadlocks. |
4164 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
4165 | */ |
4166 | if (dev->flags & IFF_UP) { | |
4167 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
4168 | ||
c773e847 | 4169 | if (txq->xmit_lock_owner != cpu) { |
97cdcf37 | 4170 | if (dev_xmit_recursion()) |
745e20f1 ED |
4171 | goto recursion_alert; |
4172 | ||
f53c7239 | 4173 | skb = validate_xmit_skb(skb, dev, &again); |
1f59533f | 4174 | if (!skb) |
d21fd63e | 4175 | goto out; |
1f59533f | 4176 | |
3744741a | 4177 | PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); |
c773e847 | 4178 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 4179 | |
73466498 | 4180 | if (!netif_xmit_stopped(txq)) { |
97cdcf37 | 4181 | dev_xmit_recursion_inc(); |
ce93718f | 4182 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
97cdcf37 | 4183 | dev_xmit_recursion_dec(); |
572a9d7b | 4184 | if (dev_xmit_complete(rc)) { |
c773e847 | 4185 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
4186 | goto out; |
4187 | } | |
4188 | } | |
c773e847 | 4189 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
4190 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
4191 | dev->name); | |
1da177e4 LT |
4192 | } else { |
4193 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
4194 | * unfortunately |
4195 | */ | |
4196 | recursion_alert: | |
e87cc472 JP |
4197 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
4198 | dev->name); | |
1da177e4 LT |
4199 | } |
4200 | } | |
4201 | ||
4202 | rc = -ENETDOWN; | |
d4828d85 | 4203 | rcu_read_unlock_bh(); |
1da177e4 | 4204 | |
015f0688 | 4205 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 4206 | kfree_skb_list(skb); |
1da177e4 LT |
4207 | return rc; |
4208 | out: | |
d4828d85 | 4209 | rcu_read_unlock_bh(); |
1da177e4 LT |
4210 | return rc; |
4211 | } | |
f663dd9a | 4212 | |
2b4aa3ce | 4213 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
4214 | { |
4215 | return __dev_queue_xmit(skb, NULL); | |
4216 | } | |
2b4aa3ce | 4217 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 4218 | |
eadec877 | 4219 | int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) |
f663dd9a | 4220 | { |
eadec877 | 4221 | return __dev_queue_xmit(skb, sb_dev); |
f663dd9a JW |
4222 | } |
4223 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
4224 | ||
36ccdf85 | 4225 | int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) |
865b03f2 MK |
4226 | { |
4227 | struct net_device *dev = skb->dev; | |
4228 | struct sk_buff *orig_skb = skb; | |
4229 | struct netdev_queue *txq; | |
4230 | int ret = NETDEV_TX_BUSY; | |
4231 | bool again = false; | |
4232 | ||
4233 | if (unlikely(!netif_running(dev) || | |
4234 | !netif_carrier_ok(dev))) | |
4235 | goto drop; | |
4236 | ||
4237 | skb = validate_xmit_skb_list(skb, dev, &again); | |
4238 | if (skb != orig_skb) | |
4239 | goto drop; | |
4240 | ||
4241 | skb_set_queue_mapping(skb, queue_id); | |
4242 | txq = skb_get_tx_queue(dev, skb); | |
3744741a | 4243 | PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); |
865b03f2 MK |
4244 | |
4245 | local_bh_disable(); | |
4246 | ||
0ad6f6e7 | 4247 | dev_xmit_recursion_inc(); |
865b03f2 MK |
4248 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
4249 | if (!netif_xmit_frozen_or_drv_stopped(txq)) | |
4250 | ret = netdev_start_xmit(skb, dev, txq, false); | |
4251 | HARD_TX_UNLOCK(dev, txq); | |
0ad6f6e7 | 4252 | dev_xmit_recursion_dec(); |
865b03f2 MK |
4253 | |
4254 | local_bh_enable(); | |
865b03f2 MK |
4255 | return ret; |
4256 | drop: | |
4257 | atomic_long_inc(&dev->tx_dropped); | |
4258 | kfree_skb_list(skb); | |
4259 | return NET_XMIT_DROP; | |
4260 | } | |
36ccdf85 | 4261 | EXPORT_SYMBOL(__dev_direct_xmit); |
1da177e4 | 4262 | |
eb13da1a | 4263 | /************************************************************************* |
4264 | * Receiver routines | |
4265 | *************************************************************************/ | |
1da177e4 | 4266 | |
6b2bedc3 | 4267 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
4268 | EXPORT_SYMBOL(netdev_max_backlog); |
4269 | ||
3b098e2d | 4270 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 | 4271 | int netdev_budget __read_mostly = 300; |
a4837980 KK |
4272 | /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ |
4273 | unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; | |
3d48b53f MT |
4274 | int weight_p __read_mostly = 64; /* old backlog weight */ |
4275 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
4276 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
4277 | int dev_rx_weight __read_mostly = 64; | |
4278 | int dev_tx_weight __read_mostly = 64; | |
323ebb61 EC |
4279 | /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ |
4280 | int gro_normal_batch __read_mostly = 8; | |
1da177e4 | 4281 | |
eecfd7c4 ED |
4282 | /* Called with irq disabled */ |
4283 | static inline void ____napi_schedule(struct softnet_data *sd, | |
4284 | struct napi_struct *napi) | |
4285 | { | |
29863d41 WW |
4286 | struct task_struct *thread; |
4287 | ||
4288 | if (test_bit(NAPI_STATE_THREADED, &napi->state)) { | |
4289 | /* Paired with smp_mb__before_atomic() in | |
5fdd2f0e WW |
4290 | * napi_enable()/dev_set_threaded(). |
4291 | * Use READ_ONCE() to guarantee a complete | |
4292 | * read on napi->thread. Only call | |
29863d41 WW |
4293 | * wake_up_process() when it's not NULL. |
4294 | */ | |
4295 | thread = READ_ONCE(napi->thread); | |
4296 | if (thread) { | |
4297 | wake_up_process(thread); | |
4298 | return; | |
4299 | } | |
4300 | } | |
4301 | ||
eecfd7c4 ED |
4302 | list_add_tail(&napi->poll_list, &sd->poll_list); |
4303 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4304 | } | |
4305 | ||
bfb564e7 KK |
4306 | #ifdef CONFIG_RPS |
4307 | ||
4308 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 4309 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 4310 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
4311 | u32 rps_cpu_mask __read_mostly; |
4312 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 4313 | |
dc05360f | 4314 | struct static_key_false rps_needed __read_mostly; |
3df97ba8 | 4315 | EXPORT_SYMBOL(rps_needed); |
dc05360f | 4316 | struct static_key_false rfs_needed __read_mostly; |
13bfff25 | 4317 | EXPORT_SYMBOL(rfs_needed); |
adc9300e | 4318 | |
c445477d BH |
4319 | static struct rps_dev_flow * |
4320 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4321 | struct rps_dev_flow *rflow, u16 next_cpu) | |
4322 | { | |
a31196b0 | 4323 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
4324 | #ifdef CONFIG_RFS_ACCEL |
4325 | struct netdev_rx_queue *rxqueue; | |
4326 | struct rps_dev_flow_table *flow_table; | |
4327 | struct rps_dev_flow *old_rflow; | |
4328 | u32 flow_id; | |
4329 | u16 rxq_index; | |
4330 | int rc; | |
4331 | ||
4332 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
4333 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
4334 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
4335 | goto out; |
4336 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
4337 | if (rxq_index == skb_get_rx_queue(skb)) | |
4338 | goto out; | |
4339 | ||
4340 | rxqueue = dev->_rx + rxq_index; | |
4341 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4342 | if (!flow_table) | |
4343 | goto out; | |
61b905da | 4344 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
4345 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
4346 | rxq_index, flow_id); | |
4347 | if (rc < 0) | |
4348 | goto out; | |
4349 | old_rflow = rflow; | |
4350 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
4351 | rflow->filter = rc; |
4352 | if (old_rflow->filter == rflow->filter) | |
4353 | old_rflow->filter = RPS_NO_FILTER; | |
4354 | out: | |
4355 | #endif | |
4356 | rflow->last_qtail = | |
09994d1b | 4357 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
4358 | } |
4359 | ||
09994d1b | 4360 | rflow->cpu = next_cpu; |
c445477d BH |
4361 | return rflow; |
4362 | } | |
4363 | ||
bfb564e7 KK |
4364 | /* |
4365 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
4366 | * CPU from the RPS map of the receiving queue for a given skb. | |
4367 | * rcu_read_lock must be held on entry. | |
4368 | */ | |
4369 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4370 | struct rps_dev_flow **rflowp) | |
4371 | { | |
567e4b79 ED |
4372 | const struct rps_sock_flow_table *sock_flow_table; |
4373 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 4374 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 4375 | struct rps_map *map; |
bfb564e7 | 4376 | int cpu = -1; |
567e4b79 | 4377 | u32 tcpu; |
61b905da | 4378 | u32 hash; |
bfb564e7 KK |
4379 | |
4380 | if (skb_rx_queue_recorded(skb)) { | |
4381 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 4382 | |
62fe0b40 BH |
4383 | if (unlikely(index >= dev->real_num_rx_queues)) { |
4384 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4385 | "%s received packet on queue %u, but number " | |
4386 | "of RX queues is %u\n", | |
4387 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
4388 | goto done; |
4389 | } | |
567e4b79 ED |
4390 | rxqueue += index; |
4391 | } | |
bfb564e7 | 4392 | |
567e4b79 ED |
4393 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
4394 | ||
4395 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 4396 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 4397 | if (!flow_table && !map) |
bfb564e7 KK |
4398 | goto done; |
4399 | ||
2d47b459 | 4400 | skb_reset_network_header(skb); |
61b905da TH |
4401 | hash = skb_get_hash(skb); |
4402 | if (!hash) | |
bfb564e7 KK |
4403 | goto done; |
4404 | ||
fec5e652 TH |
4405 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
4406 | if (flow_table && sock_flow_table) { | |
fec5e652 | 4407 | struct rps_dev_flow *rflow; |
567e4b79 ED |
4408 | u32 next_cpu; |
4409 | u32 ident; | |
4410 | ||
4411 | /* First check into global flow table if there is a match */ | |
4412 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
4413 | if ((ident ^ hash) & ~rps_cpu_mask) | |
4414 | goto try_rps; | |
fec5e652 | 4415 | |
567e4b79 ED |
4416 | next_cpu = ident & rps_cpu_mask; |
4417 | ||
4418 | /* OK, now we know there is a match, | |
4419 | * we can look at the local (per receive queue) flow table | |
4420 | */ | |
61b905da | 4421 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
4422 | tcpu = rflow->cpu; |
4423 | ||
fec5e652 TH |
4424 | /* |
4425 | * If the desired CPU (where last recvmsg was done) is | |
4426 | * different from current CPU (one in the rx-queue flow | |
4427 | * table entry), switch if one of the following holds: | |
a31196b0 | 4428 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
4429 | * - Current CPU is offline. |
4430 | * - The current CPU's queue tail has advanced beyond the | |
4431 | * last packet that was enqueued using this table entry. | |
4432 | * This guarantees that all previous packets for the flow | |
4433 | * have been dequeued, thus preserving in order delivery. | |
4434 | */ | |
4435 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 4436 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 4437 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
4438 | rflow->last_qtail)) >= 0)) { |
4439 | tcpu = next_cpu; | |
c445477d | 4440 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 4441 | } |
c445477d | 4442 | |
a31196b0 | 4443 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
4444 | *rflowp = rflow; |
4445 | cpu = tcpu; | |
4446 | goto done; | |
4447 | } | |
4448 | } | |
4449 | ||
567e4b79 ED |
4450 | try_rps: |
4451 | ||
0a9627f2 | 4452 | if (map) { |
8fc54f68 | 4453 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
4454 | if (cpu_online(tcpu)) { |
4455 | cpu = tcpu; | |
4456 | goto done; | |
4457 | } | |
4458 | } | |
4459 | ||
4460 | done: | |
0a9627f2 TH |
4461 | return cpu; |
4462 | } | |
4463 | ||
c445477d BH |
4464 | #ifdef CONFIG_RFS_ACCEL |
4465 | ||
4466 | /** | |
4467 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
4468 | * @dev: Device on which the filter was set | |
4469 | * @rxq_index: RX queue index | |
4470 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
4471 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
4472 | * | |
4473 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
4474 | * this function for each installed filter and remove the filters for | |
4475 | * which it returns %true. | |
4476 | */ | |
4477 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
4478 | u32 flow_id, u16 filter_id) | |
4479 | { | |
4480 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
4481 | struct rps_dev_flow_table *flow_table; | |
4482 | struct rps_dev_flow *rflow; | |
4483 | bool expire = true; | |
a31196b0 | 4484 | unsigned int cpu; |
c445477d BH |
4485 | |
4486 | rcu_read_lock(); | |
4487 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4488 | if (flow_table && flow_id <= flow_table->mask) { | |
4489 | rflow = &flow_table->flows[flow_id]; | |
6aa7de05 | 4490 | cpu = READ_ONCE(rflow->cpu); |
a31196b0 | 4491 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
4492 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
4493 | rflow->last_qtail) < | |
4494 | (int)(10 * flow_table->mask))) | |
4495 | expire = false; | |
4496 | } | |
4497 | rcu_read_unlock(); | |
4498 | return expire; | |
4499 | } | |
4500 | EXPORT_SYMBOL(rps_may_expire_flow); | |
4501 | ||
4502 | #endif /* CONFIG_RFS_ACCEL */ | |
4503 | ||
0a9627f2 | 4504 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 4505 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 4506 | { |
e36fa2f7 ED |
4507 | struct softnet_data *sd = data; |
4508 | ||
eecfd7c4 | 4509 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 4510 | sd->received_rps++; |
0a9627f2 | 4511 | } |
e36fa2f7 | 4512 | |
fec5e652 | 4513 | #endif /* CONFIG_RPS */ |
0a9627f2 | 4514 | |
e36fa2f7 ED |
4515 | /* |
4516 | * Check if this softnet_data structure is another cpu one | |
4517 | * If yes, queue it to our IPI list and return 1 | |
4518 | * If no, return 0 | |
4519 | */ | |
4520 | static int rps_ipi_queued(struct softnet_data *sd) | |
4521 | { | |
4522 | #ifdef CONFIG_RPS | |
903ceff7 | 4523 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
4524 | |
4525 | if (sd != mysd) { | |
4526 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
4527 | mysd->rps_ipi_list = sd; | |
4528 | ||
4529 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4530 | return 1; | |
4531 | } | |
4532 | #endif /* CONFIG_RPS */ | |
4533 | return 0; | |
4534 | } | |
4535 | ||
99bbc707 WB |
4536 | #ifdef CONFIG_NET_FLOW_LIMIT |
4537 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
4538 | #endif | |
4539 | ||
4540 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
4541 | { | |
4542 | #ifdef CONFIG_NET_FLOW_LIMIT | |
4543 | struct sd_flow_limit *fl; | |
4544 | struct softnet_data *sd; | |
4545 | unsigned int old_flow, new_flow; | |
4546 | ||
4547 | if (qlen < (netdev_max_backlog >> 1)) | |
4548 | return false; | |
4549 | ||
903ceff7 | 4550 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
4551 | |
4552 | rcu_read_lock(); | |
4553 | fl = rcu_dereference(sd->flow_limit); | |
4554 | if (fl) { | |
3958afa1 | 4555 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
4556 | old_flow = fl->history[fl->history_head]; |
4557 | fl->history[fl->history_head] = new_flow; | |
4558 | ||
4559 | fl->history_head++; | |
4560 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
4561 | ||
4562 | if (likely(fl->buckets[old_flow])) | |
4563 | fl->buckets[old_flow]--; | |
4564 | ||
4565 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
4566 | fl->count++; | |
4567 | rcu_read_unlock(); | |
4568 | return true; | |
4569 | } | |
4570 | } | |
4571 | rcu_read_unlock(); | |
4572 | #endif | |
4573 | return false; | |
4574 | } | |
4575 | ||
0a9627f2 TH |
4576 | /* |
4577 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
4578 | * queue (may be a remote CPU queue). | |
4579 | */ | |
fec5e652 TH |
4580 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
4581 | unsigned int *qtail) | |
0a9627f2 | 4582 | { |
e36fa2f7 | 4583 | struct softnet_data *sd; |
0a9627f2 | 4584 | unsigned long flags; |
99bbc707 | 4585 | unsigned int qlen; |
0a9627f2 | 4586 | |
e36fa2f7 | 4587 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
4588 | |
4589 | local_irq_save(flags); | |
0a9627f2 | 4590 | |
e36fa2f7 | 4591 | rps_lock(sd); |
e9e4dd32 JA |
4592 | if (!netif_running(skb->dev)) |
4593 | goto drop; | |
99bbc707 WB |
4594 | qlen = skb_queue_len(&sd->input_pkt_queue); |
4595 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 4596 | if (qlen) { |
0a9627f2 | 4597 | enqueue: |
e36fa2f7 | 4598 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 4599 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 4600 | rps_unlock(sd); |
152102c7 | 4601 | local_irq_restore(flags); |
0a9627f2 TH |
4602 | return NET_RX_SUCCESS; |
4603 | } | |
4604 | ||
ebda37c2 ED |
4605 | /* Schedule NAPI for backlog device |
4606 | * We can use non atomic operation since we own the queue lock | |
4607 | */ | |
4608 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 4609 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 4610 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
4611 | } |
4612 | goto enqueue; | |
4613 | } | |
4614 | ||
e9e4dd32 | 4615 | drop: |
dee42870 | 4616 | sd->dropped++; |
e36fa2f7 | 4617 | rps_unlock(sd); |
0a9627f2 | 4618 | |
0a9627f2 TH |
4619 | local_irq_restore(flags); |
4620 | ||
caf586e5 | 4621 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
4622 | kfree_skb(skb); |
4623 | return NET_RX_DROP; | |
4624 | } | |
1da177e4 | 4625 | |
e817f856 JDB |
4626 | static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) |
4627 | { | |
4628 | struct net_device *dev = skb->dev; | |
4629 | struct netdev_rx_queue *rxqueue; | |
4630 | ||
4631 | rxqueue = dev->_rx; | |
4632 | ||
4633 | if (skb_rx_queue_recorded(skb)) { | |
4634 | u16 index = skb_get_rx_queue(skb); | |
4635 | ||
4636 | if (unlikely(index >= dev->real_num_rx_queues)) { | |
4637 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4638 | "%s received packet on queue %u, but number " | |
4639 | "of RX queues is %u\n", | |
4640 | dev->name, index, dev->real_num_rx_queues); | |
4641 | ||
4642 | return rxqueue; /* Return first rxqueue */ | |
4643 | } | |
4644 | rxqueue += index; | |
4645 | } | |
4646 | return rxqueue; | |
4647 | } | |
4648 | ||
d4455169 | 4649 | static u32 netif_receive_generic_xdp(struct sk_buff *skb, |
02671e23 | 4650 | struct xdp_buff *xdp, |
d4455169 JF |
4651 | struct bpf_prog *xdp_prog) |
4652 | { | |
be9df4af | 4653 | void *orig_data, *orig_data_end, *hard_start; |
e817f856 | 4654 | struct netdev_rx_queue *rxqueue; |
de8f3a83 | 4655 | u32 metalen, act = XDP_DROP; |
43b5169d | 4656 | u32 mac_len, frame_sz; |
29724956 JDB |
4657 | __be16 orig_eth_type; |
4658 | struct ethhdr *eth; | |
4659 | bool orig_bcast; | |
be9df4af | 4660 | int off; |
d4455169 JF |
4661 | |
4662 | /* Reinjected packets coming from act_mirred or similar should | |
4663 | * not get XDP generic processing. | |
4664 | */ | |
2c64605b | 4665 | if (skb_is_redirected(skb)) |
d4455169 JF |
4666 | return XDP_PASS; |
4667 | ||
de8f3a83 DB |
4668 | /* XDP packets must be linear and must have sufficient headroom |
4669 | * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also | |
4670 | * native XDP provides, thus we need to do it here as well. | |
4671 | */ | |
ad1e03b2 | 4672 | if (skb_cloned(skb) || skb_is_nonlinear(skb) || |
de8f3a83 DB |
4673 | skb_headroom(skb) < XDP_PACKET_HEADROOM) { |
4674 | int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); | |
4675 | int troom = skb->tail + skb->data_len - skb->end; | |
4676 | ||
4677 | /* In case we have to go down the path and also linearize, | |
4678 | * then lets do the pskb_expand_head() work just once here. | |
4679 | */ | |
4680 | if (pskb_expand_head(skb, | |
4681 | hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, | |
4682 | troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) | |
4683 | goto do_drop; | |
2d17d8d7 | 4684 | if (skb_linearize(skb)) |
de8f3a83 DB |
4685 | goto do_drop; |
4686 | } | |
d4455169 JF |
4687 | |
4688 | /* The XDP program wants to see the packet starting at the MAC | |
4689 | * header. | |
4690 | */ | |
4691 | mac_len = skb->data - skb_mac_header(skb); | |
be9df4af | 4692 | hard_start = skb->data - skb_headroom(skb); |
a075767b JDB |
4693 | |
4694 | /* SKB "head" area always have tailroom for skb_shared_info */ | |
be9df4af | 4695 | frame_sz = (void *)skb_end_pointer(skb) - hard_start; |
43b5169d | 4696 | frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
a075767b | 4697 | |
be9df4af LB |
4698 | rxqueue = netif_get_rxqueue(skb); |
4699 | xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); | |
4700 | xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, | |
4701 | skb_headlen(skb) + mac_len, true); | |
a075767b | 4702 | |
02671e23 BT |
4703 | orig_data_end = xdp->data_end; |
4704 | orig_data = xdp->data; | |
29724956 JDB |
4705 | eth = (struct ethhdr *)xdp->data; |
4706 | orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); | |
4707 | orig_eth_type = eth->h_proto; | |
d4455169 | 4708 | |
02671e23 | 4709 | act = bpf_prog_run_xdp(xdp_prog, xdp); |
d4455169 | 4710 | |
065af355 | 4711 | /* check if bpf_xdp_adjust_head was used */ |
02671e23 | 4712 | off = xdp->data - orig_data; |
065af355 JDB |
4713 | if (off) { |
4714 | if (off > 0) | |
4715 | __skb_pull(skb, off); | |
4716 | else if (off < 0) | |
4717 | __skb_push(skb, -off); | |
4718 | ||
4719 | skb->mac_header += off; | |
4720 | skb_reset_network_header(skb); | |
4721 | } | |
d4455169 | 4722 | |
a075767b JDB |
4723 | /* check if bpf_xdp_adjust_tail was used */ |
4724 | off = xdp->data_end - orig_data_end; | |
f7613120 | 4725 | if (off != 0) { |
02671e23 | 4726 | skb_set_tail_pointer(skb, xdp->data_end - xdp->data); |
a075767b | 4727 | skb->len += off; /* positive on grow, negative on shrink */ |
f7613120 | 4728 | } |
198d83bb | 4729 | |
29724956 JDB |
4730 | /* check if XDP changed eth hdr such SKB needs update */ |
4731 | eth = (struct ethhdr *)xdp->data; | |
4732 | if ((orig_eth_type != eth->h_proto) || | |
4733 | (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { | |
4734 | __skb_push(skb, ETH_HLEN); | |
4735 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4736 | } | |
4737 | ||
d4455169 | 4738 | switch (act) { |
6103aa96 | 4739 | case XDP_REDIRECT: |
d4455169 JF |
4740 | case XDP_TX: |
4741 | __skb_push(skb, mac_len); | |
de8f3a83 | 4742 | break; |
d4455169 | 4743 | case XDP_PASS: |
02671e23 | 4744 | metalen = xdp->data - xdp->data_meta; |
de8f3a83 DB |
4745 | if (metalen) |
4746 | skb_metadata_set(skb, metalen); | |
d4455169 | 4747 | break; |
d4455169 JF |
4748 | default: |
4749 | bpf_warn_invalid_xdp_action(act); | |
df561f66 | 4750 | fallthrough; |
d4455169 JF |
4751 | case XDP_ABORTED: |
4752 | trace_xdp_exception(skb->dev, xdp_prog, act); | |
df561f66 | 4753 | fallthrough; |
d4455169 JF |
4754 | case XDP_DROP: |
4755 | do_drop: | |
4756 | kfree_skb(skb); | |
4757 | break; | |
4758 | } | |
4759 | ||
4760 | return act; | |
4761 | } | |
4762 | ||
4763 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
4764 | * network taps in order to match in-driver-XDP behavior. | |
4765 | */ | |
7c497478 | 4766 | void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) |
d4455169 JF |
4767 | { |
4768 | struct net_device *dev = skb->dev; | |
4769 | struct netdev_queue *txq; | |
4770 | bool free_skb = true; | |
4771 | int cpu, rc; | |
4772 | ||
4bd97d51 | 4773 | txq = netdev_core_pick_tx(dev, skb, NULL); |
d4455169 JF |
4774 | cpu = smp_processor_id(); |
4775 | HARD_TX_LOCK(dev, txq, cpu); | |
4776 | if (!netif_xmit_stopped(txq)) { | |
4777 | rc = netdev_start_xmit(skb, dev, txq, 0); | |
4778 | if (dev_xmit_complete(rc)) | |
4779 | free_skb = false; | |
4780 | } | |
4781 | HARD_TX_UNLOCK(dev, txq); | |
4782 | if (free_skb) { | |
4783 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
4784 | kfree_skb(skb); | |
4785 | } | |
4786 | } | |
4787 | ||
02786475 | 4788 | static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); |
d4455169 | 4789 | |
7c497478 | 4790 | int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) |
d4455169 | 4791 | { |
d4455169 | 4792 | if (xdp_prog) { |
02671e23 BT |
4793 | struct xdp_buff xdp; |
4794 | u32 act; | |
6103aa96 | 4795 | int err; |
d4455169 | 4796 | |
02671e23 | 4797 | act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); |
d4455169 | 4798 | if (act != XDP_PASS) { |
6103aa96 JF |
4799 | switch (act) { |
4800 | case XDP_REDIRECT: | |
2facaad6 | 4801 | err = xdp_do_generic_redirect(skb->dev, skb, |
02671e23 | 4802 | &xdp, xdp_prog); |
6103aa96 JF |
4803 | if (err) |
4804 | goto out_redir; | |
02671e23 | 4805 | break; |
6103aa96 | 4806 | case XDP_TX: |
d4455169 | 4807 | generic_xdp_tx(skb, xdp_prog); |
6103aa96 JF |
4808 | break; |
4809 | } | |
d4455169 JF |
4810 | return XDP_DROP; |
4811 | } | |
4812 | } | |
4813 | return XDP_PASS; | |
6103aa96 | 4814 | out_redir: |
6103aa96 JF |
4815 | kfree_skb(skb); |
4816 | return XDP_DROP; | |
d4455169 | 4817 | } |
7c497478 | 4818 | EXPORT_SYMBOL_GPL(do_xdp_generic); |
d4455169 | 4819 | |
ae78dbfa | 4820 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 4821 | { |
b0e28f1e | 4822 | int ret; |
1da177e4 | 4823 | |
588f0330 | 4824 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 4825 | |
cf66ba58 | 4826 | trace_netif_rx(skb); |
d4455169 | 4827 | |
df334545 | 4828 | #ifdef CONFIG_RPS |
dc05360f | 4829 | if (static_branch_unlikely(&rps_needed)) { |
fec5e652 | 4830 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
4831 | int cpu; |
4832 | ||
cece1945 | 4833 | preempt_disable(); |
b0e28f1e | 4834 | rcu_read_lock(); |
fec5e652 TH |
4835 | |
4836 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
4837 | if (cpu < 0) |
4838 | cpu = smp_processor_id(); | |
fec5e652 TH |
4839 | |
4840 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4841 | ||
b0e28f1e | 4842 | rcu_read_unlock(); |
cece1945 | 4843 | preempt_enable(); |
adc9300e ED |
4844 | } else |
4845 | #endif | |
fec5e652 TH |
4846 | { |
4847 | unsigned int qtail; | |
f4563a75 | 4848 | |
fec5e652 TH |
4849 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); |
4850 | put_cpu(); | |
4851 | } | |
b0e28f1e | 4852 | return ret; |
1da177e4 | 4853 | } |
ae78dbfa BH |
4854 | |
4855 | /** | |
4856 | * netif_rx - post buffer to the network code | |
4857 | * @skb: buffer to post | |
4858 | * | |
4859 | * This function receives a packet from a device driver and queues it for | |
4860 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
4861 | * may be dropped during processing for congestion control or by the | |
4862 | * protocol layers. | |
4863 | * | |
4864 | * return values: | |
4865 | * NET_RX_SUCCESS (no congestion) | |
4866 | * NET_RX_DROP (packet was dropped) | |
4867 | * | |
4868 | */ | |
4869 | ||
4870 | int netif_rx(struct sk_buff *skb) | |
4871 | { | |
b0e3f1bd GB |
4872 | int ret; |
4873 | ||
ae78dbfa BH |
4874 | trace_netif_rx_entry(skb); |
4875 | ||
b0e3f1bd GB |
4876 | ret = netif_rx_internal(skb); |
4877 | trace_netif_rx_exit(ret); | |
4878 | ||
4879 | return ret; | |
ae78dbfa | 4880 | } |
d1b19dff | 4881 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
4882 | |
4883 | int netif_rx_ni(struct sk_buff *skb) | |
4884 | { | |
4885 | int err; | |
4886 | ||
ae78dbfa BH |
4887 | trace_netif_rx_ni_entry(skb); |
4888 | ||
1da177e4 | 4889 | preempt_disable(); |
ae78dbfa | 4890 | err = netif_rx_internal(skb); |
1da177e4 LT |
4891 | if (local_softirq_pending()) |
4892 | do_softirq(); | |
4893 | preempt_enable(); | |
b0e3f1bd | 4894 | trace_netif_rx_ni_exit(err); |
1da177e4 LT |
4895 | |
4896 | return err; | |
4897 | } | |
1da177e4 LT |
4898 | EXPORT_SYMBOL(netif_rx_ni); |
4899 | ||
c11171a4 SAS |
4900 | int netif_rx_any_context(struct sk_buff *skb) |
4901 | { | |
4902 | /* | |
4903 | * If invoked from contexts which do not invoke bottom half | |
4904 | * processing either at return from interrupt or when softrqs are | |
4905 | * reenabled, use netif_rx_ni() which invokes bottomhalf processing | |
4906 | * directly. | |
4907 | */ | |
4908 | if (in_interrupt()) | |
4909 | return netif_rx(skb); | |
4910 | else | |
4911 | return netif_rx_ni(skb); | |
4912 | } | |
4913 | EXPORT_SYMBOL(netif_rx_any_context); | |
4914 | ||
0766f788 | 4915 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 4916 | { |
903ceff7 | 4917 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
4918 | |
4919 | if (sd->completion_queue) { | |
4920 | struct sk_buff *clist; | |
4921 | ||
4922 | local_irq_disable(); | |
4923 | clist = sd->completion_queue; | |
4924 | sd->completion_queue = NULL; | |
4925 | local_irq_enable(); | |
4926 | ||
4927 | while (clist) { | |
4928 | struct sk_buff *skb = clist; | |
f4563a75 | 4929 | |
1da177e4 LT |
4930 | clist = clist->next; |
4931 | ||
63354797 | 4932 | WARN_ON(refcount_read(&skb->users)); |
e6247027 ED |
4933 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
4934 | trace_consume_skb(skb); | |
4935 | else | |
4936 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
4937 | |
4938 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
4939 | __kfree_skb(skb); | |
4940 | else | |
4941 | __kfree_skb_defer(skb); | |
1da177e4 LT |
4942 | } |
4943 | } | |
4944 | ||
4945 | if (sd->output_queue) { | |
37437bb2 | 4946 | struct Qdisc *head; |
1da177e4 LT |
4947 | |
4948 | local_irq_disable(); | |
4949 | head = sd->output_queue; | |
4950 | sd->output_queue = NULL; | |
a9cbd588 | 4951 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
4952 | local_irq_enable(); |
4953 | ||
4954 | while (head) { | |
37437bb2 | 4955 | struct Qdisc *q = head; |
6b3ba914 | 4956 | spinlock_t *root_lock = NULL; |
37437bb2 | 4957 | |
1da177e4 LT |
4958 | head = head->next_sched; |
4959 | ||
6b3ba914 JF |
4960 | if (!(q->flags & TCQ_F_NOLOCK)) { |
4961 | root_lock = qdisc_lock(q); | |
4962 | spin_lock(root_lock); | |
4963 | } | |
3bcb846c ED |
4964 | /* We need to make sure head->next_sched is read |
4965 | * before clearing __QDISC_STATE_SCHED | |
4966 | */ | |
4967 | smp_mb__before_atomic(); | |
4968 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
4969 | qdisc_run(q); | |
6b3ba914 JF |
4970 | if (root_lock) |
4971 | spin_unlock(root_lock); | |
1da177e4 LT |
4972 | } |
4973 | } | |
f53c7239 SK |
4974 | |
4975 | xfrm_dev_backlog(sd); | |
1da177e4 LT |
4976 | } |
4977 | ||
181402a5 | 4978 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
4979 | /* This hook is defined here for ATM LANE */ |
4980 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
4981 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 4982 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 4983 | #endif |
1da177e4 | 4984 | |
1f211a1b DB |
4985 | static inline struct sk_buff * |
4986 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
9aa1206e | 4987 | struct net_device *orig_dev, bool *another) |
f697c3e8 | 4988 | { |
e7582bab | 4989 | #ifdef CONFIG_NET_CLS_ACT |
46209401 | 4990 | struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); |
d2788d34 | 4991 | struct tcf_result cl_res; |
24824a09 | 4992 | |
c9e99fd0 DB |
4993 | /* If there's at least one ingress present somewhere (so |
4994 | * we get here via enabled static key), remaining devices | |
4995 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 4996 | * out here. |
c9e99fd0 | 4997 | */ |
46209401 | 4998 | if (!miniq) |
4577139b | 4999 | return skb; |
46209401 | 5000 | |
f697c3e8 HX |
5001 | if (*pt_prev) { |
5002 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
5003 | *pt_prev = NULL; | |
1da177e4 LT |
5004 | } |
5005 | ||
3365495c | 5006 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
aadaca9e | 5007 | qdisc_skb_cb(skb)->mru = 0; |
7baf2429 | 5008 | qdisc_skb_cb(skb)->post_ct = false; |
8dc07fdb | 5009 | skb->tc_at_ingress = 1; |
46209401 | 5010 | mini_qdisc_bstats_cpu_update(miniq, skb); |
c9e99fd0 | 5011 | |
7d17c544 PB |
5012 | switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list, |
5013 | &cl_res, false)) { | |
d2788d34 DB |
5014 | case TC_ACT_OK: |
5015 | case TC_ACT_RECLASSIFY: | |
5016 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
5017 | break; | |
5018 | case TC_ACT_SHOT: | |
46209401 | 5019 | mini_qdisc_qstats_cpu_drop(miniq); |
8a3a4c6e ED |
5020 | kfree_skb(skb); |
5021 | return NULL; | |
d2788d34 DB |
5022 | case TC_ACT_STOLEN: |
5023 | case TC_ACT_QUEUED: | |
e25ea21f | 5024 | case TC_ACT_TRAP: |
8a3a4c6e | 5025 | consume_skb(skb); |
d2788d34 | 5026 | return NULL; |
27b29f63 AS |
5027 | case TC_ACT_REDIRECT: |
5028 | /* skb_mac_header check was done by cls/act_bpf, so | |
5029 | * we can safely push the L2 header back before | |
5030 | * redirecting to another netdev | |
5031 | */ | |
5032 | __skb_push(skb, skb->mac_len); | |
9aa1206e DB |
5033 | if (skb_do_redirect(skb) == -EAGAIN) { |
5034 | __skb_pull(skb, skb->mac_len); | |
5035 | *another = true; | |
5036 | break; | |
5037 | } | |
27b29f63 | 5038 | return NULL; |
720f22fe | 5039 | case TC_ACT_CONSUMED: |
cd11b164 | 5040 | return NULL; |
d2788d34 DB |
5041 | default: |
5042 | break; | |
f697c3e8 | 5043 | } |
e7582bab | 5044 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
5045 | return skb; |
5046 | } | |
1da177e4 | 5047 | |
24b27fc4 MB |
5048 | /** |
5049 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
5050 | * @dev: device to check | |
5051 | * | |
5052 | * Check if a receive handler is already registered for a given device. | |
5053 | * Return true if there one. | |
5054 | * | |
5055 | * The caller must hold the rtnl_mutex. | |
5056 | */ | |
5057 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
5058 | { | |
5059 | ASSERT_RTNL(); | |
5060 | return dev && rtnl_dereference(dev->rx_handler); | |
5061 | } | |
5062 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
5063 | ||
ab95bfe0 JP |
5064 | /** |
5065 | * netdev_rx_handler_register - register receive handler | |
5066 | * @dev: device to register a handler for | |
5067 | * @rx_handler: receive handler to register | |
93e2c32b | 5068 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 5069 | * |
e227867f | 5070 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
5071 | * called from __netif_receive_skb. A negative errno code is returned |
5072 | * on a failure. | |
5073 | * | |
5074 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
5075 | * |
5076 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
5077 | */ |
5078 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
5079 | rx_handler_func_t *rx_handler, |
5080 | void *rx_handler_data) | |
ab95bfe0 | 5081 | { |
1b7cd004 | 5082 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
5083 | return -EBUSY; |
5084 | ||
f5426250 PA |
5085 | if (dev->priv_flags & IFF_NO_RX_HANDLER) |
5086 | return -EINVAL; | |
5087 | ||
00cfec37 | 5088 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 5089 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
5090 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
5091 | ||
5092 | return 0; | |
5093 | } | |
5094 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
5095 | ||
5096 | /** | |
5097 | * netdev_rx_handler_unregister - unregister receive handler | |
5098 | * @dev: device to unregister a handler from | |
5099 | * | |
166ec369 | 5100 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
5101 | * |
5102 | * The caller must hold the rtnl_mutex. | |
5103 | */ | |
5104 | void netdev_rx_handler_unregister(struct net_device *dev) | |
5105 | { | |
5106 | ||
5107 | ASSERT_RTNL(); | |
a9b3cd7f | 5108 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
5109 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
5110 | * section has a guarantee to see a non NULL rx_handler_data | |
5111 | * as well. | |
5112 | */ | |
5113 | synchronize_net(); | |
a9b3cd7f | 5114 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
5115 | } |
5116 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
5117 | ||
b4b9e355 MG |
5118 | /* |
5119 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
5120 | * the special handling of PFMEMALLOC skbs. | |
5121 | */ | |
5122 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
5123 | { | |
5124 | switch (skb->protocol) { | |
2b8837ae JP |
5125 | case htons(ETH_P_ARP): |
5126 | case htons(ETH_P_IP): | |
5127 | case htons(ETH_P_IPV6): | |
5128 | case htons(ETH_P_8021Q): | |
5129 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
5130 | return true; |
5131 | default: | |
5132 | return false; | |
5133 | } | |
5134 | } | |
5135 | ||
e687ad60 PN |
5136 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
5137 | int *ret, struct net_device *orig_dev) | |
5138 | { | |
5139 | if (nf_hook_ingress_active(skb)) { | |
2c1e2703 AC |
5140 | int ingress_retval; |
5141 | ||
e687ad60 PN |
5142 | if (*pt_prev) { |
5143 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
5144 | *pt_prev = NULL; | |
5145 | } | |
5146 | ||
2c1e2703 AC |
5147 | rcu_read_lock(); |
5148 | ingress_retval = nf_hook_ingress(skb); | |
5149 | rcu_read_unlock(); | |
5150 | return ingress_retval; | |
e687ad60 PN |
5151 | } |
5152 | return 0; | |
5153 | } | |
e687ad60 | 5154 | |
c0bbbdc3 | 5155 | static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, |
88eb1944 | 5156 | struct packet_type **ppt_prev) |
1da177e4 LT |
5157 | { |
5158 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 5159 | rx_handler_func_t *rx_handler; |
c0bbbdc3 | 5160 | struct sk_buff *skb = *pskb; |
f2ccd8fa | 5161 | struct net_device *orig_dev; |
8a4eb573 | 5162 | bool deliver_exact = false; |
1da177e4 | 5163 | int ret = NET_RX_DROP; |
252e3346 | 5164 | __be16 type; |
1da177e4 | 5165 | |
588f0330 | 5166 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 5167 | |
cf66ba58 | 5168 | trace_netif_receive_skb(skb); |
9b22ea56 | 5169 | |
cc9bd5ce | 5170 | orig_dev = skb->dev; |
8f903c70 | 5171 | |
c1d2bbe1 | 5172 | skb_reset_network_header(skb); |
fda55eca ED |
5173 | if (!skb_transport_header_was_set(skb)) |
5174 | skb_reset_transport_header(skb); | |
0b5c9db1 | 5175 | skb_reset_mac_len(skb); |
1da177e4 LT |
5176 | |
5177 | pt_prev = NULL; | |
5178 | ||
63d8ea7f | 5179 | another_round: |
b6858177 | 5180 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
5181 | |
5182 | __this_cpu_inc(softnet_data.processed); | |
5183 | ||
458bf2f2 SH |
5184 | if (static_branch_unlikely(&generic_xdp_needed_key)) { |
5185 | int ret2; | |
5186 | ||
5187 | preempt_disable(); | |
5188 | ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
5189 | preempt_enable(); | |
5190 | ||
c0bbbdc3 BS |
5191 | if (ret2 != XDP_PASS) { |
5192 | ret = NET_RX_DROP; | |
5193 | goto out; | |
5194 | } | |
458bf2f2 SH |
5195 | skb_reset_mac_len(skb); |
5196 | } | |
5197 | ||
324cefaf | 5198 | if (eth_type_vlan(skb->protocol)) { |
0d5501c1 | 5199 | skb = skb_vlan_untag(skb); |
bcc6d479 | 5200 | if (unlikely(!skb)) |
2c17d27c | 5201 | goto out; |
bcc6d479 JP |
5202 | } |
5203 | ||
e7246e12 WB |
5204 | if (skb_skip_tc_classify(skb)) |
5205 | goto skip_classify; | |
1da177e4 | 5206 | |
9754e293 | 5207 | if (pfmemalloc) |
b4b9e355 MG |
5208 | goto skip_taps; |
5209 | ||
1da177e4 | 5210 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
5211 | if (pt_prev) |
5212 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5213 | pt_prev = ptype; | |
5214 | } | |
5215 | ||
5216 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
5217 | if (pt_prev) | |
5218 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5219 | pt_prev = ptype; | |
1da177e4 LT |
5220 | } |
5221 | ||
b4b9e355 | 5222 | skip_taps: |
1cf51900 | 5223 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 5224 | if (static_branch_unlikely(&ingress_needed_key)) { |
9aa1206e DB |
5225 | bool another = false; |
5226 | ||
5227 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, | |
5228 | &another); | |
5229 | if (another) | |
5230 | goto another_round; | |
4577139b | 5231 | if (!skb) |
2c17d27c | 5232 | goto out; |
e687ad60 PN |
5233 | |
5234 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 5235 | goto out; |
4577139b | 5236 | } |
1cf51900 | 5237 | #endif |
2c64605b | 5238 | skb_reset_redirect(skb); |
e7246e12 | 5239 | skip_classify: |
9754e293 | 5240 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
5241 | goto drop; |
5242 | ||
df8a39de | 5243 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
5244 | if (pt_prev) { |
5245 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5246 | pt_prev = NULL; | |
5247 | } | |
48cc32d3 | 5248 | if (vlan_do_receive(&skb)) |
2425717b JF |
5249 | goto another_round; |
5250 | else if (unlikely(!skb)) | |
2c17d27c | 5251 | goto out; |
2425717b JF |
5252 | } |
5253 | ||
48cc32d3 | 5254 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
5255 | if (rx_handler) { |
5256 | if (pt_prev) { | |
5257 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
5258 | pt_prev = NULL; | |
5259 | } | |
8a4eb573 JP |
5260 | switch (rx_handler(&skb)) { |
5261 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 5262 | ret = NET_RX_SUCCESS; |
2c17d27c | 5263 | goto out; |
8a4eb573 | 5264 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 5265 | goto another_round; |
8a4eb573 JP |
5266 | case RX_HANDLER_EXACT: |
5267 | deliver_exact = true; | |
b1866bff | 5268 | break; |
8a4eb573 JP |
5269 | case RX_HANDLER_PASS: |
5270 | break; | |
5271 | default: | |
5272 | BUG(); | |
5273 | } | |
ab95bfe0 | 5274 | } |
1da177e4 | 5275 | |
b14a9fc4 | 5276 | if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { |
36b2f61a GV |
5277 | check_vlan_id: |
5278 | if (skb_vlan_tag_get_id(skb)) { | |
5279 | /* Vlan id is non 0 and vlan_do_receive() above couldn't | |
5280 | * find vlan device. | |
5281 | */ | |
d4b812de | 5282 | skb->pkt_type = PACKET_OTHERHOST; |
324cefaf | 5283 | } else if (eth_type_vlan(skb->protocol)) { |
36b2f61a GV |
5284 | /* Outer header is 802.1P with vlan 0, inner header is |
5285 | * 802.1Q or 802.1AD and vlan_do_receive() above could | |
5286 | * not find vlan dev for vlan id 0. | |
5287 | */ | |
5288 | __vlan_hwaccel_clear_tag(skb); | |
5289 | skb = skb_vlan_untag(skb); | |
5290 | if (unlikely(!skb)) | |
5291 | goto out; | |
5292 | if (vlan_do_receive(&skb)) | |
5293 | /* After stripping off 802.1P header with vlan 0 | |
5294 | * vlan dev is found for inner header. | |
5295 | */ | |
5296 | goto another_round; | |
5297 | else if (unlikely(!skb)) | |
5298 | goto out; | |
5299 | else | |
5300 | /* We have stripped outer 802.1P vlan 0 header. | |
5301 | * But could not find vlan dev. | |
5302 | * check again for vlan id to set OTHERHOST. | |
5303 | */ | |
5304 | goto check_vlan_id; | |
5305 | } | |
d4b812de ED |
5306 | /* Note: we might in the future use prio bits |
5307 | * and set skb->priority like in vlan_do_receive() | |
5308 | * For the time being, just ignore Priority Code Point | |
5309 | */ | |
b1817524 | 5310 | __vlan_hwaccel_clear_tag(skb); |
d4b812de | 5311 | } |
48cc32d3 | 5312 | |
7866a621 SN |
5313 | type = skb->protocol; |
5314 | ||
63d8ea7f | 5315 | /* deliver only exact match when indicated */ |
7866a621 SN |
5316 | if (likely(!deliver_exact)) { |
5317 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
5318 | &ptype_base[ntohs(type) & | |
5319 | PTYPE_HASH_MASK]); | |
5320 | } | |
1f3c8804 | 5321 | |
7866a621 SN |
5322 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
5323 | &orig_dev->ptype_specific); | |
5324 | ||
5325 | if (unlikely(skb->dev != orig_dev)) { | |
5326 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
5327 | &skb->dev->ptype_specific); | |
1da177e4 LT |
5328 | } |
5329 | ||
5330 | if (pt_prev) { | |
1f8b977a | 5331 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
0e698bf6 | 5332 | goto drop; |
88eb1944 | 5333 | *ppt_prev = pt_prev; |
1da177e4 | 5334 | } else { |
b4b9e355 | 5335 | drop: |
6e7333d3 JW |
5336 | if (!deliver_exact) |
5337 | atomic_long_inc(&skb->dev->rx_dropped); | |
5338 | else | |
5339 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
5340 | kfree_skb(skb); |
5341 | /* Jamal, now you will not able to escape explaining | |
5342 | * me how you were going to use this. :-) | |
5343 | */ | |
5344 | ret = NET_RX_DROP; | |
5345 | } | |
5346 | ||
2c17d27c | 5347 | out: |
c0bbbdc3 BS |
5348 | /* The invariant here is that if *ppt_prev is not NULL |
5349 | * then skb should also be non-NULL. | |
5350 | * | |
5351 | * Apparently *ppt_prev assignment above holds this invariant due to | |
5352 | * skb dereferencing near it. | |
5353 | */ | |
5354 | *pskb = skb; | |
9754e293 DM |
5355 | return ret; |
5356 | } | |
5357 | ||
88eb1944 EC |
5358 | static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) |
5359 | { | |
5360 | struct net_device *orig_dev = skb->dev; | |
5361 | struct packet_type *pt_prev = NULL; | |
5362 | int ret; | |
5363 | ||
c0bbbdc3 | 5364 | ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); |
88eb1944 | 5365 | if (pt_prev) |
f5737cba PA |
5366 | ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, |
5367 | skb->dev, pt_prev, orig_dev); | |
88eb1944 EC |
5368 | return ret; |
5369 | } | |
5370 | ||
1c601d82 JDB |
5371 | /** |
5372 | * netif_receive_skb_core - special purpose version of netif_receive_skb | |
5373 | * @skb: buffer to process | |
5374 | * | |
5375 | * More direct receive version of netif_receive_skb(). It should | |
5376 | * only be used by callers that have a need to skip RPS and Generic XDP. | |
2de9780f | 5377 | * Caller must also take care of handling if ``(page_is_)pfmemalloc``. |
1c601d82 JDB |
5378 | * |
5379 | * This function may only be called from softirq context and interrupts | |
5380 | * should be enabled. | |
5381 | * | |
5382 | * Return values (usually ignored): | |
5383 | * NET_RX_SUCCESS: no congestion | |
5384 | * NET_RX_DROP: packet was dropped | |
5385 | */ | |
5386 | int netif_receive_skb_core(struct sk_buff *skb) | |
5387 | { | |
5388 | int ret; | |
5389 | ||
5390 | rcu_read_lock(); | |
88eb1944 | 5391 | ret = __netif_receive_skb_one_core(skb, false); |
1c601d82 JDB |
5392 | rcu_read_unlock(); |
5393 | ||
5394 | return ret; | |
5395 | } | |
5396 | EXPORT_SYMBOL(netif_receive_skb_core); | |
5397 | ||
88eb1944 EC |
5398 | static inline void __netif_receive_skb_list_ptype(struct list_head *head, |
5399 | struct packet_type *pt_prev, | |
5400 | struct net_device *orig_dev) | |
4ce0017a EC |
5401 | { |
5402 | struct sk_buff *skb, *next; | |
5403 | ||
88eb1944 EC |
5404 | if (!pt_prev) |
5405 | return; | |
5406 | if (list_empty(head)) | |
5407 | return; | |
17266ee9 | 5408 | if (pt_prev->list_func != NULL) |
fdf71426 PA |
5409 | INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, |
5410 | ip_list_rcv, head, pt_prev, orig_dev); | |
17266ee9 | 5411 | else |
9a5a90d1 AL |
5412 | list_for_each_entry_safe(skb, next, head, list) { |
5413 | skb_list_del_init(skb); | |
fdf71426 | 5414 | pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
9a5a90d1 | 5415 | } |
88eb1944 EC |
5416 | } |
5417 | ||
5418 | static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) | |
5419 | { | |
5420 | /* Fast-path assumptions: | |
5421 | * - There is no RX handler. | |
5422 | * - Only one packet_type matches. | |
5423 | * If either of these fails, we will end up doing some per-packet | |
5424 | * processing in-line, then handling the 'last ptype' for the whole | |
5425 | * sublist. This can't cause out-of-order delivery to any single ptype, | |
5426 | * because the 'last ptype' must be constant across the sublist, and all | |
5427 | * other ptypes are handled per-packet. | |
5428 | */ | |
5429 | /* Current (common) ptype of sublist */ | |
5430 | struct packet_type *pt_curr = NULL; | |
5431 | /* Current (common) orig_dev of sublist */ | |
5432 | struct net_device *od_curr = NULL; | |
5433 | struct list_head sublist; | |
5434 | struct sk_buff *skb, *next; | |
5435 | ||
9af86f93 | 5436 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5437 | list_for_each_entry_safe(skb, next, head, list) { |
5438 | struct net_device *orig_dev = skb->dev; | |
5439 | struct packet_type *pt_prev = NULL; | |
5440 | ||
22f6bbb7 | 5441 | skb_list_del_init(skb); |
c0bbbdc3 | 5442 | __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); |
9af86f93 EC |
5443 | if (!pt_prev) |
5444 | continue; | |
88eb1944 EC |
5445 | if (pt_curr != pt_prev || od_curr != orig_dev) { |
5446 | /* dispatch old sublist */ | |
88eb1944 EC |
5447 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
5448 | /* start new sublist */ | |
9af86f93 | 5449 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5450 | pt_curr = pt_prev; |
5451 | od_curr = orig_dev; | |
5452 | } | |
9af86f93 | 5453 | list_add_tail(&skb->list, &sublist); |
88eb1944 EC |
5454 | } |
5455 | ||
5456 | /* dispatch final sublist */ | |
9af86f93 | 5457 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
4ce0017a EC |
5458 | } |
5459 | ||
9754e293 DM |
5460 | static int __netif_receive_skb(struct sk_buff *skb) |
5461 | { | |
5462 | int ret; | |
5463 | ||
5464 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 5465 | unsigned int noreclaim_flag; |
9754e293 DM |
5466 | |
5467 | /* | |
5468 | * PFMEMALLOC skbs are special, they should | |
5469 | * - be delivered to SOCK_MEMALLOC sockets only | |
5470 | * - stay away from userspace | |
5471 | * - have bounded memory usage | |
5472 | * | |
5473 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
5474 | * context down to all allocation sites. | |
5475 | */ | |
f1083048 | 5476 | noreclaim_flag = memalloc_noreclaim_save(); |
88eb1944 | 5477 | ret = __netif_receive_skb_one_core(skb, true); |
f1083048 | 5478 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 | 5479 | } else |
88eb1944 | 5480 | ret = __netif_receive_skb_one_core(skb, false); |
9754e293 | 5481 | |
1da177e4 LT |
5482 | return ret; |
5483 | } | |
0a9627f2 | 5484 | |
4ce0017a EC |
5485 | static void __netif_receive_skb_list(struct list_head *head) |
5486 | { | |
5487 | unsigned long noreclaim_flag = 0; | |
5488 | struct sk_buff *skb, *next; | |
5489 | bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ | |
5490 | ||
5491 | list_for_each_entry_safe(skb, next, head, list) { | |
5492 | if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { | |
5493 | struct list_head sublist; | |
5494 | ||
5495 | /* Handle the previous sublist */ | |
5496 | list_cut_before(&sublist, head, &skb->list); | |
b9f463d6 EC |
5497 | if (!list_empty(&sublist)) |
5498 | __netif_receive_skb_list_core(&sublist, pfmemalloc); | |
4ce0017a EC |
5499 | pfmemalloc = !pfmemalloc; |
5500 | /* See comments in __netif_receive_skb */ | |
5501 | if (pfmemalloc) | |
5502 | noreclaim_flag = memalloc_noreclaim_save(); | |
5503 | else | |
5504 | memalloc_noreclaim_restore(noreclaim_flag); | |
5505 | } | |
5506 | } | |
5507 | /* Handle the remaining sublist */ | |
b9f463d6 EC |
5508 | if (!list_empty(head)) |
5509 | __netif_receive_skb_list_core(head, pfmemalloc); | |
4ce0017a EC |
5510 | /* Restore pflags */ |
5511 | if (pfmemalloc) | |
5512 | memalloc_noreclaim_restore(noreclaim_flag); | |
5513 | } | |
5514 | ||
f4e63525 | 5515 | static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) |
b5cdae32 | 5516 | { |
58038695 | 5517 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); |
b5cdae32 DM |
5518 | struct bpf_prog *new = xdp->prog; |
5519 | int ret = 0; | |
5520 | ||
fbee97fe DA |
5521 | if (new) { |
5522 | u32 i; | |
5523 | ||
984fe94f YZ |
5524 | mutex_lock(&new->aux->used_maps_mutex); |
5525 | ||
fbee97fe DA |
5526 | /* generic XDP does not work with DEVMAPs that can |
5527 | * have a bpf_prog installed on an entry | |
5528 | */ | |
5529 | for (i = 0; i < new->aux->used_map_cnt; i++) { | |
984fe94f YZ |
5530 | if (dev_map_can_have_prog(new->aux->used_maps[i]) || |
5531 | cpu_map_prog_allowed(new->aux->used_maps[i])) { | |
5532 | mutex_unlock(&new->aux->used_maps_mutex); | |
92164774 | 5533 | return -EINVAL; |
984fe94f | 5534 | } |
fbee97fe | 5535 | } |
984fe94f YZ |
5536 | |
5537 | mutex_unlock(&new->aux->used_maps_mutex); | |
fbee97fe DA |
5538 | } |
5539 | ||
b5cdae32 | 5540 | switch (xdp->command) { |
58038695 | 5541 | case XDP_SETUP_PROG: |
b5cdae32 DM |
5542 | rcu_assign_pointer(dev->xdp_prog, new); |
5543 | if (old) | |
5544 | bpf_prog_put(old); | |
5545 | ||
5546 | if (old && !new) { | |
02786475 | 5547 | static_branch_dec(&generic_xdp_needed_key); |
b5cdae32 | 5548 | } else if (new && !old) { |
02786475 | 5549 | static_branch_inc(&generic_xdp_needed_key); |
b5cdae32 | 5550 | dev_disable_lro(dev); |
56f5aa77 | 5551 | dev_disable_gro_hw(dev); |
b5cdae32 DM |
5552 | } |
5553 | break; | |
b5cdae32 | 5554 | |
b5cdae32 DM |
5555 | default: |
5556 | ret = -EINVAL; | |
5557 | break; | |
5558 | } | |
5559 | ||
5560 | return ret; | |
5561 | } | |
5562 | ||
ae78dbfa | 5563 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 5564 | { |
2c17d27c JA |
5565 | int ret; |
5566 | ||
588f0330 | 5567 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 5568 | |
c1f19b51 RC |
5569 | if (skb_defer_rx_timestamp(skb)) |
5570 | return NET_RX_SUCCESS; | |
5571 | ||
bbbe211c | 5572 | rcu_read_lock(); |
df334545 | 5573 | #ifdef CONFIG_RPS |
dc05360f | 5574 | if (static_branch_unlikely(&rps_needed)) { |
3b098e2d | 5575 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 5576 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 5577 | |
3b098e2d ED |
5578 | if (cpu >= 0) { |
5579 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
5580 | rcu_read_unlock(); | |
adc9300e | 5581 | return ret; |
3b098e2d | 5582 | } |
fec5e652 | 5583 | } |
1e94d72f | 5584 | #endif |
2c17d27c JA |
5585 | ret = __netif_receive_skb(skb); |
5586 | rcu_read_unlock(); | |
5587 | return ret; | |
0a9627f2 | 5588 | } |
ae78dbfa | 5589 | |
7da517a3 EC |
5590 | static void netif_receive_skb_list_internal(struct list_head *head) |
5591 | { | |
7da517a3 | 5592 | struct sk_buff *skb, *next; |
8c057efa | 5593 | struct list_head sublist; |
7da517a3 | 5594 | |
8c057efa | 5595 | INIT_LIST_HEAD(&sublist); |
7da517a3 EC |
5596 | list_for_each_entry_safe(skb, next, head, list) { |
5597 | net_timestamp_check(netdev_tstamp_prequeue, skb); | |
22f6bbb7 | 5598 | skb_list_del_init(skb); |
8c057efa EC |
5599 | if (!skb_defer_rx_timestamp(skb)) |
5600 | list_add_tail(&skb->list, &sublist); | |
7da517a3 | 5601 | } |
8c057efa | 5602 | list_splice_init(&sublist, head); |
7da517a3 | 5603 | |
7da517a3 EC |
5604 | rcu_read_lock(); |
5605 | #ifdef CONFIG_RPS | |
dc05360f | 5606 | if (static_branch_unlikely(&rps_needed)) { |
7da517a3 EC |
5607 | list_for_each_entry_safe(skb, next, head, list) { |
5608 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
5609 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
5610 | ||
5611 | if (cpu >= 0) { | |
8c057efa | 5612 | /* Will be handled, remove from list */ |
22f6bbb7 | 5613 | skb_list_del_init(skb); |
8c057efa | 5614 | enqueue_to_backlog(skb, cpu, &rflow->last_qtail); |
7da517a3 EC |
5615 | } |
5616 | } | |
5617 | } | |
5618 | #endif | |
5619 | __netif_receive_skb_list(head); | |
5620 | rcu_read_unlock(); | |
5621 | } | |
5622 | ||
ae78dbfa BH |
5623 | /** |
5624 | * netif_receive_skb - process receive buffer from network | |
5625 | * @skb: buffer to process | |
5626 | * | |
5627 | * netif_receive_skb() is the main receive data processing function. | |
5628 | * It always succeeds. The buffer may be dropped during processing | |
5629 | * for congestion control or by the protocol layers. | |
5630 | * | |
5631 | * This function may only be called from softirq context and interrupts | |
5632 | * should be enabled. | |
5633 | * | |
5634 | * Return values (usually ignored): | |
5635 | * NET_RX_SUCCESS: no congestion | |
5636 | * NET_RX_DROP: packet was dropped | |
5637 | */ | |
04eb4489 | 5638 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa | 5639 | { |
b0e3f1bd GB |
5640 | int ret; |
5641 | ||
ae78dbfa BH |
5642 | trace_netif_receive_skb_entry(skb); |
5643 | ||
b0e3f1bd GB |
5644 | ret = netif_receive_skb_internal(skb); |
5645 | trace_netif_receive_skb_exit(ret); | |
5646 | ||
5647 | return ret; | |
ae78dbfa | 5648 | } |
04eb4489 | 5649 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 5650 | |
f6ad8c1b EC |
5651 | /** |
5652 | * netif_receive_skb_list - process many receive buffers from network | |
5653 | * @head: list of skbs to process. | |
5654 | * | |
7da517a3 EC |
5655 | * Since return value of netif_receive_skb() is normally ignored, and |
5656 | * wouldn't be meaningful for a list, this function returns void. | |
f6ad8c1b EC |
5657 | * |
5658 | * This function may only be called from softirq context and interrupts | |
5659 | * should be enabled. | |
5660 | */ | |
5661 | void netif_receive_skb_list(struct list_head *head) | |
5662 | { | |
7da517a3 | 5663 | struct sk_buff *skb; |
f6ad8c1b | 5664 | |
b9f463d6 EC |
5665 | if (list_empty(head)) |
5666 | return; | |
b0e3f1bd GB |
5667 | if (trace_netif_receive_skb_list_entry_enabled()) { |
5668 | list_for_each_entry(skb, head, list) | |
5669 | trace_netif_receive_skb_list_entry(skb); | |
5670 | } | |
7da517a3 | 5671 | netif_receive_skb_list_internal(head); |
b0e3f1bd | 5672 | trace_netif_receive_skb_list_exit(0); |
f6ad8c1b EC |
5673 | } |
5674 | EXPORT_SYMBOL(netif_receive_skb_list); | |
5675 | ||
ce1e2a77 | 5676 | static DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
5677 | |
5678 | /* Network device is going away, flush any packets still pending */ | |
5679 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 5680 | { |
6e583ce5 | 5681 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
5682 | struct softnet_data *sd; |
5683 | ||
5684 | local_bh_disable(); | |
5685 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 5686 | |
145dd5f9 | 5687 | local_irq_disable(); |
e36fa2f7 | 5688 | rps_lock(sd); |
6e7676c1 | 5689 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 5690 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 5691 | __skb_unlink(skb, &sd->input_pkt_queue); |
7df5cb75 | 5692 | dev_kfree_skb_irq(skb); |
76cc8b13 | 5693 | input_queue_head_incr(sd); |
6e583ce5 | 5694 | } |
6e7676c1 | 5695 | } |
e36fa2f7 | 5696 | rps_unlock(sd); |
145dd5f9 | 5697 | local_irq_enable(); |
6e7676c1 CG |
5698 | |
5699 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 5700 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
5701 | __skb_unlink(skb, &sd->process_queue); |
5702 | kfree_skb(skb); | |
76cc8b13 | 5703 | input_queue_head_incr(sd); |
6e7676c1 CG |
5704 | } |
5705 | } | |
145dd5f9 PA |
5706 | local_bh_enable(); |
5707 | } | |
5708 | ||
2de79ee2 PA |
5709 | static bool flush_required(int cpu) |
5710 | { | |
5711 | #if IS_ENABLED(CONFIG_RPS) | |
5712 | struct softnet_data *sd = &per_cpu(softnet_data, cpu); | |
5713 | bool do_flush; | |
5714 | ||
5715 | local_irq_disable(); | |
5716 | rps_lock(sd); | |
5717 | ||
5718 | /* as insertion into process_queue happens with the rps lock held, | |
5719 | * process_queue access may race only with dequeue | |
5720 | */ | |
5721 | do_flush = !skb_queue_empty(&sd->input_pkt_queue) || | |
5722 | !skb_queue_empty_lockless(&sd->process_queue); | |
5723 | rps_unlock(sd); | |
5724 | local_irq_enable(); | |
5725 | ||
5726 | return do_flush; | |
5727 | #endif | |
5728 | /* without RPS we can't safely check input_pkt_queue: during a | |
5729 | * concurrent remote skb_queue_splice() we can detect as empty both | |
5730 | * input_pkt_queue and process_queue even if the latter could end-up | |
5731 | * containing a lot of packets. | |
5732 | */ | |
5733 | return true; | |
5734 | } | |
5735 | ||
41852497 | 5736 | static void flush_all_backlogs(void) |
145dd5f9 | 5737 | { |
2de79ee2 | 5738 | static cpumask_t flush_cpus; |
145dd5f9 PA |
5739 | unsigned int cpu; |
5740 | ||
2de79ee2 PA |
5741 | /* since we are under rtnl lock protection we can use static data |
5742 | * for the cpumask and avoid allocating on stack the possibly | |
5743 | * large mask | |
5744 | */ | |
5745 | ASSERT_RTNL(); | |
5746 | ||
145dd5f9 PA |
5747 | get_online_cpus(); |
5748 | ||
2de79ee2 PA |
5749 | cpumask_clear(&flush_cpus); |
5750 | for_each_online_cpu(cpu) { | |
5751 | if (flush_required(cpu)) { | |
5752 | queue_work_on(cpu, system_highpri_wq, | |
5753 | per_cpu_ptr(&flush_works, cpu)); | |
5754 | cpumask_set_cpu(cpu, &flush_cpus); | |
5755 | } | |
5756 | } | |
145dd5f9 | 5757 | |
2de79ee2 | 5758 | /* we can have in flight packet[s] on the cpus we are not flushing, |
0cbe1e57 | 5759 | * synchronize_net() in unregister_netdevice_many() will take care of |
2de79ee2 PA |
5760 | * them |
5761 | */ | |
5762 | for_each_cpu(cpu, &flush_cpus) | |
41852497 | 5763 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 PA |
5764 | |
5765 | put_online_cpus(); | |
6e583ce5 SH |
5766 | } |
5767 | ||
c8079432 MM |
5768 | /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ |
5769 | static void gro_normal_list(struct napi_struct *napi) | |
5770 | { | |
5771 | if (!napi->rx_count) | |
5772 | return; | |
5773 | netif_receive_skb_list_internal(&napi->rx_list); | |
5774 | INIT_LIST_HEAD(&napi->rx_list); | |
5775 | napi->rx_count = 0; | |
5776 | } | |
5777 | ||
5778 | /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, | |
5779 | * pass the whole batch up to the stack. | |
5780 | */ | |
8dc1c444 | 5781 | static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs) |
c8079432 MM |
5782 | { |
5783 | list_add_tail(&skb->list, &napi->rx_list); | |
8dc1c444 ED |
5784 | napi->rx_count += segs; |
5785 | if (napi->rx_count >= gro_normal_batch) | |
c8079432 MM |
5786 | gro_normal_list(napi); |
5787 | } | |
5788 | ||
c8079432 | 5789 | static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 | 5790 | { |
22061d80 | 5791 | struct packet_offload *ptype; |
d565b0a1 | 5792 | __be16 type = skb->protocol; |
22061d80 | 5793 | struct list_head *head = &offload_base; |
d565b0a1 HX |
5794 | int err = -ENOENT; |
5795 | ||
c3c7c254 ED |
5796 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
5797 | ||
fc59f9a3 HX |
5798 | if (NAPI_GRO_CB(skb)->count == 1) { |
5799 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 5800 | goto out; |
fc59f9a3 | 5801 | } |
d565b0a1 HX |
5802 | |
5803 | rcu_read_lock(); | |
5804 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 5805 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
5806 | continue; |
5807 | ||
aaa5d90b PA |
5808 | err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, |
5809 | ipv6_gro_complete, inet_gro_complete, | |
5810 | skb, 0); | |
d565b0a1 HX |
5811 | break; |
5812 | } | |
5813 | rcu_read_unlock(); | |
5814 | ||
5815 | if (err) { | |
5816 | WARN_ON(&ptype->list == head); | |
5817 | kfree_skb(skb); | |
5818 | return NET_RX_SUCCESS; | |
5819 | } | |
5820 | ||
5821 | out: | |
8dc1c444 | 5822 | gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count); |
c8079432 | 5823 | return NET_RX_SUCCESS; |
d565b0a1 HX |
5824 | } |
5825 | ||
6312fe77 | 5826 | static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, |
07d78363 | 5827 | bool flush_old) |
d565b0a1 | 5828 | { |
6312fe77 | 5829 | struct list_head *head = &napi->gro_hash[index].list; |
d4546c25 | 5830 | struct sk_buff *skb, *p; |
2e71a6f8 | 5831 | |
07d78363 | 5832 | list_for_each_entry_safe_reverse(skb, p, head, list) { |
2e71a6f8 ED |
5833 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) |
5834 | return; | |
992cba7e | 5835 | skb_list_del_init(skb); |
c8079432 | 5836 | napi_gro_complete(napi, skb); |
6312fe77 | 5837 | napi->gro_hash[index].count--; |
d565b0a1 | 5838 | } |
d9f37d01 LR |
5839 | |
5840 | if (!napi->gro_hash[index].count) | |
5841 | __clear_bit(index, &napi->gro_bitmask); | |
d565b0a1 | 5842 | } |
07d78363 | 5843 | |
6312fe77 | 5844 | /* napi->gro_hash[].list contains packets ordered by age. |
07d78363 DM |
5845 | * youngest packets at the head of it. |
5846 | * Complete skbs in reverse order to reduce latencies. | |
5847 | */ | |
5848 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
5849 | { | |
42519ede ED |
5850 | unsigned long bitmask = napi->gro_bitmask; |
5851 | unsigned int i, base = ~0U; | |
07d78363 | 5852 | |
42519ede ED |
5853 | while ((i = ffs(bitmask)) != 0) { |
5854 | bitmask >>= i; | |
5855 | base += i; | |
5856 | __napi_gro_flush_chain(napi, base, flush_old); | |
d9f37d01 | 5857 | } |
07d78363 | 5858 | } |
86cac58b | 5859 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 5860 | |
07d78363 DM |
5861 | static struct list_head *gro_list_prepare(struct napi_struct *napi, |
5862 | struct sk_buff *skb) | |
89c5fa33 | 5863 | { |
89c5fa33 | 5864 | unsigned int maclen = skb->dev->hard_header_len; |
0b4cec8c | 5865 | u32 hash = skb_get_hash_raw(skb); |
07d78363 | 5866 | struct list_head *head; |
d4546c25 | 5867 | struct sk_buff *p; |
89c5fa33 | 5868 | |
6312fe77 | 5869 | head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list; |
07d78363 | 5870 | list_for_each_entry(p, head, list) { |
89c5fa33 ED |
5871 | unsigned long diffs; |
5872 | ||
0b4cec8c TH |
5873 | NAPI_GRO_CB(p)->flush = 0; |
5874 | ||
5875 | if (hash != skb_get_hash_raw(p)) { | |
5876 | NAPI_GRO_CB(p)->same_flow = 0; | |
5877 | continue; | |
5878 | } | |
5879 | ||
89c5fa33 | 5880 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
b1817524 MM |
5881 | diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); |
5882 | if (skb_vlan_tag_present(p)) | |
fc5141cb | 5883 | diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); |
ce87fc6c | 5884 | diffs |= skb_metadata_dst_cmp(p, skb); |
de8f3a83 | 5885 | diffs |= skb_metadata_differs(p, skb); |
89c5fa33 ED |
5886 | if (maclen == ETH_HLEN) |
5887 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 5888 | skb_mac_header(skb)); |
89c5fa33 ED |
5889 | else if (!diffs) |
5890 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 5891 | skb_mac_header(skb), |
89c5fa33 ED |
5892 | maclen); |
5893 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 | 5894 | } |
07d78363 DM |
5895 | |
5896 | return head; | |
89c5fa33 ED |
5897 | } |
5898 | ||
299603e8 JC |
5899 | static void skb_gro_reset_offset(struct sk_buff *skb) |
5900 | { | |
5901 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
5902 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
5903 | ||
5904 | NAPI_GRO_CB(skb)->data_offset = 0; | |
5905 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
5906 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
5907 | ||
8aef998d | 5908 | if (!skb_headlen(skb) && pinfo->nr_frags && |
299603e8 JC |
5909 | !PageHighMem(skb_frag_page(frag0))) { |
5910 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
7cfd5fd5 ED |
5911 | NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, |
5912 | skb_frag_size(frag0), | |
5913 | skb->end - skb->tail); | |
89c5fa33 ED |
5914 | } |
5915 | } | |
5916 | ||
a50e233c ED |
5917 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
5918 | { | |
5919 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
5920 | ||
5921 | BUG_ON(skb->end - skb->tail < grow); | |
5922 | ||
5923 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
5924 | ||
5925 | skb->data_len -= grow; | |
5926 | skb->tail += grow; | |
5927 | ||
b54c9d5b | 5928 | skb_frag_off_add(&pinfo->frags[0], grow); |
a50e233c ED |
5929 | skb_frag_size_sub(&pinfo->frags[0], grow); |
5930 | ||
5931 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
5932 | skb_frag_unref(skb, 0); | |
5933 | memmove(pinfo->frags, pinfo->frags + 1, | |
5934 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
5935 | } | |
5936 | } | |
5937 | ||
c8079432 | 5938 | static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) |
07d78363 | 5939 | { |
6312fe77 | 5940 | struct sk_buff *oldest; |
07d78363 | 5941 | |
6312fe77 | 5942 | oldest = list_last_entry(head, struct sk_buff, list); |
07d78363 | 5943 | |
6312fe77 | 5944 | /* We are called with head length >= MAX_GRO_SKBS, so this is |
07d78363 DM |
5945 | * impossible. |
5946 | */ | |
5947 | if (WARN_ON_ONCE(!oldest)) | |
5948 | return; | |
5949 | ||
d9f37d01 LR |
5950 | /* Do not adjust napi->gro_hash[].count, caller is adding a new |
5951 | * SKB to the chain. | |
07d78363 | 5952 | */ |
ece23711 | 5953 | skb_list_del_init(oldest); |
c8079432 | 5954 | napi_gro_complete(napi, oldest); |
07d78363 DM |
5955 | } |
5956 | ||
bb728820 | 5957 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 | 5958 | { |
6312fe77 | 5959 | u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); |
d4546c25 | 5960 | struct list_head *head = &offload_base; |
22061d80 | 5961 | struct packet_offload *ptype; |
d565b0a1 | 5962 | __be16 type = skb->protocol; |
07d78363 | 5963 | struct list_head *gro_head; |
d4546c25 | 5964 | struct sk_buff *pp = NULL; |
5b252f0c | 5965 | enum gro_result ret; |
d4546c25 | 5966 | int same_flow; |
a50e233c | 5967 | int grow; |
d565b0a1 | 5968 | |
b5cdae32 | 5969 | if (netif_elide_gro(skb->dev)) |
d565b0a1 HX |
5970 | goto normal; |
5971 | ||
07d78363 | 5972 | gro_head = gro_list_prepare(napi, skb); |
89c5fa33 | 5973 | |
d565b0a1 HX |
5974 | rcu_read_lock(); |
5975 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 5976 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
5977 | continue; |
5978 | ||
86911732 | 5979 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 5980 | skb_reset_mac_len(skb); |
d565b0a1 | 5981 | NAPI_GRO_CB(skb)->same_flow = 0; |
d61d072e | 5982 | NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); |
5d38a079 | 5983 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 5984 | NAPI_GRO_CB(skb)->encap_mark = 0; |
fcd91dd4 | 5985 | NAPI_GRO_CB(skb)->recursion_counter = 0; |
a0ca153f | 5986 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 5987 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 5988 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 5989 | |
662880f4 TH |
5990 | /* Setup for GRO checksum validation */ |
5991 | switch (skb->ip_summed) { | |
5992 | case CHECKSUM_COMPLETE: | |
5993 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
5994 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
5995 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
5996 | break; | |
5997 | case CHECKSUM_UNNECESSARY: | |
5998 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
5999 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
6000 | break; | |
6001 | default: | |
6002 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
6003 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
6004 | } | |
d565b0a1 | 6005 | |
aaa5d90b PA |
6006 | pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, |
6007 | ipv6_gro_receive, inet_gro_receive, | |
6008 | gro_head, skb); | |
d565b0a1 HX |
6009 | break; |
6010 | } | |
6011 | rcu_read_unlock(); | |
6012 | ||
6013 | if (&ptype->list == head) | |
6014 | goto normal; | |
6015 | ||
45586c70 | 6016 | if (PTR_ERR(pp) == -EINPROGRESS) { |
25393d3f SK |
6017 | ret = GRO_CONSUMED; |
6018 | goto ok; | |
6019 | } | |
6020 | ||
0da2afd5 | 6021 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 6022 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 6023 | |
d565b0a1 | 6024 | if (pp) { |
992cba7e | 6025 | skb_list_del_init(pp); |
c8079432 | 6026 | napi_gro_complete(napi, pp); |
6312fe77 | 6027 | napi->gro_hash[hash].count--; |
d565b0a1 HX |
6028 | } |
6029 | ||
0da2afd5 | 6030 | if (same_flow) |
d565b0a1 HX |
6031 | goto ok; |
6032 | ||
600adc18 | 6033 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 6034 | goto normal; |
d565b0a1 | 6035 | |
6312fe77 | 6036 | if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) { |
c8079432 | 6037 | gro_flush_oldest(napi, gro_head); |
600adc18 | 6038 | } else { |
6312fe77 | 6039 | napi->gro_hash[hash].count++; |
600adc18 | 6040 | } |
d565b0a1 | 6041 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 6042 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 6043 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 6044 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
07d78363 | 6045 | list_add(&skb->list, gro_head); |
5d0d9be8 | 6046 | ret = GRO_HELD; |
d565b0a1 | 6047 | |
ad0f9904 | 6048 | pull: |
a50e233c ED |
6049 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
6050 | if (grow > 0) | |
6051 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 6052 | ok: |
d9f37d01 LR |
6053 | if (napi->gro_hash[hash].count) { |
6054 | if (!test_bit(hash, &napi->gro_bitmask)) | |
6055 | __set_bit(hash, &napi->gro_bitmask); | |
6056 | } else if (test_bit(hash, &napi->gro_bitmask)) { | |
6057 | __clear_bit(hash, &napi->gro_bitmask); | |
6058 | } | |
6059 | ||
5d0d9be8 | 6060 | return ret; |
d565b0a1 HX |
6061 | |
6062 | normal: | |
ad0f9904 HX |
6063 | ret = GRO_NORMAL; |
6064 | goto pull; | |
5d38a079 | 6065 | } |
96e93eab | 6066 | |
bf5a755f JC |
6067 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
6068 | { | |
6069 | struct list_head *offload_head = &offload_base; | |
6070 | struct packet_offload *ptype; | |
6071 | ||
6072 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
6073 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
6074 | continue; | |
6075 | return ptype; | |
6076 | } | |
6077 | return NULL; | |
6078 | } | |
e27a2f83 | 6079 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
6080 | |
6081 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
6082 | { | |
6083 | struct list_head *offload_head = &offload_base; | |
6084 | struct packet_offload *ptype; | |
6085 | ||
6086 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
6087 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
6088 | continue; | |
6089 | return ptype; | |
6090 | } | |
6091 | return NULL; | |
6092 | } | |
e27a2f83 | 6093 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 6094 | |
6570bc79 AL |
6095 | static gro_result_t napi_skb_finish(struct napi_struct *napi, |
6096 | struct sk_buff *skb, | |
6097 | gro_result_t ret) | |
5d38a079 | 6098 | { |
5d0d9be8 HX |
6099 | switch (ret) { |
6100 | case GRO_NORMAL: | |
8dc1c444 | 6101 | gro_normal_one(napi, skb, 1); |
c7c4b3b6 | 6102 | break; |
5d38a079 | 6103 | |
daa86548 | 6104 | case GRO_MERGED_FREE: |
e44699d2 MK |
6105 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
6106 | napi_skb_free_stolen_head(skb); | |
6107 | else | |
9243adfc | 6108 | __kfree_skb_defer(skb); |
daa86548 ED |
6109 | break; |
6110 | ||
5b252f0c BH |
6111 | case GRO_HELD: |
6112 | case GRO_MERGED: | |
25393d3f | 6113 | case GRO_CONSUMED: |
5b252f0c | 6114 | break; |
5d38a079 HX |
6115 | } |
6116 | ||
c7c4b3b6 | 6117 | return ret; |
5d0d9be8 | 6118 | } |
5d0d9be8 | 6119 | |
c7c4b3b6 | 6120 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 6121 | { |
b0e3f1bd GB |
6122 | gro_result_t ret; |
6123 | ||
93f93a44 | 6124 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 6125 | trace_napi_gro_receive_entry(skb); |
86911732 | 6126 | |
a50e233c ED |
6127 | skb_gro_reset_offset(skb); |
6128 | ||
6570bc79 | 6129 | ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); |
b0e3f1bd GB |
6130 | trace_napi_gro_receive_exit(ret); |
6131 | ||
6132 | return ret; | |
d565b0a1 HX |
6133 | } |
6134 | EXPORT_SYMBOL(napi_gro_receive); | |
6135 | ||
d0c2b0d2 | 6136 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 6137 | { |
93a35f59 ED |
6138 | if (unlikely(skb->pfmemalloc)) { |
6139 | consume_skb(skb); | |
6140 | return; | |
6141 | } | |
96e93eab | 6142 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
6143 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
6144 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
b1817524 | 6145 | __vlan_hwaccel_clear_tag(skb); |
66c46d74 | 6146 | skb->dev = napi->dev; |
6d152e23 | 6147 | skb->skb_iif = 0; |
33d9a2c7 ED |
6148 | |
6149 | /* eth_type_trans() assumes pkt_type is PACKET_HOST */ | |
6150 | skb->pkt_type = PACKET_HOST; | |
6151 | ||
c3caf119 JC |
6152 | skb->encapsulation = 0; |
6153 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 6154 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
174e2381 | 6155 | skb_ext_reset(skb); |
96e93eab HX |
6156 | |
6157 | napi->skb = skb; | |
6158 | } | |
96e93eab | 6159 | |
76620aaf | 6160 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 6161 | { |
5d38a079 | 6162 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
6163 | |
6164 | if (!skb) { | |
fd11a83d | 6165 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
6166 | if (skb) { |
6167 | napi->skb = skb; | |
6168 | skb_mark_napi_id(skb, napi); | |
6169 | } | |
80595d59 | 6170 | } |
96e93eab HX |
6171 | return skb; |
6172 | } | |
76620aaf | 6173 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 6174 | |
a50e233c ED |
6175 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
6176 | struct sk_buff *skb, | |
6177 | gro_result_t ret) | |
96e93eab | 6178 | { |
5d0d9be8 HX |
6179 | switch (ret) { |
6180 | case GRO_NORMAL: | |
a50e233c ED |
6181 | case GRO_HELD: |
6182 | __skb_push(skb, ETH_HLEN); | |
6183 | skb->protocol = eth_type_trans(skb, skb->dev); | |
323ebb61 | 6184 | if (ret == GRO_NORMAL) |
8dc1c444 | 6185 | gro_normal_one(napi, skb, 1); |
86911732 | 6186 | break; |
5d38a079 | 6187 | |
e44699d2 MK |
6188 | case GRO_MERGED_FREE: |
6189 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) | |
6190 | napi_skb_free_stolen_head(skb); | |
6191 | else | |
6192 | napi_reuse_skb(napi, skb); | |
6193 | break; | |
6194 | ||
5b252f0c | 6195 | case GRO_MERGED: |
25393d3f | 6196 | case GRO_CONSUMED: |
5b252f0c | 6197 | break; |
5d0d9be8 | 6198 | } |
5d38a079 | 6199 | |
c7c4b3b6 | 6200 | return ret; |
5d38a079 | 6201 | } |
5d0d9be8 | 6202 | |
a50e233c ED |
6203 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
6204 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
6205 | * We copy ethernet header into skb->data to have a common layout. | |
6206 | */ | |
4adb9c4a | 6207 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
6208 | { |
6209 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
6210 | const struct ethhdr *eth; |
6211 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
6212 | |
6213 | napi->skb = NULL; | |
6214 | ||
a50e233c ED |
6215 | skb_reset_mac_header(skb); |
6216 | skb_gro_reset_offset(skb); | |
6217 | ||
a50e233c ED |
6218 | if (unlikely(skb_gro_header_hard(skb, hlen))) { |
6219 | eth = skb_gro_header_slow(skb, hlen, 0); | |
6220 | if (unlikely(!eth)) { | |
4da46ceb AC |
6221 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
6222 | __func__, napi->dev->name); | |
a50e233c ED |
6223 | napi_reuse_skb(napi, skb); |
6224 | return NULL; | |
6225 | } | |
6226 | } else { | |
a4270d67 | 6227 | eth = (const struct ethhdr *)skb->data; |
a50e233c ED |
6228 | gro_pull_from_frag0(skb, hlen); |
6229 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
6230 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 6231 | } |
a50e233c ED |
6232 | __skb_pull(skb, hlen); |
6233 | ||
6234 | /* | |
6235 | * This works because the only protocols we care about don't require | |
6236 | * special handling. | |
6237 | * We'll fix it up properly in napi_frags_finish() | |
6238 | */ | |
6239 | skb->protocol = eth->h_proto; | |
76620aaf | 6240 | |
76620aaf HX |
6241 | return skb; |
6242 | } | |
76620aaf | 6243 | |
c7c4b3b6 | 6244 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 6245 | { |
b0e3f1bd | 6246 | gro_result_t ret; |
76620aaf | 6247 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 | 6248 | |
ae78dbfa BH |
6249 | trace_napi_gro_frags_entry(skb); |
6250 | ||
b0e3f1bd GB |
6251 | ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
6252 | trace_napi_gro_frags_exit(ret); | |
6253 | ||
6254 | return ret; | |
5d0d9be8 | 6255 | } |
5d38a079 HX |
6256 | EXPORT_SYMBOL(napi_gro_frags); |
6257 | ||
573e8fca TH |
6258 | /* Compute the checksum from gro_offset and return the folded value |
6259 | * after adding in any pseudo checksum. | |
6260 | */ | |
6261 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
6262 | { | |
6263 | __wsum wsum; | |
6264 | __sum16 sum; | |
6265 | ||
6266 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
6267 | ||
6268 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
6269 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
14641931 | 6270 | /* See comments in __skb_checksum_complete(). */ |
573e8fca TH |
6271 | if (likely(!sum)) { |
6272 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
6273 | !skb->csum_complete_sw) | |
7fe50ac8 | 6274 | netdev_rx_csum_fault(skb->dev, skb); |
573e8fca TH |
6275 | } |
6276 | ||
6277 | NAPI_GRO_CB(skb)->csum = wsum; | |
6278 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
6279 | ||
6280 | return sum; | |
6281 | } | |
6282 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
6283 | ||
773fc8f6 | 6284 | static void net_rps_send_ipi(struct softnet_data *remsd) |
6285 | { | |
6286 | #ifdef CONFIG_RPS | |
6287 | while (remsd) { | |
6288 | struct softnet_data *next = remsd->rps_ipi_next; | |
6289 | ||
6290 | if (cpu_online(remsd->cpu)) | |
6291 | smp_call_function_single_async(remsd->cpu, &remsd->csd); | |
6292 | remsd = next; | |
6293 | } | |
6294 | #endif | |
6295 | } | |
6296 | ||
e326bed2 | 6297 | /* |
855abcf0 | 6298 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
6299 | * Note: called with local irq disabled, but exits with local irq enabled. |
6300 | */ | |
6301 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
6302 | { | |
6303 | #ifdef CONFIG_RPS | |
6304 | struct softnet_data *remsd = sd->rps_ipi_list; | |
6305 | ||
6306 | if (remsd) { | |
6307 | sd->rps_ipi_list = NULL; | |
6308 | ||
6309 | local_irq_enable(); | |
6310 | ||
6311 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
773fc8f6 | 6312 | net_rps_send_ipi(remsd); |
e326bed2 ED |
6313 | } else |
6314 | #endif | |
6315 | local_irq_enable(); | |
6316 | } | |
6317 | ||
d75b1ade ED |
6318 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
6319 | { | |
6320 | #ifdef CONFIG_RPS | |
6321 | return sd->rps_ipi_list != NULL; | |
6322 | #else | |
6323 | return false; | |
6324 | #endif | |
6325 | } | |
6326 | ||
bea3348e | 6327 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 6328 | { |
eecfd7c4 | 6329 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
6330 | bool again = true; |
6331 | int work = 0; | |
1da177e4 | 6332 | |
e326bed2 ED |
6333 | /* Check if we have pending ipi, its better to send them now, |
6334 | * not waiting net_rx_action() end. | |
6335 | */ | |
d75b1ade | 6336 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
6337 | local_irq_disable(); |
6338 | net_rps_action_and_irq_enable(sd); | |
6339 | } | |
d75b1ade | 6340 | |
3d48b53f | 6341 | napi->weight = dev_rx_weight; |
145dd5f9 | 6342 | while (again) { |
1da177e4 | 6343 | struct sk_buff *skb; |
6e7676c1 CG |
6344 | |
6345 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 6346 | rcu_read_lock(); |
6e7676c1 | 6347 | __netif_receive_skb(skb); |
2c17d27c | 6348 | rcu_read_unlock(); |
76cc8b13 | 6349 | input_queue_head_incr(sd); |
145dd5f9 | 6350 | if (++work >= quota) |
76cc8b13 | 6351 | return work; |
145dd5f9 | 6352 | |
6e7676c1 | 6353 | } |
1da177e4 | 6354 | |
145dd5f9 | 6355 | local_irq_disable(); |
e36fa2f7 | 6356 | rps_lock(sd); |
11ef7a89 | 6357 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
6358 | /* |
6359 | * Inline a custom version of __napi_complete(). | |
6360 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
6361 | * and NAPI_STATE_SCHED is the only possible flag set |
6362 | * on backlog. | |
6363 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
6364 | * and we dont need an smp_mb() memory barrier. |
6365 | */ | |
eecfd7c4 | 6366 | napi->state = 0; |
145dd5f9 PA |
6367 | again = false; |
6368 | } else { | |
6369 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
6370 | &sd->process_queue); | |
bea3348e | 6371 | } |
e36fa2f7 | 6372 | rps_unlock(sd); |
145dd5f9 | 6373 | local_irq_enable(); |
6e7676c1 | 6374 | } |
1da177e4 | 6375 | |
bea3348e SH |
6376 | return work; |
6377 | } | |
1da177e4 | 6378 | |
bea3348e SH |
6379 | /** |
6380 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 6381 | * @n: entry to schedule |
bea3348e | 6382 | * |
bc9ad166 ED |
6383 | * The entry's receive function will be scheduled to run. |
6384 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 6385 | */ |
b5606c2d | 6386 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
6387 | { |
6388 | unsigned long flags; | |
1da177e4 | 6389 | |
bea3348e | 6390 | local_irq_save(flags); |
903ceff7 | 6391 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 6392 | local_irq_restore(flags); |
1da177e4 | 6393 | } |
bea3348e SH |
6394 | EXPORT_SYMBOL(__napi_schedule); |
6395 | ||
39e6c820 ED |
6396 | /** |
6397 | * napi_schedule_prep - check if napi can be scheduled | |
6398 | * @n: napi context | |
6399 | * | |
6400 | * Test if NAPI routine is already running, and if not mark | |
ee1a4c84 | 6401 | * it as running. This is used as a condition variable to |
39e6c820 ED |
6402 | * insure only one NAPI poll instance runs. We also make |
6403 | * sure there is no pending NAPI disable. | |
6404 | */ | |
6405 | bool napi_schedule_prep(struct napi_struct *n) | |
6406 | { | |
6407 | unsigned long val, new; | |
6408 | ||
6409 | do { | |
6410 | val = READ_ONCE(n->state); | |
6411 | if (unlikely(val & NAPIF_STATE_DISABLE)) | |
6412 | return false; | |
6413 | new = val | NAPIF_STATE_SCHED; | |
6414 | ||
6415 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
6416 | * This was suggested by Alexander Duyck, as compiler | |
6417 | * emits better code than : | |
6418 | * if (val & NAPIF_STATE_SCHED) | |
6419 | * new |= NAPIF_STATE_MISSED; | |
6420 | */ | |
6421 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
6422 | NAPIF_STATE_MISSED; | |
6423 | } while (cmpxchg(&n->state, val, new) != val); | |
6424 | ||
6425 | return !(val & NAPIF_STATE_SCHED); | |
6426 | } | |
6427 | EXPORT_SYMBOL(napi_schedule_prep); | |
6428 | ||
bc9ad166 ED |
6429 | /** |
6430 | * __napi_schedule_irqoff - schedule for receive | |
6431 | * @n: entry to schedule | |
6432 | * | |
6433 | * Variant of __napi_schedule() assuming hard irqs are masked | |
6434 | */ | |
6435 | void __napi_schedule_irqoff(struct napi_struct *n) | |
6436 | { | |
6437 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
6438 | } | |
6439 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
6440 | ||
364b6055 | 6441 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 6442 | { |
6f8b12d6 ED |
6443 | unsigned long flags, val, new, timeout = 0; |
6444 | bool ret = true; | |
d565b0a1 HX |
6445 | |
6446 | /* | |
217f6974 ED |
6447 | * 1) Don't let napi dequeue from the cpu poll list |
6448 | * just in case its running on a different cpu. | |
6449 | * 2) If we are busy polling, do nothing here, we have | |
6450 | * the guarantee we will be called later. | |
d565b0a1 | 6451 | */ |
217f6974 ED |
6452 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
6453 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 6454 | return false; |
d565b0a1 | 6455 | |
6f8b12d6 ED |
6456 | if (work_done) { |
6457 | if (n->gro_bitmask) | |
7e417a66 ED |
6458 | timeout = READ_ONCE(n->dev->gro_flush_timeout); |
6459 | n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); | |
6f8b12d6 ED |
6460 | } |
6461 | if (n->defer_hard_irqs_count > 0) { | |
6462 | n->defer_hard_irqs_count--; | |
7e417a66 | 6463 | timeout = READ_ONCE(n->dev->gro_flush_timeout); |
6f8b12d6 ED |
6464 | if (timeout) |
6465 | ret = false; | |
6466 | } | |
6467 | if (n->gro_bitmask) { | |
605108ac PA |
6468 | /* When the NAPI instance uses a timeout and keeps postponing |
6469 | * it, we need to bound somehow the time packets are kept in | |
6470 | * the GRO layer | |
6471 | */ | |
6472 | napi_gro_flush(n, !!timeout); | |
3b47d303 | 6473 | } |
c8079432 MM |
6474 | |
6475 | gro_normal_list(n); | |
6476 | ||
02c1602e | 6477 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
6478 | /* If n->poll_list is not empty, we need to mask irqs */ |
6479 | local_irq_save(flags); | |
02c1602e | 6480 | list_del_init(&n->poll_list); |
d75b1ade ED |
6481 | local_irq_restore(flags); |
6482 | } | |
39e6c820 ED |
6483 | |
6484 | do { | |
6485 | val = READ_ONCE(n->state); | |
6486 | ||
6487 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); | |
6488 | ||
7fd3253a BT |
6489 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | |
6490 | NAPIF_STATE_PREFER_BUSY_POLL); | |
39e6c820 ED |
6491 | |
6492 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
6493 | * because we will call napi->poll() one more time. | |
6494 | * This C code was suggested by Alexander Duyck to help gcc. | |
6495 | */ | |
6496 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
6497 | NAPIF_STATE_SCHED; | |
6498 | } while (cmpxchg(&n->state, val, new) != val); | |
6499 | ||
6500 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
6501 | __napi_schedule(n); | |
6502 | return false; | |
6503 | } | |
6504 | ||
6f8b12d6 ED |
6505 | if (timeout) |
6506 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
6507 | HRTIMER_MODE_REL_PINNED); | |
6508 | return ret; | |
d565b0a1 | 6509 | } |
3b47d303 | 6510 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 6511 | |
af12fa6e | 6512 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 6513 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
6514 | { |
6515 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
6516 | struct napi_struct *napi; | |
6517 | ||
6518 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
6519 | if (napi->napi_id == napi_id) | |
6520 | return napi; | |
6521 | ||
6522 | return NULL; | |
6523 | } | |
02d62e86 ED |
6524 | |
6525 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
217f6974 | 6526 | |
7fd3253a | 6527 | static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) |
217f6974 | 6528 | { |
7fd3253a BT |
6529 | if (!skip_schedule) { |
6530 | gro_normal_list(napi); | |
6531 | __napi_schedule(napi); | |
6532 | return; | |
6533 | } | |
217f6974 | 6534 | |
7fd3253a BT |
6535 | if (napi->gro_bitmask) { |
6536 | /* flush too old packets | |
6537 | * If HZ < 1000, flush all packets. | |
6538 | */ | |
6539 | napi_gro_flush(napi, HZ >= 1000); | |
6540 | } | |
217f6974 | 6541 | |
7fd3253a BT |
6542 | gro_normal_list(napi); |
6543 | clear_bit(NAPI_STATE_SCHED, &napi->state); | |
6544 | } | |
6545 | ||
7c951caf BT |
6546 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, |
6547 | u16 budget) | |
217f6974 | 6548 | { |
7fd3253a BT |
6549 | bool skip_schedule = false; |
6550 | unsigned long timeout; | |
217f6974 ED |
6551 | int rc; |
6552 | ||
39e6c820 ED |
6553 | /* Busy polling means there is a high chance device driver hard irq |
6554 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
6555 | * set in napi_schedule_prep(). | |
6556 | * Since we are about to call napi->poll() once more, we can safely | |
6557 | * clear NAPI_STATE_MISSED. | |
6558 | * | |
6559 | * Note: x86 could use a single "lock and ..." instruction | |
6560 | * to perform these two clear_bit() | |
6561 | */ | |
6562 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
6563 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
6564 | ||
6565 | local_bh_disable(); | |
6566 | ||
7fd3253a BT |
6567 | if (prefer_busy_poll) { |
6568 | napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); | |
6569 | timeout = READ_ONCE(napi->dev->gro_flush_timeout); | |
6570 | if (napi->defer_hard_irqs_count && timeout) { | |
6571 | hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); | |
6572 | skip_schedule = true; | |
6573 | } | |
6574 | } | |
6575 | ||
217f6974 ED |
6576 | /* All we really want here is to re-enable device interrupts. |
6577 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
6578 | */ | |
7c951caf | 6579 | rc = napi->poll(napi, budget); |
323ebb61 EC |
6580 | /* We can't gro_normal_list() here, because napi->poll() might have |
6581 | * rearmed the napi (napi_complete_done()) in which case it could | |
6582 | * already be running on another CPU. | |
6583 | */ | |
7c951caf | 6584 | trace_napi_poll(napi, rc, budget); |
217f6974 | 6585 | netpoll_poll_unlock(have_poll_lock); |
7c951caf | 6586 | if (rc == budget) |
7fd3253a | 6587 | __busy_poll_stop(napi, skip_schedule); |
217f6974 | 6588 | local_bh_enable(); |
217f6974 ED |
6589 | } |
6590 | ||
7db6b048 SS |
6591 | void napi_busy_loop(unsigned int napi_id, |
6592 | bool (*loop_end)(void *, unsigned long), | |
7c951caf | 6593 | void *loop_end_arg, bool prefer_busy_poll, u16 budget) |
02d62e86 | 6594 | { |
7db6b048 | 6595 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 6596 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 6597 | void *have_poll_lock = NULL; |
02d62e86 | 6598 | struct napi_struct *napi; |
217f6974 ED |
6599 | |
6600 | restart: | |
217f6974 | 6601 | napi_poll = NULL; |
02d62e86 | 6602 | |
2a028ecb | 6603 | rcu_read_lock(); |
02d62e86 | 6604 | |
545cd5e5 | 6605 | napi = napi_by_id(napi_id); |
02d62e86 ED |
6606 | if (!napi) |
6607 | goto out; | |
6608 | ||
217f6974 ED |
6609 | preempt_disable(); |
6610 | for (;;) { | |
2b5cd0df AD |
6611 | int work = 0; |
6612 | ||
2a028ecb | 6613 | local_bh_disable(); |
217f6974 ED |
6614 | if (!napi_poll) { |
6615 | unsigned long val = READ_ONCE(napi->state); | |
6616 | ||
6617 | /* If multiple threads are competing for this napi, | |
6618 | * we avoid dirtying napi->state as much as we can. | |
6619 | */ | |
6620 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
7fd3253a BT |
6621 | NAPIF_STATE_IN_BUSY_POLL)) { |
6622 | if (prefer_busy_poll) | |
6623 | set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); | |
217f6974 | 6624 | goto count; |
7fd3253a | 6625 | } |
217f6974 ED |
6626 | if (cmpxchg(&napi->state, val, |
6627 | val | NAPIF_STATE_IN_BUSY_POLL | | |
7fd3253a BT |
6628 | NAPIF_STATE_SCHED) != val) { |
6629 | if (prefer_busy_poll) | |
6630 | set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); | |
217f6974 | 6631 | goto count; |
7fd3253a | 6632 | } |
217f6974 ED |
6633 | have_poll_lock = netpoll_poll_lock(napi); |
6634 | napi_poll = napi->poll; | |
6635 | } | |
7c951caf BT |
6636 | work = napi_poll(napi, budget); |
6637 | trace_napi_poll(napi, work, budget); | |
323ebb61 | 6638 | gro_normal_list(napi); |
217f6974 | 6639 | count: |
2b5cd0df | 6640 | if (work > 0) |
7db6b048 | 6641 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 6642 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
2a028ecb | 6643 | local_bh_enable(); |
02d62e86 | 6644 | |
7db6b048 | 6645 | if (!loop_end || loop_end(loop_end_arg, start_time)) |
217f6974 | 6646 | break; |
02d62e86 | 6647 | |
217f6974 ED |
6648 | if (unlikely(need_resched())) { |
6649 | if (napi_poll) | |
7c951caf | 6650 | busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); |
217f6974 ED |
6651 | preempt_enable(); |
6652 | rcu_read_unlock(); | |
6653 | cond_resched(); | |
7db6b048 | 6654 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 6655 | return; |
217f6974 ED |
6656 | goto restart; |
6657 | } | |
6cdf89b1 | 6658 | cpu_relax(); |
217f6974 ED |
6659 | } |
6660 | if (napi_poll) | |
7c951caf | 6661 | busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); |
217f6974 | 6662 | preempt_enable(); |
02d62e86 | 6663 | out: |
2a028ecb | 6664 | rcu_read_unlock(); |
02d62e86 | 6665 | } |
7db6b048 | 6666 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
6667 | |
6668 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 6669 | |
149d6ad8 | 6670 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 6671 | { |
4d092dd2 | 6672 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) |
52bd2d62 | 6673 | return; |
af12fa6e | 6674 | |
52bd2d62 | 6675 | spin_lock(&napi_hash_lock); |
af12fa6e | 6676 | |
545cd5e5 | 6677 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 6678 | do { |
545cd5e5 AD |
6679 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
6680 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
6681 | } while (napi_by_id(napi_gen_id)); |
6682 | napi->napi_id = napi_gen_id; | |
af12fa6e | 6683 | |
52bd2d62 ED |
6684 | hlist_add_head_rcu(&napi->napi_hash_node, |
6685 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 6686 | |
52bd2d62 | 6687 | spin_unlock(&napi_hash_lock); |
af12fa6e | 6688 | } |
af12fa6e ET |
6689 | |
6690 | /* Warning : caller is responsible to make sure rcu grace period | |
6691 | * is respected before freeing memory containing @napi | |
6692 | */ | |
5198d545 | 6693 | static void napi_hash_del(struct napi_struct *napi) |
af12fa6e ET |
6694 | { |
6695 | spin_lock(&napi_hash_lock); | |
6696 | ||
4d092dd2 | 6697 | hlist_del_init_rcu(&napi->napi_hash_node); |
5198d545 | 6698 | |
af12fa6e ET |
6699 | spin_unlock(&napi_hash_lock); |
6700 | } | |
af12fa6e | 6701 | |
3b47d303 ED |
6702 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
6703 | { | |
6704 | struct napi_struct *napi; | |
6705 | ||
6706 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
6707 | |
6708 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
6709 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
6710 | */ | |
6f8b12d6 | 6711 | if (!napi_disable_pending(napi) && |
7fd3253a BT |
6712 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { |
6713 | clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); | |
39e6c820 | 6714 | __napi_schedule_irqoff(napi); |
7fd3253a | 6715 | } |
3b47d303 ED |
6716 | |
6717 | return HRTIMER_NORESTART; | |
6718 | } | |
6719 | ||
7c4ec749 | 6720 | static void init_gro_hash(struct napi_struct *napi) |
d565b0a1 | 6721 | { |
07d78363 DM |
6722 | int i; |
6723 | ||
6312fe77 LR |
6724 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6725 | INIT_LIST_HEAD(&napi->gro_hash[i].list); | |
6726 | napi->gro_hash[i].count = 0; | |
6727 | } | |
7c4ec749 DM |
6728 | napi->gro_bitmask = 0; |
6729 | } | |
6730 | ||
5fdd2f0e WW |
6731 | int dev_set_threaded(struct net_device *dev, bool threaded) |
6732 | { | |
6733 | struct napi_struct *napi; | |
6734 | int err = 0; | |
6735 | ||
6736 | if (dev->threaded == threaded) | |
6737 | return 0; | |
6738 | ||
6739 | if (threaded) { | |
6740 | list_for_each_entry(napi, &dev->napi_list, dev_list) { | |
6741 | if (!napi->thread) { | |
6742 | err = napi_kthread_create(napi); | |
6743 | if (err) { | |
6744 | threaded = false; | |
6745 | break; | |
6746 | } | |
6747 | } | |
6748 | } | |
6749 | } | |
6750 | ||
6751 | dev->threaded = threaded; | |
6752 | ||
6753 | /* Make sure kthread is created before THREADED bit | |
6754 | * is set. | |
6755 | */ | |
6756 | smp_mb__before_atomic(); | |
6757 | ||
6758 | /* Setting/unsetting threaded mode on a napi might not immediately | |
6759 | * take effect, if the current napi instance is actively being | |
6760 | * polled. In this case, the switch between threaded mode and | |
6761 | * softirq mode will happen in the next round of napi_schedule(). | |
6762 | * This should not cause hiccups/stalls to the live traffic. | |
6763 | */ | |
6764 | list_for_each_entry(napi, &dev->napi_list, dev_list) { | |
6765 | if (threaded) | |
6766 | set_bit(NAPI_STATE_THREADED, &napi->state); | |
6767 | else | |
6768 | clear_bit(NAPI_STATE_THREADED, &napi->state); | |
6769 | } | |
6770 | ||
6771 | return err; | |
6772 | } | |
6773 | ||
7c4ec749 DM |
6774 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
6775 | int (*poll)(struct napi_struct *, int), int weight) | |
6776 | { | |
4d092dd2 JK |
6777 | if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) |
6778 | return; | |
6779 | ||
7c4ec749 | 6780 | INIT_LIST_HEAD(&napi->poll_list); |
4d092dd2 | 6781 | INIT_HLIST_NODE(&napi->napi_hash_node); |
7c4ec749 DM |
6782 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
6783 | napi->timer.function = napi_watchdog; | |
6784 | init_gro_hash(napi); | |
5d38a079 | 6785 | napi->skb = NULL; |
323ebb61 EC |
6786 | INIT_LIST_HEAD(&napi->rx_list); |
6787 | napi->rx_count = 0; | |
d565b0a1 | 6788 | napi->poll = poll; |
82dc3c63 | 6789 | if (weight > NAPI_POLL_WEIGHT) |
bf29e9e9 QC |
6790 | netdev_err_once(dev, "%s() called with weight %d\n", __func__, |
6791 | weight); | |
d565b0a1 | 6792 | napi->weight = weight; |
d565b0a1 | 6793 | napi->dev = dev; |
5d38a079 | 6794 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
6795 | napi->poll_owner = -1; |
6796 | #endif | |
6797 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
96e97bc0 JK |
6798 | set_bit(NAPI_STATE_NPSVC, &napi->state); |
6799 | list_add_rcu(&napi->dev_list, &dev->napi_list); | |
93d05d4a | 6800 | napi_hash_add(napi); |
29863d41 WW |
6801 | /* Create kthread for this napi if dev->threaded is set. |
6802 | * Clear dev->threaded if kthread creation failed so that | |
6803 | * threaded mode will not be enabled in napi_enable(). | |
6804 | */ | |
6805 | if (dev->threaded && napi_kthread_create(napi)) | |
6806 | dev->threaded = 0; | |
d565b0a1 HX |
6807 | } |
6808 | EXPORT_SYMBOL(netif_napi_add); | |
6809 | ||
3b47d303 ED |
6810 | void napi_disable(struct napi_struct *n) |
6811 | { | |
6812 | might_sleep(); | |
6813 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
6814 | ||
6815 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
6816 | msleep(1); | |
2d8bff12 NH |
6817 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
6818 | msleep(1); | |
3b47d303 ED |
6819 | |
6820 | hrtimer_cancel(&n->timer); | |
6821 | ||
7fd3253a | 6822 | clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state); |
3b47d303 | 6823 | clear_bit(NAPI_STATE_DISABLE, &n->state); |
29863d41 | 6824 | clear_bit(NAPI_STATE_THREADED, &n->state); |
3b47d303 ED |
6825 | } |
6826 | EXPORT_SYMBOL(napi_disable); | |
6827 | ||
29863d41 WW |
6828 | /** |
6829 | * napi_enable - enable NAPI scheduling | |
6830 | * @n: NAPI context | |
6831 | * | |
6832 | * Resume NAPI from being scheduled on this context. | |
6833 | * Must be paired with napi_disable. | |
6834 | */ | |
6835 | void napi_enable(struct napi_struct *n) | |
6836 | { | |
6837 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
6838 | smp_mb__before_atomic(); | |
6839 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
6840 | clear_bit(NAPI_STATE_NPSVC, &n->state); | |
6841 | if (n->dev->threaded && n->thread) | |
6842 | set_bit(NAPI_STATE_THREADED, &n->state); | |
6843 | } | |
6844 | EXPORT_SYMBOL(napi_enable); | |
6845 | ||
07d78363 | 6846 | static void flush_gro_hash(struct napi_struct *napi) |
d4546c25 | 6847 | { |
07d78363 | 6848 | int i; |
d4546c25 | 6849 | |
07d78363 DM |
6850 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6851 | struct sk_buff *skb, *n; | |
6852 | ||
6312fe77 | 6853 | list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) |
07d78363 | 6854 | kfree_skb(skb); |
6312fe77 | 6855 | napi->gro_hash[i].count = 0; |
07d78363 | 6856 | } |
d4546c25 DM |
6857 | } |
6858 | ||
93d05d4a | 6859 | /* Must be called in process context */ |
5198d545 | 6860 | void __netif_napi_del(struct napi_struct *napi) |
d565b0a1 | 6861 | { |
4d092dd2 JK |
6862 | if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) |
6863 | return; | |
6864 | ||
5198d545 | 6865 | napi_hash_del(napi); |
5251ef82 | 6866 | list_del_rcu(&napi->dev_list); |
76620aaf | 6867 | napi_free_frags(napi); |
d565b0a1 | 6868 | |
07d78363 | 6869 | flush_gro_hash(napi); |
d9f37d01 | 6870 | napi->gro_bitmask = 0; |
29863d41 WW |
6871 | |
6872 | if (napi->thread) { | |
6873 | kthread_stop(napi->thread); | |
6874 | napi->thread = NULL; | |
6875 | } | |
d565b0a1 | 6876 | } |
5198d545 | 6877 | EXPORT_SYMBOL(__netif_napi_del); |
d565b0a1 | 6878 | |
898f8015 | 6879 | static int __napi_poll(struct napi_struct *n, bool *repoll) |
726ce70e | 6880 | { |
726ce70e HX |
6881 | int work, weight; |
6882 | ||
726ce70e HX |
6883 | weight = n->weight; |
6884 | ||
6885 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
6886 | * with netpoll's poll_napi(). Only the entity which | |
6887 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
6888 | * actually make the ->poll() call. Therefore we avoid | |
6889 | * accidentally calling ->poll() when NAPI is not scheduled. | |
6890 | */ | |
6891 | work = 0; | |
6892 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
6893 | work = n->poll(n, weight); | |
1db19db7 | 6894 | trace_napi_poll(n, work, weight); |
726ce70e HX |
6895 | } |
6896 | ||
427d5838 ED |
6897 | if (unlikely(work > weight)) |
6898 | pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n", | |
6899 | n->poll, work, weight); | |
726ce70e HX |
6900 | |
6901 | if (likely(work < weight)) | |
898f8015 | 6902 | return work; |
726ce70e HX |
6903 | |
6904 | /* Drivers must not modify the NAPI state if they | |
6905 | * consume the entire weight. In such cases this code | |
6906 | * still "owns" the NAPI instance and therefore can | |
6907 | * move the instance around on the list at-will. | |
6908 | */ | |
6909 | if (unlikely(napi_disable_pending(n))) { | |
6910 | napi_complete(n); | |
898f8015 | 6911 | return work; |
726ce70e HX |
6912 | } |
6913 | ||
7fd3253a BT |
6914 | /* The NAPI context has more processing work, but busy-polling |
6915 | * is preferred. Exit early. | |
6916 | */ | |
6917 | if (napi_prefer_busy_poll(n)) { | |
6918 | if (napi_complete_done(n, work)) { | |
6919 | /* If timeout is not set, we need to make sure | |
6920 | * that the NAPI is re-scheduled. | |
6921 | */ | |
6922 | napi_schedule(n); | |
6923 | } | |
898f8015 | 6924 | return work; |
7fd3253a BT |
6925 | } |
6926 | ||
d9f37d01 | 6927 | if (n->gro_bitmask) { |
726ce70e HX |
6928 | /* flush too old packets |
6929 | * If HZ < 1000, flush all packets. | |
6930 | */ | |
6931 | napi_gro_flush(n, HZ >= 1000); | |
6932 | } | |
6933 | ||
c8079432 MM |
6934 | gro_normal_list(n); |
6935 | ||
001ce546 HX |
6936 | /* Some drivers may have called napi_schedule |
6937 | * prior to exhausting their budget. | |
6938 | */ | |
6939 | if (unlikely(!list_empty(&n->poll_list))) { | |
6940 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
6941 | n->dev ? n->dev->name : "backlog"); | |
898f8015 | 6942 | return work; |
001ce546 HX |
6943 | } |
6944 | ||
898f8015 FF |
6945 | *repoll = true; |
6946 | ||
6947 | return work; | |
6948 | } | |
6949 | ||
6950 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) | |
6951 | { | |
6952 | bool do_repoll = false; | |
6953 | void *have; | |
6954 | int work; | |
6955 | ||
6956 | list_del_init(&n->poll_list); | |
6957 | ||
6958 | have = netpoll_poll_lock(n); | |
6959 | ||
6960 | work = __napi_poll(n, &do_repoll); | |
6961 | ||
6962 | if (do_repoll) | |
6963 | list_add_tail(&n->poll_list, repoll); | |
726ce70e | 6964 | |
726ce70e HX |
6965 | netpoll_poll_unlock(have); |
6966 | ||
6967 | return work; | |
6968 | } | |
6969 | ||
29863d41 WW |
6970 | static int napi_thread_wait(struct napi_struct *napi) |
6971 | { | |
6972 | set_current_state(TASK_INTERRUPTIBLE); | |
6973 | ||
6974 | while (!kthread_should_stop() && !napi_disable_pending(napi)) { | |
6975 | if (test_bit(NAPI_STATE_SCHED, &napi->state)) { | |
6976 | WARN_ON(!list_empty(&napi->poll_list)); | |
6977 | __set_current_state(TASK_RUNNING); | |
6978 | return 0; | |
6979 | } | |
6980 | ||
6981 | schedule(); | |
6982 | set_current_state(TASK_INTERRUPTIBLE); | |
6983 | } | |
6984 | __set_current_state(TASK_RUNNING); | |
6985 | return -1; | |
6986 | } | |
6987 | ||
6988 | static int napi_threaded_poll(void *data) | |
6989 | { | |
6990 | struct napi_struct *napi = data; | |
6991 | void *have; | |
6992 | ||
6993 | while (!napi_thread_wait(napi)) { | |
6994 | for (;;) { | |
6995 | bool repoll = false; | |
6996 | ||
6997 | local_bh_disable(); | |
6998 | ||
6999 | have = netpoll_poll_lock(napi); | |
7000 | __napi_poll(napi, &repoll); | |
7001 | netpoll_poll_unlock(have); | |
7002 | ||
29863d41 WW |
7003 | local_bh_enable(); |
7004 | ||
7005 | if (!repoll) | |
7006 | break; | |
7007 | ||
7008 | cond_resched(); | |
7009 | } | |
7010 | } | |
7011 | return 0; | |
7012 | } | |
7013 | ||
0766f788 | 7014 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 7015 | { |
903ceff7 | 7016 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e MW |
7017 | unsigned long time_limit = jiffies + |
7018 | usecs_to_jiffies(netdev_budget_usecs); | |
51b0bded | 7019 | int budget = netdev_budget; |
d75b1ade ED |
7020 | LIST_HEAD(list); |
7021 | LIST_HEAD(repoll); | |
53fb95d3 | 7022 | |
1da177e4 | 7023 | local_irq_disable(); |
d75b1ade ED |
7024 | list_splice_init(&sd->poll_list, &list); |
7025 | local_irq_enable(); | |
1da177e4 | 7026 | |
ceb8d5bf | 7027 | for (;;) { |
bea3348e | 7028 | struct napi_struct *n; |
1da177e4 | 7029 | |
ceb8d5bf HX |
7030 | if (list_empty(&list)) { |
7031 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
fec6e49b | 7032 | return; |
ceb8d5bf HX |
7033 | break; |
7034 | } | |
7035 | ||
6bd373eb HX |
7036 | n = list_first_entry(&list, struct napi_struct, poll_list); |
7037 | budget -= napi_poll(n, &repoll); | |
7038 | ||
d75b1ade | 7039 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
7040 | * Allow this to run for 2 jiffies since which will allow |
7041 | * an average latency of 1.5/HZ. | |
bea3348e | 7042 | */ |
ceb8d5bf HX |
7043 | if (unlikely(budget <= 0 || |
7044 | time_after_eq(jiffies, time_limit))) { | |
7045 | sd->time_squeeze++; | |
7046 | break; | |
7047 | } | |
1da177e4 | 7048 | } |
d75b1ade | 7049 | |
d75b1ade ED |
7050 | local_irq_disable(); |
7051 | ||
7052 | list_splice_tail_init(&sd->poll_list, &list); | |
7053 | list_splice_tail(&repoll, &list); | |
7054 | list_splice(&list, &sd->poll_list); | |
7055 | if (!list_empty(&sd->poll_list)) | |
7056 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
7057 | ||
e326bed2 | 7058 | net_rps_action_and_irq_enable(sd); |
1da177e4 LT |
7059 | } |
7060 | ||
aa9d8560 | 7061 | struct netdev_adjacent { |
9ff162a8 | 7062 | struct net_device *dev; |
5d261913 VF |
7063 | |
7064 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 7065 | bool master; |
5d261913 | 7066 | |
32b6d34f TY |
7067 | /* lookup ignore flag */ |
7068 | bool ignore; | |
7069 | ||
5d261913 VF |
7070 | /* counter for the number of times this device was added to us */ |
7071 | u16 ref_nr; | |
7072 | ||
402dae96 VF |
7073 | /* private field for the users */ |
7074 | void *private; | |
7075 | ||
9ff162a8 JP |
7076 | struct list_head list; |
7077 | struct rcu_head rcu; | |
9ff162a8 JP |
7078 | }; |
7079 | ||
6ea29da1 | 7080 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 7081 | struct list_head *adj_list) |
9ff162a8 | 7082 | { |
5d261913 | 7083 | struct netdev_adjacent *adj; |
5d261913 | 7084 | |
2f268f12 | 7085 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
7086 | if (adj->dev == adj_dev) |
7087 | return adj; | |
9ff162a8 JP |
7088 | } |
7089 | return NULL; | |
7090 | } | |
7091 | ||
eff74233 TY |
7092 | static int ____netdev_has_upper_dev(struct net_device *upper_dev, |
7093 | struct netdev_nested_priv *priv) | |
f1170fd4 | 7094 | { |
eff74233 | 7095 | struct net_device *dev = (struct net_device *)priv->data; |
f1170fd4 DA |
7096 | |
7097 | return upper_dev == dev; | |
7098 | } | |
7099 | ||
9ff162a8 JP |
7100 | /** |
7101 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
7102 | * @dev: device | |
7103 | * @upper_dev: upper device to check | |
7104 | * | |
7105 | * Find out if a device is linked to specified upper device and return true | |
7106 | * in case it is. Note that this checks only immediate upper device, | |
7107 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
7108 | */ | |
7109 | bool netdev_has_upper_dev(struct net_device *dev, | |
7110 | struct net_device *upper_dev) | |
7111 | { | |
eff74233 TY |
7112 | struct netdev_nested_priv priv = { |
7113 | .data = (void *)upper_dev, | |
7114 | }; | |
7115 | ||
9ff162a8 JP |
7116 | ASSERT_RTNL(); |
7117 | ||
32b6d34f | 7118 | return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, |
eff74233 | 7119 | &priv); |
9ff162a8 JP |
7120 | } |
7121 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
7122 | ||
1a3f060c | 7123 | /** |
c1639be9 | 7124 | * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device |
1a3f060c DA |
7125 | * @dev: device |
7126 | * @upper_dev: upper device to check | |
7127 | * | |
7128 | * Find out if a device is linked to specified upper device and return true | |
7129 | * in case it is. Note that this checks the entire upper device chain. | |
7130 | * The caller must hold rcu lock. | |
7131 | */ | |
7132 | ||
1a3f060c DA |
7133 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
7134 | struct net_device *upper_dev) | |
7135 | { | |
eff74233 TY |
7136 | struct netdev_nested_priv priv = { |
7137 | .data = (void *)upper_dev, | |
7138 | }; | |
7139 | ||
32b6d34f | 7140 | return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, |
eff74233 | 7141 | &priv); |
1a3f060c DA |
7142 | } |
7143 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
7144 | ||
9ff162a8 JP |
7145 | /** |
7146 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
7147 | * @dev: device | |
7148 | * | |
7149 | * Find out if a device is linked to an upper device and return true in case | |
7150 | * it is. The caller must hold the RTNL lock. | |
7151 | */ | |
25cc72a3 | 7152 | bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
7153 | { |
7154 | ASSERT_RTNL(); | |
7155 | ||
f1170fd4 | 7156 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 7157 | } |
25cc72a3 | 7158 | EXPORT_SYMBOL(netdev_has_any_upper_dev); |
9ff162a8 JP |
7159 | |
7160 | /** | |
7161 | * netdev_master_upper_dev_get - Get master upper device | |
7162 | * @dev: device | |
7163 | * | |
7164 | * Find a master upper device and return pointer to it or NULL in case | |
7165 | * it's not there. The caller must hold the RTNL lock. | |
7166 | */ | |
7167 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
7168 | { | |
aa9d8560 | 7169 | struct netdev_adjacent *upper; |
9ff162a8 JP |
7170 | |
7171 | ASSERT_RTNL(); | |
7172 | ||
2f268f12 | 7173 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
7174 | return NULL; |
7175 | ||
2f268f12 | 7176 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 7177 | struct netdev_adjacent, list); |
9ff162a8 JP |
7178 | if (likely(upper->master)) |
7179 | return upper->dev; | |
7180 | return NULL; | |
7181 | } | |
7182 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
7183 | ||
32b6d34f TY |
7184 | static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) |
7185 | { | |
7186 | struct netdev_adjacent *upper; | |
7187 | ||
7188 | ASSERT_RTNL(); | |
7189 | ||
7190 | if (list_empty(&dev->adj_list.upper)) | |
7191 | return NULL; | |
7192 | ||
7193 | upper = list_first_entry(&dev->adj_list.upper, | |
7194 | struct netdev_adjacent, list); | |
7195 | if (likely(upper->master) && !upper->ignore) | |
7196 | return upper->dev; | |
7197 | return NULL; | |
7198 | } | |
7199 | ||
0f524a80 DA |
7200 | /** |
7201 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
7202 | * @dev: device | |
7203 | * | |
7204 | * Find out if a device is linked to a lower device and return true in case | |
7205 | * it is. The caller must hold the RTNL lock. | |
7206 | */ | |
7207 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
7208 | { | |
7209 | ASSERT_RTNL(); | |
7210 | ||
7211 | return !list_empty(&dev->adj_list.lower); | |
7212 | } | |
7213 | ||
b6ccba4c VF |
7214 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
7215 | { | |
7216 | struct netdev_adjacent *adj; | |
7217 | ||
7218 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
7219 | ||
7220 | return adj->private; | |
7221 | } | |
7222 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
7223 | ||
44a40855 VY |
7224 | /** |
7225 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
7226 | * @dev: device | |
7227 | * @iter: list_head ** of the current position | |
7228 | * | |
7229 | * Gets the next device from the dev's upper list, starting from iter | |
7230 | * position. The caller must hold RCU read lock. | |
7231 | */ | |
7232 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
7233 | struct list_head **iter) | |
7234 | { | |
7235 | struct netdev_adjacent *upper; | |
7236 | ||
7237 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
7238 | ||
7239 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7240 | ||
7241 | if (&upper->list == &dev->adj_list.upper) | |
7242 | return NULL; | |
7243 | ||
7244 | *iter = &upper->list; | |
7245 | ||
7246 | return upper->dev; | |
7247 | } | |
7248 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
7249 | ||
32b6d34f TY |
7250 | static struct net_device *__netdev_next_upper_dev(struct net_device *dev, |
7251 | struct list_head **iter, | |
7252 | bool *ignore) | |
5343da4c TY |
7253 | { |
7254 | struct netdev_adjacent *upper; | |
7255 | ||
7256 | upper = list_entry((*iter)->next, struct netdev_adjacent, list); | |
7257 | ||
7258 | if (&upper->list == &dev->adj_list.upper) | |
7259 | return NULL; | |
7260 | ||
7261 | *iter = &upper->list; | |
32b6d34f | 7262 | *ignore = upper->ignore; |
5343da4c TY |
7263 | |
7264 | return upper->dev; | |
7265 | } | |
7266 | ||
1a3f060c DA |
7267 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
7268 | struct list_head **iter) | |
7269 | { | |
7270 | struct netdev_adjacent *upper; | |
7271 | ||
7272 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
7273 | ||
7274 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7275 | ||
7276 | if (&upper->list == &dev->adj_list.upper) | |
7277 | return NULL; | |
7278 | ||
7279 | *iter = &upper->list; | |
7280 | ||
7281 | return upper->dev; | |
7282 | } | |
7283 | ||
32b6d34f TY |
7284 | static int __netdev_walk_all_upper_dev(struct net_device *dev, |
7285 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7286 | struct netdev_nested_priv *priv), |
7287 | struct netdev_nested_priv *priv) | |
5343da4c TY |
7288 | { |
7289 | struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7290 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7291 | int ret, cur = 0; | |
32b6d34f | 7292 | bool ignore; |
5343da4c TY |
7293 | |
7294 | now = dev; | |
7295 | iter = &dev->adj_list.upper; | |
7296 | ||
7297 | while (1) { | |
7298 | if (now != dev) { | |
eff74233 | 7299 | ret = fn(now, priv); |
5343da4c TY |
7300 | if (ret) |
7301 | return ret; | |
7302 | } | |
7303 | ||
7304 | next = NULL; | |
7305 | while (1) { | |
32b6d34f | 7306 | udev = __netdev_next_upper_dev(now, &iter, &ignore); |
5343da4c TY |
7307 | if (!udev) |
7308 | break; | |
32b6d34f TY |
7309 | if (ignore) |
7310 | continue; | |
5343da4c TY |
7311 | |
7312 | next = udev; | |
7313 | niter = &udev->adj_list.upper; | |
7314 | dev_stack[cur] = now; | |
7315 | iter_stack[cur++] = iter; | |
7316 | break; | |
7317 | } | |
7318 | ||
7319 | if (!next) { | |
7320 | if (!cur) | |
7321 | return 0; | |
7322 | next = dev_stack[--cur]; | |
7323 | niter = iter_stack[cur]; | |
7324 | } | |
7325 | ||
7326 | now = next; | |
7327 | iter = niter; | |
7328 | } | |
7329 | ||
7330 | return 0; | |
7331 | } | |
7332 | ||
1a3f060c DA |
7333 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, |
7334 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7335 | struct netdev_nested_priv *priv), |
7336 | struct netdev_nested_priv *priv) | |
1a3f060c | 7337 | { |
5343da4c TY |
7338 | struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; |
7339 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7340 | int ret, cur = 0; | |
1a3f060c | 7341 | |
5343da4c TY |
7342 | now = dev; |
7343 | iter = &dev->adj_list.upper; | |
1a3f060c | 7344 | |
5343da4c TY |
7345 | while (1) { |
7346 | if (now != dev) { | |
eff74233 | 7347 | ret = fn(now, priv); |
5343da4c TY |
7348 | if (ret) |
7349 | return ret; | |
7350 | } | |
7351 | ||
7352 | next = NULL; | |
7353 | while (1) { | |
7354 | udev = netdev_next_upper_dev_rcu(now, &iter); | |
7355 | if (!udev) | |
7356 | break; | |
7357 | ||
7358 | next = udev; | |
7359 | niter = &udev->adj_list.upper; | |
7360 | dev_stack[cur] = now; | |
7361 | iter_stack[cur++] = iter; | |
7362 | break; | |
7363 | } | |
7364 | ||
7365 | if (!next) { | |
7366 | if (!cur) | |
7367 | return 0; | |
7368 | next = dev_stack[--cur]; | |
7369 | niter = iter_stack[cur]; | |
7370 | } | |
7371 | ||
7372 | now = next; | |
7373 | iter = niter; | |
1a3f060c DA |
7374 | } |
7375 | ||
7376 | return 0; | |
7377 | } | |
7378 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
7379 | ||
32b6d34f TY |
7380 | static bool __netdev_has_upper_dev(struct net_device *dev, |
7381 | struct net_device *upper_dev) | |
7382 | { | |
eff74233 | 7383 | struct netdev_nested_priv priv = { |
1fc70edb | 7384 | .flags = 0, |
eff74233 TY |
7385 | .data = (void *)upper_dev, |
7386 | }; | |
7387 | ||
32b6d34f TY |
7388 | ASSERT_RTNL(); |
7389 | ||
7390 | return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, | |
eff74233 | 7391 | &priv); |
32b6d34f TY |
7392 | } |
7393 | ||
31088a11 VF |
7394 | /** |
7395 | * netdev_lower_get_next_private - Get the next ->private from the | |
7396 | * lower neighbour list | |
7397 | * @dev: device | |
7398 | * @iter: list_head ** of the current position | |
7399 | * | |
7400 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
7401 | * list, starting from iter position. The caller must hold either hold the | |
7402 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 7403 | * list will remain unchanged. |
31088a11 VF |
7404 | */ |
7405 | void *netdev_lower_get_next_private(struct net_device *dev, | |
7406 | struct list_head **iter) | |
7407 | { | |
7408 | struct netdev_adjacent *lower; | |
7409 | ||
7410 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
7411 | ||
7412 | if (&lower->list == &dev->adj_list.lower) | |
7413 | return NULL; | |
7414 | ||
6859e7df | 7415 | *iter = lower->list.next; |
31088a11 VF |
7416 | |
7417 | return lower->private; | |
7418 | } | |
7419 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
7420 | ||
7421 | /** | |
7422 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
7423 | * lower neighbour list, RCU | |
7424 | * variant | |
7425 | * @dev: device | |
7426 | * @iter: list_head ** of the current position | |
7427 | * | |
7428 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
7429 | * list, starting from iter position. The caller must hold RCU read lock. | |
7430 | */ | |
7431 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
7432 | struct list_head **iter) | |
7433 | { | |
7434 | struct netdev_adjacent *lower; | |
7435 | ||
7436 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
7437 | ||
7438 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7439 | ||
7440 | if (&lower->list == &dev->adj_list.lower) | |
7441 | return NULL; | |
7442 | ||
6859e7df | 7443 | *iter = &lower->list; |
31088a11 VF |
7444 | |
7445 | return lower->private; | |
7446 | } | |
7447 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
7448 | ||
4085ebe8 VY |
7449 | /** |
7450 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
7451 | * list | |
7452 | * @dev: device | |
7453 | * @iter: list_head ** of the current position | |
7454 | * | |
7455 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
7456 | * list, starting from iter position. The caller must hold RTNL lock or | |
7457 | * its own locking that guarantees that the neighbour lower | |
b469139e | 7458 | * list will remain unchanged. |
4085ebe8 VY |
7459 | */ |
7460 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
7461 | { | |
7462 | struct netdev_adjacent *lower; | |
7463 | ||
cfdd28be | 7464 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
7465 | |
7466 | if (&lower->list == &dev->adj_list.lower) | |
7467 | return NULL; | |
7468 | ||
cfdd28be | 7469 | *iter = lower->list.next; |
4085ebe8 VY |
7470 | |
7471 | return lower->dev; | |
7472 | } | |
7473 | EXPORT_SYMBOL(netdev_lower_get_next); | |
7474 | ||
1a3f060c DA |
7475 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
7476 | struct list_head **iter) | |
7477 | { | |
7478 | struct netdev_adjacent *lower; | |
7479 | ||
46b5ab1a | 7480 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
7481 | |
7482 | if (&lower->list == &dev->adj_list.lower) | |
7483 | return NULL; | |
7484 | ||
46b5ab1a | 7485 | *iter = &lower->list; |
1a3f060c DA |
7486 | |
7487 | return lower->dev; | |
7488 | } | |
7489 | ||
32b6d34f TY |
7490 | static struct net_device *__netdev_next_lower_dev(struct net_device *dev, |
7491 | struct list_head **iter, | |
7492 | bool *ignore) | |
7493 | { | |
7494 | struct netdev_adjacent *lower; | |
7495 | ||
7496 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); | |
7497 | ||
7498 | if (&lower->list == &dev->adj_list.lower) | |
7499 | return NULL; | |
7500 | ||
7501 | *iter = &lower->list; | |
7502 | *ignore = lower->ignore; | |
7503 | ||
7504 | return lower->dev; | |
7505 | } | |
7506 | ||
1a3f060c DA |
7507 | int netdev_walk_all_lower_dev(struct net_device *dev, |
7508 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7509 | struct netdev_nested_priv *priv), |
7510 | struct netdev_nested_priv *priv) | |
1a3f060c | 7511 | { |
5343da4c TY |
7512 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; |
7513 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7514 | int ret, cur = 0; | |
1a3f060c | 7515 | |
5343da4c TY |
7516 | now = dev; |
7517 | iter = &dev->adj_list.lower; | |
1a3f060c | 7518 | |
5343da4c TY |
7519 | while (1) { |
7520 | if (now != dev) { | |
eff74233 | 7521 | ret = fn(now, priv); |
5343da4c TY |
7522 | if (ret) |
7523 | return ret; | |
7524 | } | |
7525 | ||
7526 | next = NULL; | |
7527 | while (1) { | |
7528 | ldev = netdev_next_lower_dev(now, &iter); | |
7529 | if (!ldev) | |
7530 | break; | |
7531 | ||
7532 | next = ldev; | |
7533 | niter = &ldev->adj_list.lower; | |
7534 | dev_stack[cur] = now; | |
7535 | iter_stack[cur++] = iter; | |
7536 | break; | |
7537 | } | |
7538 | ||
7539 | if (!next) { | |
7540 | if (!cur) | |
7541 | return 0; | |
7542 | next = dev_stack[--cur]; | |
7543 | niter = iter_stack[cur]; | |
7544 | } | |
7545 | ||
7546 | now = next; | |
7547 | iter = niter; | |
1a3f060c DA |
7548 | } |
7549 | ||
7550 | return 0; | |
7551 | } | |
7552 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
7553 | ||
32b6d34f TY |
7554 | static int __netdev_walk_all_lower_dev(struct net_device *dev, |
7555 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7556 | struct netdev_nested_priv *priv), |
7557 | struct netdev_nested_priv *priv) | |
32b6d34f TY |
7558 | { |
7559 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7560 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7561 | int ret, cur = 0; | |
7562 | bool ignore; | |
7563 | ||
7564 | now = dev; | |
7565 | iter = &dev->adj_list.lower; | |
7566 | ||
7567 | while (1) { | |
7568 | if (now != dev) { | |
eff74233 | 7569 | ret = fn(now, priv); |
32b6d34f TY |
7570 | if (ret) |
7571 | return ret; | |
7572 | } | |
7573 | ||
7574 | next = NULL; | |
7575 | while (1) { | |
7576 | ldev = __netdev_next_lower_dev(now, &iter, &ignore); | |
7577 | if (!ldev) | |
7578 | break; | |
7579 | if (ignore) | |
7580 | continue; | |
7581 | ||
7582 | next = ldev; | |
7583 | niter = &ldev->adj_list.lower; | |
7584 | dev_stack[cur] = now; | |
7585 | iter_stack[cur++] = iter; | |
7586 | break; | |
7587 | } | |
7588 | ||
7589 | if (!next) { | |
7590 | if (!cur) | |
7591 | return 0; | |
7592 | next = dev_stack[--cur]; | |
7593 | niter = iter_stack[cur]; | |
7594 | } | |
7595 | ||
7596 | now = next; | |
7597 | iter = niter; | |
7598 | } | |
7599 | ||
7600 | return 0; | |
7601 | } | |
7602 | ||
7151affe TY |
7603 | struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
7604 | struct list_head **iter) | |
1a3f060c DA |
7605 | { |
7606 | struct netdev_adjacent *lower; | |
7607 | ||
7608 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
7609 | if (&lower->list == &dev->adj_list.lower) | |
7610 | return NULL; | |
7611 | ||
7612 | *iter = &lower->list; | |
7613 | ||
7614 | return lower->dev; | |
7615 | } | |
7151affe | 7616 | EXPORT_SYMBOL(netdev_next_lower_dev_rcu); |
1a3f060c | 7617 | |
5343da4c TY |
7618 | static u8 __netdev_upper_depth(struct net_device *dev) |
7619 | { | |
7620 | struct net_device *udev; | |
7621 | struct list_head *iter; | |
7622 | u8 max_depth = 0; | |
32b6d34f | 7623 | bool ignore; |
5343da4c TY |
7624 | |
7625 | for (iter = &dev->adj_list.upper, | |
32b6d34f | 7626 | udev = __netdev_next_upper_dev(dev, &iter, &ignore); |
5343da4c | 7627 | udev; |
32b6d34f TY |
7628 | udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { |
7629 | if (ignore) | |
7630 | continue; | |
5343da4c TY |
7631 | if (max_depth < udev->upper_level) |
7632 | max_depth = udev->upper_level; | |
7633 | } | |
7634 | ||
7635 | return max_depth; | |
7636 | } | |
7637 | ||
7638 | static u8 __netdev_lower_depth(struct net_device *dev) | |
1a3f060c DA |
7639 | { |
7640 | struct net_device *ldev; | |
7641 | struct list_head *iter; | |
5343da4c | 7642 | u8 max_depth = 0; |
32b6d34f | 7643 | bool ignore; |
1a3f060c DA |
7644 | |
7645 | for (iter = &dev->adj_list.lower, | |
32b6d34f | 7646 | ldev = __netdev_next_lower_dev(dev, &iter, &ignore); |
1a3f060c | 7647 | ldev; |
32b6d34f TY |
7648 | ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { |
7649 | if (ignore) | |
7650 | continue; | |
5343da4c TY |
7651 | if (max_depth < ldev->lower_level) |
7652 | max_depth = ldev->lower_level; | |
7653 | } | |
1a3f060c | 7654 | |
5343da4c TY |
7655 | return max_depth; |
7656 | } | |
7657 | ||
eff74233 TY |
7658 | static int __netdev_update_upper_level(struct net_device *dev, |
7659 | struct netdev_nested_priv *__unused) | |
5343da4c TY |
7660 | { |
7661 | dev->upper_level = __netdev_upper_depth(dev) + 1; | |
7662 | return 0; | |
7663 | } | |
7664 | ||
eff74233 | 7665 | static int __netdev_update_lower_level(struct net_device *dev, |
1fc70edb | 7666 | struct netdev_nested_priv *priv) |
5343da4c TY |
7667 | { |
7668 | dev->lower_level = __netdev_lower_depth(dev) + 1; | |
1fc70edb TY |
7669 | |
7670 | #ifdef CONFIG_LOCKDEP | |
7671 | if (!priv) | |
7672 | return 0; | |
7673 | ||
7674 | if (priv->flags & NESTED_SYNC_IMM) | |
7675 | dev->nested_level = dev->lower_level - 1; | |
7676 | if (priv->flags & NESTED_SYNC_TODO) | |
7677 | net_unlink_todo(dev); | |
7678 | #endif | |
5343da4c TY |
7679 | return 0; |
7680 | } | |
7681 | ||
7682 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
7683 | int (*fn)(struct net_device *dev, | |
eff74233 TY |
7684 | struct netdev_nested_priv *priv), |
7685 | struct netdev_nested_priv *priv) | |
5343da4c TY |
7686 | { |
7687 | struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; | |
7688 | struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; | |
7689 | int ret, cur = 0; | |
7690 | ||
7691 | now = dev; | |
7692 | iter = &dev->adj_list.lower; | |
7693 | ||
7694 | while (1) { | |
7695 | if (now != dev) { | |
eff74233 | 7696 | ret = fn(now, priv); |
5343da4c TY |
7697 | if (ret) |
7698 | return ret; | |
7699 | } | |
7700 | ||
7701 | next = NULL; | |
7702 | while (1) { | |
7703 | ldev = netdev_next_lower_dev_rcu(now, &iter); | |
7704 | if (!ldev) | |
7705 | break; | |
7706 | ||
7707 | next = ldev; | |
7708 | niter = &ldev->adj_list.lower; | |
7709 | dev_stack[cur] = now; | |
7710 | iter_stack[cur++] = iter; | |
7711 | break; | |
7712 | } | |
7713 | ||
7714 | if (!next) { | |
7715 | if (!cur) | |
7716 | return 0; | |
7717 | next = dev_stack[--cur]; | |
7718 | niter = iter_stack[cur]; | |
7719 | } | |
7720 | ||
7721 | now = next; | |
7722 | iter = niter; | |
1a3f060c DA |
7723 | } |
7724 | ||
7725 | return 0; | |
7726 | } | |
7727 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
7728 | ||
e001bfad | 7729 | /** |
7730 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
7731 | * lower neighbour list, RCU | |
7732 | * variant | |
7733 | * @dev: device | |
7734 | * | |
7735 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
7736 | * list. The caller must hold RCU read lock. | |
7737 | */ | |
7738 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
7739 | { | |
7740 | struct netdev_adjacent *lower; | |
7741 | ||
7742 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
7743 | struct netdev_adjacent, list); | |
7744 | if (lower) | |
7745 | return lower->private; | |
7746 | return NULL; | |
7747 | } | |
7748 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
7749 | ||
9ff162a8 JP |
7750 | /** |
7751 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
7752 | * @dev: device | |
7753 | * | |
7754 | * Find a master upper device and return pointer to it or NULL in case | |
7755 | * it's not there. The caller must hold the RCU read lock. | |
7756 | */ | |
7757 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
7758 | { | |
aa9d8560 | 7759 | struct netdev_adjacent *upper; |
9ff162a8 | 7760 | |
2f268f12 | 7761 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 7762 | struct netdev_adjacent, list); |
9ff162a8 JP |
7763 | if (upper && likely(upper->master)) |
7764 | return upper->dev; | |
7765 | return NULL; | |
7766 | } | |
7767 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
7768 | ||
0a59f3a9 | 7769 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
7770 | struct net_device *adj_dev, |
7771 | struct list_head *dev_list) | |
7772 | { | |
7773 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 7774 | |
3ee32707 VF |
7775 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
7776 | "upper_%s" : "lower_%s", adj_dev->name); | |
7777 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
7778 | linkname); | |
7779 | } | |
0a59f3a9 | 7780 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
7781 | char *name, |
7782 | struct list_head *dev_list) | |
7783 | { | |
7784 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 7785 | |
3ee32707 VF |
7786 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
7787 | "upper_%s" : "lower_%s", name); | |
7788 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
7789 | } | |
7790 | ||
7ce64c79 AF |
7791 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
7792 | struct net_device *adj_dev, | |
7793 | struct list_head *dev_list) | |
7794 | { | |
7795 | return (dev_list == &dev->adj_list.upper || | |
7796 | dev_list == &dev->adj_list.lower) && | |
7797 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
7798 | } | |
3ee32707 | 7799 | |
5d261913 VF |
7800 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
7801 | struct net_device *adj_dev, | |
7863c054 | 7802 | struct list_head *dev_list, |
402dae96 | 7803 | void *private, bool master) |
5d261913 VF |
7804 | { |
7805 | struct netdev_adjacent *adj; | |
842d67a7 | 7806 | int ret; |
5d261913 | 7807 | |
6ea29da1 | 7808 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
7809 | |
7810 | if (adj) { | |
790510d9 | 7811 | adj->ref_nr += 1; |
67b62f98 DA |
7812 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
7813 | dev->name, adj_dev->name, adj->ref_nr); | |
7814 | ||
5d261913 VF |
7815 | return 0; |
7816 | } | |
7817 | ||
7818 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
7819 | if (!adj) | |
7820 | return -ENOMEM; | |
7821 | ||
7822 | adj->dev = adj_dev; | |
7823 | adj->master = master; | |
790510d9 | 7824 | adj->ref_nr = 1; |
402dae96 | 7825 | adj->private = private; |
32b6d34f | 7826 | adj->ignore = false; |
5d261913 | 7827 | dev_hold(adj_dev); |
2f268f12 | 7828 | |
67b62f98 DA |
7829 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
7830 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 7831 | |
7ce64c79 | 7832 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 7833 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
7834 | if (ret) |
7835 | goto free_adj; | |
7836 | } | |
7837 | ||
7863c054 | 7838 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
7839 | if (master) { |
7840 | ret = sysfs_create_link(&(dev->dev.kobj), | |
7841 | &(adj_dev->dev.kobj), "master"); | |
7842 | if (ret) | |
5831d66e | 7843 | goto remove_symlinks; |
842d67a7 | 7844 | |
7863c054 | 7845 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 7846 | } else { |
7863c054 | 7847 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 7848 | } |
5d261913 VF |
7849 | |
7850 | return 0; | |
842d67a7 | 7851 | |
5831d66e | 7852 | remove_symlinks: |
7ce64c79 | 7853 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 7854 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
7855 | free_adj: |
7856 | kfree(adj); | |
974daef7 | 7857 | dev_put(adj_dev); |
842d67a7 VF |
7858 | |
7859 | return ret; | |
5d261913 VF |
7860 | } |
7861 | ||
1d143d9f | 7862 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
7863 | struct net_device *adj_dev, | |
93409033 | 7864 | u16 ref_nr, |
1d143d9f | 7865 | struct list_head *dev_list) |
5d261913 VF |
7866 | { |
7867 | struct netdev_adjacent *adj; | |
7868 | ||
67b62f98 DA |
7869 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
7870 | dev->name, adj_dev->name, ref_nr); | |
7871 | ||
6ea29da1 | 7872 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 7873 | |
2f268f12 | 7874 | if (!adj) { |
67b62f98 | 7875 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 7876 | dev->name, adj_dev->name); |
67b62f98 DA |
7877 | WARN_ON(1); |
7878 | return; | |
2f268f12 | 7879 | } |
5d261913 | 7880 | |
93409033 | 7881 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
7882 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
7883 | dev->name, adj_dev->name, ref_nr, | |
7884 | adj->ref_nr - ref_nr); | |
93409033 | 7885 | adj->ref_nr -= ref_nr; |
5d261913 VF |
7886 | return; |
7887 | } | |
7888 | ||
842d67a7 VF |
7889 | if (adj->master) |
7890 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
7891 | ||
7ce64c79 | 7892 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 7893 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 7894 | |
5d261913 | 7895 | list_del_rcu(&adj->list); |
67b62f98 | 7896 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 7897 | adj_dev->name, dev->name, adj_dev->name); |
5d261913 VF |
7898 | dev_put(adj_dev); |
7899 | kfree_rcu(adj, rcu); | |
7900 | } | |
7901 | ||
1d143d9f | 7902 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
7903 | struct net_device *upper_dev, | |
7904 | struct list_head *up_list, | |
7905 | struct list_head *down_list, | |
7906 | void *private, bool master) | |
5d261913 VF |
7907 | { |
7908 | int ret; | |
7909 | ||
790510d9 | 7910 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 7911 | private, master); |
5d261913 VF |
7912 | if (ret) |
7913 | return ret; | |
7914 | ||
790510d9 | 7915 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 7916 | private, false); |
5d261913 | 7917 | if (ret) { |
790510d9 | 7918 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
7919 | return ret; |
7920 | } | |
7921 | ||
7922 | return 0; | |
7923 | } | |
7924 | ||
1d143d9f | 7925 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
7926 | struct net_device *upper_dev, | |
93409033 | 7927 | u16 ref_nr, |
1d143d9f | 7928 | struct list_head *up_list, |
7929 | struct list_head *down_list) | |
5d261913 | 7930 | { |
93409033 AC |
7931 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
7932 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
7933 | } |
7934 | ||
1d143d9f | 7935 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
7936 | struct net_device *upper_dev, | |
7937 | void *private, bool master) | |
2f268f12 | 7938 | { |
f1170fd4 DA |
7939 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
7940 | &dev->adj_list.upper, | |
7941 | &upper_dev->adj_list.lower, | |
7942 | private, master); | |
5d261913 VF |
7943 | } |
7944 | ||
1d143d9f | 7945 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
7946 | struct net_device *upper_dev) | |
2f268f12 | 7947 | { |
93409033 | 7948 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
7949 | &dev->adj_list.upper, |
7950 | &upper_dev->adj_list.lower); | |
7951 | } | |
5d261913 | 7952 | |
9ff162a8 | 7953 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 7954 | struct net_device *upper_dev, bool master, |
42ab19ee | 7955 | void *upper_priv, void *upper_info, |
1fc70edb | 7956 | struct netdev_nested_priv *priv, |
42ab19ee | 7957 | struct netlink_ext_ack *extack) |
9ff162a8 | 7958 | { |
51d0c047 DA |
7959 | struct netdev_notifier_changeupper_info changeupper_info = { |
7960 | .info = { | |
7961 | .dev = dev, | |
42ab19ee | 7962 | .extack = extack, |
51d0c047 DA |
7963 | }, |
7964 | .upper_dev = upper_dev, | |
7965 | .master = master, | |
7966 | .linking = true, | |
7967 | .upper_info = upper_info, | |
7968 | }; | |
50d629e7 | 7969 | struct net_device *master_dev; |
5d261913 | 7970 | int ret = 0; |
9ff162a8 JP |
7971 | |
7972 | ASSERT_RTNL(); | |
7973 | ||
7974 | if (dev == upper_dev) | |
7975 | return -EBUSY; | |
7976 | ||
7977 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
32b6d34f | 7978 | if (__netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
7979 | return -EBUSY; |
7980 | ||
5343da4c TY |
7981 | if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) |
7982 | return -EMLINK; | |
7983 | ||
50d629e7 | 7984 | if (!master) { |
32b6d34f | 7985 | if (__netdev_has_upper_dev(dev, upper_dev)) |
50d629e7 MM |
7986 | return -EEXIST; |
7987 | } else { | |
32b6d34f | 7988 | master_dev = __netdev_master_upper_dev_get(dev); |
50d629e7 MM |
7989 | if (master_dev) |
7990 | return master_dev == upper_dev ? -EEXIST : -EBUSY; | |
7991 | } | |
9ff162a8 | 7992 | |
51d0c047 | 7993 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
7994 | &changeupper_info.info); |
7995 | ret = notifier_to_errno(ret); | |
7996 | if (ret) | |
7997 | return ret; | |
7998 | ||
6dffb044 | 7999 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 8000 | master); |
5d261913 VF |
8001 | if (ret) |
8002 | return ret; | |
9ff162a8 | 8003 | |
51d0c047 | 8004 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
b03804e7 IS |
8005 | &changeupper_info.info); |
8006 | ret = notifier_to_errno(ret); | |
8007 | if (ret) | |
f1170fd4 | 8008 | goto rollback; |
b03804e7 | 8009 | |
5343da4c | 8010 | __netdev_update_upper_level(dev, NULL); |
32b6d34f | 8011 | __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); |
5343da4c | 8012 | |
1fc70edb | 8013 | __netdev_update_lower_level(upper_dev, priv); |
32b6d34f | 8014 | __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, |
1fc70edb | 8015 | priv); |
5343da4c | 8016 | |
9ff162a8 | 8017 | return 0; |
5d261913 | 8018 | |
f1170fd4 | 8019 | rollback: |
2f268f12 | 8020 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
8021 | |
8022 | return ret; | |
9ff162a8 JP |
8023 | } |
8024 | ||
8025 | /** | |
8026 | * netdev_upper_dev_link - Add a link to the upper device | |
8027 | * @dev: device | |
8028 | * @upper_dev: new upper device | |
7a006d59 | 8029 | * @extack: netlink extended ack |
9ff162a8 JP |
8030 | * |
8031 | * Adds a link to device which is upper to this one. The caller must hold | |
8032 | * the RTNL lock. On a failure a negative errno code is returned. | |
8033 | * On success the reference counts are adjusted and the function | |
8034 | * returns zero. | |
8035 | */ | |
8036 | int netdev_upper_dev_link(struct net_device *dev, | |
42ab19ee DA |
8037 | struct net_device *upper_dev, |
8038 | struct netlink_ext_ack *extack) | |
9ff162a8 | 8039 | { |
1fc70edb TY |
8040 | struct netdev_nested_priv priv = { |
8041 | .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, | |
8042 | .data = NULL, | |
8043 | }; | |
8044 | ||
42ab19ee | 8045 | return __netdev_upper_dev_link(dev, upper_dev, false, |
1fc70edb | 8046 | NULL, NULL, &priv, extack); |
9ff162a8 JP |
8047 | } |
8048 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
8049 | ||
8050 | /** | |
8051 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
8052 | * @dev: device | |
8053 | * @upper_dev: new upper device | |
6dffb044 | 8054 | * @upper_priv: upper device private |
29bf24af | 8055 | * @upper_info: upper info to be passed down via notifier |
7a006d59 | 8056 | * @extack: netlink extended ack |
9ff162a8 JP |
8057 | * |
8058 | * Adds a link to device which is upper to this one. In this case, only | |
8059 | * one master upper device can be linked, although other non-master devices | |
8060 | * might be linked as well. The caller must hold the RTNL lock. | |
8061 | * On a failure a negative errno code is returned. On success the reference | |
8062 | * counts are adjusted and the function returns zero. | |
8063 | */ | |
8064 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 8065 | struct net_device *upper_dev, |
42ab19ee DA |
8066 | void *upper_priv, void *upper_info, |
8067 | struct netlink_ext_ack *extack) | |
9ff162a8 | 8068 | { |
1fc70edb TY |
8069 | struct netdev_nested_priv priv = { |
8070 | .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, | |
8071 | .data = NULL, | |
8072 | }; | |
8073 | ||
29bf24af | 8074 | return __netdev_upper_dev_link(dev, upper_dev, true, |
1fc70edb | 8075 | upper_priv, upper_info, &priv, extack); |
9ff162a8 JP |
8076 | } |
8077 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
8078 | ||
fe8300fd | 8079 | static void __netdev_upper_dev_unlink(struct net_device *dev, |
1fc70edb TY |
8080 | struct net_device *upper_dev, |
8081 | struct netdev_nested_priv *priv) | |
9ff162a8 | 8082 | { |
51d0c047 DA |
8083 | struct netdev_notifier_changeupper_info changeupper_info = { |
8084 | .info = { | |
8085 | .dev = dev, | |
8086 | }, | |
8087 | .upper_dev = upper_dev, | |
8088 | .linking = false, | |
8089 | }; | |
f4563a75 | 8090 | |
9ff162a8 JP |
8091 | ASSERT_RTNL(); |
8092 | ||
0e4ead9d | 8093 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; |
0e4ead9d | 8094 | |
51d0c047 | 8095 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
8096 | &changeupper_info.info); |
8097 | ||
2f268f12 | 8098 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 8099 | |
51d0c047 | 8100 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
0e4ead9d | 8101 | &changeupper_info.info); |
5343da4c TY |
8102 | |
8103 | __netdev_update_upper_level(dev, NULL); | |
32b6d34f | 8104 | __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); |
5343da4c | 8105 | |
1fc70edb | 8106 | __netdev_update_lower_level(upper_dev, priv); |
32b6d34f | 8107 | __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, |
1fc70edb | 8108 | priv); |
9ff162a8 | 8109 | } |
fe8300fd TY |
8110 | |
8111 | /** | |
8112 | * netdev_upper_dev_unlink - Removes a link to upper device | |
8113 | * @dev: device | |
8114 | * @upper_dev: new upper device | |
8115 | * | |
8116 | * Removes a link to device which is upper to this one. The caller must hold | |
8117 | * the RTNL lock. | |
8118 | */ | |
8119 | void netdev_upper_dev_unlink(struct net_device *dev, | |
8120 | struct net_device *upper_dev) | |
8121 | { | |
1fc70edb TY |
8122 | struct netdev_nested_priv priv = { |
8123 | .flags = NESTED_SYNC_TODO, | |
8124 | .data = NULL, | |
8125 | }; | |
8126 | ||
8127 | __netdev_upper_dev_unlink(dev, upper_dev, &priv); | |
9ff162a8 JP |
8128 | } |
8129 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
8130 | ||
32b6d34f TY |
8131 | static void __netdev_adjacent_dev_set(struct net_device *upper_dev, |
8132 | struct net_device *lower_dev, | |
8133 | bool val) | |
8134 | { | |
8135 | struct netdev_adjacent *adj; | |
8136 | ||
8137 | adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); | |
8138 | if (adj) | |
8139 | adj->ignore = val; | |
8140 | ||
8141 | adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); | |
8142 | if (adj) | |
8143 | adj->ignore = val; | |
8144 | } | |
8145 | ||
8146 | static void netdev_adjacent_dev_disable(struct net_device *upper_dev, | |
8147 | struct net_device *lower_dev) | |
8148 | { | |
8149 | __netdev_adjacent_dev_set(upper_dev, lower_dev, true); | |
8150 | } | |
8151 | ||
8152 | static void netdev_adjacent_dev_enable(struct net_device *upper_dev, | |
8153 | struct net_device *lower_dev) | |
8154 | { | |
8155 | __netdev_adjacent_dev_set(upper_dev, lower_dev, false); | |
8156 | } | |
8157 | ||
8158 | int netdev_adjacent_change_prepare(struct net_device *old_dev, | |
8159 | struct net_device *new_dev, | |
8160 | struct net_device *dev, | |
8161 | struct netlink_ext_ack *extack) | |
8162 | { | |
1fc70edb TY |
8163 | struct netdev_nested_priv priv = { |
8164 | .flags = 0, | |
8165 | .data = NULL, | |
8166 | }; | |
32b6d34f TY |
8167 | int err; |
8168 | ||
8169 | if (!new_dev) | |
8170 | return 0; | |
8171 | ||
8172 | if (old_dev && new_dev != old_dev) | |
8173 | netdev_adjacent_dev_disable(dev, old_dev); | |
1fc70edb TY |
8174 | err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, |
8175 | extack); | |
32b6d34f TY |
8176 | if (err) { |
8177 | if (old_dev && new_dev != old_dev) | |
8178 | netdev_adjacent_dev_enable(dev, old_dev); | |
8179 | return err; | |
8180 | } | |
8181 | ||
8182 | return 0; | |
8183 | } | |
8184 | EXPORT_SYMBOL(netdev_adjacent_change_prepare); | |
8185 | ||
8186 | void netdev_adjacent_change_commit(struct net_device *old_dev, | |
8187 | struct net_device *new_dev, | |
8188 | struct net_device *dev) | |
8189 | { | |
1fc70edb TY |
8190 | struct netdev_nested_priv priv = { |
8191 | .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, | |
8192 | .data = NULL, | |
8193 | }; | |
8194 | ||
32b6d34f TY |
8195 | if (!new_dev || !old_dev) |
8196 | return; | |
8197 | ||
8198 | if (new_dev == old_dev) | |
8199 | return; | |
8200 | ||
8201 | netdev_adjacent_dev_enable(dev, old_dev); | |
1fc70edb | 8202 | __netdev_upper_dev_unlink(old_dev, dev, &priv); |
32b6d34f TY |
8203 | } |
8204 | EXPORT_SYMBOL(netdev_adjacent_change_commit); | |
8205 | ||
8206 | void netdev_adjacent_change_abort(struct net_device *old_dev, | |
8207 | struct net_device *new_dev, | |
8208 | struct net_device *dev) | |
8209 | { | |
1fc70edb TY |
8210 | struct netdev_nested_priv priv = { |
8211 | .flags = 0, | |
8212 | .data = NULL, | |
8213 | }; | |
8214 | ||
32b6d34f TY |
8215 | if (!new_dev) |
8216 | return; | |
8217 | ||
8218 | if (old_dev && new_dev != old_dev) | |
8219 | netdev_adjacent_dev_enable(dev, old_dev); | |
8220 | ||
1fc70edb | 8221 | __netdev_upper_dev_unlink(new_dev, dev, &priv); |
32b6d34f TY |
8222 | } |
8223 | EXPORT_SYMBOL(netdev_adjacent_change_abort); | |
8224 | ||
61bd3857 MS |
8225 | /** |
8226 | * netdev_bonding_info_change - Dispatch event about slave change | |
8227 | * @dev: device | |
4a26e453 | 8228 | * @bonding_info: info to dispatch |
61bd3857 MS |
8229 | * |
8230 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
8231 | * The caller must hold the RTNL lock. | |
8232 | */ | |
8233 | void netdev_bonding_info_change(struct net_device *dev, | |
8234 | struct netdev_bonding_info *bonding_info) | |
8235 | { | |
51d0c047 DA |
8236 | struct netdev_notifier_bonding_info info = { |
8237 | .info.dev = dev, | |
8238 | }; | |
61bd3857 MS |
8239 | |
8240 | memcpy(&info.bonding_info, bonding_info, | |
8241 | sizeof(struct netdev_bonding_info)); | |
51d0c047 | 8242 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, |
61bd3857 MS |
8243 | &info.info); |
8244 | } | |
8245 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
8246 | ||
cff9f12b MG |
8247 | /** |
8248 | * netdev_get_xmit_slave - Get the xmit slave of master device | |
8842500d | 8249 | * @dev: device |
cff9f12b MG |
8250 | * @skb: The packet |
8251 | * @all_slaves: assume all the slaves are active | |
8252 | * | |
8253 | * The reference counters are not incremented so the caller must be | |
8254 | * careful with locks. The caller must hold RCU lock. | |
8255 | * %NULL is returned if no slave is found. | |
8256 | */ | |
8257 | ||
8258 | struct net_device *netdev_get_xmit_slave(struct net_device *dev, | |
8259 | struct sk_buff *skb, | |
8260 | bool all_slaves) | |
8261 | { | |
8262 | const struct net_device_ops *ops = dev->netdev_ops; | |
8263 | ||
8264 | if (!ops->ndo_get_xmit_slave) | |
8265 | return NULL; | |
8266 | return ops->ndo_get_xmit_slave(dev, skb, all_slaves); | |
8267 | } | |
8268 | EXPORT_SYMBOL(netdev_get_xmit_slave); | |
8269 | ||
719a402c TT |
8270 | static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, |
8271 | struct sock *sk) | |
8272 | { | |
8273 | const struct net_device_ops *ops = dev->netdev_ops; | |
8274 | ||
8275 | if (!ops->ndo_sk_get_lower_dev) | |
8276 | return NULL; | |
8277 | return ops->ndo_sk_get_lower_dev(dev, sk); | |
8278 | } | |
8279 | ||
8280 | /** | |
8281 | * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket | |
8282 | * @dev: device | |
8283 | * @sk: the socket | |
8284 | * | |
8285 | * %NULL is returned if no lower device is found. | |
8286 | */ | |
8287 | ||
8288 | struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, | |
8289 | struct sock *sk) | |
8290 | { | |
8291 | struct net_device *lower; | |
8292 | ||
8293 | lower = netdev_sk_get_lower_dev(dev, sk); | |
8294 | while (lower) { | |
8295 | dev = lower; | |
8296 | lower = netdev_sk_get_lower_dev(dev, sk); | |
8297 | } | |
8298 | ||
8299 | return dev; | |
8300 | } | |
8301 | EXPORT_SYMBOL(netdev_sk_get_lowest_dev); | |
8302 | ||
2ce1ee17 | 8303 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
8304 | { |
8305 | struct netdev_adjacent *iter; | |
8306 | ||
8307 | struct net *net = dev_net(dev); | |
8308 | ||
8309 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 8310 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
8311 | continue; |
8312 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
8313 | &iter->dev->adj_list.lower); | |
8314 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
8315 | &dev->adj_list.upper); | |
8316 | } | |
8317 | ||
8318 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 8319 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
8320 | continue; |
8321 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
8322 | &iter->dev->adj_list.upper); | |
8323 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
8324 | &dev->adj_list.lower); | |
8325 | } | |
8326 | } | |
8327 | ||
2ce1ee17 | 8328 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
8329 | { |
8330 | struct netdev_adjacent *iter; | |
8331 | ||
8332 | struct net *net = dev_net(dev); | |
8333 | ||
8334 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 8335 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
8336 | continue; |
8337 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
8338 | &iter->dev->adj_list.lower); | |
8339 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
8340 | &dev->adj_list.upper); | |
8341 | } | |
8342 | ||
8343 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 8344 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
8345 | continue; |
8346 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
8347 | &iter->dev->adj_list.upper); | |
8348 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
8349 | &dev->adj_list.lower); | |
8350 | } | |
8351 | } | |
8352 | ||
5bb025fa | 8353 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 8354 | { |
5bb025fa | 8355 | struct netdev_adjacent *iter; |
402dae96 | 8356 | |
4c75431a AF |
8357 | struct net *net = dev_net(dev); |
8358 | ||
5bb025fa | 8359 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 8360 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 8361 | continue; |
5bb025fa VF |
8362 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
8363 | &iter->dev->adj_list.lower); | |
8364 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
8365 | &iter->dev->adj_list.lower); | |
8366 | } | |
402dae96 | 8367 | |
5bb025fa | 8368 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 8369 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 8370 | continue; |
5bb025fa VF |
8371 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
8372 | &iter->dev->adj_list.upper); | |
8373 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
8374 | &iter->dev->adj_list.upper); | |
8375 | } | |
402dae96 | 8376 | } |
402dae96 VF |
8377 | |
8378 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
8379 | struct net_device *lower_dev) | |
8380 | { | |
8381 | struct netdev_adjacent *lower; | |
8382 | ||
8383 | if (!lower_dev) | |
8384 | return NULL; | |
6ea29da1 | 8385 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
8386 | if (!lower) |
8387 | return NULL; | |
8388 | ||
8389 | return lower->private; | |
8390 | } | |
8391 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
8392 | ||
4085ebe8 | 8393 | |
04d48266 | 8394 | /** |
c1639be9 | 8395 | * netdev_lower_state_changed - Dispatch event about lower device state change |
04d48266 JP |
8396 | * @lower_dev: device |
8397 | * @lower_state_info: state to dispatch | |
8398 | * | |
8399 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
8400 | * The caller must hold the RTNL lock. | |
8401 | */ | |
8402 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
8403 | void *lower_state_info) | |
8404 | { | |
51d0c047 DA |
8405 | struct netdev_notifier_changelowerstate_info changelowerstate_info = { |
8406 | .info.dev = lower_dev, | |
8407 | }; | |
04d48266 JP |
8408 | |
8409 | ASSERT_RTNL(); | |
8410 | changelowerstate_info.lower_state_info = lower_state_info; | |
51d0c047 | 8411 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, |
04d48266 JP |
8412 | &changelowerstate_info.info); |
8413 | } | |
8414 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
8415 | ||
b6c40d68 PM |
8416 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
8417 | { | |
d314774c SH |
8418 | const struct net_device_ops *ops = dev->netdev_ops; |
8419 | ||
d2615bf4 | 8420 | if (ops->ndo_change_rx_flags) |
d314774c | 8421 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
8422 | } |
8423 | ||
991fb3f7 | 8424 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 8425 | { |
b536db93 | 8426 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
8427 | kuid_t uid; |
8428 | kgid_t gid; | |
1da177e4 | 8429 | |
24023451 PM |
8430 | ASSERT_RTNL(); |
8431 | ||
dad9b335 WC |
8432 | dev->flags |= IFF_PROMISC; |
8433 | dev->promiscuity += inc; | |
8434 | if (dev->promiscuity == 0) { | |
8435 | /* | |
8436 | * Avoid overflow. | |
8437 | * If inc causes overflow, untouch promisc and return error. | |
8438 | */ | |
8439 | if (inc < 0) | |
8440 | dev->flags &= ~IFF_PROMISC; | |
8441 | else { | |
8442 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
8443 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
8444 | dev->name); | |
dad9b335 WC |
8445 | return -EOVERFLOW; |
8446 | } | |
8447 | } | |
52609c0b | 8448 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
8449 | pr_info("device %s %s promiscuous mode\n", |
8450 | dev->name, | |
8451 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
8452 | if (audit_enabled) { |
8453 | current_uid_gid(&uid, &gid); | |
cdfb6b34 RGB |
8454 | audit_log(audit_context(), GFP_ATOMIC, |
8455 | AUDIT_ANOM_PROMISCUOUS, | |
8456 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
8457 | dev->name, (dev->flags & IFF_PROMISC), | |
8458 | (old_flags & IFF_PROMISC), | |
8459 | from_kuid(&init_user_ns, audit_get_loginuid(current)), | |
8460 | from_kuid(&init_user_ns, uid), | |
8461 | from_kgid(&init_user_ns, gid), | |
8462 | audit_get_sessionid(current)); | |
8192b0c4 | 8463 | } |
24023451 | 8464 | |
b6c40d68 | 8465 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 8466 | } |
991fb3f7 ND |
8467 | if (notify) |
8468 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 8469 | return 0; |
1da177e4 LT |
8470 | } |
8471 | ||
4417da66 PM |
8472 | /** |
8473 | * dev_set_promiscuity - update promiscuity count on a device | |
8474 | * @dev: device | |
8475 | * @inc: modifier | |
8476 | * | |
8477 | * Add or remove promiscuity from a device. While the count in the device | |
8478 | * remains above zero the interface remains promiscuous. Once it hits zero | |
8479 | * the device reverts back to normal filtering operation. A negative inc | |
8480 | * value is used to drop promiscuity on the device. | |
dad9b335 | 8481 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 8482 | */ |
dad9b335 | 8483 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 8484 | { |
b536db93 | 8485 | unsigned int old_flags = dev->flags; |
dad9b335 | 8486 | int err; |
4417da66 | 8487 | |
991fb3f7 | 8488 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 8489 | if (err < 0) |
dad9b335 | 8490 | return err; |
4417da66 PM |
8491 | if (dev->flags != old_flags) |
8492 | dev_set_rx_mode(dev); | |
dad9b335 | 8493 | return err; |
4417da66 | 8494 | } |
d1b19dff | 8495 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 8496 | |
991fb3f7 | 8497 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 8498 | { |
991fb3f7 | 8499 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 8500 | |
24023451 PM |
8501 | ASSERT_RTNL(); |
8502 | ||
1da177e4 | 8503 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
8504 | dev->allmulti += inc; |
8505 | if (dev->allmulti == 0) { | |
8506 | /* | |
8507 | * Avoid overflow. | |
8508 | * If inc causes overflow, untouch allmulti and return error. | |
8509 | */ | |
8510 | if (inc < 0) | |
8511 | dev->flags &= ~IFF_ALLMULTI; | |
8512 | else { | |
8513 | dev->allmulti -= inc; | |
7b6cd1ce JP |
8514 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
8515 | dev->name); | |
dad9b335 WC |
8516 | return -EOVERFLOW; |
8517 | } | |
8518 | } | |
24023451 | 8519 | if (dev->flags ^ old_flags) { |
b6c40d68 | 8520 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 8521 | dev_set_rx_mode(dev); |
991fb3f7 ND |
8522 | if (notify) |
8523 | __dev_notify_flags(dev, old_flags, | |
8524 | dev->gflags ^ old_gflags); | |
24023451 | 8525 | } |
dad9b335 | 8526 | return 0; |
4417da66 | 8527 | } |
991fb3f7 ND |
8528 | |
8529 | /** | |
8530 | * dev_set_allmulti - update allmulti count on a device | |
8531 | * @dev: device | |
8532 | * @inc: modifier | |
8533 | * | |
8534 | * Add or remove reception of all multicast frames to a device. While the | |
8535 | * count in the device remains above zero the interface remains listening | |
8536 | * to all interfaces. Once it hits zero the device reverts back to normal | |
8537 | * filtering operation. A negative @inc value is used to drop the counter | |
8538 | * when releasing a resource needing all multicasts. | |
8539 | * Return 0 if successful or a negative errno code on error. | |
8540 | */ | |
8541 | ||
8542 | int dev_set_allmulti(struct net_device *dev, int inc) | |
8543 | { | |
8544 | return __dev_set_allmulti(dev, inc, true); | |
8545 | } | |
d1b19dff | 8546 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
8547 | |
8548 | /* | |
8549 | * Upload unicast and multicast address lists to device and | |
8550 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 8551 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
8552 | * are present. |
8553 | */ | |
8554 | void __dev_set_rx_mode(struct net_device *dev) | |
8555 | { | |
d314774c SH |
8556 | const struct net_device_ops *ops = dev->netdev_ops; |
8557 | ||
4417da66 PM |
8558 | /* dev_open will call this function so the list will stay sane. */ |
8559 | if (!(dev->flags&IFF_UP)) | |
8560 | return; | |
8561 | ||
8562 | if (!netif_device_present(dev)) | |
40b77c94 | 8563 | return; |
4417da66 | 8564 | |
01789349 | 8565 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
8566 | /* Unicast addresses changes may only happen under the rtnl, |
8567 | * therefore calling __dev_set_promiscuity here is safe. | |
8568 | */ | |
32e7bfc4 | 8569 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 8570 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 8571 | dev->uc_promisc = true; |
32e7bfc4 | 8572 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 8573 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 8574 | dev->uc_promisc = false; |
4417da66 | 8575 | } |
4417da66 | 8576 | } |
01789349 JP |
8577 | |
8578 | if (ops->ndo_set_rx_mode) | |
8579 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
8580 | } |
8581 | ||
8582 | void dev_set_rx_mode(struct net_device *dev) | |
8583 | { | |
b9e40857 | 8584 | netif_addr_lock_bh(dev); |
4417da66 | 8585 | __dev_set_rx_mode(dev); |
b9e40857 | 8586 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
8587 | } |
8588 | ||
f0db275a SH |
8589 | /** |
8590 | * dev_get_flags - get flags reported to userspace | |
8591 | * @dev: device | |
8592 | * | |
8593 | * Get the combination of flag bits exported through APIs to userspace. | |
8594 | */ | |
95c96174 | 8595 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 8596 | { |
95c96174 | 8597 | unsigned int flags; |
1da177e4 LT |
8598 | |
8599 | flags = (dev->flags & ~(IFF_PROMISC | | |
8600 | IFF_ALLMULTI | | |
b00055aa SR |
8601 | IFF_RUNNING | |
8602 | IFF_LOWER_UP | | |
8603 | IFF_DORMANT)) | | |
1da177e4 LT |
8604 | (dev->gflags & (IFF_PROMISC | |
8605 | IFF_ALLMULTI)); | |
8606 | ||
b00055aa SR |
8607 | if (netif_running(dev)) { |
8608 | if (netif_oper_up(dev)) | |
8609 | flags |= IFF_RUNNING; | |
8610 | if (netif_carrier_ok(dev)) | |
8611 | flags |= IFF_LOWER_UP; | |
8612 | if (netif_dormant(dev)) | |
8613 | flags |= IFF_DORMANT; | |
8614 | } | |
1da177e4 LT |
8615 | |
8616 | return flags; | |
8617 | } | |
d1b19dff | 8618 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 8619 | |
6d040321 PM |
8620 | int __dev_change_flags(struct net_device *dev, unsigned int flags, |
8621 | struct netlink_ext_ack *extack) | |
1da177e4 | 8622 | { |
b536db93 | 8623 | unsigned int old_flags = dev->flags; |
bd380811 | 8624 | int ret; |
1da177e4 | 8625 | |
24023451 PM |
8626 | ASSERT_RTNL(); |
8627 | ||
1da177e4 LT |
8628 | /* |
8629 | * Set the flags on our device. | |
8630 | */ | |
8631 | ||
8632 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
8633 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
8634 | IFF_AUTOMEDIA)) | | |
8635 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
8636 | IFF_ALLMULTI)); | |
8637 | ||
8638 | /* | |
8639 | * Load in the correct multicast list now the flags have changed. | |
8640 | */ | |
8641 | ||
b6c40d68 PM |
8642 | if ((old_flags ^ flags) & IFF_MULTICAST) |
8643 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 8644 | |
4417da66 | 8645 | dev_set_rx_mode(dev); |
1da177e4 LT |
8646 | |
8647 | /* | |
8648 | * Have we downed the interface. We handle IFF_UP ourselves | |
8649 | * according to user attempts to set it, rather than blindly | |
8650 | * setting it. | |
8651 | */ | |
8652 | ||
8653 | ret = 0; | |
7051b88a | 8654 | if ((old_flags ^ flags) & IFF_UP) { |
8655 | if (old_flags & IFF_UP) | |
8656 | __dev_close(dev); | |
8657 | else | |
40c900aa | 8658 | ret = __dev_open(dev, extack); |
7051b88a | 8659 | } |
1da177e4 | 8660 | |
1da177e4 | 8661 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 8662 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 8663 | unsigned int old_flags = dev->flags; |
d1b19dff | 8664 | |
1da177e4 | 8665 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
8666 | |
8667 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
8668 | if (dev->flags != old_flags) | |
8669 | dev_set_rx_mode(dev); | |
1da177e4 LT |
8670 | } |
8671 | ||
8672 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 8673 | * is important. Some (broken) drivers set IFF_PROMISC, when |
8674 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
8675 | */ |
8676 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
8677 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
8678 | ||
1da177e4 | 8679 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 8680 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
8681 | } |
8682 | ||
bd380811 PM |
8683 | return ret; |
8684 | } | |
8685 | ||
a528c219 ND |
8686 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
8687 | unsigned int gchanges) | |
bd380811 PM |
8688 | { |
8689 | unsigned int changes = dev->flags ^ old_flags; | |
8690 | ||
a528c219 | 8691 | if (gchanges) |
7f294054 | 8692 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 8693 | |
bd380811 PM |
8694 | if (changes & IFF_UP) { |
8695 | if (dev->flags & IFF_UP) | |
8696 | call_netdevice_notifiers(NETDEV_UP, dev); | |
8697 | else | |
8698 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
8699 | } | |
8700 | ||
8701 | if (dev->flags & IFF_UP && | |
be9efd36 | 8702 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
51d0c047 DA |
8703 | struct netdev_notifier_change_info change_info = { |
8704 | .info = { | |
8705 | .dev = dev, | |
8706 | }, | |
8707 | .flags_changed = changes, | |
8708 | }; | |
be9efd36 | 8709 | |
51d0c047 | 8710 | call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); |
be9efd36 | 8711 | } |
bd380811 PM |
8712 | } |
8713 | ||
8714 | /** | |
8715 | * dev_change_flags - change device settings | |
8716 | * @dev: device | |
8717 | * @flags: device state flags | |
567c5e13 | 8718 | * @extack: netlink extended ack |
bd380811 PM |
8719 | * |
8720 | * Change settings on device based state flags. The flags are | |
8721 | * in the userspace exported format. | |
8722 | */ | |
567c5e13 PM |
8723 | int dev_change_flags(struct net_device *dev, unsigned int flags, |
8724 | struct netlink_ext_ack *extack) | |
bd380811 | 8725 | { |
b536db93 | 8726 | int ret; |
991fb3f7 | 8727 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 | 8728 | |
6d040321 | 8729 | ret = __dev_change_flags(dev, flags, extack); |
bd380811 PM |
8730 | if (ret < 0) |
8731 | return ret; | |
8732 | ||
991fb3f7 | 8733 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 8734 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
8735 | return ret; |
8736 | } | |
d1b19dff | 8737 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 8738 | |
f51048c3 | 8739 | int __dev_set_mtu(struct net_device *dev, int new_mtu) |
2315dc91 VF |
8740 | { |
8741 | const struct net_device_ops *ops = dev->netdev_ops; | |
8742 | ||
8743 | if (ops->ndo_change_mtu) | |
8744 | return ops->ndo_change_mtu(dev, new_mtu); | |
8745 | ||
501a90c9 ED |
8746 | /* Pairs with all the lockless reads of dev->mtu in the stack */ |
8747 | WRITE_ONCE(dev->mtu, new_mtu); | |
2315dc91 VF |
8748 | return 0; |
8749 | } | |
f51048c3 | 8750 | EXPORT_SYMBOL(__dev_set_mtu); |
2315dc91 | 8751 | |
d836f5c6 ED |
8752 | int dev_validate_mtu(struct net_device *dev, int new_mtu, |
8753 | struct netlink_ext_ack *extack) | |
8754 | { | |
8755 | /* MTU must be positive, and in range */ | |
8756 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
8757 | NL_SET_ERR_MSG(extack, "mtu less than device minimum"); | |
8758 | return -EINVAL; | |
8759 | } | |
8760 | ||
8761 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
8762 | NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); | |
8763 | return -EINVAL; | |
8764 | } | |
8765 | return 0; | |
8766 | } | |
8767 | ||
f0db275a | 8768 | /** |
7a4c53be | 8769 | * dev_set_mtu_ext - Change maximum transfer unit |
f0db275a SH |
8770 | * @dev: device |
8771 | * @new_mtu: new transfer unit | |
7a4c53be | 8772 | * @extack: netlink extended ack |
f0db275a SH |
8773 | * |
8774 | * Change the maximum transfer size of the network device. | |
8775 | */ | |
7a4c53be SH |
8776 | int dev_set_mtu_ext(struct net_device *dev, int new_mtu, |
8777 | struct netlink_ext_ack *extack) | |
1da177e4 | 8778 | { |
2315dc91 | 8779 | int err, orig_mtu; |
1da177e4 LT |
8780 | |
8781 | if (new_mtu == dev->mtu) | |
8782 | return 0; | |
8783 | ||
d836f5c6 ED |
8784 | err = dev_validate_mtu(dev, new_mtu, extack); |
8785 | if (err) | |
8786 | return err; | |
1da177e4 LT |
8787 | |
8788 | if (!netif_device_present(dev)) | |
8789 | return -ENODEV; | |
8790 | ||
1d486bfb VF |
8791 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
8792 | err = notifier_to_errno(err); | |
8793 | if (err) | |
8794 | return err; | |
d314774c | 8795 | |
2315dc91 VF |
8796 | orig_mtu = dev->mtu; |
8797 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 8798 | |
2315dc91 | 8799 | if (!err) { |
af7d6cce SD |
8800 | err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
8801 | orig_mtu); | |
2315dc91 VF |
8802 | err = notifier_to_errno(err); |
8803 | if (err) { | |
8804 | /* setting mtu back and notifying everyone again, | |
8805 | * so that they have a chance to revert changes. | |
8806 | */ | |
8807 | __dev_set_mtu(dev, orig_mtu); | |
af7d6cce SD |
8808 | call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
8809 | new_mtu); | |
2315dc91 VF |
8810 | } |
8811 | } | |
1da177e4 LT |
8812 | return err; |
8813 | } | |
7a4c53be SH |
8814 | |
8815 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
8816 | { | |
8817 | struct netlink_ext_ack extack; | |
8818 | int err; | |
8819 | ||
a6bcfc89 | 8820 | memset(&extack, 0, sizeof(extack)); |
7a4c53be | 8821 | err = dev_set_mtu_ext(dev, new_mtu, &extack); |
a6bcfc89 | 8822 | if (err && extack._msg) |
7a4c53be SH |
8823 | net_err_ratelimited("%s: %s\n", dev->name, extack._msg); |
8824 | return err; | |
8825 | } | |
d1b19dff | 8826 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 8827 | |
6a643ddb CW |
8828 | /** |
8829 | * dev_change_tx_queue_len - Change TX queue length of a netdevice | |
8830 | * @dev: device | |
8831 | * @new_len: new tx queue length | |
8832 | */ | |
8833 | int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) | |
8834 | { | |
8835 | unsigned int orig_len = dev->tx_queue_len; | |
8836 | int res; | |
8837 | ||
8838 | if (new_len != (unsigned int)new_len) | |
8839 | return -ERANGE; | |
8840 | ||
8841 | if (new_len != orig_len) { | |
8842 | dev->tx_queue_len = new_len; | |
8843 | res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); | |
8844 | res = notifier_to_errno(res); | |
7effaf06 TT |
8845 | if (res) |
8846 | goto err_rollback; | |
8847 | res = dev_qdisc_change_tx_queue_len(dev); | |
8848 | if (res) | |
8849 | goto err_rollback; | |
6a643ddb CW |
8850 | } |
8851 | ||
8852 | return 0; | |
7effaf06 TT |
8853 | |
8854 | err_rollback: | |
8855 | netdev_err(dev, "refused to change device tx_queue_len\n"); | |
8856 | dev->tx_queue_len = orig_len; | |
8857 | return res; | |
6a643ddb CW |
8858 | } |
8859 | ||
cbda10fa VD |
8860 | /** |
8861 | * dev_set_group - Change group this device belongs to | |
8862 | * @dev: device | |
8863 | * @new_group: group this device should belong to | |
8864 | */ | |
8865 | void dev_set_group(struct net_device *dev, int new_group) | |
8866 | { | |
8867 | dev->group = new_group; | |
8868 | } | |
8869 | EXPORT_SYMBOL(dev_set_group); | |
8870 | ||
d59cdf94 PM |
8871 | /** |
8872 | * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. | |
8873 | * @dev: device | |
8874 | * @addr: new address | |
8875 | * @extack: netlink extended ack | |
8876 | */ | |
8877 | int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, | |
8878 | struct netlink_ext_ack *extack) | |
8879 | { | |
8880 | struct netdev_notifier_pre_changeaddr_info info = { | |
8881 | .info.dev = dev, | |
8882 | .info.extack = extack, | |
8883 | .dev_addr = addr, | |
8884 | }; | |
8885 | int rc; | |
8886 | ||
8887 | rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); | |
8888 | return notifier_to_errno(rc); | |
8889 | } | |
8890 | EXPORT_SYMBOL(dev_pre_changeaddr_notify); | |
8891 | ||
f0db275a SH |
8892 | /** |
8893 | * dev_set_mac_address - Change Media Access Control Address | |
8894 | * @dev: device | |
8895 | * @sa: new address | |
3a37a963 | 8896 | * @extack: netlink extended ack |
f0db275a SH |
8897 | * |
8898 | * Change the hardware (MAC) address of the device | |
8899 | */ | |
3a37a963 PM |
8900 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, |
8901 | struct netlink_ext_ack *extack) | |
1da177e4 | 8902 | { |
d314774c | 8903 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
8904 | int err; |
8905 | ||
d314774c | 8906 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
8907 | return -EOPNOTSUPP; |
8908 | if (sa->sa_family != dev->type) | |
8909 | return -EINVAL; | |
8910 | if (!netif_device_present(dev)) | |
8911 | return -ENODEV; | |
d59cdf94 PM |
8912 | err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); |
8913 | if (err) | |
8914 | return err; | |
d314774c | 8915 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
8916 | if (err) |
8917 | return err; | |
fbdeca2d | 8918 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 8919 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 8920 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 8921 | return 0; |
1da177e4 | 8922 | } |
d1b19dff | 8923 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 8924 | |
3b23a32a CW |
8925 | static DECLARE_RWSEM(dev_addr_sem); |
8926 | ||
8927 | int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, | |
8928 | struct netlink_ext_ack *extack) | |
8929 | { | |
8930 | int ret; | |
8931 | ||
8932 | down_write(&dev_addr_sem); | |
8933 | ret = dev_set_mac_address(dev, sa, extack); | |
8934 | up_write(&dev_addr_sem); | |
8935 | return ret; | |
8936 | } | |
8937 | EXPORT_SYMBOL(dev_set_mac_address_user); | |
8938 | ||
8939 | int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) | |
8940 | { | |
8941 | size_t size = sizeof(sa->sa_data); | |
8942 | struct net_device *dev; | |
8943 | int ret = 0; | |
8944 | ||
8945 | down_read(&dev_addr_sem); | |
8946 | rcu_read_lock(); | |
8947 | ||
8948 | dev = dev_get_by_name_rcu(net, dev_name); | |
8949 | if (!dev) { | |
8950 | ret = -ENODEV; | |
8951 | goto unlock; | |
8952 | } | |
8953 | if (!dev->addr_len) | |
8954 | memset(sa->sa_data, 0, size); | |
8955 | else | |
8956 | memcpy(sa->sa_data, dev->dev_addr, | |
8957 | min_t(size_t, size, dev->addr_len)); | |
8958 | sa->sa_family = dev->type; | |
8959 | ||
8960 | unlock: | |
8961 | rcu_read_unlock(); | |
8962 | up_read(&dev_addr_sem); | |
8963 | return ret; | |
8964 | } | |
8965 | EXPORT_SYMBOL(dev_get_mac_address); | |
8966 | ||
4bf84c35 JP |
8967 | /** |
8968 | * dev_change_carrier - Change device carrier | |
8969 | * @dev: device | |
691b3b7e | 8970 | * @new_carrier: new value |
4bf84c35 JP |
8971 | * |
8972 | * Change device carrier | |
8973 | */ | |
8974 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
8975 | { | |
8976 | const struct net_device_ops *ops = dev->netdev_ops; | |
8977 | ||
8978 | if (!ops->ndo_change_carrier) | |
8979 | return -EOPNOTSUPP; | |
8980 | if (!netif_device_present(dev)) | |
8981 | return -ENODEV; | |
8982 | return ops->ndo_change_carrier(dev, new_carrier); | |
8983 | } | |
8984 | EXPORT_SYMBOL(dev_change_carrier); | |
8985 | ||
66b52b0d JP |
8986 | /** |
8987 | * dev_get_phys_port_id - Get device physical port ID | |
8988 | * @dev: device | |
8989 | * @ppid: port ID | |
8990 | * | |
8991 | * Get device physical port ID | |
8992 | */ | |
8993 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 8994 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
8995 | { |
8996 | const struct net_device_ops *ops = dev->netdev_ops; | |
8997 | ||
8998 | if (!ops->ndo_get_phys_port_id) | |
8999 | return -EOPNOTSUPP; | |
9000 | return ops->ndo_get_phys_port_id(dev, ppid); | |
9001 | } | |
9002 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
9003 | ||
db24a904 DA |
9004 | /** |
9005 | * dev_get_phys_port_name - Get device physical port name | |
9006 | * @dev: device | |
9007 | * @name: port name | |
ed49e650 | 9008 | * @len: limit of bytes to copy to name |
db24a904 DA |
9009 | * |
9010 | * Get device physical port name | |
9011 | */ | |
9012 | int dev_get_phys_port_name(struct net_device *dev, | |
9013 | char *name, size_t len) | |
9014 | { | |
9015 | const struct net_device_ops *ops = dev->netdev_ops; | |
af3836df | 9016 | int err; |
db24a904 | 9017 | |
af3836df JP |
9018 | if (ops->ndo_get_phys_port_name) { |
9019 | err = ops->ndo_get_phys_port_name(dev, name, len); | |
9020 | if (err != -EOPNOTSUPP) | |
9021 | return err; | |
9022 | } | |
9023 | return devlink_compat_phys_port_name_get(dev, name, len); | |
db24a904 DA |
9024 | } |
9025 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
9026 | ||
d6abc596 FF |
9027 | /** |
9028 | * dev_get_port_parent_id - Get the device's port parent identifier | |
9029 | * @dev: network device | |
9030 | * @ppid: pointer to a storage for the port's parent identifier | |
9031 | * @recurse: allow/disallow recursion to lower devices | |
9032 | * | |
9033 | * Get the devices's port parent identifier | |
9034 | */ | |
9035 | int dev_get_port_parent_id(struct net_device *dev, | |
9036 | struct netdev_phys_item_id *ppid, | |
9037 | bool recurse) | |
9038 | { | |
9039 | const struct net_device_ops *ops = dev->netdev_ops; | |
9040 | struct netdev_phys_item_id first = { }; | |
9041 | struct net_device *lower_dev; | |
9042 | struct list_head *iter; | |
7e1146e8 JP |
9043 | int err; |
9044 | ||
9045 | if (ops->ndo_get_port_parent_id) { | |
9046 | err = ops->ndo_get_port_parent_id(dev, ppid); | |
9047 | if (err != -EOPNOTSUPP) | |
9048 | return err; | |
9049 | } | |
d6abc596 | 9050 | |
7e1146e8 JP |
9051 | err = devlink_compat_switch_id_get(dev, ppid); |
9052 | if (!err || err != -EOPNOTSUPP) | |
9053 | return err; | |
d6abc596 FF |
9054 | |
9055 | if (!recurse) | |
7e1146e8 | 9056 | return -EOPNOTSUPP; |
d6abc596 FF |
9057 | |
9058 | netdev_for_each_lower_dev(dev, lower_dev, iter) { | |
9059 | err = dev_get_port_parent_id(lower_dev, ppid, recurse); | |
9060 | if (err) | |
9061 | break; | |
9062 | if (!first.id_len) | |
9063 | first = *ppid; | |
9064 | else if (memcmp(&first, ppid, sizeof(*ppid))) | |
e1b9efe6 | 9065 | return -EOPNOTSUPP; |
d6abc596 FF |
9066 | } |
9067 | ||
9068 | return err; | |
9069 | } | |
9070 | EXPORT_SYMBOL(dev_get_port_parent_id); | |
9071 | ||
9072 | /** | |
9073 | * netdev_port_same_parent_id - Indicate if two network devices have | |
9074 | * the same port parent identifier | |
9075 | * @a: first network device | |
9076 | * @b: second network device | |
9077 | */ | |
9078 | bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) | |
9079 | { | |
9080 | struct netdev_phys_item_id a_id = { }; | |
9081 | struct netdev_phys_item_id b_id = { }; | |
9082 | ||
9083 | if (dev_get_port_parent_id(a, &a_id, true) || | |
9084 | dev_get_port_parent_id(b, &b_id, true)) | |
9085 | return false; | |
9086 | ||
9087 | return netdev_phys_item_id_same(&a_id, &b_id); | |
9088 | } | |
9089 | EXPORT_SYMBOL(netdev_port_same_parent_id); | |
9090 | ||
d746d707 AK |
9091 | /** |
9092 | * dev_change_proto_down - update protocol port state information | |
9093 | * @dev: device | |
9094 | * @proto_down: new value | |
9095 | * | |
9096 | * This info can be used by switch drivers to set the phys state of the | |
9097 | * port. | |
9098 | */ | |
9099 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
9100 | { | |
9101 | const struct net_device_ops *ops = dev->netdev_ops; | |
9102 | ||
9103 | if (!ops->ndo_change_proto_down) | |
9104 | return -EOPNOTSUPP; | |
9105 | if (!netif_device_present(dev)) | |
9106 | return -ENODEV; | |
9107 | return ops->ndo_change_proto_down(dev, proto_down); | |
9108 | } | |
9109 | EXPORT_SYMBOL(dev_change_proto_down); | |
9110 | ||
b5899679 AR |
9111 | /** |
9112 | * dev_change_proto_down_generic - generic implementation for | |
9113 | * ndo_change_proto_down that sets carrier according to | |
9114 | * proto_down. | |
9115 | * | |
9116 | * @dev: device | |
9117 | * @proto_down: new value | |
9118 | */ | |
9119 | int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) | |
9120 | { | |
9121 | if (proto_down) | |
9122 | netif_carrier_off(dev); | |
9123 | else | |
9124 | netif_carrier_on(dev); | |
9125 | dev->proto_down = proto_down; | |
9126 | return 0; | |
9127 | } | |
9128 | EXPORT_SYMBOL(dev_change_proto_down_generic); | |
9129 | ||
829eb208 RP |
9130 | /** |
9131 | * dev_change_proto_down_reason - proto down reason | |
9132 | * | |
9133 | * @dev: device | |
9134 | * @mask: proto down mask | |
9135 | * @value: proto down value | |
9136 | */ | |
9137 | void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, | |
9138 | u32 value) | |
9139 | { | |
9140 | int b; | |
9141 | ||
9142 | if (!mask) { | |
9143 | dev->proto_down_reason = value; | |
9144 | } else { | |
9145 | for_each_set_bit(b, &mask, 32) { | |
9146 | if (value & (1 << b)) | |
9147 | dev->proto_down_reason |= BIT(b); | |
9148 | else | |
9149 | dev->proto_down_reason &= ~BIT(b); | |
9150 | } | |
9151 | } | |
9152 | } | |
9153 | EXPORT_SYMBOL(dev_change_proto_down_reason); | |
9154 | ||
aa8d3a71 AN |
9155 | struct bpf_xdp_link { |
9156 | struct bpf_link link; | |
9157 | struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ | |
9158 | int flags; | |
9159 | }; | |
9160 | ||
c8a36f19 | 9161 | static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) |
d67b9cd2 | 9162 | { |
7f0a8382 AN |
9163 | if (flags & XDP_FLAGS_HW_MODE) |
9164 | return XDP_MODE_HW; | |
9165 | if (flags & XDP_FLAGS_DRV_MODE) | |
9166 | return XDP_MODE_DRV; | |
c8a36f19 AN |
9167 | if (flags & XDP_FLAGS_SKB_MODE) |
9168 | return XDP_MODE_SKB; | |
9169 | return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; | |
7f0a8382 | 9170 | } |
d67b9cd2 | 9171 | |
7f0a8382 AN |
9172 | static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) |
9173 | { | |
9174 | switch (mode) { | |
9175 | case XDP_MODE_SKB: | |
9176 | return generic_xdp_install; | |
9177 | case XDP_MODE_DRV: | |
9178 | case XDP_MODE_HW: | |
9179 | return dev->netdev_ops->ndo_bpf; | |
9180 | default: | |
9181 | return NULL; | |
5d867245 | 9182 | } |
7f0a8382 | 9183 | } |
118b4aa2 | 9184 | |
aa8d3a71 AN |
9185 | static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, |
9186 | enum bpf_xdp_mode mode) | |
9187 | { | |
9188 | return dev->xdp_state[mode].link; | |
9189 | } | |
9190 | ||
7f0a8382 AN |
9191 | static struct bpf_prog *dev_xdp_prog(struct net_device *dev, |
9192 | enum bpf_xdp_mode mode) | |
9193 | { | |
aa8d3a71 AN |
9194 | struct bpf_xdp_link *link = dev_xdp_link(dev, mode); |
9195 | ||
9196 | if (link) | |
9197 | return link->link.prog; | |
7f0a8382 AN |
9198 | return dev->xdp_state[mode].prog; |
9199 | } | |
9200 | ||
998f1729 THJ |
9201 | static u8 dev_xdp_prog_count(struct net_device *dev) |
9202 | { | |
9203 | u8 count = 0; | |
9204 | int i; | |
9205 | ||
9206 | for (i = 0; i < __MAX_XDP_MODE; i++) | |
9207 | if (dev->xdp_state[i].prog || dev->xdp_state[i].link) | |
9208 | count++; | |
9209 | return count; | |
9210 | } | |
9211 | ||
7f0a8382 AN |
9212 | u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) |
9213 | { | |
9214 | struct bpf_prog *prog = dev_xdp_prog(dev, mode); | |
118b4aa2 | 9215 | |
7f0a8382 AN |
9216 | return prog ? prog->aux->id : 0; |
9217 | } | |
58038695 | 9218 | |
aa8d3a71 AN |
9219 | static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, |
9220 | struct bpf_xdp_link *link) | |
9221 | { | |
9222 | dev->xdp_state[mode].link = link; | |
9223 | dev->xdp_state[mode].prog = NULL; | |
d67b9cd2 DB |
9224 | } |
9225 | ||
7f0a8382 AN |
9226 | static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, |
9227 | struct bpf_prog *prog) | |
9228 | { | |
aa8d3a71 | 9229 | dev->xdp_state[mode].link = NULL; |
7f0a8382 | 9230 | dev->xdp_state[mode].prog = prog; |
d67b9cd2 DB |
9231 | } |
9232 | ||
7f0a8382 AN |
9233 | static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, |
9234 | bpf_op_t bpf_op, struct netlink_ext_ack *extack, | |
9235 | u32 flags, struct bpf_prog *prog) | |
d67b9cd2 | 9236 | { |
f4e63525 | 9237 | struct netdev_bpf xdp; |
7e6897f9 BT |
9238 | int err; |
9239 | ||
d67b9cd2 | 9240 | memset(&xdp, 0, sizeof(xdp)); |
7f0a8382 | 9241 | xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; |
d67b9cd2 | 9242 | xdp.extack = extack; |
32d60277 | 9243 | xdp.flags = flags; |
d67b9cd2 DB |
9244 | xdp.prog = prog; |
9245 | ||
7f0a8382 AN |
9246 | /* Drivers assume refcnt is already incremented (i.e, prog pointer is |
9247 | * "moved" into driver), so they don't increment it on their own, but | |
9248 | * they do decrement refcnt when program is detached or replaced. | |
9249 | * Given net_device also owns link/prog, we need to bump refcnt here | |
9250 | * to prevent drivers from underflowing it. | |
9251 | */ | |
9252 | if (prog) | |
9253 | bpf_prog_inc(prog); | |
7e6897f9 | 9254 | err = bpf_op(dev, &xdp); |
7f0a8382 AN |
9255 | if (err) { |
9256 | if (prog) | |
9257 | bpf_prog_put(prog); | |
9258 | return err; | |
9259 | } | |
7e6897f9 | 9260 | |
7f0a8382 AN |
9261 | if (mode != XDP_MODE_HW) |
9262 | bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); | |
7e6897f9 | 9263 | |
7f0a8382 | 9264 | return 0; |
d67b9cd2 DB |
9265 | } |
9266 | ||
bd0b2e7f JK |
9267 | static void dev_xdp_uninstall(struct net_device *dev) |
9268 | { | |
aa8d3a71 | 9269 | struct bpf_xdp_link *link; |
7f0a8382 AN |
9270 | struct bpf_prog *prog; |
9271 | enum bpf_xdp_mode mode; | |
9272 | bpf_op_t bpf_op; | |
bd0b2e7f | 9273 | |
7f0a8382 | 9274 | ASSERT_RTNL(); |
bd0b2e7f | 9275 | |
7f0a8382 AN |
9276 | for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { |
9277 | prog = dev_xdp_prog(dev, mode); | |
9278 | if (!prog) | |
9279 | continue; | |
bd0b2e7f | 9280 | |
7f0a8382 AN |
9281 | bpf_op = dev_xdp_bpf_op(dev, mode); |
9282 | if (!bpf_op) | |
9283 | continue; | |
bd0b2e7f | 9284 | |
7f0a8382 AN |
9285 | WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); |
9286 | ||
aa8d3a71 AN |
9287 | /* auto-detach link from net device */ |
9288 | link = dev_xdp_link(dev, mode); | |
9289 | if (link) | |
9290 | link->dev = NULL; | |
9291 | else | |
9292 | bpf_prog_put(prog); | |
9293 | ||
9294 | dev_xdp_set_link(dev, mode, NULL); | |
7f0a8382 | 9295 | } |
bd0b2e7f JK |
9296 | } |
9297 | ||
d4baa936 | 9298 | static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, |
aa8d3a71 AN |
9299 | struct bpf_xdp_link *link, struct bpf_prog *new_prog, |
9300 | struct bpf_prog *old_prog, u32 flags) | |
a7862b45 | 9301 | { |
998f1729 | 9302 | unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); |
d4baa936 AN |
9303 | struct bpf_prog *cur_prog; |
9304 | enum bpf_xdp_mode mode; | |
7f0a8382 | 9305 | bpf_op_t bpf_op; |
a7862b45 BB |
9306 | int err; |
9307 | ||
85de8576 DB |
9308 | ASSERT_RTNL(); |
9309 | ||
aa8d3a71 AN |
9310 | /* either link or prog attachment, never both */ |
9311 | if (link && (new_prog || old_prog)) | |
9312 | return -EINVAL; | |
9313 | /* link supports only XDP mode flags */ | |
9314 | if (link && (flags & ~XDP_FLAGS_MODES)) { | |
9315 | NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); | |
9316 | return -EINVAL; | |
9317 | } | |
998f1729 THJ |
9318 | /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ |
9319 | if (num_modes > 1) { | |
d4baa936 AN |
9320 | NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); |
9321 | return -EINVAL; | |
9322 | } | |
998f1729 THJ |
9323 | /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ |
9324 | if (!num_modes && dev_xdp_prog_count(dev) > 1) { | |
9325 | NL_SET_ERR_MSG(extack, | |
9326 | "More than one program loaded, unset mode is ambiguous"); | |
9327 | return -EINVAL; | |
9328 | } | |
d4baa936 AN |
9329 | /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ |
9330 | if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { | |
9331 | NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); | |
9332 | return -EINVAL; | |
01dde20c | 9333 | } |
a25717d2 | 9334 | |
c8a36f19 | 9335 | mode = dev_xdp_mode(dev, flags); |
aa8d3a71 AN |
9336 | /* can't replace attached link */ |
9337 | if (dev_xdp_link(dev, mode)) { | |
9338 | NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); | |
9339 | return -EBUSY; | |
01dde20c | 9340 | } |
c14a9f63 | 9341 | |
d4baa936 | 9342 | cur_prog = dev_xdp_prog(dev, mode); |
aa8d3a71 AN |
9343 | /* can't replace attached prog with link */ |
9344 | if (link && cur_prog) { | |
9345 | NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); | |
9346 | return -EBUSY; | |
9347 | } | |
d4baa936 AN |
9348 | if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { |
9349 | NL_SET_ERR_MSG(extack, "Active program does not match expected"); | |
9350 | return -EEXIST; | |
92234c8f | 9351 | } |
c14a9f63 | 9352 | |
aa8d3a71 AN |
9353 | /* put effective new program into new_prog */ |
9354 | if (link) | |
9355 | new_prog = link->link.prog; | |
85de8576 | 9356 | |
d4baa936 AN |
9357 | if (new_prog) { |
9358 | bool offload = mode == XDP_MODE_HW; | |
7f0a8382 AN |
9359 | enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB |
9360 | ? XDP_MODE_DRV : XDP_MODE_SKB; | |
441a3303 | 9361 | |
068d9d1e AN |
9362 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { |
9363 | NL_SET_ERR_MSG(extack, "XDP program already attached"); | |
9364 | return -EBUSY; | |
9365 | } | |
d4baa936 | 9366 | if (!offload && dev_xdp_prog(dev, other_mode)) { |
7f0a8382 | 9367 | NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); |
d67b9cd2 | 9368 | return -EEXIST; |
01dde20c | 9369 | } |
d4baa936 | 9370 | if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { |
7f0a8382 | 9371 | NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); |
441a3303 JK |
9372 | return -EINVAL; |
9373 | } | |
d4baa936 | 9374 | if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { |
fbee97fe | 9375 | NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); |
fbee97fe DA |
9376 | return -EINVAL; |
9377 | } | |
d4baa936 AN |
9378 | if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { |
9379 | NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); | |
92164774 LB |
9380 | return -EINVAL; |
9381 | } | |
d4baa936 | 9382 | } |
92164774 | 9383 | |
d4baa936 AN |
9384 | /* don't call drivers if the effective program didn't change */ |
9385 | if (new_prog != cur_prog) { | |
9386 | bpf_op = dev_xdp_bpf_op(dev, mode); | |
9387 | if (!bpf_op) { | |
9388 | NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); | |
9389 | return -EOPNOTSUPP; | |
c14a9f63 | 9390 | } |
a7862b45 | 9391 | |
d4baa936 AN |
9392 | err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); |
9393 | if (err) | |
9394 | return err; | |
7f0a8382 | 9395 | } |
d4baa936 | 9396 | |
aa8d3a71 AN |
9397 | if (link) |
9398 | dev_xdp_set_link(dev, mode, link); | |
9399 | else | |
9400 | dev_xdp_set_prog(dev, mode, new_prog); | |
d4baa936 AN |
9401 | if (cur_prog) |
9402 | bpf_prog_put(cur_prog); | |
a7862b45 | 9403 | |
7f0a8382 | 9404 | return 0; |
a7862b45 | 9405 | } |
a7862b45 | 9406 | |
aa8d3a71 AN |
9407 | static int dev_xdp_attach_link(struct net_device *dev, |
9408 | struct netlink_ext_ack *extack, | |
9409 | struct bpf_xdp_link *link) | |
9410 | { | |
9411 | return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); | |
9412 | } | |
9413 | ||
9414 | static int dev_xdp_detach_link(struct net_device *dev, | |
9415 | struct netlink_ext_ack *extack, | |
9416 | struct bpf_xdp_link *link) | |
9417 | { | |
9418 | enum bpf_xdp_mode mode; | |
9419 | bpf_op_t bpf_op; | |
9420 | ||
9421 | ASSERT_RTNL(); | |
9422 | ||
c8a36f19 | 9423 | mode = dev_xdp_mode(dev, link->flags); |
aa8d3a71 AN |
9424 | if (dev_xdp_link(dev, mode) != link) |
9425 | return -EINVAL; | |
9426 | ||
9427 | bpf_op = dev_xdp_bpf_op(dev, mode); | |
9428 | WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); | |
9429 | dev_xdp_set_link(dev, mode, NULL); | |
9430 | return 0; | |
9431 | } | |
9432 | ||
9433 | static void bpf_xdp_link_release(struct bpf_link *link) | |
9434 | { | |
9435 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9436 | ||
9437 | rtnl_lock(); | |
9438 | ||
9439 | /* if racing with net_device's tear down, xdp_link->dev might be | |
9440 | * already NULL, in which case link was already auto-detached | |
9441 | */ | |
73b11c2a | 9442 | if (xdp_link->dev) { |
aa8d3a71 | 9443 | WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); |
73b11c2a AN |
9444 | xdp_link->dev = NULL; |
9445 | } | |
aa8d3a71 AN |
9446 | |
9447 | rtnl_unlock(); | |
9448 | } | |
9449 | ||
73b11c2a AN |
9450 | static int bpf_xdp_link_detach(struct bpf_link *link) |
9451 | { | |
9452 | bpf_xdp_link_release(link); | |
9453 | return 0; | |
9454 | } | |
9455 | ||
aa8d3a71 AN |
9456 | static void bpf_xdp_link_dealloc(struct bpf_link *link) |
9457 | { | |
9458 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9459 | ||
9460 | kfree(xdp_link); | |
9461 | } | |
9462 | ||
c1931c97 AN |
9463 | static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, |
9464 | struct seq_file *seq) | |
9465 | { | |
9466 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9467 | u32 ifindex = 0; | |
9468 | ||
9469 | rtnl_lock(); | |
9470 | if (xdp_link->dev) | |
9471 | ifindex = xdp_link->dev->ifindex; | |
9472 | rtnl_unlock(); | |
9473 | ||
9474 | seq_printf(seq, "ifindex:\t%u\n", ifindex); | |
9475 | } | |
9476 | ||
9477 | static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, | |
9478 | struct bpf_link_info *info) | |
9479 | { | |
9480 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9481 | u32 ifindex = 0; | |
9482 | ||
9483 | rtnl_lock(); | |
9484 | if (xdp_link->dev) | |
9485 | ifindex = xdp_link->dev->ifindex; | |
9486 | rtnl_unlock(); | |
9487 | ||
9488 | info->xdp.ifindex = ifindex; | |
9489 | return 0; | |
9490 | } | |
9491 | ||
026a4c28 AN |
9492 | static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, |
9493 | struct bpf_prog *old_prog) | |
9494 | { | |
9495 | struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); | |
9496 | enum bpf_xdp_mode mode; | |
9497 | bpf_op_t bpf_op; | |
9498 | int err = 0; | |
9499 | ||
9500 | rtnl_lock(); | |
9501 | ||
9502 | /* link might have been auto-released already, so fail */ | |
9503 | if (!xdp_link->dev) { | |
9504 | err = -ENOLINK; | |
9505 | goto out_unlock; | |
9506 | } | |
9507 | ||
9508 | if (old_prog && link->prog != old_prog) { | |
9509 | err = -EPERM; | |
9510 | goto out_unlock; | |
9511 | } | |
9512 | old_prog = link->prog; | |
9513 | if (old_prog == new_prog) { | |
9514 | /* no-op, don't disturb drivers */ | |
9515 | bpf_prog_put(new_prog); | |
9516 | goto out_unlock; | |
9517 | } | |
9518 | ||
c8a36f19 | 9519 | mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); |
026a4c28 AN |
9520 | bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); |
9521 | err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, | |
9522 | xdp_link->flags, new_prog); | |
9523 | if (err) | |
9524 | goto out_unlock; | |
9525 | ||
9526 | old_prog = xchg(&link->prog, new_prog); | |
9527 | bpf_prog_put(old_prog); | |
9528 | ||
9529 | out_unlock: | |
9530 | rtnl_unlock(); | |
9531 | return err; | |
9532 | } | |
9533 | ||
aa8d3a71 AN |
9534 | static const struct bpf_link_ops bpf_xdp_link_lops = { |
9535 | .release = bpf_xdp_link_release, | |
9536 | .dealloc = bpf_xdp_link_dealloc, | |
73b11c2a | 9537 | .detach = bpf_xdp_link_detach, |
c1931c97 AN |
9538 | .show_fdinfo = bpf_xdp_link_show_fdinfo, |
9539 | .fill_link_info = bpf_xdp_link_fill_link_info, | |
026a4c28 | 9540 | .update_prog = bpf_xdp_link_update, |
aa8d3a71 AN |
9541 | }; |
9542 | ||
9543 | int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) | |
9544 | { | |
9545 | struct net *net = current->nsproxy->net_ns; | |
9546 | struct bpf_link_primer link_primer; | |
9547 | struct bpf_xdp_link *link; | |
9548 | struct net_device *dev; | |
9549 | int err, fd; | |
9550 | ||
9551 | dev = dev_get_by_index(net, attr->link_create.target_ifindex); | |
9552 | if (!dev) | |
9553 | return -EINVAL; | |
9554 | ||
9555 | link = kzalloc(sizeof(*link), GFP_USER); | |
9556 | if (!link) { | |
9557 | err = -ENOMEM; | |
9558 | goto out_put_dev; | |
9559 | } | |
9560 | ||
9561 | bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); | |
9562 | link->dev = dev; | |
9563 | link->flags = attr->link_create.flags; | |
9564 | ||
9565 | err = bpf_link_prime(&link->link, &link_primer); | |
9566 | if (err) { | |
9567 | kfree(link); | |
9568 | goto out_put_dev; | |
9569 | } | |
9570 | ||
9571 | rtnl_lock(); | |
9572 | err = dev_xdp_attach_link(dev, NULL, link); | |
9573 | rtnl_unlock(); | |
9574 | ||
9575 | if (err) { | |
9576 | bpf_link_cleanup(&link_primer); | |
9577 | goto out_put_dev; | |
9578 | } | |
9579 | ||
9580 | fd = bpf_link_settle(&link_primer); | |
9581 | /* link itself doesn't hold dev's refcnt to not complicate shutdown */ | |
9582 | dev_put(dev); | |
9583 | return fd; | |
9584 | ||
9585 | out_put_dev: | |
9586 | dev_put(dev); | |
9587 | return err; | |
9588 | } | |
9589 | ||
d4baa936 AN |
9590 | /** |
9591 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
9592 | * @dev: device | |
9593 | * @extack: netlink extended ack | |
9594 | * @fd: new program fd or negative value to clear | |
9595 | * @expected_fd: old program fd that userspace expects to replace or clear | |
9596 | * @flags: xdp-related flags | |
9597 | * | |
9598 | * Set or clear a bpf program for a device | |
9599 | */ | |
9600 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, | |
9601 | int fd, int expected_fd, u32 flags) | |
9602 | { | |
c8a36f19 | 9603 | enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); |
d4baa936 AN |
9604 | struct bpf_prog *new_prog = NULL, *old_prog = NULL; |
9605 | int err; | |
9606 | ||
9607 | ASSERT_RTNL(); | |
9608 | ||
9609 | if (fd >= 0) { | |
9610 | new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, | |
9611 | mode != XDP_MODE_SKB); | |
9612 | if (IS_ERR(new_prog)) | |
9613 | return PTR_ERR(new_prog); | |
9614 | } | |
9615 | ||
9616 | if (expected_fd >= 0) { | |
9617 | old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, | |
9618 | mode != XDP_MODE_SKB); | |
9619 | if (IS_ERR(old_prog)) { | |
9620 | err = PTR_ERR(old_prog); | |
9621 | old_prog = NULL; | |
9622 | goto err_out; | |
c14a9f63 | 9623 | } |
a7862b45 BB |
9624 | } |
9625 | ||
aa8d3a71 | 9626 | err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); |
a7862b45 | 9627 | |
d4baa936 AN |
9628 | err_out: |
9629 | if (err && new_prog) | |
9630 | bpf_prog_put(new_prog); | |
9631 | if (old_prog) | |
9632 | bpf_prog_put(old_prog); | |
a7862b45 BB |
9633 | return err; |
9634 | } | |
a7862b45 | 9635 | |
1da177e4 LT |
9636 | /** |
9637 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 9638 | * @net: the applicable net namespace |
1da177e4 LT |
9639 | * |
9640 | * Returns a suitable unique value for a new device interface | |
9641 | * number. The caller must hold the rtnl semaphore or the | |
9642 | * dev_base_lock to be sure it remains unique. | |
9643 | */ | |
881d966b | 9644 | static int dev_new_index(struct net *net) |
1da177e4 | 9645 | { |
aa79e66e | 9646 | int ifindex = net->ifindex; |
f4563a75 | 9647 | |
1da177e4 LT |
9648 | for (;;) { |
9649 | if (++ifindex <= 0) | |
9650 | ifindex = 1; | |
881d966b | 9651 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 9652 | return net->ifindex = ifindex; |
1da177e4 LT |
9653 | } |
9654 | } | |
9655 | ||
1da177e4 | 9656 | /* Delayed registration/unregisteration */ |
3b5b34fd | 9657 | static LIST_HEAD(net_todo_list); |
200b916f | 9658 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 9659 | |
6f05f629 | 9660 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 9661 | { |
1da177e4 | 9662 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 9663 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
9664 | } |
9665 | ||
fd867d51 JW |
9666 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
9667 | struct net_device *upper, netdev_features_t features) | |
9668 | { | |
9669 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
9670 | netdev_features_t feature; | |
5ba3f7d6 | 9671 | int feature_bit; |
fd867d51 | 9672 | |
3b89ea9c | 9673 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 9674 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
9675 | if (!(upper->wanted_features & feature) |
9676 | && (features & feature)) { | |
9677 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
9678 | &feature, upper->name); | |
9679 | features &= ~feature; | |
9680 | } | |
9681 | } | |
9682 | ||
9683 | return features; | |
9684 | } | |
9685 | ||
9686 | static void netdev_sync_lower_features(struct net_device *upper, | |
9687 | struct net_device *lower, netdev_features_t features) | |
9688 | { | |
9689 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
9690 | netdev_features_t feature; | |
5ba3f7d6 | 9691 | int feature_bit; |
fd867d51 | 9692 | |
3b89ea9c | 9693 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 9694 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
9695 | if (!(features & feature) && (lower->features & feature)) { |
9696 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
9697 | &feature, lower->name); | |
9698 | lower->wanted_features &= ~feature; | |
dd912306 | 9699 | __netdev_update_features(lower); |
fd867d51 JW |
9700 | |
9701 | if (unlikely(lower->features & feature)) | |
9702 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
9703 | &feature, lower->name); | |
dd912306 CW |
9704 | else |
9705 | netdev_features_change(lower); | |
fd867d51 JW |
9706 | } |
9707 | } | |
9708 | } | |
9709 | ||
c8f44aff MM |
9710 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
9711 | netdev_features_t features) | |
b63365a2 | 9712 | { |
57422dc5 MM |
9713 | /* Fix illegal checksum combinations */ |
9714 | if ((features & NETIF_F_HW_CSUM) && | |
9715 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 9716 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
9717 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
9718 | } | |
9719 | ||
b63365a2 | 9720 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 9721 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 9722 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 9723 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
9724 | } |
9725 | ||
ec5f0615 PS |
9726 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
9727 | !(features & NETIF_F_IP_CSUM)) { | |
9728 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
9729 | features &= ~NETIF_F_TSO; | |
9730 | features &= ~NETIF_F_TSO_ECN; | |
9731 | } | |
9732 | ||
9733 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
9734 | !(features & NETIF_F_IPV6_CSUM)) { | |
9735 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
9736 | features &= ~NETIF_F_TSO6; | |
9737 | } | |
9738 | ||
b1dc497b AD |
9739 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
9740 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
9741 | features &= ~NETIF_F_TSO_MANGLEID; | |
9742 | ||
31d8b9e0 BH |
9743 | /* TSO ECN requires that TSO is present as well. */ |
9744 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
9745 | features &= ~NETIF_F_TSO_ECN; | |
9746 | ||
212b573f MM |
9747 | /* Software GSO depends on SG. */ |
9748 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 9749 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
9750 | features &= ~NETIF_F_GSO; |
9751 | } | |
9752 | ||
802ab55a AD |
9753 | /* GSO partial features require GSO partial be set */ |
9754 | if ((features & dev->gso_partial_features) && | |
9755 | !(features & NETIF_F_GSO_PARTIAL)) { | |
9756 | netdev_dbg(dev, | |
9757 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
9758 | features &= ~dev->gso_partial_features; | |
9759 | } | |
9760 | ||
fb1f5f79 MC |
9761 | if (!(features & NETIF_F_RXCSUM)) { |
9762 | /* NETIF_F_GRO_HW implies doing RXCSUM since every packet | |
9763 | * successfully merged by hardware must also have the | |
9764 | * checksum verified by hardware. If the user does not | |
9765 | * want to enable RXCSUM, logically, we should disable GRO_HW. | |
9766 | */ | |
9767 | if (features & NETIF_F_GRO_HW) { | |
9768 | netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); | |
9769 | features &= ~NETIF_F_GRO_HW; | |
9770 | } | |
9771 | } | |
9772 | ||
de8d5ab2 GP |
9773 | /* LRO/HW-GRO features cannot be combined with RX-FCS */ |
9774 | if (features & NETIF_F_RXFCS) { | |
9775 | if (features & NETIF_F_LRO) { | |
9776 | netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); | |
9777 | features &= ~NETIF_F_LRO; | |
9778 | } | |
9779 | ||
9780 | if (features & NETIF_F_GRO_HW) { | |
9781 | netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); | |
9782 | features &= ~NETIF_F_GRO_HW; | |
9783 | } | |
e6c6a929 GP |
9784 | } |
9785 | ||
25537d71 TT |
9786 | if (features & NETIF_F_HW_TLS_TX) { |
9787 | bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == | |
9788 | (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
9789 | bool hw_csum = features & NETIF_F_HW_CSUM; | |
9790 | ||
9791 | if (!ip_csum && !hw_csum) { | |
9792 | netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); | |
9793 | features &= ~NETIF_F_HW_TLS_TX; | |
9794 | } | |
ae0b04b2 TT |
9795 | } |
9796 | ||
a3eb4e9d TT |
9797 | if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { |
9798 | netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); | |
9799 | features &= ~NETIF_F_HW_TLS_RX; | |
9800 | } | |
9801 | ||
b63365a2 HX |
9802 | return features; |
9803 | } | |
b63365a2 | 9804 | |
6cb6a27c | 9805 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 9806 | { |
fd867d51 | 9807 | struct net_device *upper, *lower; |
c8f44aff | 9808 | netdev_features_t features; |
fd867d51 | 9809 | struct list_head *iter; |
e7868a85 | 9810 | int err = -1; |
5455c699 | 9811 | |
87267485 MM |
9812 | ASSERT_RTNL(); |
9813 | ||
5455c699 MM |
9814 | features = netdev_get_wanted_features(dev); |
9815 | ||
9816 | if (dev->netdev_ops->ndo_fix_features) | |
9817 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
9818 | ||
9819 | /* driver might be less strict about feature dependencies */ | |
9820 | features = netdev_fix_features(dev, features); | |
9821 | ||
4250b75b | 9822 | /* some features can't be enabled if they're off on an upper device */ |
fd867d51 JW |
9823 | netdev_for_each_upper_dev_rcu(dev, upper, iter) |
9824 | features = netdev_sync_upper_features(dev, upper, features); | |
9825 | ||
5455c699 | 9826 | if (dev->features == features) |
e7868a85 | 9827 | goto sync_lower; |
5455c699 | 9828 | |
c8f44aff MM |
9829 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
9830 | &dev->features, &features); | |
5455c699 MM |
9831 | |
9832 | if (dev->netdev_ops->ndo_set_features) | |
9833 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
9834 | else |
9835 | err = 0; | |
5455c699 | 9836 | |
6cb6a27c | 9837 | if (unlikely(err < 0)) { |
5455c699 | 9838 | netdev_err(dev, |
c8f44aff MM |
9839 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
9840 | err, &features, &dev->features); | |
17b85d29 NA |
9841 | /* return non-0 since some features might have changed and |
9842 | * it's better to fire a spurious notification than miss it | |
9843 | */ | |
9844 | return -1; | |
6cb6a27c MM |
9845 | } |
9846 | ||
e7868a85 | 9847 | sync_lower: |
fd867d51 JW |
9848 | /* some features must be disabled on lower devices when disabled |
9849 | * on an upper device (think: bonding master or bridge) | |
9850 | */ | |
9851 | netdev_for_each_lower_dev(dev, lower, iter) | |
9852 | netdev_sync_lower_features(dev, lower, features); | |
9853 | ||
ae847f40 SD |
9854 | if (!err) { |
9855 | netdev_features_t diff = features ^ dev->features; | |
9856 | ||
9857 | if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
9858 | /* udp_tunnel_{get,drop}_rx_info both need | |
9859 | * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the | |
9860 | * device, or they won't do anything. | |
9861 | * Thus we need to update dev->features | |
9862 | * *before* calling udp_tunnel_get_rx_info, | |
9863 | * but *after* calling udp_tunnel_drop_rx_info. | |
9864 | */ | |
9865 | if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
9866 | dev->features = features; | |
9867 | udp_tunnel_get_rx_info(dev); | |
9868 | } else { | |
9869 | udp_tunnel_drop_rx_info(dev); | |
9870 | } | |
9871 | } | |
9872 | ||
9daae9bd GP |
9873 | if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { |
9874 | if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { | |
9875 | dev->features = features; | |
9876 | err |= vlan_get_rx_ctag_filter_info(dev); | |
9877 | } else { | |
9878 | vlan_drop_rx_ctag_filter_info(dev); | |
9879 | } | |
9880 | } | |
9881 | ||
9882 | if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { | |
9883 | if (features & NETIF_F_HW_VLAN_STAG_FILTER) { | |
9884 | dev->features = features; | |
9885 | err |= vlan_get_rx_stag_filter_info(dev); | |
9886 | } else { | |
9887 | vlan_drop_rx_stag_filter_info(dev); | |
9888 | } | |
9889 | } | |
9890 | ||
6cb6a27c | 9891 | dev->features = features; |
ae847f40 | 9892 | } |
6cb6a27c | 9893 | |
e7868a85 | 9894 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
9895 | } |
9896 | ||
afe12cc8 MM |
9897 | /** |
9898 | * netdev_update_features - recalculate device features | |
9899 | * @dev: the device to check | |
9900 | * | |
9901 | * Recalculate dev->features set and send notifications if it | |
9902 | * has changed. Should be called after driver or hardware dependent | |
9903 | * conditions might have changed that influence the features. | |
9904 | */ | |
6cb6a27c MM |
9905 | void netdev_update_features(struct net_device *dev) |
9906 | { | |
9907 | if (__netdev_update_features(dev)) | |
9908 | netdev_features_change(dev); | |
5455c699 MM |
9909 | } |
9910 | EXPORT_SYMBOL(netdev_update_features); | |
9911 | ||
afe12cc8 MM |
9912 | /** |
9913 | * netdev_change_features - recalculate device features | |
9914 | * @dev: the device to check | |
9915 | * | |
9916 | * Recalculate dev->features set and send notifications even | |
9917 | * if they have not changed. Should be called instead of | |
9918 | * netdev_update_features() if also dev->vlan_features might | |
9919 | * have changed to allow the changes to be propagated to stacked | |
9920 | * VLAN devices. | |
9921 | */ | |
9922 | void netdev_change_features(struct net_device *dev) | |
9923 | { | |
9924 | __netdev_update_features(dev); | |
9925 | netdev_features_change(dev); | |
9926 | } | |
9927 | EXPORT_SYMBOL(netdev_change_features); | |
9928 | ||
fc4a7489 PM |
9929 | /** |
9930 | * netif_stacked_transfer_operstate - transfer operstate | |
9931 | * @rootdev: the root or lower level device to transfer state from | |
9932 | * @dev: the device to transfer operstate to | |
9933 | * | |
9934 | * Transfer operational state from root to device. This is normally | |
9935 | * called when a stacking relationship exists between the root | |
9936 | * device and the device(a leaf device). | |
9937 | */ | |
9938 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
9939 | struct net_device *dev) | |
9940 | { | |
9941 | if (rootdev->operstate == IF_OPER_DORMANT) | |
9942 | netif_dormant_on(dev); | |
9943 | else | |
9944 | netif_dormant_off(dev); | |
9945 | ||
eec517cd AL |
9946 | if (rootdev->operstate == IF_OPER_TESTING) |
9947 | netif_testing_on(dev); | |
9948 | else | |
9949 | netif_testing_off(dev); | |
9950 | ||
0575c86b ZS |
9951 | if (netif_carrier_ok(rootdev)) |
9952 | netif_carrier_on(dev); | |
9953 | else | |
9954 | netif_carrier_off(dev); | |
fc4a7489 PM |
9955 | } |
9956 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
9957 | ||
1b4bf461 ED |
9958 | static int netif_alloc_rx_queues(struct net_device *dev) |
9959 | { | |
1b4bf461 | 9960 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 9961 | struct netdev_rx_queue *rx; |
10595902 | 9962 | size_t sz = count * sizeof(*rx); |
e817f856 | 9963 | int err = 0; |
1b4bf461 | 9964 | |
bd25fa7b | 9965 | BUG_ON(count < 1); |
1b4bf461 | 9966 | |
dcda9b04 | 9967 | rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
9968 | if (!rx) |
9969 | return -ENOMEM; | |
9970 | ||
bd25fa7b TH |
9971 | dev->_rx = rx; |
9972 | ||
e817f856 | 9973 | for (i = 0; i < count; i++) { |
fe822240 | 9974 | rx[i].dev = dev; |
e817f856 JDB |
9975 | |
9976 | /* XDP RX-queue setup */ | |
b02e5a0e | 9977 | err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); |
e817f856 JDB |
9978 | if (err < 0) |
9979 | goto err_rxq_info; | |
9980 | } | |
1b4bf461 | 9981 | return 0; |
e817f856 JDB |
9982 | |
9983 | err_rxq_info: | |
9984 | /* Rollback successful reg's and free other resources */ | |
9985 | while (i--) | |
9986 | xdp_rxq_info_unreg(&rx[i].xdp_rxq); | |
141b52a9 | 9987 | kvfree(dev->_rx); |
e817f856 JDB |
9988 | dev->_rx = NULL; |
9989 | return err; | |
9990 | } | |
9991 | ||
9992 | static void netif_free_rx_queues(struct net_device *dev) | |
9993 | { | |
9994 | unsigned int i, count = dev->num_rx_queues; | |
e817f856 JDB |
9995 | |
9996 | /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ | |
9997 | if (!dev->_rx) | |
9998 | return; | |
9999 | ||
e817f856 | 10000 | for (i = 0; i < count; i++) |
82aaff2f JK |
10001 | xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); |
10002 | ||
10003 | kvfree(dev->_rx); | |
1b4bf461 ED |
10004 | } |
10005 | ||
aa942104 CG |
10006 | static void netdev_init_one_queue(struct net_device *dev, |
10007 | struct netdev_queue *queue, void *_unused) | |
10008 | { | |
10009 | /* Initialize queue lock */ | |
10010 | spin_lock_init(&queue->_xmit_lock); | |
1a33e10e | 10011 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); |
aa942104 | 10012 | queue->xmit_lock_owner = -1; |
b236da69 | 10013 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 10014 | queue->dev = dev; |
114cf580 TH |
10015 | #ifdef CONFIG_BQL |
10016 | dql_init(&queue->dql, HZ); | |
10017 | #endif | |
aa942104 CG |
10018 | } |
10019 | ||
60877a32 ED |
10020 | static void netif_free_tx_queues(struct net_device *dev) |
10021 | { | |
4cb28970 | 10022 | kvfree(dev->_tx); |
60877a32 ED |
10023 | } |
10024 | ||
e6484930 TH |
10025 | static int netif_alloc_netdev_queues(struct net_device *dev) |
10026 | { | |
10027 | unsigned int count = dev->num_tx_queues; | |
10028 | struct netdev_queue *tx; | |
60877a32 | 10029 | size_t sz = count * sizeof(*tx); |
e6484930 | 10030 | |
d339727c ED |
10031 | if (count < 1 || count > 0xffff) |
10032 | return -EINVAL; | |
62b5942a | 10033 | |
dcda9b04 | 10034 | tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
10035 | if (!tx) |
10036 | return -ENOMEM; | |
10037 | ||
e6484930 | 10038 | dev->_tx = tx; |
1d24eb48 | 10039 | |
e6484930 TH |
10040 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
10041 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
10042 | |
10043 | return 0; | |
e6484930 TH |
10044 | } |
10045 | ||
a2029240 DV |
10046 | void netif_tx_stop_all_queues(struct net_device *dev) |
10047 | { | |
10048 | unsigned int i; | |
10049 | ||
10050 | for (i = 0; i < dev->num_tx_queues; i++) { | |
10051 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 10052 | |
a2029240 DV |
10053 | netif_tx_stop_queue(txq); |
10054 | } | |
10055 | } | |
10056 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
10057 | ||
1da177e4 LT |
10058 | /** |
10059 | * register_netdevice - register a network device | |
10060 | * @dev: device to register | |
10061 | * | |
10062 | * Take a completed network device structure and add it to the kernel | |
10063 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
10064 | * chain. 0 is returned on success. A negative errno code is returned | |
10065 | * on a failure to set up the device, or if the name is a duplicate. | |
10066 | * | |
10067 | * Callers must hold the rtnl semaphore. You may want | |
10068 | * register_netdev() instead of this. | |
10069 | * | |
10070 | * BUGS: | |
10071 | * The locking appears insufficient to guarantee two parallel registers | |
10072 | * will not get the same name. | |
10073 | */ | |
10074 | ||
10075 | int register_netdevice(struct net_device *dev) | |
10076 | { | |
1da177e4 | 10077 | int ret; |
d314774c | 10078 | struct net *net = dev_net(dev); |
1da177e4 | 10079 | |
e283de3a FF |
10080 | BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < |
10081 | NETDEV_FEATURE_COUNT); | |
1da177e4 LT |
10082 | BUG_ON(dev_boot_phase); |
10083 | ASSERT_RTNL(); | |
10084 | ||
b17a7c17 SH |
10085 | might_sleep(); |
10086 | ||
1da177e4 LT |
10087 | /* When net_device's are persistent, this will be fatal. */ |
10088 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 10089 | BUG_ON(!net); |
1da177e4 | 10090 | |
9000edb7 JK |
10091 | ret = ethtool_check_ops(dev->ethtool_ops); |
10092 | if (ret) | |
10093 | return ret; | |
10094 | ||
f1f28aa3 | 10095 | spin_lock_init(&dev->addr_list_lock); |
845e0ebb | 10096 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 10097 | |
828de4f6 | 10098 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
10099 | if (ret < 0) |
10100 | goto out; | |
10101 | ||
9077f052 | 10102 | ret = -ENOMEM; |
ff927412 JP |
10103 | dev->name_node = netdev_name_node_head_alloc(dev); |
10104 | if (!dev->name_node) | |
10105 | goto out; | |
10106 | ||
1da177e4 | 10107 | /* Init, if this function is available */ |
d314774c SH |
10108 | if (dev->netdev_ops->ndo_init) { |
10109 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
10110 | if (ret) { |
10111 | if (ret > 0) | |
10112 | ret = -EIO; | |
42c17fa6 | 10113 | goto err_free_name; |
1da177e4 LT |
10114 | } |
10115 | } | |
4ec93edb | 10116 | |
f646968f PM |
10117 | if (((dev->hw_features | dev->features) & |
10118 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
10119 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
10120 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
10121 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
10122 | ret = -EINVAL; | |
10123 | goto err_uninit; | |
10124 | } | |
10125 | ||
9c7dafbf PE |
10126 | ret = -EBUSY; |
10127 | if (!dev->ifindex) | |
10128 | dev->ifindex = dev_new_index(net); | |
10129 | else if (__dev_get_by_index(net, dev->ifindex)) | |
10130 | goto err_uninit; | |
10131 | ||
5455c699 MM |
10132 | /* Transfer changeable features to wanted_features and enable |
10133 | * software offloads (GSO and GRO). | |
10134 | */ | |
1a3c998f | 10135 | dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); |
14d1232f | 10136 | dev->features |= NETIF_F_SOFT_FEATURES; |
d764a122 | 10137 | |
876c4384 | 10138 | if (dev->udp_tunnel_nic_info) { |
d764a122 SD |
10139 | dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; |
10140 | dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
10141 | } | |
10142 | ||
14d1232f | 10143 | dev->wanted_features = dev->features & dev->hw_features; |
1da177e4 | 10144 | |
cbc53e08 | 10145 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 10146 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 10147 | |
7f348a60 AD |
10148 | /* If IPv4 TCP segmentation offload is supported we should also |
10149 | * allow the device to enable segmenting the frame with the option | |
10150 | * of ignoring a static IP ID value. This doesn't enable the | |
10151 | * feature itself but allows the user to enable it later. | |
10152 | */ | |
cbc53e08 AD |
10153 | if (dev->hw_features & NETIF_F_TSO) |
10154 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
10155 | if (dev->vlan_features & NETIF_F_TSO) |
10156 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
10157 | if (dev->mpls_features & NETIF_F_TSO) | |
10158 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
10159 | if (dev->hw_enc_features & NETIF_F_TSO) | |
10160 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 10161 | |
1180e7d6 | 10162 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 10163 | */ |
1180e7d6 | 10164 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 10165 | |
ee579677 PS |
10166 | /* Make NETIF_F_SG inheritable to tunnel devices. |
10167 | */ | |
802ab55a | 10168 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 10169 | |
0d89d203 SH |
10170 | /* Make NETIF_F_SG inheritable to MPLS. |
10171 | */ | |
10172 | dev->mpls_features |= NETIF_F_SG; | |
10173 | ||
7ffbe3fd JB |
10174 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
10175 | ret = notifier_to_errno(ret); | |
10176 | if (ret) | |
10177 | goto err_uninit; | |
10178 | ||
8b41d188 | 10179 | ret = netdev_register_kobject(dev); |
cb626bf5 JH |
10180 | if (ret) { |
10181 | dev->reg_state = NETREG_UNREGISTERED; | |
7ce1b0ed | 10182 | goto err_uninit; |
cb626bf5 | 10183 | } |
b17a7c17 SH |
10184 | dev->reg_state = NETREG_REGISTERED; |
10185 | ||
6cb6a27c | 10186 | __netdev_update_features(dev); |
8e9b59b2 | 10187 | |
1da177e4 LT |
10188 | /* |
10189 | * Default initial state at registry is that the | |
10190 | * device is present. | |
10191 | */ | |
10192 | ||
10193 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
10194 | ||
8f4cccbb BH |
10195 | linkwatch_init_dev(dev); |
10196 | ||
1da177e4 | 10197 | dev_init_scheduler(dev); |
1da177e4 | 10198 | dev_hold(dev); |
ce286d32 | 10199 | list_netdevice(dev); |
7bf23575 | 10200 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 10201 | |
948b337e JP |
10202 | /* If the device has permanent device address, driver should |
10203 | * set dev_addr and also addr_assign_type should be set to | |
10204 | * NET_ADDR_PERM (default value). | |
10205 | */ | |
10206 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
10207 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
10208 | ||
1da177e4 | 10209 | /* Notify protocols, that a new device appeared. */ |
056925ab | 10210 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 10211 | ret = notifier_to_errno(ret); |
93ee31f1 | 10212 | if (ret) { |
766b0515 JK |
10213 | /* Expect explicit free_netdev() on failure */ |
10214 | dev->needs_free_netdev = false; | |
037e56bd | 10215 | unregister_netdevice_queue(dev, NULL); |
766b0515 | 10216 | goto out; |
93ee31f1 | 10217 | } |
d90a909e EB |
10218 | /* |
10219 | * Prevent userspace races by waiting until the network | |
10220 | * device is fully setup before sending notifications. | |
10221 | */ | |
a2835763 PM |
10222 | if (!dev->rtnl_link_ops || |
10223 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 10224 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
10225 | |
10226 | out: | |
10227 | return ret; | |
7ce1b0ed HX |
10228 | |
10229 | err_uninit: | |
d314774c SH |
10230 | if (dev->netdev_ops->ndo_uninit) |
10231 | dev->netdev_ops->ndo_uninit(dev); | |
cf124db5 DM |
10232 | if (dev->priv_destructor) |
10233 | dev->priv_destructor(dev); | |
42c17fa6 DC |
10234 | err_free_name: |
10235 | netdev_name_node_free(dev->name_node); | |
7ce1b0ed | 10236 | goto out; |
1da177e4 | 10237 | } |
d1b19dff | 10238 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 10239 | |
937f1ba5 BH |
10240 | /** |
10241 | * init_dummy_netdev - init a dummy network device for NAPI | |
10242 | * @dev: device to init | |
10243 | * | |
10244 | * This takes a network device structure and initialize the minimum | |
10245 | * amount of fields so it can be used to schedule NAPI polls without | |
10246 | * registering a full blown interface. This is to be used by drivers | |
10247 | * that need to tie several hardware interfaces to a single NAPI | |
10248 | * poll scheduler due to HW limitations. | |
10249 | */ | |
10250 | int init_dummy_netdev(struct net_device *dev) | |
10251 | { | |
10252 | /* Clear everything. Note we don't initialize spinlocks | |
10253 | * are they aren't supposed to be taken by any of the | |
10254 | * NAPI code and this dummy netdev is supposed to be | |
10255 | * only ever used for NAPI polls | |
10256 | */ | |
10257 | memset(dev, 0, sizeof(struct net_device)); | |
10258 | ||
10259 | /* make sure we BUG if trying to hit standard | |
10260 | * register/unregister code path | |
10261 | */ | |
10262 | dev->reg_state = NETREG_DUMMY; | |
10263 | ||
937f1ba5 BH |
10264 | /* NAPI wants this */ |
10265 | INIT_LIST_HEAD(&dev->napi_list); | |
10266 | ||
10267 | /* a dummy interface is started by default */ | |
10268 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
10269 | set_bit(__LINK_STATE_START, &dev->state); | |
10270 | ||
35edfdc7 JE |
10271 | /* napi_busy_loop stats accounting wants this */ |
10272 | dev_net_set(dev, &init_net); | |
10273 | ||
29b4433d ED |
10274 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
10275 | * because users of this 'device' dont need to change | |
10276 | * its refcount. | |
10277 | */ | |
10278 | ||
937f1ba5 BH |
10279 | return 0; |
10280 | } | |
10281 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
10282 | ||
10283 | ||
1da177e4 LT |
10284 | /** |
10285 | * register_netdev - register a network device | |
10286 | * @dev: device to register | |
10287 | * | |
10288 | * Take a completed network device structure and add it to the kernel | |
10289 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
10290 | * chain. 0 is returned on success. A negative errno code is returned | |
10291 | * on a failure to set up the device, or if the name is a duplicate. | |
10292 | * | |
38b4da38 | 10293 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
10294 | * and expands the device name if you passed a format string to |
10295 | * alloc_netdev. | |
10296 | */ | |
10297 | int register_netdev(struct net_device *dev) | |
10298 | { | |
10299 | int err; | |
10300 | ||
b0f3debc KT |
10301 | if (rtnl_lock_killable()) |
10302 | return -EINTR; | |
1da177e4 | 10303 | err = register_netdevice(dev); |
1da177e4 LT |
10304 | rtnl_unlock(); |
10305 | return err; | |
10306 | } | |
10307 | EXPORT_SYMBOL(register_netdev); | |
10308 | ||
29b4433d ED |
10309 | int netdev_refcnt_read(const struct net_device *dev) |
10310 | { | |
10311 | int i, refcnt = 0; | |
10312 | ||
10313 | for_each_possible_cpu(i) | |
10314 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
10315 | return refcnt; | |
10316 | } | |
10317 | EXPORT_SYMBOL(netdev_refcnt_read); | |
10318 | ||
de2b541b MCC |
10319 | #define WAIT_REFS_MIN_MSECS 1 |
10320 | #define WAIT_REFS_MAX_MSECS 250 | |
2c53040f | 10321 | /** |
1da177e4 | 10322 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 10323 | * @dev: target net_device |
1da177e4 LT |
10324 | * |
10325 | * This is called when unregistering network devices. | |
10326 | * | |
10327 | * Any protocol or device that holds a reference should register | |
10328 | * for netdevice notification, and cleanup and put back the | |
10329 | * reference if they receive an UNREGISTER event. | |
10330 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 10331 | * call dev_put. |
1da177e4 LT |
10332 | */ |
10333 | static void netdev_wait_allrefs(struct net_device *dev) | |
10334 | { | |
10335 | unsigned long rebroadcast_time, warning_time; | |
0e4be9e5 | 10336 | int wait = 0, refcnt; |
1da177e4 | 10337 | |
e014debe ED |
10338 | linkwatch_forget_dev(dev); |
10339 | ||
1da177e4 | 10340 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
10341 | refcnt = netdev_refcnt_read(dev); |
10342 | ||
10343 | while (refcnt != 0) { | |
1da177e4 | 10344 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 10345 | rtnl_lock(); |
1da177e4 LT |
10346 | |
10347 | /* Rebroadcast unregister notification */ | |
056925ab | 10348 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 10349 | |
748e2d93 | 10350 | __rtnl_unlock(); |
0115e8e3 | 10351 | rcu_barrier(); |
748e2d93 ED |
10352 | rtnl_lock(); |
10353 | ||
1da177e4 LT |
10354 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
10355 | &dev->state)) { | |
10356 | /* We must not have linkwatch events | |
10357 | * pending on unregister. If this | |
10358 | * happens, we simply run the queue | |
10359 | * unscheduled, resulting in a noop | |
10360 | * for this device. | |
10361 | */ | |
10362 | linkwatch_run_queue(); | |
10363 | } | |
10364 | ||
6756ae4b | 10365 | __rtnl_unlock(); |
1da177e4 LT |
10366 | |
10367 | rebroadcast_time = jiffies; | |
10368 | } | |
10369 | ||
0e4be9e5 FR |
10370 | if (!wait) { |
10371 | rcu_barrier(); | |
10372 | wait = WAIT_REFS_MIN_MSECS; | |
10373 | } else { | |
10374 | msleep(wait); | |
10375 | wait = min(wait << 1, WAIT_REFS_MAX_MSECS); | |
10376 | } | |
1da177e4 | 10377 | |
29b4433d ED |
10378 | refcnt = netdev_refcnt_read(dev); |
10379 | ||
d7c04b05 | 10380 | if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
10381 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
10382 | dev->name, refcnt); | |
1da177e4 LT |
10383 | warning_time = jiffies; |
10384 | } | |
10385 | } | |
10386 | } | |
10387 | ||
10388 | /* The sequence is: | |
10389 | * | |
10390 | * rtnl_lock(); | |
10391 | * ... | |
10392 | * register_netdevice(x1); | |
10393 | * register_netdevice(x2); | |
10394 | * ... | |
10395 | * unregister_netdevice(y1); | |
10396 | * unregister_netdevice(y2); | |
10397 | * ... | |
10398 | * rtnl_unlock(); | |
10399 | * free_netdev(y1); | |
10400 | * free_netdev(y2); | |
10401 | * | |
58ec3b4d | 10402 | * We are invoked by rtnl_unlock(). |
1da177e4 | 10403 | * This allows us to deal with problems: |
b17a7c17 | 10404 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
10405 | * without deadlocking with linkwatch via keventd. |
10406 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
10407 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
10408 | * |
10409 | * We must not return until all unregister events added during | |
10410 | * the interval the lock was held have been completed. | |
1da177e4 | 10411 | */ |
1da177e4 LT |
10412 | void netdev_run_todo(void) |
10413 | { | |
626ab0e6 | 10414 | struct list_head list; |
1fc70edb TY |
10415 | #ifdef CONFIG_LOCKDEP |
10416 | struct list_head unlink_list; | |
10417 | ||
10418 | list_replace_init(&net_unlink_list, &unlink_list); | |
10419 | ||
10420 | while (!list_empty(&unlink_list)) { | |
10421 | struct net_device *dev = list_first_entry(&unlink_list, | |
10422 | struct net_device, | |
10423 | unlink_list); | |
0e8b8d6a | 10424 | list_del_init(&dev->unlink_list); |
1fc70edb TY |
10425 | dev->nested_level = dev->lower_level - 1; |
10426 | } | |
10427 | #endif | |
1da177e4 | 10428 | |
1da177e4 | 10429 | /* Snapshot list, allow later requests */ |
626ab0e6 | 10430 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
10431 | |
10432 | __rtnl_unlock(); | |
626ab0e6 | 10433 | |
0115e8e3 ED |
10434 | |
10435 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
10436 | if (!list_empty(&list)) |
10437 | rcu_barrier(); | |
10438 | ||
1da177e4 LT |
10439 | while (!list_empty(&list)) { |
10440 | struct net_device *dev | |
e5e26d75 | 10441 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
10442 | list_del(&dev->todo_list); |
10443 | ||
b17a7c17 | 10444 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 10445 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
10446 | dev->name, dev->reg_state); |
10447 | dump_stack(); | |
10448 | continue; | |
10449 | } | |
1da177e4 | 10450 | |
b17a7c17 | 10451 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 10452 | |
b17a7c17 | 10453 | netdev_wait_allrefs(dev); |
1da177e4 | 10454 | |
b17a7c17 | 10455 | /* paranoia */ |
29b4433d | 10456 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
10457 | BUG_ON(!list_empty(&dev->ptype_all)); |
10458 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
10459 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
10460 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
330c7272 | 10461 | #if IS_ENABLED(CONFIG_DECNET) |
547b792c | 10462 | WARN_ON(dev->dn_ptr); |
330c7272 | 10463 | #endif |
cf124db5 DM |
10464 | if (dev->priv_destructor) |
10465 | dev->priv_destructor(dev); | |
10466 | if (dev->needs_free_netdev) | |
10467 | free_netdev(dev); | |
9093bbb2 | 10468 | |
50624c93 EB |
10469 | /* Report a network device has been unregistered */ |
10470 | rtnl_lock(); | |
10471 | dev_net(dev)->dev_unreg_count--; | |
10472 | __rtnl_unlock(); | |
10473 | wake_up(&netdev_unregistering_wq); | |
10474 | ||
9093bbb2 SH |
10475 | /* Free network device */ |
10476 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 10477 | } |
1da177e4 LT |
10478 | } |
10479 | ||
9256645a JW |
10480 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
10481 | * all the same fields in the same order as net_device_stats, with only | |
10482 | * the type differing, but rtnl_link_stats64 may have additional fields | |
10483 | * at the end for newer counters. | |
3cfde79c | 10484 | */ |
77a1abf5 ED |
10485 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
10486 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
10487 | { |
10488 | #if BITS_PER_LONG == 64 | |
9256645a | 10489 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
9af9959e | 10490 | memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); |
9256645a JW |
10491 | /* zero out counters that only exist in rtnl_link_stats64 */ |
10492 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
10493 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 10494 | #else |
9256645a | 10495 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
10496 | const unsigned long *src = (const unsigned long *)netdev_stats; |
10497 | u64 *dst = (u64 *)stats64; | |
10498 | ||
9256645a | 10499 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
10500 | for (i = 0; i < n; i++) |
10501 | dst[i] = src[i]; | |
9256645a JW |
10502 | /* zero out counters that only exist in rtnl_link_stats64 */ |
10503 | memset((char *)stats64 + n * sizeof(u64), 0, | |
10504 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
10505 | #endif |
10506 | } | |
77a1abf5 | 10507 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 10508 | |
eeda3fd6 SH |
10509 | /** |
10510 | * dev_get_stats - get network device statistics | |
10511 | * @dev: device to get statistics from | |
28172739 | 10512 | * @storage: place to store stats |
eeda3fd6 | 10513 | * |
d7753516 BH |
10514 | * Get network statistics from device. Return @storage. |
10515 | * The device driver may provide its own method by setting | |
10516 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
10517 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 10518 | */ |
d7753516 BH |
10519 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
10520 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 10521 | { |
eeda3fd6 SH |
10522 | const struct net_device_ops *ops = dev->netdev_ops; |
10523 | ||
28172739 ED |
10524 | if (ops->ndo_get_stats64) { |
10525 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
10526 | ops->ndo_get_stats64(dev, storage); |
10527 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 10528 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
10529 | } else { |
10530 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 10531 | } |
6f64ec74 ED |
10532 | storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); |
10533 | storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); | |
10534 | storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); | |
28172739 | 10535 | return storage; |
c45d286e | 10536 | } |
eeda3fd6 | 10537 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 10538 | |
44fa32f0 HK |
10539 | /** |
10540 | * dev_fetch_sw_netstats - get per-cpu network device statistics | |
10541 | * @s: place to store stats | |
10542 | * @netstats: per-cpu network stats to read from | |
10543 | * | |
10544 | * Read per-cpu network statistics and populate the related fields in @s. | |
10545 | */ | |
10546 | void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, | |
10547 | const struct pcpu_sw_netstats __percpu *netstats) | |
10548 | { | |
10549 | int cpu; | |
10550 | ||
10551 | for_each_possible_cpu(cpu) { | |
10552 | const struct pcpu_sw_netstats *stats; | |
10553 | struct pcpu_sw_netstats tmp; | |
10554 | unsigned int start; | |
10555 | ||
10556 | stats = per_cpu_ptr(netstats, cpu); | |
10557 | do { | |
10558 | start = u64_stats_fetch_begin_irq(&stats->syncp); | |
10559 | tmp.rx_packets = stats->rx_packets; | |
10560 | tmp.rx_bytes = stats->rx_bytes; | |
10561 | tmp.tx_packets = stats->tx_packets; | |
10562 | tmp.tx_bytes = stats->tx_bytes; | |
10563 | } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); | |
10564 | ||
10565 | s->rx_packets += tmp.rx_packets; | |
10566 | s->rx_bytes += tmp.rx_bytes; | |
10567 | s->tx_packets += tmp.tx_packets; | |
10568 | s->tx_bytes += tmp.tx_bytes; | |
10569 | } | |
10570 | } | |
10571 | EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); | |
10572 | ||
a1839426 HK |
10573 | /** |
10574 | * dev_get_tstats64 - ndo_get_stats64 implementation | |
10575 | * @dev: device to get statistics from | |
10576 | * @s: place to store stats | |
10577 | * | |
10578 | * Populate @s from dev->stats and dev->tstats. Can be used as | |
10579 | * ndo_get_stats64() callback. | |
10580 | */ | |
10581 | void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) | |
10582 | { | |
10583 | netdev_stats_to_stats64(s, &dev->stats); | |
10584 | dev_fetch_sw_netstats(s, dev->tstats); | |
10585 | } | |
10586 | EXPORT_SYMBOL_GPL(dev_get_tstats64); | |
10587 | ||
24824a09 | 10588 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 10589 | { |
24824a09 | 10590 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 10591 | |
24824a09 ED |
10592 | #ifdef CONFIG_NET_CLS_ACT |
10593 | if (queue) | |
10594 | return queue; | |
10595 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
10596 | if (!queue) | |
10597 | return NULL; | |
10598 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 10599 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
10600 | queue->qdisc_sleeping = &noop_qdisc; |
10601 | rcu_assign_pointer(dev->ingress_queue, queue); | |
10602 | #endif | |
10603 | return queue; | |
bb949fbd DM |
10604 | } |
10605 | ||
2c60db03 ED |
10606 | static const struct ethtool_ops default_ethtool_ops; |
10607 | ||
d07d7507 SG |
10608 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
10609 | const struct ethtool_ops *ops) | |
10610 | { | |
10611 | if (dev->ethtool_ops == &default_ethtool_ops) | |
10612 | dev->ethtool_ops = ops; | |
10613 | } | |
10614 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
10615 | ||
74d332c1 ED |
10616 | void netdev_freemem(struct net_device *dev) |
10617 | { | |
10618 | char *addr = (char *)dev - dev->padded; | |
10619 | ||
4cb28970 | 10620 | kvfree(addr); |
74d332c1 ED |
10621 | } |
10622 | ||
1da177e4 | 10623 | /** |
722c9a0c | 10624 | * alloc_netdev_mqs - allocate network device |
10625 | * @sizeof_priv: size of private data to allocate space for | |
10626 | * @name: device name format string | |
10627 | * @name_assign_type: origin of device name | |
10628 | * @setup: callback to initialize device | |
10629 | * @txqs: the number of TX subqueues to allocate | |
10630 | * @rxqs: the number of RX subqueues to allocate | |
10631 | * | |
10632 | * Allocates a struct net_device with private data area for driver use | |
10633 | * and performs basic initialization. Also allocates subqueue structs | |
10634 | * for each queue on the device. | |
1da177e4 | 10635 | */ |
36909ea4 | 10636 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 10637 | unsigned char name_assign_type, |
36909ea4 TH |
10638 | void (*setup)(struct net_device *), |
10639 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 10640 | { |
1da177e4 | 10641 | struct net_device *dev; |
52a59bd5 | 10642 | unsigned int alloc_size; |
1ce8e7b5 | 10643 | struct net_device *p; |
1da177e4 | 10644 | |
b6fe17d6 SH |
10645 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
10646 | ||
36909ea4 | 10647 | if (txqs < 1) { |
7b6cd1ce | 10648 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
10649 | return NULL; |
10650 | } | |
10651 | ||
36909ea4 | 10652 | if (rxqs < 1) { |
7b6cd1ce | 10653 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
10654 | return NULL; |
10655 | } | |
36909ea4 | 10656 | |
fd2ea0a7 | 10657 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
10658 | if (sizeof_priv) { |
10659 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 10660 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
10661 | alloc_size += sizeof_priv; |
10662 | } | |
10663 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 10664 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 10665 | |
dcda9b04 | 10666 | p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
62b5942a | 10667 | if (!p) |
1da177e4 | 10668 | return NULL; |
1da177e4 | 10669 | |
1ce8e7b5 | 10670 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 10671 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 10672 | |
29b4433d ED |
10673 | dev->pcpu_refcnt = alloc_percpu(int); |
10674 | if (!dev->pcpu_refcnt) | |
74d332c1 | 10675 | goto free_dev; |
ab9c73cc | 10676 | |
ab9c73cc | 10677 | if (dev_addr_init(dev)) |
29b4433d | 10678 | goto free_pcpu; |
ab9c73cc | 10679 | |
22bedad3 | 10680 | dev_mc_init(dev); |
a748ee24 | 10681 | dev_uc_init(dev); |
ccffad25 | 10682 | |
c346dca1 | 10683 | dev_net_set(dev, &init_net); |
1da177e4 | 10684 | |
8d3bdbd5 | 10685 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 10686 | dev->gso_max_segs = GSO_MAX_SEGS; |
5343da4c TY |
10687 | dev->upper_level = 1; |
10688 | dev->lower_level = 1; | |
1fc70edb TY |
10689 | #ifdef CONFIG_LOCKDEP |
10690 | dev->nested_level = 0; | |
10691 | INIT_LIST_HEAD(&dev->unlink_list); | |
10692 | #endif | |
8d3bdbd5 | 10693 | |
8d3bdbd5 DM |
10694 | INIT_LIST_HEAD(&dev->napi_list); |
10695 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 10696 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 10697 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
10698 | INIT_LIST_HEAD(&dev->adj_list.upper); |
10699 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
10700 | INIT_LIST_HEAD(&dev->ptype_all); |
10701 | INIT_LIST_HEAD(&dev->ptype_specific); | |
93642e14 | 10702 | INIT_LIST_HEAD(&dev->net_notifier_list); |
59cc1f61 JK |
10703 | #ifdef CONFIG_NET_SCHED |
10704 | hash_init(dev->qdisc_hash); | |
10705 | #endif | |
02875878 | 10706 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
10707 | setup(dev); |
10708 | ||
a813104d | 10709 | if (!dev->tx_queue_len) { |
f84bb1ea | 10710 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 10711 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 10712 | } |
906470c1 | 10713 | |
36909ea4 TH |
10714 | dev->num_tx_queues = txqs; |
10715 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 10716 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 10717 | goto free_all; |
e8a0464c | 10718 | |
36909ea4 TH |
10719 | dev->num_rx_queues = rxqs; |
10720 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 10721 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 10722 | goto free_all; |
0a9627f2 | 10723 | |
1da177e4 | 10724 | strcpy(dev->name, name); |
c835a677 | 10725 | dev->name_assign_type = name_assign_type; |
cbda10fa | 10726 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
10727 | if (!dev->ethtool_ops) |
10728 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 | 10729 | |
357b6cc5 | 10730 | nf_hook_ingress_init(dev); |
e687ad60 | 10731 | |
1da177e4 | 10732 | return dev; |
ab9c73cc | 10733 | |
8d3bdbd5 DM |
10734 | free_all: |
10735 | free_netdev(dev); | |
10736 | return NULL; | |
10737 | ||
29b4433d ED |
10738 | free_pcpu: |
10739 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
10740 | free_dev: |
10741 | netdev_freemem(dev); | |
ab9c73cc | 10742 | return NULL; |
1da177e4 | 10743 | } |
36909ea4 | 10744 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
10745 | |
10746 | /** | |
722c9a0c | 10747 | * free_netdev - free network device |
10748 | * @dev: device | |
1da177e4 | 10749 | * |
722c9a0c | 10750 | * This function does the last stage of destroying an allocated device |
10751 | * interface. The reference to the device object is released. If this | |
10752 | * is the last reference then it will be freed.Must be called in process | |
10753 | * context. | |
1da177e4 LT |
10754 | */ |
10755 | void free_netdev(struct net_device *dev) | |
10756 | { | |
d565b0a1 HX |
10757 | struct napi_struct *p, *n; |
10758 | ||
93d05d4a | 10759 | might_sleep(); |
c269a24c JK |
10760 | |
10761 | /* When called immediately after register_netdevice() failed the unwind | |
10762 | * handling may still be dismantling the device. Handle that case by | |
10763 | * deferring the free. | |
10764 | */ | |
10765 | if (dev->reg_state == NETREG_UNREGISTERING) { | |
10766 | ASSERT_RTNL(); | |
10767 | dev->needs_free_netdev = true; | |
10768 | return; | |
10769 | } | |
10770 | ||
60877a32 | 10771 | netif_free_tx_queues(dev); |
e817f856 | 10772 | netif_free_rx_queues(dev); |
e8a0464c | 10773 | |
33d480ce | 10774 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 10775 | |
f001fde5 JP |
10776 | /* Flush device addresses */ |
10777 | dev_addr_flush(dev); | |
10778 | ||
d565b0a1 HX |
10779 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
10780 | netif_napi_del(p); | |
10781 | ||
29b4433d ED |
10782 | free_percpu(dev->pcpu_refcnt); |
10783 | dev->pcpu_refcnt = NULL; | |
75ccae62 THJ |
10784 | free_percpu(dev->xdp_bulkq); |
10785 | dev->xdp_bulkq = NULL; | |
29b4433d | 10786 | |
3041a069 | 10787 | /* Compatibility with error handling in drivers */ |
1da177e4 | 10788 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 10789 | netdev_freemem(dev); |
1da177e4 LT |
10790 | return; |
10791 | } | |
10792 | ||
10793 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
10794 | dev->reg_state = NETREG_RELEASED; | |
10795 | ||
43cb76d9 GKH |
10796 | /* will free via device release */ |
10797 | put_device(&dev->dev); | |
1da177e4 | 10798 | } |
d1b19dff | 10799 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 10800 | |
f0db275a SH |
10801 | /** |
10802 | * synchronize_net - Synchronize with packet receive processing | |
10803 | * | |
10804 | * Wait for packets currently being received to be done. | |
10805 | * Does not block later packets from starting. | |
10806 | */ | |
4ec93edb | 10807 | void synchronize_net(void) |
1da177e4 LT |
10808 | { |
10809 | might_sleep(); | |
be3fc413 ED |
10810 | if (rtnl_is_locked()) |
10811 | synchronize_rcu_expedited(); | |
10812 | else | |
10813 | synchronize_rcu(); | |
1da177e4 | 10814 | } |
d1b19dff | 10815 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
10816 | |
10817 | /** | |
44a0873d | 10818 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 10819 | * @dev: device |
44a0873d | 10820 | * @head: list |
6ebfbc06 | 10821 | * |
1da177e4 | 10822 | * This function shuts down a device interface and removes it |
d59b54b1 | 10823 | * from the kernel tables. |
44a0873d | 10824 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
10825 | * |
10826 | * Callers must hold the rtnl semaphore. You may want | |
10827 | * unregister_netdev() instead of this. | |
10828 | */ | |
10829 | ||
44a0873d | 10830 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 10831 | { |
a6620712 HX |
10832 | ASSERT_RTNL(); |
10833 | ||
44a0873d | 10834 | if (head) { |
9fdce099 | 10835 | list_move_tail(&dev->unreg_list, head); |
44a0873d | 10836 | } else { |
037e56bd JK |
10837 | LIST_HEAD(single); |
10838 | ||
10839 | list_add(&dev->unreg_list, &single); | |
0cbe1e57 | 10840 | unregister_netdevice_many(&single); |
44a0873d | 10841 | } |
1da177e4 | 10842 | } |
44a0873d | 10843 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 10844 | |
9b5e383c ED |
10845 | /** |
10846 | * unregister_netdevice_many - unregister many devices | |
10847 | * @head: list of devices | |
87757a91 ED |
10848 | * |
10849 | * Note: As most callers use a stack allocated list_head, | |
10850 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
10851 | */ |
10852 | void unregister_netdevice_many(struct list_head *head) | |
bcfe2f1a JK |
10853 | { |
10854 | struct net_device *dev, *tmp; | |
10855 | LIST_HEAD(close_head); | |
10856 | ||
10857 | BUG_ON(dev_boot_phase); | |
10858 | ASSERT_RTNL(); | |
10859 | ||
0cbe1e57 JK |
10860 | if (list_empty(head)) |
10861 | return; | |
10862 | ||
bcfe2f1a JK |
10863 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
10864 | /* Some devices call without registering | |
10865 | * for initialization unwind. Remove those | |
10866 | * devices and proceed with the remaining. | |
10867 | */ | |
10868 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
10869 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", | |
10870 | dev->name, dev); | |
10871 | ||
10872 | WARN_ON(1); | |
10873 | list_del(&dev->unreg_list); | |
10874 | continue; | |
10875 | } | |
10876 | dev->dismantle = true; | |
10877 | BUG_ON(dev->reg_state != NETREG_REGISTERED); | |
10878 | } | |
10879 | ||
10880 | /* If device is running, close it first. */ | |
10881 | list_for_each_entry(dev, head, unreg_list) | |
10882 | list_add_tail(&dev->close_list, &close_head); | |
10883 | dev_close_many(&close_head, true); | |
10884 | ||
10885 | list_for_each_entry(dev, head, unreg_list) { | |
10886 | /* And unlink it from device chain. */ | |
10887 | unlist_netdevice(dev); | |
10888 | ||
10889 | dev->reg_state = NETREG_UNREGISTERING; | |
10890 | } | |
10891 | flush_all_backlogs(); | |
10892 | ||
10893 | synchronize_net(); | |
10894 | ||
10895 | list_for_each_entry(dev, head, unreg_list) { | |
10896 | struct sk_buff *skb = NULL; | |
10897 | ||
10898 | /* Shutdown queueing discipline. */ | |
10899 | dev_shutdown(dev); | |
10900 | ||
10901 | dev_xdp_uninstall(dev); | |
10902 | ||
10903 | /* Notify protocols, that we are about to destroy | |
10904 | * this device. They should clean all the things. | |
10905 | */ | |
10906 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
10907 | ||
10908 | if (!dev->rtnl_link_ops || | |
10909 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
10910 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, | |
10911 | GFP_KERNEL, NULL, 0); | |
10912 | ||
10913 | /* | |
10914 | * Flush the unicast and multicast chains | |
10915 | */ | |
10916 | dev_uc_flush(dev); | |
10917 | dev_mc_flush(dev); | |
10918 | ||
10919 | netdev_name_node_alt_flush(dev); | |
10920 | netdev_name_node_free(dev->name_node); | |
10921 | ||
10922 | if (dev->netdev_ops->ndo_uninit) | |
10923 | dev->netdev_ops->ndo_uninit(dev); | |
10924 | ||
10925 | if (skb) | |
10926 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
10927 | ||
10928 | /* Notifier chain MUST detach us all upper devices. */ | |
10929 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
10930 | WARN_ON(netdev_has_any_lower_dev(dev)); | |
10931 | ||
10932 | /* Remove entries from kobject tree */ | |
10933 | netdev_unregister_kobject(dev); | |
10934 | #ifdef CONFIG_XPS | |
10935 | /* Remove XPS queueing entries */ | |
10936 | netif_reset_xps_queues_gt(dev, 0); | |
10937 | #endif | |
10938 | } | |
10939 | ||
10940 | synchronize_net(); | |
10941 | ||
10942 | list_for_each_entry(dev, head, unreg_list) { | |
10943 | dev_put(dev); | |
10944 | net_set_todo(dev); | |
10945 | } | |
0cbe1e57 JK |
10946 | |
10947 | list_del(head); | |
bcfe2f1a | 10948 | } |
0cbe1e57 | 10949 | EXPORT_SYMBOL(unregister_netdevice_many); |
bcfe2f1a | 10950 | |
1da177e4 LT |
10951 | /** |
10952 | * unregister_netdev - remove device from the kernel | |
10953 | * @dev: device | |
10954 | * | |
10955 | * This function shuts down a device interface and removes it | |
d59b54b1 | 10956 | * from the kernel tables. |
1da177e4 LT |
10957 | * |
10958 | * This is just a wrapper for unregister_netdevice that takes | |
10959 | * the rtnl semaphore. In general you want to use this and not | |
10960 | * unregister_netdevice. | |
10961 | */ | |
10962 | void unregister_netdev(struct net_device *dev) | |
10963 | { | |
10964 | rtnl_lock(); | |
10965 | unregister_netdevice(dev); | |
10966 | rtnl_unlock(); | |
10967 | } | |
1da177e4 LT |
10968 | EXPORT_SYMBOL(unregister_netdev); |
10969 | ||
ce286d32 EB |
10970 | /** |
10971 | * dev_change_net_namespace - move device to different nethost namespace | |
10972 | * @dev: device | |
10973 | * @net: network namespace | |
10974 | * @pat: If not NULL name pattern to try if the current device name | |
10975 | * is already taken in the destination network namespace. | |
10976 | * | |
10977 | * This function shuts down a device interface and moves it | |
10978 | * to a new network namespace. On success 0 is returned, on | |
10979 | * a failure a netagive errno code is returned. | |
10980 | * | |
10981 | * Callers must hold the rtnl semaphore. | |
10982 | */ | |
10983 | ||
10984 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
10985 | { | |
ef6a4c88 | 10986 | struct net *net_old = dev_net(dev); |
38e01b30 | 10987 | int err, new_nsid, new_ifindex; |
ce286d32 EB |
10988 | |
10989 | ASSERT_RTNL(); | |
10990 | ||
10991 | /* Don't allow namespace local devices to be moved. */ | |
10992 | err = -EINVAL; | |
10993 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
10994 | goto out; | |
10995 | ||
10996 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
10997 | if (dev->reg_state != NETREG_REGISTERED) |
10998 | goto out; | |
10999 | ||
11000 | /* Get out if there is nothing todo */ | |
11001 | err = 0; | |
ef6a4c88 | 11002 | if (net_eq(net_old, net)) |
ce286d32 EB |
11003 | goto out; |
11004 | ||
11005 | /* Pick the destination device name, and ensure | |
11006 | * we can use it in the destination network namespace. | |
11007 | */ | |
11008 | err = -EEXIST; | |
d9031024 | 11009 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
11010 | /* We get here if we can't use the current device name */ |
11011 | if (!pat) | |
11012 | goto out; | |
7892bd08 LR |
11013 | err = dev_get_valid_name(net, dev, pat); |
11014 | if (err < 0) | |
ce286d32 EB |
11015 | goto out; |
11016 | } | |
11017 | ||
11018 | /* | |
11019 | * And now a mini version of register_netdevice unregister_netdevice. | |
11020 | */ | |
11021 | ||
11022 | /* If device is running close it first. */ | |
9b772652 | 11023 | dev_close(dev); |
ce286d32 EB |
11024 | |
11025 | /* And unlink it from device chain */ | |
ce286d32 EB |
11026 | unlist_netdevice(dev); |
11027 | ||
11028 | synchronize_net(); | |
11029 | ||
11030 | /* Shutdown queueing discipline. */ | |
11031 | dev_shutdown(dev); | |
11032 | ||
11033 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 11034 | * this device. They should clean all the things. |
11035 | * | |
11036 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
11037 | * This is wanted because this way 8021q and macvlan know | |
11038 | * the device is just moving and can keep their slaves up. | |
11039 | */ | |
ce286d32 | 11040 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 | 11041 | rcu_barrier(); |
38e01b30 | 11042 | |
d4e4fdf9 | 11043 | new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); |
38e01b30 ND |
11044 | /* If there is an ifindex conflict assign a new one */ |
11045 | if (__dev_get_by_index(net, dev->ifindex)) | |
11046 | new_ifindex = dev_new_index(net); | |
11047 | else | |
11048 | new_ifindex = dev->ifindex; | |
11049 | ||
11050 | rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, | |
11051 | new_ifindex); | |
ce286d32 EB |
11052 | |
11053 | /* | |
11054 | * Flush the unicast and multicast chains | |
11055 | */ | |
a748ee24 | 11056 | dev_uc_flush(dev); |
22bedad3 | 11057 | dev_mc_flush(dev); |
ce286d32 | 11058 | |
4e66ae2e SH |
11059 | /* Send a netdev-removed uevent to the old namespace */ |
11060 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 11061 | netdev_adjacent_del_links(dev); |
4e66ae2e | 11062 | |
93642e14 JP |
11063 | /* Move per-net netdevice notifiers that are following the netdevice */ |
11064 | move_netdevice_notifiers_dev_net(dev, net); | |
11065 | ||
ce286d32 | 11066 | /* Actually switch the network namespace */ |
c346dca1 | 11067 | dev_net_set(dev, net); |
38e01b30 | 11068 | dev->ifindex = new_ifindex; |
ce286d32 | 11069 | |
4e66ae2e SH |
11070 | /* Send a netdev-add uevent to the new namespace */ |
11071 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 11072 | netdev_adjacent_add_links(dev); |
4e66ae2e | 11073 | |
8b41d188 | 11074 | /* Fixup kobjects */ |
a1b3f594 | 11075 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 11076 | WARN_ON(err); |
ce286d32 | 11077 | |
ef6a4c88 CB |
11078 | /* Adapt owner in case owning user namespace of target network |
11079 | * namespace is different from the original one. | |
11080 | */ | |
11081 | err = netdev_change_owner(dev, net_old, net); | |
11082 | WARN_ON(err); | |
11083 | ||
ce286d32 EB |
11084 | /* Add the device back in the hashes */ |
11085 | list_netdevice(dev); | |
11086 | ||
11087 | /* Notify protocols, that a new device appeared. */ | |
11088 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
11089 | ||
d90a909e EB |
11090 | /* |
11091 | * Prevent userspace races by waiting until the network | |
11092 | * device is fully setup before sending notifications. | |
11093 | */ | |
7f294054 | 11094 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 11095 | |
ce286d32 EB |
11096 | synchronize_net(); |
11097 | err = 0; | |
11098 | out: | |
11099 | return err; | |
11100 | } | |
463d0183 | 11101 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 11102 | |
f0bf90de | 11103 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
11104 | { |
11105 | struct sk_buff **list_skb; | |
1da177e4 | 11106 | struct sk_buff *skb; |
f0bf90de | 11107 | unsigned int cpu; |
97d8b6e3 | 11108 | struct softnet_data *sd, *oldsd, *remsd = NULL; |
1da177e4 | 11109 | |
1da177e4 LT |
11110 | local_irq_disable(); |
11111 | cpu = smp_processor_id(); | |
11112 | sd = &per_cpu(softnet_data, cpu); | |
11113 | oldsd = &per_cpu(softnet_data, oldcpu); | |
11114 | ||
11115 | /* Find end of our completion_queue. */ | |
11116 | list_skb = &sd->completion_queue; | |
11117 | while (*list_skb) | |
11118 | list_skb = &(*list_skb)->next; | |
11119 | /* Append completion queue from offline CPU. */ | |
11120 | *list_skb = oldsd->completion_queue; | |
11121 | oldsd->completion_queue = NULL; | |
11122 | ||
1da177e4 | 11123 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
11124 | if (oldsd->output_queue) { |
11125 | *sd->output_queue_tailp = oldsd->output_queue; | |
11126 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
11127 | oldsd->output_queue = NULL; | |
11128 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
11129 | } | |
ac64da0b ED |
11130 | /* Append NAPI poll list from offline CPU, with one exception : |
11131 | * process_backlog() must be called by cpu owning percpu backlog. | |
11132 | * We properly handle process_queue & input_pkt_queue later. | |
11133 | */ | |
11134 | while (!list_empty(&oldsd->poll_list)) { | |
11135 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
11136 | struct napi_struct, | |
11137 | poll_list); | |
11138 | ||
11139 | list_del_init(&napi->poll_list); | |
11140 | if (napi->poll == process_backlog) | |
11141 | napi->state = 0; | |
11142 | else | |
11143 | ____napi_schedule(sd, napi); | |
264524d5 | 11144 | } |
1da177e4 LT |
11145 | |
11146 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
11147 | local_irq_enable(); | |
11148 | ||
773fc8f6 | 11149 | #ifdef CONFIG_RPS |
11150 | remsd = oldsd->rps_ipi_list; | |
11151 | oldsd->rps_ipi_list = NULL; | |
11152 | #endif | |
11153 | /* send out pending IPI's on offline CPU */ | |
11154 | net_rps_send_ipi(remsd); | |
11155 | ||
1da177e4 | 11156 | /* Process offline CPU's input_pkt_queue */ |
76cc8b13 | 11157 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 11158 | netif_rx_ni(skb); |
76cc8b13 | 11159 | input_queue_head_incr(oldsd); |
fec5e652 | 11160 | } |
ac64da0b | 11161 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 11162 | netif_rx_ni(skb); |
76cc8b13 TH |
11163 | input_queue_head_incr(oldsd); |
11164 | } | |
1da177e4 | 11165 | |
f0bf90de | 11166 | return 0; |
1da177e4 | 11167 | } |
1da177e4 | 11168 | |
7f353bf2 | 11169 | /** |
b63365a2 HX |
11170 | * netdev_increment_features - increment feature set by one |
11171 | * @all: current feature set | |
11172 | * @one: new feature set | |
11173 | * @mask: mask feature set | |
7f353bf2 HX |
11174 | * |
11175 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
11176 | * @one to the master device with current feature set @all. Will not |
11177 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 11178 | */ |
c8f44aff MM |
11179 | netdev_features_t netdev_increment_features(netdev_features_t all, |
11180 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 11181 | { |
c8cd0989 | 11182 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 11183 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 11184 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 11185 | |
a188222b | 11186 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 11187 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 11188 | |
1742f183 | 11189 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
11190 | if (all & NETIF_F_HW_CSUM) |
11191 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
11192 | |
11193 | return all; | |
11194 | } | |
b63365a2 | 11195 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 11196 | |
430f03cd | 11197 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
11198 | { |
11199 | int i; | |
11200 | struct hlist_head *hash; | |
11201 | ||
6da2ec56 | 11202 | hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); |
30d97d35 PE |
11203 | if (hash != NULL) |
11204 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
11205 | INIT_HLIST_HEAD(&hash[i]); | |
11206 | ||
11207 | return hash; | |
11208 | } | |
11209 | ||
881d966b | 11210 | /* Initialize per network namespace state */ |
4665079c | 11211 | static int __net_init netdev_init(struct net *net) |
881d966b | 11212 | { |
d9f37d01 | 11213 | BUILD_BUG_ON(GRO_HASH_BUCKETS > |
c593642c | 11214 | 8 * sizeof_field(struct napi_struct, gro_bitmask)); |
d9f37d01 | 11215 | |
734b6541 RM |
11216 | if (net != &init_net) |
11217 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 11218 | |
30d97d35 PE |
11219 | net->dev_name_head = netdev_create_hash(); |
11220 | if (net->dev_name_head == NULL) | |
11221 | goto err_name; | |
881d966b | 11222 | |
30d97d35 PE |
11223 | net->dev_index_head = netdev_create_hash(); |
11224 | if (net->dev_index_head == NULL) | |
11225 | goto err_idx; | |
881d966b | 11226 | |
a30c7b42 JP |
11227 | RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); |
11228 | ||
881d966b | 11229 | return 0; |
30d97d35 PE |
11230 | |
11231 | err_idx: | |
11232 | kfree(net->dev_name_head); | |
11233 | err_name: | |
11234 | return -ENOMEM; | |
881d966b EB |
11235 | } |
11236 | ||
f0db275a SH |
11237 | /** |
11238 | * netdev_drivername - network driver for the device | |
11239 | * @dev: network device | |
f0db275a SH |
11240 | * |
11241 | * Determine network driver for device. | |
11242 | */ | |
3019de12 | 11243 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 11244 | { |
cf04a4c7 SH |
11245 | const struct device_driver *driver; |
11246 | const struct device *parent; | |
3019de12 | 11247 | const char *empty = ""; |
6579e57b AV |
11248 | |
11249 | parent = dev->dev.parent; | |
6579e57b | 11250 | if (!parent) |
3019de12 | 11251 | return empty; |
6579e57b AV |
11252 | |
11253 | driver = parent->driver; | |
11254 | if (driver && driver->name) | |
3019de12 DM |
11255 | return driver->name; |
11256 | return empty; | |
6579e57b AV |
11257 | } |
11258 | ||
6ea754eb JP |
11259 | static void __netdev_printk(const char *level, const struct net_device *dev, |
11260 | struct va_format *vaf) | |
256df2f3 | 11261 | { |
b004ff49 | 11262 | if (dev && dev->dev.parent) { |
6ea754eb JP |
11263 | dev_printk_emit(level[1] - '0', |
11264 | dev->dev.parent, | |
11265 | "%s %s %s%s: %pV", | |
11266 | dev_driver_string(dev->dev.parent), | |
11267 | dev_name(dev->dev.parent), | |
11268 | netdev_name(dev), netdev_reg_state(dev), | |
11269 | vaf); | |
b004ff49 | 11270 | } else if (dev) { |
6ea754eb JP |
11271 | printk("%s%s%s: %pV", |
11272 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 11273 | } else { |
6ea754eb | 11274 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 11275 | } |
256df2f3 JP |
11276 | } |
11277 | ||
6ea754eb JP |
11278 | void netdev_printk(const char *level, const struct net_device *dev, |
11279 | const char *format, ...) | |
256df2f3 JP |
11280 | { |
11281 | struct va_format vaf; | |
11282 | va_list args; | |
256df2f3 JP |
11283 | |
11284 | va_start(args, format); | |
11285 | ||
11286 | vaf.fmt = format; | |
11287 | vaf.va = &args; | |
11288 | ||
6ea754eb | 11289 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 11290 | |
256df2f3 | 11291 | va_end(args); |
256df2f3 JP |
11292 | } |
11293 | EXPORT_SYMBOL(netdev_printk); | |
11294 | ||
11295 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 11296 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 11297 | { \ |
256df2f3 JP |
11298 | struct va_format vaf; \ |
11299 | va_list args; \ | |
11300 | \ | |
11301 | va_start(args, fmt); \ | |
11302 | \ | |
11303 | vaf.fmt = fmt; \ | |
11304 | vaf.va = &args; \ | |
11305 | \ | |
6ea754eb | 11306 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 11307 | \ |
256df2f3 | 11308 | va_end(args); \ |
256df2f3 JP |
11309 | } \ |
11310 | EXPORT_SYMBOL(func); | |
11311 | ||
11312 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
11313 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
11314 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
11315 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
11316 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
11317 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
11318 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
11319 | ||
4665079c | 11320 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
11321 | { |
11322 | kfree(net->dev_name_head); | |
11323 | kfree(net->dev_index_head); | |
ee21b18b VA |
11324 | if (net != &init_net) |
11325 | WARN_ON_ONCE(!list_empty(&net->dev_base_head)); | |
881d966b EB |
11326 | } |
11327 | ||
022cbae6 | 11328 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
11329 | .init = netdev_init, |
11330 | .exit = netdev_exit, | |
11331 | }; | |
11332 | ||
4665079c | 11333 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 11334 | { |
e008b5fc | 11335 | struct net_device *dev, *aux; |
ce286d32 | 11336 | /* |
e008b5fc | 11337 | * Push all migratable network devices back to the |
ce286d32 EB |
11338 | * initial network namespace |
11339 | */ | |
11340 | rtnl_lock(); | |
e008b5fc | 11341 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 11342 | int err; |
aca51397 | 11343 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
11344 | |
11345 | /* Ignore unmoveable devices (i.e. loopback) */ | |
11346 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
11347 | continue; | |
11348 | ||
e008b5fc EB |
11349 | /* Leave virtual devices for the generic cleanup */ |
11350 | if (dev->rtnl_link_ops) | |
11351 | continue; | |
d0c082ce | 11352 | |
25985edc | 11353 | /* Push remaining network devices to init_net */ |
aca51397 | 11354 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
55b40dbf JP |
11355 | if (__dev_get_by_name(&init_net, fb_name)) |
11356 | snprintf(fb_name, IFNAMSIZ, "dev%%d"); | |
aca51397 | 11357 | err = dev_change_net_namespace(dev, &init_net, fb_name); |
ce286d32 | 11358 | if (err) { |
7b6cd1ce JP |
11359 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
11360 | __func__, dev->name, err); | |
aca51397 | 11361 | BUG(); |
ce286d32 EB |
11362 | } |
11363 | } | |
11364 | rtnl_unlock(); | |
11365 | } | |
11366 | ||
50624c93 EB |
11367 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
11368 | { | |
11369 | /* Return with the rtnl_lock held when there are no network | |
11370 | * devices unregistering in any network namespace in net_list. | |
11371 | */ | |
11372 | struct net *net; | |
11373 | bool unregistering; | |
ff960a73 | 11374 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 11375 | |
ff960a73 | 11376 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 11377 | for (;;) { |
50624c93 EB |
11378 | unregistering = false; |
11379 | rtnl_lock(); | |
11380 | list_for_each_entry(net, net_list, exit_list) { | |
11381 | if (net->dev_unreg_count > 0) { | |
11382 | unregistering = true; | |
11383 | break; | |
11384 | } | |
11385 | } | |
11386 | if (!unregistering) | |
11387 | break; | |
11388 | __rtnl_unlock(); | |
ff960a73 PZ |
11389 | |
11390 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 11391 | } |
ff960a73 | 11392 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
11393 | } |
11394 | ||
04dc7f6b EB |
11395 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
11396 | { | |
11397 | /* At exit all network devices most be removed from a network | |
b595076a | 11398 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
11399 | * Do this across as many network namespaces as possible to |
11400 | * improve batching efficiency. | |
11401 | */ | |
11402 | struct net_device *dev; | |
11403 | struct net *net; | |
11404 | LIST_HEAD(dev_kill_list); | |
11405 | ||
50624c93 EB |
11406 | /* To prevent network device cleanup code from dereferencing |
11407 | * loopback devices or network devices that have been freed | |
11408 | * wait here for all pending unregistrations to complete, | |
11409 | * before unregistring the loopback device and allowing the | |
11410 | * network namespace be freed. | |
11411 | * | |
11412 | * The netdev todo list containing all network devices | |
11413 | * unregistrations that happen in default_device_exit_batch | |
11414 | * will run in the rtnl_unlock() at the end of | |
11415 | * default_device_exit_batch. | |
11416 | */ | |
11417 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
11418 | list_for_each_entry(net, net_list, exit_list) { |
11419 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 11420 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
11421 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
11422 | else | |
11423 | unregister_netdevice_queue(dev, &dev_kill_list); | |
11424 | } | |
11425 | } | |
11426 | unregister_netdevice_many(&dev_kill_list); | |
11427 | rtnl_unlock(); | |
11428 | } | |
11429 | ||
022cbae6 | 11430 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 11431 | .exit = default_device_exit, |
04dc7f6b | 11432 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
11433 | }; |
11434 | ||
1da177e4 LT |
11435 | /* |
11436 | * Initialize the DEV module. At boot time this walks the device list and | |
11437 | * unhooks any devices that fail to initialise (normally hardware not | |
11438 | * present) and leaves us with a valid list of present and active devices. | |
11439 | * | |
11440 | */ | |
11441 | ||
11442 | /* | |
11443 | * This is called single threaded during boot, so no need | |
11444 | * to take the rtnl semaphore. | |
11445 | */ | |
11446 | static int __init net_dev_init(void) | |
11447 | { | |
11448 | int i, rc = -ENOMEM; | |
11449 | ||
11450 | BUG_ON(!dev_boot_phase); | |
11451 | ||
1da177e4 LT |
11452 | if (dev_proc_init()) |
11453 | goto out; | |
11454 | ||
8b41d188 | 11455 | if (netdev_kobject_init()) |
1da177e4 LT |
11456 | goto out; |
11457 | ||
11458 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 11459 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
11460 | INIT_LIST_HEAD(&ptype_base[i]); |
11461 | ||
62532da9 VY |
11462 | INIT_LIST_HEAD(&offload_base); |
11463 | ||
881d966b EB |
11464 | if (register_pernet_subsys(&netdev_net_ops)) |
11465 | goto out; | |
1da177e4 LT |
11466 | |
11467 | /* | |
11468 | * Initialise the packet receive queues. | |
11469 | */ | |
11470 | ||
6f912042 | 11471 | for_each_possible_cpu(i) { |
41852497 | 11472 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 11473 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 11474 | |
41852497 ED |
11475 | INIT_WORK(flush, flush_backlog); |
11476 | ||
e36fa2f7 | 11477 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 11478 | skb_queue_head_init(&sd->process_queue); |
f53c7239 SK |
11479 | #ifdef CONFIG_XFRM_OFFLOAD |
11480 | skb_queue_head_init(&sd->xfrm_backlog); | |
11481 | #endif | |
e36fa2f7 | 11482 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 11483 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 11484 | #ifdef CONFIG_RPS |
545b8c8d | 11485 | INIT_CSD(&sd->csd, rps_trigger_softirq, sd); |
e36fa2f7 | 11486 | sd->cpu = i; |
1e94d72f | 11487 | #endif |
0a9627f2 | 11488 | |
7c4ec749 | 11489 | init_gro_hash(&sd->backlog); |
e36fa2f7 ED |
11490 | sd->backlog.poll = process_backlog; |
11491 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
11492 | } |
11493 | ||
1da177e4 LT |
11494 | dev_boot_phase = 0; |
11495 | ||
505d4f73 EB |
11496 | /* The loopback device is special if any other network devices |
11497 | * is present in a network namespace the loopback device must | |
11498 | * be present. Since we now dynamically allocate and free the | |
11499 | * loopback device ensure this invariant is maintained by | |
11500 | * keeping the loopback device as the first device on the | |
11501 | * list of network devices. Ensuring the loopback devices | |
11502 | * is the first device that appears and the last network device | |
11503 | * that disappears. | |
11504 | */ | |
11505 | if (register_pernet_device(&loopback_net_ops)) | |
11506 | goto out; | |
11507 | ||
11508 | if (register_pernet_device(&default_device_ops)) | |
11509 | goto out; | |
11510 | ||
962cf36c CM |
11511 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
11512 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 11513 | |
f0bf90de SAS |
11514 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
11515 | NULL, dev_cpu_dead); | |
11516 | WARN_ON(rc < 0); | |
1da177e4 LT |
11517 | rc = 0; |
11518 | out: | |
11519 | return rc; | |
11520 | } | |
11521 | ||
11522 | subsys_initcall(net_dev_init); |